1 /******************************************************************* 2 * This file is part of the Emulex Linux Device Driver for * 3 * Fibre Channel Host Bus Adapters. * 4 * Copyright (C) 2017 Broadcom. All Rights Reserved. The term * 5 * “Broadcom” refers to Broadcom Limited and/or its subsidiaries. * 6 * Copyright (C) 2004-2016 Emulex. All rights reserved. * 7 * EMULEX and SLI are trademarks of Emulex. * 8 * www.broadcom.com * 9 * Portions Copyright (C) 2004-2005 Christoph Hellwig * 10 * * 11 * This program is free software; you can redistribute it and/or * 12 * modify it under the terms of version 2 of the GNU General * 13 * Public License as published by the Free Software Foundation. * 14 * This program is distributed in the hope that it will be useful. * 15 * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND * 16 * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, * 17 * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE * 18 * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD * 19 * TO BE LEGALLY INVALID. See the GNU General Public License for * 20 * more details, a copy of which can be found in the file COPYING * 21 * included with this package. * 22 ********************************************************************/ 23 #include <linux/pci.h> 24 #include <linux/slab.h> 25 #include <linux/interrupt.h> 26 #include <linux/delay.h> 27 #include <asm/unaligned.h> 28 #include <linux/crc-t10dif.h> 29 #include <net/checksum.h> 30 31 #include <scsi/scsi.h> 32 #include <scsi/scsi_device.h> 33 #include <scsi/scsi_eh.h> 34 #include <scsi/scsi_host.h> 35 #include <scsi/scsi_tcq.h> 36 #include <scsi/scsi_transport_fc.h> 37 #include <scsi/fc/fc_fs.h> 38 39 #include <linux/nvme.h> 40 #include <linux/nvme-fc-driver.h> 41 #include <linux/nvme-fc.h> 42 #include "lpfc_version.h" 43 #include "lpfc_hw4.h" 44 #include "lpfc_hw.h" 45 #include "lpfc_sli.h" 46 #include "lpfc_sli4.h" 47 #include "lpfc_nl.h" 48 #include "lpfc_disc.h" 49 #include "lpfc.h" 50 #include "lpfc_nvme.h" 51 #include "lpfc_scsi.h" 52 #include "lpfc_logmsg.h" 53 #include "lpfc_crtn.h" 54 #include "lpfc_vport.h" 55 #include "lpfc_debugfs.h" 56 57 /* NVME initiator-based functions */ 58 59 static struct lpfc_nvme_buf * 60 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp); 61 62 static void 63 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_nvme_buf *); 64 65 66 /** 67 * lpfc_nvme_create_queue - 68 * @lpfc_pnvme: Pointer to the driver's nvme instance data 69 * @qidx: An cpu index used to affinitize IO queues and MSIX vectors. 70 * @handle: An opaque driver handle used in follow-up calls. 71 * 72 * Driver registers this routine to preallocate and initialize any 73 * internal data structures to bind the @qidx to its internal IO queues. 74 * A hardware queue maps (qidx) to a specific driver MSI-X vector/EQ/CQ/WQ. 75 * 76 * Return value : 77 * 0 - Success 78 * -EINVAL - Unsupported input value. 79 * -ENOMEM - Could not alloc necessary memory 80 **/ 81 static int 82 lpfc_nvme_create_queue(struct nvme_fc_local_port *pnvme_lport, 83 unsigned int qidx, u16 qsize, 84 void **handle) 85 { 86 struct lpfc_nvme_lport *lport; 87 struct lpfc_vport *vport; 88 struct lpfc_nvme_qhandle *qhandle; 89 char *str; 90 91 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 92 vport = lport->vport; 93 qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL); 94 if (qhandle == NULL) 95 return -ENOMEM; 96 97 qhandle->cpu_id = smp_processor_id(); 98 qhandle->qidx = qidx; 99 /* 100 * NVME qidx == 0 is the admin queue, so both admin queue 101 * and first IO queue will use MSI-X vector and associated 102 * EQ/CQ/WQ at index 0. After that they are sequentially assigned. 103 */ 104 if (qidx) { 105 str = "IO "; /* IO queue */ 106 qhandle->index = ((qidx - 1) % 107 vport->phba->cfg_nvme_io_channel); 108 } else { 109 str = "ADM"; /* Admin queue */ 110 qhandle->index = qidx; 111 } 112 113 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME, 114 "6073 Binding %s HdwQueue %d (cpu %d) to " 115 "io_channel %d qhandle %p\n", str, 116 qidx, qhandle->cpu_id, qhandle->index, qhandle); 117 *handle = (void *)qhandle; 118 return 0; 119 } 120 121 /** 122 * lpfc_nvme_delete_queue - 123 * @lpfc_pnvme: Pointer to the driver's nvme instance data 124 * @qidx: An cpu index used to affinitize IO queues and MSIX vectors. 125 * @handle: An opaque driver handle from lpfc_nvme_create_queue 126 * 127 * Driver registers this routine to free 128 * any internal data structures to bind the @qidx to its internal 129 * IO queues. 130 * 131 * Return value : 132 * 0 - Success 133 * TODO: What are the failure codes. 134 **/ 135 static void 136 lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport, 137 unsigned int qidx, 138 void *handle) 139 { 140 struct lpfc_nvme_lport *lport; 141 struct lpfc_vport *vport; 142 143 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 144 vport = lport->vport; 145 146 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 147 "6001 ENTER. lpfc_pnvme %p, qidx x%xi qhandle %p\n", 148 lport, qidx, handle); 149 kfree(handle); 150 } 151 152 static void 153 lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport) 154 { 155 struct lpfc_nvme_lport *lport = localport->private; 156 157 /* release any threads waiting for the unreg to complete */ 158 complete(&lport->lport_unreg_done); 159 } 160 161 /* lpfc_nvme_remoteport_delete 162 * 163 * @remoteport: Pointer to an nvme transport remoteport instance. 164 * 165 * This is a template downcall. NVME transport calls this function 166 * when it has completed the unregistration of a previously 167 * registered remoteport. 168 * 169 * Return value : 170 * None 171 */ 172 void 173 lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport) 174 { 175 struct lpfc_nvme_rport *rport = remoteport->private; 176 struct lpfc_vport *vport; 177 struct lpfc_nodelist *ndlp; 178 179 ndlp = rport->ndlp; 180 if (!ndlp) 181 goto rport_err; 182 183 vport = ndlp->vport; 184 if (!vport) 185 goto rport_err; 186 187 /* Remove this rport from the lport's list - memory is owned by the 188 * transport. Remove the ndlp reference for the NVME transport before 189 * calling state machine to remove the node. 190 */ 191 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 192 "6146 remoteport delete complete %p\n", 193 remoteport); 194 ndlp->nrport = NULL; 195 lpfc_nlp_put(ndlp); 196 197 rport_err: 198 /* This call has to execute as long as the rport is valid. 199 * Release any threads waiting for the unreg to complete. 200 */ 201 complete(&rport->rport_unreg_done); 202 } 203 204 static void 205 lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe, 206 struct lpfc_wcqe_complete *wcqe) 207 { 208 struct lpfc_vport *vport = cmdwqe->vport; 209 uint32_t status; 210 struct nvmefc_ls_req *pnvme_lsreq; 211 struct lpfc_dmabuf *buf_ptr; 212 struct lpfc_nodelist *ndlp; 213 214 atomic_inc(&vport->phba->fc4NvmeLsCmpls); 215 216 pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2; 217 status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK; 218 ndlp = (struct lpfc_nodelist *)cmdwqe->context1; 219 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 220 "6047 nvme cmpl Enter " 221 "Data %p DID %x Xri: %x status %x cmd:%p lsreg:%p " 222 "bmp:%p ndlp:%p\n", 223 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0, 224 cmdwqe->sli4_xritag, status, 225 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp); 226 227 lpfc_nvmeio_data(phba, "NVME LS CMPL: xri x%x stat x%x parm x%x\n", 228 cmdwqe->sli4_xritag, status, wcqe->parameter); 229 230 if (cmdwqe->context3) { 231 buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3; 232 lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys); 233 kfree(buf_ptr); 234 cmdwqe->context3 = NULL; 235 } 236 if (pnvme_lsreq->done) 237 pnvme_lsreq->done(pnvme_lsreq, status); 238 else 239 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 240 "6046 nvme cmpl without done call back? " 241 "Data %p DID %x Xri: %x status %x\n", 242 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0, 243 cmdwqe->sli4_xritag, status); 244 if (ndlp) { 245 lpfc_nlp_put(ndlp); 246 cmdwqe->context1 = NULL; 247 } 248 lpfc_sli_release_iocbq(phba, cmdwqe); 249 } 250 251 static int 252 lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp, 253 struct lpfc_dmabuf *inp, 254 struct nvmefc_ls_req *pnvme_lsreq, 255 void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *, 256 struct lpfc_wcqe_complete *), 257 struct lpfc_nodelist *ndlp, uint32_t num_entry, 258 uint32_t tmo, uint8_t retry) 259 { 260 struct lpfc_hba *phba = vport->phba; 261 union lpfc_wqe *wqe; 262 struct lpfc_iocbq *genwqe; 263 struct ulp_bde64 *bpl; 264 struct ulp_bde64 bde; 265 int i, rc, xmit_len, first_len; 266 267 /* Allocate buffer for command WQE */ 268 genwqe = lpfc_sli_get_iocbq(phba); 269 if (genwqe == NULL) 270 return 1; 271 272 wqe = &genwqe->wqe; 273 memset(wqe, 0, sizeof(union lpfc_wqe)); 274 275 genwqe->context3 = (uint8_t *)bmp; 276 genwqe->iocb_flag |= LPFC_IO_NVME_LS; 277 278 /* Save for completion so we can release these resources */ 279 genwqe->context1 = lpfc_nlp_get(ndlp); 280 genwqe->context2 = (uint8_t *)pnvme_lsreq; 281 /* Fill in payload, bp points to frame payload */ 282 283 if (!tmo) 284 /* FC spec states we need 3 * ratov for CT requests */ 285 tmo = (3 * phba->fc_ratov); 286 287 /* For this command calculate the xmit length of the request bde. */ 288 xmit_len = 0; 289 first_len = 0; 290 bpl = (struct ulp_bde64 *)bmp->virt; 291 for (i = 0; i < num_entry; i++) { 292 bde.tus.w = bpl[i].tus.w; 293 if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64) 294 break; 295 xmit_len += bde.tus.f.bdeSize; 296 if (i == 0) 297 first_len = xmit_len; 298 } 299 300 genwqe->rsvd2 = num_entry; 301 genwqe->hba_wqidx = 0; 302 303 /* Words 0 - 2 */ 304 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64; 305 wqe->generic.bde.tus.f.bdeSize = first_len; 306 wqe->generic.bde.addrLow = bpl[0].addrLow; 307 wqe->generic.bde.addrHigh = bpl[0].addrHigh; 308 309 /* Word 3 */ 310 wqe->gen_req.request_payload_len = first_len; 311 312 /* Word 4 */ 313 314 /* Word 5 */ 315 bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0); 316 bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1); 317 bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1); 318 bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_ELS4_REQ); 319 bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME); 320 321 /* Word 6 */ 322 bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com, 323 phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]); 324 bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag); 325 326 /* Word 7 */ 327 bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1)); 328 bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3); 329 bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE); 330 bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI); 331 332 /* Word 8 */ 333 wqe->gen_req.wqe_com.abort_tag = genwqe->iotag; 334 335 /* Word 9 */ 336 bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag); 337 338 /* Word 10 */ 339 bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1); 340 bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ); 341 bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1); 342 bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE); 343 bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0); 344 345 /* Word 11 */ 346 bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 347 bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND); 348 349 350 /* Issue GEN REQ WQE for NPORT <did> */ 351 lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS, 352 "6050 Issue GEN REQ WQE to NPORT x%x " 353 "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n", 354 ndlp->nlp_DID, genwqe->iotag, 355 vport->port_state, 356 genwqe, pnvme_lsreq, bmp, xmit_len, first_len); 357 genwqe->wqe_cmpl = cmpl; 358 genwqe->iocb_cmpl = NULL; 359 genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT; 360 genwqe->vport = vport; 361 genwqe->retry = retry; 362 363 lpfc_nvmeio_data(phba, "NVME LS XMIT: xri x%x iotag x%x to x%06x\n", 364 genwqe->sli4_xritag, genwqe->iotag, ndlp->nlp_DID); 365 366 rc = lpfc_sli4_issue_wqe(phba, LPFC_ELS_RING, genwqe); 367 if (rc == WQE_ERROR) { 368 lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS, 369 "6045 Issue GEN REQ WQE to NPORT x%x " 370 "Data: x%x x%x\n", 371 ndlp->nlp_DID, genwqe->iotag, 372 vport->port_state); 373 lpfc_sli_release_iocbq(phba, genwqe); 374 return 1; 375 } 376 return 0; 377 } 378 379 /** 380 * lpfc_nvme_ls_req - Issue an Link Service request 381 * @lpfc_pnvme: Pointer to the driver's nvme instance data 382 * @lpfc_nvme_lport: Pointer to the driver's local port data 383 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 384 * 385 * Driver registers this routine to handle any link service request 386 * from the nvme_fc transport to a remote nvme-aware port. 387 * 388 * Return value : 389 * 0 - Success 390 * TODO: What are the failure codes. 391 **/ 392 static int 393 lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport, 394 struct nvme_fc_remote_port *pnvme_rport, 395 struct nvmefc_ls_req *pnvme_lsreq) 396 { 397 int ret = 0; 398 struct lpfc_nvme_lport *lport; 399 struct lpfc_vport *vport; 400 struct lpfc_nodelist *ndlp; 401 struct ulp_bde64 *bpl; 402 struct lpfc_dmabuf *bmp; 403 uint16_t ntype, nstate; 404 405 /* there are two dma buf in the request, actually there is one and 406 * the second one is just the start address + cmd size. 407 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped 408 * in a lpfc_dmabuf struct. When freeing we just free the wrapper 409 * because the nvem layer owns the data bufs. 410 * We do not have to break these packets open, we don't care what is in 411 * them. And we do not have to look at the resonse data, we only care 412 * that we got a response. All of the caring is going to happen in the 413 * nvme-fc layer. 414 */ 415 416 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 417 vport = lport->vport; 418 419 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id); 420 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 421 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 422 "6051 DID x%06x not an active rport.\n", 423 pnvme_rport->port_id); 424 return -ENODEV; 425 } 426 427 /* The remote node has to be a mapped nvme target or an 428 * unmapped nvme initiator or it's an error. 429 */ 430 ntype = ndlp->nlp_type; 431 nstate = ndlp->nlp_state; 432 if ((ntype & NLP_NVME_TARGET && nstate != NLP_STE_MAPPED_NODE) || 433 (ntype & NLP_NVME_INITIATOR && nstate != NLP_STE_UNMAPPED_NODE)) { 434 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 435 "6088 DID x%06x not ready for " 436 "IO. State x%x, Type x%x\n", 437 pnvme_rport->port_id, 438 ndlp->nlp_state, ndlp->nlp_type); 439 return -ENODEV; 440 } 441 bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL); 442 if (!bmp) { 443 444 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 445 "6044 Could not find node for DID %x\n", 446 pnvme_rport->port_id); 447 return 2; 448 } 449 INIT_LIST_HEAD(&bmp->list); 450 bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys)); 451 if (!bmp->virt) { 452 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 453 "6042 Could not find node for DID %x\n", 454 pnvme_rport->port_id); 455 kfree(bmp); 456 return 3; 457 } 458 bpl = (struct ulp_bde64 *)bmp->virt; 459 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma)); 460 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma)); 461 bpl->tus.f.bdeFlags = 0; 462 bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen; 463 bpl->tus.w = le32_to_cpu(bpl->tus.w); 464 bpl++; 465 466 bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma)); 467 bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma)); 468 bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I; 469 bpl->tus.f.bdeSize = pnvme_lsreq->rsplen; 470 bpl->tus.w = le32_to_cpu(bpl->tus.w); 471 472 /* Expand print to include key fields. */ 473 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 474 "6149 ENTER. lport %p, rport %p lsreq%p rqstlen:%d " 475 "rsplen:%d %pad %pad\n", 476 pnvme_lport, pnvme_rport, 477 pnvme_lsreq, pnvme_lsreq->rqstlen, 478 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma, 479 &pnvme_lsreq->rspdma); 480 481 atomic_inc(&vport->phba->fc4NvmeLsRequests); 482 483 /* Hardcode the wait to 30 seconds. Connections are failing otherwise. 484 * This code allows it all to work. 485 */ 486 ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr, 487 pnvme_lsreq, lpfc_nvme_cmpl_gen_req, 488 ndlp, 2, 30, 0); 489 if (ret != WQE_SUCCESS) { 490 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 491 "6052 EXIT. issue ls wqe failed lport %p, " 492 "rport %p lsreq%p Status %x DID %x\n", 493 pnvme_lport, pnvme_rport, pnvme_lsreq, 494 ret, ndlp->nlp_DID); 495 lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys); 496 kfree(bmp); 497 return ret; 498 } 499 500 /* Stub in routine and return 0 for now. */ 501 return ret; 502 } 503 504 /** 505 * lpfc_nvme_ls_abort - Issue an Link Service request 506 * @lpfc_pnvme: Pointer to the driver's nvme instance data 507 * @lpfc_nvme_lport: Pointer to the driver's local port data 508 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 509 * 510 * Driver registers this routine to handle any link service request 511 * from the nvme_fc transport to a remote nvme-aware port. 512 * 513 * Return value : 514 * 0 - Success 515 * TODO: What are the failure codes. 516 **/ 517 static void 518 lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport, 519 struct nvme_fc_remote_port *pnvme_rport, 520 struct nvmefc_ls_req *pnvme_lsreq) 521 { 522 struct lpfc_nvme_lport *lport; 523 struct lpfc_vport *vport; 524 struct lpfc_hba *phba; 525 struct lpfc_nodelist *ndlp; 526 LIST_HEAD(abort_list); 527 struct lpfc_sli_ring *pring; 528 struct lpfc_iocbq *wqe, *next_wqe; 529 530 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 531 vport = lport->vport; 532 phba = vport->phba; 533 534 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id); 535 if (!ndlp) { 536 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 537 "6049 Could not find node for DID %x\n", 538 pnvme_rport->port_id); 539 return; 540 } 541 542 /* Expand print to include key fields. */ 543 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 544 "6040 ENTER. lport %p, rport %p lsreq %p rqstlen:%d " 545 "rsplen:%d %pad %pad\n", 546 pnvme_lport, pnvme_rport, 547 pnvme_lsreq, pnvme_lsreq->rqstlen, 548 pnvme_lsreq->rsplen, &pnvme_lsreq->rqstdma, 549 &pnvme_lsreq->rspdma); 550 551 /* 552 * Lock the ELS ring txcmplq and build a local list of all ELS IOs 553 * that need an ABTS. The IOs need to stay on the txcmplq so that 554 * the abort operation completes them successfully. 555 */ 556 pring = phba->sli4_hba.nvmels_wq->pring; 557 spin_lock_irq(&phba->hbalock); 558 spin_lock(&pring->ring_lock); 559 list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) { 560 /* Add to abort_list on on NDLP match. */ 561 if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) { 562 wqe->iocb_flag |= LPFC_DRIVER_ABORTED; 563 list_add_tail(&wqe->dlist, &abort_list); 564 } 565 } 566 spin_unlock(&pring->ring_lock); 567 spin_unlock_irq(&phba->hbalock); 568 569 /* Abort the targeted IOs and remove them from the abort list. */ 570 list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) { 571 spin_lock_irq(&phba->hbalock); 572 list_del_init(&wqe->dlist); 573 lpfc_sli_issue_abort_iotag(phba, pring, wqe); 574 spin_unlock_irq(&phba->hbalock); 575 } 576 } 577 578 /* Fix up the existing sgls for NVME IO. */ 579 static void 580 lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport, 581 struct lpfc_nvme_buf *lpfc_ncmd, 582 struct nvmefc_fcp_req *nCmd) 583 { 584 struct sli4_sge *sgl; 585 union lpfc_wqe128 *wqe; 586 uint32_t *wptr, *dptr; 587 588 /* 589 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to 590 * match NVME. NVME sends 96 bytes. Also, use the 591 * nvme commands command and response dma addresses 592 * rather than the virtual memory to ease the restore 593 * operation. 594 */ 595 sgl = lpfc_ncmd->nvme_sgl; 596 sgl->sge_len = cpu_to_le32(nCmd->cmdlen); 597 598 sgl++; 599 600 /* Setup the physical region for the FCP RSP */ 601 sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma)); 602 sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma)); 603 sgl->word2 = le32_to_cpu(sgl->word2); 604 if (nCmd->sg_cnt) 605 bf_set(lpfc_sli4_sge_last, sgl, 0); 606 else 607 bf_set(lpfc_sli4_sge_last, sgl, 1); 608 sgl->word2 = cpu_to_le32(sgl->word2); 609 sgl->sge_len = cpu_to_le32(nCmd->rsplen); 610 611 /* 612 * Get a local pointer to the built-in wqe and correct 613 * the cmd size to match NVME's 96 bytes and fix 614 * the dma address. 615 */ 616 617 /* 128 byte wqe support here */ 618 wqe = (union lpfc_wqe128 *)&lpfc_ncmd->cur_iocbq.wqe; 619 620 /* Word 0-2 - NVME CMND IU (embedded payload) */ 621 wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED; 622 wqe->generic.bde.tus.f.bdeSize = 60; 623 wqe->generic.bde.addrHigh = 0; 624 wqe->generic.bde.addrLow = 64; /* Word 16 */ 625 626 /* Word 3 */ 627 bf_set(payload_offset_len, &wqe->fcp_icmd, 628 (nCmd->rsplen + nCmd->cmdlen)); 629 630 /* Word 10 */ 631 bf_set(wqe_nvme, &wqe->fcp_icmd.wqe_com, 1); 632 bf_set(wqe_wqes, &wqe->fcp_icmd.wqe_com, 1); 633 634 /* 635 * Embed the payload in the last half of the WQE 636 * WQE words 16-30 get the NVME CMD IU payload 637 * 638 * WQE words 16-19 get payload Words 1-4 639 * WQE words 20-21 get payload Words 6-7 640 * WQE words 22-29 get payload Words 16-23 641 */ 642 wptr = &wqe->words[16]; /* WQE ptr */ 643 dptr = (uint32_t *)nCmd->cmdaddr; /* payload ptr */ 644 dptr++; /* Skip Word 0 in payload */ 645 646 *wptr++ = *dptr++; /* Word 1 */ 647 *wptr++ = *dptr++; /* Word 2 */ 648 *wptr++ = *dptr++; /* Word 3 */ 649 *wptr++ = *dptr++; /* Word 4 */ 650 dptr++; /* Skip Word 5 in payload */ 651 *wptr++ = *dptr++; /* Word 6 */ 652 *wptr++ = *dptr++; /* Word 7 */ 653 dptr += 8; /* Skip Words 8-15 in payload */ 654 *wptr++ = *dptr++; /* Word 16 */ 655 *wptr++ = *dptr++; /* Word 17 */ 656 *wptr++ = *dptr++; /* Word 18 */ 657 *wptr++ = *dptr++; /* Word 19 */ 658 *wptr++ = *dptr++; /* Word 20 */ 659 *wptr++ = *dptr++; /* Word 21 */ 660 *wptr++ = *dptr++; /* Word 22 */ 661 *wptr = *dptr; /* Word 23 */ 662 } 663 664 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 665 static void 666 lpfc_nvme_ktime(struct lpfc_hba *phba, 667 struct lpfc_nvme_buf *lpfc_ncmd) 668 { 669 uint64_t seg1, seg2, seg3, seg4; 670 671 if (!phba->ktime_on) 672 return; 673 if (!lpfc_ncmd->ts_last_cmd || 674 !lpfc_ncmd->ts_cmd_start || 675 !lpfc_ncmd->ts_cmd_wqput || 676 !lpfc_ncmd->ts_isr_cmpl || 677 !lpfc_ncmd->ts_data_nvme) 678 return; 679 if (lpfc_ncmd->ts_cmd_start < lpfc_ncmd->ts_last_cmd) 680 return; 681 if (lpfc_ncmd->ts_cmd_wqput < lpfc_ncmd->ts_cmd_start) 682 return; 683 if (lpfc_ncmd->ts_isr_cmpl < lpfc_ncmd->ts_cmd_wqput) 684 return; 685 if (lpfc_ncmd->ts_data_nvme < lpfc_ncmd->ts_isr_cmpl) 686 return; 687 /* 688 * Segment 1 - Time from Last FCP command cmpl is handed 689 * off to NVME Layer to start of next command. 690 * Segment 2 - Time from Driver receives a IO cmd start 691 * from NVME Layer to WQ put is done on IO cmd. 692 * Segment 3 - Time from Driver WQ put is done on IO cmd 693 * to MSI-X ISR for IO cmpl. 694 * Segment 4 - Time from MSI-X ISR for IO cmpl to when 695 * cmpl is handled off to the NVME Layer. 696 */ 697 seg1 = lpfc_ncmd->ts_cmd_start - lpfc_ncmd->ts_last_cmd; 698 if (seg1 > 5000000) /* 5 ms - for sequential IOs */ 699 return; 700 701 /* Calculate times relative to start of IO */ 702 seg2 = (lpfc_ncmd->ts_cmd_wqput - lpfc_ncmd->ts_cmd_start); 703 seg3 = (lpfc_ncmd->ts_isr_cmpl - 704 lpfc_ncmd->ts_cmd_start) - seg2; 705 seg4 = (lpfc_ncmd->ts_data_nvme - 706 lpfc_ncmd->ts_cmd_start) - seg2 - seg3; 707 phba->ktime_data_samples++; 708 phba->ktime_seg1_total += seg1; 709 if (seg1 < phba->ktime_seg1_min) 710 phba->ktime_seg1_min = seg1; 711 else if (seg1 > phba->ktime_seg1_max) 712 phba->ktime_seg1_max = seg1; 713 phba->ktime_seg2_total += seg2; 714 if (seg2 < phba->ktime_seg2_min) 715 phba->ktime_seg2_min = seg2; 716 else if (seg2 > phba->ktime_seg2_max) 717 phba->ktime_seg2_max = seg2; 718 phba->ktime_seg3_total += seg3; 719 if (seg3 < phba->ktime_seg3_min) 720 phba->ktime_seg3_min = seg3; 721 else if (seg3 > phba->ktime_seg3_max) 722 phba->ktime_seg3_max = seg3; 723 phba->ktime_seg4_total += seg4; 724 if (seg4 < phba->ktime_seg4_min) 725 phba->ktime_seg4_min = seg4; 726 else if (seg4 > phba->ktime_seg4_max) 727 phba->ktime_seg4_max = seg4; 728 729 lpfc_ncmd->ts_last_cmd = 0; 730 lpfc_ncmd->ts_cmd_start = 0; 731 lpfc_ncmd->ts_cmd_wqput = 0; 732 lpfc_ncmd->ts_isr_cmpl = 0; 733 lpfc_ncmd->ts_data_nvme = 0; 734 } 735 #endif 736 737 /** 738 * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO 739 * @lpfc_pnvme: Pointer to the driver's nvme instance data 740 * @lpfc_nvme_lport: Pointer to the driver's local port data 741 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 742 * 743 * Driver registers this routine as it io request handler. This 744 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 745 * data structure to the rport indicated in @lpfc_nvme_rport. 746 * 747 * Return value : 748 * 0 - Success 749 * TODO: What are the failure codes. 750 **/ 751 static void 752 lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn, 753 struct lpfc_wcqe_complete *wcqe) 754 { 755 struct lpfc_nvme_buf *lpfc_ncmd = 756 (struct lpfc_nvme_buf *)pwqeIn->context1; 757 struct lpfc_vport *vport = pwqeIn->vport; 758 struct nvmefc_fcp_req *nCmd; 759 struct nvme_fc_ersp_iu *ep; 760 struct nvme_fc_cmd_iu *cp; 761 struct lpfc_nvme_rport *rport; 762 struct lpfc_nodelist *ndlp; 763 struct lpfc_nvme_fcpreq_priv *freqpriv; 764 unsigned long flags; 765 uint32_t code; 766 uint16_t cid, sqhd, data; 767 uint32_t *ptr; 768 769 /* Sanity check on return of outstanding command */ 770 if (!lpfc_ncmd || !lpfc_ncmd->nvmeCmd || !lpfc_ncmd->nrport) { 771 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 772 "6071 Completion pointers bad on wqe %p.\n", 773 wcqe); 774 return; 775 } 776 atomic_inc(&phba->fc4NvmeIoCmpls); 777 778 nCmd = lpfc_ncmd->nvmeCmd; 779 rport = lpfc_ncmd->nrport; 780 781 lpfc_nvmeio_data(phba, "NVME FCP CMPL: xri x%x stat x%x parm x%x\n", 782 lpfc_ncmd->cur_iocbq.sli4_xritag, 783 bf_get(lpfc_wcqe_c_status, wcqe), wcqe->parameter); 784 /* 785 * Catch race where our node has transitioned, but the 786 * transport is still transitioning. 787 */ 788 ndlp = rport->ndlp; 789 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 790 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 791 "6061 rport %p, DID x%06x node not ready.\n", 792 rport, rport->remoteport->port_id); 793 794 ndlp = lpfc_findnode_did(vport, rport->remoteport->port_id); 795 if (!ndlp) { 796 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 797 "6062 Ignoring NVME cmpl. No ndlp\n"); 798 goto out_err; 799 } 800 } 801 802 code = bf_get(lpfc_wcqe_c_code, wcqe); 803 if (code == CQE_CODE_NVME_ERSP) { 804 /* For this type of CQE, we need to rebuild the rsp */ 805 ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr; 806 807 /* 808 * Get Command Id from cmd to plug into response. This 809 * code is not needed in the next NVME Transport drop. 810 */ 811 cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr; 812 cid = cp->sqe.common.command_id; 813 814 /* 815 * RSN is in CQE word 2 816 * SQHD is in CQE Word 3 bits 15:0 817 * Cmd Specific info is in CQE Word 1 818 * and in CQE Word 0 bits 15:0 819 */ 820 sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe); 821 822 /* Now lets build the NVME ERSP IU */ 823 ep->iu_len = cpu_to_be16(8); 824 ep->rsn = wcqe->parameter; 825 ep->xfrd_len = cpu_to_be32(nCmd->payload_length); 826 ep->rsvd12 = 0; 827 ptr = (uint32_t *)&ep->cqe.result.u64; 828 *ptr++ = wcqe->total_data_placed; 829 data = bf_get(lpfc_wcqe_c_ersp0, wcqe); 830 *ptr = (uint32_t)data; 831 ep->cqe.sq_head = sqhd; 832 ep->cqe.sq_id = nCmd->sqid; 833 ep->cqe.command_id = cid; 834 ep->cqe.status = 0; 835 836 lpfc_ncmd->status = IOSTAT_SUCCESS; 837 lpfc_ncmd->result = 0; 838 nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN; 839 nCmd->transferred_length = nCmd->payload_length; 840 } else { 841 lpfc_ncmd->status = (bf_get(lpfc_wcqe_c_status, wcqe) & 842 LPFC_IOCB_STATUS_MASK); 843 lpfc_ncmd->result = wcqe->parameter; 844 845 /* For NVME, the only failure path that results in an 846 * IO error is when the adapter rejects it. All other 847 * conditions are a success case and resolved by the 848 * transport. 849 * IOSTAT_FCP_RSP_ERROR means: 850 * 1. Length of data received doesn't match total 851 * transfer length in WQE 852 * 2. If the RSP payload does NOT match these cases: 853 * a. RSP length 12/24 bytes and all zeros 854 * b. NVME ERSP 855 */ 856 switch (lpfc_ncmd->status) { 857 case IOSTAT_SUCCESS: 858 nCmd->transferred_length = wcqe->total_data_placed; 859 nCmd->rcv_rsplen = 0; 860 nCmd->status = 0; 861 break; 862 case IOSTAT_FCP_RSP_ERROR: 863 nCmd->transferred_length = wcqe->total_data_placed; 864 nCmd->rcv_rsplen = wcqe->parameter; 865 nCmd->status = 0; 866 /* Sanity check */ 867 if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN) 868 break; 869 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 870 "6081 NVME Completion Protocol Error: " 871 "xri %x status x%x result x%x " 872 "placed x%x\n", 873 lpfc_ncmd->cur_iocbq.sli4_xritag, 874 lpfc_ncmd->status, lpfc_ncmd->result, 875 wcqe->total_data_placed); 876 break; 877 default: 878 out_err: 879 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 880 "6072 NVME Completion Error: xri %x " 881 "status x%x result x%x placed x%x\n", 882 lpfc_ncmd->cur_iocbq.sli4_xritag, 883 lpfc_ncmd->status, lpfc_ncmd->result, 884 wcqe->total_data_placed); 885 nCmd->transferred_length = 0; 886 nCmd->rcv_rsplen = 0; 887 nCmd->status = NVME_SC_FC_TRANSPORT_ERROR; 888 } 889 } 890 891 /* pick up SLI4 exhange busy condition */ 892 if (bf_get(lpfc_wcqe_c_xb, wcqe)) 893 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 894 else 895 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 896 897 if (ndlp && NLP_CHK_NODE_ACT(ndlp)) 898 atomic_dec(&ndlp->cmd_pending); 899 900 /* Update stats and complete the IO. There is 901 * no need for dma unprep because the nvme_transport 902 * owns the dma address. 903 */ 904 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 905 if (phba->ktime_on) { 906 lpfc_ncmd->ts_isr_cmpl = pwqeIn->isr_timestamp; 907 lpfc_ncmd->ts_data_nvme = ktime_get_ns(); 908 phba->ktime_last_cmd = lpfc_ncmd->ts_data_nvme; 909 lpfc_nvme_ktime(phba, lpfc_ncmd); 910 } 911 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 912 if (lpfc_ncmd->cpu != smp_processor_id()) 913 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 914 "6701 CPU Check cmpl: " 915 "cpu %d expect %d\n", 916 smp_processor_id(), lpfc_ncmd->cpu); 917 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT) 918 phba->cpucheck_cmpl_io[lpfc_ncmd->cpu]++; 919 } 920 #endif 921 freqpriv = nCmd->private; 922 freqpriv->nvme_buf = NULL; 923 nCmd->done(nCmd); 924 925 spin_lock_irqsave(&phba->hbalock, flags); 926 lpfc_ncmd->nrport = NULL; 927 spin_unlock_irqrestore(&phba->hbalock, flags); 928 929 lpfc_release_nvme_buf(phba, lpfc_ncmd); 930 } 931 932 933 /** 934 * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO 935 * @lpfc_pnvme: Pointer to the driver's nvme instance data 936 * @lpfc_nvme_lport: Pointer to the driver's local port data 937 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 938 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 939 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 940 * 941 * Driver registers this routine as it io request handler. This 942 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 943 * data structure to the rport indicated in @lpfc_nvme_rport. 944 * 945 * Return value : 946 * 0 - Success 947 * TODO: What are the failure codes. 948 **/ 949 static int 950 lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport, 951 struct lpfc_nvme_buf *lpfc_ncmd, 952 struct lpfc_nodelist *pnode) 953 { 954 struct lpfc_hba *phba = vport->phba; 955 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 956 struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq); 957 union lpfc_wqe128 *wqe = (union lpfc_wqe128 *)&pwqeq->wqe; 958 uint32_t req_len; 959 960 if (!pnode || !NLP_CHK_NODE_ACT(pnode)) 961 return -EINVAL; 962 963 /* 964 * There are three possibilities here - use scatter-gather segment, use 965 * the single mapping, or neither. 966 */ 967 wqe->fcp_iwrite.initial_xfer_len = 0; 968 if (nCmd->sg_cnt) { 969 if (nCmd->io_dir == NVMEFC_FCP_WRITE) { 970 /* Word 5 */ 971 if ((phba->cfg_nvme_enable_fb) && 972 (pnode->nlp_flag & NLP_FIRSTBURST)) { 973 req_len = lpfc_ncmd->nvmeCmd->payload_length; 974 if (req_len < pnode->nvme_fb_size) 975 wqe->fcp_iwrite.initial_xfer_len = 976 req_len; 977 else 978 wqe->fcp_iwrite.initial_xfer_len = 979 pnode->nvme_fb_size; 980 } 981 982 /* Word 7 */ 983 bf_set(wqe_cmnd, &wqe->generic.wqe_com, 984 CMD_FCP_IWRITE64_WQE); 985 bf_set(wqe_pu, &wqe->generic.wqe_com, 986 PARM_READ_CHECK); 987 988 /* Word 10 */ 989 bf_set(wqe_qosd, &wqe->fcp_iwrite.wqe_com, 0); 990 bf_set(wqe_iod, &wqe->fcp_iwrite.wqe_com, 991 LPFC_WQE_IOD_WRITE); 992 bf_set(wqe_lenloc, &wqe->fcp_iwrite.wqe_com, 993 LPFC_WQE_LENLOC_WORD4); 994 if (phba->cfg_nvme_oas) 995 bf_set(wqe_oas, &wqe->fcp_iwrite.wqe_com, 1); 996 997 /* Word 11 */ 998 bf_set(wqe_cmd_type, &wqe->generic.wqe_com, 999 NVME_WRITE_CMD); 1000 1001 atomic_inc(&phba->fc4NvmeOutputRequests); 1002 } else { 1003 /* Word 7 */ 1004 bf_set(wqe_cmnd, &wqe->generic.wqe_com, 1005 CMD_FCP_IREAD64_WQE); 1006 bf_set(wqe_pu, &wqe->generic.wqe_com, 1007 PARM_READ_CHECK); 1008 1009 /* Word 10 */ 1010 bf_set(wqe_qosd, &wqe->fcp_iread.wqe_com, 0); 1011 bf_set(wqe_iod, &wqe->fcp_iread.wqe_com, 1012 LPFC_WQE_IOD_READ); 1013 bf_set(wqe_lenloc, &wqe->fcp_iread.wqe_com, 1014 LPFC_WQE_LENLOC_WORD4); 1015 if (phba->cfg_nvme_oas) 1016 bf_set(wqe_oas, &wqe->fcp_iread.wqe_com, 1); 1017 1018 /* Word 11 */ 1019 bf_set(wqe_cmd_type, &wqe->generic.wqe_com, 1020 NVME_READ_CMD); 1021 1022 atomic_inc(&phba->fc4NvmeInputRequests); 1023 } 1024 } else { 1025 /* Word 4 */ 1026 wqe->fcp_icmd.rsrvd4 = 0; 1027 1028 /* Word 7 */ 1029 bf_set(wqe_cmnd, &wqe->generic.wqe_com, CMD_FCP_ICMND64_WQE); 1030 bf_set(wqe_pu, &wqe->generic.wqe_com, 0); 1031 1032 /* Word 10 */ 1033 bf_set(wqe_qosd, &wqe->fcp_icmd.wqe_com, 1); 1034 bf_set(wqe_iod, &wqe->fcp_icmd.wqe_com, LPFC_WQE_IOD_WRITE); 1035 bf_set(wqe_lenloc, &wqe->fcp_icmd.wqe_com, 1036 LPFC_WQE_LENLOC_NONE); 1037 if (phba->cfg_nvme_oas) 1038 bf_set(wqe_oas, &wqe->fcp_icmd.wqe_com, 1); 1039 1040 /* Word 11 */ 1041 bf_set(wqe_cmd_type, &wqe->generic.wqe_com, NVME_READ_CMD); 1042 1043 atomic_inc(&phba->fc4NvmeControlRequests); 1044 } 1045 /* 1046 * Finish initializing those WQE fields that are independent 1047 * of the nvme_cmnd request_buffer 1048 */ 1049 1050 /* Word 6 */ 1051 bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com, 1052 phba->sli4_hba.rpi_ids[pnode->nlp_rpi]); 1053 bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag); 1054 1055 /* Word 7 */ 1056 /* Preserve Class data in the ndlp. */ 1057 bf_set(wqe_class, &wqe->generic.wqe_com, 1058 (pnode->nlp_fcp_info & 0x0f)); 1059 1060 /* Word 8 */ 1061 wqe->generic.wqe_com.abort_tag = pwqeq->iotag; 1062 1063 /* Word 9 */ 1064 bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag); 1065 1066 /* Word 11 */ 1067 bf_set(wqe_cqid, &wqe->generic.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 1068 1069 pwqeq->vport = vport; 1070 return 0; 1071 } 1072 1073 1074 /** 1075 * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO 1076 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1077 * @lpfc_nvme_lport: Pointer to the driver's local port data 1078 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1079 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1080 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1081 * 1082 * Driver registers this routine as it io request handler. This 1083 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1084 * data structure to the rport indicated in @lpfc_nvme_rport. 1085 * 1086 * Return value : 1087 * 0 - Success 1088 * TODO: What are the failure codes. 1089 **/ 1090 static int 1091 lpfc_nvme_prep_io_dma(struct lpfc_vport *vport, 1092 struct lpfc_nvme_buf *lpfc_ncmd) 1093 { 1094 struct lpfc_hba *phba = vport->phba; 1095 struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd; 1096 union lpfc_wqe128 *wqe = (union lpfc_wqe128 *)&lpfc_ncmd->cur_iocbq.wqe; 1097 struct sli4_sge *sgl = lpfc_ncmd->nvme_sgl; 1098 struct scatterlist *data_sg; 1099 struct sli4_sge *first_data_sgl; 1100 dma_addr_t physaddr; 1101 uint32_t num_bde = 0; 1102 uint32_t dma_len; 1103 uint32_t dma_offset = 0; 1104 int nseg, i; 1105 1106 /* Fix up the command and response DMA stuff. */ 1107 lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd); 1108 1109 /* 1110 * There are three possibilities here - use scatter-gather segment, use 1111 * the single mapping, or neither. 1112 */ 1113 if (nCmd->sg_cnt) { 1114 /* 1115 * Jump over the cmd and rsp SGEs. The fix routine 1116 * has already adjusted for this. 1117 */ 1118 sgl += 2; 1119 1120 first_data_sgl = sgl; 1121 lpfc_ncmd->seg_cnt = nCmd->sg_cnt; 1122 if (lpfc_ncmd->seg_cnt > phba->cfg_nvme_seg_cnt) { 1123 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1124 "6058 Too many sg segments from " 1125 "NVME Transport. Max %d, " 1126 "nvmeIO sg_cnt %d\n", 1127 phba->cfg_nvme_seg_cnt, 1128 lpfc_ncmd->seg_cnt); 1129 lpfc_ncmd->seg_cnt = 0; 1130 return 1; 1131 } 1132 1133 /* 1134 * The driver established a maximum scatter-gather segment count 1135 * during probe that limits the number of sg elements in any 1136 * single nvme command. Just run through the seg_cnt and format 1137 * the sge's. 1138 */ 1139 nseg = nCmd->sg_cnt; 1140 data_sg = nCmd->first_sgl; 1141 for (i = 0; i < nseg; i++) { 1142 if (data_sg == NULL) { 1143 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1144 "6059 dptr err %d, nseg %d\n", 1145 i, nseg); 1146 lpfc_ncmd->seg_cnt = 0; 1147 return 1; 1148 } 1149 physaddr = data_sg->dma_address; 1150 dma_len = data_sg->length; 1151 sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr)); 1152 sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr)); 1153 sgl->word2 = le32_to_cpu(sgl->word2); 1154 if ((num_bde + 1) == nseg) 1155 bf_set(lpfc_sli4_sge_last, sgl, 1); 1156 else 1157 bf_set(lpfc_sli4_sge_last, sgl, 0); 1158 bf_set(lpfc_sli4_sge_offset, sgl, dma_offset); 1159 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA); 1160 sgl->word2 = cpu_to_le32(sgl->word2); 1161 sgl->sge_len = cpu_to_le32(dma_len); 1162 1163 dma_offset += dma_len; 1164 data_sg = sg_next(data_sg); 1165 sgl++; 1166 } 1167 } else { 1168 /* For this clause to be valid, the payload_length 1169 * and sg_cnt must zero. 1170 */ 1171 if (nCmd->payload_length != 0) { 1172 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1173 "6063 NVME DMA Prep Err: sg_cnt %d " 1174 "payload_length x%x\n", 1175 nCmd->sg_cnt, nCmd->payload_length); 1176 return 1; 1177 } 1178 } 1179 1180 /* 1181 * Due to difference in data length between DIF/non-DIF paths, 1182 * we need to set word 4 of WQE here 1183 */ 1184 wqe->fcp_iread.total_xfer_len = nCmd->payload_length; 1185 return 0; 1186 } 1187 1188 /** 1189 * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO 1190 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1191 * @lpfc_nvme_lport: Pointer to the driver's local port data 1192 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1193 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1194 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1195 * 1196 * Driver registers this routine as it io request handler. This 1197 * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq 1198 * data structure to the rport 1199 indicated in @lpfc_nvme_rport. 1200 * 1201 * Return value : 1202 * 0 - Success 1203 * TODO: What are the failure codes. 1204 **/ 1205 static int 1206 lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport, 1207 struct nvme_fc_remote_port *pnvme_rport, 1208 void *hw_queue_handle, 1209 struct nvmefc_fcp_req *pnvme_fcreq) 1210 { 1211 int ret = 0; 1212 struct lpfc_nvme_lport *lport; 1213 struct lpfc_vport *vport; 1214 struct lpfc_hba *phba; 1215 struct lpfc_nodelist *ndlp; 1216 struct lpfc_nvme_buf *lpfc_ncmd; 1217 struct lpfc_nvme_rport *rport; 1218 struct lpfc_nvme_qhandle *lpfc_queue_info; 1219 struct lpfc_nvme_fcpreq_priv *freqpriv = pnvme_fcreq->private; 1220 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1221 uint64_t start = 0; 1222 #endif 1223 1224 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1225 vport = lport->vport; 1226 phba = vport->phba; 1227 1228 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1229 if (phba->ktime_on) 1230 start = ktime_get_ns(); 1231 #endif 1232 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 1233 lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle; 1234 1235 /* 1236 * Catch race where our node has transitioned, but the 1237 * transport is still transitioning. 1238 */ 1239 ndlp = rport->ndlp; 1240 if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) { 1241 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 1242 "6053 rport %p, ndlp %p, DID x%06x " 1243 "ndlp not ready.\n", 1244 rport, ndlp, pnvme_rport->port_id); 1245 1246 ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id); 1247 if (!ndlp) { 1248 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 1249 "6066 Missing node for DID %x\n", 1250 pnvme_rport->port_id); 1251 ret = -ENODEV; 1252 goto out_fail; 1253 } 1254 } 1255 1256 /* The remote node has to be a mapped target or it's an error. */ 1257 if ((ndlp->nlp_type & NLP_NVME_TARGET) && 1258 (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) { 1259 lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR, 1260 "6036 rport %p, DID x%06x not ready for " 1261 "IO. State x%x, Type x%x\n", 1262 rport, pnvme_rport->port_id, 1263 ndlp->nlp_state, ndlp->nlp_type); 1264 ret = -ENODEV; 1265 goto out_fail; 1266 1267 } 1268 1269 /* The node is shared with FCP IO, make sure the IO pending count does 1270 * not exceed the programmed depth. 1271 */ 1272 if (atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) { 1273 ret = -EAGAIN; 1274 goto out_fail; 1275 } 1276 1277 lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp); 1278 if (lpfc_ncmd == NULL) { 1279 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR, 1280 "6065 driver's buffer pool is empty, " 1281 "IO failed\n"); 1282 ret = -ENOMEM; 1283 goto out_fail; 1284 } 1285 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1286 if (phba->ktime_on) { 1287 lpfc_ncmd->ts_cmd_start = start; 1288 lpfc_ncmd->ts_last_cmd = phba->ktime_last_cmd; 1289 } 1290 #endif 1291 1292 /* 1293 * Store the data needed by the driver to issue, abort, and complete 1294 * an IO. 1295 * Do not let the IO hang out forever. There is no midlayer issuing 1296 * an abort so inform the FW of the maximum IO pending time. 1297 */ 1298 freqpriv->nvme_buf = lpfc_ncmd; 1299 lpfc_ncmd->nvmeCmd = pnvme_fcreq; 1300 lpfc_ncmd->nrport = rport; 1301 lpfc_ncmd->ndlp = ndlp; 1302 lpfc_ncmd->start_time = jiffies; 1303 1304 lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp); 1305 ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd); 1306 if (ret) { 1307 ret = -ENOMEM; 1308 goto out_free_nvme_buf; 1309 } 1310 1311 atomic_inc(&ndlp->cmd_pending); 1312 1313 /* 1314 * Issue the IO on the WQ indicated by index in the hw_queue_handle. 1315 * This identfier was create in our hardware queue create callback 1316 * routine. The driver now is dependent on the IO queue steering from 1317 * the transport. We are trusting the upper NVME layers know which 1318 * index to use and that they have affinitized a CPU to this hardware 1319 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ. 1320 */ 1321 lpfc_ncmd->cur_iocbq.hba_wqidx = lpfc_queue_info->index; 1322 1323 lpfc_nvmeio_data(phba, "NVME FCP XMIT: xri x%x idx %d to %06x\n", 1324 lpfc_ncmd->cur_iocbq.sli4_xritag, 1325 lpfc_queue_info->index, ndlp->nlp_DID); 1326 1327 ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq); 1328 if (ret) { 1329 atomic_dec(&ndlp->cmd_pending); 1330 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR, 1331 "6113 FCP could not issue WQE err %x " 1332 "sid: x%x did: x%x oxid: x%x\n", 1333 ret, vport->fc_myDID, ndlp->nlp_DID, 1334 lpfc_ncmd->cur_iocbq.sli4_xritag); 1335 ret = -EBUSY; 1336 goto out_free_nvme_buf; 1337 } 1338 1339 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS 1340 if (phba->ktime_on) 1341 lpfc_ncmd->ts_cmd_wqput = ktime_get_ns(); 1342 1343 if (phba->cpucheck_on & LPFC_CHECK_NVME_IO) { 1344 lpfc_ncmd->cpu = smp_processor_id(); 1345 if (lpfc_ncmd->cpu != lpfc_queue_info->index) { 1346 /* Check for admin queue */ 1347 if (lpfc_queue_info->qidx) { 1348 lpfc_printf_vlog(vport, 1349 KERN_ERR, LOG_NVME_IOERR, 1350 "6702 CPU Check cmd: " 1351 "cpu %d wq %d\n", 1352 lpfc_ncmd->cpu, 1353 lpfc_queue_info->index); 1354 } 1355 lpfc_ncmd->cpu = lpfc_queue_info->index; 1356 } 1357 if (lpfc_ncmd->cpu < LPFC_CHECK_CPU_CNT) 1358 phba->cpucheck_xmt_io[lpfc_ncmd->cpu]++; 1359 } 1360 #endif 1361 return 0; 1362 1363 out_free_nvme_buf: 1364 if (lpfc_ncmd->nvmeCmd->sg_cnt) { 1365 if (lpfc_ncmd->nvmeCmd->io_dir == NVMEFC_FCP_WRITE) 1366 atomic_dec(&phba->fc4NvmeOutputRequests); 1367 else 1368 atomic_dec(&phba->fc4NvmeInputRequests); 1369 } else 1370 atomic_dec(&phba->fc4NvmeControlRequests); 1371 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1372 out_fail: 1373 return ret; 1374 } 1375 1376 /** 1377 * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request. 1378 * @phba: Pointer to HBA context object 1379 * @cmdiocb: Pointer to command iocb object. 1380 * @rspiocb: Pointer to response iocb object. 1381 * 1382 * This is the callback function for any NVME FCP IO that was aborted. 1383 * 1384 * Return value: 1385 * None 1386 **/ 1387 void 1388 lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb, 1389 struct lpfc_wcqe_complete *abts_cmpl) 1390 { 1391 lpfc_printf_log(phba, KERN_ERR, LOG_NVME, 1392 "6145 ABORT_XRI_CN completing on rpi x%x " 1393 "original iotag x%x, abort cmd iotag x%x " 1394 "req_tag x%x, status x%x, hwstatus x%x\n", 1395 cmdiocb->iocb.un.acxri.abortContextTag, 1396 cmdiocb->iocb.un.acxri.abortIoTag, 1397 cmdiocb->iotag, 1398 bf_get(lpfc_wcqe_c_request_tag, abts_cmpl), 1399 bf_get(lpfc_wcqe_c_status, abts_cmpl), 1400 bf_get(lpfc_wcqe_c_hw_status, abts_cmpl)); 1401 lpfc_sli_release_iocbq(phba, cmdiocb); 1402 } 1403 1404 /** 1405 * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS 1406 * @lpfc_pnvme: Pointer to the driver's nvme instance data 1407 * @lpfc_nvme_lport: Pointer to the driver's local port data 1408 * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq 1409 * @lpfc_nvme_fcreq: IO request from nvme fc to driver. 1410 * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue 1411 * 1412 * Driver registers this routine as its nvme request io abort handler. This 1413 * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq 1414 * data structure to the rport indicated in @lpfc_nvme_rport. This routine 1415 * is executed asynchronously - one the target is validated as "MAPPED" and 1416 * ready for IO, the driver issues the abort request and returns. 1417 * 1418 * Return value: 1419 * None 1420 **/ 1421 static void 1422 lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport, 1423 struct nvme_fc_remote_port *pnvme_rport, 1424 void *hw_queue_handle, 1425 struct nvmefc_fcp_req *pnvme_fcreq) 1426 { 1427 struct lpfc_nvme_lport *lport; 1428 struct lpfc_vport *vport; 1429 struct lpfc_hba *phba; 1430 struct lpfc_nvme_rport *rport; 1431 struct lpfc_nvme_buf *lpfc_nbuf; 1432 struct lpfc_iocbq *abts_buf; 1433 struct lpfc_iocbq *nvmereq_wqe; 1434 struct lpfc_nvme_fcpreq_priv *freqpriv = pnvme_fcreq->private; 1435 union lpfc_wqe *abts_wqe; 1436 unsigned long flags; 1437 int ret_val; 1438 1439 lport = (struct lpfc_nvme_lport *)pnvme_lport->private; 1440 rport = (struct lpfc_nvme_rport *)pnvme_rport->private; 1441 vport = lport->vport; 1442 phba = vport->phba; 1443 1444 /* Announce entry to new IO submit field. */ 1445 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1446 "6002 Abort Request to rport DID x%06x " 1447 "for nvme_fc_req %p\n", 1448 pnvme_rport->port_id, 1449 pnvme_fcreq); 1450 1451 /* If the hba is getting reset, this flag is set. It is 1452 * cleared when the reset is complete and rings reestablished. 1453 */ 1454 spin_lock_irqsave(&phba->hbalock, flags); 1455 /* driver queued commands are in process of being flushed */ 1456 if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) { 1457 spin_unlock_irqrestore(&phba->hbalock, flags); 1458 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1459 "6139 Driver in reset cleanup - flushing " 1460 "NVME Req now. hba_flag x%x\n", 1461 phba->hba_flag); 1462 return; 1463 } 1464 1465 lpfc_nbuf = freqpriv->nvme_buf; 1466 if (!lpfc_nbuf) { 1467 spin_unlock_irqrestore(&phba->hbalock, flags); 1468 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1469 "6140 NVME IO req has no matching lpfc nvme " 1470 "io buffer. Skipping abort req.\n"); 1471 return; 1472 } else if (!lpfc_nbuf->nvmeCmd) { 1473 spin_unlock_irqrestore(&phba->hbalock, flags); 1474 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1475 "6141 lpfc NVME IO req has no nvme_fcreq " 1476 "io buffer. Skipping abort req.\n"); 1477 return; 1478 } 1479 nvmereq_wqe = &lpfc_nbuf->cur_iocbq; 1480 1481 /* 1482 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's 1483 * state must match the nvme_fcreq passed by the nvme 1484 * transport. If they don't match, it is likely the driver 1485 * has already completed the NVME IO and the nvme transport 1486 * has not seen it yet. 1487 */ 1488 if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) { 1489 spin_unlock_irqrestore(&phba->hbalock, flags); 1490 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1491 "6143 NVME req mismatch: " 1492 "lpfc_nbuf %p nvmeCmd %p, " 1493 "pnvme_fcreq %p. Skipping Abort xri x%x\n", 1494 lpfc_nbuf, lpfc_nbuf->nvmeCmd, 1495 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1496 return; 1497 } 1498 1499 /* Don't abort IOs no longer on the pending queue. */ 1500 if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) { 1501 spin_unlock_irqrestore(&phba->hbalock, flags); 1502 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1503 "6142 NVME IO req %p not queued - skipping " 1504 "abort req xri x%x\n", 1505 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1506 return; 1507 } 1508 1509 lpfc_nvmeio_data(phba, "NVME FCP ABORT: xri x%x idx %d to %06x\n", 1510 nvmereq_wqe->sli4_xritag, 1511 nvmereq_wqe->hba_wqidx, pnvme_rport->port_id); 1512 1513 /* Outstanding abort is in progress */ 1514 if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) { 1515 spin_unlock_irqrestore(&phba->hbalock, flags); 1516 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1517 "6144 Outstanding NVME I/O Abort Request " 1518 "still pending on nvme_fcreq %p, " 1519 "lpfc_ncmd %p xri x%x\n", 1520 pnvme_fcreq, lpfc_nbuf, 1521 nvmereq_wqe->sli4_xritag); 1522 return; 1523 } 1524 1525 abts_buf = __lpfc_sli_get_iocbq(phba); 1526 if (!abts_buf) { 1527 spin_unlock_irqrestore(&phba->hbalock, flags); 1528 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1529 "6136 No available abort wqes. Skipping " 1530 "Abts req for nvme_fcreq %p xri x%x\n", 1531 pnvme_fcreq, nvmereq_wqe->sli4_xritag); 1532 return; 1533 } 1534 1535 /* Ready - mark outstanding as aborted by driver. */ 1536 nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED; 1537 1538 /* Complete prepping the abort wqe and issue to the FW. */ 1539 abts_wqe = &abts_buf->wqe; 1540 1541 /* WQEs are reused. Clear stale data and set key fields to 1542 * zero like ia, iaab, iaar, xri_tag, and ctxt_tag. 1543 */ 1544 memset(abts_wqe, 0, sizeof(union lpfc_wqe)); 1545 bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG); 1546 1547 /* word 7 */ 1548 bf_set(wqe_ct, &abts_wqe->abort_cmd.wqe_com, 0); 1549 bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX); 1550 bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com, 1551 nvmereq_wqe->iocb.ulpClass); 1552 1553 /* word 8 - tell the FW to abort the IO associated with this 1554 * outstanding exchange ID. 1555 */ 1556 abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag; 1557 1558 /* word 9 - this is the iotag for the abts_wqe completion. */ 1559 bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com, 1560 abts_buf->iotag); 1561 1562 /* word 10 */ 1563 bf_set(wqe_wqid, &abts_wqe->abort_cmd.wqe_com, nvmereq_wqe->hba_wqidx); 1564 bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1); 1565 bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE); 1566 1567 /* word 11 */ 1568 bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND); 1569 bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1); 1570 bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT); 1571 1572 /* ABTS WQE must go to the same WQ as the WQE to be aborted */ 1573 abts_buf->iocb_flag |= LPFC_IO_NVME; 1574 abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx; 1575 abts_buf->vport = vport; 1576 abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl; 1577 ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf); 1578 spin_unlock_irqrestore(&phba->hbalock, flags); 1579 if (ret_val == IOCB_ERROR) { 1580 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS, 1581 "6137 Failed abts issue_wqe with status x%x " 1582 "for nvme_fcreq %p.\n", 1583 ret_val, pnvme_fcreq); 1584 lpfc_sli_release_iocbq(phba, abts_buf); 1585 return; 1586 } 1587 1588 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS, 1589 "6138 Transport Abort NVME Request Issued for " 1590 "ox_id x%x on reqtag x%x\n", 1591 nvmereq_wqe->sli4_xritag, 1592 abts_buf->iotag); 1593 } 1594 1595 /* Declare and initialization an instance of the FC NVME template. */ 1596 static struct nvme_fc_port_template lpfc_nvme_template = { 1597 /* initiator-based functions */ 1598 .localport_delete = lpfc_nvme_localport_delete, 1599 .remoteport_delete = lpfc_nvme_remoteport_delete, 1600 .create_queue = lpfc_nvme_create_queue, 1601 .delete_queue = lpfc_nvme_delete_queue, 1602 .ls_req = lpfc_nvme_ls_req, 1603 .fcp_io = lpfc_nvme_fcp_io_submit, 1604 .ls_abort = lpfc_nvme_ls_abort, 1605 .fcp_abort = lpfc_nvme_fcp_abort, 1606 1607 .max_hw_queues = 1, 1608 .max_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1609 .max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS, 1610 .dma_boundary = 0xFFFFFFFF, 1611 1612 /* Sizes of additional private data for data structures. 1613 * No use for the last two sizes at this time. 1614 */ 1615 .local_priv_sz = sizeof(struct lpfc_nvme_lport), 1616 .remote_priv_sz = sizeof(struct lpfc_nvme_rport), 1617 .lsrqst_priv_sz = 0, 1618 .fcprqst_priv_sz = sizeof(struct lpfc_nvme_fcpreq_priv), 1619 }; 1620 1621 /** 1622 * lpfc_sli4_post_nvme_sgl_block - post a block of nvme sgl list to firmware 1623 * @phba: pointer to lpfc hba data structure. 1624 * @nblist: pointer to nvme buffer list. 1625 * @count: number of scsi buffers on the list. 1626 * 1627 * This routine is invoked to post a block of @count scsi sgl pages from a 1628 * SCSI buffer list @nblist to the HBA using non-embedded mailbox command. 1629 * No Lock is held. 1630 * 1631 **/ 1632 static int 1633 lpfc_sli4_post_nvme_sgl_block(struct lpfc_hba *phba, 1634 struct list_head *nblist, 1635 int count) 1636 { 1637 struct lpfc_nvme_buf *lpfc_ncmd; 1638 struct lpfc_mbx_post_uembed_sgl_page1 *sgl; 1639 struct sgl_page_pairs *sgl_pg_pairs; 1640 void *viraddr; 1641 LPFC_MBOXQ_t *mbox; 1642 uint32_t reqlen, alloclen, pg_pairs; 1643 uint32_t mbox_tmo; 1644 uint16_t xritag_start = 0; 1645 int rc = 0; 1646 uint32_t shdr_status, shdr_add_status; 1647 dma_addr_t pdma_phys_bpl1; 1648 union lpfc_sli4_cfg_shdr *shdr; 1649 1650 /* Calculate the requested length of the dma memory */ 1651 reqlen = count * sizeof(struct sgl_page_pairs) + 1652 sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t); 1653 if (reqlen > SLI4_PAGE_SIZE) { 1654 lpfc_printf_log(phba, KERN_WARNING, LOG_INIT, 1655 "6118 Block sgl registration required DMA " 1656 "size (%d) great than a page\n", reqlen); 1657 return -ENOMEM; 1658 } 1659 mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL); 1660 if (!mbox) { 1661 lpfc_printf_log(phba, KERN_ERR, LOG_INIT, 1662 "6119 Failed to allocate mbox cmd memory\n"); 1663 return -ENOMEM; 1664 } 1665 1666 /* Allocate DMA memory and set up the non-embedded mailbox command */ 1667 alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE, 1668 LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen, 1669 LPFC_SLI4_MBX_NEMBED); 1670 1671 if (alloclen < reqlen) { 1672 lpfc_printf_log(phba, KERN_ERR, LOG_INIT, 1673 "6120 Allocated DMA memory size (%d) is " 1674 "less than the requested DMA memory " 1675 "size (%d)\n", alloclen, reqlen); 1676 lpfc_sli4_mbox_cmd_free(phba, mbox); 1677 return -ENOMEM; 1678 } 1679 1680 /* Get the first SGE entry from the non-embedded DMA memory */ 1681 viraddr = mbox->sge_array->addr[0]; 1682 1683 /* Set up the SGL pages in the non-embedded DMA pages */ 1684 sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr; 1685 sgl_pg_pairs = &sgl->sgl_pg_pairs; 1686 1687 pg_pairs = 0; 1688 list_for_each_entry(lpfc_ncmd, nblist, list) { 1689 /* Set up the sge entry */ 1690 sgl_pg_pairs->sgl_pg0_addr_lo = 1691 cpu_to_le32(putPaddrLow(lpfc_ncmd->dma_phys_sgl)); 1692 sgl_pg_pairs->sgl_pg0_addr_hi = 1693 cpu_to_le32(putPaddrHigh(lpfc_ncmd->dma_phys_sgl)); 1694 if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE) 1695 pdma_phys_bpl1 = lpfc_ncmd->dma_phys_sgl + 1696 SGL_PAGE_SIZE; 1697 else 1698 pdma_phys_bpl1 = 0; 1699 sgl_pg_pairs->sgl_pg1_addr_lo = 1700 cpu_to_le32(putPaddrLow(pdma_phys_bpl1)); 1701 sgl_pg_pairs->sgl_pg1_addr_hi = 1702 cpu_to_le32(putPaddrHigh(pdma_phys_bpl1)); 1703 /* Keep the first xritag on the list */ 1704 if (pg_pairs == 0) 1705 xritag_start = lpfc_ncmd->cur_iocbq.sli4_xritag; 1706 sgl_pg_pairs++; 1707 pg_pairs++; 1708 } 1709 bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start); 1710 bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs); 1711 /* Perform endian conversion if necessary */ 1712 sgl->word0 = cpu_to_le32(sgl->word0); 1713 1714 if (!phba->sli4_hba.intr_enable) 1715 rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL); 1716 else { 1717 mbox_tmo = lpfc_mbox_tmo_val(phba, mbox); 1718 rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo); 1719 } 1720 shdr = (union lpfc_sli4_cfg_shdr *)&sgl->cfg_shdr; 1721 shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response); 1722 shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response); 1723 if (rc != MBX_TIMEOUT) 1724 lpfc_sli4_mbox_cmd_free(phba, mbox); 1725 if (shdr_status || shdr_add_status || rc) { 1726 lpfc_printf_log(phba, KERN_ERR, LOG_SLI, 1727 "6125 POST_SGL_BLOCK mailbox command failed " 1728 "status x%x add_status x%x mbx status x%x\n", 1729 shdr_status, shdr_add_status, rc); 1730 rc = -ENXIO; 1731 } 1732 return rc; 1733 } 1734 1735 /** 1736 * lpfc_post_nvme_sgl_list - Post blocks of nvme buffer sgls from a list 1737 * @phba: pointer to lpfc hba data structure. 1738 * @post_nblist: pointer to the nvme buffer list. 1739 * 1740 * This routine walks a list of nvme buffers that was passed in. It attempts 1741 * to construct blocks of nvme buffer sgls which contains contiguous xris and 1742 * uses the non-embedded SGL block post mailbox commands to post to the port. 1743 * For single NVME buffer sgl with non-contiguous xri, if any, it shall use 1744 * embedded SGL post mailbox command for posting. The @post_nblist passed in 1745 * must be local list, thus no lock is needed when manipulate the list. 1746 * 1747 * Returns: 0 = failure, non-zero number of successfully posted buffers. 1748 **/ 1749 static int 1750 lpfc_post_nvme_sgl_list(struct lpfc_hba *phba, 1751 struct list_head *post_nblist, int sb_count) 1752 { 1753 struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next; 1754 int status, sgl_size; 1755 int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0; 1756 dma_addr_t pdma_phys_sgl1; 1757 int last_xritag = NO_XRI; 1758 int cur_xritag; 1759 LIST_HEAD(prep_nblist); 1760 LIST_HEAD(blck_nblist); 1761 LIST_HEAD(nvme_nblist); 1762 1763 /* sanity check */ 1764 if (sb_count <= 0) 1765 return -EINVAL; 1766 1767 sgl_size = phba->cfg_sg_dma_buf_size; 1768 1769 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, post_nblist, list) { 1770 list_del_init(&lpfc_ncmd->list); 1771 block_cnt++; 1772 if ((last_xritag != NO_XRI) && 1773 (lpfc_ncmd->cur_iocbq.sli4_xritag != last_xritag + 1)) { 1774 /* a hole in xri block, form a sgl posting block */ 1775 list_splice_init(&prep_nblist, &blck_nblist); 1776 post_cnt = block_cnt - 1; 1777 /* prepare list for next posting block */ 1778 list_add_tail(&lpfc_ncmd->list, &prep_nblist); 1779 block_cnt = 1; 1780 } else { 1781 /* prepare list for next posting block */ 1782 list_add_tail(&lpfc_ncmd->list, &prep_nblist); 1783 /* enough sgls for non-embed sgl mbox command */ 1784 if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) { 1785 list_splice_init(&prep_nblist, &blck_nblist); 1786 post_cnt = block_cnt; 1787 block_cnt = 0; 1788 } 1789 } 1790 num_posting++; 1791 last_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag; 1792 1793 /* end of repost sgl list condition for NVME buffers */ 1794 if (num_posting == sb_count) { 1795 if (post_cnt == 0) { 1796 /* last sgl posting block */ 1797 list_splice_init(&prep_nblist, &blck_nblist); 1798 post_cnt = block_cnt; 1799 } else if (block_cnt == 1) { 1800 /* last single sgl with non-contiguous xri */ 1801 if (sgl_size > SGL_PAGE_SIZE) 1802 pdma_phys_sgl1 = 1803 lpfc_ncmd->dma_phys_sgl + 1804 SGL_PAGE_SIZE; 1805 else 1806 pdma_phys_sgl1 = 0; 1807 cur_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag; 1808 status = lpfc_sli4_post_sgl(phba, 1809 lpfc_ncmd->dma_phys_sgl, 1810 pdma_phys_sgl1, cur_xritag); 1811 if (status) { 1812 /* failure, put on abort nvme list */ 1813 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 1814 } else { 1815 /* success, put on NVME buffer list */ 1816 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 1817 lpfc_ncmd->status = IOSTAT_SUCCESS; 1818 num_posted++; 1819 } 1820 /* success, put on NVME buffer sgl list */ 1821 list_add_tail(&lpfc_ncmd->list, &nvme_nblist); 1822 } 1823 } 1824 1825 /* continue until a nembed page worth of sgls */ 1826 if (post_cnt == 0) 1827 continue; 1828 1829 /* post block of NVME buffer list sgls */ 1830 status = lpfc_sli4_post_nvme_sgl_block(phba, &blck_nblist, 1831 post_cnt); 1832 1833 /* don't reset xirtag due to hole in xri block */ 1834 if (block_cnt == 0) 1835 last_xritag = NO_XRI; 1836 1837 /* reset NVME buffer post count for next round of posting */ 1838 post_cnt = 0; 1839 1840 /* put posted NVME buffer-sgl posted on NVME buffer sgl list */ 1841 while (!list_empty(&blck_nblist)) { 1842 list_remove_head(&blck_nblist, lpfc_ncmd, 1843 struct lpfc_nvme_buf, list); 1844 if (status) { 1845 /* failure, put on abort nvme list */ 1846 lpfc_ncmd->flags |= LPFC_SBUF_XBUSY; 1847 } else { 1848 /* success, put on NVME buffer list */ 1849 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 1850 lpfc_ncmd->status = IOSTAT_SUCCESS; 1851 num_posted++; 1852 } 1853 list_add_tail(&lpfc_ncmd->list, &nvme_nblist); 1854 } 1855 } 1856 /* Push NVME buffers with sgl posted to the available list */ 1857 while (!list_empty(&nvme_nblist)) { 1858 list_remove_head(&nvme_nblist, lpfc_ncmd, 1859 struct lpfc_nvme_buf, list); 1860 lpfc_release_nvme_buf(phba, lpfc_ncmd); 1861 } 1862 return num_posted; 1863 } 1864 1865 /** 1866 * lpfc_repost_nvme_sgl_list - Repost all the allocated nvme buffer sgls 1867 * @phba: pointer to lpfc hba data structure. 1868 * 1869 * This routine walks the list of nvme buffers that have been allocated and 1870 * repost them to the port by using SGL block post. This is needed after a 1871 * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine 1872 * is responsible for moving all nvme buffers on the lpfc_abts_nvme_sgl_list 1873 * to the lpfc_nvme_buf_list. If the repost fails, reject all nvme buffers. 1874 * 1875 * Returns: 0 = success, non-zero failure. 1876 **/ 1877 int 1878 lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba) 1879 { 1880 LIST_HEAD(post_nblist); 1881 int num_posted, rc = 0; 1882 1883 /* get all NVME buffers need to repost to a local list */ 1884 spin_lock_irq(&phba->nvme_buf_list_get_lock); 1885 spin_lock(&phba->nvme_buf_list_put_lock); 1886 list_splice_init(&phba->lpfc_nvme_buf_list_get, &post_nblist); 1887 list_splice(&phba->lpfc_nvme_buf_list_put, &post_nblist); 1888 spin_unlock(&phba->nvme_buf_list_put_lock); 1889 spin_unlock_irq(&phba->nvme_buf_list_get_lock); 1890 1891 /* post the list of nvme buffer sgls to port if available */ 1892 if (!list_empty(&post_nblist)) { 1893 num_posted = lpfc_post_nvme_sgl_list(phba, &post_nblist, 1894 phba->sli4_hba.nvme_xri_cnt); 1895 /* failed to post any nvme buffer, return error */ 1896 if (num_posted == 0) 1897 rc = -EIO; 1898 } 1899 return rc; 1900 } 1901 1902 /** 1903 * lpfc_new_nvme_buf - Scsi buffer allocator for HBA with SLI4 IF spec 1904 * @vport: The virtual port for which this call being executed. 1905 * @num_to_allocate: The requested number of buffers to allocate. 1906 * 1907 * This routine allocates nvme buffers for device with SLI-4 interface spec, 1908 * the nvme buffer contains all the necessary information needed to initiate 1909 * a NVME I/O. After allocating up to @num_to_allocate NVME buffers and put 1910 * them on a list, it post them to the port by using SGL block post. 1911 * 1912 * Return codes: 1913 * int - number of nvme buffers that were allocated and posted. 1914 * 0 = failure, less than num_to_alloc is a partial failure. 1915 **/ 1916 static int 1917 lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc) 1918 { 1919 struct lpfc_hba *phba = vport->phba; 1920 struct lpfc_nvme_buf *lpfc_ncmd; 1921 struct lpfc_iocbq *pwqeq; 1922 union lpfc_wqe128 *wqe; 1923 struct sli4_sge *sgl; 1924 dma_addr_t pdma_phys_sgl; 1925 uint16_t iotag, lxri = 0; 1926 int bcnt, num_posted, sgl_size; 1927 LIST_HEAD(prep_nblist); 1928 LIST_HEAD(post_nblist); 1929 LIST_HEAD(nvme_nblist); 1930 1931 sgl_size = phba->cfg_sg_dma_buf_size; 1932 1933 for (bcnt = 0; bcnt < num_to_alloc; bcnt++) { 1934 lpfc_ncmd = kzalloc(sizeof(struct lpfc_nvme_buf), GFP_KERNEL); 1935 if (!lpfc_ncmd) 1936 break; 1937 /* 1938 * Get memory from the pci pool to map the virt space to 1939 * pci bus space for an I/O. The DMA buffer includes the 1940 * number of SGE's necessary to support the sg_tablesize. 1941 */ 1942 lpfc_ncmd->data = pci_pool_alloc(phba->lpfc_sg_dma_buf_pool, 1943 GFP_KERNEL, 1944 &lpfc_ncmd->dma_handle); 1945 if (!lpfc_ncmd->data) { 1946 kfree(lpfc_ncmd); 1947 break; 1948 } 1949 memset(lpfc_ncmd->data, 0, phba->cfg_sg_dma_buf_size); 1950 1951 lxri = lpfc_sli4_next_xritag(phba); 1952 if (lxri == NO_XRI) { 1953 pci_pool_free(phba->lpfc_sg_dma_buf_pool, 1954 lpfc_ncmd->data, lpfc_ncmd->dma_handle); 1955 kfree(lpfc_ncmd); 1956 break; 1957 } 1958 pwqeq = &(lpfc_ncmd->cur_iocbq); 1959 wqe = (union lpfc_wqe128 *)&pwqeq->wqe; 1960 1961 /* Allocate iotag for lpfc_ncmd->cur_iocbq. */ 1962 iotag = lpfc_sli_next_iotag(phba, pwqeq); 1963 if (iotag == 0) { 1964 pci_pool_free(phba->lpfc_sg_dma_buf_pool, 1965 lpfc_ncmd->data, lpfc_ncmd->dma_handle); 1966 kfree(lpfc_ncmd); 1967 lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR, 1968 "6121 Failed to allocated IOTAG for" 1969 " XRI:0x%x\n", lxri); 1970 lpfc_sli4_free_xri(phba, lxri); 1971 break; 1972 } 1973 pwqeq->sli4_lxritag = lxri; 1974 pwqeq->sli4_xritag = phba->sli4_hba.xri_ids[lxri]; 1975 pwqeq->iocb_flag |= LPFC_IO_NVME; 1976 pwqeq->context1 = lpfc_ncmd; 1977 pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl; 1978 1979 /* Initialize local short-hand pointers. */ 1980 lpfc_ncmd->nvme_sgl = lpfc_ncmd->data; 1981 sgl = lpfc_ncmd->nvme_sgl; 1982 pdma_phys_sgl = lpfc_ncmd->dma_handle; 1983 lpfc_ncmd->dma_phys_sgl = pdma_phys_sgl; 1984 1985 /* Rsp SGE will be filled in when we rcv an IO 1986 * from the NVME Layer to be sent. 1987 * The cmd is going to be embedded so we need a SKIP SGE. 1988 */ 1989 bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP); 1990 bf_set(lpfc_sli4_sge_last, sgl, 0); 1991 sgl->word2 = cpu_to_le32(sgl->word2); 1992 /* Fill in word 3 / sgl_len during cmd submission */ 1993 1994 lpfc_ncmd->cur_iocbq.context1 = lpfc_ncmd; 1995 1996 /* Word 7 */ 1997 bf_set(wqe_erp, &wqe->generic.wqe_com, 0); 1998 /* NVME upper layers will time things out, if needed */ 1999 bf_set(wqe_tmo, &wqe->generic.wqe_com, 0); 2000 2001 /* Word 10 */ 2002 bf_set(wqe_ebde_cnt, &wqe->generic.wqe_com, 0); 2003 bf_set(wqe_dbde, &wqe->generic.wqe_com, 1); 2004 2005 /* add the nvme buffer to a post list */ 2006 list_add_tail(&lpfc_ncmd->list, &post_nblist); 2007 spin_lock_irq(&phba->nvme_buf_list_get_lock); 2008 phba->sli4_hba.nvme_xri_cnt++; 2009 spin_unlock_irq(&phba->nvme_buf_list_get_lock); 2010 } 2011 lpfc_printf_log(phba, KERN_INFO, LOG_NVME, 2012 "6114 Allocate %d out of %d requested new NVME " 2013 "buffers\n", bcnt, num_to_alloc); 2014 2015 /* post the list of nvme buffer sgls to port if available */ 2016 if (!list_empty(&post_nblist)) 2017 num_posted = lpfc_post_nvme_sgl_list(phba, 2018 &post_nblist, bcnt); 2019 else 2020 num_posted = 0; 2021 2022 return num_posted; 2023 } 2024 2025 /** 2026 * lpfc_get_nvme_buf - Get a nvme buffer from lpfc_nvme_buf_list of the HBA 2027 * @phba: The HBA for which this call is being executed. 2028 * 2029 * This routine removes a nvme buffer from head of @phba lpfc_nvme_buf_list list 2030 * and returns to caller. 2031 * 2032 * Return codes: 2033 * NULL - Error 2034 * Pointer to lpfc_nvme_buf - Success 2035 **/ 2036 static struct lpfc_nvme_buf * 2037 lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp) 2038 { 2039 struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next; 2040 unsigned long iflag = 0; 2041 int found = 0; 2042 2043 spin_lock_irqsave(&phba->nvme_buf_list_get_lock, iflag); 2044 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, 2045 &phba->lpfc_nvme_buf_list_get, list) { 2046 if (lpfc_test_rrq_active(phba, ndlp, 2047 lpfc_ncmd->cur_iocbq.sli4_lxritag)) 2048 continue; 2049 list_del_init(&lpfc_ncmd->list); 2050 found = 1; 2051 break; 2052 } 2053 if (!found) { 2054 spin_lock(&phba->nvme_buf_list_put_lock); 2055 list_splice(&phba->lpfc_nvme_buf_list_put, 2056 &phba->lpfc_nvme_buf_list_get); 2057 INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put); 2058 spin_unlock(&phba->nvme_buf_list_put_lock); 2059 list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, 2060 &phba->lpfc_nvme_buf_list_get, list) { 2061 if (lpfc_test_rrq_active( 2062 phba, ndlp, lpfc_ncmd->cur_iocbq.sli4_lxritag)) 2063 continue; 2064 list_del_init(&lpfc_ncmd->list); 2065 found = 1; 2066 break; 2067 } 2068 } 2069 spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag); 2070 if (!found) 2071 return NULL; 2072 return lpfc_ncmd; 2073 } 2074 2075 /** 2076 * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list. 2077 * @phba: The Hba for which this call is being executed. 2078 * @lpfc_ncmd: The nvme buffer which is being released. 2079 * 2080 * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba 2081 * lpfc_nvme_buf_list list. For SLI4 XRI's are tied to the nvme buffer 2082 * and cannot be reused for at least RA_TOV amount of time if it was 2083 * aborted. 2084 **/ 2085 static void 2086 lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd) 2087 { 2088 unsigned long iflag = 0; 2089 2090 lpfc_ncmd->nonsg_phys = 0; 2091 if (lpfc_ncmd->flags & LPFC_SBUF_XBUSY) { 2092 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2093 "6310 XB release deferred for " 2094 "ox_id x%x on reqtag x%x\n", 2095 lpfc_ncmd->cur_iocbq.sli4_xritag, 2096 lpfc_ncmd->cur_iocbq.iotag); 2097 2098 spin_lock_irqsave(&phba->sli4_hba.abts_nvme_buf_list_lock, 2099 iflag); 2100 lpfc_ncmd->nvmeCmd = NULL; 2101 list_add_tail(&lpfc_ncmd->list, 2102 &phba->sli4_hba.lpfc_abts_nvme_buf_list); 2103 spin_unlock_irqrestore(&phba->sli4_hba.abts_nvme_buf_list_lock, 2104 iflag); 2105 } else { 2106 lpfc_ncmd->nvmeCmd = NULL; 2107 lpfc_ncmd->cur_iocbq.iocb_flag = LPFC_IO_NVME; 2108 spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag); 2109 list_add_tail(&lpfc_ncmd->list, &phba->lpfc_nvme_buf_list_put); 2110 spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag); 2111 } 2112 } 2113 2114 /** 2115 * lpfc_nvme_create_localport - Create/Bind an nvme localport instance. 2116 * @pvport - the lpfc_vport instance requesting a localport. 2117 * 2118 * This routine is invoked to create an nvme localport instance to bind 2119 * to the nvme_fc_transport. It is called once during driver load 2120 * like lpfc_create_shost after all other services are initialized. 2121 * It requires a vport, vpi, and wwns at call time. Other localport 2122 * parameters are modified as the driver's FCID and the Fabric WWN 2123 * are established. 2124 * 2125 * Return codes 2126 * 0 - successful 2127 * -ENOMEM - no heap memory available 2128 * other values - from nvme registration upcall 2129 **/ 2130 int 2131 lpfc_nvme_create_localport(struct lpfc_vport *vport) 2132 { 2133 int ret = 0; 2134 struct lpfc_hba *phba = vport->phba; 2135 struct nvme_fc_port_info nfcp_info; 2136 struct nvme_fc_local_port *localport; 2137 struct lpfc_nvme_lport *lport; 2138 int len; 2139 2140 /* Initialize this localport instance. The vport wwn usage ensures 2141 * that NPIV is accounted for. 2142 */ 2143 memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info)); 2144 nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR; 2145 nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn); 2146 nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn); 2147 2148 /* Limit to LPFC_MAX_NVME_SEG_CNT. 2149 * For now need + 1 to get around NVME transport logic. 2150 */ 2151 if (phba->cfg_sg_seg_cnt > LPFC_MAX_NVME_SEG_CNT) { 2152 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_INIT, 2153 "6300 Reducing sg segment cnt to %d\n", 2154 LPFC_MAX_NVME_SEG_CNT); 2155 phba->cfg_nvme_seg_cnt = LPFC_MAX_NVME_SEG_CNT; 2156 } else { 2157 phba->cfg_nvme_seg_cnt = phba->cfg_sg_seg_cnt; 2158 } 2159 lpfc_nvme_template.max_sgl_segments = phba->cfg_nvme_seg_cnt + 1; 2160 lpfc_nvme_template.max_hw_queues = phba->cfg_nvme_io_channel; 2161 2162 /* localport is allocated from the stack, but the registration 2163 * call allocates heap memory as well as the private area. 2164 */ 2165 #if (IS_ENABLED(CONFIG_NVME_FC)) 2166 ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template, 2167 &vport->phba->pcidev->dev, &localport); 2168 #else 2169 ret = -ENOMEM; 2170 #endif 2171 if (!ret) { 2172 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC, 2173 "6005 Successfully registered local " 2174 "NVME port num %d, localP %p, private %p, " 2175 "sg_seg %d\n", 2176 localport->port_num, localport, 2177 localport->private, 2178 lpfc_nvme_template.max_sgl_segments); 2179 2180 /* Private is our lport size declared in the template. */ 2181 lport = (struct lpfc_nvme_lport *)localport->private; 2182 vport->localport = localport; 2183 lport->vport = vport; 2184 vport->nvmei_support = 1; 2185 len = lpfc_new_nvme_buf(vport, phba->sli4_hba.nvme_xri_max); 2186 vport->phba->total_nvme_bufs += len; 2187 } 2188 2189 return ret; 2190 } 2191 2192 /** 2193 * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport. 2194 * @pnvme: pointer to lpfc nvme data structure. 2195 * 2196 * This routine is invoked to destroy all lports bound to the phba. 2197 * The lport memory was allocated by the nvme fc transport and is 2198 * released there. This routine ensures all rports bound to the 2199 * lport have been disconnected. 2200 * 2201 **/ 2202 void 2203 lpfc_nvme_destroy_localport(struct lpfc_vport *vport) 2204 { 2205 #if (IS_ENABLED(CONFIG_NVME_FC)) 2206 struct nvme_fc_local_port *localport; 2207 struct lpfc_nvme_lport *lport; 2208 int ret; 2209 2210 if (vport->nvmei_support == 0) 2211 return; 2212 2213 localport = vport->localport; 2214 vport->localport = NULL; 2215 lport = (struct lpfc_nvme_lport *)localport->private; 2216 2217 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2218 "6011 Destroying NVME localport %p\n", 2219 localport); 2220 2221 /* lport's rport list is clear. Unregister 2222 * lport and release resources. 2223 */ 2224 init_completion(&lport->lport_unreg_done); 2225 ret = nvme_fc_unregister_localport(localport); 2226 wait_for_completion_timeout(&lport->lport_unreg_done, 5); 2227 2228 /* Regardless of the unregister upcall response, clear 2229 * nvmei_support. All rports are unregistered and the 2230 * driver will clean up. 2231 */ 2232 vport->nvmei_support = 0; 2233 if (ret == 0) { 2234 lpfc_printf_vlog(vport, 2235 KERN_INFO, LOG_NVME_DISC, 2236 "6009 Unregistered lport Success\n"); 2237 } else { 2238 lpfc_printf_vlog(vport, 2239 KERN_INFO, LOG_NVME_DISC, 2240 "6010 Unregistered lport " 2241 "Failed, status x%x\n", 2242 ret); 2243 } 2244 #endif 2245 } 2246 2247 void 2248 lpfc_nvme_update_localport(struct lpfc_vport *vport) 2249 { 2250 #if (IS_ENABLED(CONFIG_NVME_FC)) 2251 struct nvme_fc_local_port *localport; 2252 struct lpfc_nvme_lport *lport; 2253 2254 localport = vport->localport; 2255 if (!localport) { 2256 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2257 "6710 Update NVME fail. No localport\n"); 2258 return; 2259 } 2260 lport = (struct lpfc_nvme_lport *)localport->private; 2261 if (!lport) { 2262 lpfc_printf_vlog(vport, KERN_WARNING, LOG_NVME, 2263 "6171 Update NVME fail. localP %p, No lport\n", 2264 localport); 2265 return; 2266 } 2267 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME, 2268 "6012 Update NVME lport %p did x%x\n", 2269 localport, vport->fc_myDID); 2270 2271 localport->port_id = vport->fc_myDID; 2272 if (localport->port_id == 0) 2273 localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY; 2274 else 2275 localport->port_role = FC_PORT_ROLE_NVME_INITIATOR; 2276 2277 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2278 "6030 bound lport %p to DID x%06x\n", 2279 lport, localport->port_id); 2280 #endif 2281 } 2282 2283 int 2284 lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2285 { 2286 #if (IS_ENABLED(CONFIG_NVME_FC)) 2287 int ret = 0; 2288 struct nvme_fc_local_port *localport; 2289 struct lpfc_nvme_lport *lport; 2290 struct lpfc_nvme_rport *rport; 2291 struct nvme_fc_remote_port *remote_port; 2292 struct nvme_fc_port_info rpinfo; 2293 2294 lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC, 2295 "6006 Register NVME PORT. DID x%06x nlptype x%x\n", 2296 ndlp->nlp_DID, ndlp->nlp_type); 2297 2298 localport = vport->localport; 2299 lport = (struct lpfc_nvme_lport *)localport->private; 2300 2301 /* NVME rports are not preserved across devloss. 2302 * Just register this instance. Note, rpinfo->dev_loss_tmo 2303 * is left 0 to indicate accept transport defaults. The 2304 * driver communicates port role capabilities consistent 2305 * with the PRLI response data. 2306 */ 2307 memset(&rpinfo, 0, sizeof(struct nvme_fc_port_info)); 2308 rpinfo.port_id = ndlp->nlp_DID; 2309 if (ndlp->nlp_type & NLP_NVME_TARGET) 2310 rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET; 2311 if (ndlp->nlp_type & NLP_NVME_INITIATOR) 2312 rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR; 2313 2314 if (ndlp->nlp_type & NLP_NVME_DISCOVERY) 2315 rpinfo.port_role |= FC_PORT_ROLE_NVME_DISCOVERY; 2316 2317 rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn); 2318 rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn); 2319 ret = nvme_fc_register_remoteport(localport, &rpinfo, &remote_port); 2320 if (!ret) { 2321 /* If the ndlp already has an nrport, this is just 2322 * a resume of the existing rport. Else this is a 2323 * new rport. 2324 */ 2325 rport = remote_port->private; 2326 if (ndlp->nrport == rport) { 2327 lpfc_printf_vlog(ndlp->vport, KERN_INFO, 2328 LOG_NVME_DISC, 2329 "6014 Rebinding lport to " 2330 "rport wwpn 0x%llx, " 2331 "Data: x%x x%x x%x x%06x\n", 2332 remote_port->port_name, 2333 remote_port->port_id, 2334 remote_port->port_role, 2335 ndlp->nlp_type, 2336 ndlp->nlp_DID); 2337 } else { 2338 /* New rport. */ 2339 rport->remoteport = remote_port; 2340 rport->lport = lport; 2341 rport->ndlp = lpfc_nlp_get(ndlp); 2342 if (!rport->ndlp) 2343 return -1; 2344 ndlp->nrport = rport; 2345 lpfc_printf_vlog(vport, KERN_INFO, 2346 LOG_NVME_DISC | LOG_NODE, 2347 "6022 Binding new rport to " 2348 "lport %p Rport WWNN 0x%llx, " 2349 "Rport WWPN 0x%llx DID " 2350 "x%06x Role x%x\n", 2351 lport, 2352 rpinfo.node_name, rpinfo.port_name, 2353 rpinfo.port_id, rpinfo.port_role); 2354 } 2355 } else { 2356 lpfc_printf_vlog(vport, KERN_ERR, 2357 LOG_NVME_DISC | LOG_NODE, 2358 "6031 RemotePort Registration failed " 2359 "err: %d, DID x%06x\n", 2360 ret, ndlp->nlp_DID); 2361 } 2362 2363 return ret; 2364 #else 2365 return 0; 2366 #endif 2367 } 2368 2369 /* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport. 2370 * 2371 * There is no notion of Devloss or rport recovery from the current 2372 * nvme_transport perspective. Loss of an rport just means IO cannot 2373 * be sent and recovery is completely up to the initator. 2374 * For now, the driver just unbinds the DID and port_role so that 2375 * no further IO can be issued. Changes are planned for later. 2376 * 2377 * Notes - the ndlp reference count is not decremented here since 2378 * since there is no nvme_transport api for devloss. Node ref count 2379 * is only adjusted in driver unload. 2380 */ 2381 void 2382 lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) 2383 { 2384 #if (IS_ENABLED(CONFIG_NVME_FC)) 2385 int ret; 2386 struct nvme_fc_local_port *localport; 2387 struct lpfc_nvme_lport *lport; 2388 struct lpfc_nvme_rport *rport; 2389 struct nvme_fc_remote_port *remoteport; 2390 2391 localport = vport->localport; 2392 2393 /* This is fundamental error. The localport is always 2394 * available until driver unload. Just exit. 2395 */ 2396 if (!localport) 2397 return; 2398 2399 lport = (struct lpfc_nvme_lport *)localport->private; 2400 if (!lport) 2401 goto input_err; 2402 2403 rport = ndlp->nrport; 2404 if (!rport) 2405 goto input_err; 2406 2407 remoteport = rport->remoteport; 2408 lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC, 2409 "6033 Unreg nvme remoteport %p, portname x%llx, " 2410 "port_id x%06x, portstate x%x port type x%x\n", 2411 remoteport, remoteport->port_name, 2412 remoteport->port_id, remoteport->port_state, 2413 ndlp->nlp_type); 2414 2415 /* Sanity check ndlp type. Only call for NVME ports. Don't 2416 * clear any rport state until the transport calls back. 2417 */ 2418 if (ndlp->nlp_type & (NLP_NVME_TARGET | NLP_NVME_INITIATOR)) { 2419 init_completion(&rport->rport_unreg_done); 2420 2421 /* No concern about the role change on the nvme remoteport. 2422 * The transport will update it. 2423 */ 2424 ret = nvme_fc_unregister_remoteport(remoteport); 2425 if (ret != 0) { 2426 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2427 "6167 NVME unregister failed %d " 2428 "port_state x%x\n", 2429 ret, remoteport->port_state); 2430 } 2431 2432 } 2433 return; 2434 2435 input_err: 2436 #endif 2437 lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC, 2438 "6168 State error: lport %p, rport%p FCID x%06x\n", 2439 vport->localport, ndlp->rport, ndlp->nlp_DID); 2440 } 2441 2442 /** 2443 * lpfc_sli4_nvme_xri_aborted - Fast-path process of NVME xri abort 2444 * @phba: pointer to lpfc hba data structure. 2445 * @axri: pointer to the fcp xri abort wcqe structure. 2446 * 2447 * This routine is invoked by the worker thread to process a SLI4 fast-path 2448 * FCP aborted xri. 2449 **/ 2450 void 2451 lpfc_sli4_nvme_xri_aborted(struct lpfc_hba *phba, 2452 struct sli4_wcqe_xri_aborted *axri) 2453 { 2454 uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri); 2455 uint16_t rxid = bf_get(lpfc_wcqe_xa_remote_xid, axri); 2456 struct lpfc_nvme_buf *lpfc_ncmd, *next_lpfc_ncmd; 2457 struct lpfc_nodelist *ndlp; 2458 unsigned long iflag = 0; 2459 int rrq_empty = 0; 2460 2461 if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME)) 2462 return; 2463 spin_lock_irqsave(&phba->hbalock, iflag); 2464 spin_lock(&phba->sli4_hba.abts_nvme_buf_list_lock); 2465 list_for_each_entry_safe(lpfc_ncmd, next_lpfc_ncmd, 2466 &phba->sli4_hba.lpfc_abts_nvme_buf_list, 2467 list) { 2468 if (lpfc_ncmd->cur_iocbq.sli4_xritag == xri) { 2469 list_del_init(&lpfc_ncmd->list); 2470 lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY; 2471 lpfc_ncmd->status = IOSTAT_SUCCESS; 2472 spin_unlock( 2473 &phba->sli4_hba.abts_nvme_buf_list_lock); 2474 2475 rrq_empty = list_empty(&phba->active_rrq_list); 2476 spin_unlock_irqrestore(&phba->hbalock, iflag); 2477 ndlp = lpfc_ncmd->ndlp; 2478 if (ndlp) { 2479 lpfc_set_rrq_active( 2480 phba, ndlp, 2481 lpfc_ncmd->cur_iocbq.sli4_lxritag, 2482 rxid, 1); 2483 lpfc_sli4_abts_err_handler(phba, ndlp, axri); 2484 } 2485 2486 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2487 "6311 XRI Aborted xri x%x tag x%x " 2488 "released\n", 2489 xri, lpfc_ncmd->cur_iocbq.iotag); 2490 2491 lpfc_release_nvme_buf(phba, lpfc_ncmd); 2492 if (rrq_empty) 2493 lpfc_worker_wake_up(phba); 2494 return; 2495 } 2496 } 2497 spin_unlock(&phba->sli4_hba.abts_nvme_buf_list_lock); 2498 spin_unlock_irqrestore(&phba->hbalock, iflag); 2499 2500 lpfc_printf_log(phba, KERN_INFO, LOG_NVME_ABTS, 2501 "6312 XRI Aborted xri x%x not found\n", xri); 2502 2503 } 2504