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 /* Update stats and complete the IO. There is 1139 * no need for dma unprep because the nvme_transport 1140 * owns the dma address. 1141 */ 1142 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1143 if (lpfc_ncmd->ts_cmd_start) { 1144 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp; 1145 lpfc_ncmd->ts_data_nvme = ktime_get_ns(); 1146 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme; 1147 lpfc_nvme_ktime(phba, lpfc_ncmd); 1148 } 1149 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1150 if (lpfc_ncmd->cpu != smp_processor_id()) 1151 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 1152 "6701 CPU Check cmpl: " 1153 "cpu %d expect %d\n", 1154 smp_processor_id(), lpfc_ncmd->cpu); 1155 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT) 1156 phba->cpucheck_cmpl_io[lpfc_ncmd->cpu]++; 1157 } 1158 #endif 1159 1160 /* NVME targets need completion held off until the abort exchange 1161 * completes unless the NVME Rport is getting unregistered. 1162 */ 1163 1164 if (!(lpfc_ncmd->flags & LPFC_SBUF_XBUSY)) { 1165 freqpriv = nCmd->private; 1166 freqpriv->nvme_buf = NULL; 1167 nCmd->done(nCmd); 1168 lpfc_ncmd->nvmeCmd = NULL; 1169 } 1170 1171 spin_lock_irqsave(&phba->hbalock, flags); 1172 lpfc_ncmd->nrport = NULL; 1173 spin_unlock_irqrestore(&phba->hbalock, flags); 1174 1175 /* Call release with XB=1 to queue the IO into the abort list. */ 1176 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1177 } 1178 1179 1180 /** 1181 * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO 1182 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1183 * @lpfc_nvme_lport: Pointer to the driver's local port data 1184 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1185 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1186 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1187 * 1188 * Driver registers this routine as it io request handler. This 1189 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1190 * data structure to the rport indicated in @lpfc_nvme_rport. 1191 * 1192 * Return value : 1193 * 0 - Success 1194 * TODO: What are the failure codes. 1195 **/ 1196 static int 1197 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport, 1198 struct lpfc_nvme_buf *lpfc_ncmd, 1199 struct lpfc_nodelist *pnode, 1200 struct lpfc_nvme_ctrl_stat *cstat) 1201 { 1202 struct lpfc_hba *phba = vport->phba; 1203 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1204 struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq); 1205 union lpfc_wqe128 *wqe = &pwqeq->wqe; 1206 uint32_t req_len; 1207 1208 if (!pnode || !NLP_CHK_NODE_ACT(pnode)) 1209 return -EINVAL; 1210 1211 /* 1212 * There are three possibilities here - use scatter-gather segment, use 1213 * the single mapping, or neither. 1214 */ 1215 if (nCmd->sg_cnt) { 1216 if (nCmd->io_dir == NVMEFC_FCP_WRITE) { 1217 /* From the iwrite template, initialize words 7 - 11 */ 1218 memcpy(&wqe->words[7], 1219 &lpfc_iwrite_cmd_template.words[7], 1220 sizeof(uint32_t) * 5); 1221 1222 /* Word 4 */ 1223 wqe->fcp_iwrite.total_xfer_len = nCmd->payload_length; 1224 1225 /* Word 5 */ 1226 if ((phba->cfg_nvme_enable_fb) && 1227 (pnode->nlp_flag & NLP_FIRSTBURST)) { 1228 req_len = lpfc_ncmd->nvmeCmd->payload_length; 1229 if (req_len < pnode->nvme_fb_size) 1230 wqe->fcp_iwrite.initial_xfer_len = 1231 req_len; 1232 else 1233 wqe->fcp_iwrite.initial_xfer_len = 1234 pnode->nvme_fb_size; 1235 } else { 1236 wqe->fcp_iwrite.initial_xfer_len = 0; 1237 } 1238 atomic_inc(&cstat->fc4NvmeOutputRequests); 1239 } else { 1240 /* From the iread template, initialize words 7 - 11 */ 1241 memcpy(&wqe->words[7], 1242 &lpfc_iread_cmd_template.words[7], 1243 sizeof(uint32_t) * 5); 1244 1245 /* Word 4 */ 1246 wqe->fcp_iread.total_xfer_len = nCmd->payload_length; 1247 1248 /* Word 5 */ 1249 wqe->fcp_iread.rsrvd5 = 0; 1250 1251 atomic_inc(&cstat->fc4NvmeInputRequests); 1252 } 1253 } else { 1254 /* From the icmnd template, initialize words 4 - 11 */ 1255 memcpy(&wqe->words[4], &lpfc_icmnd_cmd_template.words[4], 1256 sizeof(uint32_t) * 8); 1257 atomic_inc(&cstat->fc4NvmeControlRequests); 1258 } 1259 /* 1260 * Finish initializing those WQE fields that are independent 1261 * of the nvme_cmnd request_buffer 1262 */ 1263 1264 /* Word 3 */ 1265 bf_set(payload_offset_len, &wqe->fcp_icmd, 1266 (nCmd->rsplen + nCmd->cmdlen)); 1267 1268 /* Word 6 */ 1269 bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com, 1270 phba->sli4_hba.rpi_ids[pnode->nlp_rpi]); 1271 bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag); 1272 1273 /* Word 8 */ 1274 wqe->generic.wqe_com.abort_tag = pwqeq->iotag; 1275 1276 /* Word 9 */ 1277 bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag); 1278 1279 /* Words 13 14 15 are for PBDE support */ 1280 1281 pwqeq->vport = vport; 1282 return 0; 1283 } 1284 1285 1286 /** 1287 * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO 1288 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1289 * @lpfc_nvme_lport: Pointer to the driver's local port data 1290 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1291 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1292 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1293 * 1294 * Driver registers this routine as it io request handler. This 1295 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1296 * data structure to the rport indicated in @lpfc_nvme_rport. 1297 * 1298 * Return value : 1299 * 0 - Success 1300 * TODO: What are the failure codes. 1301 **/ 1302 static int 1303 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport, 1304 struct lpfc_nvme_buf *lpfc_ncmd) 1305 { 1306 struct lpfc_hba *phba = vport->phba; 1307 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1308 union lpfc_wqe128 *wqe = &lpfc_ncmd->cur_iocbq.wqe; 1309 struct sli4_sge *sgl = lpfc_ncmd->nvme_sgl; 1310 struct scatterlist *data_sg; 1311 struct sli4_sge *first_data_sgl; 1312 struct ulp_bde64 *bde; 1313 dma_addr_t physaddr; 1314 uint32_t num_bde = 0; 1315 uint32_t dma_len; 1316 uint32_t dma_offset = 0; 1317 int nseg, i; 1318 1319 /* Fix up the command and response DMA stuff. */ 1320 lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd); 1321 1322 /* 1323 * There are three possibilities here - use scatter-gather segment, use 1324 * the single mapping, or neither. 1325 */ 1326 if (nCmd->sg_cnt) { 1327 /* 1328 * Jump over the cmd and rsp SGEs. The fix routine 1329 * has already adjusted for this. 1330 */ 1331 sgl += 2; 1332 1333 first_data_sgl = sgl; 1334 lpfc_ncmd->seg_cnt = nCmd->sg_cnt; 1335 if (lpfc_ncmd->seg_cnt > lpfc_nvme_template.max_sgl_segments) { 1336 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1337 "6058 Too many sg segments from " 1338 "NVME Transport. Max %d, " 1339 "nvmeIO sg_cnt %d\n", 1340 phba->cfg_nvme_seg_cnt + 1, 1341 lpfc_ncmd->seg_cnt); 1342 lpfc_ncmd->seg_cnt = 0; 1343 return 1; 1344 } 1345 1346 /* 1347 * The driver established a maximum scatter-gather segment count 1348 * during probe that limits the number of sg elements in any 1349 * single nvme command. Just run through the seg_cnt and format 1350 * the sge's. 1351 */ 1352 nseg = nCmd->sg_cnt; 1353 data_sg = nCmd->first_sgl; 1354 for (i = 0; i < nseg; i++) { 1355 if (data_sg == NULL) { 1356 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1357 "6059 dptr err %d, nseg %d\n", 1358 i, nseg); 1359 lpfc_ncmd->seg_cnt = 0; 1360 return 1; 1361 } 1362 physaddr = data_sg->dma_address; 1363 dma_len = data_sg->length; 1364 sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr)); 1365 sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr)); 1366 sgl->word2 = le32_to_cpu(sgl->word2); 1367 if ((num_bde + 1) == nseg) 1368 bf_set(lpfc_sli4_sge_last, sgl, 1); 1369 else 1370 bf_set(lpfc_sli4_sge_last, sgl, 0); 1371 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset); 1372 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA); 1373 sgl->word2 = cpu_to_le32(sgl->word2); 1374 sgl->sge_len = cpu_to_le32(dma_len); 1375 1376 dma_offset += dma_len; 1377 data_sg = sg_next(data_sg); 1378 sgl++; 1379 } 1380 if (phba->cfg_enable_pbde) { 1381 /* Use PBDE support for first SGL only, offset == 0 */ 1382 /* Words 13-15 */ 1383 bde = (struct ulp_bde64 *) 1384 &wqe->words[13]; 1385 bde->addrLow = first_data_sgl->addr_lo; 1386 bde->addrHigh = first_data_sgl->addr_hi; 1387 bde->tus.f.bdeSize = 1388 le32_to_cpu(first_data_sgl->sge_len); 1389 bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64; 1390 bde->tus.w = cpu_to_le32(bde->tus.w); 1391 /* wqe_pbde is 1 in template */ 1392 } else { 1393 memset(&wqe->words[13], 0, (sizeof(uint32_t) * 3)); 1394 bf_set(wqe_pbde, &wqe->generic.wqe_com, 0); 1395 } 1396 1397 } else { 1398 /* For this clause to be valid, the payload_length 1399 * and sg_cnt must zero. 1400 */ 1401 if (nCmd->payload_length != 0) { 1402 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1403 "6063 NVME DMA Prep Err: sg_cnt %d " 1404 "payload_length x%x\n", 1405 nCmd->sg_cnt, nCmd->payload_length); 1406 return 1; 1407 } 1408 } 1409 return 0; 1410 } 1411 1412 /** 1413 * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO 1414 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1415 * @lpfc_nvme_lport: Pointer to the driver's local port data 1416 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1417 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1418 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1419 * 1420 * Driver registers this routine as it io request handler. This 1421 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1422 * data structure to the rport 1423 indicated in @lpfc_nvme_rport. 1424 * 1425 * Return value : 1426 * 0 - Success 1427 * TODO: What are the failure codes. 1428 **/ 1429 static int 1430 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport, 1431 struct nvme_fc_remote_port *pnvme_rport, 1432 void *hw_queue_handle, 1433 struct nvmefc_fcp_req *pnvme_fcreq) 1434 { 1435 int ret = 0; 1436 int expedite = 0; 1437 int idx; 1438 struct lpfc_nvme_lport *lport; 1439 struct lpfc_nvme_ctrl_stat *cstat; 1440 struct lpfc_vport *vport; 1441 struct lpfc_hba *phba; 1442 struct lpfc_nodelist *ndlp; 1443 struct lpfc_nvme_buf *lpfc_ncmd; 1444 struct lpfc_nvme_rport *rport; 1445 struct lpfc_nvme_qhandle *lpfc_queue_info; 1446 struct lpfc_nvme_fcpreq_priv *freqpriv; 1447 struct nvme_common_command *sqe; 1448 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1449 uint64_t start = 0; 1450 #endif 1451 1452 /* Validate pointers. LLDD fault handling with transport does 1453 * have timing races. 1454 */ 1455 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1456 if (unlikely(!lport)) { 1457 ret = -EINVAL; 1458 goto out_fail; 1459 } 1460 1461 vport = lport->vport; 1462 1463 if (unlikely(!hw_queue_handle)) { 1464 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1465 "6117 Fail IO, NULL hw_queue_handle\n"); 1466 atomic_inc(&lport->xmt_fcp_err); 1467 ret = -EBUSY; 1468 goto out_fail; 1469 } 1470 1471 phba = vport->phba; 1472 1473 if (vport->load_flag & FC_UNLOADING) { 1474 ret = -ENODEV; 1475 goto out_fail; 1476 } 1477 1478 if (vport->load_flag & FC_UNLOADING) { 1479 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1480 "6124 Fail IO, Driver unload\n"); 1481 atomic_inc(&lport->xmt_fcp_err); 1482 ret = -ENODEV; 1483 goto out_fail; 1484 } 1485 1486 freqpriv = pnvme_fcreq->private; 1487 if (unlikely(!freqpriv)) { 1488 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1489 "6158 Fail IO, NULL request data\n"); 1490 atomic_inc(&lport->xmt_fcp_err); 1491 ret = -EINVAL; 1492 goto out_fail; 1493 } 1494 1495 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1496 if (phba->ktime_on) 1497 start = ktime_get_ns(); 1498 #endif 1499 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 1500 lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle; 1501 1502 /* 1503 * Catch race where our node has transitioned, but the 1504 * transport is still transitioning. 1505 */ 1506 ndlp = rport->ndlp; 1507 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 1508 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1509 "6053 Fail IO, ndlp not ready: rport %p " 1510 "ndlp %p, DID x%06x\n", 1511 rport, ndlp, pnvme_rport->port_id); 1512 atomic_inc(&lport->xmt_fcp_err); 1513 ret = -EBUSY; 1514 goto out_fail; 1515 } 1516 1517 /* The remote node has to be a mapped target or it's an error. */ 1518 if ((ndlp->nlp_type & NLP_NVME_TARGET) && 1519 (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) { 1520 lpfc_printf_vlog(vport, KERN_INFO, LOG_NODE | LOG_NVME_IOERR, 1521 "6036 Fail IO, DID x%06x not ready for " 1522 "IO. State x%x, Type x%x Flg x%x\n", 1523 pnvme_rport->port_id, 1524 ndlp->nlp_state, ndlp->nlp_type, 1525 ndlp->upcall_flags); 1526 atomic_inc(&lport->xmt_fcp_bad_ndlp); 1527 ret = -EBUSY; 1528 goto out_fail; 1529 1530 } 1531 1532 /* Currently only NVME Keep alive commands should be expedited 1533 * if the driver runs out of a resource. These should only be 1534 * issued on the admin queue, qidx 0 1535 */ 1536 if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) { 1537 sqe = &((struct nvme_fc_cmd_iu *) 1538 pnvme_fcreq->cmdaddr)->sqe.common; 1539 if (sqe->opcode == nvme_admin_keep_alive) 1540 expedite = 1; 1541 } 1542 1543 /* The node is shared with FCP IO, make sure the IO pending count does 1544 * not exceed the programmed depth. 1545 */ 1546 if (lpfc_ndlp_check_qdepth(phba, ndlp)) { 1547 if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) && 1548 !expedite) { 1549 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1550 "6174 Fail IO, ndlp qdepth exceeded: " 1551 "idx %d DID %x pend %d qdepth %d\n", 1552 lpfc_queue_info->index, ndlp->nlp_DID, 1553 atomic_read(&ndlp->cmd_pending), 1554 ndlp->cmd_qdepth); 1555 atomic_inc(&lport->xmt_fcp_qdepth); 1556 ret = -EBUSY; 1557 goto out_fail; 1558 } 1559 } 1560 1561 lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, expedite); 1562 if (lpfc_ncmd == NULL) { 1563 atomic_inc(&lport->xmt_fcp_noxri); 1564 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1565 "6065 Fail IO, driver buffer pool is empty: " 1566 "idx %d DID %x\n", 1567 lpfc_queue_info->index, ndlp->nlp_DID); 1568 ret = -EBUSY; 1569 goto out_fail; 1570 } 1571 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1572 if (start) { 1573 lpfc_ncmd->ts_cmd_start = start; 1574 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd; 1575 } else { 1576 lpfc_ncmd->ts_cmd_start = 0; 1577 } 1578 #endif 1579 1580 /* 1581 * Store the data needed by the driver to issue, abort, and complete 1582 * an IO. 1583 * Do not let the IO hang out forever. There is no midlayer issuing 1584 * an abort so inform the FW of the maximum IO pending time. 1585 */ 1586 freqpriv->nvme_buf = lpfc_ncmd; 1587 lpfc_ncmd->nvmeCmd = pnvme_fcreq; 1588 lpfc_ncmd->nrport = rport; 1589 lpfc_ncmd->ndlp = ndlp; 1590 lpfc_ncmd->start_time = jiffies; 1591 1592 /* 1593 * Issue the IO on the WQ indicated by index in the hw_queue_handle. 1594 * This identfier was create in our hardware queue create callback 1595 * routine. The driver now is dependent on the IO queue steering from 1596 * the transport. We are trusting the upper NVME layers know which 1597 * index to use and that they have affinitized a CPU to this hardware 1598 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ. 1599 */ 1600 idx = lpfc_queue_info->index; 1601 lpfc_ncmd->cur_iocbq.hba_wqidx = idx; 1602 cstat = &lport->cstat[idx]; 1603 1604 lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp, cstat); 1605 ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd); 1606 if (ret) { 1607 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1608 "6175 Fail IO, Prep DMA: " 1609 "idx %d DID %x\n", 1610 lpfc_queue_info->index, ndlp->nlp_DID); 1611 atomic_inc(&lport->xmt_fcp_err); 1612 ret = -ENOMEM; 1613 goto out_free_nvme_buf; 1614 } 1615 1616 lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n", 1617 lpfc_ncmd->cur_iocbq.sli4_xritag, 1618 lpfc_queue_info->index, ndlp->nlp_DID); 1619 1620 ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq); 1621 if (ret) { 1622 atomic_inc(&lport->xmt_fcp_wqerr); 1623 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1624 "6113 Fail IO, Could not issue WQE err %x " 1625 "sid: x%x did: x%x oxid: x%x\n", 1626 ret, vport->fc_myDID, ndlp->nlp_DID, 1627 lpfc_ncmd->cur_iocbq.sli4_xritag); 1628 goto out_free_nvme_buf; 1629 } 1630 1631 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1632 if (lpfc_ncmd->ts_cmd_start) 1633 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns(); 1634 1635 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1636 lpfc_ncmd->cpu = smp_processor_id(); 1637 if (lpfc_ncmd->cpu != lpfc_queue_info->index) { 1638 /* Check for admin queue */ 1639 if (lpfc_queue_info->qidx) { 1640 lpfc_printf_vlog(vport, 1641 KERN_ERR, LOG_NVME_IOERR, 1642 "6702 CPU Check cmd: " 1643 "cpu %d wq %d\n", 1644 lpfc_ncmd->cpu, 1645 lpfc_queue_info->index); 1646 } 1647 lpfc_ncmd->cpu = lpfc_queue_info->index; 1648 } 1649 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT) 1650 phba->cpucheck_xmt_io[lpfc_ncmd->cpu]++; 1651 } 1652 #endif 1653 return 0; 1654 1655 out_free_nvme_buf: 1656 if (lpfc_ncmd->nvmeCmd->sg_cnt) { 1657 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE) 1658 atomic_dec(&cstat->fc4NvmeOutputRequests); 1659 else 1660 atomic_dec(&cstat->fc4NvmeInputRequests); 1661 } else 1662 atomic_dec(&cstat->fc4NvmeControlRequests); 1663 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1664 out_fail: 1665 return ret; 1666 } 1667 1668 /** 1669 * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request. 1670 * @phba: Pointer to HBA context object 1671 * @cmdiocb: Pointer to command iocb object. 1672 * @rspiocb: Pointer to response iocb object. 1673 * 1674 * This is the callback function for any NVME FCP IO that was aborted. 1675 * 1676 * Return value: 1677 * None 1678 **/ 1679 void 1680 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb, 1681 struct lpfc_wcqe_complete *abts_cmpl) 1682 { 1683 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 1684 "6145 ABORT_XRI_CN completing on rpi x%x " 1685 "original iotag x%x, abort cmd iotag x%x " 1686 "req_tag x%x, status x%x, hwstatus x%x\n", 1687 cmdiocb->iocb.un.acxri.abortContextTag, 1688 cmdiocb->iocb.un.acxri.abortIoTag, 1689 cmdiocb->iotag, 1690 bf_get(lpfc_wcqe_c_request_tag, abts_cmpl), 1691 bf_get(lpfc_wcqe_c_status, abts_cmpl), 1692 bf_get(lpfc_wcqe_c_hw_status, abts_cmpl)); 1693 lpfc_sli_release_iocbq(phba, cmdiocb); 1694 } 1695 1696 /** 1697 * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS 1698 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1699 * @lpfc_nvme_lport: Pointer to the driver's local port data 1700 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1701 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1702 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1703 * 1704 * Driver registers this routine as its nvme request io abort handler. This 1705 * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq 1706 * data structure to the rport indicated in @lpfc_nvme_rport. This routine 1707 * is executed asynchronously - one the target is validated as "MAPPED" and 1708 * ready for IO, the driver issues the abort request and returns. 1709 * 1710 * Return value: 1711 * None 1712 **/ 1713 static void 1714 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport, 1715 struct nvme_fc_remote_port *pnvme_rport, 1716 void *hw_queue_handle, 1717 struct nvmefc_fcp_req *pnvme_fcreq) 1718 { 1719 struct lpfc_nvme_lport *lport; 1720 struct lpfc_vport *vport; 1721 struct lpfc_hba *phba; 1722 struct lpfc_nvme_buf *lpfc_nbuf; 1723 struct lpfc_iocbq *abts_buf; 1724 struct lpfc_iocbq *nvmereq_wqe; 1725 struct lpfc_nvme_fcpreq_priv *freqpriv; 1726 union lpfc_wqe128 *abts_wqe; 1727 unsigned long flags; 1728 int ret_val; 1729 1730 /* Validate pointers. LLDD fault handling with transport does 1731 * have timing races. 1732 */ 1733 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1734 if (unlikely(!lport)) 1735 return; 1736 1737 vport = lport->vport; 1738 1739 if (unlikely(!hw_queue_handle)) { 1740 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1741 "6129 Fail Abort, HW Queue Handle NULL.\n"); 1742 return; 1743 } 1744 1745 phba = vport->phba; 1746 freqpriv = pnvme_fcreq->private; 1747 1748 if (unlikely(!freqpriv)) 1749 return; 1750 if (vport->load_flag & FC_UNLOADING) 1751 return; 1752 1753 /* Announce entry to new IO submit field. */ 1754 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1755 "6002 Abort Request to rport DID x%06x " 1756 "for nvme_fc_req %p\n", 1757 pnvme_rport->port_id, 1758 pnvme_fcreq); 1759 1760 /* If the hba is getting reset, this flag is set. It is 1761 * cleared when the reset is complete and rings reestablished. 1762 */ 1763 spin_lock_irqsave(&phba->hbalock, flags); 1764 /* driver queued commands are in process of being flushed */ 1765 if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) { 1766 spin_unlock_irqrestore(&phba->hbalock, flags); 1767 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1768 "6139 Driver in reset cleanup - flushing " 1769 "NVME Req now. hba_flag x%x\n", 1770 phba->hba_flag); 1771 return; 1772 } 1773 1774 lpfc_nbuf = freqpriv->nvme_buf; 1775 if (!lpfc_nbuf) { 1776 spin_unlock_irqrestore(&phba->hbalock, flags); 1777 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1778 "6140 NVME IO req has no matching lpfc nvme " 1779 "io buffer. Skipping abort req.\n"); 1780 return; 1781 } else if (!lpfc_nbuf->nvmeCmd) { 1782 spin_unlock_irqrestore(&phba->hbalock, flags); 1783 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1784 "6141 lpfc NVME IO req has no nvme_fcreq " 1785 "io buffer. Skipping abort req.\n"); 1786 return; 1787 } 1788 nvmereq_wqe = &lpfc_nbuf->cur_iocbq; 1789 1790 /* 1791 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's 1792 * state must match the nvme_fcreq passed by the nvme 1793 * transport. If they don't match, it is likely the driver 1794 * has already completed the NVME IO and the nvme transport 1795 * has not seen it yet. 1796 */ 1797 if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) { 1798 spin_unlock_irqrestore(&phba->hbalock, flags); 1799 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1800 "6143 NVME req mismatch: " 1801 "lpfc_nbuf %p nvmeCmd %p, " 1802 "pnvme_fcreq %p. Skipping Abort xri x%x\n", 1803 lpfc_nbuf, lpfc_nbuf->nvmeCmd, 1804 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1805 return; 1806 } 1807 1808 /* Don't abort IOs no longer on the pending queue. */ 1809 if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) { 1810 spin_unlock_irqrestore(&phba->hbalock, flags); 1811 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1812 "6142 NVME IO req %p not queued - skipping " 1813 "abort req xri x%x\n", 1814 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1815 return; 1816 } 1817 1818 atomic_inc(&lport->xmt_fcp_abort); 1819 lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n", 1820 nvmereq_wqe->sli4_xritag, 1821 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id); 1822 1823 /* Outstanding abort is in progress */ 1824 if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) { 1825 spin_unlock_irqrestore(&phba->hbalock, flags); 1826 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1827 "6144 Outstanding NVME I/O Abort Request " 1828 "still pending on nvme_fcreq %p, " 1829 "lpfc_ncmd %p xri x%x\n", 1830 pnvme_fcreq, lpfc_nbuf, 1831 nvmereq_wqe->sli4_xritag); 1832 return; 1833 } 1834 1835 abts_buf = __lpfc_sli_get_iocbq(phba); 1836 if (!abts_buf) { 1837 spin_unlock_irqrestore(&phba->hbalock, flags); 1838 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1839 "6136 No available abort wqes. Skipping " 1840 "Abts req for nvme_fcreq %p xri x%x\n", 1841 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1842 return; 1843 } 1844 1845 /* Ready - mark outstanding as aborted by driver. */ 1846 nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED; 1847 1848 /* Complete prepping the abort wqe and issue to the FW. */ 1849 abts_wqe = &abts_buf->wqe; 1850 1851 /* WQEs are reused. Clear stale data and set key fields to 1852 * zero like ia, iaab, iaar, xri_tag, and ctxt_tag. 1853 */ 1854 memset(abts_wqe, 0, sizeof(union lpfc_wqe)); 1855 bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG); 1856 1857 /* word 7 */ 1858 bf_set(wqe_ct, &abts_wqe->abort_cmd.wqe_com, 0); 1859 bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX); 1860 bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com, 1861 nvmereq_wqe->iocb.ulpClass); 1862 1863 /* word 8 - tell the FW to abort the IO associated with this 1864 * outstanding exchange ID. 1865 */ 1866 abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag; 1867 1868 /* word 9 - this is the iotag for the abts_wqe completion. */ 1869 bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com, 1870 abts_buf->iotag); 1871 1872 /* word 10 */ 1873 bf_set(wqe_wqid, &abts_wqe->abort_cmd.wqe_com, nvmereq_wqe->hba_wqidx); 1874 bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1); 1875 bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE); 1876 1877 /* word 11 */ 1878 bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND); 1879 bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1); 1880 bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 1881 1882 /* ABTS WQE must go to the same WQ as the WQE to be aborted */ 1883 abts_buf->iocb_flag |= LPFC_IO_NVME; 1884 abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx; 1885 abts_buf->vport = vport; 1886 abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl; 1887 ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf); 1888 spin_unlock_irqrestore(&phba->hbalock, flags); 1889 if (ret_val) { 1890 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1891 "6137 Failed abts issue_wqe with status x%x " 1892 "for nvme_fcreq %p.\n", 1893 ret_val, pnvme_fcreq); 1894 lpfc_sli_release_iocbq(phba, abts_buf); 1895 return; 1896 } 1897 1898 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1899 "6138 Transport Abort NVME Request Issued for " 1900 "ox_id x%x on reqtag x%x\n", 1901 nvmereq_wqe->sli4_xritag, 1902 abts_buf->iotag); 1903 } 1904 1905 /* Declare and initialization an instance of the FC NVME template. */ 1906 static struct nvme_fc_port_template lpfc_nvme_template = { 1907 /* initiator-based functions */ 1908 .localport_delete = lpfc_nvme_localport_delete, 1909 .remoteport_delete = lpfc_nvme_remoteport_delete, 1910 .create_queue = lpfc_nvme_create_queue, 1911 .delete_queue = lpfc_nvme_delete_queue, 1912 .ls_req = lpfc_nvme_ls_req, 1913 .fcp_io = lpfc_nvme_fcp_io_submit, 1914 .ls_abort = lpfc_nvme_ls_abort, 1915 .fcp_abort = lpfc_nvme_fcp_abort, 1916 1917 .max_hw_queues = 1, 1918 .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1919 .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1920 .dma_boundary = 0xFFFFFFFF, 1921 1922 /* Sizes of additional private data for data structures. 1923 * No use for the last two sizes at this time. 1924 */ 1925 .local_priv_sz = sizeof(struct lpfc_nvme_lport), 1926 .remote_priv_sz = sizeof(struct lpfc_nvme_rport), 1927 .lsrqst_priv_sz = 0, 1928 .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv), 1929 }; 1930 1931 /** 1932 * lpfc_sli4_post_nvme_sgl_block - post a block of nvme sgl list to firmware 1933 * @phba: pointer to lpfc hba data structure. 1934 * @nblist: pointer to nvme buffer list. 1935 * @count: number of scsi buffers on the list. 1936 * 1937 * This routine is invoked to post a block of @count scsi sgl pages from a 1938 * SCSI buffer list @nblist to the HBA using non-embedded mailbox command. 1939 * No Lock is held. 1940 * 1941 **/ 1942 static int 1943 lpfc_sli4_post_nvme_sgl_block(struct lpfc_hba *phba, 1944 struct list_head *nblist, 1945 int count) 1946 { 1947 struct lpfc_nvme_buf *lpfc_ncmd; 1948 struct lpfc_mbx_post_uembed_sgl_page1 *sgl; 1949 struct sgl_page_pairs *sgl_pg_pairs; 1950 void *viraddr; 1951 LPFC_MBOXQ_t *mbox; 1952 uint32_t reqlen, alloclen, pg_pairs; 1953 uint32_t mbox_tmo; 1954 uint16_t xritag_start = 0; 1955 int rc = 0; 1956 uint32_t shdr_status, shdr_add_status; 1957 dma_addr_t pdma_phys_bpl1; 1958 union lpfc_sli4_cfg_shdr *shdr; 1959 1960 /* Calculate the requested length of the dma memory */ 1961 reqlen = count * sizeof(struct sgl_page_pairs) + 1962 sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t); 1963 if (reqlen > SLI4_PAGE_SIZE) { 1964 lpfc_printf_log(phba, KERN_WARNING, LOG_INIT, 1965 "6118 Block sgl registration required DMA " 1966 "size (%d) great than a page\n", reqlen); 1967 return -ENOMEM; 1968 } 1969 mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL); 1970 if (!mbox) { 1971 lpfc_printf_log(phba, KERN_ERR, LOG_INIT, 1972 "6119 Failed to allocate mbox cmd memory\n"); 1973 return -ENOMEM; 1974 } 1975 1976 /* Allocate DMA memory and set up the non-embedded mailbox command */ 1977 alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE, 1978 LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen, 1979 LPFC_SLI4_MBX_NEMBED); 1980 1981 if (alloclen < reqlen) { 1982 lpfc_printf_log(phba, KERN_ERR, LOG_INIT, 1983 "6120 Allocated DMA memory size (%d) is " 1984 "less than the requested DMA memory " 1985 "size (%d)\n", alloclen, reqlen); 1986 lpfc_sli4_mbox_cmd_free(phba, mbox); 1987 return -ENOMEM; 1988 } 1989 1990 /* Get the first SGE entry from the non-embedded DMA memory */ 1991 viraddr = mbox->sge_array->addr[0]; 1992 1993 /* Set up the SGL pages in the non-embedded DMA pages */ 1994 sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr; 1995 sgl_pg_pairs = &sgl->sgl_pg_pairs; 1996 1997 pg_pairs = 0; 1998 list_for_each_entry(lpfc_ncmd, nblist, list) { 1999 /* Set up the sge entry */ 2000 sgl_pg_pairs->sgl_pg0_addr_lo = 2001 cpu_to_le32(putPaddrLow(lpfc_ncmd->dma_phys_sgl)); 2002 sgl_pg_pairs->sgl_pg0_addr_hi = 2003 cpu_to_le32(putPaddrHigh(lpfc_ncmd->dma_phys_sgl)); 2004 if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE) 2005 pdma_phys_bpl1 = lpfc_ncmd->dma_phys_sgl + 2006 SGL_PAGE_SIZE; 2007 else 2008 pdma_phys_bpl1 = 0; 2009 sgl_pg_pairs->sgl_pg1_addr_lo = 2010 cpu_to_le32(putPaddrLow(pdma_phys_bpl1)); 2011 sgl_pg_pairs->sgl_pg1_addr_hi = 2012 cpu_to_le32(putPaddrHigh(pdma_phys_bpl1)); 2013 /* Keep the first xritag on the list */ 2014 if (pg_pairs == 0) 2015 xritag_start = lpfc_ncmd->cur_iocbq.sli4_xritag; 2016 sgl_pg_pairs++; 2017 pg_pairs++; 2018 } 2019 bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start); 2020 bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs); 2021 /* Perform endian conversion if necessary */ 2022 sgl->word0 = cpu_to_le32(sgl->word0); 2023 2024 if (!phba->sli4_hba.intr_enable) 2025 rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL); 2026 else { 2027 mbox_tmo = lpfc_mbox_tmo_val(phba, mbox); 2028 rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo); 2029 } 2030 shdr = (union lpfc_sli4_cfg_shdr *)&sgl->cfg_shdr; 2031 shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response); 2032 shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response); 2033 if (rc != MBX_TIMEOUT) 2034 lpfc_sli4_mbox_cmd_free(phba, mbox); 2035 if (shdr_status || shdr_add_status || rc) { 2036 lpfc_printf_log(phba, KERN_ERR, LOG_SLI, 2037 "6125 POST_SGL_BLOCK mailbox command failed " 2038 "status x%x add_status x%x mbx status x%x\n", 2039 shdr_status, shdr_add_status, rc); 2040 rc = -ENXIO; 2041 } 2042 return rc; 2043 } 2044 2045 /** 2046 * lpfc_post_nvme_sgl_list - Post blocks of nvme buffer sgls from a list 2047 * @phba: pointer to lpfc hba data structure. 2048 * @post_nblist: pointer to the nvme buffer list. 2049 * 2050 * This routine walks a list of nvme buffers that was passed in. It attempts 2051 * to construct blocks of nvme buffer sgls which contains contiguous xris and 2052 * uses the non-embedded SGL block post mailbox commands to post to the port. 2053 * For single NVME buffer sgl with non-contiguous xri, if any, it shall use 2054 * embedded SGL post mailbox command for posting. The @post_nblist passed in 2055 * must be local list, thus no lock is needed when manipulate the list. 2056 * 2057 * Returns: 0 = failure, non-zero number of successfully posted buffers. 2058 **/ 2059 static int 2060 lpfc_post_nvme_sgl_list(struct lpfc_hba *phba, 2061 struct list_head *post_nblist, int sb_count) 2062 { 2063 struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next; 2064 int status, sgl_size; 2065 int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0; 2066 dma_addr_t pdma_phys_sgl1; 2067 int last_xritag = NO_XRI; 2068 int cur_xritag; 2069 LIST_HEAD(prep_nblist); 2070 LIST_HEAD(blck_nblist); 2071 LIST_HEAD(nvme_nblist); 2072 2073 /* sanity check */ 2074 if (sb_count <= 0) 2075 return -EINVAL; 2076 2077 sgl_size = phba->cfg_sg_dma_buf_size; 2078 2079 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, post_nblist, list) { 2080 list_del_init(&lpfc_ncmd->list); 2081 block_cnt++; 2082 if ((last_xritag != NO_XRI) && 2083 (lpfc_ncmd->cur_iocbq.sli4_xritag != last_xritag + 1)) { 2084 /* a hole in xri block, form a sgl posting block */ 2085 list_splice_init(&prep_nblist, &blck_nblist); 2086 post_cnt = block_cnt - 1; 2087 /* prepare list for next posting block */ 2088 list_add_tail(&lpfc_ncmd->list, &prep_nblist); 2089 block_cnt = 1; 2090 } else { 2091 /* prepare list for next posting block */ 2092 list_add_tail(&lpfc_ncmd->list, &prep_nblist); 2093 /* enough sgls for non-embed sgl mbox command */ 2094 if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) { 2095 list_splice_init(&prep_nblist, &blck_nblist); 2096 post_cnt = block_cnt; 2097 block_cnt = 0; 2098 } 2099 } 2100 num_posting++; 2101 last_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag; 2102 2103 /* end of repost sgl list condition for NVME buffers */ 2104 if (num_posting == sb_count) { 2105 if (post_cnt == 0) { 2106 /* last sgl posting block */ 2107 list_splice_init(&prep_nblist, &blck_nblist); 2108 post_cnt = block_cnt; 2109 } else if (block_cnt == 1) { 2110 /* last single sgl with non-contiguous xri */ 2111 if (sgl_size > SGL_PAGE_SIZE) 2112 pdma_phys_sgl1 = 2113 lpfc_ncmd->dma_phys_sgl + 2114 SGL_PAGE_SIZE; 2115 else 2116 pdma_phys_sgl1 = 0; 2117 cur_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag; 2118 status = lpfc_sli4_post_sgl(phba, 2119 lpfc_ncmd->dma_phys_sgl, 2120 pdma_phys_sgl1, cur_xritag); 2121 if (status) { 2122 /* failure, put on abort nvme list */ 2123 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 2124 } else { 2125 /* success, put on NVME buffer list */ 2126 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2127 lpfc_ncmd->status = IOSTAT_SUCCESS; 2128 num_posted++; 2129 } 2130 /* success, put on NVME buffer sgl list */ 2131 list_add_tail(&lpfc_ncmd->list, &nvme_nblist); 2132 } 2133 } 2134 2135 /* continue until a nembed page worth of sgls */ 2136 if (post_cnt == 0) 2137 continue; 2138 2139 /* post block of NVME buffer list sgls */ 2140 status = lpfc_sli4_post_nvme_sgl_block(phba, &blck_nblist, 2141 post_cnt); 2142 2143 /* don't reset xirtag due to hole in xri block */ 2144 if (block_cnt == 0) 2145 last_xritag = NO_XRI; 2146 2147 /* reset NVME buffer post count for next round of posting */ 2148 post_cnt = 0; 2149 2150 /* put posted NVME buffer-sgl posted on NVME buffer sgl list */ 2151 while (!list_empty(&blck_nblist)) { 2152 list_remove_head(&blck_nblist, lpfc_ncmd, 2153 struct lpfc_nvme_buf, list); 2154 if (status) { 2155 /* failure, put on abort nvme list */ 2156 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 2157 } else { 2158 /* success, put on NVME buffer list */ 2159 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2160 lpfc_ncmd->status = IOSTAT_SUCCESS; 2161 num_posted++; 2162 } 2163 list_add_tail(&lpfc_ncmd->list, &nvme_nblist); 2164 } 2165 } 2166 /* Push NVME buffers with sgl posted to the available list */ 2167 while (!list_empty(&nvme_nblist)) { 2168 list_remove_head(&nvme_nblist, lpfc_ncmd, 2169 struct lpfc_nvme_buf, list); 2170 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2171 } 2172 return num_posted; 2173 } 2174 2175 /** 2176 * lpfc_repost_nvme_sgl_list - Repost all the allocated nvme buffer sgls 2177 * @phba: pointer to lpfc hba data structure. 2178 * 2179 * This routine walks the list of nvme buffers that have been allocated and 2180 * repost them to the port by using SGL block post. This is needed after a 2181 * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine 2182 * is responsible for moving all nvme buffers on the lpfc_abts_nvme_sgl_list 2183 * to the lpfc_nvme_buf_list. If the repost fails, reject all nvme buffers. 2184 * 2185 * Returns: 0 = success, non-zero failure. 2186 **/ 2187 int 2188 lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba) 2189 { 2190 LIST_HEAD(post_nblist); 2191 int num_posted, rc = 0; 2192 2193 /* get all NVME buffers need to repost to a local list */ 2194 spin_lock_irq(&phba->nvme_buf_list_get_lock); 2195 spin_lock(&phba->nvme_buf_list_put_lock); 2196 list_splice_init(&phba->lpfc_nvme_buf_list_get, &post_nblist); 2197 list_splice(&phba->lpfc_nvme_buf_list_put, &post_nblist); 2198 phba->get_nvme_bufs = 0; 2199 phba->put_nvme_bufs = 0; 2200 spin_unlock(&phba->nvme_buf_list_put_lock); 2201 spin_unlock_irq(&phba->nvme_buf_list_get_lock); 2202 2203 /* post the list of nvme buffer sgls to port if available */ 2204 if (!list_empty(&post_nblist)) { 2205 num_posted = lpfc_post_nvme_sgl_list(phba, &post_nblist, 2206 phba->sli4_hba.nvme_xri_cnt); 2207 /* failed to post any nvme buffer, return error */ 2208 if (num_posted == 0) 2209 rc = -EIO; 2210 } 2211 return rc; 2212 } 2213 2214 /** 2215 * lpfc_new_nvme_buf - Scsi buffer allocator for HBA with SLI4 IF spec 2216 * @vport: The virtual port for which this call being executed. 2217 * @num_to_allocate: The requested number of buffers to allocate. 2218 * 2219 * This routine allocates nvme buffers for device with SLI-4 interface spec, 2220 * the nvme buffer contains all the necessary information needed to initiate 2221 * a NVME I/O. After allocating up to @num_to_allocate NVME buffers and put 2222 * them on a list, it post them to the port by using SGL block post. 2223 * 2224 * Return codes: 2225 * int - number of nvme buffers that were allocated and posted. 2226 * 0 = failure, less than num_to_alloc is a partial failure. 2227 **/ 2228 static int 2229 lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc) 2230 { 2231 struct lpfc_hba *phba = vport->phba; 2232 struct lpfc_nvme_buf *lpfc_ncmd; 2233 struct lpfc_iocbq *pwqeq; 2234 union lpfc_wqe128 *wqe; 2235 struct sli4_sge *sgl; 2236 dma_addr_t pdma_phys_sgl; 2237 uint16_t iotag, lxri = 0; 2238 int bcnt, num_posted, sgl_size; 2239 LIST_HEAD(prep_nblist); 2240 LIST_HEAD(post_nblist); 2241 LIST_HEAD(nvme_nblist); 2242 2243 sgl_size = phba->cfg_sg_dma_buf_size; 2244 2245 for (bcnt = 0; bcnt < num_to_alloc; bcnt++) { 2246 lpfc_ncmd = kzalloc(sizeof(struct lpfc_nvme_buf), GFP_KERNEL); 2247 if (!lpfc_ncmd) 2248 break; 2249 /* 2250 * Get memory from the pci pool to map the virt space to 2251 * pci bus space for an I/O. The DMA buffer includes the 2252 * number of SGE's necessary to support the sg_tablesize. 2253 */ 2254 lpfc_ncmd->data = dma_pool_zalloc(phba->lpfc_sg_dma_buf_pool, 2255 GFP_KERNEL, 2256 &lpfc_ncmd->dma_handle); 2257 if (!lpfc_ncmd->data) { 2258 kfree(lpfc_ncmd); 2259 break; 2260 } 2261 2262 lxri = lpfc_sli4_next_xritag(phba); 2263 if (lxri == NO_XRI) { 2264 dma_pool_free(phba->lpfc_sg_dma_buf_pool, 2265 lpfc_ncmd->data, lpfc_ncmd->dma_handle); 2266 kfree(lpfc_ncmd); 2267 break; 2268 } 2269 pwqeq = &(lpfc_ncmd->cur_iocbq); 2270 wqe = &pwqeq->wqe; 2271 2272 /* Allocate iotag for lpfc_ncmd->cur_iocbq. */ 2273 iotag = lpfc_sli_next_iotag(phba, pwqeq); 2274 if (iotag == 0) { 2275 dma_pool_free(phba->lpfc_sg_dma_buf_pool, 2276 lpfc_ncmd->data, lpfc_ncmd->dma_handle); 2277 kfree(lpfc_ncmd); 2278 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 2279 "6121 Failed to allocated IOTAG for" 2280 " XRI:0x%x\n", lxri); 2281 lpfc_sli4_free_xri(phba, lxri); 2282 break; 2283 } 2284 pwqeq->sli4_lxritag = lxri; 2285 pwqeq->sli4_xritag = phba->sli4_hba.xri_ids[lxri]; 2286 pwqeq->iocb_flag |= LPFC_IO_NVME; 2287 pwqeq->context1 = lpfc_ncmd; 2288 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl; 2289 2290 /* Initialize local short-hand pointers. */ 2291 lpfc_ncmd->nvme_sgl = lpfc_ncmd->data; 2292 sgl = lpfc_ncmd->nvme_sgl; 2293 pdma_phys_sgl = lpfc_ncmd->dma_handle; 2294 lpfc_ncmd->dma_phys_sgl = pdma_phys_sgl; 2295 2296 /* Rsp SGE will be filled in when we rcv an IO 2297 * from the NVME Layer to be sent. 2298 * The cmd is going to be embedded so we need a SKIP SGE. 2299 */ 2300 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP); 2301 bf_set(lpfc_sli4_sge_last, sgl, 0); 2302 sgl->word2 = cpu_to_le32(sgl->word2); 2303 /* Fill in word 3 / sgl_len during cmd submission */ 2304 2305 lpfc_ncmd->cur_iocbq.context1 = lpfc_ncmd; 2306 2307 /* Initialize WQE */ 2308 memset(wqe, 0, sizeof(union lpfc_wqe)); 2309 2310 /* add the nvme buffer to a post list */ 2311 list_add_tail(&lpfc_ncmd->list, &post_nblist); 2312 spin_lock_irq(&phba->nvme_buf_list_get_lock); 2313 phba->sli4_hba.nvme_xri_cnt++; 2314 spin_unlock_irq(&phba->nvme_buf_list_get_lock); 2315 } 2316 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 2317 "6114 Allocate %d out of %d requested new NVME " 2318 "buffers\n", bcnt, num_to_alloc); 2319 2320 /* post the list of nvme buffer sgls to port if available */ 2321 if (!list_empty(&post_nblist)) 2322 num_posted = lpfc_post_nvme_sgl_list(phba, 2323 &post_nblist, bcnt); 2324 else 2325 num_posted = 0; 2326 2327 return num_posted; 2328 } 2329 2330 static inline struct lpfc_nvme_buf * 2331 lpfc_nvme_buf(struct lpfc_hba *phba) 2332 { 2333 struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next; 2334 2335 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, 2336 &phba->lpfc_nvme_buf_list_get, list) { 2337 list_del_init(&lpfc_ncmd->list); 2338 phba->get_nvme_bufs--; 2339 return lpfc_ncmd; 2340 } 2341 return NULL; 2342 } 2343 2344 /** 2345 * lpfc_get_nvme_buf - Get a nvme buffer from lpfc_nvme_buf_list of the HBA 2346 * @phba: The HBA for which this call is being executed. 2347 * 2348 * This routine removes a nvme buffer from head of @phba lpfc_nvme_buf_list list 2349 * and returns to caller. 2350 * 2351 * Return codes: 2352 * NULL - Error 2353 * Pointer to lpfc_nvme_buf - Success 2354 **/ 2355 static struct lpfc_nvme_buf * 2356 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp, 2357 int expedite) 2358 { 2359 struct lpfc_nvme_buf *lpfc_ncmd = NULL; 2360 unsigned long iflag = 0; 2361 2362 spin_lock_irqsave(&phba->nvme_buf_list_get_lock, iflag); 2363 if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite) 2364 lpfc_ncmd = lpfc_nvme_buf(phba); 2365 if (!lpfc_ncmd) { 2366 spin_lock(&phba->nvme_buf_list_put_lock); 2367 list_splice(&phba->lpfc_nvme_buf_list_put, 2368 &phba->lpfc_nvme_buf_list_get); 2369 phba->get_nvme_bufs += phba->put_nvme_bufs; 2370 INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put); 2371 phba->put_nvme_bufs = 0; 2372 spin_unlock(&phba->nvme_buf_list_put_lock); 2373 if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite) 2374 lpfc_ncmd = lpfc_nvme_buf(phba); 2375 } 2376 spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag); 2377 2378 if (lpfc_ndlp_check_qdepth(phba, ndlp) && lpfc_ncmd) { 2379 atomic_inc(&ndlp->cmd_pending); 2380 lpfc_ncmd->flags |= LPFC_BUMP_QDEPTH; 2381 } 2382 return lpfc_ncmd; 2383 } 2384 2385 /** 2386 * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list. 2387 * @phba: The Hba for which this call is being executed. 2388 * @lpfc_ncmd: The nvme buffer which is being released. 2389 * 2390 * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba 2391 * lpfc_nvme_buf_list list. For SLI4 XRI's are tied to the nvme buffer 2392 * and cannot be reused for at least RA_TOV amount of time if it was 2393 * aborted. 2394 **/ 2395 static void 2396 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd) 2397 { 2398 unsigned long iflag = 0; 2399 2400 if ((lpfc_ncmd->flags & LPFC_BUMP_QDEPTH) && lpfc_ncmd->ndlp) 2401 atomic_dec(&lpfc_ncmd->ndlp->cmd_pending); 2402 2403 lpfc_ncmd->nonsg_phys = 0; 2404 lpfc_ncmd->ndlp = NULL; 2405 lpfc_ncmd->flags &= ~LPFC_BUMP_QDEPTH; 2406 2407 if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) { 2408 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2409 "6310 XB release deferred for " 2410 "ox_id x%x on reqtag x%x\n", 2411 lpfc_ncmd->cur_iocbq.sli4_xritag, 2412 lpfc_ncmd->cur_iocbq.iotag); 2413 2414 spin_lock_irqsave(&phba->sli4_hba.abts_nvme_buf_list_lock, 2415 iflag); 2416 list_add_tail(&lpfc_ncmd->list, 2417 &phba->sli4_hba.lpfc_abts_nvme_buf_list); 2418 spin_unlock_irqrestore(&phba->sli4_hba.abts_nvme_buf_list_lock, 2419 iflag); 2420 } else { 2421 lpfc_ncmd->nvmeCmd = NULL; 2422 lpfc_ncmd->cur_iocbq.iocb_flag = LPFC_IO_NVME; 2423 spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag); 2424 list_add_tail(&lpfc_ncmd->list, &phba->lpfc_nvme_buf_list_put); 2425 phba->put_nvme_bufs++; 2426 spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag); 2427 } 2428 } 2429 2430 /** 2431 * lpfc_nvme_create_localport - Create/Bind an nvme localport instance. 2432 * @pvport - the lpfc_vport instance requesting a localport. 2433 * 2434 * This routine is invoked to create an nvme localport instance to bind 2435 * to the nvme_fc_transport. It is called once during driver load 2436 * like lpfc_create_shost after all other services are initialized. 2437 * It requires a vport, vpi, and wwns at call time. Other localport 2438 * parameters are modified as the driver's FCID and the Fabric WWN 2439 * are established. 2440 * 2441 * Return codes 2442 * 0 - successful 2443 * -ENOMEM - no heap memory available 2444 * other values - from nvme registration upcall 2445 **/ 2446 int 2447 lpfc_nvme_create_localport(struct lpfc_vport *vport) 2448 { 2449 int ret = 0; 2450 struct lpfc_hba *phba = vport->phba; 2451 struct nvme_fc_port_info nfcp_info; 2452 struct nvme_fc_local_port *localport; 2453 struct lpfc_nvme_lport *lport; 2454 struct lpfc_nvme_ctrl_stat *cstat; 2455 int len, i; 2456 2457 /* Initialize this localport instance. The vport wwn usage ensures 2458 * that NPIV is accounted for. 2459 */ 2460 memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info)); 2461 nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR; 2462 nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn); 2463 nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn); 2464 2465 /* Limit to LPFC_MAX_NVME_SEG_CNT. 2466 * For now need + 1 to get around NVME transport logic. 2467 */ 2468 if (phba->cfg_sg_seg_cnt > LPFC_MAX_NVME_SEG_CNT) { 2469 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_INIT, 2470 "6300 Reducing sg segment cnt to %d\n", 2471 LPFC_MAX_NVME_SEG_CNT); 2472 phba->cfg_nvme_seg_cnt = LPFC_MAX_NVME_SEG_CNT; 2473 } else { 2474 phba->cfg_nvme_seg_cnt = phba->cfg_sg_seg_cnt; 2475 } 2476 lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1; 2477 lpfc_nvme_template.max_hw_queues = phba->cfg_nvme_io_channel; 2478 2479 cstat = kmalloc((sizeof(struct lpfc_nvme_ctrl_stat) * 2480 phba->cfg_nvme_io_channel), GFP_KERNEL); 2481 if (!cstat) 2482 return -ENOMEM; 2483 2484 /* localport is allocated from the stack, but the registration 2485 * call allocates heap memory as well as the private area. 2486 */ 2487 #if (IS_ENABLED(CONFIG_NVME_FC)) 2488 ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template, 2489 &vport->phba->pcidev->dev, &localport); 2490 #else 2491 ret = -ENOMEM; 2492 #endif 2493 if (!ret) { 2494 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC, 2495 "6005 Successfully registered local " 2496 "NVME port num %d, localP %p, private %p, " 2497 "sg_seg %d\n", 2498 localport->port_num, localport, 2499 localport->private, 2500 lpfc_nvme_template.max_sgl_segments); 2501 2502 /* Private is our lport size declared in the template. */ 2503 lport = (struct lpfc_nvme_lport *)localport->private; 2504 vport->localport = localport; 2505 lport->vport = vport; 2506 lport->cstat = cstat; 2507 vport->nvmei_support = 1; 2508 2509 atomic_set(&lport->xmt_fcp_noxri, 0); 2510 atomic_set(&lport->xmt_fcp_bad_ndlp, 0); 2511 atomic_set(&lport->xmt_fcp_qdepth, 0); 2512 atomic_set(&lport->xmt_fcp_err, 0); 2513 atomic_set(&lport->xmt_fcp_wqerr, 0); 2514 atomic_set(&lport->xmt_fcp_abort, 0); 2515 atomic_set(&lport->xmt_ls_abort, 0); 2516 atomic_set(&lport->xmt_ls_err, 0); 2517 atomic_set(&lport->cmpl_fcp_xb, 0); 2518 atomic_set(&lport->cmpl_fcp_err, 0); 2519 atomic_set(&lport->cmpl_ls_xb, 0); 2520 atomic_set(&lport->cmpl_ls_err, 0); 2521 atomic_set(&lport->fc4NvmeLsRequests, 0); 2522 atomic_set(&lport->fc4NvmeLsCmpls, 0); 2523 2524 for (i = 0; i < phba->cfg_nvme_io_channel; i++) { 2525 cstat = &lport->cstat[i]; 2526 atomic_set(&cstat->fc4NvmeInputRequests, 0); 2527 atomic_set(&cstat->fc4NvmeOutputRequests, 0); 2528 atomic_set(&cstat->fc4NvmeControlRequests, 0); 2529 atomic_set(&cstat->fc4NvmeIoCmpls, 0); 2530 } 2531 2532 /* Don't post more new bufs if repost already recovered 2533 * the nvme sgls. 2534 */ 2535 if (phba->sli4_hba.nvme_xri_cnt == 0) { 2536 len = lpfc_new_nvme_buf(vport, 2537 phba->sli4_hba.nvme_xri_max); 2538 vport->phba->total_nvme_bufs += len; 2539 } 2540 } else { 2541 kfree(cstat); 2542 } 2543 2544 return ret; 2545 } 2546 2547 /* lpfc_nvme_lport_unreg_wait - Wait for the host to complete an lport unreg. 2548 * 2549 * The driver has to wait for the host nvme transport to callback 2550 * indicating the localport has successfully unregistered all 2551 * resources. Since this is an uninterruptible wait, loop every ten 2552 * seconds and print a message indicating no progress. 2553 * 2554 * An uninterruptible wait is used because of the risk of transport-to- 2555 * driver state mismatch. 2556 */ 2557 void 2558 lpfc_nvme_lport_unreg_wait(struct lpfc_vport *vport, 2559 struct lpfc_nvme_lport *lport) 2560 { 2561 #if (IS_ENABLED(CONFIG_NVME_FC)) 2562 u32 wait_tmo; 2563 int ret; 2564 2565 /* Host transport has to clean up and confirm requiring an indefinite 2566 * wait. Print a message if a 10 second wait expires and renew the 2567 * wait. This is unexpected. 2568 */ 2569 wait_tmo = msecs_to_jiffies(LPFC_NVME_WAIT_TMO * 1000); 2570 while (true) { 2571 ret = wait_for_completion_timeout(&lport->lport_unreg_done, 2572 wait_tmo); 2573 if (unlikely(!ret)) { 2574 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 2575 "6176 Lport %p Localport %p wait " 2576 "timed out. Renewing.\n", 2577 lport, vport->localport); 2578 continue; 2579 } 2580 break; 2581 } 2582 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 2583 "6177 Lport %p Localport %p Complete Success\n", 2584 lport, vport->localport); 2585 #endif 2586 } 2587 2588 /** 2589 * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport. 2590 * @pnvme: pointer to lpfc nvme data structure. 2591 * 2592 * This routine is invoked to destroy all lports bound to the phba. 2593 * The lport memory was allocated by the nvme fc transport and is 2594 * released there. This routine ensures all rports bound to the 2595 * lport have been disconnected. 2596 * 2597 **/ 2598 void 2599 lpfc_nvme_destroy_localport(struct lpfc_vport *vport) 2600 { 2601 #if (IS_ENABLED(CONFIG_NVME_FC)) 2602 struct nvme_fc_local_port *localport; 2603 struct lpfc_nvme_lport *lport; 2604 struct lpfc_nvme_ctrl_stat *cstat; 2605 int ret; 2606 2607 if (vport->nvmei_support == 0) 2608 return; 2609 2610 localport = vport->localport; 2611 vport->localport = NULL; 2612 lport = (struct lpfc_nvme_lport *)localport->private; 2613 cstat = lport->cstat; 2614 2615 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2616 "6011 Destroying NVME localport %p\n", 2617 localport); 2618 2619 /* lport's rport list is clear. Unregister 2620 * lport and release resources. 2621 */ 2622 init_completion(&lport->lport_unreg_done); 2623 ret = nvme_fc_unregister_localport(localport); 2624 2625 /* Wait for completion. This either blocks 2626 * indefinitely or succeeds 2627 */ 2628 lpfc_nvme_lport_unreg_wait(vport, lport); 2629 kfree(cstat); 2630 2631 /* Regardless of the unregister upcall response, clear 2632 * nvmei_support. All rports are unregistered and the 2633 * driver will clean up. 2634 */ 2635 vport->nvmei_support = 0; 2636 if (ret == 0) { 2637 lpfc_printf_vlog(vport, 2638 KERN_INFO, LOG_NVME_DISC, 2639 "6009 Unregistered lport Success\n"); 2640 } else { 2641 lpfc_printf_vlog(vport, 2642 KERN_INFO, LOG_NVME_DISC, 2643 "6010 Unregistered lport " 2644 "Failed, status x%x\n", 2645 ret); 2646 } 2647 #endif 2648 } 2649 2650 void 2651 lpfc_nvme_update_localport(struct lpfc_vport *vport) 2652 { 2653 #if (IS_ENABLED(CONFIG_NVME_FC)) 2654 struct nvme_fc_local_port *localport; 2655 struct lpfc_nvme_lport *lport; 2656 2657 localport = vport->localport; 2658 if (!localport) { 2659 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2660 "6710 Update NVME fail. No localport\n"); 2661 return; 2662 } 2663 lport = (struct lpfc_nvme_lport *)localport->private; 2664 if (!lport) { 2665 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2666 "6171 Update NVME fail. localP %p, No lport\n", 2667 localport); 2668 return; 2669 } 2670 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2671 "6012 Update NVME lport %p did x%x\n", 2672 localport, vport->fc_myDID); 2673 2674 localport->port_id = vport->fc_myDID; 2675 if (localport->port_id == 0) 2676 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY; 2677 else 2678 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR; 2679 2680 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2681 "6030 bound lport %p to DID x%06x\n", 2682 lport, localport->port_id); 2683 #endif 2684 } 2685 2686 int 2687 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2688 { 2689 #if (IS_ENABLED(CONFIG_NVME_FC)) 2690 int ret = 0; 2691 struct nvme_fc_local_port *localport; 2692 struct lpfc_nvme_lport *lport; 2693 struct lpfc_nvme_rport *rport; 2694 struct lpfc_nvme_rport *oldrport; 2695 struct nvme_fc_remote_port *remote_port; 2696 struct nvme_fc_port_info rpinfo; 2697 struct lpfc_nodelist *prev_ndlp = NULL; 2698 2699 lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC, 2700 "6006 Register NVME PORT. DID x%06x nlptype x%x\n", 2701 ndlp->nlp_DID, ndlp->nlp_type); 2702 2703 localport = vport->localport; 2704 if (!localport) 2705 return 0; 2706 2707 lport = (struct lpfc_nvme_lport *)localport->private; 2708 2709 /* NVME rports are not preserved across devloss. 2710 * Just register this instance. Note, rpinfo->dev_loss_tmo 2711 * is left 0 to indicate accept transport defaults. The 2712 * driver communicates port role capabilities consistent 2713 * with the PRLI response data. 2714 */ 2715 memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info)); 2716 rpinfo.port_id = ndlp->nlp_DID; 2717 if (ndlp->nlp_type & NLP_NVME_TARGET) 2718 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET; 2719 if (ndlp->nlp_type & NLP_NVME_INITIATOR) 2720 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR; 2721 2722 if (ndlp->nlp_type & NLP_NVME_DISCOVERY) 2723 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY; 2724 2725 rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn); 2726 rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn); 2727 2728 spin_lock_irq(&vport->phba->hbalock); 2729 oldrport = lpfc_ndlp_get_nrport(ndlp); 2730 spin_unlock_irq(&vport->phba->hbalock); 2731 if (!oldrport) 2732 lpfc_nlp_get(ndlp); 2733 2734 ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port); 2735 if (!ret) { 2736 /* If the ndlp already has an nrport, this is just 2737 * a resume of the existing rport. Else this is a 2738 * new rport. 2739 */ 2740 /* Guard against an unregister/reregister 2741 * race that leaves the WAIT flag set. 2742 */ 2743 spin_lock_irq(&vport->phba->hbalock); 2744 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 2745 spin_unlock_irq(&vport->phba->hbalock); 2746 rport = remote_port->private; 2747 if (oldrport) { 2748 /* New remoteport record does not guarantee valid 2749 * host private memory area. 2750 */ 2751 prev_ndlp = oldrport->ndlp; 2752 if (oldrport == remote_port->private) { 2753 /* Same remoteport - ndlp should match. 2754 * Just reuse. 2755 */ 2756 lpfc_printf_vlog(ndlp->vport, KERN_INFO, 2757 LOG_NVME_DISC, 2758 "6014 Rebinding lport to " 2759 "remoteport %p wwpn 0x%llx, " 2760 "Data: x%x x%x %p %p x%x x%06x\n", 2761 remote_port, 2762 remote_port->port_name, 2763 remote_port->port_id, 2764 remote_port->port_role, 2765 prev_ndlp, 2766 ndlp, 2767 ndlp->nlp_type, 2768 ndlp->nlp_DID); 2769 return 0; 2770 } 2771 2772 /* Sever the ndlp<->rport association 2773 * before dropping the ndlp ref from 2774 * register. 2775 */ 2776 spin_lock_irq(&vport->phba->hbalock); 2777 ndlp->nrport = NULL; 2778 ndlp->upcall_flags &= ~NLP_WAIT_FOR_UNREG; 2779 spin_unlock_irq(&vport->phba->hbalock); 2780 rport->ndlp = NULL; 2781 rport->remoteport = NULL; 2782 2783 /* Reference only removed if previous NDLP is no longer 2784 * active. It might be just a swap and removing the 2785 * reference would cause a premature cleanup. 2786 */ 2787 if (prev_ndlp && prev_ndlp != ndlp) { 2788 if ((!NLP_CHK_NODE_ACT(prev_ndlp)) || 2789 (!prev_ndlp->nrport)) 2790 lpfc_nlp_put(prev_ndlp); 2791 } 2792 } 2793 2794 /* Clean bind the rport to the ndlp. */ 2795 rport->remoteport = remote_port; 2796 rport->lport = lport; 2797 rport->ndlp = ndlp; 2798 spin_lock_irq(&vport->phba->hbalock); 2799 ndlp->nrport = rport; 2800 spin_unlock_irq(&vport->phba->hbalock); 2801 lpfc_printf_vlog(vport, KERN_INFO, 2802 LOG_NVME_DISC | LOG_NODE, 2803 "6022 Binding new rport to " 2804 "lport %p Remoteport %p rport %p WWNN 0x%llx, " 2805 "Rport WWPN 0x%llx DID " 2806 "x%06x Role x%x, ndlp %p prev_ndlp %p\n", 2807 lport, remote_port, rport, 2808 rpinfo.node_name, rpinfo.port_name, 2809 rpinfo.port_id, rpinfo.port_role, 2810 ndlp, prev_ndlp); 2811 } else { 2812 lpfc_printf_vlog(vport, KERN_ERR, 2813 LOG_NVME_DISC | LOG_NODE, 2814 "6031 RemotePort Registration failed " 2815 "err: %d, DID x%06x\n", 2816 ret, ndlp->nlp_DID); 2817 } 2818 2819 return ret; 2820 #else 2821 return 0; 2822 #endif 2823 } 2824 2825 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport. 2826 * 2827 * There is no notion of Devloss or rport recovery from the current 2828 * nvme_transport perspective. Loss of an rport just means IO cannot 2829 * be sent and recovery is completely up to the initator. 2830 * For now, the driver just unbinds the DID and port_role so that 2831 * no further IO can be issued. Changes are planned for later. 2832 * 2833 * Notes - the ndlp reference count is not decremented here since 2834 * since there is no nvme_transport api for devloss. Node ref count 2835 * is only adjusted in driver unload. 2836 */ 2837 void 2838 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2839 { 2840 #if (IS_ENABLED(CONFIG_NVME_FC)) 2841 int ret; 2842 struct nvme_fc_local_port *localport; 2843 struct lpfc_nvme_lport *lport; 2844 struct lpfc_nvme_rport *rport; 2845 struct nvme_fc_remote_port *remoteport = NULL; 2846 2847 localport = vport->localport; 2848 2849 /* This is fundamental error. The localport is always 2850 * available until driver unload. Just exit. 2851 */ 2852 if (!localport) 2853 return; 2854 2855 lport = (struct lpfc_nvme_lport *)localport->private; 2856 if (!lport) 2857 goto input_err; 2858 2859 spin_lock_irq(&vport->phba->hbalock); 2860 rport = lpfc_ndlp_get_nrport(ndlp); 2861 if (rport) 2862 remoteport = rport->remoteport; 2863 spin_unlock_irq(&vport->phba->hbalock); 2864 if (!remoteport) 2865 goto input_err; 2866 2867 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2868 "6033 Unreg nvme remoteport %p, portname x%llx, " 2869 "port_id x%06x, portstate x%x port type x%x\n", 2870 remoteport, remoteport->port_name, 2871 remoteport->port_id, remoteport->port_state, 2872 ndlp->nlp_type); 2873 2874 /* Sanity check ndlp type. Only call for NVME ports. Don't 2875 * clear any rport state until the transport calls back. 2876 */ 2877 2878 if (ndlp->nlp_type & NLP_NVME_TARGET) { 2879 /* No concern about the role change on the nvme remoteport. 2880 * The transport will update it. 2881 */ 2882 ndlp->upcall_flags |= NLP_WAIT_FOR_UNREG; 2883 2884 /* Don't let the host nvme transport keep sending keep-alives 2885 * on this remoteport. Vport is unloading, no recovery. The 2886 * return values is ignored. The upcall is a courtesy to the 2887 * transport. 2888 */ 2889 if (vport->load_flag & FC_UNLOADING) 2890 (void)nvme_fc_set_remoteport_devloss(remoteport, 0); 2891 2892 ret = nvme_fc_unregister_remoteport(remoteport); 2893 if (ret != 0) { 2894 lpfc_nlp_put(ndlp); 2895 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2896 "6167 NVME unregister failed %d " 2897 "port_state x%x\n", 2898 ret, remoteport->port_state); 2899 } 2900 } 2901 return; 2902 2903 input_err: 2904 #endif 2905 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2906 "6168 State error: lport %p, rport%p FCID x%06x\n", 2907 vport->localport, ndlp->rport, ndlp->nlp_DID); 2908 } 2909 2910 /** 2911 * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort 2912 * @phba: pointer to lpfc hba data structure. 2913 * @axri: pointer to the fcp xri abort wcqe structure. 2914 * 2915 * This routine is invoked by the worker thread to process a SLI4 fast-path 2916 * NVME aborted xri. Aborted NVME IO commands are completed to the transport 2917 * here. 2918 **/ 2919 void 2920 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba, 2921 struct sli4_wcqe_xri_aborted *axri) 2922 { 2923 uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri); 2924 struct lpfc_nvme_buf *lpfc_ncmd, *next_lpfc_ncmd; 2925 struct nvmefc_fcp_req *nvme_cmd = NULL; 2926 struct lpfc_nodelist *ndlp; 2927 unsigned long iflag = 0; 2928 2929 if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME)) 2930 return; 2931 spin_lock_irqsave(&phba->hbalock, iflag); 2932 spin_lock(&phba->sli4_hba.abts_nvme_buf_list_lock); 2933 list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd, 2934 &phba->sli4_hba.lpfc_abts_nvme_buf_list, 2935 list) { 2936 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) { 2937 list_del_init(&lpfc_ncmd->list); 2938 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2939 lpfc_ncmd->status = IOSTAT_SUCCESS; 2940 spin_unlock( 2941 &phba->sli4_hba.abts_nvme_buf_list_lock); 2942 2943 spin_unlock_irqrestore(&phba->hbalock, iflag); 2944 ndlp = lpfc_ncmd->ndlp; 2945 if (ndlp) 2946 lpfc_sli4_abts_err_handler(phba, ndlp, axri); 2947 2948 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2949 "6311 nvme_cmd %p xri x%x tag x%x " 2950 "abort complete and xri released\n", 2951 lpfc_ncmd->nvmeCmd, xri, 2952 lpfc_ncmd->cur_iocbq.iotag); 2953 2954 /* Aborted NVME commands are required to not complete 2955 * before the abort exchange command fully completes. 2956 * Once completed, it is available via the put list. 2957 */ 2958 if (lpfc_ncmd->nvmeCmd) { 2959 nvme_cmd = lpfc_ncmd->nvmeCmd; 2960 nvme_cmd->done(nvme_cmd); 2961 lpfc_ncmd->nvmeCmd = NULL; 2962 } 2963 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2964 return; 2965 } 2966 } 2967 spin_unlock(&phba->sli4_hba.abts_nvme_buf_list_lock); 2968 spin_unlock_irqrestore(&phba->hbalock, iflag); 2969 2970 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2971 "6312 XRI Aborted xri x%x not found\n", xri); 2972 2973 } 2974 2975 /** 2976 * lpfc_nvme_wait_for_io_drain - Wait for all NVME wqes to complete 2977 * @phba: Pointer to HBA context object. 2978 * 2979 * This function flushes all wqes in the nvme rings and frees all resources 2980 * in the txcmplq. This function does not issue abort wqes for the IO 2981 * commands in txcmplq, they will just be returned with 2982 * IOERR_SLI_DOWN. This function is invoked with EEH when device's PCI 2983 * slot has been permanently disabled. 2984 **/ 2985 void 2986 lpfc_nvme_wait_for_io_drain(struct lpfc_hba *phba) 2987 { 2988 struct lpfc_sli_ring *pring; 2989 u32 i, wait_cnt = 0; 2990 2991 if (phba->sli_rev < LPFC_SLI_REV4 || !phba->sli4_hba.nvme_wq) 2992 return; 2993 2994 /* Cycle through all NVME rings and make sure all outstanding 2995 * WQEs have been removed from the txcmplqs. 2996 */ 2997 for (i = 0; i < phba->cfg_nvme_io_channel; i++) { 2998 pring = phba->sli4_hba.nvme_wq[i]->pring; 2999 3000 if (!pring) 3001 continue; 3002 3003 /* Retrieve everything on the txcmplq */ 3004 while (!list_empty(&pring->txcmplq)) { 3005 msleep(LPFC_XRI_EXCH_BUSY_WAIT_T1); 3006 wait_cnt++; 3007 3008 /* The sleep is 10mS. Every ten seconds, 3009 * dump a message. Something is wrong. 3010 */ 3011 if ((wait_cnt % 1000) == 0) { 3012 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 3013 "6178 NVME IO not empty, " 3014 "cnt %d\n", wait_cnt); 3015 } 3016 } 3017 } 3018 } 3019