xref: /openbmc/linux/drivers/scsi/lpfc/lpfc_nvme.c (revision bacf743e)
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 	qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
97 	if (qhandle == NULL)
98 		return -ENOMEM;
99 
100 	qhandle->cpu_id = raw_smp_processor_id();
101 	qhandle->qidx = qidx;
102 	/*
103 	 * NVME qidx == 0 is the admin queue, so both admin queue
104 	 * and first IO queue will use MSI-X vector and associated
105 	 * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
106 	 */
107 	if (qidx) {
108 		str = "IO ";  /* IO queue */
109 		qhandle->index = ((qidx - 1) %
110 			lpfc_nvme_template.max_hw_queues);
111 	} else {
112 		str = "ADM";  /* Admin queue */
113 		qhandle->index = qidx;
114 	}
115 
116 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
117 			 "6073 Binding %s HdwQueue %d  (cpu %d) to "
118 			 "hdw_queue %d qhandle x%px\n", str,
119 			 qidx, qhandle->cpu_id, qhandle->index, qhandle);
120 	*handle = (void *)qhandle;
121 	return 0;
122 }
123 
124 /**
125  * lpfc_nvme_delete_queue -
126  * @pnvme_lport: Transport localport that LS is to be issued from
127  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
128  * @handle: An opaque driver handle from lpfc_nvme_create_queue
129  *
130  * Driver registers this routine to free
131  * any internal data structures to bind the @qidx to its internal
132  * IO queues.
133  *
134  * Return value :
135  *   0 - Success
136  *   TODO:  What are the failure codes.
137  **/
138 static void
139 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
140 		       unsigned int qidx,
141 		       void *handle)
142 {
143 	struct lpfc_nvme_lport *lport;
144 	struct lpfc_vport *vport;
145 
146 	if (!pnvme_lport->private)
147 		return;
148 
149 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
150 	vport = lport->vport;
151 
152 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
153 			"6001 ENTER.  lpfc_pnvme x%px, qidx x%x qhandle x%px\n",
154 			lport, qidx, handle);
155 	kfree(handle);
156 }
157 
158 static void
159 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
160 {
161 	struct lpfc_nvme_lport *lport = localport->private;
162 
163 	lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME,
164 			 "6173 localport x%px delete complete\n",
165 			 lport);
166 
167 	/* release any threads waiting for the unreg to complete */
168 	if (lport->vport->localport)
169 		complete(lport->lport_unreg_cmp);
170 }
171 
172 /* lpfc_nvme_remoteport_delete
173  *
174  * @remoteport: Pointer to an nvme transport remoteport instance.
175  *
176  * This is a template downcall.  NVME transport calls this function
177  * when it has completed the unregistration of a previously
178  * registered remoteport.
179  *
180  * Return value :
181  * None
182  */
183 static void
184 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
185 {
186 	struct lpfc_nvme_rport *rport = remoteport->private;
187 	struct lpfc_vport *vport;
188 	struct lpfc_nodelist *ndlp;
189 	u32 fc4_xpt_flags;
190 
191 	ndlp = rport->ndlp;
192 	if (!ndlp) {
193 		pr_err("**** %s: NULL ndlp on rport x%px remoteport x%px\n",
194 		       __func__, rport, remoteport);
195 		goto rport_err;
196 	}
197 
198 	vport = ndlp->vport;
199 	if (!vport) {
200 		pr_err("**** %s: Null vport on ndlp x%px, ste x%x rport x%px\n",
201 		       __func__, ndlp, ndlp->nlp_state, rport);
202 		goto rport_err;
203 	}
204 
205 	fc4_xpt_flags = NVME_XPT_REGD | SCSI_XPT_REGD;
206 
207 	/* Remove this rport from the lport's list - memory is owned by the
208 	 * transport. Remove the ndlp reference for the NVME transport before
209 	 * calling state machine to remove the node.
210 	 */
211 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
212 			 "6146 remoteport delete of remoteport x%px, ndlp x%px "
213 			 "DID x%x xflags x%x\n",
214 			 remoteport, ndlp, ndlp->nlp_DID, ndlp->fc4_xpt_flags);
215 	spin_lock_irq(&ndlp->lock);
216 
217 	/* The register rebind might have occurred before the delete
218 	 * downcall.  Guard against this race.
219 	 */
220 	if (ndlp->fc4_xpt_flags & NVME_XPT_UNREG_WAIT)
221 		ndlp->fc4_xpt_flags &= ~(NVME_XPT_UNREG_WAIT | NVME_XPT_REGD);
222 
223 	spin_unlock_irq(&ndlp->lock);
224 
225 	/* On a devloss timeout event, one more put is executed provided the
226 	 * NVME and SCSI rport unregister requests are complete.  If the vport
227 	 * is unloading, this extra put is executed by lpfc_drop_node.
228 	 */
229 	if (!(ndlp->fc4_xpt_flags & fc4_xpt_flags))
230 		lpfc_disc_state_machine(vport, ndlp, NULL, NLP_EVT_DEVICE_RM);
231 
232  rport_err:
233 	return;
234 }
235 
236 /**
237  * lpfc_nvme_handle_lsreq - Process an unsolicited NVME LS request
238  * @phba: pointer to lpfc hba data structure.
239  * @axchg: pointer to exchange context for the NVME LS request
240  *
241  * This routine is used for processing an asychronously received NVME LS
242  * request. Any remaining validation is done and the LS is then forwarded
243  * to the nvme-fc transport via nvme_fc_rcv_ls_req().
244  *
245  * The calling sequence should be: nvme_fc_rcv_ls_req() -> (processing)
246  * -> lpfc_nvme_xmt_ls_rsp/cmp -> req->done.
247  * __lpfc_nvme_xmt_ls_rsp_cmp should free the allocated axchg.
248  *
249  * Returns 0 if LS was handled and delivered to the transport
250  * Returns 1 if LS failed to be handled and should be dropped
251  */
252 int
253 lpfc_nvme_handle_lsreq(struct lpfc_hba *phba,
254 			struct lpfc_async_xchg_ctx *axchg)
255 {
256 #if (IS_ENABLED(CONFIG_NVME_FC))
257 	struct lpfc_vport *vport;
258 	struct lpfc_nvme_rport *lpfc_rport;
259 	struct nvme_fc_remote_port *remoteport;
260 	struct lpfc_nvme_lport *lport;
261 	uint32_t *payload = axchg->payload;
262 	int rc;
263 
264 	vport = axchg->ndlp->vport;
265 	lpfc_rport = axchg->ndlp->nrport;
266 	if (!lpfc_rport)
267 		return -EINVAL;
268 
269 	remoteport = lpfc_rport->remoteport;
270 	if (!vport->localport)
271 		return -EINVAL;
272 
273 	lport = vport->localport->private;
274 	if (!lport)
275 		return -EINVAL;
276 
277 	rc = nvme_fc_rcv_ls_req(remoteport, &axchg->ls_rsp, axchg->payload,
278 				axchg->size);
279 
280 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_DISC,
281 			"6205 NVME Unsol rcv: sz %d rc %d: %08x %08x %08x "
282 			"%08x %08x %08x\n",
283 			axchg->size, rc,
284 			*payload, *(payload+1), *(payload+2),
285 			*(payload+3), *(payload+4), *(payload+5));
286 
287 	if (!rc)
288 		return 0;
289 #endif
290 	return 1;
291 }
292 
293 /**
294  * __lpfc_nvme_ls_req_cmp - Generic completion handler for a NVME
295  *        LS request.
296  * @phba: Pointer to HBA context object
297  * @vport: The local port that issued the LS
298  * @cmdwqe: Pointer to driver command WQE object.
299  * @wcqe: Pointer to driver response CQE object.
300  *
301  * This function is the generic completion handler for NVME LS requests.
302  * The function updates any states and statistics, calls the transport
303  * ls_req done() routine, then tears down the command and buffers used
304  * for the LS request.
305  **/
306 void
307 __lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba,  struct lpfc_vport *vport,
308 			struct lpfc_iocbq *cmdwqe,
309 			struct lpfc_wcqe_complete *wcqe)
310 {
311 	struct nvmefc_ls_req *pnvme_lsreq;
312 	struct lpfc_dmabuf *buf_ptr;
313 	struct lpfc_nodelist *ndlp;
314 	uint32_t status;
315 
316 	pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
317 	ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
318 	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
319 
320 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
321 			 "6047 NVMEx LS REQ x%px cmpl DID %x Xri: %x "
322 			 "status %x reason x%x cmd:x%px lsreg:x%px bmp:x%px "
323 			 "ndlp:x%px\n",
324 			 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
325 			 cmdwqe->sli4_xritag, status,
326 			 (wcqe->parameter & 0xffff),
327 			 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
328 
329 	lpfc_nvmeio_data(phba, "NVMEx LS CMPL: xri x%x stat x%x parm x%x\n",
330 			 cmdwqe->sli4_xritag, status, wcqe->parameter);
331 
332 	if (cmdwqe->context3) {
333 		buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
334 		lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
335 		kfree(buf_ptr);
336 		cmdwqe->context3 = NULL;
337 	}
338 	if (pnvme_lsreq->done)
339 		pnvme_lsreq->done(pnvme_lsreq, status);
340 	else
341 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
342 				 "6046 NVMEx cmpl without done call back? "
343 				 "Data x%px DID %x Xri: %x status %x\n",
344 				pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
345 				cmdwqe->sli4_xritag, status);
346 	if (ndlp) {
347 		lpfc_nlp_put(ndlp);
348 		cmdwqe->context1 = NULL;
349 	}
350 	lpfc_sli_release_iocbq(phba, cmdwqe);
351 }
352 
353 static void
354 lpfc_nvme_ls_req_cmp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
355 		     struct lpfc_iocbq *rspwqe)
356 {
357 	struct lpfc_vport *vport = cmdwqe->vport;
358 	struct lpfc_nvme_lport *lport;
359 	uint32_t status;
360 	struct lpfc_wcqe_complete *wcqe = &rspwqe->wcqe_cmpl;
361 
362 	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
363 
364 	if (vport->localport) {
365 		lport = (struct lpfc_nvme_lport *)vport->localport->private;
366 		if (lport) {
367 			atomic_inc(&lport->fc4NvmeLsCmpls);
368 			if (status) {
369 				if (bf_get(lpfc_wcqe_c_xb, wcqe))
370 					atomic_inc(&lport->cmpl_ls_xb);
371 				atomic_inc(&lport->cmpl_ls_err);
372 			}
373 		}
374 	}
375 
376 	__lpfc_nvme_ls_req_cmp(phba, vport, cmdwqe, wcqe);
377 }
378 
379 static int
380 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
381 		  struct lpfc_dmabuf *inp,
382 		  struct nvmefc_ls_req *pnvme_lsreq,
383 		  void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
384 			       struct lpfc_iocbq *),
385 		  struct lpfc_nodelist *ndlp, uint32_t num_entry,
386 		  uint32_t tmo, uint8_t retry)
387 {
388 	struct lpfc_hba *phba = vport->phba;
389 	union lpfc_wqe128 *wqe;
390 	struct lpfc_iocbq *genwqe;
391 	struct ulp_bde64 *bpl;
392 	struct ulp_bde64 bde;
393 	int i, rc, xmit_len, first_len;
394 
395 	/* Allocate buffer for  command WQE */
396 	genwqe = lpfc_sli_get_iocbq(phba);
397 	if (genwqe == NULL)
398 		return 1;
399 
400 	wqe = &genwqe->wqe;
401 	/* Initialize only 64 bytes */
402 	memset(wqe, 0, sizeof(union lpfc_wqe));
403 
404 	genwqe->context3 = (uint8_t *)bmp;
405 	genwqe->cmd_flag |= LPFC_IO_NVME_LS;
406 
407 	/* Save for completion so we can release these resources */
408 	genwqe->context1 = lpfc_nlp_get(ndlp);
409 	if (!genwqe->context1) {
410 		dev_warn(&phba->pcidev->dev,
411 			 "Warning: Failed node ref, not sending LS_REQ\n");
412 		lpfc_sli_release_iocbq(phba, genwqe);
413 		return 1;
414 	}
415 
416 	genwqe->context2 = (uint8_t *)pnvme_lsreq;
417 	/* Fill in payload, bp points to frame payload */
418 
419 	if (!tmo)
420 		/* FC spec states we need 3 * ratov for CT requests */
421 		tmo = (3 * phba->fc_ratov);
422 
423 	/* For this command calculate the xmit length of the request bde. */
424 	xmit_len = 0;
425 	first_len = 0;
426 	bpl = (struct ulp_bde64 *)bmp->virt;
427 	for (i = 0; i < num_entry; i++) {
428 		bde.tus.w = bpl[i].tus.w;
429 		if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
430 			break;
431 		xmit_len += bde.tus.f.bdeSize;
432 		if (i == 0)
433 			first_len = xmit_len;
434 	}
435 
436 	genwqe->num_bdes = num_entry;
437 	genwqe->hba_wqidx = 0;
438 
439 	/* Words 0 - 2 */
440 	wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
441 	wqe->generic.bde.tus.f.bdeSize = first_len;
442 	wqe->generic.bde.addrLow = bpl[0].addrLow;
443 	wqe->generic.bde.addrHigh = bpl[0].addrHigh;
444 
445 	/* Word 3 */
446 	wqe->gen_req.request_payload_len = first_len;
447 
448 	/* Word 4 */
449 
450 	/* Word 5 */
451 	bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
452 	bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
453 	bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
454 	bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
455 	bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
456 
457 	/* Word 6 */
458 	bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
459 	       phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
460 	bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
461 
462 	/* Word 7 */
463 	bf_set(wqe_tmo, &wqe->gen_req.wqe_com, tmo);
464 	bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
465 	bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
466 	bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
467 
468 	/* Word 8 */
469 	wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
470 
471 	/* Word 9 */
472 	bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
473 
474 	/* Word 10 */
475 	bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
476 	bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
477 	bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
478 	bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
479 	bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
480 
481 	/* Word 11 */
482 	bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
483 	bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
484 
485 
486 	/* Issue GEN REQ WQE for NPORT <did> */
487 	genwqe->cmd_cmpl = cmpl;
488 	genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
489 	genwqe->vport = vport;
490 	genwqe->retry = retry;
491 
492 	lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
493 			 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
494 
495 	rc = lpfc_sli4_issue_wqe(phba, &phba->sli4_hba.hdwq[0], genwqe);
496 	if (rc) {
497 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
498 				 "6045 Issue GEN REQ WQE to NPORT x%x "
499 				 "Data: x%x x%x  rc x%x\n",
500 				 ndlp->nlp_DID, genwqe->iotag,
501 				 vport->port_state, rc);
502 		lpfc_nlp_put(ndlp);
503 		lpfc_sli_release_iocbq(phba, genwqe);
504 		return 1;
505 	}
506 
507 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_ELS,
508 			 "6050 Issue GEN REQ WQE to NPORT x%x "
509 			 "Data: oxid: x%x state: x%x wq:x%px lsreq:x%px "
510 			 "bmp:x%px xmit:%d 1st:%d\n",
511 			 ndlp->nlp_DID, genwqe->sli4_xritag,
512 			 vport->port_state,
513 			 genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
514 	return 0;
515 }
516 
517 
518 /**
519  * __lpfc_nvme_ls_req - Generic service routine to issue an NVME LS request
520  * @vport: The local port issuing the LS
521  * @ndlp: The remote port to send the LS to
522  * @pnvme_lsreq: Pointer to LS request structure from the transport
523  * @gen_req_cmp: Completion call-back
524  *
525  * Routine validates the ndlp, builds buffers and sends a GEN_REQUEST
526  * WQE to perform the LS operation.
527  *
528  * Return value :
529  *   0 - Success
530  *   non-zero: various error codes, in form of -Exxx
531  **/
532 int
533 __lpfc_nvme_ls_req(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
534 		      struct nvmefc_ls_req *pnvme_lsreq,
535 		      void (*gen_req_cmp)(struct lpfc_hba *phba,
536 				struct lpfc_iocbq *cmdwqe,
537 				struct lpfc_iocbq *rspwqe))
538 {
539 	struct lpfc_dmabuf *bmp;
540 	struct ulp_bde64 *bpl;
541 	int ret;
542 	uint16_t ntype, nstate;
543 
544 	if (!ndlp) {
545 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
546 				 "6051 NVMEx LS REQ: Bad NDLP x%px, Failing "
547 				 "LS Req\n",
548 				 ndlp);
549 		return -ENODEV;
550 	}
551 
552 	ntype = ndlp->nlp_type;
553 	nstate = ndlp->nlp_state;
554 	if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
555 	    (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
556 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
557 				 "6088 NVMEx LS REQ: Fail DID x%06x not "
558 				 "ready for IO. Type x%x, State x%x\n",
559 				 ndlp->nlp_DID, ntype, nstate);
560 		return -ENODEV;
561 	}
562 
563 	if (!vport->phba->sli4_hba.nvmels_wq)
564 		return -ENOMEM;
565 
566 	/*
567 	 * there are two dma buf in the request, actually there is one and
568 	 * the second one is just the start address + cmd size.
569 	 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
570 	 * in a lpfc_dmabuf struct. When freeing we just free the wrapper
571 	 * because the nvem layer owns the data bufs.
572 	 * We do not have to break these packets open, we don't care what is
573 	 * in them. And we do not have to look at the resonse data, we only
574 	 * care that we got a response. All of the caring is going to happen
575 	 * in the nvme-fc layer.
576 	 */
577 
578 	bmp = kmalloc(sizeof(*bmp), GFP_KERNEL);
579 	if (!bmp) {
580 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
581 				 "6044 NVMEx LS REQ: Could not alloc LS buf "
582 				 "for DID %x\n",
583 				 ndlp->nlp_DID);
584 		return -ENOMEM;
585 	}
586 
587 	bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
588 	if (!bmp->virt) {
589 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
590 				 "6042 NVMEx LS REQ: Could not alloc mbuf "
591 				 "for DID %x\n",
592 				 ndlp->nlp_DID);
593 		kfree(bmp);
594 		return -ENOMEM;
595 	}
596 
597 	INIT_LIST_HEAD(&bmp->list);
598 
599 	bpl = (struct ulp_bde64 *)bmp->virt;
600 	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
601 	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
602 	bpl->tus.f.bdeFlags = 0;
603 	bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
604 	bpl->tus.w = le32_to_cpu(bpl->tus.w);
605 	bpl++;
606 
607 	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
608 	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
609 	bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
610 	bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
611 	bpl->tus.w = le32_to_cpu(bpl->tus.w);
612 
613 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
614 			"6149 NVMEx LS REQ: Issue to DID 0x%06x lsreq x%px, "
615 			"rqstlen:%d rsplen:%d %pad %pad\n",
616 			ndlp->nlp_DID, pnvme_lsreq, pnvme_lsreq->rqstlen,
617 			pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
618 			&pnvme_lsreq->rspdma);
619 
620 	ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
621 				pnvme_lsreq, gen_req_cmp, ndlp, 2,
622 				pnvme_lsreq->timeout, 0);
623 	if (ret != WQE_SUCCESS) {
624 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
625 				 "6052 NVMEx REQ: EXIT. issue ls wqe failed "
626 				 "lsreq x%px Status %x DID %x\n",
627 				 pnvme_lsreq, ret, ndlp->nlp_DID);
628 		lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
629 		kfree(bmp);
630 		return -EIO;
631 	}
632 
633 	return 0;
634 }
635 
636 /**
637  * lpfc_nvme_ls_req - Issue an NVME Link Service request
638  * @pnvme_lport: Transport localport that LS is to be issued from.
639  * @pnvme_rport: Transport remoteport that LS is to be sent to.
640  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
641  *
642  * Driver registers this routine to handle any link service request
643  * from the nvme_fc transport to a remote nvme-aware port.
644  *
645  * Return value :
646  *   0 - Success
647  *   non-zero: various error codes, in form of -Exxx
648  **/
649 static int
650 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
651 		 struct nvme_fc_remote_port *pnvme_rport,
652 		 struct nvmefc_ls_req *pnvme_lsreq)
653 {
654 	struct lpfc_nvme_lport *lport;
655 	struct lpfc_nvme_rport *rport;
656 	struct lpfc_vport *vport;
657 	int ret;
658 
659 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
660 	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
661 	if (unlikely(!lport) || unlikely(!rport))
662 		return -EINVAL;
663 
664 	vport = lport->vport;
665 	if (vport->load_flag & FC_UNLOADING)
666 		return -ENODEV;
667 
668 	atomic_inc(&lport->fc4NvmeLsRequests);
669 
670 	ret = __lpfc_nvme_ls_req(vport, rport->ndlp, pnvme_lsreq,
671 				 lpfc_nvme_ls_req_cmp);
672 	if (ret)
673 		atomic_inc(&lport->xmt_ls_err);
674 
675 	return ret;
676 }
677 
678 /**
679  * __lpfc_nvme_ls_abort - Generic service routine to abort a prior
680  *         NVME LS request
681  * @vport: The local port that issued the LS
682  * @ndlp: The remote port the LS was sent to
683  * @pnvme_lsreq: Pointer to LS request structure from the transport
684  *
685  * The driver validates the ndlp, looks for the LS, and aborts the
686  * LS if found.
687  *
688  * Returns:
689  * 0 : if LS found and aborted
690  * non-zero: various error conditions in form -Exxx
691  **/
692 int
693 __lpfc_nvme_ls_abort(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
694 			struct nvmefc_ls_req *pnvme_lsreq)
695 {
696 	struct lpfc_hba *phba = vport->phba;
697 	struct lpfc_sli_ring *pring;
698 	struct lpfc_iocbq *wqe, *next_wqe;
699 	bool foundit = false;
700 
701 	if (!ndlp) {
702 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
703 				"6049 NVMEx LS REQ Abort: Bad NDLP x%px DID "
704 				"x%06x, Failing LS Req\n",
705 				ndlp, ndlp ? ndlp->nlp_DID : 0);
706 		return -EINVAL;
707 	}
708 
709 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
710 			 "6040 NVMEx LS REQ Abort: Issue LS_ABORT for lsreq "
711 			 "x%px rqstlen:%d rsplen:%d %pad %pad\n",
712 			 pnvme_lsreq, pnvme_lsreq->rqstlen,
713 			 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
714 			 &pnvme_lsreq->rspdma);
715 
716 	/*
717 	 * Lock the ELS ring txcmplq and look for the wqe that matches
718 	 * this ELS. If found, issue an abort on the wqe.
719 	 */
720 	pring = phba->sli4_hba.nvmels_wq->pring;
721 	spin_lock_irq(&phba->hbalock);
722 	spin_lock(&pring->ring_lock);
723 	list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
724 		if (wqe->context2 == pnvme_lsreq) {
725 			wqe->cmd_flag |= LPFC_DRIVER_ABORTED;
726 			foundit = true;
727 			break;
728 		}
729 	}
730 	spin_unlock(&pring->ring_lock);
731 
732 	if (foundit)
733 		lpfc_sli_issue_abort_iotag(phba, pring, wqe, NULL);
734 	spin_unlock_irq(&phba->hbalock);
735 
736 	if (foundit)
737 		return 0;
738 
739 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC | LOG_NVME_ABTS,
740 			 "6213 NVMEx LS REQ Abort: Unable to locate req x%px\n",
741 			 pnvme_lsreq);
742 	return -EINVAL;
743 }
744 
745 static int
746 lpfc_nvme_xmt_ls_rsp(struct nvme_fc_local_port *localport,
747 		     struct nvme_fc_remote_port *remoteport,
748 		     struct nvmefc_ls_rsp *ls_rsp)
749 {
750 	struct lpfc_async_xchg_ctx *axchg =
751 		container_of(ls_rsp, struct lpfc_async_xchg_ctx, ls_rsp);
752 	struct lpfc_nvme_lport *lport;
753 	int rc;
754 
755 	if (axchg->phba->pport->load_flag & FC_UNLOADING)
756 		return -ENODEV;
757 
758 	lport = (struct lpfc_nvme_lport *)localport->private;
759 
760 	rc = __lpfc_nvme_xmt_ls_rsp(axchg, ls_rsp, __lpfc_nvme_xmt_ls_rsp_cmp);
761 
762 	if (rc) {
763 		/*
764 		 * unless the failure is due to having already sent
765 		 * the response, an abort will be generated for the
766 		 * exchange if the rsp can't be sent.
767 		 */
768 		if (rc != -EALREADY)
769 			atomic_inc(&lport->xmt_ls_abort);
770 		return rc;
771 	}
772 
773 	return 0;
774 }
775 
776 /**
777  * lpfc_nvme_ls_abort - Abort a prior NVME LS request
778  * @pnvme_lport: Transport localport that LS is to be issued from.
779  * @pnvme_rport: Transport remoteport that LS is to be sent to.
780  * @pnvme_lsreq: the transport nvme_ls_req structure for the LS
781  *
782  * Driver registers this routine to abort a NVME LS request that is
783  * in progress (from the transports perspective).
784  **/
785 static void
786 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
787 		   struct nvme_fc_remote_port *pnvme_rport,
788 		   struct nvmefc_ls_req *pnvme_lsreq)
789 {
790 	struct lpfc_nvme_lport *lport;
791 	struct lpfc_vport *vport;
792 	struct lpfc_nodelist *ndlp;
793 	int ret;
794 
795 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
796 	if (unlikely(!lport))
797 		return;
798 	vport = lport->vport;
799 
800 	if (vport->load_flag & FC_UNLOADING)
801 		return;
802 
803 	ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
804 
805 	ret = __lpfc_nvme_ls_abort(vport, ndlp, pnvme_lsreq);
806 	if (!ret)
807 		atomic_inc(&lport->xmt_ls_abort);
808 }
809 
810 /* Fix up the existing sgls for NVME IO. */
811 static inline void
812 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
813 		       struct lpfc_io_buf *lpfc_ncmd,
814 		       struct nvmefc_fcp_req *nCmd)
815 {
816 	struct lpfc_hba  *phba = vport->phba;
817 	struct sli4_sge *sgl;
818 	union lpfc_wqe128 *wqe;
819 	uint32_t *wptr, *dptr;
820 
821 	/*
822 	 * Get a local pointer to the built-in wqe and correct
823 	 * the cmd size to match NVME's 96 bytes and fix
824 	 * the dma address.
825 	 */
826 
827 	wqe = &lpfc_ncmd->cur_iocbq.wqe;
828 
829 	/*
830 	 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
831 	 * match NVME.  NVME sends 96 bytes. Also, use the
832 	 * nvme commands command and response dma addresses
833 	 * rather than the virtual memory to ease the restore
834 	 * operation.
835 	 */
836 	sgl = lpfc_ncmd->dma_sgl;
837 	sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
838 	if (phba->cfg_nvme_embed_cmd) {
839 		sgl->addr_hi = 0;
840 		sgl->addr_lo = 0;
841 
842 		/* Word 0-2 - NVME CMND IU (embedded payload) */
843 		wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
844 		wqe->generic.bde.tus.f.bdeSize = 56;
845 		wqe->generic.bde.addrHigh = 0;
846 		wqe->generic.bde.addrLow =  64;  /* Word 16 */
847 
848 		/* Word 10  - dbde is 0, wqes is 1 in template */
849 
850 		/*
851 		 * Embed the payload in the last half of the WQE
852 		 * WQE words 16-30 get the NVME CMD IU payload
853 		 *
854 		 * WQE words 16-19 get payload Words 1-4
855 		 * WQE words 20-21 get payload Words 6-7
856 		 * WQE words 22-29 get payload Words 16-23
857 		 */
858 		wptr = &wqe->words[16];  /* WQE ptr */
859 		dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
860 		dptr++;			/* Skip Word 0 in payload */
861 
862 		*wptr++ = *dptr++;	/* Word 1 */
863 		*wptr++ = *dptr++;	/* Word 2 */
864 		*wptr++ = *dptr++;	/* Word 3 */
865 		*wptr++ = *dptr++;	/* Word 4 */
866 		dptr++;			/* Skip Word 5 in payload */
867 		*wptr++ = *dptr++;	/* Word 6 */
868 		*wptr++ = *dptr++;	/* Word 7 */
869 		dptr += 8;		/* Skip Words 8-15 in payload */
870 		*wptr++ = *dptr++;	/* Word 16 */
871 		*wptr++ = *dptr++;	/* Word 17 */
872 		*wptr++ = *dptr++;	/* Word 18 */
873 		*wptr++ = *dptr++;	/* Word 19 */
874 		*wptr++ = *dptr++;	/* Word 20 */
875 		*wptr++ = *dptr++;	/* Word 21 */
876 		*wptr++ = *dptr++;	/* Word 22 */
877 		*wptr   = *dptr;	/* Word 23 */
878 	} else {
879 		sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
880 		sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
881 
882 		/* Word 0-2 - NVME CMND IU Inline BDE */
883 		wqe->generic.bde.tus.f.bdeFlags =  BUFF_TYPE_BDE_64;
884 		wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
885 		wqe->generic.bde.addrHigh = sgl->addr_hi;
886 		wqe->generic.bde.addrLow =  sgl->addr_lo;
887 
888 		/* Word 10 */
889 		bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
890 		bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
891 	}
892 
893 	sgl++;
894 
895 	/* Setup the physical region for the FCP RSP */
896 	sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
897 	sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
898 	sgl->word2 = le32_to_cpu(sgl->word2);
899 	if (nCmd->sg_cnt)
900 		bf_set(lpfc_sli4_sge_last, sgl, 0);
901 	else
902 		bf_set(lpfc_sli4_sge_last, sgl, 1);
903 	sgl->word2 = cpu_to_le32(sgl->word2);
904 	sgl->sge_len = cpu_to_le32(nCmd->rsplen);
905 }
906 
907 
908 /*
909  * lpfc_nvme_io_cmd_cmpl - Complete an NVME-over-FCP IO
910  *
911  * Driver registers this routine as it io request handler.  This
912  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
913  * data structure to the rport indicated in @lpfc_nvme_rport.
914  *
915  * Return value :
916  *   0 - Success
917  *   TODO: What are the failure codes.
918  **/
919 static void
920 lpfc_nvme_io_cmd_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
921 		      struct lpfc_iocbq *pwqeOut)
922 {
923 	struct lpfc_io_buf *lpfc_ncmd =
924 		(struct lpfc_io_buf *)pwqeIn->context1;
925 	struct lpfc_wcqe_complete *wcqe = &pwqeOut->wcqe_cmpl;
926 	struct lpfc_vport *vport = pwqeIn->vport;
927 	struct nvmefc_fcp_req *nCmd;
928 	struct nvme_fc_ersp_iu *ep;
929 	struct nvme_fc_cmd_iu *cp;
930 	struct lpfc_nodelist *ndlp;
931 	struct lpfc_nvme_fcpreq_priv *freqpriv;
932 	struct lpfc_nvme_lport *lport;
933 	uint32_t code, status, idx;
934 	uint16_t cid, sqhd, data;
935 	uint32_t *ptr;
936 	uint32_t lat;
937 	bool call_done = false;
938 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
939 	int cpu;
940 #endif
941 	int offline = 0;
942 
943 	/* Sanity check on return of outstanding command */
944 	if (!lpfc_ncmd) {
945 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
946 				 "6071 Null lpfc_ncmd pointer. No "
947 				 "release, skip completion\n");
948 		return;
949 	}
950 
951 	/* Guard against abort handler being called at same time */
952 	spin_lock(&lpfc_ncmd->buf_lock);
953 
954 	if (!lpfc_ncmd->nvmeCmd) {
955 		spin_unlock(&lpfc_ncmd->buf_lock);
956 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
957 				 "6066 Missing cmpl ptrs: lpfc_ncmd x%px, "
958 				 "nvmeCmd x%px\n",
959 				 lpfc_ncmd, lpfc_ncmd->nvmeCmd);
960 
961 		/* Release the lpfc_ncmd regardless of the missing elements. */
962 		lpfc_release_nvme_buf(phba, lpfc_ncmd);
963 		return;
964 	}
965 	nCmd = lpfc_ncmd->nvmeCmd;
966 	status = bf_get(lpfc_wcqe_c_status, wcqe);
967 
968 	idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
969 	phba->sli4_hba.hdwq[idx].nvme_cstat.io_cmpls++;
970 
971 	if (unlikely(status && vport->localport)) {
972 		lport = (struct lpfc_nvme_lport *)vport->localport->private;
973 		if (lport) {
974 			if (bf_get(lpfc_wcqe_c_xb, wcqe))
975 				atomic_inc(&lport->cmpl_fcp_xb);
976 			atomic_inc(&lport->cmpl_fcp_err);
977 		}
978 	}
979 
980 	lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
981 			 lpfc_ncmd->cur_iocbq.sli4_xritag,
982 			 status, wcqe->parameter);
983 	/*
984 	 * Catch race where our node has transitioned, but the
985 	 * transport is still transitioning.
986 	 */
987 	ndlp = lpfc_ncmd->ndlp;
988 	if (!ndlp) {
989 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
990 				 "6062 Ignoring NVME cmpl.  No ndlp\n");
991 		goto out_err;
992 	}
993 
994 	code = bf_get(lpfc_wcqe_c_code, wcqe);
995 	if (code == CQE_CODE_NVME_ERSP) {
996 		/* For this type of CQE, we need to rebuild the rsp */
997 		ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
998 
999 		/*
1000 		 * Get Command Id from cmd to plug into response. This
1001 		 * code is not needed in the next NVME Transport drop.
1002 		 */
1003 		cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1004 		cid = cp->sqe.common.command_id;
1005 
1006 		/*
1007 		 * RSN is in CQE word 2
1008 		 * SQHD is in CQE Word 3 bits 15:0
1009 		 * Cmd Specific info is in CQE Word 1
1010 		 * and in CQE Word 0 bits 15:0
1011 		 */
1012 		sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1013 
1014 		/* Now lets build the NVME ERSP IU */
1015 		ep->iu_len = cpu_to_be16(8);
1016 		ep->rsn = wcqe->parameter;
1017 		ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1018 		ep->rsvd12 = 0;
1019 		ptr = (uint32_t *)&ep->cqe.result.u64;
1020 		*ptr++ = wcqe->total_data_placed;
1021 		data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1022 		*ptr = (uint32_t)data;
1023 		ep->cqe.sq_head = sqhd;
1024 		ep->cqe.sq_id =  nCmd->sqid;
1025 		ep->cqe.command_id = cid;
1026 		ep->cqe.status = 0;
1027 
1028 		lpfc_ncmd->status = IOSTAT_SUCCESS;
1029 		lpfc_ncmd->result = 0;
1030 		nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1031 		nCmd->transferred_length = nCmd->payload_length;
1032 	} else {
1033 		lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1034 		lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1035 
1036 		/* For NVME, the only failure path that results in an
1037 		 * IO error is when the adapter rejects it.  All other
1038 		 * conditions are a success case and resolved by the
1039 		 * transport.
1040 		 * IOSTAT_FCP_RSP_ERROR means:
1041 		 * 1. Length of data received doesn't match total
1042 		 *    transfer length in WQE
1043 		 * 2. If the RSP payload does NOT match these cases:
1044 		 *    a. RSP length 12/24 bytes and all zeros
1045 		 *    b. NVME ERSP
1046 		 */
1047 		switch (lpfc_ncmd->status) {
1048 		case IOSTAT_SUCCESS:
1049 			nCmd->transferred_length = wcqe->total_data_placed;
1050 			nCmd->rcv_rsplen = 0;
1051 			nCmd->status = 0;
1052 			break;
1053 		case IOSTAT_FCP_RSP_ERROR:
1054 			nCmd->transferred_length = wcqe->total_data_placed;
1055 			nCmd->rcv_rsplen = wcqe->parameter;
1056 			nCmd->status = 0;
1057 
1058 			/* Check if this is really an ERSP */
1059 			if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) {
1060 				lpfc_ncmd->status = IOSTAT_SUCCESS;
1061 				lpfc_ncmd->result = 0;
1062 
1063 				lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
1064 					 "6084 NVME Completion ERSP: "
1065 					 "xri %x placed x%x\n",
1066 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1067 					 wcqe->total_data_placed);
1068 				break;
1069 			}
1070 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1071 					 "6081 NVME Completion Protocol Error: "
1072 					 "xri %x status x%x result x%x "
1073 					 "placed x%x\n",
1074 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1075 					 lpfc_ncmd->status, lpfc_ncmd->result,
1076 					 wcqe->total_data_placed);
1077 			break;
1078 		case IOSTAT_LOCAL_REJECT:
1079 			/* Let fall through to set command final state. */
1080 			if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1081 				lpfc_printf_vlog(vport, KERN_INFO,
1082 					 LOG_NVME_IOERR,
1083 					 "6032 Delay Aborted cmd x%px "
1084 					 "nvme cmd x%px, xri x%x, "
1085 					 "xb %d\n",
1086 					 lpfc_ncmd, nCmd,
1087 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1088 					 bf_get(lpfc_wcqe_c_xb, wcqe));
1089 			fallthrough;
1090 		default:
1091 out_err:
1092 			lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1093 					 "6072 NVME Completion Error: xri %x "
1094 					 "status x%x result x%x [x%x] "
1095 					 "placed x%x\n",
1096 					 lpfc_ncmd->cur_iocbq.sli4_xritag,
1097 					 lpfc_ncmd->status, lpfc_ncmd->result,
1098 					 wcqe->parameter,
1099 					 wcqe->total_data_placed);
1100 			nCmd->transferred_length = 0;
1101 			nCmd->rcv_rsplen = 0;
1102 			nCmd->status = NVME_SC_INTERNAL;
1103 			offline = pci_channel_offline(vport->phba->pcidev);
1104 		}
1105 	}
1106 
1107 	/* pick up SLI4 exhange busy condition */
1108 	if (bf_get(lpfc_wcqe_c_xb, wcqe) && !offline)
1109 		lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1110 	else
1111 		lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1112 
1113 	/* Update stats and complete the IO.  There is
1114 	 * no need for dma unprep because the nvme_transport
1115 	 * owns the dma address.
1116 	 */
1117 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1118 	if (lpfc_ncmd->ts_cmd_start) {
1119 		lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1120 		lpfc_ncmd->ts_data_io = ktime_get_ns();
1121 		phba->ktime_last_cmd = lpfc_ncmd->ts_data_io;
1122 		lpfc_io_ktime(phba, lpfc_ncmd);
1123 	}
1124 	if (unlikely(phba->hdwqstat_on & LPFC_CHECK_NVME_IO)) {
1125 		cpu = raw_smp_processor_id();
1126 		this_cpu_inc(phba->sli4_hba.c_stat->cmpl_io);
1127 		if (lpfc_ncmd->cpu != cpu)
1128 			lpfc_printf_vlog(vport,
1129 					 KERN_INFO, LOG_NVME_IOERR,
1130 					 "6701 CPU Check cmpl: "
1131 					 "cpu %d expect %d\n",
1132 					 cpu, lpfc_ncmd->cpu);
1133 	}
1134 #endif
1135 
1136 	/* NVME targets need completion held off until the abort exchange
1137 	 * completes unless the NVME Rport is getting unregistered.
1138 	 */
1139 
1140 	if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1141 		freqpriv = nCmd->private;
1142 		freqpriv->nvme_buf = NULL;
1143 		lpfc_ncmd->nvmeCmd = NULL;
1144 		call_done = true;
1145 	}
1146 	spin_unlock(&lpfc_ncmd->buf_lock);
1147 
1148 	/* Check if IO qualified for CMF */
1149 	if (phba->cmf_active_mode != LPFC_CFG_OFF &&
1150 	    nCmd->io_dir == NVMEFC_FCP_READ &&
1151 	    nCmd->payload_length) {
1152 		/* Used when calculating average latency */
1153 		lat = ktime_get_ns() - lpfc_ncmd->rx_cmd_start;
1154 		lpfc_update_cmf_cmpl(phba, lat, nCmd->payload_length, NULL);
1155 	}
1156 
1157 	if (call_done)
1158 		nCmd->done(nCmd);
1159 
1160 	/* Call release with XB=1 to queue the IO into the abort list. */
1161 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
1162 }
1163 
1164 
1165 /**
1166  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1167  * @vport: pointer to a host virtual N_Port data structure
1168  * @lpfc_ncmd: Pointer to lpfc scsi command
1169  * @pnode: pointer to a node-list data structure
1170  * @cstat: pointer to the control status structure
1171  *
1172  * Driver registers this routine as it io request handler.  This
1173  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1174  * data structure to the rport indicated in @lpfc_nvme_rport.
1175  *
1176  * Return value :
1177  *   0 - Success
1178  *   TODO: What are the failure codes.
1179  **/
1180 static int
1181 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1182 		      struct lpfc_io_buf *lpfc_ncmd,
1183 		      struct lpfc_nodelist *pnode,
1184 		      struct lpfc_fc4_ctrl_stat *cstat)
1185 {
1186 	struct lpfc_hba *phba = vport->phba;
1187 	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1188 	struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
1189 	union lpfc_wqe128 *wqe = &pwqeq->wqe;
1190 	uint32_t req_len;
1191 
1192 	/*
1193 	 * There are three possibilities here - use scatter-gather segment, use
1194 	 * the single mapping, or neither.
1195 	 */
1196 	if (nCmd->sg_cnt) {
1197 		if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1198 			/* From the iwrite template, initialize words 7 - 11 */
1199 			memcpy(&wqe->words[7],
1200 			       &lpfc_iwrite_cmd_template.words[7],
1201 			       sizeof(uint32_t) * 5);
1202 
1203 			/* Word 4 */
1204 			wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1205 
1206 			/* Word 5 */
1207 			if ((phba->cfg_nvme_enable_fb) &&
1208 			    (pnode->nlp_flag & NLP_FIRSTBURST)) {
1209 				req_len = lpfc_ncmd->nvmeCmd->payload_length;
1210 				if (req_len < pnode->nvme_fb_size)
1211 					wqe->fcp_iwrite.initial_xfer_len =
1212 						req_len;
1213 				else
1214 					wqe->fcp_iwrite.initial_xfer_len =
1215 						pnode->nvme_fb_size;
1216 			} else {
1217 				wqe->fcp_iwrite.initial_xfer_len = 0;
1218 			}
1219 			cstat->output_requests++;
1220 		} else {
1221 			/* From the iread template, initialize words 7 - 11 */
1222 			memcpy(&wqe->words[7],
1223 			       &lpfc_iread_cmd_template.words[7],
1224 			       sizeof(uint32_t) * 5);
1225 
1226 			/* Word 4 */
1227 			wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1228 
1229 			/* Word 5 */
1230 			wqe->fcp_iread.rsrvd5 = 0;
1231 
1232 			/* For a CMF Managed port, iod must be zero'ed */
1233 			if (phba->cmf_active_mode == LPFC_CFG_MANAGED)
1234 				bf_set(wqe_iod, &wqe->fcp_iread.wqe_com,
1235 				       LPFC_WQE_IOD_NONE);
1236 			cstat->input_requests++;
1237 		}
1238 	} else {
1239 		/* From the icmnd template, initialize words 4 - 11 */
1240 		memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1241 		       sizeof(uint32_t) * 8);
1242 		cstat->control_requests++;
1243 	}
1244 
1245 	if (pnode->nlp_nvme_info & NLP_NVME_NSLER)
1246 		bf_set(wqe_erp, &wqe->generic.wqe_com, 1);
1247 	/*
1248 	 * Finish initializing those WQE fields that are independent
1249 	 * of the nvme_cmnd request_buffer
1250 	 */
1251 
1252 	/* Word 3 */
1253 	bf_set(payload_offset_len, &wqe->fcp_icmd,
1254 	       (nCmd->rsplen + nCmd->cmdlen));
1255 
1256 	/* Word 6 */
1257 	bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1258 	       phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1259 	bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1260 
1261 	/* Word 8 */
1262 	wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1263 
1264 	/* Word 9 */
1265 	bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1266 
1267 	/* Word 10 */
1268 	bf_set(wqe_xchg, &wqe->fcp_iwrite.wqe_com, LPFC_NVME_XCHG);
1269 
1270 	/* Words 13 14 15 are for PBDE support */
1271 
1272 	pwqeq->vport = vport;
1273 	return 0;
1274 }
1275 
1276 
1277 /**
1278  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1279  * @vport: pointer to a host virtual N_Port data structure
1280  * @lpfc_ncmd: Pointer to lpfc scsi command
1281  *
1282  * Driver registers this routine as it io request handler.  This
1283  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1284  * data structure to the rport indicated in @lpfc_nvme_rport.
1285  *
1286  * Return value :
1287  *   0 - Success
1288  *   TODO: What are the failure codes.
1289  **/
1290 static int
1291 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1292 		      struct lpfc_io_buf *lpfc_ncmd)
1293 {
1294 	struct lpfc_hba *phba = vport->phba;
1295 	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1296 	union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1297 	struct sli4_sge *sgl = lpfc_ncmd->dma_sgl;
1298 	struct sli4_hybrid_sgl *sgl_xtra = NULL;
1299 	struct scatterlist *data_sg;
1300 	struct sli4_sge *first_data_sgl;
1301 	struct ulp_bde64 *bde;
1302 	dma_addr_t physaddr = 0;
1303 	uint32_t dma_len = 0;
1304 	uint32_t dma_offset = 0;
1305 	int nseg, i, j;
1306 	bool lsp_just_set = false;
1307 
1308 	/* Fix up the command and response DMA stuff. */
1309 	lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1310 
1311 	/*
1312 	 * There are three possibilities here - use scatter-gather segment, use
1313 	 * the single mapping, or neither.
1314 	 */
1315 	if (nCmd->sg_cnt) {
1316 		/*
1317 		 * Jump over the cmd and rsp SGEs.  The fix routine
1318 		 * has already adjusted for this.
1319 		 */
1320 		sgl += 2;
1321 
1322 		first_data_sgl = sgl;
1323 		lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1324 		if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1325 			lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1326 					"6058 Too many sg segments from "
1327 					"NVME Transport.  Max %d, "
1328 					"nvmeIO sg_cnt %d\n",
1329 					phba->cfg_nvme_seg_cnt + 1,
1330 					lpfc_ncmd->seg_cnt);
1331 			lpfc_ncmd->seg_cnt = 0;
1332 			return 1;
1333 		}
1334 
1335 		/*
1336 		 * The driver established a maximum scatter-gather segment count
1337 		 * during probe that limits the number of sg elements in any
1338 		 * single nvme command.  Just run through the seg_cnt and format
1339 		 * the sge's.
1340 		 */
1341 		nseg = nCmd->sg_cnt;
1342 		data_sg = nCmd->first_sgl;
1343 
1344 		/* for tracking the segment boundaries */
1345 		j = 2;
1346 		for (i = 0; i < nseg; i++) {
1347 			if (data_sg == NULL) {
1348 				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1349 						"6059 dptr err %d, nseg %d\n",
1350 						i, nseg);
1351 				lpfc_ncmd->seg_cnt = 0;
1352 				return 1;
1353 			}
1354 
1355 			sgl->word2 = 0;
1356 			if (nseg == 1) {
1357 				bf_set(lpfc_sli4_sge_last, sgl, 1);
1358 				bf_set(lpfc_sli4_sge_type, sgl,
1359 				       LPFC_SGE_TYPE_DATA);
1360 			} else {
1361 				bf_set(lpfc_sli4_sge_last, sgl, 0);
1362 
1363 				/* expand the segment */
1364 				if (!lsp_just_set &&
1365 				    !((j + 1) % phba->border_sge_num) &&
1366 				    ((nseg - 1) != i)) {
1367 					/* set LSP type */
1368 					bf_set(lpfc_sli4_sge_type, sgl,
1369 					       LPFC_SGE_TYPE_LSP);
1370 
1371 					sgl_xtra = lpfc_get_sgl_per_hdwq(
1372 							phba, lpfc_ncmd);
1373 
1374 					if (unlikely(!sgl_xtra)) {
1375 						lpfc_ncmd->seg_cnt = 0;
1376 						return 1;
1377 					}
1378 					sgl->addr_lo = cpu_to_le32(putPaddrLow(
1379 						       sgl_xtra->dma_phys_sgl));
1380 					sgl->addr_hi = cpu_to_le32(putPaddrHigh(
1381 						       sgl_xtra->dma_phys_sgl));
1382 
1383 				} else {
1384 					bf_set(lpfc_sli4_sge_type, sgl,
1385 					       LPFC_SGE_TYPE_DATA);
1386 				}
1387 			}
1388 
1389 			if (!(bf_get(lpfc_sli4_sge_type, sgl) &
1390 				     LPFC_SGE_TYPE_LSP)) {
1391 				if ((nseg - 1) == i)
1392 					bf_set(lpfc_sli4_sge_last, sgl, 1);
1393 
1394 				physaddr = data_sg->dma_address;
1395 				dma_len = data_sg->length;
1396 				sgl->addr_lo = cpu_to_le32(
1397 							 putPaddrLow(physaddr));
1398 				sgl->addr_hi = cpu_to_le32(
1399 							putPaddrHigh(physaddr));
1400 
1401 				bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1402 				sgl->word2 = cpu_to_le32(sgl->word2);
1403 				sgl->sge_len = cpu_to_le32(dma_len);
1404 
1405 				dma_offset += dma_len;
1406 				data_sg = sg_next(data_sg);
1407 
1408 				sgl++;
1409 
1410 				lsp_just_set = false;
1411 			} else {
1412 				sgl->word2 = cpu_to_le32(sgl->word2);
1413 
1414 				sgl->sge_len = cpu_to_le32(
1415 						     phba->cfg_sg_dma_buf_size);
1416 
1417 				sgl = (struct sli4_sge *)sgl_xtra->dma_sgl;
1418 				i = i - 1;
1419 
1420 				lsp_just_set = true;
1421 			}
1422 
1423 			j++;
1424 		}
1425 
1426 		/* PBDE support for first data SGE only */
1427 		if (nseg == 1 && phba->cfg_enable_pbde) {
1428 			/* Words 13-15 */
1429 			bde = (struct ulp_bde64 *)
1430 				&wqe->words[13];
1431 			bde->addrLow = first_data_sgl->addr_lo;
1432 			bde->addrHigh = first_data_sgl->addr_hi;
1433 			bde->tus.f.bdeSize =
1434 				le32_to_cpu(first_data_sgl->sge_len);
1435 			bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1436 			bde->tus.w = cpu_to_le32(bde->tus.w);
1437 
1438 			/* Word 11 - set PBDE bit */
1439 			bf_set(wqe_pbde, &wqe->generic.wqe_com, 1);
1440 		} else {
1441 			memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1442 			/* Word 11 - PBDE bit disabled by default template */
1443 		}
1444 
1445 	} else {
1446 		lpfc_ncmd->seg_cnt = 0;
1447 
1448 		/* For this clause to be valid, the payload_length
1449 		 * and sg_cnt must zero.
1450 		 */
1451 		if (nCmd->payload_length != 0) {
1452 			lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
1453 					"6063 NVME DMA Prep Err: sg_cnt %d "
1454 					"payload_length x%x\n",
1455 					nCmd->sg_cnt, nCmd->payload_length);
1456 			return 1;
1457 		}
1458 	}
1459 	return 0;
1460 }
1461 
1462 /**
1463  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1464  * @pnvme_lport: Pointer to the driver's local port data
1465  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1466  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1467  * @pnvme_fcreq: IO request from nvme fc to driver.
1468  *
1469  * Driver registers this routine as it io request handler.  This
1470  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1471  * data structure to the rport indicated in @lpfc_nvme_rport.
1472  *
1473  * Return value :
1474  *   0 - Success
1475  *   TODO: What are the failure codes.
1476  **/
1477 static int
1478 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1479 			struct nvme_fc_remote_port *pnvme_rport,
1480 			void *hw_queue_handle,
1481 			struct nvmefc_fcp_req *pnvme_fcreq)
1482 {
1483 	int ret = 0;
1484 	int expedite = 0;
1485 	int idx, cpu;
1486 	struct lpfc_nvme_lport *lport;
1487 	struct lpfc_fc4_ctrl_stat *cstat;
1488 	struct lpfc_vport *vport;
1489 	struct lpfc_hba *phba;
1490 	struct lpfc_nodelist *ndlp;
1491 	struct lpfc_io_buf *lpfc_ncmd;
1492 	struct lpfc_nvme_rport *rport;
1493 	struct lpfc_nvme_qhandle *lpfc_queue_info;
1494 	struct lpfc_nvme_fcpreq_priv *freqpriv;
1495 	struct nvme_common_command *sqe;
1496 	uint64_t start = 0;
1497 
1498 	/* Validate pointers. LLDD fault handling with transport does
1499 	 * have timing races.
1500 	 */
1501 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1502 	if (unlikely(!lport)) {
1503 		ret = -EINVAL;
1504 		goto out_fail;
1505 	}
1506 
1507 	vport = lport->vport;
1508 
1509 	if (unlikely(!hw_queue_handle)) {
1510 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1511 				 "6117 Fail IO, NULL hw_queue_handle\n");
1512 		atomic_inc(&lport->xmt_fcp_err);
1513 		ret = -EBUSY;
1514 		goto out_fail;
1515 	}
1516 
1517 	phba = vport->phba;
1518 
1519 	if (unlikely(vport->load_flag & FC_UNLOADING)) {
1520 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1521 				 "6124 Fail IO, Driver unload\n");
1522 		atomic_inc(&lport->xmt_fcp_err);
1523 		ret = -ENODEV;
1524 		goto out_fail;
1525 	}
1526 
1527 	freqpriv = pnvme_fcreq->private;
1528 	if (unlikely(!freqpriv)) {
1529 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1530 				 "6158 Fail IO, NULL request data\n");
1531 		atomic_inc(&lport->xmt_fcp_err);
1532 		ret = -EINVAL;
1533 		goto out_fail;
1534 	}
1535 
1536 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1537 	if (phba->ktime_on)
1538 		start = ktime_get_ns();
1539 #endif
1540 	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1541 	lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1542 
1543 	/*
1544 	 * Catch race where our node has transitioned, but the
1545 	 * transport is still transitioning.
1546 	 */
1547 	ndlp = rport->ndlp;
1548 	if (!ndlp) {
1549 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1550 				 "6053 Busy IO, ndlp not ready: rport x%px "
1551 				  "ndlp x%px, DID x%06x\n",
1552 				 rport, ndlp, pnvme_rport->port_id);
1553 		atomic_inc(&lport->xmt_fcp_err);
1554 		ret = -EBUSY;
1555 		goto out_fail;
1556 	}
1557 
1558 	/* The remote node has to be a mapped target or it's an error. */
1559 	if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1560 	    (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1561 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1562 				 "6036 Fail IO, DID x%06x not ready for "
1563 				 "IO. State x%x, Type x%x Flg x%x\n",
1564 				 pnvme_rport->port_id,
1565 				 ndlp->nlp_state, ndlp->nlp_type,
1566 				 ndlp->fc4_xpt_flags);
1567 		atomic_inc(&lport->xmt_fcp_bad_ndlp);
1568 		ret = -EBUSY;
1569 		goto out_fail;
1570 
1571 	}
1572 
1573 	/* Currently only NVME Keep alive commands should be expedited
1574 	 * if the driver runs out of a resource. These should only be
1575 	 * issued on the admin queue, qidx 0
1576 	 */
1577 	if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1578 		sqe = &((struct nvme_fc_cmd_iu *)
1579 			pnvme_fcreq->cmdaddr)->sqe.common;
1580 		if (sqe->opcode == nvme_admin_keep_alive)
1581 			expedite = 1;
1582 	}
1583 
1584 	/* Check if IO qualifies for CMF */
1585 	if (phba->cmf_active_mode != LPFC_CFG_OFF &&
1586 	    pnvme_fcreq->io_dir == NVMEFC_FCP_READ &&
1587 	    pnvme_fcreq->payload_length) {
1588 		ret = lpfc_update_cmf_cmd(phba, pnvme_fcreq->payload_length);
1589 		if (ret) {
1590 			ret = -EBUSY;
1591 			goto out_fail;
1592 		}
1593 		/* Get start time for IO latency */
1594 		start = ktime_get_ns();
1595 	}
1596 
1597 	/* The node is shared with FCP IO, make sure the IO pending count does
1598 	 * not exceed the programmed depth.
1599 	 */
1600 	if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1601 		if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1602 		    !expedite) {
1603 			lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1604 					 "6174 Fail IO, ndlp qdepth exceeded: "
1605 					 "idx %d DID %x pend %d qdepth %d\n",
1606 					 lpfc_queue_info->index, ndlp->nlp_DID,
1607 					 atomic_read(&ndlp->cmd_pending),
1608 					 ndlp->cmd_qdepth);
1609 			atomic_inc(&lport->xmt_fcp_qdepth);
1610 			ret = -EBUSY;
1611 			goto out_fail1;
1612 		}
1613 	}
1614 
1615 	/* Lookup Hardware Queue index based on fcp_io_sched module parameter */
1616 	if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) {
1617 		idx = lpfc_queue_info->index;
1618 	} else {
1619 		cpu = raw_smp_processor_id();
1620 		idx = phba->sli4_hba.cpu_map[cpu].hdwq;
1621 	}
1622 
1623 	lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite);
1624 	if (lpfc_ncmd == NULL) {
1625 		atomic_inc(&lport->xmt_fcp_noxri);
1626 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1627 				 "6065 Fail IO, driver buffer pool is empty: "
1628 				 "idx %d DID %x\n",
1629 				 lpfc_queue_info->index, ndlp->nlp_DID);
1630 		ret = -EBUSY;
1631 		goto out_fail1;
1632 	}
1633 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1634 	if (start) {
1635 		lpfc_ncmd->ts_cmd_start = start;
1636 		lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1637 	} else {
1638 		lpfc_ncmd->ts_cmd_start = 0;
1639 	}
1640 #endif
1641 	lpfc_ncmd->rx_cmd_start = start;
1642 
1643 	/*
1644 	 * Store the data needed by the driver to issue, abort, and complete
1645 	 * an IO.
1646 	 * Do not let the IO hang out forever.  There is no midlayer issuing
1647 	 * an abort so inform the FW of the maximum IO pending time.
1648 	 */
1649 	freqpriv->nvme_buf = lpfc_ncmd;
1650 	lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1651 	lpfc_ncmd->ndlp = ndlp;
1652 	lpfc_ncmd->qidx = lpfc_queue_info->qidx;
1653 
1654 	/*
1655 	 * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1656 	 * This identfier was create in our hardware queue create callback
1657 	 * routine. The driver now is dependent on the IO queue steering from
1658 	 * the transport.  We are trusting the upper NVME layers know which
1659 	 * index to use and that they have affinitized a CPU to this hardware
1660 	 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1661 	 */
1662 	lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1663 	cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat;
1664 
1665 	lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1666 	ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1667 	if (ret) {
1668 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1669 				 "6175 Fail IO, Prep DMA: "
1670 				 "idx %d DID %x\n",
1671 				 lpfc_queue_info->index, ndlp->nlp_DID);
1672 		atomic_inc(&lport->xmt_fcp_err);
1673 		ret = -ENOMEM;
1674 		goto out_free_nvme_buf;
1675 	}
1676 
1677 	lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1678 			 lpfc_ncmd->cur_iocbq.sli4_xritag,
1679 			 lpfc_queue_info->index, ndlp->nlp_DID);
1680 
1681 	ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq);
1682 	if (ret) {
1683 		atomic_inc(&lport->xmt_fcp_wqerr);
1684 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1685 				 "6113 Fail IO, Could not issue WQE err %x "
1686 				 "sid: x%x did: x%x oxid: x%x\n",
1687 				 ret, vport->fc_myDID, ndlp->nlp_DID,
1688 				 lpfc_ncmd->cur_iocbq.sli4_xritag);
1689 		goto out_free_nvme_buf;
1690 	}
1691 
1692 	if (phba->cfg_xri_rebalancing)
1693 		lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no);
1694 
1695 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1696 	if (lpfc_ncmd->ts_cmd_start)
1697 		lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1698 
1699 	if (phba->hdwqstat_on & LPFC_CHECK_NVME_IO) {
1700 		cpu = raw_smp_processor_id();
1701 		this_cpu_inc(phba->sli4_hba.c_stat->xmt_io);
1702 		lpfc_ncmd->cpu = cpu;
1703 		if (idx != cpu)
1704 			lpfc_printf_vlog(vport,
1705 					 KERN_INFO, LOG_NVME_IOERR,
1706 					"6702 CPU Check cmd: "
1707 					"cpu %d wq %d\n",
1708 					lpfc_ncmd->cpu,
1709 					lpfc_queue_info->index);
1710 	}
1711 #endif
1712 	return 0;
1713 
1714  out_free_nvme_buf:
1715 	if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1716 		if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1717 			cstat->output_requests--;
1718 		else
1719 			cstat->input_requests--;
1720 	} else
1721 		cstat->control_requests--;
1722 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
1723  out_fail1:
1724 	lpfc_update_cmf_cmpl(phba, LPFC_CGN_NOT_SENT,
1725 			     pnvme_fcreq->payload_length, NULL);
1726  out_fail:
1727 	return ret;
1728 }
1729 
1730 /**
1731  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1732  * @phba: Pointer to HBA context object
1733  * @cmdiocb: Pointer to command iocb object.
1734  * @abts_cmpl: Pointer to wcqe complete object.
1735  *
1736  * This is the callback function for any NVME FCP IO that was aborted.
1737  *
1738  * Return value:
1739  *   None
1740  **/
1741 void
1742 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1743 			   struct lpfc_wcqe_complete *abts_cmpl)
1744 {
1745 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1746 			"6145 ABORT_XRI_CN completing on rpi x%x "
1747 			"original iotag x%x, abort cmd iotag x%x "
1748 			"req_tag x%x, status x%x, hwstatus x%x\n",
1749 			bf_get(wqe_ctxt_tag, &cmdiocb->wqe.generic.wqe_com),
1750 			get_job_abtsiotag(phba, cmdiocb), cmdiocb->iotag,
1751 			bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1752 			bf_get(lpfc_wcqe_c_status, abts_cmpl),
1753 			bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1754 	lpfc_sli_release_iocbq(phba, cmdiocb);
1755 }
1756 
1757 /**
1758  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1759  * @pnvme_lport: Pointer to the driver's local port data
1760  * @pnvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1761  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1762  * @pnvme_fcreq: IO request from nvme fc to driver.
1763  *
1764  * Driver registers this routine as its nvme request io abort handler.  This
1765  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1766  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1767  * is executed asynchronously - one the target is validated as "MAPPED" and
1768  * ready for IO, the driver issues the abort request and returns.
1769  *
1770  * Return value:
1771  *   None
1772  **/
1773 static void
1774 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1775 		    struct nvme_fc_remote_port *pnvme_rport,
1776 		    void *hw_queue_handle,
1777 		    struct nvmefc_fcp_req *pnvme_fcreq)
1778 {
1779 	struct lpfc_nvme_lport *lport;
1780 	struct lpfc_vport *vport;
1781 	struct lpfc_hba *phba;
1782 	struct lpfc_io_buf *lpfc_nbuf;
1783 	struct lpfc_iocbq *nvmereq_wqe;
1784 	struct lpfc_nvme_fcpreq_priv *freqpriv;
1785 	unsigned long flags;
1786 	int ret_val;
1787 
1788 	/* Validate pointers. LLDD fault handling with transport does
1789 	 * have timing races.
1790 	 */
1791 	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1792 	if (unlikely(!lport))
1793 		return;
1794 
1795 	vport = lport->vport;
1796 
1797 	if (unlikely(!hw_queue_handle)) {
1798 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1799 				 "6129 Fail Abort, HW Queue Handle NULL.\n");
1800 		return;
1801 	}
1802 
1803 	phba = vport->phba;
1804 	freqpriv = pnvme_fcreq->private;
1805 
1806 	if (unlikely(!freqpriv))
1807 		return;
1808 	if (vport->load_flag & FC_UNLOADING)
1809 		return;
1810 
1811 	/* Announce entry to new IO submit field. */
1812 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1813 			 "6002 Abort Request to rport DID x%06x "
1814 			 "for nvme_fc_req x%px\n",
1815 			 pnvme_rport->port_id,
1816 			 pnvme_fcreq);
1817 
1818 	/* If the hba is getting reset, this flag is set.  It is
1819 	 * cleared when the reset is complete and rings reestablished.
1820 	 */
1821 	spin_lock_irqsave(&phba->hbalock, flags);
1822 	/* driver queued commands are in process of being flushed */
1823 	if (phba->hba_flag & HBA_IOQ_FLUSH) {
1824 		spin_unlock_irqrestore(&phba->hbalock, flags);
1825 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1826 				 "6139 Driver in reset cleanup - flushing "
1827 				 "NVME Req now.  hba_flag x%x\n",
1828 				 phba->hba_flag);
1829 		return;
1830 	}
1831 
1832 	lpfc_nbuf = freqpriv->nvme_buf;
1833 	if (!lpfc_nbuf) {
1834 		spin_unlock_irqrestore(&phba->hbalock, flags);
1835 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1836 				 "6140 NVME IO req has no matching lpfc nvme "
1837 				 "io buffer.  Skipping abort req.\n");
1838 		return;
1839 	} else if (!lpfc_nbuf->nvmeCmd) {
1840 		spin_unlock_irqrestore(&phba->hbalock, flags);
1841 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1842 				 "6141 lpfc NVME IO req has no nvme_fcreq "
1843 				 "io buffer.  Skipping abort req.\n");
1844 		return;
1845 	}
1846 	nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1847 
1848 	/* Guard against IO completion being called at same time */
1849 	spin_lock(&lpfc_nbuf->buf_lock);
1850 
1851 	/*
1852 	 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1853 	 * state must match the nvme_fcreq passed by the nvme
1854 	 * transport.  If they don't match, it is likely the driver
1855 	 * has already completed the NVME IO and the nvme transport
1856 	 * has not seen it yet.
1857 	 */
1858 	if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1859 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1860 				 "6143 NVME req mismatch: "
1861 				 "lpfc_nbuf x%px nvmeCmd x%px, "
1862 				 "pnvme_fcreq x%px.  Skipping Abort xri x%x\n",
1863 				 lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1864 				 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1865 		goto out_unlock;
1866 	}
1867 
1868 	/* Don't abort IOs no longer on the pending queue. */
1869 	if (!(nvmereq_wqe->cmd_flag & LPFC_IO_ON_TXCMPLQ)) {
1870 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1871 				 "6142 NVME IO req x%px not queued - skipping "
1872 				 "abort req xri x%x\n",
1873 				 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1874 		goto out_unlock;
1875 	}
1876 
1877 	atomic_inc(&lport->xmt_fcp_abort);
1878 	lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1879 			 nvmereq_wqe->sli4_xritag,
1880 			 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1881 
1882 	/* Outstanding abort is in progress */
1883 	if (nvmereq_wqe->cmd_flag & LPFC_DRIVER_ABORTED) {
1884 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1885 				 "6144 Outstanding NVME I/O Abort Request "
1886 				 "still pending on nvme_fcreq x%px, "
1887 				 "lpfc_ncmd x%px xri x%x\n",
1888 				 pnvme_fcreq, lpfc_nbuf,
1889 				 nvmereq_wqe->sli4_xritag);
1890 		goto out_unlock;
1891 	}
1892 
1893 	ret_val = lpfc_sli4_issue_abort_iotag(phba, nvmereq_wqe,
1894 					      lpfc_nvme_abort_fcreq_cmpl);
1895 
1896 	spin_unlock(&lpfc_nbuf->buf_lock);
1897 	spin_unlock_irqrestore(&phba->hbalock, flags);
1898 
1899 	/* Make sure HBA is alive */
1900 	lpfc_issue_hb_tmo(phba);
1901 
1902 	if (ret_val != WQE_SUCCESS) {
1903 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
1904 				 "6137 Failed abts issue_wqe with status x%x "
1905 				 "for nvme_fcreq x%px.\n",
1906 				 ret_val, pnvme_fcreq);
1907 		return;
1908 	}
1909 
1910 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1911 			 "6138 Transport Abort NVME Request Issued for "
1912 			 "ox_id x%x\n",
1913 			 nvmereq_wqe->sli4_xritag);
1914 	return;
1915 
1916 out_unlock:
1917 	spin_unlock(&lpfc_nbuf->buf_lock);
1918 	spin_unlock_irqrestore(&phba->hbalock, flags);
1919 	return;
1920 }
1921 
1922 /* Declare and initialization an instance of the FC NVME template. */
1923 static struct nvme_fc_port_template lpfc_nvme_template = {
1924 	/* initiator-based functions */
1925 	.localport_delete  = lpfc_nvme_localport_delete,
1926 	.remoteport_delete = lpfc_nvme_remoteport_delete,
1927 	.create_queue = lpfc_nvme_create_queue,
1928 	.delete_queue = lpfc_nvme_delete_queue,
1929 	.ls_req       = lpfc_nvme_ls_req,
1930 	.fcp_io       = lpfc_nvme_fcp_io_submit,
1931 	.ls_abort     = lpfc_nvme_ls_abort,
1932 	.fcp_abort    = lpfc_nvme_fcp_abort,
1933 	.xmt_ls_rsp   = lpfc_nvme_xmt_ls_rsp,
1934 
1935 	.max_hw_queues = 1,
1936 	.max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1937 	.max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1938 	.dma_boundary = 0xFFFFFFFF,
1939 
1940 	/* Sizes of additional private data for data structures.
1941 	 * No use for the last two sizes at this time.
1942 	 */
1943 	.local_priv_sz = sizeof(struct lpfc_nvme_lport),
1944 	.remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1945 	.lsrqst_priv_sz = 0,
1946 	.fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1947 };
1948 
1949 /*
1950  * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA
1951  *
1952  * This routine removes a nvme buffer from head of @hdwq io_buf_list
1953  * and returns to caller.
1954  *
1955  * Return codes:
1956  *   NULL - Error
1957  *   Pointer to lpfc_nvme_buf - Success
1958  **/
1959 static struct lpfc_io_buf *
1960 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
1961 		  int idx, int expedite)
1962 {
1963 	struct lpfc_io_buf *lpfc_ncmd;
1964 	struct lpfc_sli4_hdw_queue *qp;
1965 	struct sli4_sge *sgl;
1966 	struct lpfc_iocbq *pwqeq;
1967 	union lpfc_wqe128 *wqe;
1968 
1969 	lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite);
1970 
1971 	if (lpfc_ncmd) {
1972 		pwqeq = &(lpfc_ncmd->cur_iocbq);
1973 		wqe = &pwqeq->wqe;
1974 
1975 		/* Setup key fields in buffer that may have been changed
1976 		 * if other protocols used this buffer.
1977 		 */
1978 		pwqeq->cmd_flag = LPFC_IO_NVME;
1979 		pwqeq->cmd_cmpl = lpfc_nvme_io_cmd_cmpl;
1980 		lpfc_ncmd->start_time = jiffies;
1981 		lpfc_ncmd->flags = 0;
1982 
1983 		/* Rsp SGE will be filled in when we rcv an IO
1984 		 * from the NVME Layer to be sent.
1985 		 * The cmd is going to be embedded so we need a SKIP SGE.
1986 		 */
1987 		sgl = lpfc_ncmd->dma_sgl;
1988 		bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
1989 		bf_set(lpfc_sli4_sge_last, sgl, 0);
1990 		sgl->word2 = cpu_to_le32(sgl->word2);
1991 		/* Fill in word 3 / sgl_len during cmd submission */
1992 
1993 		/* Initialize 64 bytes only */
1994 		memset(wqe, 0, sizeof(union lpfc_wqe));
1995 
1996 		if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1997 			atomic_inc(&ndlp->cmd_pending);
1998 			lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH;
1999 		}
2000 
2001 	} else {
2002 		qp = &phba->sli4_hba.hdwq[idx];
2003 		qp->empty_io_bufs++;
2004 	}
2005 
2006 	return  lpfc_ncmd;
2007 }
2008 
2009 /**
2010  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2011  * @phba: The Hba for which this call is being executed.
2012  * @lpfc_ncmd: The nvme buffer which is being released.
2013  *
2014  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2015  * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2016  * and cannot be reused for at least RA_TOV amount of time if it was
2017  * aborted.
2018  **/
2019 static void
2020 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd)
2021 {
2022 	struct lpfc_sli4_hdw_queue *qp;
2023 	unsigned long iflag = 0;
2024 
2025 	if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
2026 		atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
2027 
2028 	lpfc_ncmd->ndlp = NULL;
2029 	lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH;
2030 
2031 	qp = lpfc_ncmd->hdwq;
2032 	if (unlikely(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
2033 		lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2034 				"6310 XB release deferred for "
2035 				"ox_id x%x on reqtag x%x\n",
2036 				lpfc_ncmd->cur_iocbq.sli4_xritag,
2037 				lpfc_ncmd->cur_iocbq.iotag);
2038 
2039 		spin_lock_irqsave(&qp->abts_io_buf_list_lock, iflag);
2040 		list_add_tail(&lpfc_ncmd->list,
2041 			&qp->lpfc_abts_io_buf_list);
2042 		qp->abts_nvme_io_bufs++;
2043 		spin_unlock_irqrestore(&qp->abts_io_buf_list_lock, iflag);
2044 	} else
2045 		lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp);
2046 }
2047 
2048 /**
2049  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2050  * @vport: the lpfc_vport instance requesting a localport.
2051  *
2052  * This routine is invoked to create an nvme localport instance to bind
2053  * to the nvme_fc_transport.  It is called once during driver load
2054  * like lpfc_create_shost after all other services are initialized.
2055  * It requires a vport, vpi, and wwns at call time.  Other localport
2056  * parameters are modified as the driver's FCID and the Fabric WWN
2057  * are established.
2058  *
2059  * Return codes
2060  *      0 - successful
2061  *      -ENOMEM - no heap memory available
2062  *      other values - from nvme registration upcall
2063  **/
2064 int
2065 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2066 {
2067 	int ret = 0;
2068 	struct lpfc_hba  *phba = vport->phba;
2069 	struct nvme_fc_port_info nfcp_info;
2070 	struct nvme_fc_local_port *localport;
2071 	struct lpfc_nvme_lport *lport;
2072 
2073 	/* Initialize this localport instance.  The vport wwn usage ensures
2074 	 * that NPIV is accounted for.
2075 	 */
2076 	memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2077 	nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2078 	nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2079 	nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2080 
2081 	/* We need to tell the transport layer + 1 because it takes page
2082 	 * alignment into account. When space for the SGL is allocated we
2083 	 * allocate + 3, one for cmd, one for rsp and one for this alignment
2084 	 */
2085 	lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2086 
2087 	/* Advertise how many hw queues we support based on cfg_hdw_queue,
2088 	 * which will not exceed cpu count.
2089 	 */
2090 	lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue;
2091 
2092 	if (!IS_ENABLED(CONFIG_NVME_FC))
2093 		return ret;
2094 
2095 	/* localport is allocated from the stack, but the registration
2096 	 * call allocates heap memory as well as the private area.
2097 	 */
2098 
2099 	ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2100 					 &vport->phba->pcidev->dev, &localport);
2101 	if (!ret) {
2102 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2103 				 "6005 Successfully registered local "
2104 				 "NVME port num %d, localP x%px, private "
2105 				 "x%px, sg_seg %d\n",
2106 				 localport->port_num, localport,
2107 				 localport->private,
2108 				 lpfc_nvme_template.max_sgl_segments);
2109 
2110 		/* Private is our lport size declared in the template. */
2111 		lport = (struct lpfc_nvme_lport *)localport->private;
2112 		vport->localport = localport;
2113 		lport->vport = vport;
2114 		vport->nvmei_support = 1;
2115 
2116 		atomic_set(&lport->xmt_fcp_noxri, 0);
2117 		atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2118 		atomic_set(&lport->xmt_fcp_qdepth, 0);
2119 		atomic_set(&lport->xmt_fcp_err, 0);
2120 		atomic_set(&lport->xmt_fcp_wqerr, 0);
2121 		atomic_set(&lport->xmt_fcp_abort, 0);
2122 		atomic_set(&lport->xmt_ls_abort, 0);
2123 		atomic_set(&lport->xmt_ls_err, 0);
2124 		atomic_set(&lport->cmpl_fcp_xb, 0);
2125 		atomic_set(&lport->cmpl_fcp_err, 0);
2126 		atomic_set(&lport->cmpl_ls_xb, 0);
2127 		atomic_set(&lport->cmpl_ls_err, 0);
2128 
2129 		atomic_set(&lport->fc4NvmeLsRequests, 0);
2130 		atomic_set(&lport->fc4NvmeLsCmpls, 0);
2131 	}
2132 
2133 	return ret;
2134 }
2135 
2136 #if (IS_ENABLED(CONFIG_NVME_FC))
2137 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2138  *
2139  * The driver has to wait for the host nvme transport to callback
2140  * indicating the localport has successfully unregistered all
2141  * resources.  Since this is an uninterruptible wait, loop every ten
2142  * seconds and print a message indicating no progress.
2143  *
2144  * An uninterruptible wait is used because of the risk of transport-to-
2145  * driver state mismatch.
2146  */
2147 static void
2148 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2149 			   struct lpfc_nvme_lport *lport,
2150 			   struct completion *lport_unreg_cmp)
2151 {
2152 	u32 wait_tmo;
2153 	int ret, i, pending = 0;
2154 	struct lpfc_sli_ring  *pring;
2155 	struct lpfc_hba  *phba = vport->phba;
2156 	struct lpfc_sli4_hdw_queue *qp;
2157 	int abts_scsi, abts_nvme;
2158 
2159 	/* Host transport has to clean up and confirm requiring an indefinite
2160 	 * wait. Print a message if a 10 second wait expires and renew the
2161 	 * wait. This is unexpected.
2162 	 */
2163 	wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2164 	while (true) {
2165 		ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo);
2166 		if (unlikely(!ret)) {
2167 			pending = 0;
2168 			abts_scsi = 0;
2169 			abts_nvme = 0;
2170 			for (i = 0; i < phba->cfg_hdw_queue; i++) {
2171 				qp = &phba->sli4_hba.hdwq[i];
2172 				if (!vport || !vport->localport ||
2173 				    !qp || !qp->io_wq)
2174 					return;
2175 
2176 				pring = qp->io_wq->pring;
2177 				if (!pring)
2178 					continue;
2179 				pending += pring->txcmplq_cnt;
2180 				abts_scsi += qp->abts_scsi_io_bufs;
2181 				abts_nvme += qp->abts_nvme_io_bufs;
2182 			}
2183 			if (!vport || !vport->localport ||
2184 			    vport->phba->hba_flag & HBA_PCI_ERR)
2185 				return;
2186 
2187 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2188 					 "6176 Lport x%px Localport x%px wait "
2189 					 "timed out. Pending %d [%d:%d]. "
2190 					 "Renewing.\n",
2191 					 lport, vport->localport, pending,
2192 					 abts_scsi, abts_nvme);
2193 			continue;
2194 		}
2195 		break;
2196 	}
2197 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2198 			 "6177 Lport x%px Localport x%px Complete Success\n",
2199 			 lport, vport->localport);
2200 }
2201 #endif
2202 
2203 /**
2204  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2205  * @vport: pointer to a host virtual N_Port data structure
2206  *
2207  * This routine is invoked to destroy all lports bound to the phba.
2208  * The lport memory was allocated by the nvme fc transport and is
2209  * released there.  This routine ensures all rports bound to the
2210  * lport have been disconnected.
2211  *
2212  **/
2213 void
2214 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2215 {
2216 #if (IS_ENABLED(CONFIG_NVME_FC))
2217 	struct nvme_fc_local_port *localport;
2218 	struct lpfc_nvme_lport *lport;
2219 	int ret;
2220 	DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp);
2221 
2222 	if (vport->nvmei_support == 0)
2223 		return;
2224 
2225 	localport = vport->localport;
2226 	if (!localport)
2227 		return;
2228 	lport = (struct lpfc_nvme_lport *)localport->private;
2229 
2230 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2231 			 "6011 Destroying NVME localport x%px\n",
2232 			 localport);
2233 
2234 	/* lport's rport list is clear.  Unregister
2235 	 * lport and release resources.
2236 	 */
2237 	lport->lport_unreg_cmp = &lport_unreg_cmp;
2238 	ret = nvme_fc_unregister_localport(localport);
2239 
2240 	/* Wait for completion.  This either blocks
2241 	 * indefinitely or succeeds
2242 	 */
2243 	lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp);
2244 	vport->localport = NULL;
2245 
2246 	/* Regardless of the unregister upcall response, clear
2247 	 * nvmei_support.  All rports are unregistered and the
2248 	 * driver will clean up.
2249 	 */
2250 	vport->nvmei_support = 0;
2251 	if (ret == 0) {
2252 		lpfc_printf_vlog(vport,
2253 				 KERN_INFO, LOG_NVME_DISC,
2254 				 "6009 Unregistered lport Success\n");
2255 	} else {
2256 		lpfc_printf_vlog(vport,
2257 				 KERN_INFO, LOG_NVME_DISC,
2258 				 "6010 Unregistered lport "
2259 				 "Failed, status x%x\n",
2260 				 ret);
2261 	}
2262 #endif
2263 }
2264 
2265 void
2266 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2267 {
2268 #if (IS_ENABLED(CONFIG_NVME_FC))
2269 	struct nvme_fc_local_port *localport;
2270 	struct lpfc_nvme_lport *lport;
2271 
2272 	localport = vport->localport;
2273 	if (!localport) {
2274 		lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2275 				 "6710 Update NVME fail. No localport\n");
2276 		return;
2277 	}
2278 	lport = (struct lpfc_nvme_lport *)localport->private;
2279 	if (!lport) {
2280 		lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2281 				 "6171 Update NVME fail. localP x%px, No lport\n",
2282 				 localport);
2283 		return;
2284 	}
2285 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2286 			 "6012 Update NVME lport x%px did x%x\n",
2287 			 localport, vport->fc_myDID);
2288 
2289 	localport->port_id = vport->fc_myDID;
2290 	if (localport->port_id == 0)
2291 		localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2292 	else
2293 		localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2294 
2295 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2296 			 "6030 bound lport x%px to DID x%06x\n",
2297 			 lport, localport->port_id);
2298 #endif
2299 }
2300 
2301 int
2302 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2303 {
2304 #if (IS_ENABLED(CONFIG_NVME_FC))
2305 	int ret = 0;
2306 	struct nvme_fc_local_port *localport;
2307 	struct lpfc_nvme_lport *lport;
2308 	struct lpfc_nvme_rport *rport;
2309 	struct lpfc_nvme_rport *oldrport;
2310 	struct nvme_fc_remote_port *remote_port;
2311 	struct nvme_fc_port_info rpinfo;
2312 	struct lpfc_nodelist *prev_ndlp = NULL;
2313 	struct fc_rport *srport = ndlp->rport;
2314 
2315 	lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2316 			 "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2317 			 ndlp->nlp_DID, ndlp->nlp_type);
2318 
2319 	localport = vport->localport;
2320 	if (!localport)
2321 		return 0;
2322 
2323 	lport = (struct lpfc_nvme_lport *)localport->private;
2324 
2325 	/* NVME rports are not preserved across devloss.
2326 	 * Just register this instance.  Note, rpinfo->dev_loss_tmo
2327 	 * is left 0 to indicate accept transport defaults.  The
2328 	 * driver communicates port role capabilities consistent
2329 	 * with the PRLI response data.
2330 	 */
2331 	memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2332 	rpinfo.port_id = ndlp->nlp_DID;
2333 	if (ndlp->nlp_type & NLP_NVME_TARGET)
2334 		rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2335 	if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2336 		rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2337 
2338 	if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2339 		rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2340 
2341 	rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2342 	rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2343 	if (srport)
2344 		rpinfo.dev_loss_tmo = srport->dev_loss_tmo;
2345 	else
2346 		rpinfo.dev_loss_tmo = vport->cfg_devloss_tmo;
2347 
2348 	spin_lock_irq(&ndlp->lock);
2349 	oldrport = lpfc_ndlp_get_nrport(ndlp);
2350 	if (oldrport) {
2351 		prev_ndlp = oldrport->ndlp;
2352 		spin_unlock_irq(&ndlp->lock);
2353 	} else {
2354 		spin_unlock_irq(&ndlp->lock);
2355 		if (!lpfc_nlp_get(ndlp)) {
2356 			dev_warn(&vport->phba->pcidev->dev,
2357 				 "Warning - No node ref - exit register\n");
2358 			return 0;
2359 		}
2360 	}
2361 
2362 	ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2363 	if (!ret) {
2364 		/* If the ndlp already has an nrport, this is just
2365 		 * a resume of the existing rport.  Else this is a
2366 		 * new rport.
2367 		 */
2368 		/* Guard against an unregister/reregister
2369 		 * race that leaves the WAIT flag set.
2370 		 */
2371 		spin_lock_irq(&ndlp->lock);
2372 		ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT;
2373 		ndlp->fc4_xpt_flags |= NVME_XPT_REGD;
2374 		spin_unlock_irq(&ndlp->lock);
2375 		rport = remote_port->private;
2376 		if (oldrport) {
2377 
2378 			/* Sever the ndlp<->rport association
2379 			 * before dropping the ndlp ref from
2380 			 * register.
2381 			 */
2382 			spin_lock_irq(&ndlp->lock);
2383 			ndlp->nrport = NULL;
2384 			ndlp->fc4_xpt_flags &= ~NVME_XPT_UNREG_WAIT;
2385 			spin_unlock_irq(&ndlp->lock);
2386 			rport->ndlp = NULL;
2387 			rport->remoteport = NULL;
2388 
2389 			/* Reference only removed if previous NDLP is no longer
2390 			 * active. It might be just a swap and removing the
2391 			 * reference would cause a premature cleanup.
2392 			 */
2393 			if (prev_ndlp && prev_ndlp != ndlp) {
2394 				if (!prev_ndlp->nrport)
2395 					lpfc_nlp_put(prev_ndlp);
2396 			}
2397 		}
2398 
2399 		/* Clean bind the rport to the ndlp. */
2400 		rport->remoteport = remote_port;
2401 		rport->lport = lport;
2402 		rport->ndlp = ndlp;
2403 		spin_lock_irq(&ndlp->lock);
2404 		ndlp->nrport = rport;
2405 		spin_unlock_irq(&ndlp->lock);
2406 		lpfc_printf_vlog(vport, KERN_INFO,
2407 				 LOG_NVME_DISC | LOG_NODE,
2408 				 "6022 Bind lport x%px to remoteport x%px "
2409 				 "rport x%px WWNN 0x%llx, "
2410 				 "Rport WWPN 0x%llx DID "
2411 				 "x%06x Role x%x, ndlp %p prev_ndlp x%px\n",
2412 				 lport, remote_port, rport,
2413 				 rpinfo.node_name, rpinfo.port_name,
2414 				 rpinfo.port_id, rpinfo.port_role,
2415 				 ndlp, prev_ndlp);
2416 	} else {
2417 		lpfc_printf_vlog(vport, KERN_ERR,
2418 				 LOG_TRACE_EVENT,
2419 				 "6031 RemotePort Registration failed "
2420 				 "err: %d, DID x%06x\n",
2421 				 ret, ndlp->nlp_DID);
2422 	}
2423 
2424 	return ret;
2425 #else
2426 	return 0;
2427 #endif
2428 }
2429 
2430 /*
2431  * lpfc_nvme_rescan_port - Check to see if we should rescan this remoteport
2432  *
2433  * If the ndlp represents an NVME Target, that we are logged into,
2434  * ping the NVME FC Transport layer to initiate a device rescan
2435  * on this remote NPort.
2436  */
2437 void
2438 lpfc_nvme_rescan_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2439 {
2440 #if (IS_ENABLED(CONFIG_NVME_FC))
2441 	struct lpfc_nvme_rport *nrport;
2442 	struct nvme_fc_remote_port *remoteport = NULL;
2443 
2444 	spin_lock_irq(&ndlp->lock);
2445 	nrport = lpfc_ndlp_get_nrport(ndlp);
2446 	if (nrport)
2447 		remoteport = nrport->remoteport;
2448 	spin_unlock_irq(&ndlp->lock);
2449 
2450 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2451 			 "6170 Rescan NPort DID x%06x type x%x "
2452 			 "state x%x nrport x%px remoteport x%px\n",
2453 			 ndlp->nlp_DID, ndlp->nlp_type, ndlp->nlp_state,
2454 			 nrport, remoteport);
2455 
2456 	if (!nrport || !remoteport)
2457 		goto rescan_exit;
2458 
2459 	/* Only rescan if we are an NVME target in the MAPPED state */
2460 	if (remoteport->port_role & FC_PORT_ROLE_NVME_DISCOVERY &&
2461 	    ndlp->nlp_state == NLP_STE_MAPPED_NODE) {
2462 		nvme_fc_rescan_remoteport(remoteport);
2463 
2464 		lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2465 				 "6172 NVME rescanned DID x%06x "
2466 				 "port_state x%x\n",
2467 				 ndlp->nlp_DID, remoteport->port_state);
2468 	}
2469 	return;
2470  rescan_exit:
2471 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2472 			 "6169 Skip NVME Rport Rescan, NVME remoteport "
2473 			 "unregistered\n");
2474 #endif
2475 }
2476 
2477 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2478  *
2479  * There is no notion of Devloss or rport recovery from the current
2480  * nvme_transport perspective.  Loss of an rport just means IO cannot
2481  * be sent and recovery is completely up to the initator.
2482  * For now, the driver just unbinds the DID and port_role so that
2483  * no further IO can be issued.  Changes are planned for later.
2484  *
2485  * Notes - the ndlp reference count is not decremented here since
2486  * since there is no nvme_transport api for devloss.  Node ref count
2487  * is only adjusted in driver unload.
2488  */
2489 void
2490 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2491 {
2492 #if (IS_ENABLED(CONFIG_NVME_FC))
2493 	int ret;
2494 	struct nvme_fc_local_port *localport;
2495 	struct lpfc_nvme_lport *lport;
2496 	struct lpfc_nvme_rport *rport;
2497 	struct nvme_fc_remote_port *remoteport = NULL;
2498 
2499 	localport = vport->localport;
2500 
2501 	/* This is fundamental error.  The localport is always
2502 	 * available until driver unload.  Just exit.
2503 	 */
2504 	if (!localport)
2505 		return;
2506 
2507 	lport = (struct lpfc_nvme_lport *)localport->private;
2508 	if (!lport)
2509 		goto input_err;
2510 
2511 	spin_lock_irq(&ndlp->lock);
2512 	rport = lpfc_ndlp_get_nrport(ndlp);
2513 	if (rport)
2514 		remoteport = rport->remoteport;
2515 	spin_unlock_irq(&ndlp->lock);
2516 	if (!remoteport)
2517 		goto input_err;
2518 
2519 	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2520 			 "6033 Unreg nvme remoteport x%px, portname x%llx, "
2521 			 "port_id x%06x, portstate x%x port type x%x "
2522 			 "refcnt %d\n",
2523 			 remoteport, remoteport->port_name,
2524 			 remoteport->port_id, remoteport->port_state,
2525 			 ndlp->nlp_type, kref_read(&ndlp->kref));
2526 
2527 	/* Sanity check ndlp type.  Only call for NVME ports. Don't
2528 	 * clear any rport state until the transport calls back.
2529 	 */
2530 
2531 	if (ndlp->nlp_type & NLP_NVME_TARGET) {
2532 		/* No concern about the role change on the nvme remoteport.
2533 		 * The transport will update it.
2534 		 */
2535 		spin_lock_irq(&vport->phba->hbalock);
2536 		ndlp->fc4_xpt_flags |= NVME_XPT_UNREG_WAIT;
2537 		spin_unlock_irq(&vport->phba->hbalock);
2538 
2539 		/* Don't let the host nvme transport keep sending keep-alives
2540 		 * on this remoteport. Vport is unloading, no recovery. The
2541 		 * return values is ignored.  The upcall is a courtesy to the
2542 		 * transport.
2543 		 */
2544 		if (vport->load_flag & FC_UNLOADING ||
2545 		    unlikely(vport->phba->hba_flag & HBA_PCI_ERR))
2546 			(void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2547 
2548 		ret = nvme_fc_unregister_remoteport(remoteport);
2549 
2550 		/* The driver no longer knows if the nrport memory is valid.
2551 		 * because the controller teardown process has begun and
2552 		 * is asynchronous.  Break the binding in the ndlp. Also
2553 		 * remove the register ndlp reference to setup node release.
2554 		 */
2555 		ndlp->nrport = NULL;
2556 		lpfc_nlp_put(ndlp);
2557 		if (ret != 0) {
2558 			lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2559 					 "6167 NVME unregister failed %d "
2560 					 "port_state x%x\n",
2561 					 ret, remoteport->port_state);
2562 		}
2563 	}
2564 	return;
2565 
2566  input_err:
2567 #endif
2568 	lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
2569 			 "6168 State error: lport x%px, rport x%px FCID x%06x\n",
2570 			 vport->localport, ndlp->rport, ndlp->nlp_DID);
2571 }
2572 
2573 /**
2574  * lpfc_sli4_nvme_pci_offline_aborted - Fast-path process of NVME xri abort
2575  * @phba: pointer to lpfc hba data structure.
2576  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2577  *
2578  * This routine is invoked by the worker thread to process a SLI4 fast-path
2579  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2580  * here.
2581  **/
2582 void
2583 lpfc_sli4_nvme_pci_offline_aborted(struct lpfc_hba *phba,
2584 				   struct lpfc_io_buf *lpfc_ncmd)
2585 {
2586 	struct nvmefc_fcp_req *nvme_cmd = NULL;
2587 
2588 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2589 			"6533 %s nvme_cmd %p tag x%x abort complete and "
2590 			"xri released\n", __func__,
2591 			lpfc_ncmd->nvmeCmd,
2592 			lpfc_ncmd->cur_iocbq.iotag);
2593 
2594 	/* Aborted NVME commands are required to not complete
2595 	 * before the abort exchange command fully completes.
2596 	 * Once completed, it is available via the put list.
2597 	 */
2598 	if (lpfc_ncmd->nvmeCmd) {
2599 		nvme_cmd = lpfc_ncmd->nvmeCmd;
2600 		nvme_cmd->transferred_length = 0;
2601 		nvme_cmd->rcv_rsplen = 0;
2602 		nvme_cmd->status = NVME_SC_INTERNAL;
2603 		nvme_cmd->done(nvme_cmd);
2604 		lpfc_ncmd->nvmeCmd = NULL;
2605 	}
2606 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
2607 }
2608 
2609 /**
2610  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2611  * @phba: pointer to lpfc hba data structure.
2612  * @axri: pointer to the fcp xri abort wcqe structure.
2613  * @lpfc_ncmd: The nvme job structure for the request being aborted.
2614  *
2615  * This routine is invoked by the worker thread to process a SLI4 fast-path
2616  * NVME aborted xri.  Aborted NVME IO commands are completed to the transport
2617  * here.
2618  **/
2619 void
2620 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2621 			   struct sli4_wcqe_xri_aborted *axri,
2622 			   struct lpfc_io_buf *lpfc_ncmd)
2623 {
2624 	uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2625 	struct nvmefc_fcp_req *nvme_cmd = NULL;
2626 	struct lpfc_nodelist *ndlp = lpfc_ncmd->ndlp;
2627 
2628 
2629 	if (ndlp)
2630 		lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2631 
2632 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2633 			"6311 nvme_cmd %p xri x%x tag x%x abort complete and "
2634 			"xri released\n",
2635 			lpfc_ncmd->nvmeCmd, xri,
2636 			lpfc_ncmd->cur_iocbq.iotag);
2637 
2638 	/* Aborted NVME commands are required to not complete
2639 	 * before the abort exchange command fully completes.
2640 	 * Once completed, it is available via the put list.
2641 	 */
2642 	if (lpfc_ncmd->nvmeCmd) {
2643 		nvme_cmd = lpfc_ncmd->nvmeCmd;
2644 		nvme_cmd->done(nvme_cmd);
2645 		lpfc_ncmd->nvmeCmd = NULL;
2646 	}
2647 	lpfc_release_nvme_buf(phba, lpfc_ncmd);
2648 }
2649 
2650 /**
2651  * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2652  * @phba: Pointer to HBA context object.
2653  *
2654  * This function flushes all wqes in the nvme rings and frees all resources
2655  * in the txcmplq. This function does not issue abort wqes for the IO
2656  * commands in txcmplq, they will just be returned with
2657  * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2658  * slot has been permanently disabled.
2659  **/
2660 void
2661 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2662 {
2663 	struct lpfc_sli_ring  *pring;
2664 	u32 i, wait_cnt = 0;
2665 
2666 	if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq)
2667 		return;
2668 
2669 	/* Cycle through all IO rings and make sure all outstanding
2670 	 * WQEs have been removed from the txcmplqs.
2671 	 */
2672 	for (i = 0; i < phba->cfg_hdw_queue; i++) {
2673 		if (!phba->sli4_hba.hdwq[i].io_wq)
2674 			continue;
2675 		pring = phba->sli4_hba.hdwq[i].io_wq->pring;
2676 
2677 		if (!pring)
2678 			continue;
2679 
2680 		/* Retrieve everything on the txcmplq */
2681 		while (!list_empty(&pring->txcmplq)) {
2682 			msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2683 			wait_cnt++;
2684 
2685 			/* The sleep is 10mS.  Every ten seconds,
2686 			 * dump a message.  Something is wrong.
2687 			 */
2688 			if ((wait_cnt % 1000) == 0) {
2689 				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
2690 						"6178 NVME IO not empty, "
2691 						"cnt %d\n", wait_cnt);
2692 			}
2693 		}
2694 	}
2695 
2696 	/* Make sure HBA is alive */
2697 	lpfc_issue_hb_tmo(phba);
2698 
2699 }
2700 
2701 void
2702 lpfc_nvme_cancel_iocb(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
2703 		      uint32_t stat, uint32_t param)
2704 {
2705 #if (IS_ENABLED(CONFIG_NVME_FC))
2706 	struct lpfc_io_buf *lpfc_ncmd;
2707 	struct nvmefc_fcp_req *nCmd;
2708 	struct lpfc_wcqe_complete wcqe;
2709 	struct lpfc_wcqe_complete *wcqep = &wcqe;
2710 
2711 	lpfc_ncmd = (struct lpfc_io_buf *)pwqeIn->context1;
2712 	if (!lpfc_ncmd) {
2713 		lpfc_sli_release_iocbq(phba, pwqeIn);
2714 		return;
2715 	}
2716 	/* For abort iocb just return, IO iocb will do a done call */
2717 	if (bf_get(wqe_cmnd, &pwqeIn->wqe.gen_req.wqe_com) ==
2718 	    CMD_ABORT_XRI_CX) {
2719 		lpfc_sli_release_iocbq(phba, pwqeIn);
2720 		return;
2721 	}
2722 
2723 	spin_lock(&lpfc_ncmd->buf_lock);
2724 	nCmd = lpfc_ncmd->nvmeCmd;
2725 	if (!nCmd) {
2726 		spin_unlock(&lpfc_ncmd->buf_lock);
2727 		lpfc_release_nvme_buf(phba, lpfc_ncmd);
2728 		return;
2729 	}
2730 	spin_unlock(&lpfc_ncmd->buf_lock);
2731 
2732 	lpfc_printf_log(phba, KERN_INFO, LOG_NVME_IOERR,
2733 			"6194 NVME Cancel xri %x\n",
2734 			lpfc_ncmd->cur_iocbq.sli4_xritag);
2735 
2736 	wcqep->word0 = 0;
2737 	bf_set(lpfc_wcqe_c_status, wcqep, stat);
2738 	wcqep->parameter = param;
2739 	wcqep->word3 = 0; /* xb is 0 */
2740 
2741 	/* Call release with XB=1 to queue the IO into the abort list. */
2742 	if (phba->sli.sli_flag & LPFC_SLI_ACTIVE)
2743 		bf_set(lpfc_wcqe_c_xb, wcqep, 1);
2744 
2745 	memcpy(&pwqeIn->wcqe_cmpl, wcqep, sizeof(*wcqep));
2746 	(pwqeIn->cmd_cmpl)(phba, pwqeIn, pwqeIn);
2747 #endif
2748 }
2749