1 /******************************************************************* 2 * This file is part of the Emulex Linux Device Driver for * 3 * Fibre Channel Host Bus Adapters. * 4 * Copyright (C) 2017-2018 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_nvme_buf * 60 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp, 61 int expedite); 62 63 static void 64 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_nvme_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 vport->phba->cfg_nvme_io_channel); 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 "io_channel %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%xi 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 complete(&lport->lport_unreg_done); 301 } 302 303 /* lpfc_nvme_remoteport_delete 304 * 305 * @remoteport: Pointer to an nvme transport remoteport instance. 306 * 307 * This is a template downcall. NVME transport calls this function 308 * when it has completed the unregistration of a previously 309 * registered remoteport. 310 * 311 * Return value : 312 * None 313 */ 314 void 315 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport) 316 { 317 struct lpfc_nvme_rport *rport = remoteport->private; 318 struct lpfc_vport *vport; 319 struct lpfc_nodelist *ndlp; 320 321 ndlp = rport->ndlp; 322 if (!ndlp) 323 goto rport_err; 324 325 vport = ndlp->vport; 326 if (!vport) 327 goto rport_err; 328 329 /* Remove this rport from the lport's list - memory is owned by the 330 * transport. Remove the ndlp reference for the NVME transport before 331 * calling state machine to remove the node. 332 */ 333 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 334 "6146 remoteport delete of remoteport %p\n", 335 remoteport); 336 spin_lock_irq(&vport->phba->hbalock); 337 338 /* The register rebind might have occurred before the delete 339 * downcall. Guard against this race. 340 */ 341 if (ndlp->upcall_flags & NLP_WAIT_FOR_UNREG) { 342 ndlp->nrport = NULL; 343 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 344 } 345 spin_unlock_irq(&vport->phba->hbalock); 346 347 /* Remove original register reference. The host transport 348 * won't reference this rport/remoteport any further. 349 */ 350 lpfc_nlp_put(ndlp); 351 352 rport_err: 353 return; 354 } 355 356 static void 357 lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe, 358 struct lpfc_wcqe_complete *wcqe) 359 { 360 struct lpfc_vport *vport = cmdwqe->vport; 361 struct lpfc_nvme_lport *lport; 362 uint32_t status; 363 struct nvmefc_ls_req *pnvme_lsreq; 364 struct lpfc_dmabuf *buf_ptr; 365 struct lpfc_nodelist *ndlp; 366 367 pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2; 368 status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK; 369 370 if (vport->localport) { 371 lport = (struct lpfc_nvme_lport *)vport->localport->private; 372 if (lport) { 373 atomic_inc(&lport->fc4NvmeLsCmpls); 374 if (status) { 375 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 376 atomic_inc(&lport->cmpl_ls_xb); 377 atomic_inc(&lport->cmpl_ls_err); 378 } 379 } 380 } 381 382 ndlp = (struct lpfc_nodelist *)cmdwqe->context1; 383 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 384 "6047 nvme cmpl Enter " 385 "Data %p DID %x Xri: %x status %x reason x%x cmd:%p " 386 "lsreg:%p bmp:%p ndlp:%p\n", 387 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0, 388 cmdwqe->sli4_xritag, status, 389 (wcqe->parameter & 0xffff), 390 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp); 391 392 lpfc_nvmeio_data(phba, "NVME LS CMPL: xri x%x stat x%x parm x%x\n", 393 cmdwqe->sli4_xritag, status, wcqe->parameter); 394 395 if (cmdwqe->context3) { 396 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3; 397 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys); 398 kfree(buf_ptr); 399 cmdwqe->context3 = NULL; 400 } 401 if (pnvme_lsreq->done) 402 pnvme_lsreq->done(pnvme_lsreq, status); 403 else 404 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 405 "6046 nvme cmpl without done call back? " 406 "Data %p DID %x Xri: %x status %x\n", 407 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0, 408 cmdwqe->sli4_xritag, status); 409 if (ndlp) { 410 lpfc_nlp_put(ndlp); 411 cmdwqe->context1 = NULL; 412 } 413 lpfc_sli_release_iocbq(phba, cmdwqe); 414 } 415 416 static int 417 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp, 418 struct lpfc_dmabuf *inp, 419 struct nvmefc_ls_req *pnvme_lsreq, 420 void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *, 421 struct lpfc_wcqe_complete *), 422 struct lpfc_nodelist *ndlp, uint32_t num_entry, 423 uint32_t tmo, uint8_t retry) 424 { 425 struct lpfc_hba *phba = vport->phba; 426 union lpfc_wqe128 *wqe; 427 struct lpfc_iocbq *genwqe; 428 struct ulp_bde64 *bpl; 429 struct ulp_bde64 bde; 430 int i, rc, xmit_len, first_len; 431 432 /* Allocate buffer for command WQE */ 433 genwqe = lpfc_sli_get_iocbq(phba); 434 if (genwqe == NULL) 435 return 1; 436 437 wqe = &genwqe->wqe; 438 memset(wqe, 0, sizeof(union lpfc_wqe)); 439 440 genwqe->context3 = (uint8_t *)bmp; 441 genwqe->iocb_flag |= LPFC_IO_NVME_LS; 442 443 /* Save for completion so we can release these resources */ 444 genwqe->context1 = lpfc_nlp_get(ndlp); 445 genwqe->context2 = (uint8_t *)pnvme_lsreq; 446 /* Fill in payload, bp points to frame payload */ 447 448 if (!tmo) 449 /* FC spec states we need 3 * ratov for CT requests */ 450 tmo = (3 * phba->fc_ratov); 451 452 /* For this command calculate the xmit length of the request bde. */ 453 xmit_len = 0; 454 first_len = 0; 455 bpl = (struct ulp_bde64 *)bmp->virt; 456 for (i = 0; i < num_entry; i++) { 457 bde.tus.w = bpl[i].tus.w; 458 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64) 459 break; 460 xmit_len += bde.tus.f.bdeSize; 461 if (i == 0) 462 first_len = xmit_len; 463 } 464 465 genwqe->rsvd2 = num_entry; 466 genwqe->hba_wqidx = 0; 467 468 /* Words 0 - 2 */ 469 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64; 470 wqe->generic.bde.tus.f.bdeSize = first_len; 471 wqe->generic.bde.addrLow = bpl[0].addrLow; 472 wqe->generic.bde.addrHigh = bpl[0].addrHigh; 473 474 /* Word 3 */ 475 wqe->gen_req.request_payload_len = first_len; 476 477 /* Word 4 */ 478 479 /* Word 5 */ 480 bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0); 481 bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1); 482 bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1); 483 bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ); 484 bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME); 485 486 /* Word 6 */ 487 bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com, 488 phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]); 489 bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag); 490 491 /* Word 7 */ 492 bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1)); 493 bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3); 494 bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE); 495 bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI); 496 497 /* Word 8 */ 498 wqe->gen_req.wqe_com.abort_tag = genwqe->iotag; 499 500 /* Word 9 */ 501 bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag); 502 503 /* Word 10 */ 504 bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1); 505 bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ); 506 bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1); 507 bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE); 508 bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0); 509 510 /* Word 11 */ 511 bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 512 bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND); 513 514 515 /* Issue GEN REQ WQE for NPORT <did> */ 516 lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS, 517 "6050 Issue GEN REQ WQE to NPORT x%x " 518 "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n", 519 ndlp->nlp_DID, genwqe->iotag, 520 vport->port_state, 521 genwqe, pnvme_lsreq, bmp, xmit_len, first_len); 522 genwqe->wqe_cmpl = cmpl; 523 genwqe->iocb_cmpl = NULL; 524 genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT; 525 genwqe->vport = vport; 526 genwqe->retry = retry; 527 528 lpfc_nvmeio_data(phba, "NVME LS XMIT: xri x%x iotag x%x to x%06x\n", 529 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID); 530 531 rc = lpfc_sli4_issue_wqe(phba, LPFC_ELS_RING, genwqe); 532 if (rc) { 533 lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS, 534 "6045 Issue GEN REQ WQE to NPORT x%x " 535 "Data: x%x x%x\n", 536 ndlp->nlp_DID, genwqe->iotag, 537 vport->port_state); 538 lpfc_sli_release_iocbq(phba, genwqe); 539 return 1; 540 } 541 return 0; 542 } 543 544 /** 545 * lpfc_nvme_ls_req - Issue an Link Service request 546 * @lpfc_pnvme: Pointer to the driver's nvme instance data 547 * @lpfc_nvme_lport: Pointer to the driver's local port data 548 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 549 * 550 * Driver registers this routine to handle any link service request 551 * from the nvme_fc transport to a remote nvme-aware port. 552 * 553 * Return value : 554 * 0 - Success 555 * TODO: What are the failure codes. 556 **/ 557 static int 558 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport, 559 struct nvme_fc_remote_port *pnvme_rport, 560 struct nvmefc_ls_req *pnvme_lsreq) 561 { 562 int ret = 0; 563 struct lpfc_nvme_lport *lport; 564 struct lpfc_nvme_rport *rport; 565 struct lpfc_vport *vport; 566 struct lpfc_nodelist *ndlp; 567 struct ulp_bde64 *bpl; 568 struct lpfc_dmabuf *bmp; 569 uint16_t ntype, nstate; 570 571 /* there are two dma buf in the request, actually there is one and 572 * the second one is just the start address + cmd size. 573 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped 574 * in a lpfc_dmabuf struct. When freeing we just free the wrapper 575 * because the nvem layer owns the data bufs. 576 * We do not have to break these packets open, we don't care what is in 577 * them. And we do not have to look at the resonse data, we only care 578 * that we got a response. All of the caring is going to happen in the 579 * nvme-fc layer. 580 */ 581 582 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 583 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 584 if (unlikely(!lport) || unlikely(!rport)) 585 return -EINVAL; 586 587 vport = lport->vport; 588 589 if (vport->load_flag & FC_UNLOADING) 590 return -ENODEV; 591 592 /* Need the ndlp. It is stored in the driver's rport. */ 593 ndlp = rport->ndlp; 594 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 595 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 596 "6051 Remoteport %p, rport has invalid ndlp. " 597 "Failing LS Req\n", pnvme_rport); 598 return -ENODEV; 599 } 600 601 /* The remote node has to be a mapped nvme target or an 602 * unmapped nvme initiator or it's an error. 603 */ 604 ntype = ndlp->nlp_type; 605 nstate = ndlp->nlp_state; 606 if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) || 607 (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) { 608 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 609 "6088 DID x%06x not ready for " 610 "IO. State x%x, Type x%x\n", 611 pnvme_rport->port_id, 612 ndlp->nlp_state, ndlp->nlp_type); 613 return -ENODEV; 614 } 615 bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL); 616 if (!bmp) { 617 618 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 619 "6044 Could not find node for DID %x\n", 620 pnvme_rport->port_id); 621 return 2; 622 } 623 INIT_LIST_HEAD(&bmp->list); 624 bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys)); 625 if (!bmp->virt) { 626 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 627 "6042 Could not find node for DID %x\n", 628 pnvme_rport->port_id); 629 kfree(bmp); 630 return 3; 631 } 632 bpl = (struct ulp_bde64 *)bmp->virt; 633 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma)); 634 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma)); 635 bpl->tus.f.bdeFlags = 0; 636 bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen; 637 bpl->tus.w = le32_to_cpu(bpl->tus.w); 638 bpl++; 639 640 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma)); 641 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma)); 642 bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I; 643 bpl->tus.f.bdeSize = pnvme_lsreq->rsplen; 644 bpl->tus.w = le32_to_cpu(bpl->tus.w); 645 646 /* Expand print to include key fields. */ 647 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 648 "6149 Issue LS Req to DID 0x%06x lport %p, rport %p " 649 "lsreq%p rqstlen:%d rsplen:%d %pad %pad\n", 650 ndlp->nlp_DID, 651 pnvme_lport, pnvme_rport, 652 pnvme_lsreq, pnvme_lsreq->rqstlen, 653 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma, 654 &pnvme_lsreq->rspdma); 655 656 atomic_inc(&lport->fc4NvmeLsRequests); 657 658 /* Hardcode the wait to 30 seconds. Connections are failing otherwise. 659 * This code allows it all to work. 660 */ 661 ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr, 662 pnvme_lsreq, lpfc_nvme_cmpl_gen_req, 663 ndlp, 2, 30, 0); 664 if (ret != WQE_SUCCESS) { 665 atomic_inc(&lport->xmt_ls_err); 666 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 667 "6052 EXIT. issue ls wqe failed lport %p, " 668 "rport %p lsreq%p Status %x DID %x\n", 669 pnvme_lport, pnvme_rport, pnvme_lsreq, 670 ret, ndlp->nlp_DID); 671 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys); 672 kfree(bmp); 673 return ret; 674 } 675 676 /* Stub in routine and return 0 for now. */ 677 return ret; 678 } 679 680 /** 681 * lpfc_nvme_ls_abort - Issue an Link Service request 682 * @lpfc_pnvme: Pointer to the driver's nvme instance data 683 * @lpfc_nvme_lport: Pointer to the driver's local port data 684 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 685 * 686 * Driver registers this routine to handle any link service request 687 * from the nvme_fc transport to a remote nvme-aware port. 688 * 689 * Return value : 690 * 0 - Success 691 * TODO: What are the failure codes. 692 **/ 693 static void 694 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport, 695 struct nvme_fc_remote_port *pnvme_rport, 696 struct nvmefc_ls_req *pnvme_lsreq) 697 { 698 struct lpfc_nvme_lport *lport; 699 struct lpfc_vport *vport; 700 struct lpfc_hba *phba; 701 struct lpfc_nodelist *ndlp; 702 LIST_HEAD(abort_list); 703 struct lpfc_sli_ring *pring; 704 struct lpfc_iocbq *wqe, *next_wqe; 705 706 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 707 if (unlikely(!lport)) 708 return; 709 vport = lport->vport; 710 phba = vport->phba; 711 712 if (vport->load_flag & FC_UNLOADING) 713 return; 714 715 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id); 716 if (!ndlp) { 717 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 718 "6049 Could not find node for DID %x\n", 719 pnvme_rport->port_id); 720 return; 721 } 722 723 /* Expand print to include key fields. */ 724 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 725 "6040 ENTER. lport %p, rport %p lsreq %p rqstlen:%d " 726 "rsplen:%d %pad %pad\n", 727 pnvme_lport, pnvme_rport, 728 pnvme_lsreq, pnvme_lsreq->rqstlen, 729 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma, 730 &pnvme_lsreq->rspdma); 731 732 /* 733 * Lock the ELS ring txcmplq and build a local list of all ELS IOs 734 * that need an ABTS. The IOs need to stay on the txcmplq so that 735 * the abort operation completes them successfully. 736 */ 737 pring = phba->sli4_hba.nvmels_wq->pring; 738 spin_lock_irq(&phba->hbalock); 739 spin_lock(&pring->ring_lock); 740 list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) { 741 /* Add to abort_list on on NDLP match. */ 742 if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) { 743 wqe->iocb_flag |= LPFC_DRIVER_ABORTED; 744 list_add_tail(&wqe->dlist, &abort_list); 745 } 746 } 747 spin_unlock(&pring->ring_lock); 748 spin_unlock_irq(&phba->hbalock); 749 750 /* Abort the targeted IOs and remove them from the abort list. */ 751 list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) { 752 atomic_inc(&lport->xmt_ls_abort); 753 spin_lock_irq(&phba->hbalock); 754 list_del_init(&wqe->dlist); 755 lpfc_sli_issue_abort_iotag(phba, pring, wqe); 756 spin_unlock_irq(&phba->hbalock); 757 } 758 } 759 760 /* Fix up the existing sgls for NVME IO. */ 761 static inline void 762 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport, 763 struct lpfc_nvme_buf *lpfc_ncmd, 764 struct nvmefc_fcp_req *nCmd) 765 { 766 struct lpfc_hba *phba = vport->phba; 767 struct sli4_sge *sgl; 768 union lpfc_wqe128 *wqe; 769 uint32_t *wptr, *dptr; 770 771 /* 772 * Get a local pointer to the built-in wqe and correct 773 * the cmd size to match NVME's 96 bytes and fix 774 * the dma address. 775 */ 776 777 wqe = &lpfc_ncmd->cur_iocbq.wqe; 778 779 /* 780 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to 781 * match NVME. NVME sends 96 bytes. Also, use the 782 * nvme commands command and response dma addresses 783 * rather than the virtual memory to ease the restore 784 * operation. 785 */ 786 sgl = lpfc_ncmd->nvme_sgl; 787 sgl->sge_len = cpu_to_le32(nCmd->cmdlen); 788 if (phba->cfg_nvme_embed_cmd) { 789 sgl->addr_hi = 0; 790 sgl->addr_lo = 0; 791 792 /* Word 0-2 - NVME CMND IU (embedded payload) */ 793 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED; 794 wqe->generic.bde.tus.f.bdeSize = 56; 795 wqe->generic.bde.addrHigh = 0; 796 wqe->generic.bde.addrLow = 64; /* Word 16 */ 797 798 /* Word 10 - dbde is 0, wqes is 1 in template */ 799 800 /* 801 * Embed the payload in the last half of the WQE 802 * WQE words 16-30 get the NVME CMD IU payload 803 * 804 * WQE words 16-19 get payload Words 1-4 805 * WQE words 20-21 get payload Words 6-7 806 * WQE words 22-29 get payload Words 16-23 807 */ 808 wptr = &wqe->words[16]; /* WQE ptr */ 809 dptr = (uint32_t *)nCmd->cmdaddr; /* payload ptr */ 810 dptr++; /* Skip Word 0 in payload */ 811 812 *wptr++ = *dptr++; /* Word 1 */ 813 *wptr++ = *dptr++; /* Word 2 */ 814 *wptr++ = *dptr++; /* Word 3 */ 815 *wptr++ = *dptr++; /* Word 4 */ 816 dptr++; /* Skip Word 5 in payload */ 817 *wptr++ = *dptr++; /* Word 6 */ 818 *wptr++ = *dptr++; /* Word 7 */ 819 dptr += 8; /* Skip Words 8-15 in payload */ 820 *wptr++ = *dptr++; /* Word 16 */ 821 *wptr++ = *dptr++; /* Word 17 */ 822 *wptr++ = *dptr++; /* Word 18 */ 823 *wptr++ = *dptr++; /* Word 19 */ 824 *wptr++ = *dptr++; /* Word 20 */ 825 *wptr++ = *dptr++; /* Word 21 */ 826 *wptr++ = *dptr++; /* Word 22 */ 827 *wptr = *dptr; /* Word 23 */ 828 } else { 829 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->cmddma)); 830 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->cmddma)); 831 832 /* Word 0-2 - NVME CMND IU Inline BDE */ 833 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64; 834 wqe->generic.bde.tus.f.bdeSize = nCmd->cmdlen; 835 wqe->generic.bde.addrHigh = sgl->addr_hi; 836 wqe->generic.bde.addrLow = sgl->addr_lo; 837 838 /* Word 10 */ 839 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1); 840 bf_set(wqe_wqes, &wqe->generic.wqe_com, 0); 841 } 842 843 sgl++; 844 845 /* Setup the physical region for the FCP RSP */ 846 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma)); 847 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma)); 848 sgl->word2 = le32_to_cpu(sgl->word2); 849 if (nCmd->sg_cnt) 850 bf_set(lpfc_sli4_sge_last, sgl, 0); 851 else 852 bf_set(lpfc_sli4_sge_last, sgl, 1); 853 sgl->word2 = cpu_to_le32(sgl->word2); 854 sgl->sge_len = cpu_to_le32(nCmd->rsplen); 855 } 856 857 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 858 static void 859 lpfc_nvme_ktime(struct lpfc_hba *phba, 860 struct lpfc_nvme_buf *lpfc_ncmd) 861 { 862 uint64_t seg1, seg2, seg3, seg4; 863 uint64_t segsum; 864 865 if (!lpfc_ncmd->ts_last_cmd || 866 !lpfc_ncmd->ts_cmd_start || 867 !lpfc_ncmd->ts_cmd_wqput || 868 !lpfc_ncmd->ts_isr_cmpl || 869 !lpfc_ncmd->ts_data_nvme) 870 return; 871 872 if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_cmd_start) 873 return; 874 if (lpfc_ncmd->ts_cmd_start < lpfc_ncmd->ts_last_cmd) 875 return; 876 if (lpfc_ncmd->ts_cmd_wqput < lpfc_ncmd->ts_cmd_start) 877 return; 878 if (lpfc_ncmd->ts_isr_cmpl < lpfc_ncmd->ts_cmd_wqput) 879 return; 880 if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_isr_cmpl) 881 return; 882 /* 883 * Segment 1 - Time from Last FCP command cmpl is handed 884 * off to NVME Layer to start of next command. 885 * Segment 2 - Time from Driver receives a IO cmd start 886 * from NVME Layer to WQ put is done on IO cmd. 887 * Segment 3 - Time from Driver WQ put is done on IO cmd 888 * to MSI-X ISR for IO cmpl. 889 * Segment 4 - Time from MSI-X ISR for IO cmpl to when 890 * cmpl is handled off to the NVME Layer. 891 */ 892 seg1 = lpfc_ncmd->ts_cmd_start - lpfc_ncmd->ts_last_cmd; 893 if (seg1 > 5000000) /* 5 ms - for sequential IOs only */ 894 seg1 = 0; 895 896 /* Calculate times relative to start of IO */ 897 seg2 = (lpfc_ncmd->ts_cmd_wqput - lpfc_ncmd->ts_cmd_start); 898 segsum = seg2; 899 seg3 = lpfc_ncmd->ts_isr_cmpl - lpfc_ncmd->ts_cmd_start; 900 if (segsum > seg3) 901 return; 902 seg3 -= segsum; 903 segsum += seg3; 904 905 seg4 = lpfc_ncmd->ts_data_nvme - lpfc_ncmd->ts_cmd_start; 906 if (segsum > seg4) 907 return; 908 seg4 -= segsum; 909 910 phba->ktime_data_samples++; 911 phba->ktime_seg1_total += seg1; 912 if (seg1 < phba->ktime_seg1_min) 913 phba->ktime_seg1_min = seg1; 914 else if (seg1 > phba->ktime_seg1_max) 915 phba->ktime_seg1_max = seg1; 916 phba->ktime_seg2_total += seg2; 917 if (seg2 < phba->ktime_seg2_min) 918 phba->ktime_seg2_min = seg2; 919 else if (seg2 > phba->ktime_seg2_max) 920 phba->ktime_seg2_max = seg2; 921 phba->ktime_seg3_total += seg3; 922 if (seg3 < phba->ktime_seg3_min) 923 phba->ktime_seg3_min = seg3; 924 else if (seg3 > phba->ktime_seg3_max) 925 phba->ktime_seg3_max = seg3; 926 phba->ktime_seg4_total += seg4; 927 if (seg4 < phba->ktime_seg4_min) 928 phba->ktime_seg4_min = seg4; 929 else if (seg4 > phba->ktime_seg4_max) 930 phba->ktime_seg4_max = seg4; 931 932 lpfc_ncmd->ts_last_cmd = 0; 933 lpfc_ncmd->ts_cmd_start = 0; 934 lpfc_ncmd->ts_cmd_wqput = 0; 935 lpfc_ncmd->ts_isr_cmpl = 0; 936 lpfc_ncmd->ts_data_nvme = 0; 937 } 938 #endif 939 940 /** 941 * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO 942 * @lpfc_pnvme: Pointer to the driver's nvme instance data 943 * @lpfc_nvme_lport: Pointer to the driver's local port data 944 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 945 * 946 * Driver registers this routine as it io request handler. This 947 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 948 * data structure to the rport indicated in @lpfc_nvme_rport. 949 * 950 * Return value : 951 * 0 - Success 952 * TODO: What are the failure codes. 953 **/ 954 static void 955 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn, 956 struct lpfc_wcqe_complete *wcqe) 957 { 958 struct lpfc_nvme_buf *lpfc_ncmd = 959 (struct lpfc_nvme_buf *)pwqeIn->context1; 960 struct lpfc_vport *vport = pwqeIn->vport; 961 struct nvmefc_fcp_req *nCmd; 962 struct nvme_fc_ersp_iu *ep; 963 struct nvme_fc_cmd_iu *cp; 964 struct lpfc_nvme_rport *rport; 965 struct lpfc_nodelist *ndlp; 966 struct lpfc_nvme_fcpreq_priv *freqpriv; 967 struct lpfc_nvme_lport *lport; 968 struct lpfc_nvme_ctrl_stat *cstat; 969 unsigned long flags; 970 uint32_t code, status, idx; 971 uint16_t cid, sqhd, data; 972 uint32_t *ptr; 973 974 /* Sanity check on return of outstanding command */ 975 if (!lpfc_ncmd || !lpfc_ncmd->nvmeCmd || !lpfc_ncmd->nrport) { 976 if (!lpfc_ncmd) { 977 lpfc_printf_vlog(vport, KERN_ERR, 978 LOG_NODE | LOG_NVME_IOERR, 979 "6071 Null lpfc_ncmd pointer. No " 980 "release, skip completion\n"); 981 return; 982 } 983 984 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 985 "6066 Missing cmpl ptrs: lpfc_ncmd %p, " 986 "nvmeCmd %p nrport %p\n", 987 lpfc_ncmd, lpfc_ncmd->nvmeCmd, 988 lpfc_ncmd->nrport); 989 990 /* Release the lpfc_ncmd regardless of the missing elements. */ 991 lpfc_release_nvme_buf(phba, lpfc_ncmd); 992 return; 993 } 994 nCmd = lpfc_ncmd->nvmeCmd; 995 rport = lpfc_ncmd->nrport; 996 status = bf_get(lpfc_wcqe_c_status, wcqe); 997 998 if (vport->localport) { 999 lport = (struct lpfc_nvme_lport *)vport->localport->private; 1000 if (lport) { 1001 idx = lpfc_ncmd->cur_iocbq.hba_wqidx; 1002 cstat = &lport->cstat[idx]; 1003 atomic_inc(&cstat->fc4NvmeIoCmpls); 1004 if (status) { 1005 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 1006 atomic_inc(&lport->cmpl_fcp_xb); 1007 atomic_inc(&lport->cmpl_fcp_err); 1008 } 1009 } 1010 } 1011 1012 lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n", 1013 lpfc_ncmd->cur_iocbq.sli4_xritag, 1014 status, wcqe->parameter); 1015 /* 1016 * Catch race where our node has transitioned, but the 1017 * transport is still transitioning. 1018 */ 1019 ndlp = rport->ndlp; 1020 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 1021 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 1022 "6061 rport %p, DID x%06x node not ready.\n", 1023 rport, rport->remoteport->port_id); 1024 1025 ndlp = lpfc_findnode_did(vport, rport->remoteport->port_id); 1026 if (!ndlp) { 1027 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 1028 "6062 Ignoring NVME cmpl. No ndlp\n"); 1029 goto out_err; 1030 } 1031 } 1032 1033 code = bf_get(lpfc_wcqe_c_code, wcqe); 1034 if (code == CQE_CODE_NVME_ERSP) { 1035 /* For this type of CQE, we need to rebuild the rsp */ 1036 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr; 1037 1038 /* 1039 * Get Command Id from cmd to plug into response. This 1040 * code is not needed in the next NVME Transport drop. 1041 */ 1042 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr; 1043 cid = cp->sqe.common.command_id; 1044 1045 /* 1046 * RSN is in CQE word 2 1047 * SQHD is in CQE Word 3 bits 15:0 1048 * Cmd Specific info is in CQE Word 1 1049 * and in CQE Word 0 bits 15:0 1050 */ 1051 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe); 1052 1053 /* Now lets build the NVME ERSP IU */ 1054 ep->iu_len = cpu_to_be16(8); 1055 ep->rsn = wcqe->parameter; 1056 ep->xfrd_len = cpu_to_be32(nCmd->payload_length); 1057 ep->rsvd12 = 0; 1058 ptr = (uint32_t *)&ep->cqe.result.u64; 1059 *ptr++ = wcqe->total_data_placed; 1060 data = bf_get(lpfc_wcqe_c_ersp0, wcqe); 1061 *ptr = (uint32_t)data; 1062 ep->cqe.sq_head = sqhd; 1063 ep->cqe.sq_id = nCmd->sqid; 1064 ep->cqe.command_id = cid; 1065 ep->cqe.status = 0; 1066 1067 lpfc_ncmd->status = IOSTAT_SUCCESS; 1068 lpfc_ncmd->result = 0; 1069 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN; 1070 nCmd->transferred_length = nCmd->payload_length; 1071 } else { 1072 lpfc_ncmd->status = (status & LPFC_IOCB_STATUS_MASK); 1073 lpfc_ncmd->result = (wcqe->parameter & IOERR_PARAM_MASK); 1074 1075 /* For NVME, the only failure path that results in an 1076 * IO error is when the adapter rejects it. All other 1077 * conditions are a success case and resolved by the 1078 * transport. 1079 * IOSTAT_FCP_RSP_ERROR means: 1080 * 1. Length of data received doesn't match total 1081 * transfer length in WQE 1082 * 2. If the RSP payload does NOT match these cases: 1083 * a. RSP length 12/24 bytes and all zeros 1084 * b. NVME ERSP 1085 */ 1086 switch (lpfc_ncmd->status) { 1087 case IOSTAT_SUCCESS: 1088 nCmd->transferred_length = wcqe->total_data_placed; 1089 nCmd->rcv_rsplen = 0; 1090 nCmd->status = 0; 1091 break; 1092 case IOSTAT_FCP_RSP_ERROR: 1093 nCmd->transferred_length = wcqe->total_data_placed; 1094 nCmd->rcv_rsplen = wcqe->parameter; 1095 nCmd->status = 0; 1096 /* Sanity check */ 1097 if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) 1098 break; 1099 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 1100 "6081 NVME Completion Protocol Error: " 1101 "xri %x status x%x result x%x " 1102 "placed x%x\n", 1103 lpfc_ncmd->cur_iocbq.sli4_xritag, 1104 lpfc_ncmd->status, lpfc_ncmd->result, 1105 wcqe->total_data_placed); 1106 break; 1107 case IOSTAT_LOCAL_REJECT: 1108 /* Let fall through to set command final state. */ 1109 if (lpfc_ncmd->result == IOERR_ABORT_REQUESTED) 1110 lpfc_printf_vlog(vport, KERN_INFO, 1111 LOG_NVME_IOERR, 1112 "6032 Delay Aborted cmd %p " 1113 "nvme cmd %p, xri x%x, " 1114 "xb %d\n", 1115 lpfc_ncmd, nCmd, 1116 lpfc_ncmd->cur_iocbq.sli4_xritag, 1117 bf_get(lpfc_wcqe_c_xb, wcqe)); 1118 default: 1119 out_err: 1120 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1121 "6072 NVME Completion Error: xri %x " 1122 "status x%x result x%x placed x%x\n", 1123 lpfc_ncmd->cur_iocbq.sli4_xritag, 1124 lpfc_ncmd->status, lpfc_ncmd->result, 1125 wcqe->total_data_placed); 1126 nCmd->transferred_length = 0; 1127 nCmd->rcv_rsplen = 0; 1128 nCmd->status = NVME_SC_INTERNAL; 1129 } 1130 } 1131 1132 /* pick up SLI4 exhange busy condition */ 1133 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 1134 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 1135 else 1136 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 1137 1138 if (ndlp && NLP_CHK_NODE_ACT(ndlp)) 1139 atomic_dec(&ndlp->cmd_pending); 1140 1141 /* Update stats and complete the IO. There is 1142 * no need for dma unprep because the nvme_transport 1143 * owns the dma address. 1144 */ 1145 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1146 if (lpfc_ncmd->ts_cmd_start) { 1147 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp; 1148 lpfc_ncmd->ts_data_nvme = ktime_get_ns(); 1149 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme; 1150 lpfc_nvme_ktime(phba, lpfc_ncmd); 1151 } 1152 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1153 if (lpfc_ncmd->cpu != smp_processor_id()) 1154 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 1155 "6701 CPU Check cmpl: " 1156 "cpu %d expect %d\n", 1157 smp_processor_id(), lpfc_ncmd->cpu); 1158 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT) 1159 phba->cpucheck_cmpl_io[lpfc_ncmd->cpu]++; 1160 } 1161 #endif 1162 1163 /* NVME targets need completion held off until the abort exchange 1164 * completes unless the NVME Rport is getting unregistered. 1165 */ 1166 1167 if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) { 1168 freqpriv = nCmd->private; 1169 freqpriv->nvme_buf = NULL; 1170 nCmd->done(nCmd); 1171 lpfc_ncmd->nvmeCmd = NULL; 1172 } 1173 1174 spin_lock_irqsave(&phba->hbalock, flags); 1175 lpfc_ncmd->nrport = NULL; 1176 spin_unlock_irqrestore(&phba->hbalock, flags); 1177 1178 /* Call release with XB=1 to queue the IO into the abort list. */ 1179 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1180 } 1181 1182 1183 /** 1184 * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO 1185 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1186 * @lpfc_nvme_lport: Pointer to the driver's local port data 1187 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1188 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1189 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1190 * 1191 * Driver registers this routine as it io request handler. This 1192 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1193 * data structure to the rport indicated in @lpfc_nvme_rport. 1194 * 1195 * Return value : 1196 * 0 - Success 1197 * TODO: What are the failure codes. 1198 **/ 1199 static int 1200 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport, 1201 struct lpfc_nvme_buf *lpfc_ncmd, 1202 struct lpfc_nodelist *pnode, 1203 struct lpfc_nvme_ctrl_stat *cstat) 1204 { 1205 struct lpfc_hba *phba = vport->phba; 1206 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1207 struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq); 1208 union lpfc_wqe128 *wqe = &pwqeq->wqe; 1209 uint32_t req_len; 1210 1211 if (!pnode || !NLP_CHK_NODE_ACT(pnode)) 1212 return -EINVAL; 1213 1214 /* 1215 * There are three possibilities here - use scatter-gather segment, use 1216 * the single mapping, or neither. 1217 */ 1218 if (nCmd->sg_cnt) { 1219 if (nCmd->io_dir == NVMEFC_FCP_WRITE) { 1220 /* From the iwrite template, initialize words 7 - 11 */ 1221 memcpy(&wqe->words[7], 1222 &lpfc_iwrite_cmd_template.words[7], 1223 sizeof(uint32_t) * 5); 1224 1225 /* Word 4 */ 1226 wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length; 1227 1228 /* Word 5 */ 1229 if ((phba->cfg_nvme_enable_fb) && 1230 (pnode->nlp_flag & NLP_FIRSTBURST)) { 1231 req_len = lpfc_ncmd->nvmeCmd->payload_length; 1232 if (req_len < pnode->nvme_fb_size) 1233 wqe->fcp_iwrite.initial_xfer_len = 1234 req_len; 1235 else 1236 wqe->fcp_iwrite.initial_xfer_len = 1237 pnode->nvme_fb_size; 1238 } else { 1239 wqe->fcp_iwrite.initial_xfer_len = 0; 1240 } 1241 atomic_inc(&cstat->fc4NvmeOutputRequests); 1242 } else { 1243 /* From the iread template, initialize words 7 - 11 */ 1244 memcpy(&wqe->words[7], 1245 &lpfc_iread_cmd_template.words[7], 1246 sizeof(uint32_t) * 5); 1247 1248 /* Word 4 */ 1249 wqe->fcp_iread.total_xfer_len = nCmd->payload_length; 1250 1251 /* Word 5 */ 1252 wqe->fcp_iread.rsrvd5 = 0; 1253 1254 atomic_inc(&cstat->fc4NvmeInputRequests); 1255 } 1256 } else { 1257 /* From the icmnd template, initialize words 4 - 11 */ 1258 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4], 1259 sizeof(uint32_t) * 8); 1260 atomic_inc(&cstat->fc4NvmeControlRequests); 1261 } 1262 /* 1263 * Finish initializing those WQE fields that are independent 1264 * of the nvme_cmnd request_buffer 1265 */ 1266 1267 /* Word 3 */ 1268 bf_set(payload_offset_len, &wqe->fcp_icmd, 1269 (nCmd->rsplen + nCmd->cmdlen)); 1270 1271 /* Word 6 */ 1272 bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com, 1273 phba->sli4_hba.rpi_ids[pnode->nlp_rpi]); 1274 bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag); 1275 1276 /* Word 8 */ 1277 wqe->generic.wqe_com.abort_tag = pwqeq->iotag; 1278 1279 /* Word 9 */ 1280 bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag); 1281 1282 pwqeq->vport = vport; 1283 return 0; 1284 } 1285 1286 1287 /** 1288 * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO 1289 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1290 * @lpfc_nvme_lport: Pointer to the driver's local port data 1291 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1292 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1293 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1294 * 1295 * Driver registers this routine as it io request handler. This 1296 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1297 * data structure to the rport indicated in @lpfc_nvme_rport. 1298 * 1299 * Return value : 1300 * 0 - Success 1301 * TODO: What are the failure codes. 1302 **/ 1303 static int 1304 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport, 1305 struct lpfc_nvme_buf *lpfc_ncmd) 1306 { 1307 struct lpfc_hba *phba = vport->phba; 1308 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1309 union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe; 1310 struct sli4_sge *sgl = lpfc_ncmd->nvme_sgl; 1311 struct scatterlist *data_sg; 1312 struct sli4_sge *first_data_sgl; 1313 struct ulp_bde64 *bde; 1314 dma_addr_t physaddr; 1315 uint32_t num_bde = 0; 1316 uint32_t dma_len; 1317 uint32_t dma_offset = 0; 1318 int nseg, i; 1319 1320 /* Fix up the command and response DMA stuff. */ 1321 lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd); 1322 1323 /* 1324 * There are three possibilities here - use scatter-gather segment, use 1325 * the single mapping, or neither. 1326 */ 1327 if (nCmd->sg_cnt) { 1328 /* 1329 * Jump over the cmd and rsp SGEs. The fix routine 1330 * has already adjusted for this. 1331 */ 1332 sgl += 2; 1333 1334 first_data_sgl = sgl; 1335 lpfc_ncmd->seg_cnt = nCmd->sg_cnt; 1336 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) { 1337 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1338 "6058 Too many sg segments from " 1339 "NVME Transport. Max %d, " 1340 "nvmeIO sg_cnt %d\n", 1341 phba->cfg_nvme_seg_cnt + 1, 1342 lpfc_ncmd->seg_cnt); 1343 lpfc_ncmd->seg_cnt = 0; 1344 return 1; 1345 } 1346 1347 /* 1348 * The driver established a maximum scatter-gather segment count 1349 * during probe that limits the number of sg elements in any 1350 * single nvme command. Just run through the seg_cnt and format 1351 * the sge's. 1352 */ 1353 nseg = nCmd->sg_cnt; 1354 data_sg = nCmd->first_sgl; 1355 for (i = 0; i < nseg; i++) { 1356 if (data_sg == NULL) { 1357 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1358 "6059 dptr err %d, nseg %d\n", 1359 i, nseg); 1360 lpfc_ncmd->seg_cnt = 0; 1361 return 1; 1362 } 1363 physaddr = data_sg->dma_address; 1364 dma_len = data_sg->length; 1365 sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr)); 1366 sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr)); 1367 sgl->word2 = le32_to_cpu(sgl->word2); 1368 if ((num_bde + 1) == nseg) 1369 bf_set(lpfc_sli4_sge_last, sgl, 1); 1370 else 1371 bf_set(lpfc_sli4_sge_last, sgl, 0); 1372 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset); 1373 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA); 1374 sgl->word2 = cpu_to_le32(sgl->word2); 1375 sgl->sge_len = cpu_to_le32(dma_len); 1376 1377 dma_offset += dma_len; 1378 data_sg = sg_next(data_sg); 1379 sgl++; 1380 } 1381 if (phba->nvme_embed_pbde) { 1382 /* Use PBDE support for first SGL only, offset == 0 */ 1383 /* Words 13-15 */ 1384 bde = (struct ulp_bde64 *) 1385 &wqe->words[13]; 1386 bde->addrLow = first_data_sgl->addr_lo; 1387 bde->addrHigh = first_data_sgl->addr_hi; 1388 bde->tus.f.bdeSize = 1389 le32_to_cpu(first_data_sgl->sge_len); 1390 bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64; 1391 bde->tus.w = cpu_to_le32(bde->tus.w); 1392 /* wqe_pbde is 1 in template */ 1393 } else { 1394 memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3)); 1395 bf_set(wqe_pbde, &wqe->generic.wqe_com, 0); 1396 } 1397 } else { 1398 bf_set(wqe_pbde, &wqe->generic.wqe_com, 0); 1399 memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3)); 1400 1401 /* For this clause to be valid, the payload_length 1402 * and sg_cnt must zero. 1403 */ 1404 if (nCmd->payload_length != 0) { 1405 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1406 "6063 NVME DMA Prep Err: sg_cnt %d " 1407 "payload_length x%x\n", 1408 nCmd->sg_cnt, nCmd->payload_length); 1409 return 1; 1410 } 1411 } 1412 return 0; 1413 } 1414 1415 /** 1416 * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO 1417 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1418 * @lpfc_nvme_lport: Pointer to the driver's local port data 1419 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1420 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1421 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1422 * 1423 * Driver registers this routine as it io request handler. This 1424 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1425 * data structure to the rport 1426 indicated in @lpfc_nvme_rport. 1427 * 1428 * Return value : 1429 * 0 - Success 1430 * TODO: What are the failure codes. 1431 **/ 1432 static int 1433 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport, 1434 struct nvme_fc_remote_port *pnvme_rport, 1435 void *hw_queue_handle, 1436 struct nvmefc_fcp_req *pnvme_fcreq) 1437 { 1438 int ret = 0; 1439 int expedite = 0; 1440 int idx; 1441 struct lpfc_nvme_lport *lport; 1442 struct lpfc_nvme_ctrl_stat *cstat; 1443 struct lpfc_vport *vport; 1444 struct lpfc_hba *phba; 1445 struct lpfc_nodelist *ndlp; 1446 struct lpfc_nvme_buf *lpfc_ncmd; 1447 struct lpfc_nvme_rport *rport; 1448 struct lpfc_nvme_qhandle *lpfc_queue_info; 1449 struct lpfc_nvme_fcpreq_priv *freqpriv; 1450 struct nvme_common_command *sqe; 1451 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1452 uint64_t start = 0; 1453 #endif 1454 1455 /* Validate pointers. LLDD fault handling with transport does 1456 * have timing races. 1457 */ 1458 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1459 if (unlikely(!lport)) { 1460 ret = -EINVAL; 1461 goto out_fail; 1462 } 1463 1464 vport = lport->vport; 1465 1466 if (unlikely(!hw_queue_handle)) { 1467 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1468 "6117 Fail IO, NULL hw_queue_handle\n"); 1469 atomic_inc(&lport->xmt_fcp_err); 1470 ret = -EBUSY; 1471 goto out_fail; 1472 } 1473 1474 phba = vport->phba; 1475 1476 if (vport->load_flag & FC_UNLOADING) { 1477 ret = -ENODEV; 1478 goto out_fail; 1479 } 1480 1481 if (vport->load_flag & FC_UNLOADING) { 1482 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1483 "6124 Fail IO, Driver unload\n"); 1484 atomic_inc(&lport->xmt_fcp_err); 1485 ret = -ENODEV; 1486 goto out_fail; 1487 } 1488 1489 freqpriv = pnvme_fcreq->private; 1490 if (unlikely(!freqpriv)) { 1491 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1492 "6158 Fail IO, NULL request data\n"); 1493 atomic_inc(&lport->xmt_fcp_err); 1494 ret = -EINVAL; 1495 goto out_fail; 1496 } 1497 1498 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1499 if (phba->ktime_on) 1500 start = ktime_get_ns(); 1501 #endif 1502 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 1503 lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle; 1504 1505 /* 1506 * Catch race where our node has transitioned, but the 1507 * transport is still transitioning. 1508 */ 1509 ndlp = rport->ndlp; 1510 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 1511 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1512 "6053 Fail IO, ndlp not ready: rport %p " 1513 "ndlp %p, DID x%06x\n", 1514 rport, ndlp, pnvme_rport->port_id); 1515 atomic_inc(&lport->xmt_fcp_err); 1516 ret = -EBUSY; 1517 goto out_fail; 1518 } 1519 1520 /* The remote node has to be a mapped target or it's an error. */ 1521 if ((ndlp->nlp_type & NLP_NVME_TARGET) && 1522 (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) { 1523 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1524 "6036 Fail IO, DID x%06x not ready for " 1525 "IO. State x%x, Type x%x Flg x%x\n", 1526 pnvme_rport->port_id, 1527 ndlp->nlp_state, ndlp->nlp_type, 1528 ndlp->upcall_flags); 1529 atomic_inc(&lport->xmt_fcp_bad_ndlp); 1530 ret = -EBUSY; 1531 goto out_fail; 1532 1533 } 1534 1535 /* Currently only NVME Keep alive commands should be expedited 1536 * if the driver runs out of a resource. These should only be 1537 * issued on the admin queue, qidx 0 1538 */ 1539 if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) { 1540 sqe = &((struct nvme_fc_cmd_iu *) 1541 pnvme_fcreq->cmdaddr)->sqe.common; 1542 if (sqe->opcode == nvme_admin_keep_alive) 1543 expedite = 1; 1544 } 1545 1546 /* The node is shared with FCP IO, make sure the IO pending count does 1547 * not exceed the programmed depth. 1548 */ 1549 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) && 1550 !expedite) { 1551 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1552 "6174 Fail IO, ndlp qdepth exceeded: " 1553 "idx %d DID %x pend %d qdepth %d\n", 1554 lpfc_queue_info->index, ndlp->nlp_DID, 1555 atomic_read(&ndlp->cmd_pending), 1556 ndlp->cmd_qdepth); 1557 atomic_inc(&lport->xmt_fcp_qdepth); 1558 ret = -EBUSY; 1559 goto out_fail; 1560 } 1561 1562 lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, expedite); 1563 if (lpfc_ncmd == NULL) { 1564 atomic_inc(&lport->xmt_fcp_noxri); 1565 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1566 "6065 Fail IO, driver buffer pool is empty: " 1567 "idx %d DID %x\n", 1568 lpfc_queue_info->index, ndlp->nlp_DID); 1569 ret = -EBUSY; 1570 goto out_fail; 1571 } 1572 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1573 if (start) { 1574 lpfc_ncmd->ts_cmd_start = start; 1575 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd; 1576 } else { 1577 lpfc_ncmd->ts_cmd_start = 0; 1578 } 1579 #endif 1580 1581 /* 1582 * Store the data needed by the driver to issue, abort, and complete 1583 * an IO. 1584 * Do not let the IO hang out forever. There is no midlayer issuing 1585 * an abort so inform the FW of the maximum IO pending time. 1586 */ 1587 freqpriv->nvme_buf = lpfc_ncmd; 1588 lpfc_ncmd->nvmeCmd = pnvme_fcreq; 1589 lpfc_ncmd->nrport = rport; 1590 lpfc_ncmd->ndlp = ndlp; 1591 lpfc_ncmd->start_time = jiffies; 1592 1593 /* 1594 * Issue the IO on the WQ indicated by index in the hw_queue_handle. 1595 * This identfier was create in our hardware queue create callback 1596 * routine. The driver now is dependent on the IO queue steering from 1597 * the transport. We are trusting the upper NVME layers know which 1598 * index to use and that they have affinitized a CPU to this hardware 1599 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ. 1600 */ 1601 idx = lpfc_queue_info->index; 1602 lpfc_ncmd->cur_iocbq.hba_wqidx = idx; 1603 cstat = &lport->cstat[idx]; 1604 1605 lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat); 1606 ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd); 1607 if (ret) { 1608 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1609 "6175 Fail IO, Prep DMA: " 1610 "idx %d DID %x\n", 1611 lpfc_queue_info->index, ndlp->nlp_DID); 1612 atomic_inc(&lport->xmt_fcp_err); 1613 ret = -ENOMEM; 1614 goto out_free_nvme_buf; 1615 } 1616 1617 atomic_inc(&ndlp->cmd_pending); 1618 1619 lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n", 1620 lpfc_ncmd->cur_iocbq.sli4_xritag, 1621 lpfc_queue_info->index, ndlp->nlp_DID); 1622 1623 ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq); 1624 if (ret) { 1625 atomic_inc(&lport->xmt_fcp_wqerr); 1626 atomic_dec(&ndlp->cmd_pending); 1627 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1628 "6113 Fail IO, Could not issue WQE err %x " 1629 "sid: x%x did: x%x oxid: x%x\n", 1630 ret, vport->fc_myDID, ndlp->nlp_DID, 1631 lpfc_ncmd->cur_iocbq.sli4_xritag); 1632 goto out_free_nvme_buf; 1633 } 1634 1635 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1636 if (lpfc_ncmd->ts_cmd_start) 1637 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns(); 1638 1639 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1640 lpfc_ncmd->cpu = smp_processor_id(); 1641 if (lpfc_ncmd->cpu != lpfc_queue_info->index) { 1642 /* Check for admin queue */ 1643 if (lpfc_queue_info->qidx) { 1644 lpfc_printf_vlog(vport, 1645 KERN_ERR, LOG_NVME_IOERR, 1646 "6702 CPU Check cmd: " 1647 "cpu %d wq %d\n", 1648 lpfc_ncmd->cpu, 1649 lpfc_queue_info->index); 1650 } 1651 lpfc_ncmd->cpu = lpfc_queue_info->index; 1652 } 1653 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT) 1654 phba->cpucheck_xmt_io[lpfc_ncmd->cpu]++; 1655 } 1656 #endif 1657 return 0; 1658 1659 out_free_nvme_buf: 1660 if (lpfc_ncmd->nvmeCmd->sg_cnt) { 1661 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE) 1662 atomic_dec(&cstat->fc4NvmeOutputRequests); 1663 else 1664 atomic_dec(&cstat->fc4NvmeInputRequests); 1665 } else 1666 atomic_dec(&cstat->fc4NvmeControlRequests); 1667 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1668 out_fail: 1669 return ret; 1670 } 1671 1672 /** 1673 * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request. 1674 * @phba: Pointer to HBA context object 1675 * @cmdiocb: Pointer to command iocb object. 1676 * @rspiocb: Pointer to response iocb object. 1677 * 1678 * This is the callback function for any NVME FCP IO that was aborted. 1679 * 1680 * Return value: 1681 * None 1682 **/ 1683 void 1684 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb, 1685 struct lpfc_wcqe_complete *abts_cmpl) 1686 { 1687 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 1688 "6145 ABORT_XRI_CN completing on rpi x%x " 1689 "original iotag x%x, abort cmd iotag x%x " 1690 "req_tag x%x, status x%x, hwstatus x%x\n", 1691 cmdiocb->iocb.un.acxri.abortContextTag, 1692 cmdiocb->iocb.un.acxri.abortIoTag, 1693 cmdiocb->iotag, 1694 bf_get(lpfc_wcqe_c_request_tag, abts_cmpl), 1695 bf_get(lpfc_wcqe_c_status, abts_cmpl), 1696 bf_get(lpfc_wcqe_c_hw_status, abts_cmpl)); 1697 lpfc_sli_release_iocbq(phba, cmdiocb); 1698 } 1699 1700 /** 1701 * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS 1702 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1703 * @lpfc_nvme_lport: Pointer to the driver's local port data 1704 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1705 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1706 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1707 * 1708 * Driver registers this routine as its nvme request io abort handler. This 1709 * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq 1710 * data structure to the rport indicated in @lpfc_nvme_rport. This routine 1711 * is executed asynchronously - one the target is validated as "MAPPED" and 1712 * ready for IO, the driver issues the abort request and returns. 1713 * 1714 * Return value: 1715 * None 1716 **/ 1717 static void 1718 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport, 1719 struct nvme_fc_remote_port *pnvme_rport, 1720 void *hw_queue_handle, 1721 struct nvmefc_fcp_req *pnvme_fcreq) 1722 { 1723 struct lpfc_nvme_lport *lport; 1724 struct lpfc_vport *vport; 1725 struct lpfc_hba *phba; 1726 struct lpfc_nvme_buf *lpfc_nbuf; 1727 struct lpfc_iocbq *abts_buf; 1728 struct lpfc_iocbq *nvmereq_wqe; 1729 struct lpfc_nvme_fcpreq_priv *freqpriv; 1730 union lpfc_wqe128 *abts_wqe; 1731 unsigned long flags; 1732 int ret_val; 1733 1734 /* Validate pointers. LLDD fault handling with transport does 1735 * have timing races. 1736 */ 1737 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1738 if (unlikely(!lport)) 1739 return; 1740 1741 vport = lport->vport; 1742 1743 if (unlikely(!hw_queue_handle)) { 1744 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1745 "6129 Fail Abort, HW Queue Handle NULL.\n"); 1746 return; 1747 } 1748 1749 phba = vport->phba; 1750 freqpriv = pnvme_fcreq->private; 1751 1752 if (unlikely(!freqpriv)) 1753 return; 1754 if (vport->load_flag & FC_UNLOADING) 1755 return; 1756 1757 /* Announce entry to new IO submit field. */ 1758 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1759 "6002 Abort Request to rport DID x%06x " 1760 "for nvme_fc_req %p\n", 1761 pnvme_rport->port_id, 1762 pnvme_fcreq); 1763 1764 /* If the hba is getting reset, this flag is set. It is 1765 * cleared when the reset is complete and rings reestablished. 1766 */ 1767 spin_lock_irqsave(&phba->hbalock, flags); 1768 /* driver queued commands are in process of being flushed */ 1769 if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) { 1770 spin_unlock_irqrestore(&phba->hbalock, flags); 1771 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1772 "6139 Driver in reset cleanup - flushing " 1773 "NVME Req now. hba_flag x%x\n", 1774 phba->hba_flag); 1775 return; 1776 } 1777 1778 lpfc_nbuf = freqpriv->nvme_buf; 1779 if (!lpfc_nbuf) { 1780 spin_unlock_irqrestore(&phba->hbalock, flags); 1781 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1782 "6140 NVME IO req has no matching lpfc nvme " 1783 "io buffer. Skipping abort req.\n"); 1784 return; 1785 } else if (!lpfc_nbuf->nvmeCmd) { 1786 spin_unlock_irqrestore(&phba->hbalock, flags); 1787 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1788 "6141 lpfc NVME IO req has no nvme_fcreq " 1789 "io buffer. Skipping abort req.\n"); 1790 return; 1791 } 1792 nvmereq_wqe = &lpfc_nbuf->cur_iocbq; 1793 1794 /* 1795 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's 1796 * state must match the nvme_fcreq passed by the nvme 1797 * transport. If they don't match, it is likely the driver 1798 * has already completed the NVME IO and the nvme transport 1799 * has not seen it yet. 1800 */ 1801 if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) { 1802 spin_unlock_irqrestore(&phba->hbalock, flags); 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 return; 1810 } 1811 1812 /* Don't abort IOs no longer on the pending queue. */ 1813 if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) { 1814 spin_unlock_irqrestore(&phba->hbalock, flags); 1815 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1816 "6142 NVME IO req %p not queued - skipping " 1817 "abort req xri x%x\n", 1818 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1819 return; 1820 } 1821 1822 atomic_inc(&lport->xmt_fcp_abort); 1823 lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n", 1824 nvmereq_wqe->sli4_xritag, 1825 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id); 1826 1827 /* Outstanding abort is in progress */ 1828 if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) { 1829 spin_unlock_irqrestore(&phba->hbalock, flags); 1830 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1831 "6144 Outstanding NVME I/O Abort Request " 1832 "still pending on nvme_fcreq %p, " 1833 "lpfc_ncmd %p xri x%x\n", 1834 pnvme_fcreq, lpfc_nbuf, 1835 nvmereq_wqe->sli4_xritag); 1836 return; 1837 } 1838 1839 abts_buf = __lpfc_sli_get_iocbq(phba); 1840 if (!abts_buf) { 1841 spin_unlock_irqrestore(&phba->hbalock, flags); 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 return; 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_ct, &abts_wqe->abort_cmd.wqe_com, 0); 1863 bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX); 1864 bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com, 1865 nvmereq_wqe->iocb.ulpClass); 1866 1867 /* word 8 - tell the FW to abort the IO associated with this 1868 * outstanding exchange ID. 1869 */ 1870 abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag; 1871 1872 /* word 9 - this is the iotag for the abts_wqe completion. */ 1873 bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com, 1874 abts_buf->iotag); 1875 1876 /* word 10 */ 1877 bf_set(wqe_wqid, &abts_wqe->abort_cmd.wqe_com, nvmereq_wqe->hba_wqidx); 1878 bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1); 1879 bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE); 1880 1881 /* word 11 */ 1882 bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND); 1883 bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1); 1884 bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 1885 1886 /* ABTS WQE must go to the same WQ as the WQE to be aborted */ 1887 abts_buf->iocb_flag |= LPFC_IO_NVME; 1888 abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx; 1889 abts_buf->vport = vport; 1890 abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl; 1891 ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf); 1892 spin_unlock_irqrestore(&phba->hbalock, flags); 1893 if (ret_val) { 1894 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1895 "6137 Failed abts issue_wqe with status x%x " 1896 "for nvme_fcreq %p.\n", 1897 ret_val, pnvme_fcreq); 1898 lpfc_sli_release_iocbq(phba, abts_buf); 1899 return; 1900 } 1901 1902 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1903 "6138 Transport Abort NVME Request Issued for " 1904 "ox_id x%x on reqtag x%x\n", 1905 nvmereq_wqe->sli4_xritag, 1906 abts_buf->iotag); 1907 } 1908 1909 /* Declare and initialization an instance of the FC NVME template. */ 1910 static struct nvme_fc_port_template lpfc_nvme_template = { 1911 /* initiator-based functions */ 1912 .localport_delete = lpfc_nvme_localport_delete, 1913 .remoteport_delete = lpfc_nvme_remoteport_delete, 1914 .create_queue = lpfc_nvme_create_queue, 1915 .delete_queue = lpfc_nvme_delete_queue, 1916 .ls_req = lpfc_nvme_ls_req, 1917 .fcp_io = lpfc_nvme_fcp_io_submit, 1918 .ls_abort = lpfc_nvme_ls_abort, 1919 .fcp_abort = lpfc_nvme_fcp_abort, 1920 1921 .max_hw_queues = 1, 1922 .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1923 .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1924 .dma_boundary = 0xFFFFFFFF, 1925 1926 /* Sizes of additional private data for data structures. 1927 * No use for the last two sizes at this time. 1928 */ 1929 .local_priv_sz = sizeof(struct lpfc_nvme_lport), 1930 .remote_priv_sz = sizeof(struct lpfc_nvme_rport), 1931 .lsrqst_priv_sz = 0, 1932 .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv), 1933 }; 1934 1935 /** 1936 * lpfc_sli4_post_nvme_sgl_block - post a block of nvme sgl list to firmware 1937 * @phba: pointer to lpfc hba data structure. 1938 * @nblist: pointer to nvme buffer list. 1939 * @count: number of scsi buffers on the list. 1940 * 1941 * This routine is invoked to post a block of @count scsi sgl pages from a 1942 * SCSI buffer list @nblist to the HBA using non-embedded mailbox command. 1943 * No Lock is held. 1944 * 1945 **/ 1946 static int 1947 lpfc_sli4_post_nvme_sgl_block(struct lpfc_hba *phba, 1948 struct list_head *nblist, 1949 int count) 1950 { 1951 struct lpfc_nvme_buf *lpfc_ncmd; 1952 struct lpfc_mbx_post_uembed_sgl_page1 *sgl; 1953 struct sgl_page_pairs *sgl_pg_pairs; 1954 void *viraddr; 1955 LPFC_MBOXQ_t *mbox; 1956 uint32_t reqlen, alloclen, pg_pairs; 1957 uint32_t mbox_tmo; 1958 uint16_t xritag_start = 0; 1959 int rc = 0; 1960 uint32_t shdr_status, shdr_add_status; 1961 dma_addr_t pdma_phys_bpl1; 1962 union lpfc_sli4_cfg_shdr *shdr; 1963 1964 /* Calculate the requested length of the dma memory */ 1965 reqlen = count * sizeof(struct sgl_page_pairs) + 1966 sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t); 1967 if (reqlen > SLI4_PAGE_SIZE) { 1968 lpfc_printf_log(phba, KERN_WARNING, LOG_INIT, 1969 "6118 Block sgl registration required DMA " 1970 "size (%d) great than a page\n", reqlen); 1971 return -ENOMEM; 1972 } 1973 mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL); 1974 if (!mbox) { 1975 lpfc_printf_log(phba, KERN_ERR, LOG_INIT, 1976 "6119 Failed to allocate mbox cmd memory\n"); 1977 return -ENOMEM; 1978 } 1979 1980 /* Allocate DMA memory and set up the non-embedded mailbox command */ 1981 alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE, 1982 LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen, 1983 LPFC_SLI4_MBX_NEMBED); 1984 1985 if (alloclen < reqlen) { 1986 lpfc_printf_log(phba, KERN_ERR, LOG_INIT, 1987 "6120 Allocated DMA memory size (%d) is " 1988 "less than the requested DMA memory " 1989 "size (%d)\n", alloclen, reqlen); 1990 lpfc_sli4_mbox_cmd_free(phba, mbox); 1991 return -ENOMEM; 1992 } 1993 1994 /* Get the first SGE entry from the non-embedded DMA memory */ 1995 viraddr = mbox->sge_array->addr[0]; 1996 1997 /* Set up the SGL pages in the non-embedded DMA pages */ 1998 sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr; 1999 sgl_pg_pairs = &sgl->sgl_pg_pairs; 2000 2001 pg_pairs = 0; 2002 list_for_each_entry(lpfc_ncmd, nblist, list) { 2003 /* Set up the sge entry */ 2004 sgl_pg_pairs->sgl_pg0_addr_lo = 2005 cpu_to_le32(putPaddrLow(lpfc_ncmd->dma_phys_sgl)); 2006 sgl_pg_pairs->sgl_pg0_addr_hi = 2007 cpu_to_le32(putPaddrHigh(lpfc_ncmd->dma_phys_sgl)); 2008 if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE) 2009 pdma_phys_bpl1 = lpfc_ncmd->dma_phys_sgl + 2010 SGL_PAGE_SIZE; 2011 else 2012 pdma_phys_bpl1 = 0; 2013 sgl_pg_pairs->sgl_pg1_addr_lo = 2014 cpu_to_le32(putPaddrLow(pdma_phys_bpl1)); 2015 sgl_pg_pairs->sgl_pg1_addr_hi = 2016 cpu_to_le32(putPaddrHigh(pdma_phys_bpl1)); 2017 /* Keep the first xritag on the list */ 2018 if (pg_pairs == 0) 2019 xritag_start = lpfc_ncmd->cur_iocbq.sli4_xritag; 2020 sgl_pg_pairs++; 2021 pg_pairs++; 2022 } 2023 bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start); 2024 bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs); 2025 /* Perform endian conversion if necessary */ 2026 sgl->word0 = cpu_to_le32(sgl->word0); 2027 2028 if (!phba->sli4_hba.intr_enable) 2029 rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL); 2030 else { 2031 mbox_tmo = lpfc_mbox_tmo_val(phba, mbox); 2032 rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo); 2033 } 2034 shdr = (union lpfc_sli4_cfg_shdr *)&sgl->cfg_shdr; 2035 shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response); 2036 shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response); 2037 if (rc != MBX_TIMEOUT) 2038 lpfc_sli4_mbox_cmd_free(phba, mbox); 2039 if (shdr_status || shdr_add_status || rc) { 2040 lpfc_printf_log(phba, KERN_ERR, LOG_SLI, 2041 "6125 POST_SGL_BLOCK mailbox command failed " 2042 "status x%x add_status x%x mbx status x%x\n", 2043 shdr_status, shdr_add_status, rc); 2044 rc = -ENXIO; 2045 } 2046 return rc; 2047 } 2048 2049 /** 2050 * lpfc_post_nvme_sgl_list - Post blocks of nvme buffer sgls from a list 2051 * @phba: pointer to lpfc hba data structure. 2052 * @post_nblist: pointer to the nvme buffer list. 2053 * 2054 * This routine walks a list of nvme buffers that was passed in. It attempts 2055 * to construct blocks of nvme buffer sgls which contains contiguous xris and 2056 * uses the non-embedded SGL block post mailbox commands to post to the port. 2057 * For single NVME buffer sgl with non-contiguous xri, if any, it shall use 2058 * embedded SGL post mailbox command for posting. The @post_nblist passed in 2059 * must be local list, thus no lock is needed when manipulate the list. 2060 * 2061 * Returns: 0 = failure, non-zero number of successfully posted buffers. 2062 **/ 2063 static int 2064 lpfc_post_nvme_sgl_list(struct lpfc_hba *phba, 2065 struct list_head *post_nblist, int sb_count) 2066 { 2067 struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next; 2068 int status, sgl_size; 2069 int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0; 2070 dma_addr_t pdma_phys_sgl1; 2071 int last_xritag = NO_XRI; 2072 int cur_xritag; 2073 LIST_HEAD(prep_nblist); 2074 LIST_HEAD(blck_nblist); 2075 LIST_HEAD(nvme_nblist); 2076 2077 /* sanity check */ 2078 if (sb_count <= 0) 2079 return -EINVAL; 2080 2081 sgl_size = phba->cfg_sg_dma_buf_size; 2082 2083 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, post_nblist, list) { 2084 list_del_init(&lpfc_ncmd->list); 2085 block_cnt++; 2086 if ((last_xritag != NO_XRI) && 2087 (lpfc_ncmd->cur_iocbq.sli4_xritag != last_xritag + 1)) { 2088 /* a hole in xri block, form a sgl posting block */ 2089 list_splice_init(&prep_nblist, &blck_nblist); 2090 post_cnt = block_cnt - 1; 2091 /* prepare list for next posting block */ 2092 list_add_tail(&lpfc_ncmd->list, &prep_nblist); 2093 block_cnt = 1; 2094 } else { 2095 /* prepare list for next posting block */ 2096 list_add_tail(&lpfc_ncmd->list, &prep_nblist); 2097 /* enough sgls for non-embed sgl mbox command */ 2098 if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) { 2099 list_splice_init(&prep_nblist, &blck_nblist); 2100 post_cnt = block_cnt; 2101 block_cnt = 0; 2102 } 2103 } 2104 num_posting++; 2105 last_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag; 2106 2107 /* end of repost sgl list condition for NVME buffers */ 2108 if (num_posting == sb_count) { 2109 if (post_cnt == 0) { 2110 /* last sgl posting block */ 2111 list_splice_init(&prep_nblist, &blck_nblist); 2112 post_cnt = block_cnt; 2113 } else if (block_cnt == 1) { 2114 /* last single sgl with non-contiguous xri */ 2115 if (sgl_size > SGL_PAGE_SIZE) 2116 pdma_phys_sgl1 = 2117 lpfc_ncmd->dma_phys_sgl + 2118 SGL_PAGE_SIZE; 2119 else 2120 pdma_phys_sgl1 = 0; 2121 cur_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag; 2122 status = lpfc_sli4_post_sgl(phba, 2123 lpfc_ncmd->dma_phys_sgl, 2124 pdma_phys_sgl1, cur_xritag); 2125 if (status) { 2126 /* failure, put on abort nvme list */ 2127 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 2128 } else { 2129 /* success, put on NVME buffer list */ 2130 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2131 lpfc_ncmd->status = IOSTAT_SUCCESS; 2132 num_posted++; 2133 } 2134 /* success, put on NVME buffer sgl list */ 2135 list_add_tail(&lpfc_ncmd->list, &nvme_nblist); 2136 } 2137 } 2138 2139 /* continue until a nembed page worth of sgls */ 2140 if (post_cnt == 0) 2141 continue; 2142 2143 /* post block of NVME buffer list sgls */ 2144 status = lpfc_sli4_post_nvme_sgl_block(phba, &blck_nblist, 2145 post_cnt); 2146 2147 /* don't reset xirtag due to hole in xri block */ 2148 if (block_cnt == 0) 2149 last_xritag = NO_XRI; 2150 2151 /* reset NVME buffer post count for next round of posting */ 2152 post_cnt = 0; 2153 2154 /* put posted NVME buffer-sgl posted on NVME buffer sgl list */ 2155 while (!list_empty(&blck_nblist)) { 2156 list_remove_head(&blck_nblist, lpfc_ncmd, 2157 struct lpfc_nvme_buf, list); 2158 if (status) { 2159 /* failure, put on abort nvme list */ 2160 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 2161 } else { 2162 /* success, put on NVME buffer list */ 2163 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2164 lpfc_ncmd->status = IOSTAT_SUCCESS; 2165 num_posted++; 2166 } 2167 list_add_tail(&lpfc_ncmd->list, &nvme_nblist); 2168 } 2169 } 2170 /* Push NVME buffers with sgl posted to the available list */ 2171 while (!list_empty(&nvme_nblist)) { 2172 list_remove_head(&nvme_nblist, lpfc_ncmd, 2173 struct lpfc_nvme_buf, list); 2174 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2175 } 2176 return num_posted; 2177 } 2178 2179 /** 2180 * lpfc_repost_nvme_sgl_list - Repost all the allocated nvme buffer sgls 2181 * @phba: pointer to lpfc hba data structure. 2182 * 2183 * This routine walks the list of nvme buffers that have been allocated and 2184 * repost them to the port by using SGL block post. This is needed after a 2185 * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine 2186 * is responsible for moving all nvme buffers on the lpfc_abts_nvme_sgl_list 2187 * to the lpfc_nvme_buf_list. If the repost fails, reject all nvme buffers. 2188 * 2189 * Returns: 0 = success, non-zero failure. 2190 **/ 2191 int 2192 lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba) 2193 { 2194 LIST_HEAD(post_nblist); 2195 int num_posted, rc = 0; 2196 2197 /* get all NVME buffers need to repost to a local list */ 2198 spin_lock_irq(&phba->nvme_buf_list_get_lock); 2199 spin_lock(&phba->nvme_buf_list_put_lock); 2200 list_splice_init(&phba->lpfc_nvme_buf_list_get, &post_nblist); 2201 list_splice(&phba->lpfc_nvme_buf_list_put, &post_nblist); 2202 phba->get_nvme_bufs = 0; 2203 phba->put_nvme_bufs = 0; 2204 spin_unlock(&phba->nvme_buf_list_put_lock); 2205 spin_unlock_irq(&phba->nvme_buf_list_get_lock); 2206 2207 /* post the list of nvme buffer sgls to port if available */ 2208 if (!list_empty(&post_nblist)) { 2209 num_posted = lpfc_post_nvme_sgl_list(phba, &post_nblist, 2210 phba->sli4_hba.nvme_xri_cnt); 2211 /* failed to post any nvme buffer, return error */ 2212 if (num_posted == 0) 2213 rc = -EIO; 2214 } 2215 return rc; 2216 } 2217 2218 /** 2219 * lpfc_new_nvme_buf - Scsi buffer allocator for HBA with SLI4 IF spec 2220 * @vport: The virtual port for which this call being executed. 2221 * @num_to_allocate: The requested number of buffers to allocate. 2222 * 2223 * This routine allocates nvme buffers for device with SLI-4 interface spec, 2224 * the nvme buffer contains all the necessary information needed to initiate 2225 * a NVME I/O. After allocating up to @num_to_allocate NVME buffers and put 2226 * them on a list, it post them to the port by using SGL block post. 2227 * 2228 * Return codes: 2229 * int - number of nvme buffers that were allocated and posted. 2230 * 0 = failure, less than num_to_alloc is a partial failure. 2231 **/ 2232 static int 2233 lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc) 2234 { 2235 struct lpfc_hba *phba = vport->phba; 2236 struct lpfc_nvme_buf *lpfc_ncmd; 2237 struct lpfc_iocbq *pwqeq; 2238 union lpfc_wqe128 *wqe; 2239 struct sli4_sge *sgl; 2240 dma_addr_t pdma_phys_sgl; 2241 uint16_t iotag, lxri = 0; 2242 int bcnt, num_posted, sgl_size; 2243 LIST_HEAD(prep_nblist); 2244 LIST_HEAD(post_nblist); 2245 LIST_HEAD(nvme_nblist); 2246 2247 sgl_size = phba->cfg_sg_dma_buf_size; 2248 2249 for (bcnt = 0; bcnt < num_to_alloc; bcnt++) { 2250 lpfc_ncmd = kzalloc(sizeof(struct lpfc_nvme_buf), GFP_KERNEL); 2251 if (!lpfc_ncmd) 2252 break; 2253 /* 2254 * Get memory from the pci pool to map the virt space to 2255 * pci bus space for an I/O. The DMA buffer includes the 2256 * number of SGE's necessary to support the sg_tablesize. 2257 */ 2258 lpfc_ncmd->data = dma_pool_zalloc(phba->lpfc_sg_dma_buf_pool, 2259 GFP_KERNEL, 2260 &lpfc_ncmd->dma_handle); 2261 if (!lpfc_ncmd->data) { 2262 kfree(lpfc_ncmd); 2263 break; 2264 } 2265 2266 lxri = lpfc_sli4_next_xritag(phba); 2267 if (lxri == NO_XRI) { 2268 dma_pool_free(phba->lpfc_sg_dma_buf_pool, 2269 lpfc_ncmd->data, lpfc_ncmd->dma_handle); 2270 kfree(lpfc_ncmd); 2271 break; 2272 } 2273 pwqeq = &(lpfc_ncmd->cur_iocbq); 2274 wqe = &pwqeq->wqe; 2275 2276 /* Allocate iotag for lpfc_ncmd->cur_iocbq. */ 2277 iotag = lpfc_sli_next_iotag(phba, pwqeq); 2278 if (iotag == 0) { 2279 dma_pool_free(phba->lpfc_sg_dma_buf_pool, 2280 lpfc_ncmd->data, lpfc_ncmd->dma_handle); 2281 kfree(lpfc_ncmd); 2282 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 2283 "6121 Failed to allocated IOTAG for" 2284 " XRI:0x%x\n", lxri); 2285 lpfc_sli4_free_xri(phba, lxri); 2286 break; 2287 } 2288 pwqeq->sli4_lxritag = lxri; 2289 pwqeq->sli4_xritag = phba->sli4_hba.xri_ids[lxri]; 2290 pwqeq->iocb_flag |= LPFC_IO_NVME; 2291 pwqeq->context1 = lpfc_ncmd; 2292 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl; 2293 2294 /* Initialize local short-hand pointers. */ 2295 lpfc_ncmd->nvme_sgl = lpfc_ncmd->data; 2296 sgl = lpfc_ncmd->nvme_sgl; 2297 pdma_phys_sgl = lpfc_ncmd->dma_handle; 2298 lpfc_ncmd->dma_phys_sgl = pdma_phys_sgl; 2299 2300 /* Rsp SGE will be filled in when we rcv an IO 2301 * from the NVME Layer to be sent. 2302 * The cmd is going to be embedded so we need a SKIP SGE. 2303 */ 2304 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP); 2305 bf_set(lpfc_sli4_sge_last, sgl, 0); 2306 sgl->word2 = cpu_to_le32(sgl->word2); 2307 /* Fill in word 3 / sgl_len during cmd submission */ 2308 2309 lpfc_ncmd->cur_iocbq.context1 = lpfc_ncmd; 2310 2311 /* Initialize WQE */ 2312 memset(wqe, 0, sizeof(union lpfc_wqe)); 2313 2314 /* add the nvme buffer to a post list */ 2315 list_add_tail(&lpfc_ncmd->list, &post_nblist); 2316 spin_lock_irq(&phba->nvme_buf_list_get_lock); 2317 phba->sli4_hba.nvme_xri_cnt++; 2318 spin_unlock_irq(&phba->nvme_buf_list_get_lock); 2319 } 2320 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 2321 "6114 Allocate %d out of %d requested new NVME " 2322 "buffers\n", bcnt, num_to_alloc); 2323 2324 /* post the list of nvme buffer sgls to port if available */ 2325 if (!list_empty(&post_nblist)) 2326 num_posted = lpfc_post_nvme_sgl_list(phba, 2327 &post_nblist, bcnt); 2328 else 2329 num_posted = 0; 2330 2331 return num_posted; 2332 } 2333 2334 static inline struct lpfc_nvme_buf * 2335 lpfc_nvme_buf(struct lpfc_hba *phba) 2336 { 2337 struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next; 2338 2339 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, 2340 &phba->lpfc_nvme_buf_list_get, list) { 2341 list_del_init(&lpfc_ncmd->list); 2342 phba->get_nvme_bufs--; 2343 return lpfc_ncmd; 2344 } 2345 return NULL; 2346 } 2347 2348 /** 2349 * lpfc_get_nvme_buf - Get a nvme buffer from lpfc_nvme_buf_list of the HBA 2350 * @phba: The HBA for which this call is being executed. 2351 * 2352 * This routine removes a nvme buffer from head of @phba lpfc_nvme_buf_list list 2353 * and returns to caller. 2354 * 2355 * Return codes: 2356 * NULL - Error 2357 * Pointer to lpfc_nvme_buf - Success 2358 **/ 2359 static struct lpfc_nvme_buf * 2360 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp, 2361 int expedite) 2362 { 2363 struct lpfc_nvme_buf *lpfc_ncmd = NULL; 2364 unsigned long iflag = 0; 2365 2366 spin_lock_irqsave(&phba->nvme_buf_list_get_lock, iflag); 2367 if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite) 2368 lpfc_ncmd = lpfc_nvme_buf(phba); 2369 if (!lpfc_ncmd) { 2370 spin_lock(&phba->nvme_buf_list_put_lock); 2371 list_splice(&phba->lpfc_nvme_buf_list_put, 2372 &phba->lpfc_nvme_buf_list_get); 2373 phba->get_nvme_bufs += phba->put_nvme_bufs; 2374 INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put); 2375 phba->put_nvme_bufs = 0; 2376 spin_unlock(&phba->nvme_buf_list_put_lock); 2377 if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite) 2378 lpfc_ncmd = lpfc_nvme_buf(phba); 2379 } 2380 spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag); 2381 return lpfc_ncmd; 2382 } 2383 2384 /** 2385 * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list. 2386 * @phba: The Hba for which this call is being executed. 2387 * @lpfc_ncmd: The nvme buffer which is being released. 2388 * 2389 * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba 2390 * lpfc_nvme_buf_list list. For SLI4 XRI's are tied to the nvme buffer 2391 * and cannot be reused for at least RA_TOV amount of time if it was 2392 * aborted. 2393 **/ 2394 static void 2395 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd) 2396 { 2397 unsigned long iflag = 0; 2398 2399 lpfc_ncmd->nonsg_phys = 0; 2400 if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) { 2401 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2402 "6310 XB release deferred for " 2403 "ox_id x%x on reqtag x%x\n", 2404 lpfc_ncmd->cur_iocbq.sli4_xritag, 2405 lpfc_ncmd->cur_iocbq.iotag); 2406 2407 spin_lock_irqsave(&phba->sli4_hba.abts_nvme_buf_list_lock, 2408 iflag); 2409 list_add_tail(&lpfc_ncmd->list, 2410 &phba->sli4_hba.lpfc_abts_nvme_buf_list); 2411 spin_unlock_irqrestore(&phba->sli4_hba.abts_nvme_buf_list_lock, 2412 iflag); 2413 } else { 2414 lpfc_ncmd->nvmeCmd = NULL; 2415 lpfc_ncmd->cur_iocbq.iocb_flag = LPFC_IO_NVME; 2416 spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag); 2417 list_add_tail(&lpfc_ncmd->list, &phba->lpfc_nvme_buf_list_put); 2418 phba->put_nvme_bufs++; 2419 spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag); 2420 } 2421 } 2422 2423 /** 2424 * lpfc_nvme_create_localport - Create/Bind an nvme localport instance. 2425 * @pvport - the lpfc_vport instance requesting a localport. 2426 * 2427 * This routine is invoked to create an nvme localport instance to bind 2428 * to the nvme_fc_transport. It is called once during driver load 2429 * like lpfc_create_shost after all other services are initialized. 2430 * It requires a vport, vpi, and wwns at call time. Other localport 2431 * parameters are modified as the driver's FCID and the Fabric WWN 2432 * are established. 2433 * 2434 * Return codes 2435 * 0 - successful 2436 * -ENOMEM - no heap memory available 2437 * other values - from nvme registration upcall 2438 **/ 2439 int 2440 lpfc_nvme_create_localport(struct lpfc_vport *vport) 2441 { 2442 int ret = 0; 2443 struct lpfc_hba *phba = vport->phba; 2444 struct nvme_fc_port_info nfcp_info; 2445 struct nvme_fc_local_port *localport; 2446 struct lpfc_nvme_lport *lport; 2447 struct lpfc_nvme_ctrl_stat *cstat; 2448 int len, i; 2449 2450 /* Initialize this localport instance. The vport wwn usage ensures 2451 * that NPIV is accounted for. 2452 */ 2453 memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info)); 2454 nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR; 2455 nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn); 2456 nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn); 2457 2458 /* Limit to LPFC_MAX_NVME_SEG_CNT. 2459 * For now need + 1 to get around NVME transport logic. 2460 */ 2461 if (phba->cfg_sg_seg_cnt > LPFC_MAX_NVME_SEG_CNT) { 2462 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_INIT, 2463 "6300 Reducing sg segment cnt to %d\n", 2464 LPFC_MAX_NVME_SEG_CNT); 2465 phba->cfg_nvme_seg_cnt = LPFC_MAX_NVME_SEG_CNT; 2466 } else { 2467 phba->cfg_nvme_seg_cnt = phba->cfg_sg_seg_cnt; 2468 } 2469 lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1; 2470 lpfc_nvme_template.max_hw_queues = phba->cfg_nvme_io_channel; 2471 2472 cstat = kmalloc((sizeof(struct lpfc_nvme_ctrl_stat) * 2473 phba->cfg_nvme_io_channel), GFP_KERNEL); 2474 if (!cstat) 2475 return -ENOMEM; 2476 2477 /* localport is allocated from the stack, but the registration 2478 * call allocates heap memory as well as the private area. 2479 */ 2480 #if (IS_ENABLED(CONFIG_NVME_FC)) 2481 ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template, 2482 &vport->phba->pcidev->dev, &localport); 2483 #else 2484 ret = -ENOMEM; 2485 #endif 2486 if (!ret) { 2487 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC, 2488 "6005 Successfully registered local " 2489 "NVME port num %d, localP %p, private %p, " 2490 "sg_seg %d\n", 2491 localport->port_num, localport, 2492 localport->private, 2493 lpfc_nvme_template.max_sgl_segments); 2494 2495 /* Private is our lport size declared in the template. */ 2496 lport = (struct lpfc_nvme_lport *)localport->private; 2497 vport->localport = localport; 2498 lport->vport = vport; 2499 lport->cstat = cstat; 2500 vport->nvmei_support = 1; 2501 2502 atomic_set(&lport->xmt_fcp_noxri, 0); 2503 atomic_set(&lport->xmt_fcp_bad_ndlp, 0); 2504 atomic_set(&lport->xmt_fcp_qdepth, 0); 2505 atomic_set(&lport->xmt_fcp_err, 0); 2506 atomic_set(&lport->xmt_fcp_wqerr, 0); 2507 atomic_set(&lport->xmt_fcp_abort, 0); 2508 atomic_set(&lport->xmt_ls_abort, 0); 2509 atomic_set(&lport->xmt_ls_err, 0); 2510 atomic_set(&lport->cmpl_fcp_xb, 0); 2511 atomic_set(&lport->cmpl_fcp_err, 0); 2512 atomic_set(&lport->cmpl_ls_xb, 0); 2513 atomic_set(&lport->cmpl_ls_err, 0); 2514 atomic_set(&lport->fc4NvmeLsRequests, 0); 2515 atomic_set(&lport->fc4NvmeLsCmpls, 0); 2516 2517 for (i = 0; i < phba->cfg_nvme_io_channel; i++) { 2518 cstat = &lport->cstat[i]; 2519 atomic_set(&cstat->fc4NvmeInputRequests, 0); 2520 atomic_set(&cstat->fc4NvmeOutputRequests, 0); 2521 atomic_set(&cstat->fc4NvmeControlRequests, 0); 2522 atomic_set(&cstat->fc4NvmeIoCmpls, 0); 2523 } 2524 2525 /* Don't post more new bufs if repost already recovered 2526 * the nvme sgls. 2527 */ 2528 if (phba->sli4_hba.nvme_xri_cnt == 0) { 2529 len = lpfc_new_nvme_buf(vport, 2530 phba->sli4_hba.nvme_xri_max); 2531 vport->phba->total_nvme_bufs += len; 2532 } 2533 } else { 2534 kfree(cstat); 2535 } 2536 2537 return ret; 2538 } 2539 2540 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg. 2541 * 2542 * The driver has to wait for the host nvme transport to callback 2543 * indicating the localport has successfully unregistered all 2544 * resources. Since this is an uninterruptible wait, loop every ten 2545 * seconds and print a message indicating no progress. 2546 * 2547 * An uninterruptible wait is used because of the risk of transport-to- 2548 * driver state mismatch. 2549 */ 2550 void 2551 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport, 2552 struct lpfc_nvme_lport *lport) 2553 { 2554 #if (IS_ENABLED(CONFIG_NVME_FC)) 2555 u32 wait_tmo; 2556 int ret; 2557 2558 /* Host transport has to clean up and confirm requiring an indefinite 2559 * wait. Print a message if a 10 second wait expires and renew the 2560 * wait. This is unexpected. 2561 */ 2562 wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000); 2563 while (true) { 2564 ret = wait_for_completion_timeout(&lport->lport_unreg_done, 2565 wait_tmo); 2566 if (unlikely(!ret)) { 2567 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 2568 "6176 Lport %p Localport %p wait " 2569 "timed out. Renewing.\n", 2570 lport, vport->localport); 2571 continue; 2572 } 2573 break; 2574 } 2575 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 2576 "6177 Lport %p Localport %p Complete Success\n", 2577 lport, vport->localport); 2578 #endif 2579 } 2580 2581 /** 2582 * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport. 2583 * @pnvme: pointer to lpfc nvme data structure. 2584 * 2585 * This routine is invoked to destroy all lports bound to the phba. 2586 * The lport memory was allocated by the nvme fc transport and is 2587 * released there. This routine ensures all rports bound to the 2588 * lport have been disconnected. 2589 * 2590 **/ 2591 void 2592 lpfc_nvme_destroy_localport(struct lpfc_vport *vport) 2593 { 2594 #if (IS_ENABLED(CONFIG_NVME_FC)) 2595 struct nvme_fc_local_port *localport; 2596 struct lpfc_nvme_lport *lport; 2597 struct lpfc_nvme_ctrl_stat *cstat; 2598 int ret; 2599 2600 if (vport->nvmei_support == 0) 2601 return; 2602 2603 localport = vport->localport; 2604 vport->localport = NULL; 2605 lport = (struct lpfc_nvme_lport *)localport->private; 2606 cstat = lport->cstat; 2607 2608 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2609 "6011 Destroying NVME localport %p\n", 2610 localport); 2611 2612 /* lport's rport list is clear. Unregister 2613 * lport and release resources. 2614 */ 2615 init_completion(&lport->lport_unreg_done); 2616 ret = nvme_fc_unregister_localport(localport); 2617 2618 /* Wait for completion. This either blocks 2619 * indefinitely or succeeds 2620 */ 2621 lpfc_nvme_lport_unreg_wait(vport, lport); 2622 kfree(cstat); 2623 2624 /* Regardless of the unregister upcall response, clear 2625 * nvmei_support. All rports are unregistered and the 2626 * driver will clean up. 2627 */ 2628 vport->nvmei_support = 0; 2629 if (ret == 0) { 2630 lpfc_printf_vlog(vport, 2631 KERN_INFO, LOG_NVME_DISC, 2632 "6009 Unregistered lport Success\n"); 2633 } else { 2634 lpfc_printf_vlog(vport, 2635 KERN_INFO, LOG_NVME_DISC, 2636 "6010 Unregistered lport " 2637 "Failed, status x%x\n", 2638 ret); 2639 } 2640 #endif 2641 } 2642 2643 void 2644 lpfc_nvme_update_localport(struct lpfc_vport *vport) 2645 { 2646 #if (IS_ENABLED(CONFIG_NVME_FC)) 2647 struct nvme_fc_local_port *localport; 2648 struct lpfc_nvme_lport *lport; 2649 2650 localport = vport->localport; 2651 if (!localport) { 2652 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2653 "6710 Update NVME fail. No localport\n"); 2654 return; 2655 } 2656 lport = (struct lpfc_nvme_lport *)localport->private; 2657 if (!lport) { 2658 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2659 "6171 Update NVME fail. localP %p, No lport\n", 2660 localport); 2661 return; 2662 } 2663 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2664 "6012 Update NVME lport %p did x%x\n", 2665 localport, vport->fc_myDID); 2666 2667 localport->port_id = vport->fc_myDID; 2668 if (localport->port_id == 0) 2669 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY; 2670 else 2671 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR; 2672 2673 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2674 "6030 bound lport %p to DID x%06x\n", 2675 lport, localport->port_id); 2676 #endif 2677 } 2678 2679 int 2680 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2681 { 2682 #if (IS_ENABLED(CONFIG_NVME_FC)) 2683 int ret = 0; 2684 struct nvme_fc_local_port *localport; 2685 struct lpfc_nvme_lport *lport; 2686 struct lpfc_nvme_rport *rport; 2687 struct lpfc_nvme_rport *oldrport; 2688 struct nvme_fc_remote_port *remote_port; 2689 struct nvme_fc_port_info rpinfo; 2690 struct lpfc_nodelist *prev_ndlp; 2691 2692 lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC, 2693 "6006 Register NVME PORT. DID x%06x nlptype x%x\n", 2694 ndlp->nlp_DID, ndlp->nlp_type); 2695 2696 localport = vport->localport; 2697 if (!localport) 2698 return 0; 2699 2700 lport = (struct lpfc_nvme_lport *)localport->private; 2701 2702 /* NVME rports are not preserved across devloss. 2703 * Just register this instance. Note, rpinfo->dev_loss_tmo 2704 * is left 0 to indicate accept transport defaults. The 2705 * driver communicates port role capabilities consistent 2706 * with the PRLI response data. 2707 */ 2708 memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info)); 2709 rpinfo.port_id = ndlp->nlp_DID; 2710 if (ndlp->nlp_type & NLP_NVME_TARGET) 2711 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET; 2712 if (ndlp->nlp_type & NLP_NVME_INITIATOR) 2713 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR; 2714 2715 if (ndlp->nlp_type & NLP_NVME_DISCOVERY) 2716 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY; 2717 2718 rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn); 2719 rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn); 2720 2721 oldrport = lpfc_ndlp_get_nrport(ndlp); 2722 if (!oldrport) 2723 lpfc_nlp_get(ndlp); 2724 2725 ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port); 2726 if (!ret) { 2727 /* If the ndlp already has an nrport, this is just 2728 * a resume of the existing rport. Else this is a 2729 * new rport. 2730 */ 2731 /* Guard against an unregister/reregister 2732 * race that leaves the WAIT flag set. 2733 */ 2734 spin_lock_irq(&vport->phba->hbalock); 2735 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 2736 spin_unlock_irq(&vport->phba->hbalock); 2737 rport = remote_port->private; 2738 if (oldrport) { 2739 if (oldrport == remote_port->private) { 2740 /* Same remoteport. Just reuse. */ 2741 lpfc_printf_vlog(ndlp->vport, KERN_INFO, 2742 LOG_NVME_DISC, 2743 "6014 Rebinding lport to " 2744 "remoteport %p wwpn 0x%llx, " 2745 "Data: x%x x%x %p x%x x%06x\n", 2746 remote_port, 2747 remote_port->port_name, 2748 remote_port->port_id, 2749 remote_port->port_role, 2750 ndlp, 2751 ndlp->nlp_type, 2752 ndlp->nlp_DID); 2753 return 0; 2754 } 2755 prev_ndlp = rport->ndlp; 2756 2757 /* Sever the ndlp<->rport association 2758 * before dropping the ndlp ref from 2759 * register. 2760 */ 2761 spin_lock_irq(&vport->phba->hbalock); 2762 ndlp->nrport = NULL; 2763 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 2764 spin_unlock_irq(&vport->phba->hbalock); 2765 rport->ndlp = NULL; 2766 rport->remoteport = NULL; 2767 2768 /* Reference only removed if previous NDLP is no longer 2769 * active. It might be just a swap and removing the 2770 * reference would cause a premature cleanup. 2771 */ 2772 if (prev_ndlp && prev_ndlp != ndlp) { 2773 if ((!NLP_CHK_NODE_ACT(prev_ndlp)) || 2774 (!prev_ndlp->nrport)) 2775 lpfc_nlp_put(prev_ndlp); 2776 } 2777 } 2778 2779 /* Clean bind the rport to the ndlp. */ 2780 rport->remoteport = remote_port; 2781 rport->lport = lport; 2782 rport->ndlp = ndlp; 2783 spin_lock_irq(&vport->phba->hbalock); 2784 ndlp->nrport = rport; 2785 spin_unlock_irq(&vport->phba->hbalock); 2786 lpfc_printf_vlog(vport, KERN_INFO, 2787 LOG_NVME_DISC | LOG_NODE, 2788 "6022 Binding new rport to " 2789 "lport %p Remoteport %p WWNN 0x%llx, " 2790 "Rport WWPN 0x%llx DID " 2791 "x%06x Role x%x, ndlp %p\n", 2792 lport, remote_port, 2793 rpinfo.node_name, rpinfo.port_name, 2794 rpinfo.port_id, rpinfo.port_role, 2795 ndlp); 2796 } else { 2797 lpfc_printf_vlog(vport, KERN_ERR, 2798 LOG_NVME_DISC | LOG_NODE, 2799 "6031 RemotePort Registration failed " 2800 "err: %d, DID x%06x\n", 2801 ret, ndlp->nlp_DID); 2802 } 2803 2804 return ret; 2805 #else 2806 return 0; 2807 #endif 2808 } 2809 2810 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport. 2811 * 2812 * There is no notion of Devloss or rport recovery from the current 2813 * nvme_transport perspective. Loss of an rport just means IO cannot 2814 * be sent and recovery is completely up to the initator. 2815 * For now, the driver just unbinds the DID and port_role so that 2816 * no further IO can be issued. Changes are planned for later. 2817 * 2818 * Notes - the ndlp reference count is not decremented here since 2819 * since there is no nvme_transport api for devloss. Node ref count 2820 * is only adjusted in driver unload. 2821 */ 2822 void 2823 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2824 { 2825 #if (IS_ENABLED(CONFIG_NVME_FC)) 2826 int ret; 2827 struct nvme_fc_local_port *localport; 2828 struct lpfc_nvme_lport *lport; 2829 struct lpfc_nvme_rport *rport; 2830 struct nvme_fc_remote_port *remoteport; 2831 2832 localport = vport->localport; 2833 2834 /* This is fundamental error. The localport is always 2835 * available until driver unload. Just exit. 2836 */ 2837 if (!localport) 2838 return; 2839 2840 lport = (struct lpfc_nvme_lport *)localport->private; 2841 if (!lport) 2842 goto input_err; 2843 2844 rport = lpfc_ndlp_get_nrport(ndlp); 2845 if (!rport) 2846 goto input_err; 2847 2848 remoteport = rport->remoteport; 2849 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2850 "6033 Unreg nvme remoteport %p, portname x%llx, " 2851 "port_id x%06x, portstate x%x port type x%x\n", 2852 remoteport, remoteport->port_name, 2853 remoteport->port_id, remoteport->port_state, 2854 ndlp->nlp_type); 2855 2856 /* Sanity check ndlp type. Only call for NVME ports. Don't 2857 * clear any rport state until the transport calls back. 2858 */ 2859 2860 if (ndlp->nlp_type & NLP_NVME_TARGET) { 2861 /* No concern about the role change on the nvme remoteport. 2862 * The transport will update it. 2863 */ 2864 ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG; 2865 2866 /* Don't let the host nvme transport keep sending keep-alives 2867 * on this remoteport. Vport is unloading, no recovery. The 2868 * return values is ignored. The upcall is a courtesy to the 2869 * transport. 2870 */ 2871 if (vport->load_flag & FC_UNLOADING) 2872 (void)nvme_fc_set_remoteport_devloss(remoteport, 0); 2873 2874 ret = nvme_fc_unregister_remoteport(remoteport); 2875 if (ret != 0) { 2876 lpfc_nlp_put(ndlp); 2877 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2878 "6167 NVME unregister failed %d " 2879 "port_state x%x\n", 2880 ret, remoteport->port_state); 2881 } 2882 } 2883 return; 2884 2885 input_err: 2886 #endif 2887 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2888 "6168 State error: lport %p, rport%p FCID x%06x\n", 2889 vport->localport, ndlp->rport, ndlp->nlp_DID); 2890 } 2891 2892 /** 2893 * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort 2894 * @phba: pointer to lpfc hba data structure. 2895 * @axri: pointer to the fcp xri abort wcqe structure. 2896 * 2897 * This routine is invoked by the worker thread to process a SLI4 fast-path 2898 * NVME aborted xri. Aborted NVME IO commands are completed to the transport 2899 * here. 2900 **/ 2901 void 2902 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba, 2903 struct sli4_wcqe_xri_aborted *axri) 2904 { 2905 uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri); 2906 struct lpfc_nvme_buf *lpfc_ncmd, *next_lpfc_ncmd; 2907 struct nvmefc_fcp_req *nvme_cmd = NULL; 2908 struct lpfc_nodelist *ndlp; 2909 unsigned long iflag = 0; 2910 2911 if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME)) 2912 return; 2913 spin_lock_irqsave(&phba->hbalock, iflag); 2914 spin_lock(&phba->sli4_hba.abts_nvme_buf_list_lock); 2915 list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd, 2916 &phba->sli4_hba.lpfc_abts_nvme_buf_list, 2917 list) { 2918 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) { 2919 list_del_init(&lpfc_ncmd->list); 2920 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2921 lpfc_ncmd->status = IOSTAT_SUCCESS; 2922 spin_unlock( 2923 &phba->sli4_hba.abts_nvme_buf_list_lock); 2924 2925 spin_unlock_irqrestore(&phba->hbalock, iflag); 2926 ndlp = lpfc_ncmd->ndlp; 2927 if (ndlp) 2928 lpfc_sli4_abts_err_handler(phba, ndlp, axri); 2929 2930 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2931 "6311 nvme_cmd %p xri x%x tag x%x " 2932 "abort complete and xri released\n", 2933 lpfc_ncmd->nvmeCmd, xri, 2934 lpfc_ncmd->cur_iocbq.iotag); 2935 2936 /* Aborted NVME commands are required to not complete 2937 * before the abort exchange command fully completes. 2938 * Once completed, it is available via the put list. 2939 */ 2940 if (lpfc_ncmd->nvmeCmd) { 2941 nvme_cmd = lpfc_ncmd->nvmeCmd; 2942 nvme_cmd->done(nvme_cmd); 2943 lpfc_ncmd->nvmeCmd = NULL; 2944 } 2945 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2946 return; 2947 } 2948 } 2949 spin_unlock(&phba->sli4_hba.abts_nvme_buf_list_lock); 2950 spin_unlock_irqrestore(&phba->hbalock, iflag); 2951 2952 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2953 "6312 XRI Aborted xri x%x not found\n", xri); 2954 2955 } 2956 2957 /** 2958 * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete 2959 * @phba: Pointer to HBA context object. 2960 * 2961 * This function flushes all wqes in the nvme rings and frees all resources 2962 * in the txcmplq. This function does not issue abort wqes for the IO 2963 * commands in txcmplq, they will just be returned with 2964 * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI 2965 * slot has been permanently disabled. 2966 **/ 2967 void 2968 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba) 2969 { 2970 struct lpfc_sli_ring *pring; 2971 u32 i, wait_cnt = 0; 2972 2973 if (phba->sli_rev < LPFC_SLI_REV4) 2974 return; 2975 2976 /* Cycle through all NVME rings and make sure all outstanding 2977 * WQEs have been removed from the txcmplqs. 2978 */ 2979 for (i = 0; i < phba->cfg_nvme_io_channel; i++) { 2980 pring = phba->sli4_hba.nvme_wq[i]->pring; 2981 2982 /* Retrieve everything on the txcmplq */ 2983 while (!list_empty(&pring->txcmplq)) { 2984 msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1); 2985 wait_cnt++; 2986 2987 /* The sleep is 10mS. Every ten seconds, 2988 * dump a message. Something is wrong. 2989 */ 2990 if ((wait_cnt % 1000) == 0) { 2991 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 2992 "6178 NVME IO not empty, " 2993 "cnt %d\n", wait_cnt); 2994 } 2995 } 2996 } 2997 } 2998