?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dmaengine: imx-sdma: Let the core do the device node validation
[linux/fpc-iii.git] / drivers / scsi / lpfc / lpfc_nvme.c
blob9d99cb91539078e1cde16351a9543789df1fcd38
1 /*******************************************************************
2 * This file is part of the Emulex Linux Device Driver for *
3 * Fibre Channel Host Bus Adapters. *
4 * Copyright (C) 2017-2019 Broadcom. All Rights Reserved. The term *
5 * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries. *
6 * Copyright (C) 2004-2016 Emulex. All rights reserved. *
7 * EMULEX and SLI are trademarks of Emulex. *
8 * www.broadcom.com *
9 * Portions Copyright (C) 2004-2005 Christoph Hellwig *
10 * *
11 * This program is free software; you can redistribute it and/or *
12 * modify it under the terms of version 2 of the GNU General *
13 * Public License as published by the Free Software Foundation. *
14 * This program is distributed in the hope that it will be useful. *
15 * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND *
16 * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, *
17 * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE *
18 * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
19 * TO BE LEGALLY INVALID. See the GNU General Public License for *
20 * more details, a copy of which can be found in the file COPYING *
21 * included with this package. *
22 ********************************************************************/
23 #include <linux/pci.h>
24 #include <linux/slab.h>
25 #include <linux/interrupt.h>
26 #include <linux/delay.h>
27 #include <asm/unaligned.h>
28 #include <linux/crc-t10dif.h>
29 #include <net/checksum.h>
30
31 #include <scsi/scsi.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_tcq.h>
36 #include <scsi/scsi_transport_fc.h>
37 #include <scsi/fc/fc_fs.h>
38
39 #include <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_io_buf *
60 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
61 int idx, int expedite);
62
63 static void
64 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_io_buf *);
65
66 static struct nvme_fc_port_template lpfc_nvme_template;
67
68 static union lpfc_wqe128 lpfc_iread_cmd_template;
69 static union lpfc_wqe128 lpfc_iwrite_cmd_template;
70 static union lpfc_wqe128 lpfc_icmnd_cmd_template;
71
72 /* Setup WQE templates for NVME IOs */
73 void
74 lpfc_nvme_cmd_template(void)
75 {
76 union lpfc_wqe128 *wqe;
77
78 /* IREAD template */
79 wqe = &lpfc_iread_cmd_template;
80 memset(wqe, 0, sizeof(union lpfc_wqe128));
81
82 /* Word 0, 1, 2 - BDE is variable */
83
84 /* Word 3 - cmd_buff_len, payload_offset_len is zero */
85
86 /* Word 4 - total_xfer_len is variable */
87
88 /* Word 5 - is zero */
89
90 /* Word 6 - ctxt_tag, xri_tag is variable */
91
92 /* Word 7 */
93 bf_set(wqe_cmnd, &wqe->fcp_iread.wqe_com, CMD_FCP_IREAD64_WQE);
94 bf_set(wqe_pu, &wqe->fcp_iread.wqe_com, PARM_READ_CHECK);
95 bf_set(wqe_class, &wqe->fcp_iread.wqe_com, CLASS3);
96 bf_set(wqe_ct, &wqe->fcp_iread.wqe_com, SLI4_CT_RPI);
97
98 /* Word 8 - abort_tag is variable */
99
100 /* Word 9 - reqtag is variable */
101
102 /* Word 10 - dbde, wqes is variable */
103 bf_set(wqe_qosd, &wqe->fcp_iread.wqe_com, 0);
104 bf_set(wqe_nvme, &wqe->fcp_iread.wqe_com, 1);
105 bf_set(wqe_iod, &wqe->fcp_iread.wqe_com, LPFC_WQE_IOD_READ);
106 bf_set(wqe_lenloc, &wqe->fcp_iread.wqe_com, LPFC_WQE_LENLOC_WORD4);
107 bf_set(wqe_dbde, &wqe->fcp_iread.wqe_com, 0);
108 bf_set(wqe_wqes, &wqe->fcp_iread.wqe_com, 1);
109
110 /* Word 11 - pbde is variable */
111 bf_set(wqe_cmd_type, &wqe->fcp_iread.wqe_com, NVME_READ_CMD);
112 bf_set(wqe_cqid, &wqe->fcp_iread.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
113 bf_set(wqe_pbde, &wqe->fcp_iread.wqe_com, 1);
114
115 /* Word 12 - is zero */
116
117 /* Word 13, 14, 15 - PBDE is variable */
118
119 /* IWRITE template */
120 wqe = &lpfc_iwrite_cmd_template;
121 memset(wqe, 0, sizeof(union lpfc_wqe128));
122
123 /* Word 0, 1, 2 - BDE is variable */
124
125 /* Word 3 - cmd_buff_len, payload_offset_len is zero */
126
127 /* Word 4 - total_xfer_len is variable */
128
129 /* Word 5 - initial_xfer_len is variable */
130
131 /* Word 6 - ctxt_tag, xri_tag is variable */
132
133 /* Word 7 */
134 bf_set(wqe_cmnd, &wqe->fcp_iwrite.wqe_com, CMD_FCP_IWRITE64_WQE);
135 bf_set(wqe_pu, &wqe->fcp_iwrite.wqe_com, PARM_READ_CHECK);
136 bf_set(wqe_class, &wqe->fcp_iwrite.wqe_com, CLASS3);
137 bf_set(wqe_ct, &wqe->fcp_iwrite.wqe_com, SLI4_CT_RPI);
138
139 /* Word 8 - abort_tag is variable */
140
141 /* Word 9 - reqtag is variable */
142
143 /* Word 10 - dbde, wqes is variable */
144 bf_set(wqe_qosd, &wqe->fcp_iwrite.wqe_com, 0);
145 bf_set(wqe_nvme, &wqe->fcp_iwrite.wqe_com, 1);
146 bf_set(wqe_iod, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_IOD_WRITE);
147 bf_set(wqe_lenloc, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_LENLOC_WORD4);
148 bf_set(wqe_dbde, &wqe->fcp_iwrite.wqe_com, 0);
149 bf_set(wqe_wqes, &wqe->fcp_iwrite.wqe_com, 1);
150
151 /* Word 11 - pbde is variable */
152 bf_set(wqe_cmd_type, &wqe->fcp_iwrite.wqe_com, NVME_WRITE_CMD);
153 bf_set(wqe_cqid, &wqe->fcp_iwrite.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
154 bf_set(wqe_pbde, &wqe->fcp_iwrite.wqe_com, 1);
155
156 /* Word 12 - is zero */
157
158 /* Word 13, 14, 15 - PBDE is variable */
159
160 /* ICMND template */
161 wqe = &lpfc_icmnd_cmd_template;
162 memset(wqe, 0, sizeof(union lpfc_wqe128));
163
164 /* Word 0, 1, 2 - BDE is variable */
165
166 /* Word 3 - payload_offset_len is variable */
167
168 /* Word 4, 5 - is zero */
169
170 /* Word 6 - ctxt_tag, xri_tag is variable */
171
172 /* Word 7 */
173 bf_set(wqe_cmnd, &wqe->fcp_icmd.wqe_com, CMD_FCP_ICMND64_WQE);
174 bf_set(wqe_pu, &wqe->fcp_icmd.wqe_com, 0);
175 bf_set(wqe_class, &wqe->fcp_icmd.wqe_com, CLASS3);
176 bf_set(wqe_ct, &wqe->fcp_icmd.wqe_com, SLI4_CT_RPI);
177
178 /* Word 8 - abort_tag is variable */
179
180 /* Word 9 - reqtag is variable */
181
182 /* Word 10 - dbde, wqes is variable */
183 bf_set(wqe_qosd, &wqe->fcp_icmd.wqe_com, 1);
184 bf_set(wqe_nvme, &wqe->fcp_icmd.wqe_com, 1);
185 bf_set(wqe_iod, &wqe->fcp_icmd.wqe_com, LPFC_WQE_IOD_NONE);
186 bf_set(wqe_lenloc, &wqe->fcp_icmd.wqe_com, LPFC_WQE_LENLOC_NONE);
187 bf_set(wqe_dbde, &wqe->fcp_icmd.wqe_com, 0);
188 bf_set(wqe_wqes, &wqe->fcp_icmd.wqe_com, 1);
189
190 /* Word 11 */
191 bf_set(wqe_cmd_type, &wqe->fcp_icmd.wqe_com, FCP_COMMAND);
192 bf_set(wqe_cqid, &wqe->fcp_icmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
193 bf_set(wqe_pbde, &wqe->fcp_icmd.wqe_com, 0);
194
195 /* Word 12, 13, 14, 15 - is zero */
196 }
197
198 /**
199 * lpfc_nvme_create_queue -
200 * @lpfc_pnvme: Pointer to the driver's nvme instance data
201 * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
202 * @handle: An opaque driver handle used in follow-up calls.
203 *
204 * Driver registers this routine to preallocate and initialize any
205 * internal data structures to bind the @qidx to its internal IO queues.
206 * A hardware queue maps (qidx) to a specific driver MSI-X vector/EQ/CQ/WQ.
207 *
208 * Return value :
209 * 0 - Success
210 * -EINVAL - Unsupported input value.
211 * -ENOMEM - Could not alloc necessary memory
212 **/
213 static int
214 lpfc_nvme_create_queue(struct nvme_fc_local_port *pnvme_lport,
215 unsigned int qidx, u16 qsize,
216 void **handle)
217 {
218 struct lpfc_nvme_lport *lport;
219 struct lpfc_vport *vport;
220 struct lpfc_nvme_qhandle *qhandle;
221 char *str;
222
223 if (!pnvme_lport->private)
224 return -ENOMEM;
225
226 lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
227 vport = lport->vport;
228 qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
229 if (qhandle == NULL)
230 return -ENOMEM;
231
232 qhandle->cpu_id = raw_smp_processor_id();
233 qhandle->qidx = qidx;
234 /*
235 * NVME qidx == 0 is the admin queue, so both admin queue
236 * and first IO queue will use MSI-X vector and associated
237 * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
238 */
239 if (qidx) {
240 str = "IO "; /* IO queue */
241 qhandle->index = ((qidx - 1) %
242 lpfc_nvme_template.max_hw_queues);
243 } else {
244 str = "ADM"; /* Admin queue */
245 qhandle->index = qidx;
246 }
247
248 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
249 "6073 Binding %s HdwQueue %d (cpu %d) to "
250 "hdw_queue %d qhandle %p\n", str,
251 qidx, qhandle->cpu_id, qhandle->index, qhandle);
252 *handle = (void *)qhandle;
253 return 0;
254 }
255
256 /**
257 * lpfc_nvme_delete_queue -
258 * @lpfc_pnvme: Pointer to the driver's nvme instance data
259 * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
260 * @handle: An opaque driver handle from lpfc_nvme_create_queue
261 *
262 * Driver registers this routine to free
263 * any internal data structures to bind the @qidx to its internal
264 * IO queues.
265 *
266 * Return value :
267 * 0 - Success
268 * TODO: What are the failure codes.
269 **/
270 static void
271 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
272 unsigned int qidx,
273 void *handle)
274 {
275 struct lpfc_nvme_lport *lport;
276 struct lpfc_vport *vport;
277
278 if (!pnvme_lport->private)
279 return;
280
281 lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
282 vport = lport->vport;
283
284 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
285 "6001 ENTER. lpfc_pnvme %p, qidx x%x qhandle %p\n",
286 lport, qidx, handle);
287 kfree(handle);
288 }
289
290 static void
291 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
292 {
293 struct lpfc_nvme_lport *lport = localport->private;
294
295 lpfc_printf_vlog(lport->vport, KERN_INFO, LOG_NVME,
296 "6173 localport %p delete complete\n",
297 lport);
298
299 /* release any threads waiting for the unreg to complete */
300 if (lport->vport->localport)
301 complete(lport->lport_unreg_cmp);
302 }
303
304 /* lpfc_nvme_remoteport_delete
305 *
306 * @remoteport: Pointer to an nvme transport remoteport instance.
307 *
308 * This is a template downcall. NVME transport calls this function
309 * when it has completed the unregistration of a previously
310 * registered remoteport.
311 *
312 * Return value :
313 * None
314 */
315 static void
316 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
317 {
318 struct lpfc_nvme_rport *rport = remoteport->private;
319 struct lpfc_vport *vport;
320 struct lpfc_nodelist *ndlp;
321
322 ndlp = rport->ndlp;
323 if (!ndlp)
324 goto rport_err;
325
326 vport = ndlp->vport;
327 if (!vport)
328 goto rport_err;
329
330 /* Remove this rport from the lport's list - memory is owned by the
331 * transport. Remove the ndlp reference for the NVME transport before
332 * calling state machine to remove the node.
333 */
334 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
335 "6146 remoteport delete of remoteport %p\n",
336 remoteport);
337 spin_lock_irq(&vport->phba->hbalock);
338
339 /* The register rebind might have occurred before the delete
340 * downcall. Guard against this race.
341 */
342 if (ndlp->upcall_flags & NLP_WAIT_FOR_UNREG) {
343 ndlp->nrport = NULL;
344 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
345 }
346 spin_unlock_irq(&vport->phba->hbalock);
347
348 /* Remove original register reference. The host transport
349 * won't reference this rport/remoteport any further.
350 */
351 lpfc_nlp_put(ndlp);
352
353 rport_err:
354 return;
355 }
356
357 static void
358 lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
359 struct lpfc_wcqe_complete *wcqe)
360 {
361 struct lpfc_vport *vport = cmdwqe->vport;
362 struct lpfc_nvme_lport *lport;
363 uint32_t status;
364 struct nvmefc_ls_req *pnvme_lsreq;
365 struct lpfc_dmabuf *buf_ptr;
366 struct lpfc_nodelist *ndlp;
367
368 pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
369 status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
370
371 if (vport->localport) {
372 lport = (struct lpfc_nvme_lport *)vport->localport->private;
373 if (lport) {
374 atomic_inc(&lport->fc4NvmeLsCmpls);
375 if (status) {
376 if (bf_get(lpfc_wcqe_c_xb, wcqe))
377 atomic_inc(&lport->cmpl_ls_xb);
378 atomic_inc(&lport->cmpl_ls_err);
379 }
380 }
381 }
382
383 ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
384 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
385 "6047 nvme cmpl Enter "
386 "Data %p DID %x Xri: %x status %x reason x%x cmd:%p "
387 "lsreg:%p bmp:%p ndlp:%p\n",
388 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
389 cmdwqe->sli4_xritag, status,
390 (wcqe->parameter & 0xffff),
391 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
392
393 lpfc_nvmeio_data(phba, "NVME LS CMPL: xri x%x stat x%x parm x%x\n",
394 cmdwqe->sli4_xritag, status, wcqe->parameter);
395
396 if (cmdwqe->context3) {
397 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
398 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
399 kfree(buf_ptr);
400 cmdwqe->context3 = NULL;
401 }
402 if (pnvme_lsreq->done)
403 pnvme_lsreq->done(pnvme_lsreq, status);
404 else
405 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
406 "6046 nvme cmpl without done call back? "
407 "Data %p DID %x Xri: %x status %x\n",
408 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
409 cmdwqe->sli4_xritag, status);
410 if (ndlp) {
411 lpfc_nlp_put(ndlp);
412 cmdwqe->context1 = NULL;
413 }
414 lpfc_sli_release_iocbq(phba, cmdwqe);
415 }
416
417 static int
418 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
419 struct lpfc_dmabuf *inp,
420 struct nvmefc_ls_req *pnvme_lsreq,
421 void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
422 struct lpfc_wcqe_complete *),
423 struct lpfc_nodelist *ndlp, uint32_t num_entry,
424 uint32_t tmo, uint8_t retry)
425 {
426 struct lpfc_hba *phba = vport->phba;
427 union lpfc_wqe128 *wqe;
428 struct lpfc_iocbq *genwqe;
429 struct ulp_bde64 *bpl;
430 struct ulp_bde64 bde;
431 int i, rc, xmit_len, first_len;
432
433 /* Allocate buffer for command WQE */
434 genwqe = lpfc_sli_get_iocbq(phba);
435 if (genwqe == NULL)
436 return 1;
437
438 wqe = &genwqe->wqe;
439 memset(wqe, 0, sizeof(union lpfc_wqe));
440
441 genwqe->context3 = (uint8_t *)bmp;
442 genwqe->iocb_flag |= LPFC_IO_NVME_LS;
443
444 /* Save for completion so we can release these resources */
445 genwqe->context1 = lpfc_nlp_get(ndlp);
446 genwqe->context2 = (uint8_t *)pnvme_lsreq;
447 /* Fill in payload, bp points to frame payload */
448
449 if (!tmo)
450 /* FC spec states we need 3 * ratov for CT requests */
451 tmo = (3 * phba->fc_ratov);
452
453 /* For this command calculate the xmit length of the request bde. */
454 xmit_len = 0;
455 first_len = 0;
456 bpl = (struct ulp_bde64 *)bmp->virt;
457 for (i = 0; i < num_entry; i++) {
458 bde.tus.w = bpl[i].tus.w;
459 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
460 break;
461 xmit_len += bde.tus.f.bdeSize;
462 if (i == 0)
463 first_len = xmit_len;
464 }
465
466 genwqe->rsvd2 = num_entry;
467 genwqe->hba_wqidx = 0;
468
469 /* Words 0 - 2 */
470 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
471 wqe->generic.bde.tus.f.bdeSize = first_len;
472 wqe->generic.bde.addrLow = bpl[0].addrLow;
473 wqe->generic.bde.addrHigh = bpl[0].addrHigh;
474
475 /* Word 3 */
476 wqe->gen_req.request_payload_len = first_len;
477
478 /* Word 4 */
479
480 /* Word 5 */
481 bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
482 bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
483 bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
484 bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ);
485 bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
486
487 /* Word 6 */
488 bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
489 phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
490 bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
491
492 /* Word 7 */
493 bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1));
494 bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
495 bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
496 bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
497
498 /* Word 8 */
499 wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
500
501 /* Word 9 */
502 bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
503
504 /* Word 10 */
505 bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
506 bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
507 bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
508 bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
509 bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
510
511 /* Word 11 */
512 bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
513 bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
514
515
516 /* Issue GEN REQ WQE for NPORT <did> */
517 lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
518 "6050 Issue GEN REQ WQE to NPORT x%x "
519 "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n",
520 ndlp->nlp_DID, genwqe->iotag,
521 vport->port_state,
522 genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
523 genwqe->wqe_cmpl = cmpl;
524 genwqe->iocb_cmpl = NULL;
525 genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
526 genwqe->vport = vport;
527 genwqe->retry = retry;
528
529 lpfc_nvmeio_data(phba, "NVME LS XMIT: xri x%x iotag x%x to x%06x\n",
530 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID);
531
532 rc = lpfc_sli4_issue_wqe(phba, &phba->sli4_hba.hdwq[0], genwqe);
533 if (rc) {
534 lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
535 "6045 Issue GEN REQ WQE to NPORT x%x "
536 "Data: x%x x%x\n",
537 ndlp->nlp_DID, genwqe->iotag,
538 vport->port_state);
539 lpfc_sli_release_iocbq(phba, genwqe);
540 return 1;
541 }
542 return 0;
543 }
544
545 /**
546 * lpfc_nvme_ls_req - Issue an Link Service request
547 * @lpfc_pnvme: Pointer to the driver's nvme instance data
548 * @lpfc_nvme_lport: Pointer to the driver's local port data
549 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
550 *
551 * Driver registers this routine to handle any link service request
552 * from the nvme_fc transport to a remote nvme-aware port.
553 *
554 * Return value :
555 * 0 - Success
556 * TODO: What are the failure codes.
557 **/
558 static int
559 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
560 struct nvme_fc_remote_port *pnvme_rport,
561 struct nvmefc_ls_req *pnvme_lsreq)
562 {
563 int ret = 0;
564 struct lpfc_nvme_lport *lport;
565 struct lpfc_nvme_rport *rport;
566 struct lpfc_vport *vport;
567 struct lpfc_nodelist *ndlp;
568 struct ulp_bde64 *bpl;
569 struct lpfc_dmabuf *bmp;
570 uint16_t ntype, nstate;
571
572 /* there are two dma buf in the request, actually there is one and
573 * the second one is just the start address + cmd size.
574 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
575 * in a lpfc_dmabuf struct. When freeing we just free the wrapper
576 * because the nvem layer owns the data bufs.
577 * We do not have to break these packets open, we don't care what is in
578 * them. And we do not have to look at the resonse data, we only care
579 * that we got a response. All of the caring is going to happen in the
580 * nvme-fc layer.
581 */
582
583 lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
584 rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
585 if (unlikely(!lport) || unlikely(!rport))
586 return -EINVAL;
587
588 vport = lport->vport;
589
590 if (vport->load_flag & FC_UNLOADING)
591 return -ENODEV;
592
593 /* Need the ndlp. It is stored in the driver's rport. */
594 ndlp = rport->ndlp;
595 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
596 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
597 "6051 Remoteport %p, rport has invalid ndlp. "
598 "Failing LS Req\n", pnvme_rport);
599 return -ENODEV;
600 }
601
602 /* The remote node has to be a mapped nvme target or an
603 * unmapped nvme initiator or it's an error.
604 */
605 ntype = ndlp->nlp_type;
606 nstate = ndlp->nlp_state;
607 if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) ||
608 (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) {
609 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
610 "6088 DID x%06x not ready for "
611 "IO. State x%x, Type x%x\n",
612 pnvme_rport->port_id,
613 ndlp->nlp_state, ndlp->nlp_type);
614 return -ENODEV;
615 }
616 bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
617 if (!bmp) {
618
619 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
620 "6044 Could not find node for DID %x\n",
621 pnvme_rport->port_id);
622 return 2;
623 }
624 INIT_LIST_HEAD(&bmp->list);
625 bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
626 if (!bmp->virt) {
627 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
628 "6042 Could not find node for DID %x\n",
629 pnvme_rport->port_id);
630 kfree(bmp);
631 return 3;
632 }
633 bpl = (struct ulp_bde64 *)bmp->virt;
634 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
635 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
636 bpl->tus.f.bdeFlags = 0;
637 bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
638 bpl->tus.w = le32_to_cpu(bpl->tus.w);
639 bpl++;
640
641 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
642 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
643 bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
644 bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
645 bpl->tus.w = le32_to_cpu(bpl->tus.w);
646
647 /* Expand print to include key fields. */
648 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
649 "6149 Issue LS Req to DID 0x%06x lport %p, rport %p "
650 "lsreq%p rqstlen:%d rsplen:%d %pad %pad\n",
651 ndlp->nlp_DID,
652 pnvme_lport, pnvme_rport,
653 pnvme_lsreq, pnvme_lsreq->rqstlen,
654 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
655 &pnvme_lsreq->rspdma);
656
657 atomic_inc(&lport->fc4NvmeLsRequests);
658
659 /* Hardcode the wait to 30 seconds. Connections are failing otherwise.
660 * This code allows it all to work.
661 */
662 ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
663 pnvme_lsreq, lpfc_nvme_cmpl_gen_req,
664 ndlp, 2, 30, 0);
665 if (ret != WQE_SUCCESS) {
666 atomic_inc(&lport->xmt_ls_err);
667 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
668 "6052 EXIT. issue ls wqe failed lport %p, "
669 "rport %p lsreq%p Status %x DID %x\n",
670 pnvme_lport, pnvme_rport, pnvme_lsreq,
671 ret, ndlp->nlp_DID);
672 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
673 kfree(bmp);
674 return ret;
675 }
676
677 /* Stub in routine and return 0 for now. */
678 return ret;
679 }
680
681 /**
682 * lpfc_nvme_ls_abort - Issue an Link Service request
683 * @lpfc_pnvme: Pointer to the driver's nvme instance data
684 * @lpfc_nvme_lport: Pointer to the driver's local port data
685 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
686 *
687 * Driver registers this routine to handle any link service request
688 * from the nvme_fc transport to a remote nvme-aware port.
689 *
690 * Return value :
691 * 0 - Success
692 * TODO: What are the failure codes.
693 **/
694 static void
695 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
696 struct nvme_fc_remote_port *pnvme_rport,
697 struct nvmefc_ls_req *pnvme_lsreq)
698 {
699 struct lpfc_nvme_lport *lport;
700 struct lpfc_vport *vport;
701 struct lpfc_hba *phba;
702 struct lpfc_nodelist *ndlp;
703 LIST_HEAD(abort_list);
704 struct lpfc_sli_ring *pring;
705 struct lpfc_iocbq *wqe, *next_wqe;
706
707 lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
708 if (unlikely(!lport))
709 return;
710 vport = lport->vport;
711 phba = vport->phba;
712
713 if (vport->load_flag & FC_UNLOADING)
714 return;
715
716 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
717 if (!ndlp) {
718 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
719 "6049 Could not find node for DID %x\n",
720 pnvme_rport->port_id);
721 return;
722 }
723
724 /* Expand print to include key fields. */
725 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
726 "6040 ENTER. lport %p, rport %p lsreq %p rqstlen:%d "
727 "rsplen:%d %pad %pad\n",
728 pnvme_lport, pnvme_rport,
729 pnvme_lsreq, pnvme_lsreq->rqstlen,
730 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma,
731 &pnvme_lsreq->rspdma);
732
733 /*
734 * Lock the ELS ring txcmplq and build a local list of all ELS IOs
735 * that need an ABTS. The IOs need to stay on the txcmplq so that
736 * the abort operation completes them successfully.
737 */
738 pring = phba->sli4_hba.nvmels_wq->pring;
739 spin_lock_irq(&phba->hbalock);
740 spin_lock(&pring->ring_lock);
741 list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
742 /* Add to abort_list on on NDLP match. */
743 if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) {
744 wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
745 list_add_tail(&wqe->dlist, &abort_list);
746 }
747 }
748 spin_unlock(&pring->ring_lock);
749 spin_unlock_irq(&phba->hbalock);
750
751 /* Abort the targeted IOs and remove them from the abort list. */
752 list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) {
753 atomic_inc(&lport->xmt_ls_abort);
754 spin_lock_irq(&phba->hbalock);
755 list_del_init(&wqe->dlist);
756 lpfc_sli_issue_abort_iotag(phba, pring, wqe);
757 spin_unlock_irq(&phba->hbalock);
758 }
759 }
760
761 /* Fix up the existing sgls for NVME IO. */
762 static inline void
763 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
764 struct lpfc_io_buf *lpfc_ncmd,
765 struct nvmefc_fcp_req *nCmd)
766 {
767 struct lpfc_hba *phba = vport->phba;
768 struct sli4_sge *sgl;
769 union lpfc_wqe128 *wqe;
770 uint32_t *wptr, *dptr;
771
772 /*
773 * Get a local pointer to the built-in wqe and correct
774 * the cmd size to match NVME's 96 bytes and fix
775 * the dma address.
776 */
777
778 wqe = &lpfc_ncmd->cur_iocbq.wqe;
779
780 /*
781 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
782 * match NVME. NVME sends 96 bytes. Also, use the
783 * nvme commands command and response dma addresses
784 * rather than the virtual memory to ease the restore
785 * operation.
786 */
787 sgl = lpfc_ncmd->dma_sgl;
788 sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
789 if (phba->cfg_nvme_embed_cmd) {
790 sgl->addr_hi = 0;
791 sgl->addr_lo = 0;
792
793 /* Word 0-2 - NVME CMND IU (embedded payload) */
794 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
795 wqe->generic.bde.tus.f.bdeSize = 56;
796 wqe->generic.bde.addrHigh = 0;
797 wqe->generic.bde.addrLow = 64; /* Word 16 */
798
799 /* Word 10 - dbde is 0, wqes is 1 in template */
800
801 /*
802 * Embed the payload in the last half of the WQE
803 * WQE words 16-30 get the NVME CMD IU payload
804 *
805 * WQE words 16-19 get payload Words 1-4
806 * WQE words 20-21 get payload Words 6-7
807 * WQE words 22-29 get payload Words 16-23
808 */
809 wptr = &wqe->words[16]; /* WQE ptr */
810 dptr = (uint32_t *)nCmd->cmdaddr; /* payload ptr */
811 dptr++; /* Skip Word 0 in payload */
812
813 *wptr++ = *dptr++; /* Word 1 */
814 *wptr++ = *dptr++; /* Word 2 */
815 *wptr++ = *dptr++; /* Word 3 */
816 *wptr++ = *dptr++; /* Word 4 */
817 dptr++; /* Skip Word 5 in payload */
818 *wptr++ = *dptr++; /* Word 6 */
819 *wptr++ = *dptr++; /* Word 7 */
820 dptr += 8; /* Skip Words 8-15 in payload */
821 *wptr++ = *dptr++; /* Word 16 */
822 *wptr++ = *dptr++; /* Word 17 */
823 *wptr++ = *dptr++; /* Word 18 */
824 *wptr++ = *dptr++; /* Word 19 */
825 *wptr++ = *dptr++; /* Word 20 */
826 *wptr++ = *dptr++; /* Word 21 */
827 *wptr++ = *dptr++; /* Word 22 */
828 *wptr = *dptr; /* Word 23 */
829 } else {
830 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma));
831 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma));
832
833 /* Word 0-2 - NVME CMND IU Inline BDE */
834 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
835 wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen;
836 wqe->generic.bde.addrHigh = sgl->addr_hi;
837 wqe->generic.bde.addrLow = sgl->addr_lo;
838
839 /* Word 10 */
840 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
841 bf_set(wqe_wqes, &wqe->generic.wqe_com, 0);
842 }
843
844 sgl++;
845
846 /* Setup the physical region for the FCP RSP */
847 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
848 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
849 sgl->word2 = le32_to_cpu(sgl->word2);
850 if (nCmd->sg_cnt)
851 bf_set(lpfc_sli4_sge_last, sgl, 0);
852 else
853 bf_set(lpfc_sli4_sge_last, sgl, 1);
854 sgl->word2 = cpu_to_le32(sgl->word2);
855 sgl->sge_len = cpu_to_le32(nCmd->rsplen);
856 }
857
858 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
859 static void
860 lpfc_nvme_ktime(struct lpfc_hba *phba,
861 struct lpfc_io_buf *lpfc_ncmd)
862 {
863 uint64_t seg1, seg2, seg3, seg4;
864 uint64_t segsum;
865
866 if (!lpfc_ncmd->ts_last_cmd ||
867 !lpfc_ncmd->ts_cmd_start ||
868 !lpfc_ncmd->ts_cmd_wqput ||
869 !lpfc_ncmd->ts_isr_cmpl ||
870 !lpfc_ncmd->ts_data_nvme)
871 return;
872
873 if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_cmd_start)
874 return;
875 if (lpfc_ncmd->ts_cmd_start < lpfc_ncmd->ts_last_cmd)
876 return;
877 if (lpfc_ncmd->ts_cmd_wqput < lpfc_ncmd->ts_cmd_start)
878 return;
879 if (lpfc_ncmd->ts_isr_cmpl < lpfc_ncmd->ts_cmd_wqput)
880 return;
881 if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_isr_cmpl)
882 return;
883 /*
884 * Segment 1 - Time from Last FCP command cmpl is handed
885 * off to NVME Layer to start of next command.
886 * Segment 2 - Time from Driver receives a IO cmd start
887 * from NVME Layer to WQ put is done on IO cmd.
888 * Segment 3 - Time from Driver WQ put is done on IO cmd
889 * to MSI-X ISR for IO cmpl.
890 * Segment 4 - Time from MSI-X ISR for IO cmpl to when
891 * cmpl is handled off to the NVME Layer.
892 */
893 seg1 = lpfc_ncmd->ts_cmd_start - lpfc_ncmd->ts_last_cmd;
894 if (seg1 > 5000000) /* 5 ms - for sequential IOs only */
895 seg1 = 0;
896
897 /* Calculate times relative to start of IO */
898 seg2 = (lpfc_ncmd->ts_cmd_wqput - lpfc_ncmd->ts_cmd_start);
899 segsum = seg2;
900 seg3 = lpfc_ncmd->ts_isr_cmpl - lpfc_ncmd->ts_cmd_start;
901 if (segsum > seg3)
902 return;
903 seg3 -= segsum;
904 segsum += seg3;
905
906 seg4 = lpfc_ncmd->ts_data_nvme - lpfc_ncmd->ts_cmd_start;
907 if (segsum > seg4)
908 return;
909 seg4 -= segsum;
910
911 phba->ktime_data_samples++;
912 phba->ktime_seg1_total += seg1;
913 if (seg1 < phba->ktime_seg1_min)
914 phba->ktime_seg1_min = seg1;
915 else if (seg1 > phba->ktime_seg1_max)
916 phba->ktime_seg1_max = seg1;
917 phba->ktime_seg2_total += seg2;
918 if (seg2 < phba->ktime_seg2_min)
919 phba->ktime_seg2_min = seg2;
920 else if (seg2 > phba->ktime_seg2_max)
921 phba->ktime_seg2_max = seg2;
922 phba->ktime_seg3_total += seg3;
923 if (seg3 < phba->ktime_seg3_min)
924 phba->ktime_seg3_min = seg3;
925 else if (seg3 > phba->ktime_seg3_max)
926 phba->ktime_seg3_max = seg3;
927 phba->ktime_seg4_total += seg4;
928 if (seg4 < phba->ktime_seg4_min)
929 phba->ktime_seg4_min = seg4;
930 else if (seg4 > phba->ktime_seg4_max)
931 phba->ktime_seg4_max = seg4;
932
933 lpfc_ncmd->ts_last_cmd = 0;
934 lpfc_ncmd->ts_cmd_start = 0;
935 lpfc_ncmd->ts_cmd_wqput = 0;
936 lpfc_ncmd->ts_isr_cmpl = 0;
937 lpfc_ncmd->ts_data_nvme = 0;
938 }
939 #endif
940
941 /**
942 * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO
943 * @lpfc_pnvme: Pointer to the driver's nvme instance data
944 * @lpfc_nvme_lport: Pointer to the driver's local port data
945 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
946 *
947 * Driver registers this routine as it io request handler. This
948 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
949 * data structure to the rport indicated in @lpfc_nvme_rport.
950 *
951 * Return value :
952 * 0 - Success
953 * TODO: What are the failure codes.
954 **/
955 static void
956 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
957 struct lpfc_wcqe_complete *wcqe)
958 {
959 struct lpfc_io_buf *lpfc_ncmd =
960 (struct lpfc_io_buf *)pwqeIn->context1;
961 struct lpfc_vport *vport = pwqeIn->vport;
962 struct nvmefc_fcp_req *nCmd;
963 struct nvme_fc_ersp_iu *ep;
964 struct nvme_fc_cmd_iu *cp;
965 struct lpfc_nodelist *ndlp;
966 struct lpfc_nvme_fcpreq_priv *freqpriv;
967 struct lpfc_nvme_lport *lport;
968 uint32_t code, status, idx;
969 uint16_t cid, sqhd, data;
970 uint32_t *ptr;
971
972 /* Sanity check on return of outstanding command */
973 if (!lpfc_ncmd) {
974 lpfc_printf_vlog(vport, KERN_ERR,
975 LOG_NODE | LOG_NVME_IOERR,
976 "6071 Null lpfc_ncmd pointer. No "
977 "release, skip completion\n");
978 return;
979 }
980
981 /* Guard against abort handler being called at same time */
982 spin_lock(&lpfc_ncmd->buf_lock);
983
984 if (!lpfc_ncmd->nvmeCmd) {
985 spin_unlock(&lpfc_ncmd->buf_lock);
986 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
987 "6066 Missing cmpl ptrs: lpfc_ncmd %p, "
988 "nvmeCmd %p\n",
989 lpfc_ncmd, lpfc_ncmd->nvmeCmd);
990
991 /* Release the lpfc_ncmd regardless of the missing elements. */
992 lpfc_release_nvme_buf(phba, lpfc_ncmd);
993 return;
994 }
995 nCmd = lpfc_ncmd->nvmeCmd;
996 status = bf_get(lpfc_wcqe_c_status, wcqe);
997
998 idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
999 phba->sli4_hba.hdwq[idx].nvme_cstat.io_cmpls++;
1000
1001 if (vport->localport) {
1002 lport = (struct lpfc_nvme_lport *)vport->localport->private;
1003 if (lport && status) {
1004 if (bf_get(lpfc_wcqe_c_xb, wcqe))
1005 atomic_inc(&lport->cmpl_fcp_xb);
1006 atomic_inc(&lport->cmpl_fcp_err);
1007 }
1008 }
1009
1010 lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n",
1011 lpfc_ncmd->cur_iocbq.sli4_xritag,
1012 status, wcqe->parameter);
1013 /*
1014 * Catch race where our node has transitioned, but the
1015 * transport is still transitioning.
1016 */
1017 ndlp = lpfc_ncmd->ndlp;
1018 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1019 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1020 "6062 Ignoring NVME cmpl. No ndlp\n");
1021 goto out_err;
1022 }
1023
1024 code = bf_get(lpfc_wcqe_c_code, wcqe);
1025 if (code == CQE_CODE_NVME_ERSP) {
1026 /* For this type of CQE, we need to rebuild the rsp */
1027 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
1028
1029 /*
1030 * Get Command Id from cmd to plug into response. This
1031 * code is not needed in the next NVME Transport drop.
1032 */
1033 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
1034 cid = cp->sqe.common.command_id;
1035
1036 /*
1037 * RSN is in CQE word 2
1038 * SQHD is in CQE Word 3 bits 15:0
1039 * Cmd Specific info is in CQE Word 1
1040 * and in CQE Word 0 bits 15:0
1041 */
1042 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
1043
1044 /* Now lets build the NVME ERSP IU */
1045 ep->iu_len = cpu_to_be16(8);
1046 ep->rsn = wcqe->parameter;
1047 ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
1048 ep->rsvd12 = 0;
1049 ptr = (uint32_t *)&ep->cqe.result.u64;
1050 *ptr++ = wcqe->total_data_placed;
1051 data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
1052 *ptr = (uint32_t)data;
1053 ep->cqe.sq_head = sqhd;
1054 ep->cqe.sq_id = nCmd->sqid;
1055 ep->cqe.command_id = cid;
1056 ep->cqe.status = 0;
1057
1058 lpfc_ncmd->status = IOSTAT_SUCCESS;
1059 lpfc_ncmd->result = 0;
1060 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
1061 nCmd->transferred_length = nCmd->payload_length;
1062 } else {
1063 lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK);
1064 lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK);
1065
1066 /* For NVME, the only failure path that results in an
1067 * IO error is when the adapter rejects it. All other
1068 * conditions are a success case and resolved by the
1069 * transport.
1070 * IOSTAT_FCP_RSP_ERROR means:
1071 * 1. Length of data received doesn't match total
1072 * transfer length in WQE
1073 * 2. If the RSP payload does NOT match these cases:
1074 * a. RSP length 12/24 bytes and all zeros
1075 * b. NVME ERSP
1076 */
1077 switch (lpfc_ncmd->status) {
1078 case IOSTAT_SUCCESS:
1079 nCmd->transferred_length = wcqe->total_data_placed;
1080 nCmd->rcv_rsplen = 0;
1081 nCmd->status = 0;
1082 break;
1083 case IOSTAT_FCP_RSP_ERROR:
1084 nCmd->transferred_length = wcqe->total_data_placed;
1085 nCmd->rcv_rsplen = wcqe->parameter;
1086 nCmd->status = 0;
1087 /* Sanity check */
1088 if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN)
1089 break;
1090 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
1091 "6081 NVME Completion Protocol Error: "
1092 "xri %x status x%x result x%x "
1093 "placed x%x\n",
1094 lpfc_ncmd->cur_iocbq.sli4_xritag,
1095 lpfc_ncmd->status, lpfc_ncmd->result,
1096 wcqe->total_data_placed);
1097 break;
1098 case IOSTAT_LOCAL_REJECT:
1099 /* Let fall through to set command final state. */
1100 if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED)
1101 lpfc_printf_vlog(vport, KERN_INFO,
1102 LOG_NVME_IOERR,
1103 "6032 Delay Aborted cmd %p "
1104 "nvme cmd %p, xri x%x, "
1105 "xb %d\n",
1106 lpfc_ncmd, nCmd,
1107 lpfc_ncmd->cur_iocbq.sli4_xritag,
1108 bf_get(lpfc_wcqe_c_xb, wcqe));
1109 /* fall through */
1110 default:
1111 out_err:
1112 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1113 "6072 NVME Completion Error: xri %x "
1114 "status x%x result x%x [x%x] "
1115 "placed x%x\n",
1116 lpfc_ncmd->cur_iocbq.sli4_xritag,
1117 lpfc_ncmd->status, lpfc_ncmd->result,
1118 wcqe->parameter,
1119 wcqe->total_data_placed);
1120 nCmd->transferred_length = 0;
1121 nCmd->rcv_rsplen = 0;
1122 nCmd->status = NVME_SC_INTERNAL;
1123 }
1124 }
1125
1126 /* pick up SLI4 exhange busy condition */
1127 if (bf_get(lpfc_wcqe_c_xb, wcqe))
1128 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
1129 else
1130 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
1131
1132 /* Update stats and complete the IO. There is
1133 * no need for dma unprep because the nvme_transport
1134 * owns the dma address.
1135 */
1136 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1137 if (lpfc_ncmd->ts_cmd_start) {
1138 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp;
1139 lpfc_ncmd->ts_data_nvme = ktime_get_ns();
1140 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme;
1141 lpfc_nvme_ktime(phba, lpfc_ncmd);
1142 }
1143 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1144 uint32_t cpu;
1145 idx = lpfc_ncmd->cur_iocbq.hba_wqidx;
1146 cpu = raw_smp_processor_id();
1147 if (cpu < LPFC_CHECK_CPU_CNT) {
1148 if (lpfc_ncmd->cpu != cpu)
1149 lpfc_printf_vlog(vport,
1150 KERN_INFO, LOG_NVME_IOERR,
1151 "6701 CPU Check cmpl: "
1152 "cpu %d expect %d\n",
1153 cpu, lpfc_ncmd->cpu);
1154 phba->sli4_hba.hdwq[idx].cpucheck_cmpl_io[cpu]++;
1155 }
1156 }
1157 #endif
1158
1159 /* NVME targets need completion held off until the abort exchange
1160 * completes unless the NVME Rport is getting unregistered.
1161 */
1162
1163 if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) {
1164 freqpriv = nCmd->private;
1165 freqpriv->nvme_buf = NULL;
1166 lpfc_ncmd->nvmeCmd = NULL;
1167 spin_unlock(&lpfc_ncmd->buf_lock);
1168 nCmd->done(nCmd);
1169 } else
1170 spin_unlock(&lpfc_ncmd->buf_lock);
1171
1172 /* Call release with XB=1 to queue the IO into the abort list. */
1173 lpfc_release_nvme_buf(phba, lpfc_ncmd);
1174 }
1175
1176
1177 /**
1178 * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
1179 * @lpfc_pnvme: Pointer to the driver's nvme instance data
1180 * @lpfc_nvme_lport: Pointer to the driver's local port data
1181 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1182 * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1183 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1184 *
1185 * Driver registers this routine as it io request handler. This
1186 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1187 * data structure to the rport indicated in @lpfc_nvme_rport.
1188 *
1189 * Return value :
1190 * 0 - Success
1191 * TODO: What are the failure codes.
1192 **/
1193 static int
1194 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
1195 struct lpfc_io_buf *lpfc_ncmd,
1196 struct lpfc_nodelist *pnode,
1197 struct lpfc_fc4_ctrl_stat *cstat)
1198 {
1199 struct lpfc_hba *phba = vport->phba;
1200 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1201 struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
1202 union lpfc_wqe128 *wqe = &pwqeq->wqe;
1203 uint32_t req_len;
1204
1205 if (!NLP_CHK_NODE_ACT(pnode))
1206 return -EINVAL;
1207
1208 /*
1209 * There are three possibilities here - use scatter-gather segment, use
1210 * the single mapping, or neither.
1211 */
1212 if (nCmd->sg_cnt) {
1213 if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
1214 /* From the iwrite template, initialize words 7 - 11 */
1215 memcpy(&wqe->words[7],
1216 &lpfc_iwrite_cmd_template.words[7],
1217 sizeof(uint32_t) * 5);
1218
1219 /* Word 4 */
1220 wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length;
1221
1222 /* Word 5 */
1223 if ((phba->cfg_nvme_enable_fb) &&
1224 (pnode->nlp_flag & NLP_FIRSTBURST)) {
1225 req_len = lpfc_ncmd->nvmeCmd->payload_length;
1226 if (req_len < pnode->nvme_fb_size)
1227 wqe->fcp_iwrite.initial_xfer_len =
1228 req_len;
1229 else
1230 wqe->fcp_iwrite.initial_xfer_len =
1231 pnode->nvme_fb_size;
1232 } else {
1233 wqe->fcp_iwrite.initial_xfer_len = 0;
1234 }
1235 cstat->output_requests++;
1236 } else {
1237 /* From the iread template, initialize words 7 - 11 */
1238 memcpy(&wqe->words[7],
1239 &lpfc_iread_cmd_template.words[7],
1240 sizeof(uint32_t) * 5);
1241
1242 /* Word 4 */
1243 wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
1244
1245 /* Word 5 */
1246 wqe->fcp_iread.rsrvd5 = 0;
1247
1248 cstat->input_requests++;
1249 }
1250 } else {
1251 /* From the icmnd template, initialize words 4 - 11 */
1252 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4],
1253 sizeof(uint32_t) * 8);
1254 cstat->control_requests++;
1255 }
1256 /*
1257 * Finish initializing those WQE fields that are independent
1258 * of the nvme_cmnd request_buffer
1259 */
1260
1261 /* Word 3 */
1262 bf_set(payload_offset_len, &wqe->fcp_icmd,
1263 (nCmd->rsplen + nCmd->cmdlen));
1264
1265 /* Word 6 */
1266 bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
1267 phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
1268 bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
1269
1270 /* Word 8 */
1271 wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
1272
1273 /* Word 9 */
1274 bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
1275
1276 /* Words 13 14 15 are for PBDE support */
1277
1278 pwqeq->vport = vport;
1279 return 0;
1280 }
1281
1282
1283 /**
1284 * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
1285 * @lpfc_pnvme: Pointer to the driver's nvme instance data
1286 * @lpfc_nvme_lport: Pointer to the driver's local port data
1287 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1288 * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1289 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1290 *
1291 * Driver registers this routine as it io request handler. This
1292 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1293 * data structure to the rport indicated in @lpfc_nvme_rport.
1294 *
1295 * Return value :
1296 * 0 - Success
1297 * TODO: What are the failure codes.
1298 **/
1299 static int
1300 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
1301 struct lpfc_io_buf *lpfc_ncmd)
1302 {
1303 struct lpfc_hba *phba = vport->phba;
1304 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
1305 union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe;
1306 struct sli4_sge *sgl = lpfc_ncmd->dma_sgl;
1307 struct scatterlist *data_sg;
1308 struct sli4_sge *first_data_sgl;
1309 struct ulp_bde64 *bde;
1310 dma_addr_t physaddr;
1311 uint32_t num_bde = 0;
1312 uint32_t dma_len;
1313 uint32_t dma_offset = 0;
1314 int nseg, i;
1315
1316 /* Fix up the command and response DMA stuff. */
1317 lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
1318
1319 /*
1320 * There are three possibilities here - use scatter-gather segment, use
1321 * the single mapping, or neither.
1322 */
1323 if (nCmd->sg_cnt) {
1324 /*
1325 * Jump over the cmd and rsp SGEs. The fix routine
1326 * has already adjusted for this.
1327 */
1328 sgl += 2;
1329
1330 first_data_sgl = sgl;
1331 lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
1332 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) {
1333 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1334 "6058 Too many sg segments from "
1335 "NVME Transport. Max %d, "
1336 "nvmeIO sg_cnt %d\n",
1337 phba->cfg_nvme_seg_cnt + 1,
1338 lpfc_ncmd->seg_cnt);
1339 lpfc_ncmd->seg_cnt = 0;
1340 return 1;
1341 }
1342
1343 /*
1344 * The driver established a maximum scatter-gather segment count
1345 * during probe that limits the number of sg elements in any
1346 * single nvme command. Just run through the seg_cnt and format
1347 * the sge's.
1348 */
1349 nseg = nCmd->sg_cnt;
1350 data_sg = nCmd->first_sgl;
1351 for (i = 0; i < nseg; i++) {
1352 if (data_sg == NULL) {
1353 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1354 "6059 dptr err %d, nseg %d\n",
1355 i, nseg);
1356 lpfc_ncmd->seg_cnt = 0;
1357 return 1;
1358 }
1359 physaddr = data_sg->dma_address;
1360 dma_len = data_sg->length;
1361 sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr));
1362 sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr));
1363 sgl->word2 = le32_to_cpu(sgl->word2);
1364 if ((num_bde + 1) == nseg)
1365 bf_set(lpfc_sli4_sge_last, sgl, 1);
1366 else
1367 bf_set(lpfc_sli4_sge_last, sgl, 0);
1368 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
1369 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA);
1370 sgl->word2 = cpu_to_le32(sgl->word2);
1371 sgl->sge_len = cpu_to_le32(dma_len);
1372
1373 dma_offset += dma_len;
1374 data_sg = sg_next(data_sg);
1375 sgl++;
1376 }
1377 if (phba->cfg_enable_pbde) {
1378 /* Use PBDE support for first SGL only, offset == 0 */
1379 /* Words 13-15 */
1380 bde = (struct ulp_bde64 *)
1381 &wqe->words[13];
1382 bde->addrLow = first_data_sgl->addr_lo;
1383 bde->addrHigh = first_data_sgl->addr_hi;
1384 bde->tus.f.bdeSize =
1385 le32_to_cpu(first_data_sgl->sge_len);
1386 bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
1387 bde->tus.w = cpu_to_le32(bde->tus.w);
1388 /* wqe_pbde is 1 in template */
1389 } else {
1390 memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3));
1391 bf_set(wqe_pbde, &wqe->generic.wqe_com, 0);
1392 }
1393
1394 } else {
1395 lpfc_ncmd->seg_cnt = 0;
1396
1397 /* For this clause to be valid, the payload_length
1398 * and sg_cnt must zero.
1399 */
1400 if (nCmd->payload_length != 0) {
1401 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
1402 "6063 NVME DMA Prep Err: sg_cnt %d "
1403 "payload_length x%x\n",
1404 nCmd->sg_cnt, nCmd->payload_length);
1405 return 1;
1406 }
1407 }
1408 return 0;
1409 }
1410
1411 /**
1412 * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
1413 * @lpfc_pnvme: Pointer to the driver's nvme instance data
1414 * @lpfc_nvme_lport: Pointer to the driver's local port data
1415 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1416 * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1417 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1418 *
1419 * Driver registers this routine as it io request handler. This
1420 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
1421 * data structure to the rport
1422 indicated in @lpfc_nvme_rport.
1423 *
1424 * Return value :
1425 * 0 - Success
1426 * TODO: What are the failure codes.
1427 **/
1428 static int
1429 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
1430 struct nvme_fc_remote_port *pnvme_rport,
1431 void *hw_queue_handle,
1432 struct nvmefc_fcp_req *pnvme_fcreq)
1433 {
1434 int ret = 0;
1435 int expedite = 0;
1436 int idx, cpu;
1437 struct lpfc_nvme_lport *lport;
1438 struct lpfc_fc4_ctrl_stat *cstat;
1439 struct lpfc_vport *vport;
1440 struct lpfc_hba *phba;
1441 struct lpfc_nodelist *ndlp;
1442 struct lpfc_io_buf *lpfc_ncmd;
1443 struct lpfc_nvme_rport *rport;
1444 struct lpfc_nvme_qhandle *lpfc_queue_info;
1445 struct lpfc_nvme_fcpreq_priv *freqpriv;
1446 struct nvme_common_command *sqe;
1447 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1448 uint64_t start = 0;
1449 #endif
1450
1451 /* Validate pointers. LLDD fault handling with transport does
1452 * have timing races.
1453 */
1454 lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1455 if (unlikely(!lport)) {
1456 ret = -EINVAL;
1457 goto out_fail;
1458 }
1459
1460 vport = lport->vport;
1461
1462 if (unlikely(!hw_queue_handle)) {
1463 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1464 "6117 Fail IO, NULL hw_queue_handle\n");
1465 atomic_inc(&lport->xmt_fcp_err);
1466 ret = -EBUSY;
1467 goto out_fail;
1468 }
1469
1470 phba = vport->phba;
1471
1472 if (vport->load_flag & FC_UNLOADING) {
1473 ret = -ENODEV;
1474 goto out_fail;
1475 }
1476
1477 if (vport->load_flag & FC_UNLOADING) {
1478 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1479 "6124 Fail IO, Driver unload\n");
1480 atomic_inc(&lport->xmt_fcp_err);
1481 ret = -ENODEV;
1482 goto out_fail;
1483 }
1484
1485 freqpriv = pnvme_fcreq->private;
1486 if (unlikely(!freqpriv)) {
1487 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1488 "6158 Fail IO, NULL request data\n");
1489 atomic_inc(&lport->xmt_fcp_err);
1490 ret = -EINVAL;
1491 goto out_fail;
1492 }
1493
1494 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1495 if (phba->ktime_on)
1496 start = ktime_get_ns();
1497 #endif
1498 rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
1499 lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
1500
1501 /*
1502 * Catch race where our node has transitioned, but the
1503 * transport is still transitioning.
1504 */
1505 ndlp = rport->ndlp;
1506 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
1507 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1508 "6053 Fail IO, ndlp not ready: rport %p "
1509 "ndlp %p, DID x%06x\n",
1510 rport, ndlp, pnvme_rport->port_id);
1511 atomic_inc(&lport->xmt_fcp_err);
1512 ret = -EBUSY;
1513 goto out_fail;
1514 }
1515
1516 /* The remote node has to be a mapped target or it's an error. */
1517 if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
1518 (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
1519 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR,
1520 "6036 Fail IO, DID x%06x not ready for "
1521 "IO. State x%x, Type x%x Flg x%x\n",
1522 pnvme_rport->port_id,
1523 ndlp->nlp_state, ndlp->nlp_type,
1524 ndlp->upcall_flags);
1525 atomic_inc(&lport->xmt_fcp_bad_ndlp);
1526 ret = -EBUSY;
1527 goto out_fail;
1528
1529 }
1530
1531 /* Currently only NVME Keep alive commands should be expedited
1532 * if the driver runs out of a resource. These should only be
1533 * issued on the admin queue, qidx 0
1534 */
1535 if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
1536 sqe = &((struct nvme_fc_cmd_iu *)
1537 pnvme_fcreq->cmdaddr)->sqe.common;
1538 if (sqe->opcode == nvme_admin_keep_alive)
1539 expedite = 1;
1540 }
1541
1542 /* The node is shared with FCP IO, make sure the IO pending count does
1543 * not exceed the programmed depth.
1544 */
1545 if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1546 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
1547 !expedite) {
1548 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1549 "6174 Fail IO, ndlp qdepth exceeded: "
1550 "idx %d DID %x pend %d qdepth %d\n",
1551 lpfc_queue_info->index, ndlp->nlp_DID,
1552 atomic_read(&ndlp->cmd_pending),
1553 ndlp->cmd_qdepth);
1554 atomic_inc(&lport->xmt_fcp_qdepth);
1555 ret = -EBUSY;
1556 goto out_fail;
1557 }
1558 }
1559
1560 /* Lookup Hardware Queue index based on fcp_io_sched module parameter */
1561 if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) {
1562 idx = lpfc_queue_info->index;
1563 } else {
1564 cpu = raw_smp_processor_id();
1565 idx = phba->sli4_hba.cpu_map[cpu].hdwq;
1566 }
1567
1568 lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite);
1569 if (lpfc_ncmd == NULL) {
1570 atomic_inc(&lport->xmt_fcp_noxri);
1571 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1572 "6065 Fail IO, driver buffer pool is empty: "
1573 "idx %d DID %x\n",
1574 lpfc_queue_info->index, ndlp->nlp_DID);
1575 ret = -EBUSY;
1576 goto out_fail;
1577 }
1578 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1579 if (start) {
1580 lpfc_ncmd->ts_cmd_start = start;
1581 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd;
1582 } else {
1583 lpfc_ncmd->ts_cmd_start = 0;
1584 }
1585 #endif
1586
1587 /*
1588 * Store the data needed by the driver to issue, abort, and complete
1589 * an IO.
1590 * Do not let the IO hang out forever. There is no midlayer issuing
1591 * an abort so inform the FW of the maximum IO pending time.
1592 */
1593 freqpriv->nvme_buf = lpfc_ncmd;
1594 lpfc_ncmd->nvmeCmd = pnvme_fcreq;
1595 lpfc_ncmd->ndlp = ndlp;
1596 lpfc_ncmd->qidx = lpfc_queue_info->qidx;
1597
1598 /*
1599 * Issue the IO on the WQ indicated by index in the hw_queue_handle.
1600 * This identfier was create in our hardware queue create callback
1601 * routine. The driver now is dependent on the IO queue steering from
1602 * the transport. We are trusting the upper NVME layers know which
1603 * index to use and that they have affinitized a CPU to this hardware
1604 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
1605 */
1606 lpfc_ncmd->cur_iocbq.hba_wqidx = idx;
1607 cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat;
1608
1609 lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat);
1610 ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
1611 if (ret) {
1612 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1613 "6175 Fail IO, Prep DMA: "
1614 "idx %d DID %x\n",
1615 lpfc_queue_info->index, ndlp->nlp_DID);
1616 atomic_inc(&lport->xmt_fcp_err);
1617 ret = -ENOMEM;
1618 goto out_free_nvme_buf;
1619 }
1620
1621 lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n",
1622 lpfc_ncmd->cur_iocbq.sli4_xritag,
1623 lpfc_queue_info->index, ndlp->nlp_DID);
1624
1625 ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq);
1626 if (ret) {
1627 atomic_inc(&lport->xmt_fcp_wqerr);
1628 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
1629 "6113 Fail IO, Could not issue WQE err %x "
1630 "sid: x%x did: x%x oxid: x%x\n",
1631 ret, vport->fc_myDID, ndlp->nlp_DID,
1632 lpfc_ncmd->cur_iocbq.sli4_xritag);
1633 goto out_free_nvme_buf;
1634 }
1635
1636 if (phba->cfg_xri_rebalancing)
1637 lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no);
1638
1639 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
1640 if (lpfc_ncmd->ts_cmd_start)
1641 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns();
1642
1643 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) {
1644 cpu = raw_smp_processor_id();
1645 if (cpu < LPFC_CHECK_CPU_CNT) {
1646 lpfc_ncmd->cpu = cpu;
1647 if (idx != cpu)
1648 lpfc_printf_vlog(vport,
1649 KERN_INFO, LOG_NVME_IOERR,
1650 "6702 CPU Check cmd: "
1651 "cpu %d wq %d\n",
1652 lpfc_ncmd->cpu,
1653 lpfc_queue_info->index);
1654 phba->sli4_hba.hdwq[idx].cpucheck_xmt_io[cpu]++;
1655 }
1656 }
1657 #endif
1658 return 0;
1659
1660 out_free_nvme_buf:
1661 if (lpfc_ncmd->nvmeCmd->sg_cnt) {
1662 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE)
1663 cstat->output_requests--;
1664 else
1665 cstat->input_requests--;
1666 } else
1667 cstat->control_requests--;
1668 lpfc_release_nvme_buf(phba, lpfc_ncmd);
1669 out_fail:
1670 return ret;
1671 }
1672
1673 /**
1674 * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
1675 * @phba: Pointer to HBA context object
1676 * @cmdiocb: Pointer to command iocb object.
1677 * @rspiocb: Pointer to response iocb object.
1678 *
1679 * This is the callback function for any NVME FCP IO that was aborted.
1680 *
1681 * Return value:
1682 * None
1683 **/
1684 void
1685 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
1686 struct lpfc_wcqe_complete *abts_cmpl)
1687 {
1688 lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
1689 "6145 ABORT_XRI_CN completing on rpi x%x "
1690 "original iotag x%x, abort cmd iotag x%x "
1691 "req_tag x%x, status x%x, hwstatus x%x\n",
1692 cmdiocb->iocb.un.acxri.abortContextTag,
1693 cmdiocb->iocb.un.acxri.abortIoTag,
1694 cmdiocb->iotag,
1695 bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
1696 bf_get(lpfc_wcqe_c_status, abts_cmpl),
1697 bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
1698 lpfc_sli_release_iocbq(phba, cmdiocb);
1699 }
1700
1701 /**
1702 * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
1703 * @lpfc_pnvme: Pointer to the driver's nvme instance data
1704 * @lpfc_nvme_lport: Pointer to the driver's local port data
1705 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
1706 * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
1707 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
1708 *
1709 * Driver registers this routine as its nvme request io abort handler. This
1710 * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
1711 * data structure to the rport indicated in @lpfc_nvme_rport. This routine
1712 * is executed asynchronously - one the target is validated as "MAPPED" and
1713 * ready for IO, the driver issues the abort request and returns.
1714 *
1715 * Return value:
1716 * None
1717 **/
1718 static void
1719 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
1720 struct nvme_fc_remote_port *pnvme_rport,
1721 void *hw_queue_handle,
1722 struct nvmefc_fcp_req *pnvme_fcreq)
1723 {
1724 struct lpfc_nvme_lport *lport;
1725 struct lpfc_vport *vport;
1726 struct lpfc_hba *phba;
1727 struct lpfc_io_buf *lpfc_nbuf;
1728 struct lpfc_iocbq *abts_buf;
1729 struct lpfc_iocbq *nvmereq_wqe;
1730 struct lpfc_nvme_fcpreq_priv *freqpriv;
1731 union lpfc_wqe128 *abts_wqe;
1732 unsigned long flags;
1733 int ret_val;
1734
1735 /* Validate pointers. LLDD fault handling with transport does
1736 * have timing races.
1737 */
1738 lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
1739 if (unlikely(!lport))
1740 return;
1741
1742 vport = lport->vport;
1743
1744 if (unlikely(!hw_queue_handle)) {
1745 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1746 "6129 Fail Abort, HW Queue Handle NULL.\n");
1747 return;
1748 }
1749
1750 phba = vport->phba;
1751 freqpriv = pnvme_fcreq->private;
1752
1753 if (unlikely(!freqpriv))
1754 return;
1755 if (vport->load_flag & FC_UNLOADING)
1756 return;
1757
1758 /* Announce entry to new IO submit field. */
1759 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1760 "6002 Abort Request to rport DID x%06x "
1761 "for nvme_fc_req %p\n",
1762 pnvme_rport->port_id,
1763 pnvme_fcreq);
1764
1765 /* If the hba is getting reset, this flag is set. It is
1766 * cleared when the reset is complete and rings reestablished.
1767 */
1768 spin_lock_irqsave(&phba->hbalock, flags);
1769 /* driver queued commands are in process of being flushed */
1770 if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) {
1771 spin_unlock_irqrestore(&phba->hbalock, flags);
1772 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1773 "6139 Driver in reset cleanup - flushing "
1774 "NVME Req now. hba_flag x%x\n",
1775 phba->hba_flag);
1776 return;
1777 }
1778
1779 lpfc_nbuf = freqpriv->nvme_buf;
1780 if (!lpfc_nbuf) {
1781 spin_unlock_irqrestore(&phba->hbalock, flags);
1782 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1783 "6140 NVME IO req has no matching lpfc nvme "
1784 "io buffer. Skipping abort req.\n");
1785 return;
1786 } else if (!lpfc_nbuf->nvmeCmd) {
1787 spin_unlock_irqrestore(&phba->hbalock, flags);
1788 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1789 "6141 lpfc NVME IO req has no nvme_fcreq "
1790 "io buffer. Skipping abort req.\n");
1791 return;
1792 }
1793 nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
1794
1795 /* Guard against IO completion being called at same time */
1796 spin_lock(&lpfc_nbuf->buf_lock);
1797
1798 /*
1799 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
1800 * state must match the nvme_fcreq passed by the nvme
1801 * transport. If they don't match, it is likely the driver
1802 * has already completed the NVME IO and the nvme transport
1803 * has not seen it yet.
1804 */
1805 if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
1806 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1807 "6143 NVME req mismatch: "
1808 "lpfc_nbuf %p nvmeCmd %p, "
1809 "pnvme_fcreq %p. Skipping Abort xri x%x\n",
1810 lpfc_nbuf, lpfc_nbuf->nvmeCmd,
1811 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1812 goto out_unlock;
1813 }
1814
1815 /* Don't abort IOs no longer on the pending queue. */
1816 if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) {
1817 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1818 "6142 NVME IO req %p not queued - skipping "
1819 "abort req xri x%x\n",
1820 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1821 goto out_unlock;
1822 }
1823
1824 atomic_inc(&lport->xmt_fcp_abort);
1825 lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n",
1826 nvmereq_wqe->sli4_xritag,
1827 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id);
1828
1829 /* Outstanding abort is in progress */
1830 if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) {
1831 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1832 "6144 Outstanding NVME I/O Abort Request "
1833 "still pending on nvme_fcreq %p, "
1834 "lpfc_ncmd %p xri x%x\n",
1835 pnvme_fcreq, lpfc_nbuf,
1836 nvmereq_wqe->sli4_xritag);
1837 goto out_unlock;
1838 }
1839
1840 abts_buf = __lpfc_sli_get_iocbq(phba);
1841 if (!abts_buf) {
1842 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1843 "6136 No available abort wqes. Skipping "
1844 "Abts req for nvme_fcreq %p xri x%x\n",
1845 pnvme_fcreq, nvmereq_wqe->sli4_xritag);
1846 goto out_unlock;
1847 }
1848
1849 /* Ready - mark outstanding as aborted by driver. */
1850 nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
1851
1852 /* Complete prepping the abort wqe and issue to the FW. */
1853 abts_wqe = &abts_buf->wqe;
1854
1855 /* WQEs are reused. Clear stale data and set key fields to
1856 * zero like ia, iaab, iaar, xri_tag, and ctxt_tag.
1857 */
1858 memset(abts_wqe, 0, sizeof(union lpfc_wqe));
1859 bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG);
1860
1861 /* word 7 */
1862 bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
1863 bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com,
1864 nvmereq_wqe->iocb.ulpClass);
1865
1866 /* word 8 - tell the FW to abort the IO associated with this
1867 * outstanding exchange ID.
1868 */
1869 abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag;
1870
1871 /* word 9 - this is the iotag for the abts_wqe completion. */
1872 bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com,
1873 abts_buf->iotag);
1874
1875 /* word 10 */
1876 bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1);
1877 bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE);
1878
1879 /* word 11 */
1880 bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND);
1881 bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1);
1882 bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
1883
1884 /* ABTS WQE must go to the same WQ as the WQE to be aborted */
1885 abts_buf->iocb_flag |= LPFC_IO_NVME;
1886 abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx;
1887 abts_buf->vport = vport;
1888 abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
1889 ret_val = lpfc_sli4_issue_wqe(phba, lpfc_nbuf->hdwq, abts_buf);
1890 spin_unlock(&lpfc_nbuf->buf_lock);
1891 spin_unlock_irqrestore(&phba->hbalock, flags);
1892 if (ret_val) {
1893 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
1894 "6137 Failed abts issue_wqe with status x%x "
1895 "for nvme_fcreq %p.\n",
1896 ret_val, pnvme_fcreq);
1897 lpfc_sli_release_iocbq(phba, abts_buf);
1898 return;
1899 }
1900
1901 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
1902 "6138 Transport Abort NVME Request Issued for "
1903 "ox_id x%x on reqtag x%x\n",
1904 nvmereq_wqe->sli4_xritag,
1905 abts_buf->iotag);
1906 return;
1907
1908 out_unlock:
1909 spin_unlock(&lpfc_nbuf->buf_lock);
1910 spin_unlock_irqrestore(&phba->hbalock, flags);
1911 return;
1912 }
1913
1914 /* Declare and initialization an instance of the FC NVME template. */
1915 static struct nvme_fc_port_template lpfc_nvme_template = {
1916 /* initiator-based functions */
1917 .localport_delete = lpfc_nvme_localport_delete,
1918 .remoteport_delete = lpfc_nvme_remoteport_delete,
1919 .create_queue = lpfc_nvme_create_queue,
1920 .delete_queue = lpfc_nvme_delete_queue,
1921 .ls_req = lpfc_nvme_ls_req,
1922 .fcp_io = lpfc_nvme_fcp_io_submit,
1923 .ls_abort = lpfc_nvme_ls_abort,
1924 .fcp_abort = lpfc_nvme_fcp_abort,
1925
1926 .max_hw_queues = 1,
1927 .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1928 .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
1929 .dma_boundary = 0xFFFFFFFF,
1930
1931 /* Sizes of additional private data for data structures.
1932 * No use for the last two sizes at this time.
1933 */
1934 .local_priv_sz = sizeof(struct lpfc_nvme_lport),
1935 .remote_priv_sz = sizeof(struct lpfc_nvme_rport),
1936 .lsrqst_priv_sz = 0,
1937 .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv),
1938 };
1939
1940 /**
1941 * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA
1942 * @phba: The HBA for which this call is being executed.
1943 *
1944 * This routine removes a nvme buffer from head of @hdwq io_buf_list
1945 * and returns to caller.
1946 *
1947 * Return codes:
1948 * NULL - Error
1949 * Pointer to lpfc_nvme_buf - Success
1950 **/
1951 static struct lpfc_io_buf *
1952 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
1953 int idx, int expedite)
1954 {
1955 struct lpfc_io_buf *lpfc_ncmd;
1956 struct lpfc_sli4_hdw_queue *qp;
1957 struct sli4_sge *sgl;
1958 struct lpfc_iocbq *pwqeq;
1959 union lpfc_wqe128 *wqe;
1960
1961 lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite);
1962
1963 if (lpfc_ncmd) {
1964 pwqeq = &(lpfc_ncmd->cur_iocbq);
1965 wqe = &pwqeq->wqe;
1966
1967 /* Setup key fields in buffer that may have been changed
1968 * if other protocols used this buffer.
1969 */
1970 pwqeq->iocb_flag = LPFC_IO_NVME;
1971 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl;
1972 lpfc_ncmd->start_time = jiffies;
1973 lpfc_ncmd->flags = 0;
1974
1975 /* Rsp SGE will be filled in when we rcv an IO
1976 * from the NVME Layer to be sent.
1977 * The cmd is going to be embedded so we need a SKIP SGE.
1978 */
1979 sgl = lpfc_ncmd->dma_sgl;
1980 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
1981 bf_set(lpfc_sli4_sge_last, sgl, 0);
1982 sgl->word2 = cpu_to_le32(sgl->word2);
1983 /* Fill in word 3 / sgl_len during cmd submission */
1984
1985 /* Initialize WQE */
1986 memset(wqe, 0, sizeof(union lpfc_wqe));
1987
1988 if (lpfc_ndlp_check_qdepth(phba, ndlp)) {
1989 atomic_inc(&ndlp->cmd_pending);
1990 lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH;
1991 }
1992
1993 } else {
1994 qp = &phba->sli4_hba.hdwq[idx];
1995 qp->empty_io_bufs++;
1996 }
1997
1998 return lpfc_ncmd;
1999 }
2000
2001 /**
2002 * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
2003 * @phba: The Hba for which this call is being executed.
2004 * @lpfc_ncmd: The nvme buffer which is being released.
2005 *
2006 * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
2007 * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer
2008 * and cannot be reused for at least RA_TOV amount of time if it was
2009 * aborted.
2010 **/
2011 static void
2012 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd)
2013 {
2014 struct lpfc_sli4_hdw_queue *qp;
2015 unsigned long iflag = 0;
2016
2017 if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp)
2018 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending);
2019
2020 lpfc_ncmd->ndlp = NULL;
2021 lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH;
2022
2023 qp = lpfc_ncmd->hdwq;
2024 if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) {
2025 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2026 "6310 XB release deferred for "
2027 "ox_id x%x on reqtag x%x\n",
2028 lpfc_ncmd->cur_iocbq.sli4_xritag,
2029 lpfc_ncmd->cur_iocbq.iotag);
2030
2031 spin_lock_irqsave(&qp->abts_nvme_buf_list_lock, iflag);
2032 list_add_tail(&lpfc_ncmd->list,
2033 &qp->lpfc_abts_nvme_buf_list);
2034 qp->abts_nvme_io_bufs++;
2035 spin_unlock_irqrestore(&qp->abts_nvme_buf_list_lock, iflag);
2036 } else
2037 lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp);
2038 }
2039
2040 /**
2041 * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
2042 * @pvport - the lpfc_vport instance requesting a localport.
2043 *
2044 * This routine is invoked to create an nvme localport instance to bind
2045 * to the nvme_fc_transport. It is called once during driver load
2046 * like lpfc_create_shost after all other services are initialized.
2047 * It requires a vport, vpi, and wwns at call time. Other localport
2048 * parameters are modified as the driver's FCID and the Fabric WWN
2049 * are established.
2050 *
2051 * Return codes
2052 * 0 - successful
2053 * -ENOMEM - no heap memory available
2054 * other values - from nvme registration upcall
2055 **/
2056 int
2057 lpfc_nvme_create_localport(struct lpfc_vport *vport)
2058 {
2059 int ret = 0;
2060 struct lpfc_hba *phba = vport->phba;
2061 struct nvme_fc_port_info nfcp_info;
2062 struct nvme_fc_local_port *localport;
2063 struct lpfc_nvme_lport *lport;
2064
2065 /* Initialize this localport instance. The vport wwn usage ensures
2066 * that NPIV is accounted for.
2067 */
2068 memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
2069 nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
2070 nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
2071 nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
2072
2073 /* We need to tell the transport layer + 1 because it takes page
2074 * alignment into account. When space for the SGL is allocated we
2075 * allocate + 3, one for cmd, one for rsp and one for this alignment
2076 */
2077 lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1;
2078
2079 /* Advertise how many hw queues we support based on fcp_io_sched */
2080 if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ)
2081 lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue;
2082 else
2083 lpfc_nvme_template.max_hw_queues =
2084 phba->sli4_hba.num_present_cpu;
2085
2086 if (!IS_ENABLED(CONFIG_NVME_FC))
2087 return ret;
2088
2089 /* localport is allocated from the stack, but the registration
2090 * call allocates heap memory as well as the private area.
2091 */
2092
2093 ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
2094 &vport->phba->pcidev->dev, &localport);
2095 if (!ret) {
2096 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
2097 "6005 Successfully registered local "
2098 "NVME port num %d, localP %p, private %p, "
2099 "sg_seg %d\n",
2100 localport->port_num, localport,
2101 localport->private,
2102 lpfc_nvme_template.max_sgl_segments);
2103
2104 /* Private is our lport size declared in the template. */
2105 lport = (struct lpfc_nvme_lport *)localport->private;
2106 vport->localport = localport;
2107 lport->vport = vport;
2108 vport->nvmei_support = 1;
2109
2110 atomic_set(&lport->xmt_fcp_noxri, 0);
2111 atomic_set(&lport->xmt_fcp_bad_ndlp, 0);
2112 atomic_set(&lport->xmt_fcp_qdepth, 0);
2113 atomic_set(&lport->xmt_fcp_err, 0);
2114 atomic_set(&lport->xmt_fcp_wqerr, 0);
2115 atomic_set(&lport->xmt_fcp_abort, 0);
2116 atomic_set(&lport->xmt_ls_abort, 0);
2117 atomic_set(&lport->xmt_ls_err, 0);
2118 atomic_set(&lport->cmpl_fcp_xb, 0);
2119 atomic_set(&lport->cmpl_fcp_err, 0);
2120 atomic_set(&lport->cmpl_ls_xb, 0);
2121 atomic_set(&lport->cmpl_ls_err, 0);
2122 atomic_set(&lport->fc4NvmeLsRequests, 0);
2123 atomic_set(&lport->fc4NvmeLsCmpls, 0);
2124 }
2125
2126 return ret;
2127 }
2128
2129 #if (IS_ENABLED(CONFIG_NVME_FC))
2130 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg.
2131 *
2132 * The driver has to wait for the host nvme transport to callback
2133 * indicating the localport has successfully unregistered all
2134 * resources. Since this is an uninterruptible wait, loop every ten
2135 * seconds and print a message indicating no progress.
2136 *
2137 * An uninterruptible wait is used because of the risk of transport-to-
2138 * driver state mismatch.
2139 */
2140 static void
2141 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport,
2142 struct lpfc_nvme_lport *lport,
2143 struct completion *lport_unreg_cmp)
2144 {
2145 u32 wait_tmo;
2146 int ret;
2147
2148 /* Host transport has to clean up and confirm requiring an indefinite
2149 * wait. Print a message if a 10 second wait expires and renew the
2150 * wait. This is unexpected.
2151 */
2152 wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000);
2153 while (true) {
2154 ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo);
2155 if (unlikely(!ret)) {
2156 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
2157 "6176 Lport %p Localport %p wait "
2158 "timed out. Renewing.\n",
2159 lport, vport->localport);
2160 continue;
2161 }
2162 break;
2163 }
2164 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
2165 "6177 Lport %p Localport %p Complete Success\n",
2166 lport, vport->localport);
2167 }
2168 #endif
2169
2170 /**
2171 * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
2172 * @pnvme: pointer to lpfc nvme data structure.
2173 *
2174 * This routine is invoked to destroy all lports bound to the phba.
2175 * The lport memory was allocated by the nvme fc transport and is
2176 * released there. This routine ensures all rports bound to the
2177 * lport have been disconnected.
2178 *
2179 **/
2180 void
2181 lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
2182 {
2183 #if (IS_ENABLED(CONFIG_NVME_FC))
2184 struct nvme_fc_local_port *localport;
2185 struct lpfc_nvme_lport *lport;
2186 int ret;
2187 DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp);
2188
2189 if (vport->nvmei_support == 0)
2190 return;
2191
2192 localport = vport->localport;
2193 lport = (struct lpfc_nvme_lport *)localport->private;
2194
2195 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2196 "6011 Destroying NVME localport %p\n",
2197 localport);
2198
2199 /* lport's rport list is clear. Unregister
2200 * lport and release resources.
2201 */
2202 lport->lport_unreg_cmp = &lport_unreg_cmp;
2203 ret = nvme_fc_unregister_localport(localport);
2204
2205 /* Wait for completion. This either blocks
2206 * indefinitely or succeeds
2207 */
2208 lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp);
2209 vport->localport = NULL;
2210
2211 /* Regardless of the unregister upcall response, clear
2212 * nvmei_support. All rports are unregistered and the
2213 * driver will clean up.
2214 */
2215 vport->nvmei_support = 0;
2216 if (ret == 0) {
2217 lpfc_printf_vlog(vport,
2218 KERN_INFO, LOG_NVME_DISC,
2219 "6009 Unregistered lport Success\n");
2220 } else {
2221 lpfc_printf_vlog(vport,
2222 KERN_INFO, LOG_NVME_DISC,
2223 "6010 Unregistered lport "
2224 "Failed, status x%x\n",
2225 ret);
2226 }
2227 #endif
2228 }
2229
2230 void
2231 lpfc_nvme_update_localport(struct lpfc_vport *vport)
2232 {
2233 #if (IS_ENABLED(CONFIG_NVME_FC))
2234 struct nvme_fc_local_port *localport;
2235 struct lpfc_nvme_lport *lport;
2236
2237 localport = vport->localport;
2238 if (!localport) {
2239 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2240 "6710 Update NVME fail. No localport\n");
2241 return;
2242 }
2243 lport = (struct lpfc_nvme_lport *)localport->private;
2244 if (!lport) {
2245 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME,
2246 "6171 Update NVME fail. localP %p, No lport\n",
2247 localport);
2248 return;
2249 }
2250 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
2251 "6012 Update NVME lport %p did x%x\n",
2252 localport, vport->fc_myDID);
2253
2254 localport->port_id = vport->fc_myDID;
2255 if (localport->port_id == 0)
2256 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
2257 else
2258 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
2259
2260 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2261 "6030 bound lport %p to DID x%06x\n",
2262 lport, localport->port_id);
2263 #endif
2264 }
2265
2266 int
2267 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2268 {
2269 #if (IS_ENABLED(CONFIG_NVME_FC))
2270 int ret = 0;
2271 struct nvme_fc_local_port *localport;
2272 struct lpfc_nvme_lport *lport;
2273 struct lpfc_nvme_rport *rport;
2274 struct lpfc_nvme_rport *oldrport;
2275 struct nvme_fc_remote_port *remote_port;
2276 struct nvme_fc_port_info rpinfo;
2277 struct lpfc_nodelist *prev_ndlp = NULL;
2278
2279 lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
2280 "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
2281 ndlp->nlp_DID, ndlp->nlp_type);
2282
2283 localport = vport->localport;
2284 if (!localport)
2285 return 0;
2286
2287 lport = (struct lpfc_nvme_lport *)localport->private;
2288
2289 /* NVME rports are not preserved across devloss.
2290 * Just register this instance. Note, rpinfo->dev_loss_tmo
2291 * is left 0 to indicate accept transport defaults. The
2292 * driver communicates port role capabilities consistent
2293 * with the PRLI response data.
2294 */
2295 memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info));
2296 rpinfo.port_id = ndlp->nlp_DID;
2297 if (ndlp->nlp_type & NLP_NVME_TARGET)
2298 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
2299 if (ndlp->nlp_type & NLP_NVME_INITIATOR)
2300 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
2301
2302 if (ndlp->nlp_type & NLP_NVME_DISCOVERY)
2303 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY;
2304
2305 rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
2306 rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
2307
2308 spin_lock_irq(&vport->phba->hbalock);
2309 oldrport = lpfc_ndlp_get_nrport(ndlp);
2310 spin_unlock_irq(&vport->phba->hbalock);
2311 if (!oldrport)
2312 lpfc_nlp_get(ndlp);
2313
2314 ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port);
2315 if (!ret) {
2316 /* If the ndlp already has an nrport, this is just
2317 * a resume of the existing rport. Else this is a
2318 * new rport.
2319 */
2320 /* Guard against an unregister/reregister
2321 * race that leaves the WAIT flag set.
2322 */
2323 spin_lock_irq(&vport->phba->hbalock);
2324 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
2325 spin_unlock_irq(&vport->phba->hbalock);
2326 rport = remote_port->private;
2327 if (oldrport) {
2328 /* New remoteport record does not guarantee valid
2329 * host private memory area.
2330 */
2331 prev_ndlp = oldrport->ndlp;
2332 if (oldrport == remote_port->private) {
2333 /* Same remoteport - ndlp should match.
2334 * Just reuse.
2335 */
2336 lpfc_printf_vlog(ndlp->vport, KERN_INFO,
2337 LOG_NVME_DISC,
2338 "6014 Rebinding lport to "
2339 "remoteport %p wwpn 0x%llx, "
2340 "Data: x%x x%x %p %p x%x x%06x\n",
2341 remote_port,
2342 remote_port->port_name,
2343 remote_port->port_id,
2344 remote_port->port_role,
2345 prev_ndlp,
2346 ndlp,
2347 ndlp->nlp_type,
2348 ndlp->nlp_DID);
2349 return 0;
2350 }
2351
2352 /* Sever the ndlp<->rport association
2353 * before dropping the ndlp ref from
2354 * register.
2355 */
2356 spin_lock_irq(&vport->phba->hbalock);
2357 ndlp->nrport = NULL;
2358 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG;
2359 spin_unlock_irq(&vport->phba->hbalock);
2360 rport->ndlp = NULL;
2361 rport->remoteport = NULL;
2362
2363 /* Reference only removed if previous NDLP is no longer
2364 * active. It might be just a swap and removing the
2365 * reference would cause a premature cleanup.
2366 */
2367 if (prev_ndlp && prev_ndlp != ndlp) {
2368 if ((!NLP_CHK_NODE_ACT(prev_ndlp)) ||
2369 (!prev_ndlp->nrport))
2370 lpfc_nlp_put(prev_ndlp);
2371 }
2372 }
2373
2374 /* Clean bind the rport to the ndlp. */
2375 rport->remoteport = remote_port;
2376 rport->lport = lport;
2377 rport->ndlp = ndlp;
2378 spin_lock_irq(&vport->phba->hbalock);
2379 ndlp->nrport = rport;
2380 spin_unlock_irq(&vport->phba->hbalock);
2381 lpfc_printf_vlog(vport, KERN_INFO,
2382 LOG_NVME_DISC | LOG_NODE,
2383 "6022 Binding new rport to "
2384 "lport %p Remoteport %p rport %p WWNN 0x%llx, "
2385 "Rport WWPN 0x%llx DID "
2386 "x%06x Role x%x, ndlp %p prev_ndlp %p\n",
2387 lport, remote_port, rport,
2388 rpinfo.node_name, rpinfo.port_name,
2389 rpinfo.port_id, rpinfo.port_role,
2390 ndlp, prev_ndlp);
2391 } else {
2392 lpfc_printf_vlog(vport, KERN_ERR,
2393 LOG_NVME_DISC | LOG_NODE,
2394 "6031 RemotePort Registration failed "
2395 "err: %d, DID x%06x\n",
2396 ret, ndlp->nlp_DID);
2397 }
2398
2399 return ret;
2400 #else
2401 return 0;
2402 #endif
2403 }
2404
2405 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
2406 *
2407 * There is no notion of Devloss or rport recovery from the current
2408 * nvme_transport perspective. Loss of an rport just means IO cannot
2409 * be sent and recovery is completely up to the initator.
2410 * For now, the driver just unbinds the DID and port_role so that
2411 * no further IO can be issued. Changes are planned for later.
2412 *
2413 * Notes - the ndlp reference count is not decremented here since
2414 * since there is no nvme_transport api for devloss. Node ref count
2415 * is only adjusted in driver unload.
2416 */
2417 void
2418 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
2419 {
2420 #if (IS_ENABLED(CONFIG_NVME_FC))
2421 int ret;
2422 struct nvme_fc_local_port *localport;
2423 struct lpfc_nvme_lport *lport;
2424 struct lpfc_nvme_rport *rport;
2425 struct nvme_fc_remote_port *remoteport = NULL;
2426
2427 localport = vport->localport;
2428
2429 /* This is fundamental error. The localport is always
2430 * available until driver unload. Just exit.
2431 */
2432 if (!localport)
2433 return;
2434
2435 lport = (struct lpfc_nvme_lport *)localport->private;
2436 if (!lport)
2437 goto input_err;
2438
2439 spin_lock_irq(&vport->phba->hbalock);
2440 rport = lpfc_ndlp_get_nrport(ndlp);
2441 if (rport)
2442 remoteport = rport->remoteport;
2443 spin_unlock_irq(&vport->phba->hbalock);
2444 if (!remoteport)
2445 goto input_err;
2446
2447 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
2448 "6033 Unreg nvme remoteport %p, portname x%llx, "
2449 "port_id x%06x, portstate x%x port type x%x\n",
2450 remoteport, remoteport->port_name,
2451 remoteport->port_id, remoteport->port_state,
2452 ndlp->nlp_type);
2453
2454 /* Sanity check ndlp type. Only call for NVME ports. Don't
2455 * clear any rport state until the transport calls back.
2456 */
2457
2458 if (ndlp->nlp_type & NLP_NVME_TARGET) {
2459 /* No concern about the role change on the nvme remoteport.
2460 * The transport will update it.
2461 */
2462 ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG;
2463
2464 /* Don't let the host nvme transport keep sending keep-alives
2465 * on this remoteport. Vport is unloading, no recovery. The
2466 * return values is ignored. The upcall is a courtesy to the
2467 * transport.
2468 */
2469 if (vport->load_flag & FC_UNLOADING)
2470 (void)nvme_fc_set_remoteport_devloss(remoteport, 0);
2471
2472 ret = nvme_fc_unregister_remoteport(remoteport);
2473 if (ret != 0) {
2474 lpfc_nlp_put(ndlp);
2475 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2476 "6167 NVME unregister failed %d "
2477 "port_state x%x\n",
2478 ret, remoteport->port_state);
2479 }
2480 }
2481 return;
2482
2483 input_err:
2484 #endif
2485 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
2486 "6168 State error: lport %p, rport%p FCID x%06x\n",
2487 vport->localport, ndlp->rport, ndlp->nlp_DID);
2488 }
2489
2490 /**
2491 * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort
2492 * @phba: pointer to lpfc hba data structure.
2493 * @axri: pointer to the fcp xri abort wcqe structure.
2494 *
2495 * This routine is invoked by the worker thread to process a SLI4 fast-path
2496 * NVME aborted xri. Aborted NVME IO commands are completed to the transport
2497 * here.
2498 **/
2499 void
2500 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba,
2501 struct sli4_wcqe_xri_aborted *axri, int idx)
2502 {
2503 uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
2504 struct lpfc_io_buf *lpfc_ncmd, *next_lpfc_ncmd;
2505 struct nvmefc_fcp_req *nvme_cmd = NULL;
2506 struct lpfc_nodelist *ndlp;
2507 struct lpfc_sli4_hdw_queue *qp;
2508 unsigned long iflag = 0;
2509
2510 if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME))
2511 return;
2512 qp = &phba->sli4_hba.hdwq[idx];
2513 spin_lock_irqsave(&phba->hbalock, iflag);
2514 spin_lock(&qp->abts_nvme_buf_list_lock);
2515 list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd,
2516 &qp->lpfc_abts_nvme_buf_list, list) {
2517 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) {
2518 list_del_init(&lpfc_ncmd->list);
2519 qp->abts_nvme_io_bufs--;
2520 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
2521 lpfc_ncmd->status = IOSTAT_SUCCESS;
2522 spin_unlock(&qp->abts_nvme_buf_list_lock);
2523
2524 spin_unlock_irqrestore(&phba->hbalock, iflag);
2525 ndlp = lpfc_ncmd->ndlp;
2526 if (ndlp)
2527 lpfc_sli4_abts_err_handler(phba, ndlp, axri);
2528
2529 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2530 "6311 nvme_cmd %p xri x%x tag x%x "
2531 "abort complete and xri released\n",
2532 lpfc_ncmd->nvmeCmd, xri,
2533 lpfc_ncmd->cur_iocbq.iotag);
2534
2535 /* Aborted NVME commands are required to not complete
2536 * before the abort exchange command fully completes.
2537 * Once completed, it is available via the put list.
2538 */
2539 if (lpfc_ncmd->nvmeCmd) {
2540 nvme_cmd = lpfc_ncmd->nvmeCmd;
2541 nvme_cmd->done(nvme_cmd);
2542 lpfc_ncmd->nvmeCmd = NULL;
2543 }
2544 lpfc_release_nvme_buf(phba, lpfc_ncmd);
2545 return;
2546 }
2547 }
2548 spin_unlock(&qp->abts_nvme_buf_list_lock);
2549 spin_unlock_irqrestore(&phba->hbalock, iflag);
2550
2551 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS,
2552 "6312 XRI Aborted xri x%x not found\n", xri);
2553
2554 }
2555
2556 /**
2557 * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete
2558 * @phba: Pointer to HBA context object.
2559 *
2560 * This function flushes all wqes in the nvme rings and frees all resources
2561 * in the txcmplq. This function does not issue abort wqes for the IO
2562 * commands in txcmplq, they will just be returned with
2563 * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI
2564 * slot has been permanently disabled.
2565 **/
2566 void
2567 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba)
2568 {
2569 struct lpfc_sli_ring *pring;
2570 u32 i, wait_cnt = 0;
2571
2572 if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq)
2573 return;
2574
2575 /* Cycle through all NVME rings and make sure all outstanding
2576 * WQEs have been removed from the txcmplqs.
2577 */
2578 for (i = 0; i < phba->cfg_hdw_queue; i++) {
2579 if (!phba->sli4_hba.hdwq[i].nvme_wq)
2580 continue;
2581 pring = phba->sli4_hba.hdwq[i].nvme_wq->pring;
2582
2583 if (!pring)
2584 continue;
2585
2586 /* Retrieve everything on the txcmplq */
2587 while (!list_empty(&pring->txcmplq)) {
2588 msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1);
2589 wait_cnt++;
2590
2591 /* The sleep is 10mS. Every ten seconds,
2592 * dump a message. Something is wrong.
2593 */
2594 if ((wait_cnt % 1000) == 0) {
2595 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
2596 "6178 NVME IO not empty, "
2597 "cnt %d\n", wait_cnt);
2598 }
2599 }
2600 }
2601 }
Linux with added FPC-III support