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 = 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 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 default: 1110 out_err: 1111 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1112 "6072 NVME Completion Error: xri %x " 1113 "status x%x result x%x placed x%x\n", 1114 lpfc_ncmd->cur_iocbq.sli4_xritag, 1115 lpfc_ncmd->status, lpfc_ncmd->result, 1116 wcqe->total_data_placed); 1117 nCmd->transferred_length = 0; 1118 nCmd->rcv_rsplen = 0; 1119 nCmd->status = NVME_SC_INTERNAL; 1120 } 1121 } 1122 1123 /* pick up SLI4 exhange busy condition */ 1124 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 1125 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 1126 else 1127 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 1128 1129 /* Update stats and complete the IO. There is 1130 * no need for dma unprep because the nvme_transport 1131 * owns the dma address. 1132 */ 1133 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1134 if (lpfc_ncmd->ts_cmd_start) { 1135 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp; 1136 lpfc_ncmd->ts_data_nvme = ktime_get_ns(); 1137 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme; 1138 lpfc_nvme_ktime(phba, lpfc_ncmd); 1139 } 1140 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1141 uint32_t cpu; 1142 idx = lpfc_ncmd->cur_iocbq.hba_wqidx; 1143 cpu = smp_processor_id(); 1144 if (cpu < LPFC_CHECK_CPU_CNT) { 1145 if (lpfc_ncmd->cpu != cpu) 1146 lpfc_printf_vlog(vport, 1147 KERN_INFO, LOG_NVME_IOERR, 1148 "6701 CPU Check cmpl: " 1149 "cpu %d expect %d\n", 1150 cpu, lpfc_ncmd->cpu); 1151 phba->sli4_hba.hdwq[idx].cpucheck_cmpl_io[cpu]++; 1152 } 1153 } 1154 #endif 1155 1156 /* NVME targets need completion held off until the abort exchange 1157 * completes unless the NVME Rport is getting unregistered. 1158 */ 1159 1160 if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) { 1161 freqpriv = nCmd->private; 1162 freqpriv->nvme_buf = NULL; 1163 lpfc_ncmd->nvmeCmd = NULL; 1164 spin_unlock(&lpfc_ncmd->buf_lock); 1165 nCmd->done(nCmd); 1166 } else 1167 spin_unlock(&lpfc_ncmd->buf_lock); 1168 1169 /* Call release with XB=1 to queue the IO into the abort list. */ 1170 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1171 } 1172 1173 1174 /** 1175 * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO 1176 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1177 * @lpfc_nvme_lport: Pointer to the driver's local port data 1178 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1179 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1180 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1181 * 1182 * Driver registers this routine as it io request handler. This 1183 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1184 * data structure to the rport indicated in @lpfc_nvme_rport. 1185 * 1186 * Return value : 1187 * 0 - Success 1188 * TODO: What are the failure codes. 1189 **/ 1190 static int 1191 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport, 1192 struct lpfc_io_buf *lpfc_ncmd, 1193 struct lpfc_nodelist *pnode, 1194 struct lpfc_fc4_ctrl_stat *cstat) 1195 { 1196 struct lpfc_hba *phba = vport->phba; 1197 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1198 struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq); 1199 union lpfc_wqe128 *wqe = &pwqeq->wqe; 1200 uint32_t req_len; 1201 1202 if (!NLP_CHK_NODE_ACT(pnode)) 1203 return -EINVAL; 1204 1205 /* 1206 * There are three possibilities here - use scatter-gather segment, use 1207 * the single mapping, or neither. 1208 */ 1209 if (nCmd->sg_cnt) { 1210 if (nCmd->io_dir == NVMEFC_FCP_WRITE) { 1211 /* From the iwrite template, initialize words 7 - 11 */ 1212 memcpy(&wqe->words[7], 1213 &lpfc_iwrite_cmd_template.words[7], 1214 sizeof(uint32_t) * 5); 1215 1216 /* Word 4 */ 1217 wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length; 1218 1219 /* Word 5 */ 1220 if ((phba->cfg_nvme_enable_fb) && 1221 (pnode->nlp_flag & NLP_FIRSTBURST)) { 1222 req_len = lpfc_ncmd->nvmeCmd->payload_length; 1223 if (req_len < pnode->nvme_fb_size) 1224 wqe->fcp_iwrite.initial_xfer_len = 1225 req_len; 1226 else 1227 wqe->fcp_iwrite.initial_xfer_len = 1228 pnode->nvme_fb_size; 1229 } else { 1230 wqe->fcp_iwrite.initial_xfer_len = 0; 1231 } 1232 cstat->output_requests++; 1233 } else { 1234 /* From the iread template, initialize words 7 - 11 */ 1235 memcpy(&wqe->words[7], 1236 &lpfc_iread_cmd_template.words[7], 1237 sizeof(uint32_t) * 5); 1238 1239 /* Word 4 */ 1240 wqe->fcp_iread.total_xfer_len = nCmd->payload_length; 1241 1242 /* Word 5 */ 1243 wqe->fcp_iread.rsrvd5 = 0; 1244 1245 cstat->input_requests++; 1246 } 1247 } else { 1248 /* From the icmnd template, initialize words 4 - 11 */ 1249 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4], 1250 sizeof(uint32_t) * 8); 1251 cstat->control_requests++; 1252 } 1253 /* 1254 * Finish initializing those WQE fields that are independent 1255 * of the nvme_cmnd request_buffer 1256 */ 1257 1258 /* Word 3 */ 1259 bf_set(payload_offset_len, &wqe->fcp_icmd, 1260 (nCmd->rsplen + nCmd->cmdlen)); 1261 1262 /* Word 6 */ 1263 bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com, 1264 phba->sli4_hba.rpi_ids[pnode->nlp_rpi]); 1265 bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag); 1266 1267 /* Word 8 */ 1268 wqe->generic.wqe_com.abort_tag = pwqeq->iotag; 1269 1270 /* Word 9 */ 1271 bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag); 1272 1273 /* Words 13 14 15 are for PBDE support */ 1274 1275 pwqeq->vport = vport; 1276 return 0; 1277 } 1278 1279 1280 /** 1281 * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO 1282 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1283 * @lpfc_nvme_lport: Pointer to the driver's local port data 1284 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1285 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1286 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1287 * 1288 * Driver registers this routine as it io request handler. This 1289 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1290 * data structure to the rport indicated in @lpfc_nvme_rport. 1291 * 1292 * Return value : 1293 * 0 - Success 1294 * TODO: What are the failure codes. 1295 **/ 1296 static int 1297 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport, 1298 struct lpfc_io_buf *lpfc_ncmd) 1299 { 1300 struct lpfc_hba *phba = vport->phba; 1301 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1302 union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe; 1303 struct sli4_sge *sgl = lpfc_ncmd->dma_sgl; 1304 struct scatterlist *data_sg; 1305 struct sli4_sge *first_data_sgl; 1306 struct ulp_bde64 *bde; 1307 dma_addr_t physaddr; 1308 uint32_t num_bde = 0; 1309 uint32_t dma_len; 1310 uint32_t dma_offset = 0; 1311 int nseg, i; 1312 1313 /* Fix up the command and response DMA stuff. */ 1314 lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd); 1315 1316 /* 1317 * There are three possibilities here - use scatter-gather segment, use 1318 * the single mapping, or neither. 1319 */ 1320 if (nCmd->sg_cnt) { 1321 /* 1322 * Jump over the cmd and rsp SGEs. The fix routine 1323 * has already adjusted for this. 1324 */ 1325 sgl += 2; 1326 1327 first_data_sgl = sgl; 1328 lpfc_ncmd->seg_cnt = nCmd->sg_cnt; 1329 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) { 1330 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1331 "6058 Too many sg segments from " 1332 "NVME Transport. Max %d, " 1333 "nvmeIO sg_cnt %d\n", 1334 phba->cfg_nvme_seg_cnt + 1, 1335 lpfc_ncmd->seg_cnt); 1336 lpfc_ncmd->seg_cnt = 0; 1337 return 1; 1338 } 1339 1340 /* 1341 * The driver established a maximum scatter-gather segment count 1342 * during probe that limits the number of sg elements in any 1343 * single nvme command. Just run through the seg_cnt and format 1344 * the sge's. 1345 */ 1346 nseg = nCmd->sg_cnt; 1347 data_sg = nCmd->first_sgl; 1348 for (i = 0; i < nseg; i++) { 1349 if (data_sg == NULL) { 1350 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1351 "6059 dptr err %d, nseg %d\n", 1352 i, nseg); 1353 lpfc_ncmd->seg_cnt = 0; 1354 return 1; 1355 } 1356 physaddr = data_sg->dma_address; 1357 dma_len = data_sg->length; 1358 sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr)); 1359 sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr)); 1360 sgl->word2 = le32_to_cpu(sgl->word2); 1361 if ((num_bde + 1) == nseg) 1362 bf_set(lpfc_sli4_sge_last, sgl, 1); 1363 else 1364 bf_set(lpfc_sli4_sge_last, sgl, 0); 1365 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset); 1366 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA); 1367 sgl->word2 = cpu_to_le32(sgl->word2); 1368 sgl->sge_len = cpu_to_le32(dma_len); 1369 1370 dma_offset += dma_len; 1371 data_sg = sg_next(data_sg); 1372 sgl++; 1373 } 1374 if (phba->cfg_enable_pbde) { 1375 /* Use PBDE support for first SGL only, offset == 0 */ 1376 /* Words 13-15 */ 1377 bde = (struct ulp_bde64 *) 1378 &wqe->words[13]; 1379 bde->addrLow = first_data_sgl->addr_lo; 1380 bde->addrHigh = first_data_sgl->addr_hi; 1381 bde->tus.f.bdeSize = 1382 le32_to_cpu(first_data_sgl->sge_len); 1383 bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64; 1384 bde->tus.w = cpu_to_le32(bde->tus.w); 1385 /* wqe_pbde is 1 in template */ 1386 } else { 1387 memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3)); 1388 bf_set(wqe_pbde, &wqe->generic.wqe_com, 0); 1389 } 1390 1391 } else { 1392 lpfc_ncmd->seg_cnt = 0; 1393 1394 /* For this clause to be valid, the payload_length 1395 * and sg_cnt must zero. 1396 */ 1397 if (nCmd->payload_length != 0) { 1398 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1399 "6063 NVME DMA Prep Err: sg_cnt %d " 1400 "payload_length x%x\n", 1401 nCmd->sg_cnt, nCmd->payload_length); 1402 return 1; 1403 } 1404 } 1405 return 0; 1406 } 1407 1408 /** 1409 * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO 1410 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1411 * @lpfc_nvme_lport: Pointer to the driver's local port data 1412 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1413 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1414 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1415 * 1416 * Driver registers this routine as it io request handler. This 1417 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1418 * data structure to the rport 1419 indicated in @lpfc_nvme_rport. 1420 * 1421 * Return value : 1422 * 0 - Success 1423 * TODO: What are the failure codes. 1424 **/ 1425 static int 1426 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport, 1427 struct nvme_fc_remote_port *pnvme_rport, 1428 void *hw_queue_handle, 1429 struct nvmefc_fcp_req *pnvme_fcreq) 1430 { 1431 int ret = 0; 1432 int expedite = 0; 1433 int idx, cpu; 1434 struct lpfc_nvme_lport *lport; 1435 struct lpfc_fc4_ctrl_stat *cstat; 1436 struct lpfc_vport *vport; 1437 struct lpfc_hba *phba; 1438 struct lpfc_nodelist *ndlp; 1439 struct lpfc_io_buf *lpfc_ncmd; 1440 struct lpfc_nvme_rport *rport; 1441 struct lpfc_nvme_qhandle *lpfc_queue_info; 1442 struct lpfc_nvme_fcpreq_priv *freqpriv; 1443 struct nvme_common_command *sqe; 1444 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1445 uint64_t start = 0; 1446 #endif 1447 1448 /* Validate pointers. LLDD fault handling with transport does 1449 * have timing races. 1450 */ 1451 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1452 if (unlikely(!lport)) { 1453 ret = -EINVAL; 1454 goto out_fail; 1455 } 1456 1457 vport = lport->vport; 1458 1459 if (unlikely(!hw_queue_handle)) { 1460 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1461 "6117 Fail IO, NULL hw_queue_handle\n"); 1462 atomic_inc(&lport->xmt_fcp_err); 1463 ret = -EBUSY; 1464 goto out_fail; 1465 } 1466 1467 phba = vport->phba; 1468 1469 if (vport->load_flag & FC_UNLOADING) { 1470 ret = -ENODEV; 1471 goto out_fail; 1472 } 1473 1474 if (vport->load_flag & FC_UNLOADING) { 1475 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1476 "6124 Fail IO, Driver unload\n"); 1477 atomic_inc(&lport->xmt_fcp_err); 1478 ret = -ENODEV; 1479 goto out_fail; 1480 } 1481 1482 freqpriv = pnvme_fcreq->private; 1483 if (unlikely(!freqpriv)) { 1484 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1485 "6158 Fail IO, NULL request data\n"); 1486 atomic_inc(&lport->xmt_fcp_err); 1487 ret = -EINVAL; 1488 goto out_fail; 1489 } 1490 1491 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1492 if (phba->ktime_on) 1493 start = ktime_get_ns(); 1494 #endif 1495 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 1496 lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle; 1497 1498 /* 1499 * Catch race where our node has transitioned, but the 1500 * transport is still transitioning. 1501 */ 1502 ndlp = rport->ndlp; 1503 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 1504 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1505 "6053 Fail IO, ndlp not ready: rport %p " 1506 "ndlp %p, DID x%06x\n", 1507 rport, ndlp, pnvme_rport->port_id); 1508 atomic_inc(&lport->xmt_fcp_err); 1509 ret = -EBUSY; 1510 goto out_fail; 1511 } 1512 1513 /* The remote node has to be a mapped target or it's an error. */ 1514 if ((ndlp->nlp_type & NLP_NVME_TARGET) && 1515 (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) { 1516 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1517 "6036 Fail IO, DID x%06x not ready for " 1518 "IO. State x%x, Type x%x Flg x%x\n", 1519 pnvme_rport->port_id, 1520 ndlp->nlp_state, ndlp->nlp_type, 1521 ndlp->upcall_flags); 1522 atomic_inc(&lport->xmt_fcp_bad_ndlp); 1523 ret = -EBUSY; 1524 goto out_fail; 1525 1526 } 1527 1528 /* Currently only NVME Keep alive commands should be expedited 1529 * if the driver runs out of a resource. These should only be 1530 * issued on the admin queue, qidx 0 1531 */ 1532 if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) { 1533 sqe = &((struct nvme_fc_cmd_iu *) 1534 pnvme_fcreq->cmdaddr)->sqe.common; 1535 if (sqe->opcode == nvme_admin_keep_alive) 1536 expedite = 1; 1537 } 1538 1539 /* The node is shared with FCP IO, make sure the IO pending count does 1540 * not exceed the programmed depth. 1541 */ 1542 if (lpfc_ndlp_check_qdepth(phba, ndlp)) { 1543 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) && 1544 !expedite) { 1545 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1546 "6174 Fail IO, ndlp qdepth exceeded: " 1547 "idx %d DID %x pend %d qdepth %d\n", 1548 lpfc_queue_info->index, ndlp->nlp_DID, 1549 atomic_read(&ndlp->cmd_pending), 1550 ndlp->cmd_qdepth); 1551 atomic_inc(&lport->xmt_fcp_qdepth); 1552 ret = -EBUSY; 1553 goto out_fail; 1554 } 1555 } 1556 1557 /* Lookup Hardware Queue index based on fcp_io_sched module parameter */ 1558 if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) { 1559 idx = lpfc_queue_info->index; 1560 } else { 1561 cpu = smp_processor_id(); 1562 idx = phba->sli4_hba.cpu_map[cpu].hdwq; 1563 } 1564 1565 lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, idx, expedite); 1566 if (lpfc_ncmd == NULL) { 1567 atomic_inc(&lport->xmt_fcp_noxri); 1568 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1569 "6065 Fail IO, driver buffer pool is empty: " 1570 "idx %d DID %x\n", 1571 lpfc_queue_info->index, ndlp->nlp_DID); 1572 ret = -EBUSY; 1573 goto out_fail; 1574 } 1575 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1576 if (start) { 1577 lpfc_ncmd->ts_cmd_start = start; 1578 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd; 1579 } else { 1580 lpfc_ncmd->ts_cmd_start = 0; 1581 } 1582 #endif 1583 1584 /* 1585 * Store the data needed by the driver to issue, abort, and complete 1586 * an IO. 1587 * Do not let the IO hang out forever. There is no midlayer issuing 1588 * an abort so inform the FW of the maximum IO pending time. 1589 */ 1590 freqpriv->nvme_buf = lpfc_ncmd; 1591 lpfc_ncmd->nvmeCmd = pnvme_fcreq; 1592 lpfc_ncmd->ndlp = ndlp; 1593 lpfc_ncmd->qidx = lpfc_queue_info->qidx; 1594 1595 /* 1596 * Issue the IO on the WQ indicated by index in the hw_queue_handle. 1597 * This identfier was create in our hardware queue create callback 1598 * routine. The driver now is dependent on the IO queue steering from 1599 * the transport. We are trusting the upper NVME layers know which 1600 * index to use and that they have affinitized a CPU to this hardware 1601 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ. 1602 */ 1603 lpfc_ncmd->cur_iocbq.hba_wqidx = idx; 1604 cstat = &phba->sli4_hba.hdwq[idx].nvme_cstat; 1605 1606 lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat); 1607 ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd); 1608 if (ret) { 1609 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1610 "6175 Fail IO, Prep DMA: " 1611 "idx %d DID %x\n", 1612 lpfc_queue_info->index, ndlp->nlp_DID); 1613 atomic_inc(&lport->xmt_fcp_err); 1614 ret = -ENOMEM; 1615 goto out_free_nvme_buf; 1616 } 1617 1618 lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n", 1619 lpfc_ncmd->cur_iocbq.sli4_xritag, 1620 lpfc_queue_info->index, ndlp->nlp_DID); 1621 1622 ret = lpfc_sli4_issue_wqe(phba, lpfc_ncmd->hdwq, &lpfc_ncmd->cur_iocbq); 1623 if (ret) { 1624 atomic_inc(&lport->xmt_fcp_wqerr); 1625 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1626 "6113 Fail IO, Could not issue WQE err %x " 1627 "sid: x%x did: x%x oxid: x%x\n", 1628 ret, vport->fc_myDID, ndlp->nlp_DID, 1629 lpfc_ncmd->cur_iocbq.sli4_xritag); 1630 goto out_free_nvme_buf; 1631 } 1632 1633 if (phba->cfg_xri_rebalancing) 1634 lpfc_keep_pvt_pool_above_lowwm(phba, lpfc_ncmd->hdwq_no); 1635 1636 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1637 if (lpfc_ncmd->ts_cmd_start) 1638 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns(); 1639 1640 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1641 cpu = smp_processor_id(); 1642 if (cpu < LPFC_CHECK_CPU_CNT) { 1643 lpfc_ncmd->cpu = cpu; 1644 if (idx != cpu) 1645 lpfc_printf_vlog(vport, 1646 KERN_INFO, LOG_NVME_IOERR, 1647 "6702 CPU Check cmd: " 1648 "cpu %d wq %d\n", 1649 lpfc_ncmd->cpu, 1650 lpfc_queue_info->index); 1651 phba->sli4_hba.hdwq[idx].cpucheck_xmt_io[cpu]++; 1652 } 1653 } 1654 #endif 1655 return 0; 1656 1657 out_free_nvme_buf: 1658 if (lpfc_ncmd->nvmeCmd->sg_cnt) { 1659 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE) 1660 cstat->output_requests--; 1661 else 1662 cstat->input_requests--; 1663 } else 1664 cstat->control_requests--; 1665 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1666 out_fail: 1667 return ret; 1668 } 1669 1670 /** 1671 * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request. 1672 * @phba: Pointer to HBA context object 1673 * @cmdiocb: Pointer to command iocb object. 1674 * @rspiocb: Pointer to response iocb object. 1675 * 1676 * This is the callback function for any NVME FCP IO that was aborted. 1677 * 1678 * Return value: 1679 * None 1680 **/ 1681 void 1682 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb, 1683 struct lpfc_wcqe_complete *abts_cmpl) 1684 { 1685 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 1686 "6145 ABORT_XRI_CN completing on rpi x%x " 1687 "original iotag x%x, abort cmd iotag x%x " 1688 "req_tag x%x, status x%x, hwstatus x%x\n", 1689 cmdiocb->iocb.un.acxri.abortContextTag, 1690 cmdiocb->iocb.un.acxri.abortIoTag, 1691 cmdiocb->iotag, 1692 bf_get(lpfc_wcqe_c_request_tag, abts_cmpl), 1693 bf_get(lpfc_wcqe_c_status, abts_cmpl), 1694 bf_get(lpfc_wcqe_c_hw_status, abts_cmpl)); 1695 lpfc_sli_release_iocbq(phba, cmdiocb); 1696 } 1697 1698 /** 1699 * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS 1700 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1701 * @lpfc_nvme_lport: Pointer to the driver's local port data 1702 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1703 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1704 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1705 * 1706 * Driver registers this routine as its nvme request io abort handler. This 1707 * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq 1708 * data structure to the rport indicated in @lpfc_nvme_rport. This routine 1709 * is executed asynchronously - one the target is validated as "MAPPED" and 1710 * ready for IO, the driver issues the abort request and returns. 1711 * 1712 * Return value: 1713 * None 1714 **/ 1715 static void 1716 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport, 1717 struct nvme_fc_remote_port *pnvme_rport, 1718 void *hw_queue_handle, 1719 struct nvmefc_fcp_req *pnvme_fcreq) 1720 { 1721 struct lpfc_nvme_lport *lport; 1722 struct lpfc_vport *vport; 1723 struct lpfc_hba *phba; 1724 struct lpfc_io_buf *lpfc_nbuf; 1725 struct lpfc_iocbq *abts_buf; 1726 struct lpfc_iocbq *nvmereq_wqe; 1727 struct lpfc_nvme_fcpreq_priv *freqpriv; 1728 union lpfc_wqe128 *abts_wqe; 1729 unsigned long flags; 1730 int ret_val; 1731 1732 /* Validate pointers. LLDD fault handling with transport does 1733 * have timing races. 1734 */ 1735 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1736 if (unlikely(!lport)) 1737 return; 1738 1739 vport = lport->vport; 1740 1741 if (unlikely(!hw_queue_handle)) { 1742 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1743 "6129 Fail Abort, HW Queue Handle NULL.\n"); 1744 return; 1745 } 1746 1747 phba = vport->phba; 1748 freqpriv = pnvme_fcreq->private; 1749 1750 if (unlikely(!freqpriv)) 1751 return; 1752 if (vport->load_flag & FC_UNLOADING) 1753 return; 1754 1755 /* Announce entry to new IO submit field. */ 1756 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1757 "6002 Abort Request to rport DID x%06x " 1758 "for nvme_fc_req %p\n", 1759 pnvme_rport->port_id, 1760 pnvme_fcreq); 1761 1762 /* If the hba is getting reset, this flag is set. It is 1763 * cleared when the reset is complete and rings reestablished. 1764 */ 1765 spin_lock_irqsave(&phba->hbalock, flags); 1766 /* driver queued commands are in process of being flushed */ 1767 if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) { 1768 spin_unlock_irqrestore(&phba->hbalock, flags); 1769 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1770 "6139 Driver in reset cleanup - flushing " 1771 "NVME Req now. hba_flag x%x\n", 1772 phba->hba_flag); 1773 return; 1774 } 1775 1776 lpfc_nbuf = freqpriv->nvme_buf; 1777 if (!lpfc_nbuf) { 1778 spin_unlock_irqrestore(&phba->hbalock, flags); 1779 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1780 "6140 NVME IO req has no matching lpfc nvme " 1781 "io buffer. Skipping abort req.\n"); 1782 return; 1783 } else if (!lpfc_nbuf->nvmeCmd) { 1784 spin_unlock_irqrestore(&phba->hbalock, flags); 1785 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1786 "6141 lpfc NVME IO req has no nvme_fcreq " 1787 "io buffer. Skipping abort req.\n"); 1788 return; 1789 } 1790 nvmereq_wqe = &lpfc_nbuf->cur_iocbq; 1791 1792 /* Guard against IO completion being called at same time */ 1793 spin_lock(&lpfc_nbuf->buf_lock); 1794 1795 /* 1796 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's 1797 * state must match the nvme_fcreq passed by the nvme 1798 * transport. If they don't match, it is likely the driver 1799 * has already completed the NVME IO and the nvme transport 1800 * has not seen it yet. 1801 */ 1802 if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) { 1803 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1804 "6143 NVME req mismatch: " 1805 "lpfc_nbuf %p nvmeCmd %p, " 1806 "pnvme_fcreq %p. Skipping Abort xri x%x\n", 1807 lpfc_nbuf, lpfc_nbuf->nvmeCmd, 1808 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1809 goto out_unlock; 1810 } 1811 1812 /* Don't abort IOs no longer on the pending queue. */ 1813 if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) { 1814 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1815 "6142 NVME IO req %p not queued - skipping " 1816 "abort req xri x%x\n", 1817 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1818 goto out_unlock; 1819 } 1820 1821 atomic_inc(&lport->xmt_fcp_abort); 1822 lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n", 1823 nvmereq_wqe->sli4_xritag, 1824 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id); 1825 1826 /* Outstanding abort is in progress */ 1827 if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) { 1828 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1829 "6144 Outstanding NVME I/O Abort Request " 1830 "still pending on nvme_fcreq %p, " 1831 "lpfc_ncmd %p xri x%x\n", 1832 pnvme_fcreq, lpfc_nbuf, 1833 nvmereq_wqe->sli4_xritag); 1834 goto out_unlock; 1835 } 1836 1837 abts_buf = __lpfc_sli_get_iocbq(phba); 1838 if (!abts_buf) { 1839 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1840 "6136 No available abort wqes. Skipping " 1841 "Abts req for nvme_fcreq %p xri x%x\n", 1842 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1843 goto out_unlock; 1844 } 1845 1846 /* Ready - mark outstanding as aborted by driver. */ 1847 nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED; 1848 1849 /* Complete prepping the abort wqe and issue to the FW. */ 1850 abts_wqe = &abts_buf->wqe; 1851 1852 /* WQEs are reused. Clear stale data and set key fields to 1853 * zero like ia, iaab, iaar, xri_tag, and ctxt_tag. 1854 */ 1855 memset(abts_wqe, 0, sizeof(union lpfc_wqe)); 1856 bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG); 1857 1858 /* word 7 */ 1859 bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX); 1860 bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com, 1861 nvmereq_wqe->iocb.ulpClass); 1862 1863 /* word 8 - tell the FW to abort the IO associated with this 1864 * outstanding exchange ID. 1865 */ 1866 abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag; 1867 1868 /* word 9 - this is the iotag for the abts_wqe completion. */ 1869 bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com, 1870 abts_buf->iotag); 1871 1872 /* word 10 */ 1873 bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1); 1874 bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE); 1875 1876 /* word 11 */ 1877 bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND); 1878 bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1); 1879 bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 1880 1881 /* ABTS WQE must go to the same WQ as the WQE to be aborted */ 1882 abts_buf->iocb_flag |= LPFC_IO_NVME; 1883 abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx; 1884 abts_buf->vport = vport; 1885 abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl; 1886 ret_val = lpfc_sli4_issue_wqe(phba, lpfc_nbuf->hdwq, abts_buf); 1887 spin_unlock(&lpfc_nbuf->buf_lock); 1888 spin_unlock_irqrestore(&phba->hbalock, flags); 1889 if (ret_val) { 1890 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1891 "6137 Failed abts issue_wqe with status x%x " 1892 "for nvme_fcreq %p.\n", 1893 ret_val, pnvme_fcreq); 1894 lpfc_sli_release_iocbq(phba, abts_buf); 1895 return; 1896 } 1897 1898 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1899 "6138 Transport Abort NVME Request Issued for " 1900 "ox_id x%x on reqtag x%x\n", 1901 nvmereq_wqe->sli4_xritag, 1902 abts_buf->iotag); 1903 return; 1904 1905 out_unlock: 1906 spin_unlock(&lpfc_nbuf->buf_lock); 1907 spin_unlock_irqrestore(&phba->hbalock, flags); 1908 return; 1909 } 1910 1911 /* Declare and initialization an instance of the FC NVME template. */ 1912 static struct nvme_fc_port_template lpfc_nvme_template = { 1913 /* initiator-based functions */ 1914 .localport_delete = lpfc_nvme_localport_delete, 1915 .remoteport_delete = lpfc_nvme_remoteport_delete, 1916 .create_queue = lpfc_nvme_create_queue, 1917 .delete_queue = lpfc_nvme_delete_queue, 1918 .ls_req = lpfc_nvme_ls_req, 1919 .fcp_io = lpfc_nvme_fcp_io_submit, 1920 .ls_abort = lpfc_nvme_ls_abort, 1921 .fcp_abort = lpfc_nvme_fcp_abort, 1922 1923 .max_hw_queues = 1, 1924 .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1925 .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1926 .dma_boundary = 0xFFFFFFFF, 1927 1928 /* Sizes of additional private data for data structures. 1929 * No use for the last two sizes at this time. 1930 */ 1931 .local_priv_sz = sizeof(struct lpfc_nvme_lport), 1932 .remote_priv_sz = sizeof(struct lpfc_nvme_rport), 1933 .lsrqst_priv_sz = 0, 1934 .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv), 1935 }; 1936 1937 /** 1938 * lpfc_get_nvme_buf - Get a nvme buffer from io_buf_list of the HBA 1939 * @phba: The HBA for which this call is being executed. 1940 * 1941 * This routine removes a nvme buffer from head of @hdwq io_buf_list 1942 * and returns to caller. 1943 * 1944 * Return codes: 1945 * NULL - Error 1946 * Pointer to lpfc_nvme_buf - Success 1947 **/ 1948 static struct lpfc_io_buf * 1949 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp, 1950 int idx, int expedite) 1951 { 1952 struct lpfc_io_buf *lpfc_ncmd; 1953 struct lpfc_sli4_hdw_queue *qp; 1954 struct sli4_sge *sgl; 1955 struct lpfc_iocbq *pwqeq; 1956 union lpfc_wqe128 *wqe; 1957 1958 lpfc_ncmd = lpfc_get_io_buf(phba, NULL, idx, expedite); 1959 1960 if (lpfc_ncmd) { 1961 pwqeq = &(lpfc_ncmd->cur_iocbq); 1962 wqe = &pwqeq->wqe; 1963 1964 /* Setup key fields in buffer that may have been changed 1965 * if other protocols used this buffer. 1966 */ 1967 pwqeq->iocb_flag = LPFC_IO_NVME; 1968 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl; 1969 lpfc_ncmd->start_time = jiffies; 1970 lpfc_ncmd->flags = 0; 1971 1972 /* Rsp SGE will be filled in when we rcv an IO 1973 * from the NVME Layer to be sent. 1974 * The cmd is going to be embedded so we need a SKIP SGE. 1975 */ 1976 sgl = lpfc_ncmd->dma_sgl; 1977 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP); 1978 bf_set(lpfc_sli4_sge_last, sgl, 0); 1979 sgl->word2 = cpu_to_le32(sgl->word2); 1980 /* Fill in word 3 / sgl_len during cmd submission */ 1981 1982 /* Initialize WQE */ 1983 memset(wqe, 0, sizeof(union lpfc_wqe)); 1984 1985 if (lpfc_ndlp_check_qdepth(phba, ndlp)) { 1986 atomic_inc(&ndlp->cmd_pending); 1987 lpfc_ncmd->flags |= LPFC_SBUF_BUMP_QDEPTH; 1988 } 1989 1990 } else { 1991 qp = &phba->sli4_hba.hdwq[idx]; 1992 qp->empty_io_bufs++; 1993 } 1994 1995 return lpfc_ncmd; 1996 } 1997 1998 /** 1999 * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list. 2000 * @phba: The Hba for which this call is being executed. 2001 * @lpfc_ncmd: The nvme buffer which is being released. 2002 * 2003 * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba 2004 * lpfc_io_buf_list list. For SLI4 XRI's are tied to the nvme buffer 2005 * and cannot be reused for at least RA_TOV amount of time if it was 2006 * aborted. 2007 **/ 2008 static void 2009 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_io_buf *lpfc_ncmd) 2010 { 2011 struct lpfc_sli4_hdw_queue *qp; 2012 unsigned long iflag = 0; 2013 2014 if ((lpfc_ncmd->flags & LPFC_SBUF_BUMP_QDEPTH) && lpfc_ncmd->ndlp) 2015 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending); 2016 2017 lpfc_ncmd->ndlp = NULL; 2018 lpfc_ncmd->flags &= ~LPFC_SBUF_BUMP_QDEPTH; 2019 2020 qp = lpfc_ncmd->hdwq; 2021 if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) { 2022 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2023 "6310 XB release deferred for " 2024 "ox_id x%x on reqtag x%x\n", 2025 lpfc_ncmd->cur_iocbq.sli4_xritag, 2026 lpfc_ncmd->cur_iocbq.iotag); 2027 2028 spin_lock_irqsave(&qp->abts_nvme_buf_list_lock, iflag); 2029 list_add_tail(&lpfc_ncmd->list, 2030 &qp->lpfc_abts_nvme_buf_list); 2031 qp->abts_nvme_io_bufs++; 2032 spin_unlock_irqrestore(&qp->abts_nvme_buf_list_lock, iflag); 2033 } else 2034 lpfc_release_io_buf(phba, (struct lpfc_io_buf *)lpfc_ncmd, qp); 2035 } 2036 2037 /** 2038 * lpfc_nvme_create_localport - Create/Bind an nvme localport instance. 2039 * @pvport - the lpfc_vport instance requesting a localport. 2040 * 2041 * This routine is invoked to create an nvme localport instance to bind 2042 * to the nvme_fc_transport. It is called once during driver load 2043 * like lpfc_create_shost after all other services are initialized. 2044 * It requires a vport, vpi, and wwns at call time. Other localport 2045 * parameters are modified as the driver's FCID and the Fabric WWN 2046 * are established. 2047 * 2048 * Return codes 2049 * 0 - successful 2050 * -ENOMEM - no heap memory available 2051 * other values - from nvme registration upcall 2052 **/ 2053 int 2054 lpfc_nvme_create_localport(struct lpfc_vport *vport) 2055 { 2056 int ret = 0; 2057 struct lpfc_hba *phba = vport->phba; 2058 struct nvme_fc_port_info nfcp_info; 2059 struct nvme_fc_local_port *localport; 2060 struct lpfc_nvme_lport *lport; 2061 2062 /* Initialize this localport instance. The vport wwn usage ensures 2063 * that NPIV is accounted for. 2064 */ 2065 memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info)); 2066 nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR; 2067 nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn); 2068 nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn); 2069 2070 /* We need to tell the transport layer + 1 because it takes page 2071 * alignment into account. When space for the SGL is allocated we 2072 * allocate + 3, one for cmd, one for rsp and one for this alignment 2073 */ 2074 lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1; 2075 2076 /* Advertise how many hw queues we support based on fcp_io_sched */ 2077 if (phba->cfg_fcp_io_sched == LPFC_FCP_SCHED_BY_HDWQ) 2078 lpfc_nvme_template.max_hw_queues = phba->cfg_hdw_queue; 2079 else 2080 lpfc_nvme_template.max_hw_queues = 2081 phba->sli4_hba.num_present_cpu; 2082 2083 /* localport is allocated from the stack, but the registration 2084 * call allocates heap memory as well as the private area. 2085 */ 2086 #if (IS_ENABLED(CONFIG_NVME_FC)) 2087 ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template, 2088 &vport->phba->pcidev->dev, &localport); 2089 #else 2090 ret = -ENOMEM; 2091 #endif 2092 if (!ret) { 2093 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC, 2094 "6005 Successfully registered local " 2095 "NVME port num %d, localP %p, private %p, " 2096 "sg_seg %d\n", 2097 localport->port_num, localport, 2098 localport->private, 2099 lpfc_nvme_template.max_sgl_segments); 2100 2101 /* Private is our lport size declared in the template. */ 2102 lport = (struct lpfc_nvme_lport *)localport->private; 2103 vport->localport = localport; 2104 lport->vport = vport; 2105 vport->nvmei_support = 1; 2106 2107 atomic_set(&lport->xmt_fcp_noxri, 0); 2108 atomic_set(&lport->xmt_fcp_bad_ndlp, 0); 2109 atomic_set(&lport->xmt_fcp_qdepth, 0); 2110 atomic_set(&lport->xmt_fcp_err, 0); 2111 atomic_set(&lport->xmt_fcp_wqerr, 0); 2112 atomic_set(&lport->xmt_fcp_abort, 0); 2113 atomic_set(&lport->xmt_ls_abort, 0); 2114 atomic_set(&lport->xmt_ls_err, 0); 2115 atomic_set(&lport->cmpl_fcp_xb, 0); 2116 atomic_set(&lport->cmpl_fcp_err, 0); 2117 atomic_set(&lport->cmpl_ls_xb, 0); 2118 atomic_set(&lport->cmpl_ls_err, 0); 2119 atomic_set(&lport->fc4NvmeLsRequests, 0); 2120 atomic_set(&lport->fc4NvmeLsCmpls, 0); 2121 } 2122 2123 return ret; 2124 } 2125 2126 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg. 2127 * 2128 * The driver has to wait for the host nvme transport to callback 2129 * indicating the localport has successfully unregistered all 2130 * resources. Since this is an uninterruptible wait, loop every ten 2131 * seconds and print a message indicating no progress. 2132 * 2133 * An uninterruptible wait is used because of the risk of transport-to- 2134 * driver state mismatch. 2135 */ 2136 void 2137 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport, 2138 struct lpfc_nvme_lport *lport, 2139 struct completion *lport_unreg_cmp) 2140 { 2141 #if (IS_ENABLED(CONFIG_NVME_FC)) 2142 u32 wait_tmo; 2143 int ret; 2144 2145 /* Host transport has to clean up and confirm requiring an indefinite 2146 * wait. Print a message if a 10 second wait expires and renew the 2147 * wait. This is unexpected. 2148 */ 2149 wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000); 2150 while (true) { 2151 ret = wait_for_completion_timeout(lport_unreg_cmp, wait_tmo); 2152 if (unlikely(!ret)) { 2153 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 2154 "6176 Lport %p Localport %p wait " 2155 "timed out. Renewing.\n", 2156 lport, vport->localport); 2157 continue; 2158 } 2159 break; 2160 } 2161 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 2162 "6177 Lport %p Localport %p Complete Success\n", 2163 lport, vport->localport); 2164 #endif 2165 } 2166 2167 /** 2168 * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport. 2169 * @pnvme: pointer to lpfc nvme data structure. 2170 * 2171 * This routine is invoked to destroy all lports bound to the phba. 2172 * The lport memory was allocated by the nvme fc transport and is 2173 * released there. This routine ensures all rports bound to the 2174 * lport have been disconnected. 2175 * 2176 **/ 2177 void 2178 lpfc_nvme_destroy_localport(struct lpfc_vport *vport) 2179 { 2180 #if (IS_ENABLED(CONFIG_NVME_FC)) 2181 struct nvme_fc_local_port *localport; 2182 struct lpfc_nvme_lport *lport; 2183 int ret; 2184 DECLARE_COMPLETION_ONSTACK(lport_unreg_cmp); 2185 2186 if (vport->nvmei_support == 0) 2187 return; 2188 2189 localport = vport->localport; 2190 lport = (struct lpfc_nvme_lport *)localport->private; 2191 2192 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2193 "6011 Destroying NVME localport %p\n", 2194 localport); 2195 2196 /* lport's rport list is clear. Unregister 2197 * lport and release resources. 2198 */ 2199 lport->lport_unreg_cmp = &lport_unreg_cmp; 2200 ret = nvme_fc_unregister_localport(localport); 2201 2202 /* Wait for completion. This either blocks 2203 * indefinitely or succeeds 2204 */ 2205 lpfc_nvme_lport_unreg_wait(vport, lport, &lport_unreg_cmp); 2206 vport->localport = NULL; 2207 2208 /* Regardless of the unregister upcall response, clear 2209 * nvmei_support. All rports are unregistered and the 2210 * driver will clean up. 2211 */ 2212 vport->nvmei_support = 0; 2213 if (ret == 0) { 2214 lpfc_printf_vlog(vport, 2215 KERN_INFO, LOG_NVME_DISC, 2216 "6009 Unregistered lport Success\n"); 2217 } else { 2218 lpfc_printf_vlog(vport, 2219 KERN_INFO, LOG_NVME_DISC, 2220 "6010 Unregistered lport " 2221 "Failed, status x%x\n", 2222 ret); 2223 } 2224 #endif 2225 } 2226 2227 void 2228 lpfc_nvme_update_localport(struct lpfc_vport *vport) 2229 { 2230 #if (IS_ENABLED(CONFIG_NVME_FC)) 2231 struct nvme_fc_local_port *localport; 2232 struct lpfc_nvme_lport *lport; 2233 2234 localport = vport->localport; 2235 if (!localport) { 2236 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2237 "6710 Update NVME fail. No localport\n"); 2238 return; 2239 } 2240 lport = (struct lpfc_nvme_lport *)localport->private; 2241 if (!lport) { 2242 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2243 "6171 Update NVME fail. localP %p, No lport\n", 2244 localport); 2245 return; 2246 } 2247 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2248 "6012 Update NVME lport %p did x%x\n", 2249 localport, vport->fc_myDID); 2250 2251 localport->port_id = vport->fc_myDID; 2252 if (localport->port_id == 0) 2253 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY; 2254 else 2255 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR; 2256 2257 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2258 "6030 bound lport %p to DID x%06x\n", 2259 lport, localport->port_id); 2260 #endif 2261 } 2262 2263 int 2264 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2265 { 2266 #if (IS_ENABLED(CONFIG_NVME_FC)) 2267 int ret = 0; 2268 struct nvme_fc_local_port *localport; 2269 struct lpfc_nvme_lport *lport; 2270 struct lpfc_nvme_rport *rport; 2271 struct lpfc_nvme_rport *oldrport; 2272 struct nvme_fc_remote_port *remote_port; 2273 struct nvme_fc_port_info rpinfo; 2274 struct lpfc_nodelist *prev_ndlp = NULL; 2275 2276 lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC, 2277 "6006 Register NVME PORT. DID x%06x nlptype x%x\n", 2278 ndlp->nlp_DID, ndlp->nlp_type); 2279 2280 localport = vport->localport; 2281 if (!localport) 2282 return 0; 2283 2284 lport = (struct lpfc_nvme_lport *)localport->private; 2285 2286 /* NVME rports are not preserved across devloss. 2287 * Just register this instance. Note, rpinfo->dev_loss_tmo 2288 * is left 0 to indicate accept transport defaults. The 2289 * driver communicates port role capabilities consistent 2290 * with the PRLI response data. 2291 */ 2292 memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info)); 2293 rpinfo.port_id = ndlp->nlp_DID; 2294 if (ndlp->nlp_type & NLP_NVME_TARGET) 2295 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET; 2296 if (ndlp->nlp_type & NLP_NVME_INITIATOR) 2297 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR; 2298 2299 if (ndlp->nlp_type & NLP_NVME_DISCOVERY) 2300 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY; 2301 2302 rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn); 2303 rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn); 2304 2305 spin_lock_irq(&vport->phba->hbalock); 2306 oldrport = lpfc_ndlp_get_nrport(ndlp); 2307 spin_unlock_irq(&vport->phba->hbalock); 2308 if (!oldrport) 2309 lpfc_nlp_get(ndlp); 2310 2311 ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port); 2312 if (!ret) { 2313 /* If the ndlp already has an nrport, this is just 2314 * a resume of the existing rport. Else this is a 2315 * new rport. 2316 */ 2317 /* Guard against an unregister/reregister 2318 * race that leaves the WAIT flag set. 2319 */ 2320 spin_lock_irq(&vport->phba->hbalock); 2321 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 2322 spin_unlock_irq(&vport->phba->hbalock); 2323 rport = remote_port->private; 2324 if (oldrport) { 2325 /* New remoteport record does not guarantee valid 2326 * host private memory area. 2327 */ 2328 prev_ndlp = oldrport->ndlp; 2329 if (oldrport == remote_port->private) { 2330 /* Same remoteport - ndlp should match. 2331 * Just reuse. 2332 */ 2333 lpfc_printf_vlog(ndlp->vport, KERN_INFO, 2334 LOG_NVME_DISC, 2335 "6014 Rebinding lport to " 2336 "remoteport %p wwpn 0x%llx, " 2337 "Data: x%x x%x %p %p x%x x%06x\n", 2338 remote_port, 2339 remote_port->port_name, 2340 remote_port->port_id, 2341 remote_port->port_role, 2342 prev_ndlp, 2343 ndlp, 2344 ndlp->nlp_type, 2345 ndlp->nlp_DID); 2346 return 0; 2347 } 2348 2349 /* Sever the ndlp<->rport association 2350 * before dropping the ndlp ref from 2351 * register. 2352 */ 2353 spin_lock_irq(&vport->phba->hbalock); 2354 ndlp->nrport = NULL; 2355 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 2356 spin_unlock_irq(&vport->phba->hbalock); 2357 rport->ndlp = NULL; 2358 rport->remoteport = NULL; 2359 2360 /* Reference only removed if previous NDLP is no longer 2361 * active. It might be just a swap and removing the 2362 * reference would cause a premature cleanup. 2363 */ 2364 if (prev_ndlp && prev_ndlp != ndlp) { 2365 if ((!NLP_CHK_NODE_ACT(prev_ndlp)) || 2366 (!prev_ndlp->nrport)) 2367 lpfc_nlp_put(prev_ndlp); 2368 } 2369 } 2370 2371 /* Clean bind the rport to the ndlp. */ 2372 rport->remoteport = remote_port; 2373 rport->lport = lport; 2374 rport->ndlp = ndlp; 2375 spin_lock_irq(&vport->phba->hbalock); 2376 ndlp->nrport = rport; 2377 spin_unlock_irq(&vport->phba->hbalock); 2378 lpfc_printf_vlog(vport, KERN_INFO, 2379 LOG_NVME_DISC | LOG_NODE, 2380 "6022 Binding new rport to " 2381 "lport %p Remoteport %p rport %p WWNN 0x%llx, " 2382 "Rport WWPN 0x%llx DID " 2383 "x%06x Role x%x, ndlp %p prev_ndlp %p\n", 2384 lport, remote_port, rport, 2385 rpinfo.node_name, rpinfo.port_name, 2386 rpinfo.port_id, rpinfo.port_role, 2387 ndlp, prev_ndlp); 2388 } else { 2389 lpfc_printf_vlog(vport, KERN_ERR, 2390 LOG_NVME_DISC | LOG_NODE, 2391 "6031 RemotePort Registration failed " 2392 "err: %d, DID x%06x\n", 2393 ret, ndlp->nlp_DID); 2394 } 2395 2396 return ret; 2397 #else 2398 return 0; 2399 #endif 2400 } 2401 2402 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport. 2403 * 2404 * There is no notion of Devloss or rport recovery from the current 2405 * nvme_transport perspective. Loss of an rport just means IO cannot 2406 * be sent and recovery is completely up to the initator. 2407 * For now, the driver just unbinds the DID and port_role so that 2408 * no further IO can be issued. Changes are planned for later. 2409 * 2410 * Notes - the ndlp reference count is not decremented here since 2411 * since there is no nvme_transport api for devloss. Node ref count 2412 * is only adjusted in driver unload. 2413 */ 2414 void 2415 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2416 { 2417 #if (IS_ENABLED(CONFIG_NVME_FC)) 2418 int ret; 2419 struct nvme_fc_local_port *localport; 2420 struct lpfc_nvme_lport *lport; 2421 struct lpfc_nvme_rport *rport; 2422 struct nvme_fc_remote_port *remoteport = NULL; 2423 2424 localport = vport->localport; 2425 2426 /* This is fundamental error. The localport is always 2427 * available until driver unload. Just exit. 2428 */ 2429 if (!localport) 2430 return; 2431 2432 lport = (struct lpfc_nvme_lport *)localport->private; 2433 if (!lport) 2434 goto input_err; 2435 2436 spin_lock_irq(&vport->phba->hbalock); 2437 rport = lpfc_ndlp_get_nrport(ndlp); 2438 if (rport) 2439 remoteport = rport->remoteport; 2440 spin_unlock_irq(&vport->phba->hbalock); 2441 if (!remoteport) 2442 goto input_err; 2443 2444 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2445 "6033 Unreg nvme remoteport %p, portname x%llx, " 2446 "port_id x%06x, portstate x%x port type x%x\n", 2447 remoteport, remoteport->port_name, 2448 remoteport->port_id, remoteport->port_state, 2449 ndlp->nlp_type); 2450 2451 /* Sanity check ndlp type. Only call for NVME ports. Don't 2452 * clear any rport state until the transport calls back. 2453 */ 2454 2455 if (ndlp->nlp_type & NLP_NVME_TARGET) { 2456 /* No concern about the role change on the nvme remoteport. 2457 * The transport will update it. 2458 */ 2459 ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG; 2460 2461 /* Don't let the host nvme transport keep sending keep-alives 2462 * on this remoteport. Vport is unloading, no recovery. The 2463 * return values is ignored. The upcall is a courtesy to the 2464 * transport. 2465 */ 2466 if (vport->load_flag & FC_UNLOADING) 2467 (void)nvme_fc_set_remoteport_devloss(remoteport, 0); 2468 2469 ret = nvme_fc_unregister_remoteport(remoteport); 2470 if (ret != 0) { 2471 lpfc_nlp_put(ndlp); 2472 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2473 "6167 NVME unregister failed %d " 2474 "port_state x%x\n", 2475 ret, remoteport->port_state); 2476 } 2477 } 2478 return; 2479 2480 input_err: 2481 #endif 2482 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2483 "6168 State error: lport %p, rport%p FCID x%06x\n", 2484 vport->localport, ndlp->rport, ndlp->nlp_DID); 2485 } 2486 2487 /** 2488 * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort 2489 * @phba: pointer to lpfc hba data structure. 2490 * @axri: pointer to the fcp xri abort wcqe structure. 2491 * 2492 * This routine is invoked by the worker thread to process a SLI4 fast-path 2493 * NVME aborted xri. Aborted NVME IO commands are completed to the transport 2494 * here. 2495 **/ 2496 void 2497 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba, 2498 struct sli4_wcqe_xri_aborted *axri, int idx) 2499 { 2500 uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri); 2501 struct lpfc_io_buf *lpfc_ncmd, *next_lpfc_ncmd; 2502 struct nvmefc_fcp_req *nvme_cmd = NULL; 2503 struct lpfc_nodelist *ndlp; 2504 struct lpfc_sli4_hdw_queue *qp; 2505 unsigned long iflag = 0; 2506 2507 if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME)) 2508 return; 2509 qp = &phba->sli4_hba.hdwq[idx]; 2510 spin_lock_irqsave(&phba->hbalock, iflag); 2511 spin_lock(&qp->abts_nvme_buf_list_lock); 2512 list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd, 2513 &qp->lpfc_abts_nvme_buf_list, list) { 2514 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) { 2515 list_del_init(&lpfc_ncmd->list); 2516 qp->abts_nvme_io_bufs--; 2517 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2518 lpfc_ncmd->status = IOSTAT_SUCCESS; 2519 spin_unlock(&qp->abts_nvme_buf_list_lock); 2520 2521 spin_unlock_irqrestore(&phba->hbalock, iflag); 2522 ndlp = lpfc_ncmd->ndlp; 2523 if (ndlp) 2524 lpfc_sli4_abts_err_handler(phba, ndlp, axri); 2525 2526 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2527 "6311 nvme_cmd %p xri x%x tag x%x " 2528 "abort complete and xri released\n", 2529 lpfc_ncmd->nvmeCmd, xri, 2530 lpfc_ncmd->cur_iocbq.iotag); 2531 2532 /* Aborted NVME commands are required to not complete 2533 * before the abort exchange command fully completes. 2534 * Once completed, it is available via the put list. 2535 */ 2536 if (lpfc_ncmd->nvmeCmd) { 2537 nvme_cmd = lpfc_ncmd->nvmeCmd; 2538 nvme_cmd->done(nvme_cmd); 2539 lpfc_ncmd->nvmeCmd = NULL; 2540 } 2541 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2542 return; 2543 } 2544 } 2545 spin_unlock(&qp->abts_nvme_buf_list_lock); 2546 spin_unlock_irqrestore(&phba->hbalock, iflag); 2547 2548 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2549 "6312 XRI Aborted xri x%x not found\n", xri); 2550 2551 } 2552 2553 /** 2554 * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete 2555 * @phba: Pointer to HBA context object. 2556 * 2557 * This function flushes all wqes in the nvme rings and frees all resources 2558 * in the txcmplq. This function does not issue abort wqes for the IO 2559 * commands in txcmplq, they will just be returned with 2560 * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI 2561 * slot has been permanently disabled. 2562 **/ 2563 void 2564 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba) 2565 { 2566 struct lpfc_sli_ring *pring; 2567 u32 i, wait_cnt = 0; 2568 2569 if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.hdwq) 2570 return; 2571 2572 /* Cycle through all NVME rings and make sure all outstanding 2573 * WQEs have been removed from the txcmplqs. 2574 */ 2575 for (i = 0; i < phba->cfg_hdw_queue; i++) { 2576 if (!phba->sli4_hba.hdwq[i].nvme_wq) 2577 continue; 2578 pring = phba->sli4_hba.hdwq[i].nvme_wq->pring; 2579 2580 if (!pring) 2581 continue; 2582 2583 /* Retrieve everything on the txcmplq */ 2584 while (!list_empty(&pring->txcmplq)) { 2585 msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1); 2586 wait_cnt++; 2587 2588 /* The sleep is 10mS. Every ten seconds, 2589 * dump a message. Something is wrong. 2590 */ 2591 if ((wait_cnt % 1000) == 0) { 2592 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 2593 "6178 NVME IO not empty, " 2594 "cnt %d\n", wait_cnt); 2595 } 2596 } 2597 } 2598 } 2599