GNU Linux-libre 4.14.266-gnu1
[releases.git] / drivers / scsi / lpfc / lpfc_nvme.c
1 /*******************************************************************
2  * This file is part of the Emulex Linux Device Driver for         *
3  * Fibre Channel Host Bus Adapters.                                *
4  * Copyright (C) 2017 Broadcom. All Rights Reserved. The term      *
5  * “Broadcom” refers to Broadcom Limited and/or its subsidiaries.  *
6  * Copyright (C) 2004-2016 Emulex.  All rights reserved.           *
7  * EMULEX and SLI are trademarks of Emulex.                        *
8  * www.broadcom.com                                                *
9  * Portions Copyright (C) 2004-2005 Christoph Hellwig              *
10  *                                                                 *
11  * This program is free software; you can redistribute it and/or   *
12  * modify it under the terms of version 2 of the GNU General       *
13  * Public License as published by the Free Software Foundation.    *
14  * This program is distributed in the hope that it will be useful. *
15  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
16  * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
17  * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
18  * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
19  * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
20  * more details, a copy of which can be found in the file COPYING  *
21  * included with this package.                                     *
22  ********************************************************************/
23 #include <linux/pci.h>
24 #include <linux/slab.h>
25 #include <linux/interrupt.h>
26 #include <linux/delay.h>
27 #include <asm/unaligned.h>
28 #include <linux/crc-t10dif.h>
29 #include <net/checksum.h>
30
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_tcq.h>
36 #include <scsi/scsi_transport_fc.h>
37 #include <scsi/fc/fc_fs.h>
38
39 #include <linux/nvme.h>
40 #include <linux/nvme-fc-driver.h>
41 #include <linux/nvme-fc.h>
42 #include "lpfc_version.h"
43 #include "lpfc_hw4.h"
44 #include "lpfc_hw.h"
45 #include "lpfc_sli.h"
46 #include "lpfc_sli4.h"
47 #include "lpfc_nl.h"
48 #include "lpfc_disc.h"
49 #include "lpfc.h"
50 #include "lpfc_nvme.h"
51 #include "lpfc_scsi.h"
52 #include "lpfc_logmsg.h"
53 #include "lpfc_crtn.h"
54 #include "lpfc_vport.h"
55 #include "lpfc_debugfs.h"
56
57 /* NVME initiator-based functions */
58
59 static struct lpfc_nvme_buf *
60 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp);
61
62 static void
63 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_nvme_buf *);
64
65
66 /**
67  * lpfc_nvme_create_queue -
68  * @lpfc_pnvme: Pointer to the driver's nvme instance data
69  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
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         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
92         vport = lport->vport;
93         qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
94         if (qhandle == NULL)
95                 return -ENOMEM;
96
97         qhandle->cpu_id = smp_processor_id();
98         qhandle->qidx = qidx;
99         /*
100          * NVME qidx == 0 is the admin queue, so both admin queue
101          * and first IO queue will use MSI-X vector and associated
102          * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
103          */
104         if (qidx) {
105                 str = "IO ";  /* IO queue */
106                 qhandle->index = ((qidx - 1) %
107                         vport->phba->cfg_nvme_io_channel);
108         } else {
109                 str = "ADM";  /* Admin queue */
110                 qhandle->index = qidx;
111         }
112
113         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
114                          "6073 Binding %s HdwQueue %d  (cpu %d) to "
115                          "io_channel %d qhandle %p\n", str,
116                          qidx, qhandle->cpu_id, qhandle->index, qhandle);
117         *handle = (void *)qhandle;
118         return 0;
119 }
120
121 /**
122  * lpfc_nvme_delete_queue -
123  * @lpfc_pnvme: Pointer to the driver's nvme instance data
124  * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
125  * @handle: An opaque driver handle from lpfc_nvme_create_queue
126  *
127  * Driver registers this routine to free
128  * any internal data structures to bind the @qidx to its internal
129  * IO queues.
130  *
131  * Return value :
132  *   0 - Success
133  *   TODO:  What are the failure codes.
134  **/
135 static void
136 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
137                        unsigned int qidx,
138                        void *handle)
139 {
140         struct lpfc_nvme_lport *lport;
141         struct lpfc_vport *vport;
142
143         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
144         vport = lport->vport;
145
146         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
147                         "6001 ENTER.  lpfc_pnvme %p, qidx x%x qhandle %p\n",
148                         lport, qidx, handle);
149         kfree(handle);
150 }
151
152 static void
153 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
154 {
155         struct lpfc_nvme_lport *lport = localport->private;
156
157         /* release any threads waiting for the unreg to complete */
158         complete(&lport->lport_unreg_done);
159 }
160
161 /* lpfc_nvme_remoteport_delete
162  *
163  * @remoteport: Pointer to an nvme transport remoteport instance.
164  *
165  * This is a template downcall.  NVME transport calls this function
166  * when it has completed the unregistration of a previously
167  * registered remoteport.
168  *
169  * Return value :
170  * None
171  */
172 void
173 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
174 {
175         struct lpfc_nvme_rport *rport = remoteport->private;
176         struct lpfc_vport *vport;
177         struct lpfc_nodelist *ndlp;
178
179         ndlp = rport->ndlp;
180         if (!ndlp)
181                 goto rport_err;
182
183         vport = ndlp->vport;
184         if (!vport)
185                 goto rport_err;
186
187         /* Remove this rport from the lport's list - memory is owned by the
188          * transport. Remove the ndlp reference for the NVME transport before
189          * calling state machine to remove the node.
190          */
191         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
192                         "6146 remoteport delete complete %p\n",
193                         remoteport);
194         ndlp->nrport = NULL;
195         lpfc_nlp_put(ndlp);
196
197  rport_err:
198         /* This call has to execute as long as the rport is valid.
199          * Release any threads waiting for the unreg to complete.
200          */
201         complete(&rport->rport_unreg_done);
202 }
203
204 static void
205 lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
206                        struct lpfc_wcqe_complete *wcqe)
207 {
208         struct lpfc_vport *vport = cmdwqe->vport;
209         uint32_t status;
210         struct nvmefc_ls_req *pnvme_lsreq;
211         struct lpfc_dmabuf *buf_ptr;
212         struct lpfc_nodelist *ndlp;
213
214         atomic_inc(&vport->phba->fc4NvmeLsCmpls);
215
216         pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
217         status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
218         ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
219         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
220                          "6047 nvme cmpl Enter "
221                          "Data %p DID %x Xri: %x status %x cmd:%p lsreg:%p "
222                          "bmp:%p ndlp:%p\n",
223                          pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
224                          cmdwqe->sli4_xritag, status,
225                          cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
226
227         lpfc_nvmeio_data(phba, "NVME LS  CMPL: xri x%x stat x%x parm x%x\n",
228                          cmdwqe->sli4_xritag, status, wcqe->parameter);
229
230         if (cmdwqe->context3) {
231                 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
232                 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
233                 kfree(buf_ptr);
234                 cmdwqe->context3 = NULL;
235         }
236         if (pnvme_lsreq->done)
237                 pnvme_lsreq->done(pnvme_lsreq, status);
238         else
239                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
240                                  "6046 nvme cmpl without done call back? "
241                                  "Data %p DID %x Xri: %x status %x\n",
242                                 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
243                                 cmdwqe->sli4_xritag, status);
244         if (ndlp) {
245                 lpfc_nlp_put(ndlp);
246                 cmdwqe->context1 = NULL;
247         }
248         lpfc_sli_release_iocbq(phba, cmdwqe);
249 }
250
251 static int
252 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
253                   struct lpfc_dmabuf *inp,
254                  struct nvmefc_ls_req *pnvme_lsreq,
255              void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
256                            struct lpfc_wcqe_complete *),
257              struct lpfc_nodelist *ndlp, uint32_t num_entry,
258              uint32_t tmo, uint8_t retry)
259 {
260         struct lpfc_hba  *phba = vport->phba;
261         union lpfc_wqe *wqe;
262         struct lpfc_iocbq *genwqe;
263         struct ulp_bde64 *bpl;
264         struct ulp_bde64 bde;
265         int i, rc, xmit_len, first_len;
266
267         /* Allocate buffer for  command WQE */
268         genwqe = lpfc_sli_get_iocbq(phba);
269         if (genwqe == NULL)
270                 return 1;
271
272         wqe = &genwqe->wqe;
273         memset(wqe, 0, sizeof(union lpfc_wqe));
274
275         genwqe->context3 = (uint8_t *)bmp;
276         genwqe->iocb_flag |= LPFC_IO_NVME_LS;
277
278         /* Save for completion so we can release these resources */
279         genwqe->context1 = lpfc_nlp_get(ndlp);
280         genwqe->context2 = (uint8_t *)pnvme_lsreq;
281         /* Fill in payload, bp points to frame payload */
282
283         if (!tmo)
284                 /* FC spec states we need 3 * ratov for CT requests */
285                 tmo = (3 * phba->fc_ratov);
286
287         /* For this command calculate the xmit length of the request bde. */
288         xmit_len = 0;
289         first_len = 0;
290         bpl = (struct ulp_bde64 *)bmp->virt;
291         for (i = 0; i < num_entry; i++) {
292                 bde.tus.w = bpl[i].tus.w;
293                 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
294                         break;
295                 xmit_len += bde.tus.f.bdeSize;
296                 if (i == 0)
297                         first_len = xmit_len;
298         }
299
300         genwqe->rsvd2 = num_entry;
301         genwqe->hba_wqidx = 0;
302
303         /* Words 0 - 2 */
304         wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
305         wqe->generic.bde.tus.f.bdeSize = first_len;
306         wqe->generic.bde.addrLow = bpl[0].addrLow;
307         wqe->generic.bde.addrHigh = bpl[0].addrHigh;
308
309         /* Word 3 */
310         wqe->gen_req.request_payload_len = first_len;
311
312         /* Word 4 */
313
314         /* Word 5 */
315         bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
316         bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
317         bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
318         bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
319         bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
320
321         /* Word 6 */
322         bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
323                phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
324         bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
325
326         /* Word 7 */
327         bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1));
328         bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
329         bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
330         bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
331
332         /* Word 8 */
333         wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
334
335         /* Word 9 */
336         bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
337
338         /* Word 10 */
339         bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
340         bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
341         bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
342         bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
343         bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
344
345         /* Word 11 */
346         bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
347         bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
348
349
350         /* Issue GEN REQ WQE for NPORT <did> */
351         lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
352                          "6050 Issue GEN REQ WQE to NPORT x%x "
353                          "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n",
354                          ndlp->nlp_DID, genwqe->iotag,
355                          vport->port_state,
356                         genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
357         genwqe->wqe_cmpl = cmpl;
358         genwqe->iocb_cmpl = NULL;
359         genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
360         genwqe->vport = vport;
361         genwqe->retry = retry;
362
363         lpfc_nvmeio_data(phba, "NVME LS  XMIT: xri x%x iotag x%x to x%06x\n",
364                          genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
365
366         rc = lpfc_sli4_issue_wqe(phba, LPFC_ELS_RING, genwqe);
367         if (rc) {
368                 lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
369                                  "6045 Issue GEN REQ WQE to NPORT x%x "
370                                  "Data: x%x x%x\n",
371                                  ndlp->nlp_DID, genwqe->iotag,
372                                  vport->port_state);
373                 lpfc_sli_release_iocbq(phba, genwqe);
374                 return 1;
375         }
376         return 0;
377 }
378
379 /**
380  * lpfc_nvme_ls_req - Issue an Link Service request
381  * @lpfc_pnvme: Pointer to the driver's nvme instance data
382  * @lpfc_nvme_lport: Pointer to the driver's local port data
383  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
384  *
385  * Driver registers this routine to handle any link service request
386  * from the nvme_fc transport to a remote nvme-aware port.
387  *
388  * Return value :
389  *   0 - Success
390  *   TODO: What are the failure codes.
391  **/
392 static int
393 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
394                  struct nvme_fc_remote_port *pnvme_rport,
395                  struct nvmefc_ls_req *pnvme_lsreq)
396 {
397         int ret = 0;
398         struct lpfc_nvme_lport *lport;
399         struct lpfc_vport *vport;
400         struct lpfc_nodelist *ndlp;
401         struct ulp_bde64 *bpl;
402         struct lpfc_dmabuf *bmp;
403         uint16_t ntype, nstate;
404
405         /* there are two dma buf in the request, actually there is one and
406          * the second one is just the start address + cmd size.
407          * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
408          * in a lpfc_dmabuf struct. When freeing we just free the wrapper
409          * because the nvem layer owns the data bufs.
410          * We do not have to break these packets open, we don't care what is in
411          * them. And we do not have to look at the resonse data, we only care
412          * that we got a response. All of the caring is going to happen in the
413          * nvme-fc layer.
414          */
415
416         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
417         vport = lport->vport;
418
419         ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
420         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
421                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
422                                  "6051 DID x%06x not an active rport.\n",
423                                  pnvme_rport->port_id);
424                 return -ENODEV;
425         }
426
427         /* The remote node has to be a mapped nvme target or an
428          * unmapped nvme initiator or it's an error.
429          */
430         ntype = ndlp->nlp_type;
431         nstate = ndlp->nlp_state;
432         if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
433             (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
434                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
435                                  "6088 DID x%06x not ready for "
436                                  "IO. State x%x, Type x%x\n",
437                                  pnvme_rport->port_id,
438                                  ndlp->nlp_state, ndlp->nlp_type);
439                 return -ENODEV;
440         }
441         bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
442         if (!bmp) {
443
444                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
445                                  "6044 Could not find node for DID %x\n",
446                                  pnvme_rport->port_id);
447                 return 2;
448         }
449         INIT_LIST_HEAD(&bmp->list);
450         bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
451         if (!bmp->virt) {
452                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
453                                  "6042 Could not find node for DID %x\n",
454                                  pnvme_rport->port_id);
455                 kfree(bmp);
456                 return 3;
457         }
458         bpl = (struct ulp_bde64 *)bmp->virt;
459         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
460         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
461         bpl->tus.f.bdeFlags = 0;
462         bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
463         bpl->tus.w = le32_to_cpu(bpl->tus.w);
464         bpl++;
465
466         bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
467         bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
468         bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
469         bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
470         bpl->tus.w = le32_to_cpu(bpl->tus.w);
471
472         /* Expand print to include key fields. */
473         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
474                          "6149 ENTER.  lport %p, rport %p lsreq%p rqstlen:%d "
475                          "rsplen:%d %pad %pad\n",
476                          pnvme_lport, pnvme_rport,
477                          pnvme_lsreq, pnvme_lsreq->rqstlen,
478                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
479                          &pnvme_lsreq->rspdma);
480
481         atomic_inc(&vport->phba->fc4NvmeLsRequests);
482
483         /* Hardcode the wait to 30 seconds.  Connections are failing otherwise.
484          * This code allows it all to work.
485          */
486         ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
487                                 pnvme_lsreq, lpfc_nvme_cmpl_gen_req,
488                                 ndlp, 2, 30, 0);
489         if (ret != WQE_SUCCESS) {
490                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
491                                  "6052 EXIT. issue ls wqe failed lport %p, "
492                                  "rport %p lsreq%p Status %x DID %x\n",
493                                  pnvme_lport, pnvme_rport, pnvme_lsreq,
494                                  ret, ndlp->nlp_DID);
495                 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
496                 kfree(bmp);
497                 return ret;
498         }
499
500         /* Stub in routine and return 0 for now. */
501         return ret;
502 }
503
504 /**
505  * lpfc_nvme_ls_abort - Issue an Link Service request
506  * @lpfc_pnvme: Pointer to the driver's nvme instance data
507  * @lpfc_nvme_lport: Pointer to the driver's local port data
508  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
509  *
510  * Driver registers this routine to handle any link service request
511  * from the nvme_fc transport to a remote nvme-aware port.
512  *
513  * Return value :
514  *   0 - Success
515  *   TODO: What are the failure codes.
516  **/
517 static void
518 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
519                    struct nvme_fc_remote_port *pnvme_rport,
520                    struct nvmefc_ls_req *pnvme_lsreq)
521 {
522         struct lpfc_nvme_lport *lport;
523         struct lpfc_vport *vport;
524         struct lpfc_hba *phba;
525         struct lpfc_nodelist *ndlp;
526         LIST_HEAD(abort_list);
527         struct lpfc_sli_ring *pring;
528         struct lpfc_iocbq *wqe, *next_wqe;
529
530         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
531         vport = lport->vport;
532         phba = vport->phba;
533
534         ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
535         if (!ndlp) {
536                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
537                                  "6049 Could not find node for DID %x\n",
538                                  pnvme_rport->port_id);
539                 return;
540         }
541
542         /* Expand print to include key fields. */
543         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
544                          "6040 ENTER.  lport %p, rport %p lsreq %p rqstlen:%d "
545                          "rsplen:%d %pad %pad\n",
546                          pnvme_lport, pnvme_rport,
547                          pnvme_lsreq, pnvme_lsreq->rqstlen,
548                          pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
549                          &pnvme_lsreq->rspdma);
550
551         /*
552          * Lock the ELS ring txcmplq and build a local list of all ELS IOs
553          * that need an ABTS.  The IOs need to stay on the txcmplq so that
554          * the abort operation completes them successfully.
555          */
556         pring = phba->sli4_hba.nvmels_wq->pring;
557         spin_lock_irq(&phba->hbalock);
558         spin_lock(&pring->ring_lock);
559         list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
560                 /* Add to abort_list on on NDLP match. */
561                 if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) {
562                         wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
563                         list_add_tail(&wqe->dlist, &abort_list);
564                 }
565         }
566         spin_unlock(&pring->ring_lock);
567         spin_unlock_irq(&phba->hbalock);
568
569         /* Abort the targeted IOs and remove them from the abort list. */
570         list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) {
571                 spin_lock_irq(&phba->hbalock);
572                 list_del_init(&wqe->dlist);
573                 lpfc_sli_issue_abort_iotag(phba, pring, wqe);
574                 spin_unlock_irq(&phba->hbalock);
575         }
576 }
577
578 /* Fix up the existing sgls for NVME IO. */
579 static void
580 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
581                        struct lpfc_nvme_buf *lpfc_ncmd,
582                        struct nvmefc_fcp_req *nCmd)
583 {
584         struct sli4_sge *sgl;
585         union lpfc_wqe128 *wqe;
586         uint32_t *wptr, *dptr;
587
588         /*
589          * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
590          * match NVME.  NVME sends 96 bytes. Also, use the
591          * nvme commands command and response dma addresses
592          * rather than the virtual memory to ease the restore
593          * operation.
594          */
595         sgl = lpfc_ncmd->nvme_sgl;
596         sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
597
598         sgl++;
599
600         /* Setup the physical region for the FCP RSP */
601         sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
602         sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
603         sgl->word2 = le32_to_cpu(sgl->word2);
604         if (nCmd->sg_cnt)
605                 bf_set(lpfc_sli4_sge_last, sgl, 0);
606         else
607                 bf_set(lpfc_sli4_sge_last, sgl, 1);
608         sgl->word2 = cpu_to_le32(sgl->word2);
609         sgl->sge_len = cpu_to_le32(nCmd->rsplen);
610
611         /*
612          * Get a local pointer to the built-in wqe and correct
613          * the cmd size to match NVME's 96 bytes and fix
614          * the dma address.
615          */
616
617         /* 128 byte wqe support here */
618         wqe = (union lpfc_wqe128 *)&lpfc_ncmd->cur_iocbq.wqe;
619
620         /* Word 0-2 - NVME CMND IU (embedded payload) */
621         wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
622         wqe->generic.bde.tus.f.bdeSize = 60;
623         wqe->generic.bde.addrHigh = 0;
624         wqe->generic.bde.addrLow =  64;  /* Word 16 */
625
626         /* Word 3 */
627         bf_set(payload_offset_len, &wqe->fcp_icmd,
628                (nCmd->rsplen + nCmd->cmdlen));
629
630         /* Word 10 */
631         bf_set(wqe_nvme, &wqe->fcp_icmd.wqe_com, 1);
632         bf_set(wqe_wqes, &wqe->fcp_icmd.wqe_com, 1);
633
634         /*
635          * Embed the payload in the last half of the WQE
636          * WQE words 16-30 get the NVME CMD IU payload
637          *
638          * WQE words 16-19 get payload Words 1-4
639          * WQE words 20-21 get payload Words 6-7
640          * WQE words 22-29 get payload Words 16-23
641          */
642         wptr = &wqe->words[16];  /* WQE ptr */
643         dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
644         dptr++;                 /* Skip Word 0 in payload */
645
646         *wptr++ = *dptr++;      /* Word 1 */
647         *wptr++ = *dptr++;      /* Word 2 */
648         *wptr++ = *dptr++;      /* Word 3 */
649         *wptr++ = *dptr++;      /* Word 4 */
650         dptr++;                 /* Skip Word 5 in payload */
651         *wptr++ = *dptr++;      /* Word 6 */
652         *wptr++ = *dptr++;      /* Word 7 */
653         dptr += 8;              /* Skip Words 8-15 in payload */
654         *wptr++ = *dptr++;      /* Word 16 */
655         *wptr++ = *dptr++;      /* Word 17 */
656         *wptr++ = *dptr++;      /* Word 18 */
657         *wptr++ = *dptr++;      /* Word 19 */
658         *wptr++ = *dptr++;      /* Word 20 */
659         *wptr++ = *dptr++;      /* Word 21 */
660         *wptr++ = *dptr++;      /* Word 22 */
661         *wptr   = *dptr;        /* Word 23 */
662 }
663
664 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
665 static void
666 lpfc_nvme_ktime(struct lpfc_hba *phba,
667                 struct lpfc_nvme_buf *lpfc_ncmd)
668 {
669         uint64_t seg1, seg2, seg3, seg4;
670
671         if (!phba->ktime_on)
672                 return;
673         if (!lpfc_ncmd->ts_last_cmd ||
674             !lpfc_ncmd->ts_cmd_start ||
675             !lpfc_ncmd->ts_cmd_wqput ||
676             !lpfc_ncmd->ts_isr_cmpl ||
677             !lpfc_ncmd->ts_data_nvme)
678                 return;
679         if (lpfc_ncmd->ts_cmd_start < lpfc_ncmd->ts_last_cmd)
680                 return;
681         if (lpfc_ncmd->ts_cmd_wqput < lpfc_ncmd->ts_cmd_start)
682                 return;
683         if (lpfc_ncmd->ts_isr_cmpl < lpfc_ncmd->ts_cmd_wqput)
684                 return;
685         if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_isr_cmpl)
686                 return;
687         /*
688          * Segment 1 - Time from Last FCP command cmpl is handed
689          * off to NVME Layer to start of next command.
690          * Segment 2 - Time from Driver receives a IO cmd start
691          * from NVME Layer to WQ put is done on IO cmd.
692          * Segment 3 - Time from Driver WQ put is done on IO cmd
693          * to MSI-X ISR for IO cmpl.
694          * Segment 4 - Time from MSI-X ISR for IO cmpl to when
695          * cmpl is handled off to the NVME Layer.
696          */
697         seg1 = lpfc_ncmd->ts_cmd_start - lpfc_ncmd->ts_last_cmd;
698         if (seg1 > 5000000)  /* 5 ms - for sequential IOs */
699                 return;
700
701         /* Calculate times relative to start of IO */
702         seg2 = (lpfc_ncmd->ts_cmd_wqput - lpfc_ncmd->ts_cmd_start);
703         seg3 = (lpfc_ncmd->ts_isr_cmpl -
704                 lpfc_ncmd->ts_cmd_start) - seg2;
705         seg4 = (lpfc_ncmd->ts_data_nvme -
706                 lpfc_ncmd->ts_cmd_start) - seg2 - seg3;
707         phba->ktime_data_samples++;
708         phba->ktime_seg1_total += seg1;
709         if (seg1 < phba->ktime_seg1_min)
710                 phba->ktime_seg1_min = seg1;
711         else if (seg1 > phba->ktime_seg1_max)
712                 phba->ktime_seg1_max = seg1;
713         phba->ktime_seg2_total += seg2;
714         if (seg2 < phba->ktime_seg2_min)
715                 phba->ktime_seg2_min = seg2;
716         else if (seg2 > phba->ktime_seg2_max)
717                 phba->ktime_seg2_max = seg2;
718         phba->ktime_seg3_total += seg3;
719         if (seg3 < phba->ktime_seg3_min)
720                 phba->ktime_seg3_min = seg3;
721         else if (seg3 > phba->ktime_seg3_max)
722                 phba->ktime_seg3_max = seg3;
723         phba->ktime_seg4_total += seg4;
724         if (seg4 < phba->ktime_seg4_min)
725                 phba->ktime_seg4_min = seg4;
726         else if (seg4 > phba->ktime_seg4_max)
727                 phba->ktime_seg4_max = seg4;
728
729         lpfc_ncmd->ts_last_cmd = 0;
730         lpfc_ncmd->ts_cmd_start = 0;
731         lpfc_ncmd->ts_cmd_wqput  = 0;
732         lpfc_ncmd->ts_isr_cmpl = 0;
733         lpfc_ncmd->ts_data_nvme = 0;
734 }
735 #endif
736
737 /**
738  * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO
739  * @lpfc_pnvme: Pointer to the driver's nvme instance data
740  * @lpfc_nvme_lport: Pointer to the driver's local port data
741  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
742  *
743  * Driver registers this routine as it io request handler.  This
744  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
745  * data structure to the rport indicated in @lpfc_nvme_rport.
746  *
747  * Return value :
748  *   0 - Success
749  *   TODO: What are the failure codes.
750  **/
751 static void
752 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
753                           struct lpfc_wcqe_complete *wcqe)
754 {
755         struct lpfc_nvme_buf *lpfc_ncmd =
756                 (struct lpfc_nvme_buf *)pwqeIn->context1;
757         struct lpfc_vport *vport = pwqeIn->vport;
758         struct nvmefc_fcp_req *nCmd;
759         struct nvme_fc_ersp_iu *ep;
760         struct nvme_fc_cmd_iu *cp;
761         struct lpfc_nvme_rport *rport;
762         struct lpfc_nodelist *ndlp;
763         struct lpfc_nvme_fcpreq_priv *freqpriv;
764         unsigned long flags;
765         uint32_t code;
766         uint16_t cid, sqhd, data;
767         uint32_t *ptr;
768
769         /* Sanity check on return of outstanding command */
770         if (!lpfc_ncmd || !lpfc_ncmd->nvmeCmd || !lpfc_ncmd->nrport) {
771                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
772                                  "6071 Completion pointers bad on wqe %p.\n",
773                                  wcqe);
774                 return;
775         }
776         atomic_inc(&phba->fc4NvmeIoCmpls);
777
778         nCmd = lpfc_ncmd->nvmeCmd;
779         rport = lpfc_ncmd->nrport;
780
781         lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
782                          lpfc_ncmd->cur_iocbq.sli4_xritag,
783                          bf_get(lpfc_wcqe_c_status, wcqe), wcqe->parameter);
784         /*
785          * Catch race where our node has transitioned, but the
786          * transport is still transitioning.
787          */
788         ndlp = rport->ndlp;
789         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
790                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
791                                  "6061 rport %p,  DID x%06x node not ready.\n",
792                                  rport, rport->remoteport->port_id);
793
794                 ndlp = lpfc_findnode_did(vport, rport->remoteport->port_id);
795                 if (!ndlp) {
796                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
797                                          "6062 Ignoring NVME cmpl.  No ndlp\n");
798                         goto out_err;
799                 }
800         }
801
802         code = bf_get(lpfc_wcqe_c_code, wcqe);
803         if (code == CQE_CODE_NVME_ERSP) {
804                 /* For this type of CQE, we need to rebuild the rsp */
805                 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
806
807                 /*
808                  * Get Command Id from cmd to plug into response. This
809                  * code is not needed in the next NVME Transport drop.
810                  */
811                 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
812                 cid = cp->sqe.common.command_id;
813
814                 /*
815                  * RSN is in CQE word 2
816                  * SQHD is in CQE Word 3 bits 15:0
817                  * Cmd Specific info is in CQE Word 1
818                  * and in CQE Word 0 bits 15:0
819                  */
820                 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
821
822                 /* Now lets build the NVME ERSP IU */
823                 ep->iu_len = cpu_to_be16(8);
824                 ep->rsn = wcqe->parameter;
825                 ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
826                 ep->rsvd12 = 0;
827                 ptr = (uint32_t *)&ep->cqe.result.u64;
828                 *ptr++ = wcqe->total_data_placed;
829                 data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
830                 *ptr = (uint32_t)data;
831                 ep->cqe.sq_head = sqhd;
832                 ep->cqe.sq_id =  nCmd->sqid;
833                 ep->cqe.command_id = cid;
834                 ep->cqe.status = 0;
835
836                 lpfc_ncmd->status = IOSTAT_SUCCESS;
837                 lpfc_ncmd->result = 0;
838                 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
839                 nCmd->transferred_length = nCmd->payload_length;
840         } else {
841                 lpfc_ncmd->status = (bf_get(lpfc_wcqe_c_status, wcqe) &
842                             LPFC_IOCB_STATUS_MASK);
843                 lpfc_ncmd->result = wcqe->parameter;
844
845                 /* For NVME, the only failure path that results in an
846                  * IO error is when the adapter rejects it.  All other
847                  * conditions are a success case and resolved by the
848                  * transport.
849                  * IOSTAT_FCP_RSP_ERROR means:
850                  * 1. Length of data received doesn't match total
851                  *    transfer length in WQE
852                  * 2. If the RSP payload does NOT match these cases:
853                  *    a. RSP length 12/24 bytes and all zeros
854                  *    b. NVME ERSP
855                  */
856                 switch (lpfc_ncmd->status) {
857                 case IOSTAT_SUCCESS:
858                         nCmd->transferred_length = wcqe->total_data_placed;
859                         nCmd->rcv_rsplen = 0;
860                         nCmd->status = 0;
861                         break;
862                 case IOSTAT_FCP_RSP_ERROR:
863                         nCmd->transferred_length = wcqe->total_data_placed;
864                         nCmd->rcv_rsplen = wcqe->parameter;
865                         nCmd->status = 0;
866                         /* Sanity check */
867                         if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN)
868                                 break;
869                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
870                                          "6081 NVME Completion Protocol Error: "
871                                          "xri %x status x%x result x%x "
872                                          "placed x%x\n",
873                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
874                                          lpfc_ncmd->status, lpfc_ncmd->result,
875                                          wcqe->total_data_placed);
876                         break;
877                 default:
878 out_err:
879                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
880                                          "6072 NVME Completion Error: xri %x "
881                                          "status x%x result x%x placed x%x\n",
882                                          lpfc_ncmd->cur_iocbq.sli4_xritag,
883                                          lpfc_ncmd->status, lpfc_ncmd->result,
884                                          wcqe->total_data_placed);
885                         nCmd->transferred_length = 0;
886                         nCmd->rcv_rsplen = 0;
887                         nCmd->status = NVME_SC_INTERNAL;
888                 }
889         }
890
891         /* pick up SLI4 exhange busy condition */
892         if (bf_get(lpfc_wcqe_c_xb, wcqe))
893                 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
894         else
895                 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
896
897         if (ndlp && NLP_CHK_NODE_ACT(ndlp))
898                 atomic_dec(&ndlp->cmd_pending);
899
900         /* Update stats and complete the IO.  There is
901          * no need for dma unprep because the nvme_transport
902          * owns the dma address.
903          */
904 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
905         if (phba->ktime_on) {
906                 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
907                 lpfc_ncmd->ts_data_nvme = ktime_get_ns();
908                 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme;
909                 lpfc_nvme_ktime(phba, lpfc_ncmd);
910         }
911         if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
912                 if (lpfc_ncmd->cpu != smp_processor_id())
913                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
914                                          "6701 CPU Check cmpl: "
915                                          "cpu %d expect %d\n",
916                                          smp_processor_id(), lpfc_ncmd->cpu);
917                 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT)
918                         phba->cpucheck_cmpl_io[lpfc_ncmd->cpu]++;
919         }
920 #endif
921         freqpriv = nCmd->private;
922         freqpriv->nvme_buf = NULL;
923         nCmd->done(nCmd);
924
925         spin_lock_irqsave(&phba->hbalock, flags);
926         lpfc_ncmd->nrport = NULL;
927         spin_unlock_irqrestore(&phba->hbalock, flags);
928
929         lpfc_release_nvme_buf(phba, lpfc_ncmd);
930 }
931
932
933 /**
934  * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
935  * @lpfc_pnvme: Pointer to the driver's nvme instance data
936  * @lpfc_nvme_lport: Pointer to the driver's local port data
937  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
938  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
939  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
940  *
941  * Driver registers this routine as it io request handler.  This
942  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
943  * data structure to the rport indicated in @lpfc_nvme_rport.
944  *
945  * Return value :
946  *   0 - Success
947  *   TODO: What are the failure codes.
948  **/
949 static int
950 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
951                       struct lpfc_nvme_buf *lpfc_ncmd,
952                       struct lpfc_nodelist *pnode)
953 {
954         struct lpfc_hba *phba = vport->phba;
955         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
956         struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
957         union lpfc_wqe128 *wqe = (union lpfc_wqe128 *)&pwqeq->wqe;
958         uint32_t req_len;
959
960         if (!pnode || !NLP_CHK_NODE_ACT(pnode))
961                 return -EINVAL;
962
963         /*
964          * There are three possibilities here - use scatter-gather segment, use
965          * the single mapping, or neither.
966          */
967         wqe->fcp_iwrite.initial_xfer_len = 0;
968         if (nCmd->sg_cnt) {
969                 if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
970                         /* Word 5 */
971                         if ((phba->cfg_nvme_enable_fb) &&
972                             (pnode->nlp_flag & NLP_FIRSTBURST)) {
973                                 req_len = lpfc_ncmd->nvmeCmd->payload_length;
974                                 if (req_len < pnode->nvme_fb_size)
975                                         wqe->fcp_iwrite.initial_xfer_len =
976                                                 req_len;
977                                 else
978                                         wqe->fcp_iwrite.initial_xfer_len =
979                                                 pnode->nvme_fb_size;
980                         }
981
982                         /* Word 7 */
983                         bf_set(wqe_cmnd, &wqe->generic.wqe_com,
984                                CMD_FCP_IWRITE64_WQE);
985                         bf_set(wqe_pu, &wqe->generic.wqe_com,
986                                PARM_READ_CHECK);
987
988                         /* Word 10 */
989                         bf_set(wqe_qosd, &wqe->fcp_iwrite.wqe_com, 0);
990                         bf_set(wqe_iod, &wqe->fcp_iwrite.wqe_com,
991                                LPFC_WQE_IOD_WRITE);
992                         bf_set(wqe_lenloc, &wqe->fcp_iwrite.wqe_com,
993                                LPFC_WQE_LENLOC_WORD4);
994                         if (phba->cfg_nvme_oas)
995                                 bf_set(wqe_oas, &wqe->fcp_iwrite.wqe_com, 1);
996
997                         /* Word 11 */
998                         bf_set(wqe_cmd_type, &wqe->generic.wqe_com,
999                                NVME_WRITE_CMD);
1000
1001                         atomic_inc(&phba->fc4NvmeOutputRequests);
1002                 } else {
1003                         /* Word 7 */
1004                         bf_set(wqe_cmnd, &wqe->generic.wqe_com,
1005                                CMD_FCP_IREAD64_WQE);
1006                         bf_set(wqe_pu, &wqe->generic.wqe_com,
1007                                PARM_READ_CHECK);
1008
1009                         /* Word 10 */
1010                         bf_set(wqe_qosd, &wqe->fcp_iread.wqe_com, 0);
1011                         bf_set(wqe_iod, &wqe->fcp_iread.wqe_com,
1012                                LPFC_WQE_IOD_READ);
1013                         bf_set(wqe_lenloc, &wqe->fcp_iread.wqe_com,
1014                                LPFC_WQE_LENLOC_WORD4);
1015                         if (phba->cfg_nvme_oas)
1016                                 bf_set(wqe_oas, &wqe->fcp_iread.wqe_com, 1);
1017
1018                         /* Word 11 */
1019                         bf_set(wqe_cmd_type, &wqe->generic.wqe_com,
1020                                NVME_READ_CMD);
1021
1022                         atomic_inc(&phba->fc4NvmeInputRequests);
1023                 }
1024         } else {
1025                 /* Word 4 */
1026                 wqe->fcp_icmd.rsrvd4 = 0;
1027
1028                 /* Word 7 */
1029                 bf_set(wqe_cmnd, &wqe->generic.wqe_com, CMD_FCP_ICMND64_WQE);
1030                 bf_set(wqe_pu, &wqe->generic.wqe_com, 0);
1031
1032                 /* Word 10 */
1033                 bf_set(wqe_qosd, &wqe->fcp_icmd.wqe_com, 1);
1034                 bf_set(wqe_iod, &wqe->fcp_icmd.wqe_com, LPFC_WQE_IOD_WRITE);
1035                 bf_set(wqe_lenloc, &wqe->fcp_icmd.wqe_com,
1036                        LPFC_WQE_LENLOC_NONE);
1037                 if (phba->cfg_nvme_oas)
1038                         bf_set(wqe_oas, &wqe->fcp_icmd.wqe_com, 1);
1039
1040                 /* Word 11 */
1041                 bf_set(wqe_cmd_type, &wqe->generic.wqe_com, NVME_READ_CMD);
1042
1043                 atomic_inc(&phba->fc4NvmeControlRequests);
1044         }
1045         /*
1046          * Finish initializing those WQE fields that are independent
1047          * of the nvme_cmnd request_buffer
1048          */
1049
1050         /* Word 6 */
1051         bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1052                phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1053         bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1054
1055         /* Word 7 */
1056         /* Preserve Class data in the ndlp. */
1057         bf_set(wqe_class, &wqe->generic.wqe_com,
1058                (pnode->nlp_fcp_info & 0x0f));
1059
1060         /* Word 8 */
1061         wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1062
1063         /* Word 9 */
1064         bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1065
1066         /* Word 11 */
1067         bf_set(wqe_cqid, &wqe->generic.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
1068
1069         pwqeq->vport = vport;
1070         return 0;
1071 }
1072
1073
1074 /**
1075  * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1076  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1077  * @lpfc_nvme_lport: Pointer to the driver's local port data
1078  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1079  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1080  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1081  *
1082  * Driver registers this routine as it io request handler.  This
1083  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1084  * data structure to the rport indicated in @lpfc_nvme_rport.
1085  *
1086  * Return value :
1087  *   0 - Success
1088  *   TODO: What are the failure codes.
1089  **/
1090 static int
1091 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1092                       struct lpfc_nvme_buf *lpfc_ncmd)
1093 {
1094         struct lpfc_hba *phba = vport->phba;
1095         struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1096         union lpfc_wqe128 *wqe = (union lpfc_wqe128 *)&lpfc_ncmd->cur_iocbq.wqe;
1097         struct sli4_sge *sgl = lpfc_ncmd->nvme_sgl;
1098         struct scatterlist *data_sg;
1099         struct sli4_sge *first_data_sgl;
1100         dma_addr_t physaddr;
1101         uint32_t num_bde = 0;
1102         uint32_t dma_len;
1103         uint32_t dma_offset = 0;
1104         int nseg, i;
1105
1106         /* Fix up the command and response DMA stuff. */
1107         lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1108
1109         /*
1110          * There are three possibilities here - use scatter-gather segment, use
1111          * the single mapping, or neither.
1112          */
1113         if (nCmd->sg_cnt) {
1114                 /*
1115                  * Jump over the cmd and rsp SGEs.  The fix routine
1116                  * has already adjusted for this.
1117                  */
1118                 sgl += 2;
1119
1120                 first_data_sgl = sgl;
1121                 lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1122                 if (lpfc_ncmd->seg_cnt > phba->cfg_nvme_seg_cnt) {
1123                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1124                                         "6058 Too many sg segments from "
1125                                         "NVME Transport.  Max %d, "
1126                                         "nvmeIO sg_cnt %d\n",
1127                                         phba->cfg_nvme_seg_cnt,
1128                                         lpfc_ncmd->seg_cnt);
1129                         lpfc_ncmd->seg_cnt = 0;
1130                         return 1;
1131                 }
1132
1133                 /*
1134                  * The driver established a maximum scatter-gather segment count
1135                  * during probe that limits the number of sg elements in any
1136                  * single nvme command.  Just run through the seg_cnt and format
1137                  * the sge's.
1138                  */
1139                 nseg = nCmd->sg_cnt;
1140                 data_sg = nCmd->first_sgl;
1141                 for (i = 0; i < nseg; i++) {
1142                         if (data_sg == NULL) {
1143                                 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1144                                                 "6059 dptr err %d, nseg %d\n",
1145                                                 i, nseg);
1146                                 lpfc_ncmd->seg_cnt = 0;
1147                                 return 1;
1148                         }
1149                         physaddr = data_sg->dma_address;
1150                         dma_len = data_sg->length;
1151                         sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr));
1152                         sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr));
1153                         sgl->word2 = le32_to_cpu(sgl->word2);
1154                         if ((num_bde + 1) == nseg)
1155                                 bf_set(lpfc_sli4_sge_last, sgl, 1);
1156                         else
1157                                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1158                         bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1159                         bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA);
1160                         sgl->word2 = cpu_to_le32(sgl->word2);
1161                         sgl->sge_len = cpu_to_le32(dma_len);
1162
1163                         dma_offset += dma_len;
1164                         data_sg = sg_next(data_sg);
1165                         sgl++;
1166                 }
1167         } else {
1168                 /* For this clause to be valid, the payload_length
1169                  * and sg_cnt must zero.
1170                  */
1171                 if (nCmd->payload_length != 0) {
1172                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1173                                         "6063 NVME DMA Prep Err: sg_cnt %d "
1174                                         "payload_length x%x\n",
1175                                         nCmd->sg_cnt, nCmd->payload_length);
1176                         return 1;
1177                 }
1178         }
1179
1180         /*
1181          * Due to difference in data length between DIF/non-DIF paths,
1182          * we need to set word 4 of WQE here
1183          */
1184         wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1185         return 0;
1186 }
1187
1188 /**
1189  * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1190  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1191  * @lpfc_nvme_lport: Pointer to the driver's local port data
1192  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1193  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1194  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1195  *
1196  * Driver registers this routine as it io request handler.  This
1197  * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1198  * data structure to the rport
1199  indicated in @lpfc_nvme_rport.
1200  *
1201  * Return value :
1202  *   0 - Success
1203  *   TODO: What are the failure codes.
1204  **/
1205 static int
1206 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1207                         struct nvme_fc_remote_port *pnvme_rport,
1208                         void *hw_queue_handle,
1209                         struct nvmefc_fcp_req *pnvme_fcreq)
1210 {
1211         int ret = 0;
1212         struct lpfc_nvme_lport *lport;
1213         struct lpfc_vport *vport;
1214         struct lpfc_hba *phba;
1215         struct lpfc_nodelist *ndlp;
1216         struct lpfc_nvme_buf *lpfc_ncmd;
1217         struct lpfc_nvme_rport *rport;
1218         struct lpfc_nvme_qhandle *lpfc_queue_info;
1219         struct lpfc_nvme_fcpreq_priv *freqpriv = pnvme_fcreq->private;
1220 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1221         uint64_t start = 0;
1222 #endif
1223
1224         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1225         vport = lport->vport;
1226         phba = vport->phba;
1227
1228 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1229         if (phba->ktime_on)
1230                 start = ktime_get_ns();
1231 #endif
1232         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1233         lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1234
1235         /*
1236          * Catch race where our node has transitioned, but the
1237          * transport is still transitioning.
1238          */
1239         ndlp = rport->ndlp;
1240         if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1241                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
1242                                  "6053 rport %p, ndlp %p, DID x%06x "
1243                                  "ndlp not ready.\n",
1244                                  rport, ndlp, pnvme_rport->port_id);
1245
1246                 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
1247                 if (!ndlp) {
1248                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1249                                          "6066 Missing node for DID %x\n",
1250                                          pnvme_rport->port_id);
1251                         ret = -ENODEV;
1252                         goto out_fail;
1253                 }
1254         }
1255
1256         /* The remote node has to be a mapped target or it's an error. */
1257         if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1258             (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1259                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
1260                                  "6036 rport %p, DID x%06x not ready for "
1261                                  "IO. State x%x, Type x%x\n",
1262                                  rport, pnvme_rport->port_id,
1263                                  ndlp->nlp_state, ndlp->nlp_type);
1264                 ret = -ENODEV;
1265                 goto out_fail;
1266
1267         }
1268
1269         /* The node is shared with FCP IO, make sure the IO pending count does
1270          * not exceed the programmed depth.
1271          */
1272         if (atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) {
1273                 ret = -EBUSY;
1274                 goto out_fail;
1275         }
1276
1277         lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp);
1278         if (lpfc_ncmd == NULL) {
1279                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1280                                  "6065 driver's buffer pool is empty, "
1281                                  "IO failed\n");
1282                 ret = -EBUSY;
1283                 goto out_fail;
1284         }
1285 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1286         if (phba->ktime_on) {
1287                 lpfc_ncmd->ts_cmd_start = start;
1288                 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1289         }
1290 #endif
1291
1292         /*
1293          * Store the data needed by the driver to issue, abort, and complete
1294          * an IO.
1295          * Do not let the IO hang out forever.  There is no midlayer issuing
1296          * an abort so inform the FW of the maximum IO pending time.
1297          */
1298         freqpriv->nvme_buf = lpfc_ncmd;
1299         lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1300         lpfc_ncmd->nrport = rport;
1301         lpfc_ncmd->ndlp = ndlp;
1302         lpfc_ncmd->start_time = jiffies;
1303
1304         lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp);
1305         ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1306         if (ret) {
1307                 ret = -ENOMEM;
1308                 goto out_free_nvme_buf;
1309         }
1310
1311         atomic_inc(&ndlp->cmd_pending);
1312
1313         /*
1314          * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1315          * This identfier was create in our hardware queue create callback
1316          * routine. The driver now is dependent on the IO queue steering from
1317          * the transport.  We are trusting the upper NVME layers know which
1318          * index to use and that they have affinitized a CPU to this hardware
1319          * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1320          */
1321         lpfc_ncmd->cur_iocbq.hba_wqidx = lpfc_queue_info->index;
1322
1323         lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1324                          lpfc_ncmd->cur_iocbq.sli4_xritag,
1325                          lpfc_queue_info->index, ndlp->nlp_DID);
1326
1327         ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq);
1328         if (ret) {
1329                 atomic_dec(&ndlp->cmd_pending);
1330                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1331                                  "6113 FCP could not issue WQE err %x "
1332                                  "sid: x%x did: x%x oxid: x%x\n",
1333                                  ret, vport->fc_myDID, ndlp->nlp_DID,
1334                                  lpfc_ncmd->cur_iocbq.sli4_xritag);
1335                 goto out_free_nvme_buf;
1336         }
1337
1338 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1339         if (phba->ktime_on)
1340                 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1341
1342         if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1343                 lpfc_ncmd->cpu = smp_processor_id();
1344                 if (lpfc_ncmd->cpu != lpfc_queue_info->index) {
1345                         /* Check for admin queue */
1346                         if (lpfc_queue_info->qidx) {
1347                                 lpfc_printf_vlog(vport,
1348                                                  KERN_ERR, LOG_NVME_IOERR,
1349                                                 "6702 CPU Check cmd: "
1350                                                 "cpu %d wq %d\n",
1351                                                 lpfc_ncmd->cpu,
1352                                                 lpfc_queue_info->index);
1353                         }
1354                         lpfc_ncmd->cpu = lpfc_queue_info->index;
1355                 }
1356                 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT)
1357                         phba->cpucheck_xmt_io[lpfc_ncmd->cpu]++;
1358         }
1359 #endif
1360         return 0;
1361
1362  out_free_nvme_buf:
1363         if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1364                 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1365                         atomic_dec(&phba->fc4NvmeOutputRequests);
1366                 else
1367                         atomic_dec(&phba->fc4NvmeInputRequests);
1368         } else
1369                 atomic_dec(&phba->fc4NvmeControlRequests);
1370         lpfc_release_nvme_buf(phba, lpfc_ncmd);
1371  out_fail:
1372         return ret;
1373 }
1374
1375 /**
1376  * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1377  * @phba: Pointer to HBA context object
1378  * @cmdiocb: Pointer to command iocb object.
1379  * @rspiocb: Pointer to response iocb object.
1380  *
1381  * This is the callback function for any NVME FCP IO that was aborted.
1382  *
1383  * Return value:
1384  *   None
1385  **/
1386 void
1387 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1388                            struct lpfc_wcqe_complete *abts_cmpl)
1389 {
1390         lpfc_printf_log(phba, KERN_ERR, LOG_NVME,
1391                         "6145 ABORT_XRI_CN completing on rpi x%x "
1392                         "original iotag x%x, abort cmd iotag x%x "
1393                         "req_tag x%x, status x%x, hwstatus x%x\n",
1394                         cmdiocb->iocb.un.acxri.abortContextTag,
1395                         cmdiocb->iocb.un.acxri.abortIoTag,
1396                         cmdiocb->iotag,
1397                         bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1398                         bf_get(lpfc_wcqe_c_status, abts_cmpl),
1399                         bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1400         lpfc_sli_release_iocbq(phba, cmdiocb);
1401 }
1402
1403 /**
1404  * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1405  * @lpfc_pnvme: Pointer to the driver's nvme instance data
1406  * @lpfc_nvme_lport: Pointer to the driver's local port data
1407  * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1408  * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1409  * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1410  *
1411  * Driver registers this routine as its nvme request io abort handler.  This
1412  * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1413  * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
1414  * is executed asynchronously - one the target is validated as "MAPPED" and
1415  * ready for IO, the driver issues the abort request and returns.
1416  *
1417  * Return value:
1418  *   None
1419  **/
1420 static void
1421 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1422                     struct nvme_fc_remote_port *pnvme_rport,
1423                     void *hw_queue_handle,
1424                     struct nvmefc_fcp_req *pnvme_fcreq)
1425 {
1426         struct lpfc_nvme_lport *lport;
1427         struct lpfc_vport *vport;
1428         struct lpfc_hba *phba;
1429         struct lpfc_nvme_rport *rport;
1430         struct lpfc_nvme_buf *lpfc_nbuf;
1431         struct lpfc_iocbq *abts_buf;
1432         struct lpfc_iocbq *nvmereq_wqe;
1433         struct lpfc_nvme_fcpreq_priv *freqpriv = pnvme_fcreq->private;
1434         union lpfc_wqe *abts_wqe;
1435         unsigned long flags;
1436         int ret_val;
1437
1438         lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1439         rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1440         vport = lport->vport;
1441         phba = vport->phba;
1442
1443         /* Announce entry to new IO submit field. */
1444         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1445                          "6002 Abort Request to rport DID x%06x "
1446                          "for nvme_fc_req %p\n",
1447                          pnvme_rport->port_id,
1448                          pnvme_fcreq);
1449
1450         /* If the hba is getting reset, this flag is set.  It is
1451          * cleared when the reset is complete and rings reestablished.
1452          */
1453         spin_lock_irqsave(&phba->hbalock, flags);
1454         /* driver queued commands are in process of being flushed */
1455         if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) {
1456                 spin_unlock_irqrestore(&phba->hbalock, flags);
1457                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1458                                  "6139 Driver in reset cleanup - flushing "
1459                                  "NVME Req now.  hba_flag x%x\n",
1460                                  phba->hba_flag);
1461                 return;
1462         }
1463
1464         lpfc_nbuf = freqpriv->nvme_buf;
1465         if (!lpfc_nbuf) {
1466                 spin_unlock_irqrestore(&phba->hbalock, flags);
1467                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1468                                  "6140 NVME IO req has no matching lpfc nvme "
1469                                  "io buffer.  Skipping abort req.\n");
1470                 return;
1471         } else if (!lpfc_nbuf->nvmeCmd) {
1472                 spin_unlock_irqrestore(&phba->hbalock, flags);
1473                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1474                                  "6141 lpfc NVME IO req has no nvme_fcreq "
1475                                  "io buffer.  Skipping abort req.\n");
1476                 return;
1477         }
1478         nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1479
1480         /*
1481          * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1482          * state must match the nvme_fcreq passed by the nvme
1483          * transport.  If they don't match, it is likely the driver
1484          * has already completed the NVME IO and the nvme transport
1485          * has not seen it yet.
1486          */
1487         if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1488                 spin_unlock_irqrestore(&phba->hbalock, flags);
1489                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1490                                  "6143 NVME req mismatch: "
1491                                  "lpfc_nbuf %p nvmeCmd %p, "
1492                                  "pnvme_fcreq %p.  Skipping Abort xri x%x\n",
1493                                  lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1494                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1495                 return;
1496         }
1497
1498         /* Don't abort IOs no longer on the pending queue. */
1499         if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) {
1500                 spin_unlock_irqrestore(&phba->hbalock, flags);
1501                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1502                                  "6142 NVME IO req %p not queued - skipping "
1503                                  "abort req xri x%x\n",
1504                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1505                 return;
1506         }
1507
1508         lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1509                          nvmereq_wqe->sli4_xritag,
1510                          nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1511
1512         /* Outstanding abort is in progress */
1513         if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) {
1514                 spin_unlock_irqrestore(&phba->hbalock, flags);
1515                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1516                                  "6144 Outstanding NVME I/O Abort Request "
1517                                  "still pending on nvme_fcreq %p, "
1518                                  "lpfc_ncmd %p xri x%x\n",
1519                                  pnvme_fcreq, lpfc_nbuf,
1520                                  nvmereq_wqe->sli4_xritag);
1521                 return;
1522         }
1523
1524         abts_buf = __lpfc_sli_get_iocbq(phba);
1525         if (!abts_buf) {
1526                 spin_unlock_irqrestore(&phba->hbalock, flags);
1527                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1528                                  "6136 No available abort wqes. Skipping "
1529                                  "Abts req for nvme_fcreq %p xri x%x\n",
1530                                  pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1531                 return;
1532         }
1533
1534         /* Ready - mark outstanding as aborted by driver. */
1535         nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
1536
1537         /* Complete prepping the abort wqe and issue to the FW. */
1538         abts_wqe = &abts_buf->wqe;
1539
1540         /* WQEs are reused.  Clear stale data and set key fields to
1541          * zero like ia, iaab, iaar, xri_tag, and ctxt_tag.
1542          */
1543         memset(abts_wqe, 0, sizeof(union lpfc_wqe));
1544         bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG);
1545
1546         /* word 7 */
1547         bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
1548         bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com,
1549                nvmereq_wqe->iocb.ulpClass);
1550
1551         /* word 8 - tell the FW to abort the IO associated with this
1552          * outstanding exchange ID.
1553          */
1554         abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag;
1555
1556         /* word 9 - this is the iotag for the abts_wqe completion. */
1557         bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com,
1558                abts_buf->iotag);
1559
1560         /* word 10 */
1561         bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1);
1562         bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE);
1563
1564         /* word 11 */
1565         bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND);
1566         bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1);
1567         bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
1568
1569         /* ABTS WQE must go to the same WQ as the WQE to be aborted */
1570         abts_buf->iocb_flag |= LPFC_IO_NVME;
1571         abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx;
1572         abts_buf->vport = vport;
1573         abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
1574         ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf);
1575         spin_unlock_irqrestore(&phba->hbalock, flags);
1576         if (ret_val) {
1577                 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1578                                  "6137 Failed abts issue_wqe with status x%x "
1579                                  "for nvme_fcreq %p.\n",
1580                                  ret_val, pnvme_fcreq);
1581                 lpfc_sli_release_iocbq(phba, abts_buf);
1582                 return;
1583         }
1584
1585         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1586                          "6138 Transport Abort NVME Request Issued for "
1587                          "ox_id x%x on reqtag x%x\n",
1588                          nvmereq_wqe->sli4_xritag,
1589                          abts_buf->iotag);
1590 }
1591
1592 /* Declare and initialization an instance of the FC NVME template. */
1593 static struct nvme_fc_port_template lpfc_nvme_template = {
1594         /* initiator-based functions */
1595         .localport_delete  = lpfc_nvme_localport_delete,
1596         .remoteport_delete = lpfc_nvme_remoteport_delete,
1597         .create_queue = lpfc_nvme_create_queue,
1598         .delete_queue = lpfc_nvme_delete_queue,
1599         .ls_req       = lpfc_nvme_ls_req,
1600         .fcp_io       = lpfc_nvme_fcp_io_submit,
1601         .ls_abort     = lpfc_nvme_ls_abort,
1602         .fcp_abort    = lpfc_nvme_fcp_abort,
1603
1604         .max_hw_queues = 1,
1605         .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1606         .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1607         .dma_boundary = 0xFFFFFFFF,
1608
1609         /* Sizes of additional private data for data structures.
1610          * No use for the last two sizes at this time.
1611          */
1612         .local_priv_sz = sizeof(struct lpfc_nvme_lport),
1613         .remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1614         .lsrqst_priv_sz = 0,
1615         .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1616 };
1617
1618 /**
1619  * lpfc_sli4_post_nvme_sgl_block - post a block of nvme sgl list to firmware
1620  * @phba: pointer to lpfc hba data structure.
1621  * @nblist: pointer to nvme buffer list.
1622  * @count: number of scsi buffers on the list.
1623  *
1624  * This routine is invoked to post a block of @count scsi sgl pages from a
1625  * SCSI buffer list @nblist to the HBA using non-embedded mailbox command.
1626  * No Lock is held.
1627  *
1628  **/
1629 static int
1630 lpfc_sli4_post_nvme_sgl_block(struct lpfc_hba *phba,
1631                               struct list_head *nblist,
1632                               int count)
1633 {
1634         struct lpfc_nvme_buf *lpfc_ncmd;
1635         struct lpfc_mbx_post_uembed_sgl_page1 *sgl;
1636         struct sgl_page_pairs *sgl_pg_pairs;
1637         void *viraddr;
1638         LPFC_MBOXQ_t *mbox;
1639         uint32_t reqlen, alloclen, pg_pairs;
1640         uint32_t mbox_tmo;
1641         uint16_t xritag_start = 0;
1642         int rc = 0;
1643         uint32_t shdr_status, shdr_add_status;
1644         dma_addr_t pdma_phys_bpl1;
1645         union lpfc_sli4_cfg_shdr *shdr;
1646
1647         /* Calculate the requested length of the dma memory */
1648         reqlen = count * sizeof(struct sgl_page_pairs) +
1649                  sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t);
1650         if (reqlen > SLI4_PAGE_SIZE) {
1651                 lpfc_printf_log(phba, KERN_WARNING, LOG_INIT,
1652                                 "6118 Block sgl registration required DMA "
1653                                 "size (%d) great than a page\n", reqlen);
1654                 return -ENOMEM;
1655         }
1656         mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
1657         if (!mbox) {
1658                 lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
1659                                 "6119 Failed to allocate mbox cmd memory\n");
1660                 return -ENOMEM;
1661         }
1662
1663         /* Allocate DMA memory and set up the non-embedded mailbox command */
1664         alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
1665                                 LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen,
1666                                 LPFC_SLI4_MBX_NEMBED);
1667
1668         if (alloclen < reqlen) {
1669                 lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
1670                                 "6120 Allocated DMA memory size (%d) is "
1671                                 "less than the requested DMA memory "
1672                                 "size (%d)\n", alloclen, reqlen);
1673                 lpfc_sli4_mbox_cmd_free(phba, mbox);
1674                 return -ENOMEM;
1675         }
1676
1677         /* Get the first SGE entry from the non-embedded DMA memory */
1678         viraddr = mbox->sge_array->addr[0];
1679
1680         /* Set up the SGL pages in the non-embedded DMA pages */
1681         sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr;
1682         sgl_pg_pairs = &sgl->sgl_pg_pairs;
1683
1684         pg_pairs = 0;
1685         list_for_each_entry(lpfc_ncmd, nblist, list) {
1686                 /* Set up the sge entry */
1687                 sgl_pg_pairs->sgl_pg0_addr_lo =
1688                         cpu_to_le32(putPaddrLow(lpfc_ncmd->dma_phys_sgl));
1689                 sgl_pg_pairs->sgl_pg0_addr_hi =
1690                         cpu_to_le32(putPaddrHigh(lpfc_ncmd->dma_phys_sgl));
1691                 if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
1692                         pdma_phys_bpl1 = lpfc_ncmd->dma_phys_sgl +
1693                                                 SGL_PAGE_SIZE;
1694                 else
1695                         pdma_phys_bpl1 = 0;
1696                 sgl_pg_pairs->sgl_pg1_addr_lo =
1697                         cpu_to_le32(putPaddrLow(pdma_phys_bpl1));
1698                 sgl_pg_pairs->sgl_pg1_addr_hi =
1699                         cpu_to_le32(putPaddrHigh(pdma_phys_bpl1));
1700                 /* Keep the first xritag on the list */
1701                 if (pg_pairs == 0)
1702                         xritag_start = lpfc_ncmd->cur_iocbq.sli4_xritag;
1703                 sgl_pg_pairs++;
1704                 pg_pairs++;
1705         }
1706         bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start);
1707         bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs);
1708         /* Perform endian conversion if necessary */
1709         sgl->word0 = cpu_to_le32(sgl->word0);
1710
1711         if (!phba->sli4_hba.intr_enable)
1712                 rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
1713         else {
1714                 mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
1715                 rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
1716         }
1717         shdr = (union lpfc_sli4_cfg_shdr *)&sgl->cfg_shdr;
1718         shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
1719         shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
1720         if (rc != MBX_TIMEOUT)
1721                 lpfc_sli4_mbox_cmd_free(phba, mbox);
1722         if (shdr_status || shdr_add_status || rc) {
1723                 lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
1724                                 "6125 POST_SGL_BLOCK mailbox command failed "
1725                                 "status x%x add_status x%x mbx status x%x\n",
1726                                 shdr_status, shdr_add_status, rc);
1727                 rc = -ENXIO;
1728         }
1729         return rc;
1730 }
1731
1732 /**
1733  * lpfc_post_nvme_sgl_list - Post blocks of nvme buffer sgls from a list
1734  * @phba: pointer to lpfc hba data structure.
1735  * @post_nblist: pointer to the nvme buffer list.
1736  *
1737  * This routine walks a list of nvme buffers that was passed in. It attempts
1738  * to construct blocks of nvme buffer sgls which contains contiguous xris and
1739  * uses the non-embedded SGL block post mailbox commands to post to the port.
1740  * For single NVME buffer sgl with non-contiguous xri, if any, it shall use
1741  * embedded SGL post mailbox command for posting. The @post_nblist passed in
1742  * must be local list, thus no lock is needed when manipulate the list.
1743  *
1744  * Returns: 0 = failure, non-zero number of successfully posted buffers.
1745  **/
1746 static int
1747 lpfc_post_nvme_sgl_list(struct lpfc_hba *phba,
1748                              struct list_head *post_nblist, int sb_count)
1749 {
1750         struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
1751         int status, sgl_size;
1752         int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0;
1753         dma_addr_t pdma_phys_sgl1;
1754         int last_xritag = NO_XRI;
1755         int cur_xritag;
1756         LIST_HEAD(prep_nblist);
1757         LIST_HEAD(blck_nblist);
1758         LIST_HEAD(nvme_nblist);
1759
1760         /* sanity check */
1761         if (sb_count <= 0)
1762                 return -EINVAL;
1763
1764         sgl_size = phba->cfg_sg_dma_buf_size;
1765
1766         list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, post_nblist, list) {
1767                 list_del_init(&lpfc_ncmd->list);
1768                 block_cnt++;
1769                 if ((last_xritag != NO_XRI) &&
1770                     (lpfc_ncmd->cur_iocbq.sli4_xritag != last_xritag + 1)) {
1771                         /* a hole in xri block, form a sgl posting block */
1772                         list_splice_init(&prep_nblist, &blck_nblist);
1773                         post_cnt = block_cnt - 1;
1774                         /* prepare list for next posting block */
1775                         list_add_tail(&lpfc_ncmd->list, &prep_nblist);
1776                         block_cnt = 1;
1777                 } else {
1778                         /* prepare list for next posting block */
1779                         list_add_tail(&lpfc_ncmd->list, &prep_nblist);
1780                         /* enough sgls for non-embed sgl mbox command */
1781                         if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) {
1782                                 list_splice_init(&prep_nblist, &blck_nblist);
1783                                 post_cnt = block_cnt;
1784                                 block_cnt = 0;
1785                         }
1786                 }
1787                 num_posting++;
1788                 last_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag;
1789
1790                 /* end of repost sgl list condition for NVME buffers */
1791                 if (num_posting == sb_count) {
1792                         if (post_cnt == 0) {
1793                                 /* last sgl posting block */
1794                                 list_splice_init(&prep_nblist, &blck_nblist);
1795                                 post_cnt = block_cnt;
1796                         } else if (block_cnt == 1) {
1797                                 /* last single sgl with non-contiguous xri */
1798                                 if (sgl_size > SGL_PAGE_SIZE)
1799                                         pdma_phys_sgl1 =
1800                                                 lpfc_ncmd->dma_phys_sgl +
1801                                                 SGL_PAGE_SIZE;
1802                                 else
1803                                         pdma_phys_sgl1 = 0;
1804                                 cur_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag;
1805                                 status = lpfc_sli4_post_sgl(phba,
1806                                                 lpfc_ncmd->dma_phys_sgl,
1807                                                 pdma_phys_sgl1, cur_xritag);
1808                                 if (status) {
1809                                         /* failure, put on abort nvme list */
1810                                         lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1811                                 } else {
1812                                         /* success, put on NVME buffer list */
1813                                         lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1814                                         lpfc_ncmd->status = IOSTAT_SUCCESS;
1815                                         num_posted++;
1816                                 }
1817                                 /* success, put on NVME buffer sgl list */
1818                                 list_add_tail(&lpfc_ncmd->list, &nvme_nblist);
1819                         }
1820                 }
1821
1822                 /* continue until a nembed page worth of sgls */
1823                 if (post_cnt == 0)
1824                         continue;
1825
1826                 /* post block of NVME buffer list sgls */
1827                 status = lpfc_sli4_post_nvme_sgl_block(phba, &blck_nblist,
1828                                                        post_cnt);
1829
1830                 /* don't reset xirtag due to hole in xri block */
1831                 if (block_cnt == 0)
1832                         last_xritag = NO_XRI;
1833
1834                 /* reset NVME buffer post count for next round of posting */
1835                 post_cnt = 0;
1836
1837                 /* put posted NVME buffer-sgl posted on NVME buffer sgl list */
1838                 while (!list_empty(&blck_nblist)) {
1839                         list_remove_head(&blck_nblist, lpfc_ncmd,
1840                                          struct lpfc_nvme_buf, list);
1841                         if (status) {
1842                                 /* failure, put on abort nvme list */
1843                                 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1844                         } else {
1845                                 /* success, put on NVME buffer list */
1846                                 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1847                                 lpfc_ncmd->status = IOSTAT_SUCCESS;
1848                                 num_posted++;
1849                         }
1850                         list_add_tail(&lpfc_ncmd->list, &nvme_nblist);
1851                 }
1852         }
1853         /* Push NVME buffers with sgl posted to the available list */
1854         while (!list_empty(&nvme_nblist)) {
1855                 list_remove_head(&nvme_nblist, lpfc_ncmd,
1856                                  struct lpfc_nvme_buf, list);
1857                 lpfc_release_nvme_buf(phba, lpfc_ncmd);
1858         }
1859         return num_posted;
1860 }
1861
1862 /**
1863  * lpfc_repost_nvme_sgl_list - Repost all the allocated nvme buffer sgls
1864  * @phba: pointer to lpfc hba data structure.
1865  *
1866  * This routine walks the list of nvme buffers that have been allocated and
1867  * repost them to the port by using SGL block post. This is needed after a
1868  * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine
1869  * is responsible for moving all nvme buffers on the lpfc_abts_nvme_sgl_list
1870  * to the lpfc_nvme_buf_list. If the repost fails, reject all nvme buffers.
1871  *
1872  * Returns: 0 = success, non-zero failure.
1873  **/
1874 int
1875 lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba)
1876 {
1877         LIST_HEAD(post_nblist);
1878         int num_posted, rc = 0;
1879
1880         /* get all NVME buffers need to repost to a local list */
1881         spin_lock_irq(&phba->nvme_buf_list_get_lock);
1882         spin_lock(&phba->nvme_buf_list_put_lock);
1883         list_splice_init(&phba->lpfc_nvme_buf_list_get, &post_nblist);
1884         list_splice(&phba->lpfc_nvme_buf_list_put, &post_nblist);
1885         spin_unlock(&phba->nvme_buf_list_put_lock);
1886         spin_unlock_irq(&phba->nvme_buf_list_get_lock);
1887
1888         /* post the list of nvme buffer sgls to port if available */
1889         if (!list_empty(&post_nblist)) {
1890                 num_posted = lpfc_post_nvme_sgl_list(phba, &post_nblist,
1891                                                 phba->sli4_hba.nvme_xri_cnt);
1892                 /* failed to post any nvme buffer, return error */
1893                 if (num_posted == 0)
1894                         rc = -EIO;
1895         }
1896         return rc;
1897 }
1898
1899 /**
1900  * lpfc_new_nvme_buf - Scsi buffer allocator for HBA with SLI4 IF spec
1901  * @vport: The virtual port for which this call being executed.
1902  * @num_to_allocate: The requested number of buffers to allocate.
1903  *
1904  * This routine allocates nvme buffers for device with SLI-4 interface spec,
1905  * the nvme buffer contains all the necessary information needed to initiate
1906  * a NVME I/O. After allocating up to @num_to_allocate NVME buffers and put
1907  * them on a list, it post them to the port by using SGL block post.
1908  *
1909  * Return codes:
1910  *   int - number of nvme buffers that were allocated and posted.
1911  *   0 = failure, less than num_to_alloc is a partial failure.
1912  **/
1913 static int
1914 lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc)
1915 {
1916         struct lpfc_hba *phba = vport->phba;
1917         struct lpfc_nvme_buf *lpfc_ncmd;
1918         struct lpfc_iocbq *pwqeq;
1919         union lpfc_wqe128 *wqe;
1920         struct sli4_sge *sgl;
1921         dma_addr_t pdma_phys_sgl;
1922         uint16_t iotag, lxri = 0;
1923         int bcnt, num_posted, sgl_size;
1924         LIST_HEAD(prep_nblist);
1925         LIST_HEAD(post_nblist);
1926         LIST_HEAD(nvme_nblist);
1927
1928         sgl_size = phba->cfg_sg_dma_buf_size;
1929
1930         for (bcnt = 0; bcnt < num_to_alloc; bcnt++) {
1931                 lpfc_ncmd = kzalloc(sizeof(struct lpfc_nvme_buf), GFP_KERNEL);
1932                 if (!lpfc_ncmd)
1933                         break;
1934                 /*
1935                  * Get memory from the pci pool to map the virt space to
1936                  * pci bus space for an I/O. The DMA buffer includes the
1937                  * number of SGE's necessary to support the sg_tablesize.
1938                  */
1939                 lpfc_ncmd->data = dma_pool_alloc(phba->lpfc_sg_dma_buf_pool,
1940                                                  GFP_KERNEL,
1941                                                  &lpfc_ncmd->dma_handle);
1942                 if (!lpfc_ncmd->data) {
1943                         kfree(lpfc_ncmd);
1944                         break;
1945                 }
1946                 memset(lpfc_ncmd->data, 0, phba->cfg_sg_dma_buf_size);
1947
1948                 lxri = lpfc_sli4_next_xritag(phba);
1949                 if (lxri == NO_XRI) {
1950                         dma_pool_free(phba->lpfc_sg_dma_buf_pool,
1951                                       lpfc_ncmd->data, lpfc_ncmd->dma_handle);
1952                         kfree(lpfc_ncmd);
1953                         break;
1954                 }
1955                 pwqeq = &(lpfc_ncmd->cur_iocbq);
1956                 wqe = (union lpfc_wqe128 *)&pwqeq->wqe;
1957
1958                 /* Allocate iotag for lpfc_ncmd->cur_iocbq. */
1959                 iotag = lpfc_sli_next_iotag(phba, pwqeq);
1960                 if (iotag == 0) {
1961                         dma_pool_free(phba->lpfc_sg_dma_buf_pool,
1962                                       lpfc_ncmd->data, lpfc_ncmd->dma_handle);
1963                         kfree(lpfc_ncmd);
1964                         lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1965                                         "6121 Failed to allocated IOTAG for"
1966                                         " XRI:0x%x\n", lxri);
1967                         lpfc_sli4_free_xri(phba, lxri);
1968                         break;
1969                 }
1970                 pwqeq->sli4_lxritag = lxri;
1971                 pwqeq->sli4_xritag = phba->sli4_hba.xri_ids[lxri];
1972                 pwqeq->iocb_flag |= LPFC_IO_NVME;
1973                 pwqeq->context1 = lpfc_ncmd;
1974                 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl;
1975
1976                 /* Initialize local short-hand pointers. */
1977                 lpfc_ncmd->nvme_sgl = lpfc_ncmd->data;
1978                 sgl = lpfc_ncmd->nvme_sgl;
1979                 pdma_phys_sgl = lpfc_ncmd->dma_handle;
1980                 lpfc_ncmd->dma_phys_sgl = pdma_phys_sgl;
1981
1982                 /* Rsp SGE will be filled in when we rcv an IO
1983                  * from the NVME Layer to be sent.
1984                  * The cmd is going to be embedded so we need a SKIP SGE.
1985                  */
1986                 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
1987                 bf_set(lpfc_sli4_sge_last, sgl, 0);
1988                 sgl->word2 = cpu_to_le32(sgl->word2);
1989                 /* Fill in word 3 / sgl_len during cmd submission */
1990
1991                 lpfc_ncmd->cur_iocbq.context1 = lpfc_ncmd;
1992
1993                 /* Word 7 */
1994                 bf_set(wqe_erp, &wqe->generic.wqe_com, 0);
1995                 /* NVME upper layers will time things out, if needed */
1996                 bf_set(wqe_tmo, &wqe->generic.wqe_com, 0);
1997
1998                 /* Word 10 */
1999                 bf_set(wqe_ebde_cnt, &wqe->generic.wqe_com, 0);
2000                 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
2001
2002                 /* add the nvme buffer to a post list */
2003                 list_add_tail(&lpfc_ncmd->list, &post_nblist);
2004                 spin_lock_irq(&phba->nvme_buf_list_get_lock);
2005                 phba->sli4_hba.nvme_xri_cnt++;
2006                 spin_unlock_irq(&phba->nvme_buf_list_get_lock);
2007         }
2008         lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
2009                         "6114 Allocate %d out of %d requested new NVME "
2010                         "buffers\n", bcnt, num_to_alloc);
2011
2012         /* post the list of nvme buffer sgls to port if available */
2013         if (!list_empty(&post_nblist))
2014                 num_posted = lpfc_post_nvme_sgl_list(phba,
2015                                                      &post_nblist, bcnt);
2016         else
2017                 num_posted = 0;
2018
2019         return num_posted;
2020 }
2021
2022 /**
2023  * lpfc_get_nvme_buf - Get a nvme buffer from lpfc_nvme_buf_list of the HBA
2024  * @phba: The HBA for which this call is being executed.
2025  *
2026  * This routine removes a nvme buffer from head of @phba lpfc_nvme_buf_list list
2027  * and returns to caller.
2028  *
2029  * Return codes:
2030  *   NULL - Error
2031  *   Pointer to lpfc_nvme_buf - Success
2032  **/
2033 static struct lpfc_nvme_buf *
2034 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
2035 {
2036         struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
2037         unsigned long iflag = 0;
2038         int found = 0;
2039
2040         spin_lock_irqsave(&phba->nvme_buf_list_get_lock, iflag);
2041         list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
2042                                  &phba->lpfc_nvme_buf_list_get, list) {
2043                 if (lpfc_test_rrq_active(phba, ndlp,
2044                                          lpfc_ncmd->cur_iocbq.sli4_lxritag))
2045                         continue;
2046                 list_del_init(&lpfc_ncmd->list);
2047                 found = 1;
2048                 break;
2049         }
2050         if (!found) {
2051                 spin_lock(&phba->nvme_buf_list_put_lock);
2052                 list_splice(&phba->lpfc_nvme_buf_list_put,
2053                             &phba->lpfc_nvme_buf_list_get);
2054                 INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put);
2055                 spin_unlock(&phba->nvme_buf_list_put_lock);
2056                 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
2057                                          &phba->lpfc_nvme_buf_list_get, list) {
2058                         if (lpfc_test_rrq_active(
2059                                 phba, ndlp, lpfc_ncmd->cur_iocbq.sli4_lxritag))
2060                                 continue;
2061                         list_del_init(&lpfc_ncmd->list);
2062                         found = 1;
2063                         break;
2064                 }
2065         }
2066         spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag);
2067         if (!found)
2068                 return NULL;
2069         return  lpfc_ncmd;
2070 }
2071
2072 /**
2073  * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2074  * @phba: The Hba for which this call is being executed.
2075  * @lpfc_ncmd: The nvme buffer which is being released.
2076  *
2077  * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2078  * lpfc_nvme_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2079  * and cannot be reused for at least RA_TOV amount of time if it was
2080  * aborted.
2081  **/
2082 static void
2083 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd)
2084 {
2085         unsigned long iflag = 0;
2086
2087         lpfc_ncmd->nonsg_phys = 0;
2088         if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) {
2089                 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2090                                 "6310 XB release deferred for "
2091                                 "ox_id x%x on reqtag x%x\n",
2092                                 lpfc_ncmd->cur_iocbq.sli4_xritag,
2093                                 lpfc_ncmd->cur_iocbq.iotag);
2094
2095                 spin_lock_irqsave(&phba->sli4_hba.abts_nvme_buf_list_lock,
2096                                         iflag);
2097                 lpfc_ncmd->nvmeCmd = NULL;
2098                 list_add_tail(&lpfc_ncmd->list,
2099                         &phba->sli4_hba.lpfc_abts_nvme_buf_list);
2100                 spin_unlock_irqrestore(&phba->sli4_hba.abts_nvme_buf_list_lock,
2101                                         iflag);
2102         } else {
2103                 lpfc_ncmd->nvmeCmd = NULL;
2104                 lpfc_ncmd->cur_iocbq.iocb_flag = LPFC_IO_NVME;
2105                 spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag);
2106                 list_add_tail(&lpfc_ncmd->list, &phba->lpfc_nvme_buf_list_put);
2107                 spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag);
2108         }
2109 }
2110
2111 /**
2112  * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2113  * @pvport - the lpfc_vport instance requesting a localport.
2114  *
2115  * This routine is invoked to create an nvme localport instance to bind
2116  * to the nvme_fc_transport.  It is called once during driver load
2117  * like lpfc_create_shost after all other services are initialized.
2118  * It requires a vport, vpi, and wwns at call time.  Other localport
2119  * parameters are modified as the driver's FCID and the Fabric WWN
2120  * are established.
2121  *
2122  * Return codes
2123  *      0 - successful
2124  *      -ENOMEM - no heap memory available
2125  *      other values - from nvme registration upcall
2126  **/
2127 int
2128 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2129 {
2130         int ret = 0;
2131         struct lpfc_hba  *phba = vport->phba;
2132         struct nvme_fc_port_info nfcp_info;
2133         struct nvme_fc_local_port *localport;
2134         struct lpfc_nvme_lport *lport;
2135         int len;
2136
2137         /* Initialize this localport instance.  The vport wwn usage ensures
2138          * that NPIV is accounted for.
2139          */
2140         memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2141         nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2142         nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2143         nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2144
2145         /* Limit to LPFC_MAX_NVME_SEG_CNT.
2146          * For now need + 1 to get around NVME transport logic.
2147          */
2148         if (phba->cfg_sg_seg_cnt > LPFC_MAX_NVME_SEG_CNT) {
2149                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_INIT,
2150                                  "6300 Reducing sg segment cnt to %d\n",
2151                                  LPFC_MAX_NVME_SEG_CNT);
2152                 phba->cfg_nvme_seg_cnt = LPFC_MAX_NVME_SEG_CNT;
2153         } else {
2154                 phba->cfg_nvme_seg_cnt = phba->cfg_sg_seg_cnt;
2155         }
2156         lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2157         lpfc_nvme_template.max_hw_queues = phba->cfg_nvme_io_channel;
2158
2159         /* localport is allocated from the stack, but the registration
2160          * call allocates heap memory as well as the private area.
2161          */
2162 #if (IS_ENABLED(CONFIG_NVME_FC))
2163         ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2164                                          &vport->phba->pcidev->dev, &localport);
2165 #else
2166         ret = -ENOMEM;
2167 #endif
2168         if (!ret) {
2169                 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2170                                  "6005 Successfully registered local "
2171                                  "NVME port num %d, localP %p, private %p, "
2172                                  "sg_seg %d\n",
2173                                  localport->port_num, localport,
2174                                  localport->private,
2175                                  lpfc_nvme_template.max_sgl_segments);
2176
2177                 /* Private is our lport size declared in the template. */
2178                 lport = (struct lpfc_nvme_lport *)localport->private;
2179                 vport->localport = localport;
2180                 lport->vport = vport;
2181                 vport->nvmei_support = 1;
2182
2183                 /* Don't post more new bufs if repost already recovered
2184                  * the nvme sgls.
2185                  */
2186                 if (phba->sli4_hba.nvme_xri_cnt == 0) {
2187                         len  = lpfc_new_nvme_buf(vport,
2188                                                  phba->sli4_hba.nvme_xri_max);
2189                         vport->phba->total_nvme_bufs += len;
2190                 }
2191         }
2192
2193         return ret;
2194 }
2195
2196 /**
2197  * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2198  * @pnvme: pointer to lpfc nvme data structure.
2199  *
2200  * This routine is invoked to destroy all lports bound to the phba.
2201  * The lport memory was allocated by the nvme fc transport and is
2202  * released there.  This routine ensures all rports bound to the
2203  * lport have been disconnected.
2204  *
2205  **/
2206 void
2207 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2208 {
2209 #if (IS_ENABLED(CONFIG_NVME_FC))
2210         struct nvme_fc_local_port *localport;
2211         struct lpfc_nvme_lport *lport;
2212         int ret;
2213
2214         if (vport->nvmei_support == 0)
2215                 return;
2216
2217         localport = vport->localport;
2218         vport->localport = NULL;
2219         lport = (struct lpfc_nvme_lport *)localport->private;
2220
2221         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2222                          "6011 Destroying NVME localport %p\n",
2223                          localport);
2224
2225         /* lport's rport list is clear.  Unregister
2226          * lport and release resources.
2227          */
2228         init_completion(&lport->lport_unreg_done);
2229         ret = nvme_fc_unregister_localport(localport);
2230         wait_for_completion_timeout(&lport->lport_unreg_done, 5);
2231
2232         /* Regardless of the unregister upcall response, clear
2233          * nvmei_support.  All rports are unregistered and the
2234          * driver will clean up.
2235          */
2236         vport->nvmei_support = 0;
2237         if (ret == 0) {
2238                 lpfc_printf_vlog(vport,
2239                                  KERN_INFO, LOG_NVME_DISC,
2240                                  "6009 Unregistered lport Success\n");
2241         } else {
2242                 lpfc_printf_vlog(vport,
2243                                  KERN_INFO, LOG_NVME_DISC,
2244                                  "6010 Unregistered lport "
2245                                  "Failed, status x%x\n",
2246                                  ret);
2247         }
2248 #endif
2249 }
2250
2251 void
2252 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2253 {
2254 #if (IS_ENABLED(CONFIG_NVME_FC))
2255         struct nvme_fc_local_port *localport;
2256         struct lpfc_nvme_lport *lport;
2257
2258         localport = vport->localport;
2259         if (!localport) {
2260                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2261                                  "6710 Update NVME fail. No localport\n");
2262                 return;
2263         }
2264         lport = (struct lpfc_nvme_lport *)localport->private;
2265         if (!lport) {
2266                 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2267                                  "6171 Update NVME fail. localP %p, No lport\n",
2268                                  localport);
2269                 return;
2270         }
2271         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2272                          "6012 Update NVME lport %p did x%x\n",
2273                          localport, vport->fc_myDID);
2274
2275         localport->port_id = vport->fc_myDID;
2276         if (localport->port_id == 0)
2277                 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2278         else
2279                 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2280
2281         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2282                          "6030 bound lport %p to DID x%06x\n",
2283                          lport, localport->port_id);
2284 #endif
2285 }
2286
2287 int
2288 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2289 {
2290 #if (IS_ENABLED(CONFIG_NVME_FC))
2291         int ret = 0;
2292         struct nvme_fc_local_port *localport;
2293         struct lpfc_nvme_lport *lport;
2294         struct lpfc_nvme_rport *rport;
2295         struct nvme_fc_remote_port *remote_port;
2296         struct nvme_fc_port_info rpinfo;
2297
2298         lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2299                          "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2300                          ndlp->nlp_DID, ndlp->nlp_type);
2301
2302         localport = vport->localport;
2303         if (!localport)
2304                 return 0;
2305
2306         lport = (struct lpfc_nvme_lport *)localport->private;
2307
2308         /* NVME rports are not preserved across devloss.
2309          * Just register this instance.  Note, rpinfo->dev_loss_tmo
2310          * is left 0 to indicate accept transport defaults.  The
2311          * driver communicates port role capabilities consistent
2312          * with the PRLI response data.
2313          */
2314         memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2315         rpinfo.port_id = ndlp->nlp_DID;
2316         if (ndlp->nlp_type & NLP_NVME_TARGET)
2317                 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2318         if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2319                 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2320
2321         if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2322                 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2323
2324         rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2325         rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2326         ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2327         if (!ret) {
2328                 /* If the ndlp already has an nrport, this is just
2329                  * a resume of the existing rport.  Else this is a
2330                  * new rport.
2331                  */
2332                 rport = remote_port->private;
2333                 if (ndlp->nrport == rport) {
2334                         lpfc_printf_vlog(ndlp->vport, KERN_INFO,
2335                                          LOG_NVME_DISC,
2336                                          "6014 Rebinding lport to "
2337                                          "rport wwpn 0x%llx, "
2338                                          "Data: x%x x%x x%x x%06x\n",
2339                                          remote_port->port_name,
2340                                          remote_port->port_id,
2341                                          remote_port->port_role,
2342                                          ndlp->nlp_type,
2343                                          ndlp->nlp_DID);
2344                 } else {
2345                         /* New rport. */
2346                         rport->remoteport = remote_port;
2347                         rport->lport = lport;
2348                         rport->ndlp = lpfc_nlp_get(ndlp);
2349                         if (!rport->ndlp)
2350                                 return -1;
2351                         ndlp->nrport = rport;
2352                         lpfc_printf_vlog(vport, KERN_INFO,
2353                                          LOG_NVME_DISC | LOG_NODE,
2354                                          "6022 Binding new rport to "
2355                                          "lport %p Rport WWNN 0x%llx, "
2356                                          "Rport WWPN 0x%llx DID "
2357                                          "x%06x Role x%x\n",
2358                                          lport,
2359                                          rpinfo.node_name, rpinfo.port_name,
2360                                          rpinfo.port_id, rpinfo.port_role);
2361                 }
2362         } else {
2363                 lpfc_printf_vlog(vport, KERN_ERR,
2364                                  LOG_NVME_DISC | LOG_NODE,
2365                                  "6031 RemotePort Registration failed "
2366                                  "err: %d, DID x%06x\n",
2367                                  ret, ndlp->nlp_DID);
2368         }
2369
2370         return ret;
2371 #else
2372         return 0;
2373 #endif
2374 }
2375
2376 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2377  *
2378  * There is no notion of Devloss or rport recovery from the current
2379  * nvme_transport perspective.  Loss of an rport just means IO cannot
2380  * be sent and recovery is completely up to the initator.
2381  * For now, the driver just unbinds the DID and port_role so that
2382  * no further IO can be issued.  Changes are planned for later.
2383  *
2384  * Notes - the ndlp reference count is not decremented here since
2385  * since there is no nvme_transport api for devloss.  Node ref count
2386  * is only adjusted in driver unload.
2387  */
2388 void
2389 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2390 {
2391 #if (IS_ENABLED(CONFIG_NVME_FC))
2392         int ret;
2393         struct nvme_fc_local_port *localport;
2394         struct lpfc_nvme_lport *lport;
2395         struct lpfc_nvme_rport *rport;
2396         struct nvme_fc_remote_port *remoteport;
2397
2398         localport = vport->localport;
2399
2400         /* This is fundamental error.  The localport is always
2401          * available until driver unload.  Just exit.
2402          */
2403         if (!localport)
2404                 return;
2405
2406         lport = (struct lpfc_nvme_lport *)localport->private;
2407         if (!lport)
2408                 goto input_err;
2409
2410         rport = ndlp->nrport;
2411         if (!rport)
2412                 goto input_err;
2413
2414         remoteport = rport->remoteport;
2415         lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2416                          "6033 Unreg nvme remoteport %p, portname x%llx, "
2417                          "port_id x%06x, portstate x%x port type x%x\n",
2418                          remoteport, remoteport->port_name,
2419                          remoteport->port_id, remoteport->port_state,
2420                          ndlp->nlp_type);
2421
2422         /* Sanity check ndlp type.  Only call for NVME ports. Don't
2423          * clear any rport state until the transport calls back.
2424          */
2425         if (ndlp->nlp_type & (NLP_NVME_TARGET | NLP_NVME_INITIATOR)) {
2426                 init_completion(&rport->rport_unreg_done);
2427
2428                 /* No concern about the role change on the nvme remoteport.
2429                  * The transport will update it.
2430                  */
2431                 ret = nvme_fc_unregister_remoteport(remoteport);
2432                 if (ret != 0) {
2433                         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2434                                          "6167 NVME unregister failed %d "
2435                                          "port_state x%x\n",
2436                                          ret, remoteport->port_state);
2437                 }
2438
2439         }
2440         return;
2441
2442  input_err:
2443 #endif
2444         lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2445                          "6168 State error: lport %p, rport%p FCID x%06x\n",
2446                          vport->localport, ndlp->rport, ndlp->nlp_DID);
2447 }
2448
2449 /**
2450  * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2451  * @phba: pointer to lpfc hba data structure.
2452  * @axri: pointer to the fcp xri abort wcqe structure.
2453  *
2454  * This routine is invoked by the worker thread to process a SLI4 fast-path
2455  * FCP aborted xri.
2456  **/
2457 void
2458 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2459                            struct sli4_wcqe_xri_aborted *axri)
2460 {
2461         uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2462         uint16_t rxid = bf_get(lpfc_wcqe_xa_remote_xid, axri);
2463         struct lpfc_nvme_buf *lpfc_ncmd, *next_lpfc_ncmd;
2464         struct lpfc_nodelist *ndlp;
2465         unsigned long iflag = 0;
2466         int rrq_empty = 0;
2467
2468         if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME))
2469                 return;
2470         spin_lock_irqsave(&phba->hbalock, iflag);
2471         spin_lock(&phba->sli4_hba.abts_nvme_buf_list_lock);
2472         list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd,
2473                                  &phba->sli4_hba.lpfc_abts_nvme_buf_list,
2474                                  list) {
2475                 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) {
2476                         list_del_init(&lpfc_ncmd->list);
2477                         lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
2478                         lpfc_ncmd->status = IOSTAT_SUCCESS;
2479                         spin_unlock(
2480                                 &phba->sli4_hba.abts_nvme_buf_list_lock);
2481
2482                         rrq_empty = list_empty(&phba->active_rrq_list);
2483                         spin_unlock_irqrestore(&phba->hbalock, iflag);
2484                         ndlp = lpfc_ncmd->ndlp;
2485                         if (ndlp) {
2486                                 lpfc_set_rrq_active(
2487                                         phba, ndlp,
2488                                         lpfc_ncmd->cur_iocbq.sli4_lxritag,
2489                                         rxid, 1);
2490                                 lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2491                         }
2492
2493                         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2494                                         "6311 XRI Aborted xri x%x tag x%x "
2495                                         "released\n",
2496                                         xri, lpfc_ncmd->cur_iocbq.iotag);
2497
2498                         lpfc_release_nvme_buf(phba, lpfc_ncmd);
2499                         if (rrq_empty)
2500                                 lpfc_worker_wake_up(phba);
2501                         return;
2502                 }
2503         }
2504         spin_unlock(&phba->sli4_hba.abts_nvme_buf_list_lock);
2505         spin_unlock_irqrestore(&phba->hbalock, iflag);
2506
2507         lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2508                         "6312 XRI Aborted xri x%x not found\n", xri);
2509
2510 }