xref: /linux/drivers/scsi/elx/efct/efct_hw.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
4  * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
5  */
6 
7 #include "efct_driver.h"
8 #include "efct_hw.h"
9 #include "efct_unsol.h"
10 
11 struct efct_hw_link_stat_cb_arg {
12 	void (*cb)(int status, u32 num_counters,
13 		   struct efct_hw_link_stat_counts *counters, void *arg);
14 	void *arg;
15 };
16 
17 struct efct_hw_host_stat_cb_arg {
18 	void (*cb)(int status, u32 num_counters,
19 		   struct efct_hw_host_stat_counts *counters, void *arg);
20 	void *arg;
21 };
22 
23 struct efct_hw_fw_wr_cb_arg {
24 	void (*cb)(int status, u32 bytes_written, u32 change_status, void *arg);
25 	void *arg;
26 };
27 
28 struct efct_mbox_rqst_ctx {
29 	int (*callback)(struct efc *efc, int status, u8 *mqe, void *arg);
30 	void *arg;
31 };
32 
33 static int
34 efct_hw_link_event_init(struct efct_hw *hw)
35 {
36 	hw->link.status = SLI4_LINK_STATUS_MAX;
37 	hw->link.topology = SLI4_LINK_TOPO_NONE;
38 	hw->link.medium = SLI4_LINK_MEDIUM_MAX;
39 	hw->link.speed = 0;
40 	hw->link.loop_map = NULL;
41 	hw->link.fc_id = U32_MAX;
42 
43 	return 0;
44 }
45 
46 static int
47 efct_hw_read_max_dump_size(struct efct_hw *hw)
48 {
49 	u8 buf[SLI4_BMBX_SIZE];
50 	struct efct *efct = hw->os;
51 	int rc = 0;
52 	struct sli4_rsp_cmn_set_dump_location *rsp;
53 
54 	/* attempt to detemine the dump size for function 0 only. */
55 	if (PCI_FUNC(efct->pci->devfn) != 0)
56 		return rc;
57 
58 	if (sli_cmd_common_set_dump_location(&hw->sli, buf, 1, 0, NULL, 0))
59 		return -EIO;
60 
61 	rsp = (struct sli4_rsp_cmn_set_dump_location *)
62 	      (buf + offsetof(struct sli4_cmd_sli_config, payload.embed));
63 
64 	rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
65 	if (rc != 0) {
66 		efc_log_debug(hw->os, "set dump location cmd failed\n");
67 		return rc;
68 	}
69 
70 	hw->dump_size =
71 	  le32_to_cpu(rsp->buffer_length_dword) & SLI4_CMN_SET_DUMP_BUFFER_LEN;
72 
73 	efc_log_debug(hw->os, "Dump size %x\n",	hw->dump_size);
74 
75 	return rc;
76 }
77 
78 static int
79 __efct_read_topology_cb(struct efct_hw *hw, int status, u8 *mqe, void *arg)
80 {
81 	struct sli4_cmd_read_topology *read_topo =
82 				(struct sli4_cmd_read_topology *)mqe;
83 	u8 speed;
84 	struct efc_domain_record drec = {0};
85 	struct efct *efct = hw->os;
86 
87 	if (status || le16_to_cpu(read_topo->hdr.status)) {
88 		efc_log_debug(hw->os, "bad status cqe=%#x mqe=%#x\n", status,
89 			      le16_to_cpu(read_topo->hdr.status));
90 		return -EIO;
91 	}
92 
93 	switch (le32_to_cpu(read_topo->dw2_attentype) &
94 		SLI4_READTOPO_ATTEN_TYPE) {
95 	case SLI4_READ_TOPOLOGY_LINK_UP:
96 		hw->link.status = SLI4_LINK_STATUS_UP;
97 		break;
98 	case SLI4_READ_TOPOLOGY_LINK_DOWN:
99 		hw->link.status = SLI4_LINK_STATUS_DOWN;
100 		break;
101 	case SLI4_READ_TOPOLOGY_LINK_NO_ALPA:
102 		hw->link.status = SLI4_LINK_STATUS_NO_ALPA;
103 		break;
104 	default:
105 		hw->link.status = SLI4_LINK_STATUS_MAX;
106 		break;
107 	}
108 
109 	switch (read_topo->topology) {
110 	case SLI4_READ_TOPO_NON_FC_AL:
111 		hw->link.topology = SLI4_LINK_TOPO_NON_FC_AL;
112 		break;
113 	case SLI4_READ_TOPO_FC_AL:
114 		hw->link.topology = SLI4_LINK_TOPO_FC_AL;
115 		if (hw->link.status == SLI4_LINK_STATUS_UP)
116 			hw->link.loop_map = hw->loop_map.virt;
117 		hw->link.fc_id = read_topo->acquired_al_pa;
118 		break;
119 	default:
120 		hw->link.topology = SLI4_LINK_TOPO_MAX;
121 		break;
122 	}
123 
124 	hw->link.medium = SLI4_LINK_MEDIUM_FC;
125 
126 	speed = (le32_to_cpu(read_topo->currlink_state) &
127 		 SLI4_READTOPO_LINKSTATE_SPEED) >> 8;
128 	switch (speed) {
129 	case SLI4_READ_TOPOLOGY_SPEED_1G:
130 		hw->link.speed =  1 * 1000;
131 		break;
132 	case SLI4_READ_TOPOLOGY_SPEED_2G:
133 		hw->link.speed =  2 * 1000;
134 		break;
135 	case SLI4_READ_TOPOLOGY_SPEED_4G:
136 		hw->link.speed =  4 * 1000;
137 		break;
138 	case SLI4_READ_TOPOLOGY_SPEED_8G:
139 		hw->link.speed =  8 * 1000;
140 		break;
141 	case SLI4_READ_TOPOLOGY_SPEED_16G:
142 		hw->link.speed = 16 * 1000;
143 		break;
144 	case SLI4_READ_TOPOLOGY_SPEED_32G:
145 		hw->link.speed = 32 * 1000;
146 		break;
147 	case SLI4_READ_TOPOLOGY_SPEED_64G:
148 		hw->link.speed = 64 * 1000;
149 		break;
150 	case SLI4_READ_TOPOLOGY_SPEED_128G:
151 		hw->link.speed = 128 * 1000;
152 		break;
153 	}
154 
155 	drec.speed = hw->link.speed;
156 	drec.fc_id = hw->link.fc_id;
157 	drec.is_nport = true;
158 	efc_domain_cb(efct->efcport, EFC_HW_DOMAIN_FOUND, &drec);
159 
160 	return 0;
161 }
162 
163 static int
164 efct_hw_cb_link(void *ctx, void *e)
165 {
166 	struct efct_hw *hw = ctx;
167 	struct sli4_link_event *event = e;
168 	struct efc_domain *d = NULL;
169 	int rc = 0;
170 	struct efct *efct = hw->os;
171 
172 	efct_hw_link_event_init(hw);
173 
174 	switch (event->status) {
175 	case SLI4_LINK_STATUS_UP:
176 
177 		hw->link = *event;
178 		efct->efcport->link_status = EFC_LINK_STATUS_UP;
179 
180 		if (event->topology == SLI4_LINK_TOPO_NON_FC_AL) {
181 			struct efc_domain_record drec = {0};
182 
183 			efc_log_info(hw->os, "Link Up, NPORT, speed is %d\n",
184 				     event->speed);
185 			drec.speed = event->speed;
186 			drec.fc_id = event->fc_id;
187 			drec.is_nport = true;
188 			efc_domain_cb(efct->efcport, EFC_HW_DOMAIN_FOUND,
189 				      &drec);
190 		} else if (event->topology == SLI4_LINK_TOPO_FC_AL) {
191 			u8 buf[SLI4_BMBX_SIZE];
192 
193 			efc_log_info(hw->os, "Link Up, LOOP, speed is %d\n",
194 				     event->speed);
195 
196 			if (!sli_cmd_read_topology(&hw->sli, buf,
197 						   &hw->loop_map)) {
198 				rc = efct_hw_command(hw, buf, EFCT_CMD_NOWAIT,
199 						__efct_read_topology_cb, NULL);
200 			}
201 
202 			if (rc)
203 				efc_log_debug(hw->os, "READ_TOPOLOGY failed\n");
204 		} else {
205 			efc_log_info(hw->os, "%s(%#x), speed is %d\n",
206 				     "Link Up, unsupported topology ",
207 				     event->topology, event->speed);
208 		}
209 		break;
210 	case SLI4_LINK_STATUS_DOWN:
211 		efc_log_info(hw->os, "Link down\n");
212 
213 		hw->link.status = event->status;
214 		efct->efcport->link_status = EFC_LINK_STATUS_DOWN;
215 
216 		d = efct->efcport->domain;
217 		if (d)
218 			efc_domain_cb(efct->efcport, EFC_HW_DOMAIN_LOST, d);
219 		break;
220 	default:
221 		efc_log_debug(hw->os, "unhandled link status %#x\n",
222 			      event->status);
223 		break;
224 	}
225 
226 	return 0;
227 }
228 
229 int
230 efct_hw_setup(struct efct_hw *hw, void *os, struct pci_dev *pdev)
231 {
232 	u32 i, max_sgl, cpus;
233 
234 	if (hw->hw_setup_called)
235 		return 0;
236 
237 	/*
238 	 * efct_hw_init() relies on NULL pointers indicating that a structure
239 	 * needs allocation. If a structure is non-NULL, efct_hw_init() won't
240 	 * free/realloc that memory
241 	 */
242 	memset(hw, 0, sizeof(struct efct_hw));
243 
244 	hw->hw_setup_called = true;
245 
246 	hw->os = os;
247 
248 	mutex_init(&hw->bmbx_lock);
249 	spin_lock_init(&hw->cmd_lock);
250 	INIT_LIST_HEAD(&hw->cmd_head);
251 	INIT_LIST_HEAD(&hw->cmd_pending);
252 	hw->cmd_head_count = 0;
253 
254 	/* Create mailbox command ctx pool */
255 	hw->cmd_ctx_pool = mempool_create_kmalloc_pool(EFCT_CMD_CTX_POOL_SZ,
256 					sizeof(struct efct_command_ctx));
257 	if (!hw->cmd_ctx_pool) {
258 		efc_log_err(hw->os, "failed to allocate mailbox buffer pool\n");
259 		return -EIO;
260 	}
261 
262 	/* Create mailbox request ctx pool for library callback */
263 	hw->mbox_rqst_pool = mempool_create_kmalloc_pool(EFCT_CMD_CTX_POOL_SZ,
264 					sizeof(struct efct_mbox_rqst_ctx));
265 	if (!hw->mbox_rqst_pool) {
266 		efc_log_err(hw->os, "failed to allocate mbox request pool\n");
267 		return -EIO;
268 	}
269 
270 	spin_lock_init(&hw->io_lock);
271 	INIT_LIST_HEAD(&hw->io_inuse);
272 	INIT_LIST_HEAD(&hw->io_free);
273 	INIT_LIST_HEAD(&hw->io_wait_free);
274 
275 	atomic_set(&hw->io_alloc_failed_count, 0);
276 
277 	hw->config.speed = SLI4_LINK_SPEED_AUTO_16_8_4;
278 	if (sli_setup(&hw->sli, hw->os, pdev, ((struct efct *)os)->reg)) {
279 		efc_log_err(hw->os, "SLI setup failed\n");
280 		return -EIO;
281 	}
282 
283 	efct_hw_link_event_init(hw);
284 
285 	sli_callback(&hw->sli, SLI4_CB_LINK, efct_hw_cb_link, hw);
286 
287 	/*
288 	 * Set all the queue sizes to the maximum allowed.
289 	 */
290 	for (i = 0; i < ARRAY_SIZE(hw->num_qentries); i++)
291 		hw->num_qentries[i] = hw->sli.qinfo.max_qentries[i];
292 	/*
293 	 * Adjust the size of the WQs so that the CQ is twice as big as
294 	 * the WQ to allow for 2 completions per IO. This allows us to
295 	 * handle multi-phase as well as aborts.
296 	 */
297 	hw->num_qentries[SLI4_QTYPE_WQ] = hw->num_qentries[SLI4_QTYPE_CQ] / 2;
298 
299 	/*
300 	 * The RQ assignment for RQ pair mode.
301 	 */
302 
303 	hw->config.rq_default_buffer_size = EFCT_HW_RQ_SIZE_PAYLOAD;
304 	hw->config.n_io = hw->sli.ext[SLI4_RSRC_XRI].size;
305 
306 	cpus = num_possible_cpus();
307 	hw->config.n_eq = cpus > EFCT_HW_MAX_NUM_EQ ? EFCT_HW_MAX_NUM_EQ : cpus;
308 
309 	max_sgl = sli_get_max_sgl(&hw->sli) - SLI4_SGE_MAX_RESERVED;
310 	max_sgl = (max_sgl > EFCT_FC_MAX_SGL) ? EFCT_FC_MAX_SGL : max_sgl;
311 	hw->config.n_sgl = max_sgl;
312 
313 	(void)efct_hw_read_max_dump_size(hw);
314 
315 	return 0;
316 }
317 
318 static void
319 efct_logfcfi(struct efct_hw *hw, u32 j, u32 i, u32 id)
320 {
321 	efc_log_info(hw->os,
322 		     "REG_FCFI: filter[%d] %08X -> RQ[%d] id=%d\n",
323 		     j, hw->config.filter_def[j], i, id);
324 }
325 
326 static inline void
327 efct_hw_init_free_io(struct efct_hw_io *io)
328 {
329 	/*
330 	 * Set io->done to NULL, to avoid any callbacks, should
331 	 * a completion be received for one of these IOs
332 	 */
333 	io->done = NULL;
334 	io->abort_done = NULL;
335 	io->status_saved = false;
336 	io->abort_in_progress = false;
337 	io->type = 0xFFFF;
338 	io->wq = NULL;
339 }
340 
341 static bool efct_hw_iotype_is_originator(u16 io_type)
342 {
343 	switch (io_type) {
344 	case EFCT_HW_FC_CT:
345 	case EFCT_HW_ELS_REQ:
346 		return true;
347 	default:
348 		return false;
349 	}
350 }
351 
352 static void
353 efct_hw_io_restore_sgl(struct efct_hw *hw, struct efct_hw_io *io)
354 {
355 	/* Restore the default */
356 	io->sgl = &io->def_sgl;
357 	io->sgl_count = io->def_sgl_count;
358 }
359 
360 static void
361 efct_hw_wq_process_io(void *arg, u8 *cqe, int status)
362 {
363 	struct efct_hw_io *io = arg;
364 	struct efct_hw *hw = io->hw;
365 	struct sli4_fc_wcqe *wcqe = (void *)cqe;
366 	u32	len = 0;
367 	u32 ext = 0;
368 
369 	/* clear xbusy flag if WCQE[XB] is clear */
370 	if (io->xbusy && (wcqe->flags & SLI4_WCQE_XB) == 0)
371 		io->xbusy = false;
372 
373 	/* get extended CQE status */
374 	switch (io->type) {
375 	case EFCT_HW_BLS_ACC:
376 	case EFCT_HW_BLS_RJT:
377 		break;
378 	case EFCT_HW_ELS_REQ:
379 		sli_fc_els_did(&hw->sli, cqe, &ext);
380 		len = sli_fc_response_length(&hw->sli, cqe);
381 		break;
382 	case EFCT_HW_ELS_RSP:
383 	case EFCT_HW_FC_CT_RSP:
384 		break;
385 	case EFCT_HW_FC_CT:
386 		len = sli_fc_response_length(&hw->sli, cqe);
387 		break;
388 	case EFCT_HW_IO_TARGET_WRITE:
389 		len = sli_fc_io_length(&hw->sli, cqe);
390 		break;
391 	case EFCT_HW_IO_TARGET_READ:
392 		len = sli_fc_io_length(&hw->sli, cqe);
393 		break;
394 	case EFCT_HW_IO_TARGET_RSP:
395 		break;
396 	case EFCT_HW_IO_DNRX_REQUEUE:
397 		/* release the count for re-posting the buffer */
398 		/* efct_hw_io_free(hw, io); */
399 		break;
400 	default:
401 		efc_log_err(hw->os, "unhandled io type %#x for XRI 0x%x\n",
402 			    io->type, io->indicator);
403 		break;
404 	}
405 	if (status) {
406 		ext = sli_fc_ext_status(&hw->sli, cqe);
407 		/*
408 		 * If we're not an originator IO, and XB is set, then issue
409 		 * abort for the IO from within the HW
410 		 */
411 		if (efct_hw_iotype_is_originator(io->type) &&
412 		    wcqe->flags & SLI4_WCQE_XB) {
413 			int rc;
414 
415 			efc_log_debug(hw->os, "aborting xri=%#x tag=%#x\n",
416 				      io->indicator, io->reqtag);
417 
418 			/*
419 			 * Because targets may send a response when the IO
420 			 * completes using the same XRI, we must wait for the
421 			 * XRI_ABORTED CQE to issue the IO callback
422 			 */
423 			rc = efct_hw_io_abort(hw, io, false, NULL, NULL);
424 			if (rc == 0) {
425 				/*
426 				 * latch status to return after abort is
427 				 * complete
428 				 */
429 				io->status_saved = true;
430 				io->saved_status = status;
431 				io->saved_ext = ext;
432 				io->saved_len = len;
433 				goto exit_efct_hw_wq_process_io;
434 			} else if (rc == -EINPROGRESS) {
435 				/*
436 				 * Already being aborted by someone else (ABTS
437 				 * perhaps). Just return original
438 				 * error.
439 				 */
440 				efc_log_debug(hw->os, "%s%#x tag=%#x\n",
441 					      "abort in progress xri=",
442 					      io->indicator, io->reqtag);
443 
444 			} else {
445 				/* Failed to abort for some other reason, log
446 				 * error
447 				 */
448 				efc_log_debug(hw->os, "%s%#x tag=%#x rc=%d\n",
449 					      "Failed to abort xri=",
450 					      io->indicator, io->reqtag, rc);
451 			}
452 		}
453 	}
454 
455 	if (io->done) {
456 		efct_hw_done_t done = io->done;
457 
458 		io->done = NULL;
459 
460 		if (io->status_saved) {
461 			/* use latched status if exists */
462 			status = io->saved_status;
463 			len = io->saved_len;
464 			ext = io->saved_ext;
465 			io->status_saved = false;
466 		}
467 
468 		/* Restore default SGL */
469 		efct_hw_io_restore_sgl(hw, io);
470 		done(io, len, status, ext, io->arg);
471 	}
472 
473 exit_efct_hw_wq_process_io:
474 	return;
475 }
476 
477 static int
478 efct_hw_setup_io(struct efct_hw *hw)
479 {
480 	u32	i = 0;
481 	struct efct_hw_io	*io = NULL;
482 	uintptr_t	xfer_virt = 0;
483 	uintptr_t	xfer_phys = 0;
484 	u32	index;
485 	bool new_alloc = true;
486 	struct efc_dma *dma;
487 	struct efct *efct = hw->os;
488 
489 	if (!hw->io) {
490 		hw->io = kmalloc_array(hw->config.n_io, sizeof(io), GFP_KERNEL);
491 		if (!hw->io)
492 			return -ENOMEM;
493 
494 		memset(hw->io, 0, hw->config.n_io * sizeof(io));
495 
496 		for (i = 0; i < hw->config.n_io; i++) {
497 			hw->io[i] = kzalloc(sizeof(*io), GFP_KERNEL);
498 			if (!hw->io[i])
499 				goto error;
500 		}
501 
502 		/* Create WQE buffs for IO */
503 		hw->wqe_buffs = kzalloc((hw->config.n_io * hw->sli.wqe_size),
504 					GFP_KERNEL);
505 		if (!hw->wqe_buffs) {
506 			kfree(hw->io);
507 			return -ENOMEM;
508 		}
509 
510 	} else {
511 		/* re-use existing IOs, including SGLs */
512 		new_alloc = false;
513 	}
514 
515 	if (new_alloc) {
516 		dma = &hw->xfer_rdy;
517 		dma->size = sizeof(struct fcp_txrdy) * hw->config.n_io;
518 		dma->virt = dma_alloc_coherent(&efct->pci->dev,
519 					       dma->size, &dma->phys, GFP_KERNEL);
520 		if (!dma->virt)
521 			return -ENOMEM;
522 	}
523 	xfer_virt = (uintptr_t)hw->xfer_rdy.virt;
524 	xfer_phys = hw->xfer_rdy.phys;
525 
526 	/* Initialize the pool of HW IO objects */
527 	for (i = 0; i < hw->config.n_io; i++) {
528 		struct hw_wq_callback *wqcb;
529 
530 		io = hw->io[i];
531 
532 		/* initialize IO fields */
533 		io->hw = hw;
534 
535 		/* Assign a WQE buff */
536 		io->wqe.wqebuf = &hw->wqe_buffs[i * hw->sli.wqe_size];
537 
538 		/* Allocate the request tag for this IO */
539 		wqcb = efct_hw_reqtag_alloc(hw, efct_hw_wq_process_io, io);
540 		if (!wqcb) {
541 			efc_log_err(hw->os, "can't allocate request tag\n");
542 			return -ENOSPC;
543 		}
544 		io->reqtag = wqcb->instance_index;
545 
546 		/* Now for the fields that are initialized on each free */
547 		efct_hw_init_free_io(io);
548 
549 		/* The XB flag isn't cleared on IO free, so init to zero */
550 		io->xbusy = 0;
551 
552 		if (sli_resource_alloc(&hw->sli, SLI4_RSRC_XRI,
553 				       &io->indicator, &index)) {
554 			efc_log_err(hw->os,
555 				    "sli_resource_alloc failed @ %d\n", i);
556 			return -ENOMEM;
557 		}
558 
559 		if (new_alloc) {
560 			dma = &io->def_sgl;
561 			dma->size = hw->config.n_sgl *
562 					sizeof(struct sli4_sge);
563 			dma->virt = dma_alloc_coherent(&efct->pci->dev,
564 						       dma->size, &dma->phys,
565 						       GFP_KERNEL);
566 			if (!dma->virt) {
567 				efc_log_err(hw->os, "dma_alloc fail %d\n", i);
568 				memset(&io->def_sgl, 0,
569 				       sizeof(struct efc_dma));
570 				return -ENOMEM;
571 			}
572 		}
573 		io->def_sgl_count = hw->config.n_sgl;
574 		io->sgl = &io->def_sgl;
575 		io->sgl_count = io->def_sgl_count;
576 
577 		if (hw->xfer_rdy.size) {
578 			io->xfer_rdy.virt = (void *)xfer_virt;
579 			io->xfer_rdy.phys = xfer_phys;
580 			io->xfer_rdy.size = sizeof(struct fcp_txrdy);
581 
582 			xfer_virt += sizeof(struct fcp_txrdy);
583 			xfer_phys += sizeof(struct fcp_txrdy);
584 		}
585 	}
586 
587 	return 0;
588 error:
589 	for (i = 0; i < hw->config.n_io && hw->io[i]; i++) {
590 		kfree(hw->io[i]);
591 		hw->io[i] = NULL;
592 	}
593 
594 	kfree(hw->io);
595 	hw->io = NULL;
596 
597 	return -ENOMEM;
598 }
599 
600 static int
601 efct_hw_init_prereg_io(struct efct_hw *hw)
602 {
603 	u32 i, idx = 0;
604 	struct efct_hw_io *io = NULL;
605 	u8 cmd[SLI4_BMBX_SIZE];
606 	int rc = 0;
607 	u32 n_rem;
608 	u32 n = 0;
609 	u32 sgls_per_request = 256;
610 	struct efc_dma **sgls = NULL;
611 	struct efc_dma req;
612 	struct efct *efct = hw->os;
613 
614 	sgls = kmalloc_array(sgls_per_request, sizeof(*sgls), GFP_KERNEL);
615 	if (!sgls)
616 		return -ENOMEM;
617 
618 	memset(&req, 0, sizeof(struct efc_dma));
619 	req.size = 32 + sgls_per_request * 16;
620 	req.virt = dma_alloc_coherent(&efct->pci->dev, req.size, &req.phys,
621 				      GFP_KERNEL);
622 	if (!req.virt) {
623 		kfree(sgls);
624 		return -ENOMEM;
625 	}
626 
627 	for (n_rem = hw->config.n_io; n_rem; n_rem -= n) {
628 		/* Copy address of SGL's into local sgls[] array, break
629 		 * out if the xri is not contiguous.
630 		 */
631 		u32 min = (sgls_per_request < n_rem) ? sgls_per_request : n_rem;
632 
633 		for (n = 0; n < min; n++) {
634 			/* Check that we have contiguous xri values */
635 			if (n > 0) {
636 				if (hw->io[idx + n]->indicator !=
637 				    hw->io[idx + n - 1]->indicator + 1)
638 					break;
639 			}
640 
641 			sgls[n] = hw->io[idx + n]->sgl;
642 		}
643 
644 		if (sli_cmd_post_sgl_pages(&hw->sli, cmd,
645 				hw->io[idx]->indicator,	n, sgls, NULL, &req)) {
646 			rc = -EIO;
647 			break;
648 		}
649 
650 		rc = efct_hw_command(hw, cmd, EFCT_CMD_POLL, NULL, NULL);
651 		if (rc) {
652 			efc_log_err(hw->os, "SGL post failed, rc=%d\n", rc);
653 			break;
654 		}
655 
656 		/* Add to tail if successful */
657 		for (i = 0; i < n; i++, idx++) {
658 			io = hw->io[idx];
659 			io->state = EFCT_HW_IO_STATE_FREE;
660 			INIT_LIST_HEAD(&io->list_entry);
661 			list_add_tail(&io->list_entry, &hw->io_free);
662 		}
663 	}
664 
665 	dma_free_coherent(&efct->pci->dev, req.size, req.virt, req.phys);
666 	memset(&req, 0, sizeof(struct efc_dma));
667 	kfree(sgls);
668 
669 	return rc;
670 }
671 
672 static int
673 efct_hw_init_io(struct efct_hw *hw)
674 {
675 	u32 i, idx = 0;
676 	bool prereg = false;
677 	struct efct_hw_io *io = NULL;
678 	int rc = 0;
679 
680 	prereg = hw->sli.params.sgl_pre_registered;
681 
682 	if (prereg)
683 		return efct_hw_init_prereg_io(hw);
684 
685 	for (i = 0; i < hw->config.n_io; i++, idx++) {
686 		io = hw->io[idx];
687 		io->state = EFCT_HW_IO_STATE_FREE;
688 		INIT_LIST_HEAD(&io->list_entry);
689 		list_add_tail(&io->list_entry, &hw->io_free);
690 	}
691 
692 	return rc;
693 }
694 
695 static int
696 efct_hw_config_set_fdt_xfer_hint(struct efct_hw *hw, u32 fdt_xfer_hint)
697 {
698 	int rc = 0;
699 	u8 buf[SLI4_BMBX_SIZE];
700 	struct sli4_rqst_cmn_set_features_set_fdt_xfer_hint param;
701 
702 	memset(&param, 0, sizeof(param));
703 	param.fdt_xfer_hint = cpu_to_le32(fdt_xfer_hint);
704 	/* build the set_features command */
705 	sli_cmd_common_set_features(&hw->sli, buf,
706 		SLI4_SET_FEATURES_SET_FTD_XFER_HINT, sizeof(param), &param);
707 
708 	rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
709 	if (rc)
710 		efc_log_warn(hw->os, "set FDT hint %d failed: %d\n",
711 			     fdt_xfer_hint, rc);
712 	else
713 		efc_log_info(hw->os, "Set FTD transfer hint to %d\n",
714 			     le32_to_cpu(param.fdt_xfer_hint));
715 
716 	return rc;
717 }
718 
719 static int
720 efct_hw_config_rq(struct efct_hw *hw)
721 {
722 	u32 min_rq_count, i, rc;
723 	struct sli4_cmd_rq_cfg rq_cfg[SLI4_CMD_REG_FCFI_NUM_RQ_CFG];
724 	u8 buf[SLI4_BMBX_SIZE];
725 
726 	efc_log_info(hw->os, "using REG_FCFI standard\n");
727 
728 	/*
729 	 * Set the filter match/mask values from hw's
730 	 * filter_def values
731 	 */
732 	for (i = 0; i < SLI4_CMD_REG_FCFI_NUM_RQ_CFG; i++) {
733 		rq_cfg[i].rq_id = cpu_to_le16(0xffff);
734 		rq_cfg[i].r_ctl_mask = (u8)hw->config.filter_def[i];
735 		rq_cfg[i].r_ctl_match = (u8)(hw->config.filter_def[i] >> 8);
736 		rq_cfg[i].type_mask = (u8)(hw->config.filter_def[i] >> 16);
737 		rq_cfg[i].type_match = (u8)(hw->config.filter_def[i] >> 24);
738 	}
739 
740 	/*
741 	 * Update the rq_id's of the FCF configuration
742 	 * (don't update more than the number of rq_cfg
743 	 * elements)
744 	 */
745 	min_rq_count = (hw->hw_rq_count < SLI4_CMD_REG_FCFI_NUM_RQ_CFG)	?
746 			hw->hw_rq_count : SLI4_CMD_REG_FCFI_NUM_RQ_CFG;
747 	for (i = 0; i < min_rq_count; i++) {
748 		struct hw_rq *rq = hw->hw_rq[i];
749 		u32 j;
750 
751 		for (j = 0; j < SLI4_CMD_REG_FCFI_NUM_RQ_CFG; j++) {
752 			u32 mask = (rq->filter_mask != 0) ?
753 				rq->filter_mask : 1;
754 
755 			if (!(mask & (1U << j)))
756 				continue;
757 
758 			rq_cfg[i].rq_id = cpu_to_le16(rq->hdr->id);
759 			efct_logfcfi(hw, j, i, rq->hdr->id);
760 		}
761 	}
762 
763 	rc = -EIO;
764 	if (!sli_cmd_reg_fcfi(&hw->sli, buf, 0,	rq_cfg))
765 		rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
766 
767 	if (rc != 0) {
768 		efc_log_err(hw->os, "FCFI registration failed\n");
769 		return rc;
770 	}
771 	hw->fcf_indicator =
772 		le16_to_cpu(((struct sli4_cmd_reg_fcfi *)buf)->fcfi);
773 
774 	return rc;
775 }
776 
777 static int
778 efct_hw_config_mrq(struct efct_hw *hw, u8 mode, u16 fcf_index)
779 {
780 	u8 buf[SLI4_BMBX_SIZE], mrq_bitmask = 0;
781 	struct hw_rq *rq;
782 	struct sli4_cmd_reg_fcfi_mrq *rsp = NULL;
783 	struct sli4_cmd_rq_cfg rq_filter[SLI4_CMD_REG_FCFI_MRQ_NUM_RQ_CFG];
784 	u32 rc, i;
785 
786 	if (mode == SLI4_CMD_REG_FCFI_SET_FCFI_MODE)
787 		goto issue_cmd;
788 
789 	/* Set the filter match/mask values from hw's filter_def values */
790 	for (i = 0; i < SLI4_CMD_REG_FCFI_NUM_RQ_CFG; i++) {
791 		rq_filter[i].rq_id = cpu_to_le16(0xffff);
792 		rq_filter[i].type_mask = (u8)hw->config.filter_def[i];
793 		rq_filter[i].type_match = (u8)(hw->config.filter_def[i] >> 8);
794 		rq_filter[i].r_ctl_mask = (u8)(hw->config.filter_def[i] >> 16);
795 		rq_filter[i].r_ctl_match = (u8)(hw->config.filter_def[i] >> 24);
796 	}
797 
798 	rq = hw->hw_rq[0];
799 	rq_filter[0].rq_id = cpu_to_le16(rq->hdr->id);
800 	rq_filter[1].rq_id = cpu_to_le16(rq->hdr->id);
801 
802 	mrq_bitmask = 0x2;
803 issue_cmd:
804 	efc_log_debug(hw->os, "Issue reg_fcfi_mrq count:%d policy:%d mode:%d\n",
805 		      hw->hw_rq_count, hw->config.rq_selection_policy, mode);
806 	/* Invoke REG_FCFI_MRQ */
807 	rc = sli_cmd_reg_fcfi_mrq(&hw->sli, buf, mode, fcf_index,
808 				  hw->config.rq_selection_policy, mrq_bitmask,
809 				  hw->hw_mrq_count, rq_filter);
810 	if (rc) {
811 		efc_log_err(hw->os, "sli_cmd_reg_fcfi_mrq() failed\n");
812 		return -EIO;
813 	}
814 
815 	rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
816 
817 	rsp = (struct sli4_cmd_reg_fcfi_mrq *)buf;
818 
819 	if ((rc) || (le16_to_cpu(rsp->hdr.status))) {
820 		efc_log_err(hw->os, "FCFI MRQ reg failed. cmd=%x status=%x\n",
821 			    rsp->hdr.command, le16_to_cpu(rsp->hdr.status));
822 		return -EIO;
823 	}
824 
825 	if (mode == SLI4_CMD_REG_FCFI_SET_FCFI_MODE)
826 		hw->fcf_indicator = le16_to_cpu(rsp->fcfi);
827 
828 	return 0;
829 }
830 
831 static void
832 efct_hw_queue_hash_add(struct efct_queue_hash *hash,
833 		       u16 id, u16 index)
834 {
835 	u32 hash_index = id & (EFCT_HW_Q_HASH_SIZE - 1);
836 
837 	/*
838 	 * Since the hash is always bigger than the number of queues, then we
839 	 * never have to worry about an infinite loop.
840 	 */
841 	while (hash[hash_index].in_use)
842 		hash_index = (hash_index + 1) & (EFCT_HW_Q_HASH_SIZE - 1);
843 
844 	/* not used, claim the entry */
845 	hash[hash_index].id = id;
846 	hash[hash_index].in_use = true;
847 	hash[hash_index].index = index;
848 }
849 
850 static int
851 efct_hw_config_sli_port_health_check(struct efct_hw *hw, u8 query, u8 enable)
852 {
853 	int rc = 0;
854 	u8 buf[SLI4_BMBX_SIZE];
855 	struct sli4_rqst_cmn_set_features_health_check param;
856 	u32 health_check_flag = 0;
857 
858 	memset(&param, 0, sizeof(param));
859 
860 	if (enable)
861 		health_check_flag |= SLI4_RQ_HEALTH_CHECK_ENABLE;
862 
863 	if (query)
864 		health_check_flag |= SLI4_RQ_HEALTH_CHECK_QUERY;
865 
866 	param.health_check_dword = cpu_to_le32(health_check_flag);
867 
868 	/* build the set_features command */
869 	sli_cmd_common_set_features(&hw->sli, buf,
870 		SLI4_SET_FEATURES_SLI_PORT_HEALTH_CHECK, sizeof(param), &param);
871 
872 	rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
873 	if (rc)
874 		efc_log_err(hw->os, "efct_hw_command returns %d\n", rc);
875 	else
876 		efc_log_debug(hw->os, "SLI Port Health Check is enabled\n");
877 
878 	return rc;
879 }
880 
881 int
882 efct_hw_init(struct efct_hw *hw)
883 {
884 	int rc;
885 	u32 i = 0;
886 	int rem_count;
887 	unsigned long flags = 0;
888 	struct efct_hw_io *temp;
889 	struct efc_dma *dma;
890 
891 	/*
892 	 * Make sure the command lists are empty. If this is start-of-day,
893 	 * they'll be empty since they were just initialized in efct_hw_setup.
894 	 * If we've just gone through a reset, the command and command pending
895 	 * lists should have been cleaned up as part of the reset
896 	 * (efct_hw_reset()).
897 	 */
898 	spin_lock_irqsave(&hw->cmd_lock, flags);
899 	if (!list_empty(&hw->cmd_head)) {
900 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
901 		efc_log_err(hw->os, "command found on cmd list\n");
902 		return -EIO;
903 	}
904 	if (!list_empty(&hw->cmd_pending)) {
905 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
906 		efc_log_err(hw->os, "command found on pending list\n");
907 		return -EIO;
908 	}
909 	spin_unlock_irqrestore(&hw->cmd_lock, flags);
910 
911 	/* Free RQ buffers if prevously allocated */
912 	efct_hw_rx_free(hw);
913 
914 	/*
915 	 * The IO queues must be initialized here for the reset case. The
916 	 * efct_hw_init_io() function will re-add the IOs to the free list.
917 	 * The cmd_head list should be OK since we free all entries in
918 	 * efct_hw_command_cancel() that is called in the efct_hw_reset().
919 	 */
920 
921 	/* If we are in this function due to a reset, there may be stale items
922 	 * on lists that need to be removed.  Clean them up.
923 	 */
924 	rem_count = 0;
925 	while ((!list_empty(&hw->io_wait_free))) {
926 		rem_count++;
927 		temp = list_first_entry(&hw->io_wait_free, struct efct_hw_io,
928 					list_entry);
929 		list_del_init(&temp->list_entry);
930 	}
931 	if (rem_count > 0)
932 		efc_log_debug(hw->os, "rmvd %d items from io_wait_free list\n",
933 			      rem_count);
934 
935 	rem_count = 0;
936 	while ((!list_empty(&hw->io_inuse))) {
937 		rem_count++;
938 		temp = list_first_entry(&hw->io_inuse, struct efct_hw_io,
939 					list_entry);
940 		list_del_init(&temp->list_entry);
941 	}
942 	if (rem_count > 0)
943 		efc_log_debug(hw->os, "rmvd %d items from io_inuse list\n",
944 			      rem_count);
945 
946 	rem_count = 0;
947 	while ((!list_empty(&hw->io_free))) {
948 		rem_count++;
949 		temp = list_first_entry(&hw->io_free, struct efct_hw_io,
950 					list_entry);
951 		list_del_init(&temp->list_entry);
952 	}
953 	if (rem_count > 0)
954 		efc_log_debug(hw->os, "rmvd %d items from io_free list\n",
955 			      rem_count);
956 
957 	/* If MRQ not required, Make sure we dont request feature. */
958 	if (hw->config.n_rq == 1)
959 		hw->sli.features &= (~SLI4_REQFEAT_MRQP);
960 
961 	if (sli_init(&hw->sli)) {
962 		efc_log_err(hw->os, "SLI failed to initialize\n");
963 		return -EIO;
964 	}
965 
966 	if (hw->sliport_healthcheck) {
967 		rc = efct_hw_config_sli_port_health_check(hw, 0, 1);
968 		if (rc != 0) {
969 			efc_log_err(hw->os, "Enable port Health check fail\n");
970 			return rc;
971 		}
972 	}
973 
974 	/*
975 	 * Set FDT transfer hint, only works on Lancer
976 	 */
977 	if (hw->sli.if_type == SLI4_INTF_IF_TYPE_2) {
978 		/*
979 		 * Non-fatal error. In particular, we can disregard failure to
980 		 * set EFCT_HW_FDT_XFER_HINT on devices with legacy firmware
981 		 * that do not support EFCT_HW_FDT_XFER_HINT feature.
982 		 */
983 		efct_hw_config_set_fdt_xfer_hint(hw, EFCT_HW_FDT_XFER_HINT);
984 	}
985 
986 	/* zero the hashes */
987 	memset(hw->cq_hash, 0, sizeof(hw->cq_hash));
988 	efc_log_debug(hw->os, "Max CQs %d, hash size = %d\n",
989 		      EFCT_HW_MAX_NUM_CQ, EFCT_HW_Q_HASH_SIZE);
990 
991 	memset(hw->rq_hash, 0, sizeof(hw->rq_hash));
992 	efc_log_debug(hw->os, "Max RQs %d, hash size = %d\n",
993 		      EFCT_HW_MAX_NUM_RQ, EFCT_HW_Q_HASH_SIZE);
994 
995 	memset(hw->wq_hash, 0, sizeof(hw->wq_hash));
996 	efc_log_debug(hw->os, "Max WQs %d, hash size = %d\n",
997 		      EFCT_HW_MAX_NUM_WQ, EFCT_HW_Q_HASH_SIZE);
998 
999 	rc = efct_hw_init_queues(hw);
1000 	if (rc)
1001 		return rc;
1002 
1003 	rc = efct_hw_map_wq_cpu(hw);
1004 	if (rc)
1005 		return rc;
1006 
1007 	/* Allocate and p_st RQ buffers */
1008 	rc = efct_hw_rx_allocate(hw);
1009 	if (rc) {
1010 		efc_log_err(hw->os, "rx_allocate failed\n");
1011 		return rc;
1012 	}
1013 
1014 	rc = efct_hw_rx_post(hw);
1015 	if (rc) {
1016 		efc_log_err(hw->os, "WARNING - error posting RQ buffers\n");
1017 		return rc;
1018 	}
1019 
1020 	if (hw->config.n_eq == 1) {
1021 		rc = efct_hw_config_rq(hw);
1022 		if (rc) {
1023 			efc_log_err(hw->os, "config rq failed %d\n", rc);
1024 			return rc;
1025 		}
1026 	} else {
1027 		rc = efct_hw_config_mrq(hw, SLI4_CMD_REG_FCFI_SET_FCFI_MODE, 0);
1028 		if (rc != 0) {
1029 			efc_log_err(hw->os, "REG_FCFI_MRQ FCFI reg failed\n");
1030 			return rc;
1031 		}
1032 
1033 		rc = efct_hw_config_mrq(hw, SLI4_CMD_REG_FCFI_SET_MRQ_MODE, 0);
1034 		if (rc != 0) {
1035 			efc_log_err(hw->os, "REG_FCFI_MRQ MRQ reg failed\n");
1036 			return rc;
1037 		}
1038 	}
1039 
1040 	/*
1041 	 * Allocate the WQ request tag pool, if not previously allocated
1042 	 * (the request tag value is 16 bits, thus the pool allocation size
1043 	 * of 64k)
1044 	 */
1045 	hw->wq_reqtag_pool = efct_hw_reqtag_pool_alloc(hw);
1046 	if (!hw->wq_reqtag_pool) {
1047 		efc_log_err(hw->os, "efct_hw_reqtag_pool_alloc failed\n");
1048 		return -ENOMEM;
1049 	}
1050 
1051 	rc = efct_hw_setup_io(hw);
1052 	if (rc) {
1053 		efc_log_err(hw->os, "IO allocation failure\n");
1054 		return rc;
1055 	}
1056 
1057 	rc = efct_hw_init_io(hw);
1058 	if (rc) {
1059 		efc_log_err(hw->os, "IO initialization failure\n");
1060 		return rc;
1061 	}
1062 
1063 	dma = &hw->loop_map;
1064 	dma->size = SLI4_MIN_LOOP_MAP_BYTES;
1065 	dma->virt = dma_alloc_coherent(&hw->os->pci->dev, dma->size, &dma->phys,
1066 				       GFP_KERNEL);
1067 	if (!dma->virt)
1068 		return -EIO;
1069 
1070 	/*
1071 	 * Arming the EQ allows (e.g.) interrupts when CQ completions write EQ
1072 	 * entries
1073 	 */
1074 	for (i = 0; i < hw->eq_count; i++)
1075 		sli_queue_arm(&hw->sli, &hw->eq[i], true);
1076 
1077 	/*
1078 	 * Initialize RQ hash
1079 	 */
1080 	for (i = 0; i < hw->rq_count; i++)
1081 		efct_hw_queue_hash_add(hw->rq_hash, hw->rq[i].id, i);
1082 
1083 	/*
1084 	 * Initialize WQ hash
1085 	 */
1086 	for (i = 0; i < hw->wq_count; i++)
1087 		efct_hw_queue_hash_add(hw->wq_hash, hw->wq[i].id, i);
1088 
1089 	/*
1090 	 * Arming the CQ allows (e.g.) MQ completions to write CQ entries
1091 	 */
1092 	for (i = 0; i < hw->cq_count; i++) {
1093 		efct_hw_queue_hash_add(hw->cq_hash, hw->cq[i].id, i);
1094 		sli_queue_arm(&hw->sli, &hw->cq[i], true);
1095 	}
1096 
1097 	/* Set RQ process limit*/
1098 	for (i = 0; i < hw->hw_rq_count; i++) {
1099 		struct hw_rq *rq = hw->hw_rq[i];
1100 
1101 		hw->cq[rq->cq->instance].proc_limit = hw->config.n_io / 2;
1102 	}
1103 
1104 	/* record the fact that the queues are functional */
1105 	hw->state = EFCT_HW_STATE_ACTIVE;
1106 	/*
1107 	 * Allocate a HW IOs for send frame.
1108 	 */
1109 	hw->hw_wq[0]->send_frame_io = efct_hw_io_alloc(hw);
1110 	if (!hw->hw_wq[0]->send_frame_io)
1111 		efc_log_err(hw->os, "alloc for send_frame_io failed\n");
1112 
1113 	/* Initialize send frame sequence id */
1114 	atomic_set(&hw->send_frame_seq_id, 0);
1115 
1116 	return 0;
1117 }
1118 
1119 int
1120 efct_hw_parse_filter(struct efct_hw *hw, void *value)
1121 {
1122 	int rc = 0;
1123 	char *p = NULL;
1124 	char *token;
1125 	u32 idx = 0;
1126 
1127 	for (idx = 0; idx < ARRAY_SIZE(hw->config.filter_def); idx++)
1128 		hw->config.filter_def[idx] = 0;
1129 
1130 	p = kstrdup(value, GFP_KERNEL);
1131 	if (!p || !*p) {
1132 		efc_log_err(hw->os, "p is NULL\n");
1133 		return -ENOMEM;
1134 	}
1135 
1136 	idx = 0;
1137 	while ((token = strsep(&p, ",")) && *token) {
1138 		if (kstrtou32(token, 0, &hw->config.filter_def[idx++]))
1139 			efc_log_err(hw->os, "kstrtoint failed\n");
1140 
1141 		if (!p || !*p)
1142 			break;
1143 
1144 		if (idx == ARRAY_SIZE(hw->config.filter_def))
1145 			break;
1146 	}
1147 	kfree(p);
1148 
1149 	return rc;
1150 }
1151 
1152 u64
1153 efct_get_wwnn(struct efct_hw *hw)
1154 {
1155 	struct sli4 *sli = &hw->sli;
1156 	u8 p[8];
1157 
1158 	memcpy(p, sli->wwnn, sizeof(p));
1159 	return get_unaligned_be64(p);
1160 }
1161 
1162 u64
1163 efct_get_wwpn(struct efct_hw *hw)
1164 {
1165 	struct sli4 *sli = &hw->sli;
1166 	u8 p[8];
1167 
1168 	memcpy(p, sli->wwpn, sizeof(p));
1169 	return get_unaligned_be64(p);
1170 }
1171 
1172 static struct efc_hw_rq_buffer *
1173 efct_hw_rx_buffer_alloc(struct efct_hw *hw, u32 rqindex, u32 count,
1174 			u32 size)
1175 {
1176 	struct efct *efct = hw->os;
1177 	struct efc_hw_rq_buffer *rq_buf = NULL;
1178 	struct efc_hw_rq_buffer *prq;
1179 	u32 i;
1180 
1181 	if (!count)
1182 		return NULL;
1183 
1184 	rq_buf = kmalloc_array(count, sizeof(*rq_buf), GFP_KERNEL);
1185 	if (!rq_buf)
1186 		return NULL;
1187 	memset(rq_buf, 0, sizeof(*rq_buf) * count);
1188 
1189 	for (i = 0, prq = rq_buf; i < count; i ++, prq++) {
1190 		prq->rqindex = rqindex;
1191 		prq->dma.size = size;
1192 		prq->dma.virt = dma_alloc_coherent(&efct->pci->dev,
1193 						   prq->dma.size,
1194 						   &prq->dma.phys,
1195 						   GFP_KERNEL);
1196 		if (!prq->dma.virt) {
1197 			efc_log_err(hw->os, "DMA allocation failed\n");
1198 			kfree(rq_buf);
1199 			return NULL;
1200 		}
1201 	}
1202 	return rq_buf;
1203 }
1204 
1205 static void
1206 efct_hw_rx_buffer_free(struct efct_hw *hw,
1207 		       struct efc_hw_rq_buffer *rq_buf,
1208 			u32 count)
1209 {
1210 	struct efct *efct = hw->os;
1211 	u32 i;
1212 	struct efc_hw_rq_buffer *prq;
1213 
1214 	if (rq_buf) {
1215 		for (i = 0, prq = rq_buf; i < count; i++, prq++) {
1216 			dma_free_coherent(&efct->pci->dev,
1217 					  prq->dma.size, prq->dma.virt,
1218 					  prq->dma.phys);
1219 			memset(&prq->dma, 0, sizeof(struct efc_dma));
1220 		}
1221 
1222 		kfree(rq_buf);
1223 	}
1224 }
1225 
1226 int
1227 efct_hw_rx_allocate(struct efct_hw *hw)
1228 {
1229 	struct efct *efct = hw->os;
1230 	u32 i;
1231 	int rc = 0;
1232 	u32 rqindex = 0;
1233 	u32 hdr_size = EFCT_HW_RQ_SIZE_HDR;
1234 	u32 payload_size = hw->config.rq_default_buffer_size;
1235 
1236 	rqindex = 0;
1237 
1238 	for (i = 0; i < hw->hw_rq_count; i++) {
1239 		struct hw_rq *rq = hw->hw_rq[i];
1240 
1241 		/* Allocate header buffers */
1242 		rq->hdr_buf = efct_hw_rx_buffer_alloc(hw, rqindex,
1243 						      rq->entry_count,
1244 						      hdr_size);
1245 		if (!rq->hdr_buf) {
1246 			efc_log_err(efct, "rx_buffer_alloc hdr_buf failed\n");
1247 			rc = -EIO;
1248 			break;
1249 		}
1250 
1251 		efc_log_debug(hw->os,
1252 			      "rq[%2d] rq_id %02d header  %4d by %4d bytes\n",
1253 			      i, rq->hdr->id, rq->entry_count, hdr_size);
1254 
1255 		rqindex++;
1256 
1257 		/* Allocate payload buffers */
1258 		rq->payload_buf = efct_hw_rx_buffer_alloc(hw, rqindex,
1259 							  rq->entry_count,
1260 							  payload_size);
1261 		if (!rq->payload_buf) {
1262 			efc_log_err(efct, "rx_buffer_alloc fb_buf failed\n");
1263 			rc = -EIO;
1264 			break;
1265 		}
1266 		efc_log_debug(hw->os,
1267 			      "rq[%2d] rq_id %02d default %4d by %4d bytes\n",
1268 			      i, rq->data->id, rq->entry_count, payload_size);
1269 		rqindex++;
1270 	}
1271 
1272 	return rc ? -EIO : 0;
1273 }
1274 
1275 int
1276 efct_hw_rx_post(struct efct_hw *hw)
1277 {
1278 	u32 i;
1279 	u32 idx;
1280 	u32 rq_idx;
1281 	int rc = 0;
1282 
1283 	if (!hw->seq_pool) {
1284 		u32 count = 0;
1285 
1286 		for (i = 0; i < hw->hw_rq_count; i++)
1287 			count += hw->hw_rq[i]->entry_count;
1288 
1289 		hw->seq_pool = kmalloc_array(count,
1290 				sizeof(struct efc_hw_sequence),	GFP_KERNEL);
1291 		if (!hw->seq_pool)
1292 			return -ENOMEM;
1293 	}
1294 
1295 	/*
1296 	 * In RQ pair mode, we MUST post the header and payload buffer at the
1297 	 * same time.
1298 	 */
1299 	for (rq_idx = 0, idx = 0; rq_idx < hw->hw_rq_count; rq_idx++) {
1300 		struct hw_rq *rq = hw->hw_rq[rq_idx];
1301 
1302 		for (i = 0; i < rq->entry_count - 1; i++) {
1303 			struct efc_hw_sequence *seq;
1304 
1305 			seq = hw->seq_pool + idx;
1306 			idx++;
1307 			seq->header = &rq->hdr_buf[i];
1308 			seq->payload = &rq->payload_buf[i];
1309 			rc = efct_hw_sequence_free(hw, seq);
1310 			if (rc)
1311 				break;
1312 		}
1313 		if (rc)
1314 			break;
1315 	}
1316 
1317 	if (rc && hw->seq_pool)
1318 		kfree(hw->seq_pool);
1319 
1320 	return rc;
1321 }
1322 
1323 void
1324 efct_hw_rx_free(struct efct_hw *hw)
1325 {
1326 	u32 i;
1327 
1328 	/* Free hw_rq buffers */
1329 	for (i = 0; i < hw->hw_rq_count; i++) {
1330 		struct hw_rq *rq = hw->hw_rq[i];
1331 
1332 		if (rq) {
1333 			efct_hw_rx_buffer_free(hw, rq->hdr_buf,
1334 					       rq->entry_count);
1335 			rq->hdr_buf = NULL;
1336 			efct_hw_rx_buffer_free(hw, rq->payload_buf,
1337 					       rq->entry_count);
1338 			rq->payload_buf = NULL;
1339 		}
1340 	}
1341 }
1342 
1343 static int
1344 efct_hw_cmd_submit_pending(struct efct_hw *hw)
1345 {
1346 	int rc = 0;
1347 
1348 	/* Assumes lock held */
1349 
1350 	/* Only submit MQE if there's room */
1351 	while (hw->cmd_head_count < (EFCT_HW_MQ_DEPTH - 1) &&
1352 	       !list_empty(&hw->cmd_pending)) {
1353 		struct efct_command_ctx *ctx;
1354 
1355 		ctx = list_first_entry(&hw->cmd_pending,
1356 				       struct efct_command_ctx, list_entry);
1357 		if (!ctx)
1358 			break;
1359 
1360 		list_del_init(&ctx->list_entry);
1361 
1362 		list_add_tail(&ctx->list_entry, &hw->cmd_head);
1363 		hw->cmd_head_count++;
1364 		if (sli_mq_write(&hw->sli, hw->mq, ctx->buf) < 0) {
1365 			efc_log_debug(hw->os,
1366 				      "sli_queue_write failed: %d\n", rc);
1367 			rc = -EIO;
1368 			break;
1369 		}
1370 	}
1371 	return rc;
1372 }
1373 
1374 int
1375 efct_hw_command(struct efct_hw *hw, u8 *cmd, u32 opts, void *cb, void *arg)
1376 {
1377 	int rc = -EIO;
1378 	unsigned long flags = 0;
1379 	void *bmbx = NULL;
1380 
1381 	/*
1382 	 * If the chip is in an error state (UE'd) then reject this mailbox
1383 	 * command.
1384 	 */
1385 	if (sli_fw_error_status(&hw->sli) > 0) {
1386 		efc_log_crit(hw->os, "Chip in an error state - reset needed\n");
1387 		efc_log_crit(hw->os, "status=%#x error1=%#x error2=%#x\n",
1388 			     sli_reg_read_status(&hw->sli),
1389 			     sli_reg_read_err1(&hw->sli),
1390 			     sli_reg_read_err2(&hw->sli));
1391 
1392 		return -EIO;
1393 	}
1394 
1395 	/*
1396 	 * Send a mailbox command to the hardware, and either wait for
1397 	 * a completion (EFCT_CMD_POLL) or get an optional asynchronous
1398 	 * completion (EFCT_CMD_NOWAIT).
1399 	 */
1400 
1401 	if (opts == EFCT_CMD_POLL) {
1402 		mutex_lock(&hw->bmbx_lock);
1403 		bmbx = hw->sli.bmbx.virt;
1404 
1405 		memset(bmbx, 0, SLI4_BMBX_SIZE);
1406 		memcpy(bmbx, cmd, SLI4_BMBX_SIZE);
1407 
1408 		if (sli_bmbx_command(&hw->sli) == 0) {
1409 			rc = 0;
1410 			memcpy(cmd, bmbx, SLI4_BMBX_SIZE);
1411 		}
1412 		mutex_unlock(&hw->bmbx_lock);
1413 	} else if (opts == EFCT_CMD_NOWAIT) {
1414 		struct efct_command_ctx	*ctx = NULL;
1415 
1416 		if (hw->state != EFCT_HW_STATE_ACTIVE) {
1417 			efc_log_err(hw->os, "Can't send command, HW state=%d\n",
1418 				    hw->state);
1419 			return -EIO;
1420 		}
1421 
1422 		ctx = mempool_alloc(hw->cmd_ctx_pool, GFP_ATOMIC);
1423 		if (!ctx)
1424 			return -ENOSPC;
1425 
1426 		memset(ctx, 0, sizeof(struct efct_command_ctx));
1427 
1428 		if (cb) {
1429 			ctx->cb = cb;
1430 			ctx->arg = arg;
1431 		}
1432 
1433 		memcpy(ctx->buf, cmd, SLI4_BMBX_SIZE);
1434 		ctx->ctx = hw;
1435 
1436 		spin_lock_irqsave(&hw->cmd_lock, flags);
1437 
1438 		/* Add to pending list */
1439 		INIT_LIST_HEAD(&ctx->list_entry);
1440 		list_add_tail(&ctx->list_entry, &hw->cmd_pending);
1441 
1442 		/* Submit as much of the pending list as we can */
1443 		rc = efct_hw_cmd_submit_pending(hw);
1444 
1445 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
1446 	}
1447 
1448 	return rc;
1449 }
1450 
1451 static int
1452 efct_hw_command_process(struct efct_hw *hw, int status, u8 *mqe,
1453 			size_t size)
1454 {
1455 	struct efct_command_ctx *ctx = NULL;
1456 	unsigned long flags = 0;
1457 
1458 	spin_lock_irqsave(&hw->cmd_lock, flags);
1459 	if (!list_empty(&hw->cmd_head)) {
1460 		ctx = list_first_entry(&hw->cmd_head,
1461 				       struct efct_command_ctx, list_entry);
1462 		list_del_init(&ctx->list_entry);
1463 	}
1464 	if (!ctx) {
1465 		efc_log_err(hw->os, "no command context\n");
1466 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
1467 		return -EIO;
1468 	}
1469 
1470 	hw->cmd_head_count--;
1471 
1472 	/* Post any pending requests */
1473 	efct_hw_cmd_submit_pending(hw);
1474 
1475 	spin_unlock_irqrestore(&hw->cmd_lock, flags);
1476 
1477 	if (ctx->cb) {
1478 		memcpy(ctx->buf, mqe, size);
1479 		ctx->cb(hw, status, ctx->buf, ctx->arg);
1480 	}
1481 
1482 	mempool_free(ctx, hw->cmd_ctx_pool);
1483 
1484 	return 0;
1485 }
1486 
1487 static int
1488 efct_hw_mq_process(struct efct_hw *hw,
1489 		   int status, struct sli4_queue *mq)
1490 {
1491 	u8 mqe[SLI4_BMBX_SIZE];
1492 	int rc;
1493 
1494 	rc = sli_mq_read(&hw->sli, mq, mqe);
1495 	if (!rc)
1496 		rc = efct_hw_command_process(hw, status, mqe, mq->size);
1497 
1498 	return rc;
1499 }
1500 
1501 static int
1502 efct_hw_command_cancel(struct efct_hw *hw)
1503 {
1504 	unsigned long flags = 0;
1505 	int rc = 0;
1506 
1507 	spin_lock_irqsave(&hw->cmd_lock, flags);
1508 
1509 	/*
1510 	 * Manually clean up remaining commands. Note: since this calls
1511 	 * efct_hw_command_process(), we'll also process the cmd_pending
1512 	 * list, so no need to manually clean that out.
1513 	 */
1514 	while (!list_empty(&hw->cmd_head)) {
1515 		u8		mqe[SLI4_BMBX_SIZE] = { 0 };
1516 		struct efct_command_ctx *ctx;
1517 
1518 		ctx = list_first_entry(&hw->cmd_head,
1519 				       struct efct_command_ctx, list_entry);
1520 
1521 		efc_log_debug(hw->os, "hung command %08x\n",
1522 			      !ctx ? U32_MAX : *((u32 *)ctx->buf));
1523 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
1524 		rc = efct_hw_command_process(hw, -1, mqe, SLI4_BMBX_SIZE);
1525 		spin_lock_irqsave(&hw->cmd_lock, flags);
1526 	}
1527 
1528 	spin_unlock_irqrestore(&hw->cmd_lock, flags);
1529 
1530 	return rc;
1531 }
1532 
1533 static void
1534 efct_mbox_rsp_cb(struct efct_hw *hw, int status, u8 *mqe, void *arg)
1535 {
1536 	struct efct_mbox_rqst_ctx *ctx = arg;
1537 
1538 	if (ctx) {
1539 		if (ctx->callback)
1540 			(*ctx->callback)(hw->os->efcport, status, mqe,
1541 					 ctx->arg);
1542 
1543 		mempool_free(ctx, hw->mbox_rqst_pool);
1544 	}
1545 }
1546 
1547 int
1548 efct_issue_mbox_rqst(void *base, void *cmd, void *cb, void *arg)
1549 {
1550 	struct efct_mbox_rqst_ctx *ctx;
1551 	struct efct *efct = base;
1552 	struct efct_hw *hw = &efct->hw;
1553 	int rc;
1554 
1555 	/*
1556 	 * Allocate a callback context (which includes the mbox cmd buffer),
1557 	 * we need this to be persistent as the mbox cmd submission may be
1558 	 * queued and executed later execution.
1559 	 */
1560 	ctx = mempool_alloc(hw->mbox_rqst_pool, GFP_ATOMIC);
1561 	if (!ctx)
1562 		return -EIO;
1563 
1564 	ctx->callback = cb;
1565 	ctx->arg = arg;
1566 
1567 	rc = efct_hw_command(hw, cmd, EFCT_CMD_NOWAIT, efct_mbox_rsp_cb, ctx);
1568 	if (rc) {
1569 		efc_log_err(efct, "issue mbox rqst failure rc:%d\n", rc);
1570 		mempool_free(ctx, hw->mbox_rqst_pool);
1571 		return -EIO;
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 static inline struct efct_hw_io *
1578 _efct_hw_io_alloc(struct efct_hw *hw)
1579 {
1580 	struct efct_hw_io *io = NULL;
1581 
1582 	if (!list_empty(&hw->io_free)) {
1583 		io = list_first_entry(&hw->io_free, struct efct_hw_io,
1584 				      list_entry);
1585 		list_del(&io->list_entry);
1586 	}
1587 	if (io) {
1588 		INIT_LIST_HEAD(&io->list_entry);
1589 		list_add_tail(&io->list_entry, &hw->io_inuse);
1590 		io->state = EFCT_HW_IO_STATE_INUSE;
1591 		io->abort_reqtag = U32_MAX;
1592 		io->wq = hw->wq_cpu_array[raw_smp_processor_id()];
1593 		if (!io->wq) {
1594 			efc_log_err(hw->os, "WQ not assigned for cpu:%d\n",
1595 				    raw_smp_processor_id());
1596 			io->wq = hw->hw_wq[0];
1597 		}
1598 		kref_init(&io->ref);
1599 		io->release = efct_hw_io_free_internal;
1600 	} else {
1601 		atomic_add(1, &hw->io_alloc_failed_count);
1602 	}
1603 
1604 	return io;
1605 }
1606 
1607 struct efct_hw_io *
1608 efct_hw_io_alloc(struct efct_hw *hw)
1609 {
1610 	struct efct_hw_io *io = NULL;
1611 	unsigned long flags = 0;
1612 
1613 	spin_lock_irqsave(&hw->io_lock, flags);
1614 	io = _efct_hw_io_alloc(hw);
1615 	spin_unlock_irqrestore(&hw->io_lock, flags);
1616 
1617 	return io;
1618 }
1619 
1620 static void
1621 efct_hw_io_free_move_correct_list(struct efct_hw *hw,
1622 				  struct efct_hw_io *io)
1623 {
1624 	/*
1625 	 * When an IO is freed, depending on the exchange busy flag,
1626 	 * move it to the correct list.
1627 	 */
1628 	if (io->xbusy) {
1629 		/*
1630 		 * add to wait_free list and wait for XRI_ABORTED CQEs to clean
1631 		 * up
1632 		 */
1633 		INIT_LIST_HEAD(&io->list_entry);
1634 		list_add_tail(&io->list_entry, &hw->io_wait_free);
1635 		io->state = EFCT_HW_IO_STATE_WAIT_FREE;
1636 	} else {
1637 		/* IO not busy, add to free list */
1638 		INIT_LIST_HEAD(&io->list_entry);
1639 		list_add_tail(&io->list_entry, &hw->io_free);
1640 		io->state = EFCT_HW_IO_STATE_FREE;
1641 	}
1642 }
1643 
1644 static inline void
1645 efct_hw_io_free_common(struct efct_hw *hw, struct efct_hw_io *io)
1646 {
1647 	/* initialize IO fields */
1648 	efct_hw_init_free_io(io);
1649 
1650 	/* Restore default SGL */
1651 	efct_hw_io_restore_sgl(hw, io);
1652 }
1653 
1654 void
1655 efct_hw_io_free_internal(struct kref *arg)
1656 {
1657 	unsigned long flags = 0;
1658 	struct efct_hw_io *io =	container_of(arg, struct efct_hw_io, ref);
1659 	struct efct_hw *hw = io->hw;
1660 
1661 	/* perform common cleanup */
1662 	efct_hw_io_free_common(hw, io);
1663 
1664 	spin_lock_irqsave(&hw->io_lock, flags);
1665 	/* remove from in-use list */
1666 	if (!list_empty(&io->list_entry) && !list_empty(&hw->io_inuse)) {
1667 		list_del_init(&io->list_entry);
1668 		efct_hw_io_free_move_correct_list(hw, io);
1669 	}
1670 	spin_unlock_irqrestore(&hw->io_lock, flags);
1671 }
1672 
1673 int
1674 efct_hw_io_free(struct efct_hw *hw, struct efct_hw_io *io)
1675 {
1676 	return kref_put(&io->ref, io->release);
1677 }
1678 
1679 struct efct_hw_io *
1680 efct_hw_io_lookup(struct efct_hw *hw, u32 xri)
1681 {
1682 	u32 ioindex;
1683 
1684 	ioindex = xri - hw->sli.ext[SLI4_RSRC_XRI].base[0];
1685 	return hw->io[ioindex];
1686 }
1687 
1688 int
1689 efct_hw_io_init_sges(struct efct_hw *hw, struct efct_hw_io *io,
1690 		     enum efct_hw_io_type type)
1691 {
1692 	struct sli4_sge	*data = NULL;
1693 	u32 i = 0;
1694 	u32 skips = 0;
1695 	u32 sge_flags = 0;
1696 
1697 	if (!io) {
1698 		efc_log_err(hw->os, "bad parameter hw=%p io=%p\n", hw, io);
1699 		return -EIO;
1700 	}
1701 
1702 	/* Clear / reset the scatter-gather list */
1703 	io->sgl = &io->def_sgl;
1704 	io->sgl_count = io->def_sgl_count;
1705 	io->first_data_sge = 0;
1706 
1707 	memset(io->sgl->virt, 0, 2 * sizeof(struct sli4_sge));
1708 	io->n_sge = 0;
1709 	io->sge_offset = 0;
1710 
1711 	io->type = type;
1712 
1713 	data = io->sgl->virt;
1714 
1715 	/*
1716 	 * Some IO types have underlying hardware requirements on the order
1717 	 * of SGEs. Process all special entries here.
1718 	 */
1719 	switch (type) {
1720 	case EFCT_HW_IO_TARGET_WRITE:
1721 
1722 		/* populate host resident XFER_RDY buffer */
1723 		sge_flags = le32_to_cpu(data->dw2_flags);
1724 		sge_flags &= (~SLI4_SGE_TYPE_MASK);
1725 		sge_flags |= (SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT);
1726 		data->buffer_address_high =
1727 			cpu_to_le32(upper_32_bits(io->xfer_rdy.phys));
1728 		data->buffer_address_low  =
1729 			cpu_to_le32(lower_32_bits(io->xfer_rdy.phys));
1730 		data->buffer_length = cpu_to_le32(io->xfer_rdy.size);
1731 		data->dw2_flags = cpu_to_le32(sge_flags);
1732 		data++;
1733 
1734 		skips = EFCT_TARGET_WRITE_SKIPS;
1735 
1736 		io->n_sge = 1;
1737 		break;
1738 	case EFCT_HW_IO_TARGET_READ:
1739 		/*
1740 		 * For FCP_TSEND64, the first 2 entries are SKIP SGE's
1741 		 */
1742 		skips = EFCT_TARGET_READ_SKIPS;
1743 		break;
1744 	case EFCT_HW_IO_TARGET_RSP:
1745 		/*
1746 		 * No skips, etc. for FCP_TRSP64
1747 		 */
1748 		break;
1749 	default:
1750 		efc_log_err(hw->os, "unsupported IO type %#x\n", type);
1751 		return -EIO;
1752 	}
1753 
1754 	/*
1755 	 * Write skip entries
1756 	 */
1757 	for (i = 0; i < skips; i++) {
1758 		sge_flags = le32_to_cpu(data->dw2_flags);
1759 		sge_flags &= (~SLI4_SGE_TYPE_MASK);
1760 		sge_flags |= (SLI4_SGE_TYPE_SKIP << SLI4_SGE_TYPE_SHIFT);
1761 		data->dw2_flags = cpu_to_le32(sge_flags);
1762 		data++;
1763 	}
1764 
1765 	io->n_sge += skips;
1766 
1767 	/*
1768 	 * Set last
1769 	 */
1770 	sge_flags = le32_to_cpu(data->dw2_flags);
1771 	sge_flags |= SLI4_SGE_LAST;
1772 	data->dw2_flags = cpu_to_le32(sge_flags);
1773 
1774 	return 0;
1775 }
1776 
1777 int
1778 efct_hw_io_add_sge(struct efct_hw *hw, struct efct_hw_io *io,
1779 		   uintptr_t addr, u32 length)
1780 {
1781 	struct sli4_sge	*data = NULL;
1782 	u32 sge_flags = 0;
1783 
1784 	if (!io || !addr || !length) {
1785 		efc_log_err(hw->os,
1786 			    "bad parameter hw=%p io=%p addr=%lx length=%u\n",
1787 			    hw, io, addr, length);
1788 		return -EIO;
1789 	}
1790 
1791 	if (length > hw->sli.sge_supported_length) {
1792 		efc_log_err(hw->os,
1793 			    "length of SGE %d bigger than allowed %d\n",
1794 			    length, hw->sli.sge_supported_length);
1795 		return -EIO;
1796 	}
1797 
1798 	data = io->sgl->virt;
1799 	data += io->n_sge;
1800 
1801 	sge_flags = le32_to_cpu(data->dw2_flags);
1802 	sge_flags &= ~SLI4_SGE_TYPE_MASK;
1803 	sge_flags |= SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT;
1804 	sge_flags &= ~SLI4_SGE_DATA_OFFSET_MASK;
1805 	sge_flags |= SLI4_SGE_DATA_OFFSET_MASK & io->sge_offset;
1806 
1807 	data->buffer_address_high = cpu_to_le32(upper_32_bits(addr));
1808 	data->buffer_address_low  = cpu_to_le32(lower_32_bits(addr));
1809 	data->buffer_length = cpu_to_le32(length);
1810 
1811 	/*
1812 	 * Always assume this is the last entry and mark as such.
1813 	 * If this is not the first entry unset the "last SGE"
1814 	 * indication for the previous entry
1815 	 */
1816 	sge_flags |= SLI4_SGE_LAST;
1817 	data->dw2_flags = cpu_to_le32(sge_flags);
1818 
1819 	if (io->n_sge) {
1820 		sge_flags = le32_to_cpu(data[-1].dw2_flags);
1821 		sge_flags &= ~SLI4_SGE_LAST;
1822 		data[-1].dw2_flags = cpu_to_le32(sge_flags);
1823 	}
1824 
1825 	/* Set first_data_bde if not previously set */
1826 	if (io->first_data_sge == 0)
1827 		io->first_data_sge = io->n_sge;
1828 
1829 	io->sge_offset += length;
1830 	io->n_sge++;
1831 
1832 	return 0;
1833 }
1834 
1835 void
1836 efct_hw_io_abort_all(struct efct_hw *hw)
1837 {
1838 	struct efct_hw_io *io_to_abort	= NULL;
1839 	struct efct_hw_io *next_io = NULL;
1840 
1841 	list_for_each_entry_safe(io_to_abort, next_io,
1842 				 &hw->io_inuse, list_entry) {
1843 		efct_hw_io_abort(hw, io_to_abort, true, NULL, NULL);
1844 	}
1845 }
1846 
1847 static void
1848 efct_hw_wq_process_abort(void *arg, u8 *cqe, int status)
1849 {
1850 	struct efct_hw_io *io = arg;
1851 	struct efct_hw *hw = io->hw;
1852 	u32 ext = 0;
1853 	u32 len = 0;
1854 	struct hw_wq_callback *wqcb;
1855 
1856 	/*
1857 	 * For IOs that were aborted internally, we may need to issue the
1858 	 * callback here depending on whether a XRI_ABORTED CQE is expected ot
1859 	 * not. If the status is Local Reject/No XRI, then
1860 	 * issue the callback now.
1861 	 */
1862 	ext = sli_fc_ext_status(&hw->sli, cqe);
1863 	if (status == SLI4_FC_WCQE_STATUS_LOCAL_REJECT &&
1864 	    ext == SLI4_FC_LOCAL_REJECT_NO_XRI && io->done) {
1865 		efct_hw_done_t done = io->done;
1866 
1867 		io->done = NULL;
1868 
1869 		/*
1870 		 * Use latched status as this is always saved for an internal
1871 		 * abort Note: We won't have both a done and abort_done
1872 		 * function, so don't worry about
1873 		 *       clobbering the len, status and ext fields.
1874 		 */
1875 		status = io->saved_status;
1876 		len = io->saved_len;
1877 		ext = io->saved_ext;
1878 		io->status_saved = false;
1879 		done(io, len, status, ext, io->arg);
1880 	}
1881 
1882 	if (io->abort_done) {
1883 		efct_hw_done_t done = io->abort_done;
1884 
1885 		io->abort_done = NULL;
1886 		done(io, len, status, ext, io->abort_arg);
1887 	}
1888 
1889 	/* clear abort bit to indicate abort is complete */
1890 	io->abort_in_progress = false;
1891 
1892 	/* Free the WQ callback */
1893 	if (io->abort_reqtag == U32_MAX) {
1894 		efc_log_err(hw->os, "HW IO already freed\n");
1895 		return;
1896 	}
1897 
1898 	wqcb = efct_hw_reqtag_get_instance(hw, io->abort_reqtag);
1899 	efct_hw_reqtag_free(hw, wqcb);
1900 
1901 	/*
1902 	 * Call efct_hw_io_free() because this releases the WQ reservation as
1903 	 * well as doing the refcount put. Don't duplicate the code here.
1904 	 */
1905 	(void)efct_hw_io_free(hw, io);
1906 }
1907 
1908 static void
1909 efct_hw_fill_abort_wqe(struct efct_hw *hw, struct efct_hw_wqe *wqe)
1910 {
1911 	struct sli4_abort_wqe *abort = (void *)wqe->wqebuf;
1912 
1913 	memset(abort, 0, hw->sli.wqe_size);
1914 
1915 	abort->criteria = SLI4_ABORT_CRITERIA_XRI_TAG;
1916 	abort->ia_ir_byte |= wqe->send_abts ? 0 : 1;
1917 
1918 	/* Suppress ABTS retries */
1919 	abort->ia_ir_byte |= SLI4_ABRT_WQE_IR;
1920 
1921 	abort->t_tag  = cpu_to_le32(wqe->id);
1922 	abort->command = SLI4_WQE_ABORT;
1923 	abort->request_tag = cpu_to_le16(wqe->abort_reqtag);
1924 
1925 	abort->dw10w0_flags = cpu_to_le16(SLI4_ABRT_WQE_QOSD);
1926 
1927 	abort->cq_id = cpu_to_le16(SLI4_CQ_DEFAULT);
1928 }
1929 
1930 int
1931 efct_hw_io_abort(struct efct_hw *hw, struct efct_hw_io *io_to_abort,
1932 		 bool send_abts, void *cb, void *arg)
1933 {
1934 	struct hw_wq_callback *wqcb;
1935 	unsigned long flags = 0;
1936 
1937 	if (!io_to_abort) {
1938 		efc_log_err(hw->os, "bad parameter hw=%p io=%p\n",
1939 			    hw, io_to_abort);
1940 		return -EIO;
1941 	}
1942 
1943 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
1944 		efc_log_err(hw->os, "cannot send IO abort, HW state=%d\n",
1945 			    hw->state);
1946 		return -EIO;
1947 	}
1948 
1949 	/* take a reference on IO being aborted */
1950 	if (kref_get_unless_zero(&io_to_abort->ref) == 0) {
1951 		/* command no longer active */
1952 		efc_log_debug(hw->os,
1953 			      "io not active xri=0x%x tag=0x%x\n",
1954 			      io_to_abort->indicator, io_to_abort->reqtag);
1955 		return -ENOENT;
1956 	}
1957 
1958 	/* Must have a valid WQ reference */
1959 	if (!io_to_abort->wq) {
1960 		efc_log_debug(hw->os, "io_to_abort xri=0x%x not active on WQ\n",
1961 			      io_to_abort->indicator);
1962 		/* efct_ref_get(): same function */
1963 		kref_put(&io_to_abort->ref, io_to_abort->release);
1964 		return -ENOENT;
1965 	}
1966 
1967 	/*
1968 	 * Validation checks complete; now check to see if already being
1969 	 * aborted, if not set the flag.
1970 	 */
1971 	if (cmpxchg(&io_to_abort->abort_in_progress, false, true)) {
1972 		/* efct_ref_get(): same function */
1973 		kref_put(&io_to_abort->ref, io_to_abort->release);
1974 		efc_log_debug(hw->os,
1975 			      "io already being aborted xri=0x%x tag=0x%x\n",
1976 			      io_to_abort->indicator, io_to_abort->reqtag);
1977 		return -EINPROGRESS;
1978 	}
1979 
1980 	/*
1981 	 * If we got here, the possibilities are:
1982 	 * - host owned xri
1983 	 *	- io_to_abort->wq_index != U32_MAX
1984 	 *		- submit ABORT_WQE to same WQ
1985 	 * - port owned xri:
1986 	 *	- rxri: io_to_abort->wq_index == U32_MAX
1987 	 *		- submit ABORT_WQE to any WQ
1988 	 *	- non-rxri
1989 	 *		- io_to_abort->index != U32_MAX
1990 	 *			- submit ABORT_WQE to same WQ
1991 	 *		- io_to_abort->index == U32_MAX
1992 	 *			- submit ABORT_WQE to any WQ
1993 	 */
1994 	io_to_abort->abort_done = cb;
1995 	io_to_abort->abort_arg  = arg;
1996 
1997 	/* Allocate a request tag for the abort portion of this IO */
1998 	wqcb = efct_hw_reqtag_alloc(hw, efct_hw_wq_process_abort, io_to_abort);
1999 	if (!wqcb) {
2000 		efc_log_err(hw->os, "can't allocate request tag\n");
2001 		return -ENOSPC;
2002 	}
2003 
2004 	io_to_abort->abort_reqtag = wqcb->instance_index;
2005 	io_to_abort->wqe.send_abts = send_abts;
2006 	io_to_abort->wqe.id = io_to_abort->indicator;
2007 	io_to_abort->wqe.abort_reqtag = io_to_abort->abort_reqtag;
2008 
2009 	/*
2010 	 * If the wqe is on the pending list, then set this wqe to be
2011 	 * aborted when the IO's wqe is removed from the list.
2012 	 */
2013 	if (io_to_abort->wq) {
2014 		spin_lock_irqsave(&io_to_abort->wq->queue->lock, flags);
2015 		if (io_to_abort->wqe.list_entry.next) {
2016 			io_to_abort->wqe.abort_wqe_submit_needed = true;
2017 			spin_unlock_irqrestore(&io_to_abort->wq->queue->lock,
2018 					       flags);
2019 			return 0;
2020 		}
2021 		spin_unlock_irqrestore(&io_to_abort->wq->queue->lock, flags);
2022 	}
2023 
2024 	efct_hw_fill_abort_wqe(hw, &io_to_abort->wqe);
2025 
2026 	/* ABORT_WQE does not actually utilize an XRI on the Port,
2027 	 * therefore, keep xbusy as-is to track the exchange's state,
2028 	 * not the ABORT_WQE's state
2029 	 */
2030 	if (efct_hw_wq_write(io_to_abort->wq, &io_to_abort->wqe)) {
2031 		io_to_abort->abort_in_progress = false;
2032 		/* efct_ref_get(): same function */
2033 		kref_put(&io_to_abort->ref, io_to_abort->release);
2034 		return -EIO;
2035 	}
2036 
2037 	return 0;
2038 }
2039 
2040 void
2041 efct_hw_reqtag_pool_free(struct efct_hw *hw)
2042 {
2043 	u32 i;
2044 	struct reqtag_pool *reqtag_pool = hw->wq_reqtag_pool;
2045 	struct hw_wq_callback *wqcb = NULL;
2046 
2047 	if (reqtag_pool) {
2048 		for (i = 0; i < U16_MAX; i++) {
2049 			wqcb = reqtag_pool->tags[i];
2050 			if (!wqcb)
2051 				continue;
2052 
2053 			kfree(wqcb);
2054 		}
2055 		kfree(reqtag_pool);
2056 		hw->wq_reqtag_pool = NULL;
2057 	}
2058 }
2059 
2060 struct reqtag_pool *
2061 efct_hw_reqtag_pool_alloc(struct efct_hw *hw)
2062 {
2063 	u32 i = 0;
2064 	struct reqtag_pool *reqtag_pool;
2065 	struct hw_wq_callback *wqcb;
2066 
2067 	reqtag_pool = kzalloc(sizeof(*reqtag_pool), GFP_KERNEL);
2068 	if (!reqtag_pool)
2069 		return NULL;
2070 
2071 	INIT_LIST_HEAD(&reqtag_pool->freelist);
2072 	/* initialize reqtag pool lock */
2073 	spin_lock_init(&reqtag_pool->lock);
2074 	for (i = 0; i < U16_MAX; i++) {
2075 		wqcb = kmalloc(sizeof(*wqcb), GFP_KERNEL);
2076 		if (!wqcb)
2077 			break;
2078 
2079 		reqtag_pool->tags[i] = wqcb;
2080 		wqcb->instance_index = i;
2081 		wqcb->callback = NULL;
2082 		wqcb->arg = NULL;
2083 		INIT_LIST_HEAD(&wqcb->list_entry);
2084 		list_add_tail(&wqcb->list_entry, &reqtag_pool->freelist);
2085 	}
2086 
2087 	return reqtag_pool;
2088 }
2089 
2090 struct hw_wq_callback *
2091 efct_hw_reqtag_alloc(struct efct_hw *hw,
2092 		     void (*callback)(void *arg, u8 *cqe, int status),
2093 		     void *arg)
2094 {
2095 	struct hw_wq_callback *wqcb = NULL;
2096 	struct reqtag_pool *reqtag_pool = hw->wq_reqtag_pool;
2097 	unsigned long flags = 0;
2098 
2099 	if (!callback)
2100 		return wqcb;
2101 
2102 	spin_lock_irqsave(&reqtag_pool->lock, flags);
2103 
2104 	if (!list_empty(&reqtag_pool->freelist)) {
2105 		wqcb = list_first_entry(&reqtag_pool->freelist,
2106 					struct hw_wq_callback, list_entry);
2107 	}
2108 
2109 	if (wqcb) {
2110 		list_del_init(&wqcb->list_entry);
2111 		spin_unlock_irqrestore(&reqtag_pool->lock, flags);
2112 		wqcb->callback = callback;
2113 		wqcb->arg = arg;
2114 	} else {
2115 		spin_unlock_irqrestore(&reqtag_pool->lock, flags);
2116 	}
2117 
2118 	return wqcb;
2119 }
2120 
2121 void
2122 efct_hw_reqtag_free(struct efct_hw *hw, struct hw_wq_callback *wqcb)
2123 {
2124 	unsigned long flags = 0;
2125 	struct reqtag_pool *reqtag_pool = hw->wq_reqtag_pool;
2126 
2127 	if (!wqcb->callback)
2128 		efc_log_err(hw->os, "WQCB is already freed\n");
2129 
2130 	spin_lock_irqsave(&reqtag_pool->lock, flags);
2131 	wqcb->callback = NULL;
2132 	wqcb->arg = NULL;
2133 	INIT_LIST_HEAD(&wqcb->list_entry);
2134 	list_add(&wqcb->list_entry, &hw->wq_reqtag_pool->freelist);
2135 	spin_unlock_irqrestore(&reqtag_pool->lock, flags);
2136 }
2137 
2138 struct hw_wq_callback *
2139 efct_hw_reqtag_get_instance(struct efct_hw *hw, u32 instance_index)
2140 {
2141 	struct hw_wq_callback *wqcb;
2142 
2143 	wqcb = hw->wq_reqtag_pool->tags[instance_index];
2144 	if (!wqcb)
2145 		efc_log_err(hw->os, "wqcb for instance %d is null\n",
2146 			    instance_index);
2147 
2148 	return wqcb;
2149 }
2150 
2151 int
2152 efct_hw_queue_hash_find(struct efct_queue_hash *hash, u16 id)
2153 {
2154 	int index = -1;
2155 	int i = id & (EFCT_HW_Q_HASH_SIZE - 1);
2156 
2157 	/*
2158 	 * Since the hash is always bigger than the maximum number of Qs, then
2159 	 * we never have to worry about an infinite loop. We will always find
2160 	 * an unused entry.
2161 	 */
2162 	do {
2163 		if (hash[i].in_use && hash[i].id == id)
2164 			index = hash[i].index;
2165 		else
2166 			i = (i + 1) & (EFCT_HW_Q_HASH_SIZE - 1);
2167 	} while (index == -1 && hash[i].in_use);
2168 
2169 	return index;
2170 }
2171 
2172 int
2173 efct_hw_process(struct efct_hw *hw, u32 vector,
2174 		u32 max_isr_time_msec)
2175 {
2176 	struct hw_eq *eq;
2177 
2178 	/*
2179 	 * The caller should disable interrupts if they wish to prevent us
2180 	 * from processing during a shutdown. The following states are defined:
2181 	 *   EFCT_HW_STATE_UNINITIALIZED - No queues allocated
2182 	 *   EFCT_HW_STATE_QUEUES_ALLOCATED - The state after a chip reset,
2183 	 *                                    queues are cleared.
2184 	 *   EFCT_HW_STATE_ACTIVE - Chip and queues are operational
2185 	 *   EFCT_HW_STATE_RESET_IN_PROGRESS - reset, we still want completions
2186 	 *   EFCT_HW_STATE_TEARDOWN_IN_PROGRESS - We still want mailbox
2187 	 *                                        completions.
2188 	 */
2189 	if (hw->state == EFCT_HW_STATE_UNINITIALIZED)
2190 		return 0;
2191 
2192 	/* Get pointer to struct hw_eq */
2193 	eq = hw->hw_eq[vector];
2194 	if (!eq)
2195 		return 0;
2196 
2197 	eq->use_count++;
2198 
2199 	return efct_hw_eq_process(hw, eq, max_isr_time_msec);
2200 }
2201 
2202 int
2203 efct_hw_eq_process(struct efct_hw *hw, struct hw_eq *eq,
2204 		   u32 max_isr_time_msec)
2205 {
2206 	u8 eqe[sizeof(struct sli4_eqe)] = { 0 };
2207 	u32 tcheck_count;
2208 	u64 tstart;
2209 	u64 telapsed;
2210 	bool done = false;
2211 
2212 	tcheck_count = EFCT_HW_TIMECHECK_ITERATIONS;
2213 	tstart = jiffies_to_msecs(jiffies);
2214 
2215 	while (!done && !sli_eq_read(&hw->sli, eq->queue, eqe)) {
2216 		u16 cq_id = 0;
2217 		int rc;
2218 
2219 		rc = sli_eq_parse(&hw->sli, eqe, &cq_id);
2220 		if (unlikely(rc)) {
2221 			if (rc == SLI4_EQE_STATUS_EQ_FULL) {
2222 				u32 i;
2223 
2224 				/*
2225 				 * Received a sentinel EQE indicating the
2226 				 * EQ is full. Process all CQs
2227 				 */
2228 				for (i = 0; i < hw->cq_count; i++)
2229 					efct_hw_cq_process(hw, hw->hw_cq[i]);
2230 				continue;
2231 			} else {
2232 				return rc;
2233 			}
2234 		} else {
2235 			int index;
2236 
2237 			index  = efct_hw_queue_hash_find(hw->cq_hash, cq_id);
2238 
2239 			if (likely(index >= 0))
2240 				efct_hw_cq_process(hw, hw->hw_cq[index]);
2241 			else
2242 				efc_log_err(hw->os, "bad CQ_ID %#06x\n", cq_id);
2243 		}
2244 
2245 		if (eq->queue->n_posted > eq->queue->posted_limit)
2246 			sli_queue_arm(&hw->sli, eq->queue, false);
2247 
2248 		if (tcheck_count && (--tcheck_count == 0)) {
2249 			tcheck_count = EFCT_HW_TIMECHECK_ITERATIONS;
2250 			telapsed = jiffies_to_msecs(jiffies) - tstart;
2251 			if (telapsed >= max_isr_time_msec)
2252 				done = true;
2253 		}
2254 	}
2255 	sli_queue_eq_arm(&hw->sli, eq->queue, true);
2256 
2257 	return 0;
2258 }
2259 
2260 static int
2261 _efct_hw_wq_write(struct hw_wq *wq, struct efct_hw_wqe *wqe)
2262 {
2263 	int queue_rc;
2264 
2265 	/* Every so often, set the wqec bit to generate comsummed completions */
2266 	if (wq->wqec_count)
2267 		wq->wqec_count--;
2268 
2269 	if (wq->wqec_count == 0) {
2270 		struct sli4_generic_wqe *genwqe = (void *)wqe->wqebuf;
2271 
2272 		genwqe->cmdtype_wqec_byte |= SLI4_GEN_WQE_WQEC;
2273 		wq->wqec_count = wq->wqec_set_count;
2274 	}
2275 
2276 	/* Decrement WQ free count */
2277 	wq->free_count--;
2278 
2279 	queue_rc = sli_wq_write(&wq->hw->sli, wq->queue, wqe->wqebuf);
2280 
2281 	return (queue_rc < 0) ? -EIO : 0;
2282 }
2283 
2284 static void
2285 hw_wq_submit_pending(struct hw_wq *wq, u32 update_free_count)
2286 {
2287 	struct efct_hw_wqe *wqe;
2288 	unsigned long flags = 0;
2289 
2290 	spin_lock_irqsave(&wq->queue->lock, flags);
2291 
2292 	/* Update free count with value passed in */
2293 	wq->free_count += update_free_count;
2294 
2295 	while ((wq->free_count > 0) && (!list_empty(&wq->pending_list))) {
2296 		wqe = list_first_entry(&wq->pending_list,
2297 				       struct efct_hw_wqe, list_entry);
2298 		list_del_init(&wqe->list_entry);
2299 		_efct_hw_wq_write(wq, wqe);
2300 
2301 		if (wqe->abort_wqe_submit_needed) {
2302 			wqe->abort_wqe_submit_needed = false;
2303 			efct_hw_fill_abort_wqe(wq->hw, wqe);
2304 			INIT_LIST_HEAD(&wqe->list_entry);
2305 			list_add_tail(&wqe->list_entry, &wq->pending_list);
2306 			wq->wq_pending_count++;
2307 		}
2308 	}
2309 
2310 	spin_unlock_irqrestore(&wq->queue->lock, flags);
2311 }
2312 
2313 void
2314 efct_hw_cq_process(struct efct_hw *hw, struct hw_cq *cq)
2315 {
2316 	u8 cqe[sizeof(struct sli4_mcqe)];
2317 	u16 rid = U16_MAX;
2318 	/* completion type */
2319 	enum sli4_qentry ctype;
2320 	u32 n_processed = 0;
2321 	u32 tstart, telapsed;
2322 
2323 	tstart = jiffies_to_msecs(jiffies);
2324 
2325 	while (!sli_cq_read(&hw->sli, cq->queue, cqe)) {
2326 		int status;
2327 
2328 		status = sli_cq_parse(&hw->sli, cq->queue, cqe, &ctype, &rid);
2329 		/*
2330 		 * The sign of status is significant. If status is:
2331 		 * == 0 : call completed correctly and
2332 		 * the CQE indicated success
2333 		 * > 0 : call completed correctly and
2334 		 * the CQE indicated an error
2335 		 * < 0 : call failed and no information is available about the
2336 		 * CQE
2337 		 */
2338 		if (status < 0) {
2339 			if (status == SLI4_MCQE_STATUS_NOT_COMPLETED)
2340 				/*
2341 				 * Notification that an entry was consumed,
2342 				 * but not completed
2343 				 */
2344 				continue;
2345 
2346 			break;
2347 		}
2348 
2349 		switch (ctype) {
2350 		case SLI4_QENTRY_ASYNC:
2351 			sli_cqe_async(&hw->sli, cqe);
2352 			break;
2353 		case SLI4_QENTRY_MQ:
2354 			/*
2355 			 * Process MQ entry. Note there is no way to determine
2356 			 * the MQ_ID from the completion entry.
2357 			 */
2358 			efct_hw_mq_process(hw, status, hw->mq);
2359 			break;
2360 		case SLI4_QENTRY_WQ:
2361 			efct_hw_wq_process(hw, cq, cqe, status, rid);
2362 			break;
2363 		case SLI4_QENTRY_WQ_RELEASE: {
2364 			u32 wq_id = rid;
2365 			int index;
2366 			struct hw_wq *wq = NULL;
2367 
2368 			index = efct_hw_queue_hash_find(hw->wq_hash, wq_id);
2369 
2370 			if (likely(index >= 0)) {
2371 				wq = hw->hw_wq[index];
2372 			} else {
2373 				efc_log_err(hw->os, "bad WQ_ID %#06x\n", wq_id);
2374 				break;
2375 			}
2376 			/* Submit any HW IOs that are on the WQ pending list */
2377 			hw_wq_submit_pending(wq, wq->wqec_set_count);
2378 
2379 			break;
2380 		}
2381 
2382 		case SLI4_QENTRY_RQ:
2383 			efct_hw_rqpair_process_rq(hw, cq, cqe);
2384 			break;
2385 		case SLI4_QENTRY_XABT: {
2386 			efct_hw_xabt_process(hw, cq, cqe, rid);
2387 			break;
2388 		}
2389 		default:
2390 			efc_log_debug(hw->os, "unhandled ctype=%#x rid=%#x\n",
2391 				      ctype, rid);
2392 			break;
2393 		}
2394 
2395 		n_processed++;
2396 		if (n_processed == cq->queue->proc_limit)
2397 			break;
2398 
2399 		if (cq->queue->n_posted >= cq->queue->posted_limit)
2400 			sli_queue_arm(&hw->sli, cq->queue, false);
2401 	}
2402 
2403 	sli_queue_arm(&hw->sli, cq->queue, true);
2404 
2405 	if (n_processed > cq->queue->max_num_processed)
2406 		cq->queue->max_num_processed = n_processed;
2407 	telapsed = jiffies_to_msecs(jiffies) - tstart;
2408 	if (telapsed > cq->queue->max_process_time)
2409 		cq->queue->max_process_time = telapsed;
2410 }
2411 
2412 void
2413 efct_hw_wq_process(struct efct_hw *hw, struct hw_cq *cq,
2414 		   u8 *cqe, int status, u16 rid)
2415 {
2416 	struct hw_wq_callback *wqcb;
2417 
2418 	if (rid == EFCT_HW_REQUE_XRI_REGTAG) {
2419 		if (status)
2420 			efc_log_err(hw->os, "reque xri failed, status = %d\n",
2421 				    status);
2422 		return;
2423 	}
2424 
2425 	wqcb = efct_hw_reqtag_get_instance(hw, rid);
2426 	if (!wqcb) {
2427 		efc_log_err(hw->os, "invalid request tag: x%x\n", rid);
2428 		return;
2429 	}
2430 
2431 	if (!wqcb->callback) {
2432 		efc_log_err(hw->os, "wqcb callback is NULL\n");
2433 		return;
2434 	}
2435 
2436 	(*wqcb->callback)(wqcb->arg, cqe, status);
2437 }
2438 
2439 void
2440 efct_hw_xabt_process(struct efct_hw *hw, struct hw_cq *cq,
2441 		     u8 *cqe, u16 rid)
2442 {
2443 	/* search IOs wait free list */
2444 	struct efct_hw_io *io = NULL;
2445 	unsigned long flags = 0;
2446 
2447 	io = efct_hw_io_lookup(hw, rid);
2448 	if (!io) {
2449 		/* IO lookup failure should never happen */
2450 		efc_log_err(hw->os, "xabt io lookup failed rid=%#x\n", rid);
2451 		return;
2452 	}
2453 
2454 	if (!io->xbusy)
2455 		efc_log_debug(hw->os, "xabt io not busy rid=%#x\n", rid);
2456 	else
2457 		/* mark IO as no longer busy */
2458 		io->xbusy = false;
2459 
2460 	/*
2461 	 * For IOs that were aborted internally, we need to issue any pending
2462 	 * callback here.
2463 	 */
2464 	if (io->done) {
2465 		efct_hw_done_t done = io->done;
2466 		void		*arg = io->arg;
2467 
2468 		/*
2469 		 * Use latched status as this is always saved for an internal
2470 		 * abort
2471 		 */
2472 		int status = io->saved_status;
2473 		u32 len = io->saved_len;
2474 		u32 ext = io->saved_ext;
2475 
2476 		io->done = NULL;
2477 		io->status_saved = false;
2478 
2479 		done(io, len, status, ext, arg);
2480 	}
2481 
2482 	spin_lock_irqsave(&hw->io_lock, flags);
2483 	if (io->state == EFCT_HW_IO_STATE_INUSE ||
2484 	    io->state == EFCT_HW_IO_STATE_WAIT_FREE) {
2485 		/* if on wait_free list, caller has already freed IO;
2486 		 * remove from wait_free list and add to free list.
2487 		 * if on in-use list, already marked as no longer busy;
2488 		 * just leave there and wait for caller to free.
2489 		 */
2490 		if (io->state == EFCT_HW_IO_STATE_WAIT_FREE) {
2491 			io->state = EFCT_HW_IO_STATE_FREE;
2492 			list_del_init(&io->list_entry);
2493 			efct_hw_io_free_move_correct_list(hw, io);
2494 		}
2495 	}
2496 	spin_unlock_irqrestore(&hw->io_lock, flags);
2497 }
2498 
2499 static int
2500 efct_hw_flush(struct efct_hw *hw)
2501 {
2502 	u32 i = 0;
2503 
2504 	/* Process any remaining completions */
2505 	for (i = 0; i < hw->eq_count; i++)
2506 		efct_hw_process(hw, i, ~0);
2507 
2508 	return 0;
2509 }
2510 
2511 int
2512 efct_hw_wq_write(struct hw_wq *wq, struct efct_hw_wqe *wqe)
2513 {
2514 	int rc = 0;
2515 	unsigned long flags = 0;
2516 
2517 	spin_lock_irqsave(&wq->queue->lock, flags);
2518 	if (list_empty(&wq->pending_list)) {
2519 		if (wq->free_count > 0) {
2520 			rc = _efct_hw_wq_write(wq, wqe);
2521 		} else {
2522 			INIT_LIST_HEAD(&wqe->list_entry);
2523 			list_add_tail(&wqe->list_entry, &wq->pending_list);
2524 			wq->wq_pending_count++;
2525 		}
2526 
2527 		spin_unlock_irqrestore(&wq->queue->lock, flags);
2528 		return rc;
2529 	}
2530 
2531 	INIT_LIST_HEAD(&wqe->list_entry);
2532 	list_add_tail(&wqe->list_entry, &wq->pending_list);
2533 	wq->wq_pending_count++;
2534 	while (wq->free_count > 0) {
2535 		wqe = list_first_entry(&wq->pending_list, struct efct_hw_wqe,
2536 				       list_entry);
2537 		if (!wqe)
2538 			break;
2539 
2540 		list_del_init(&wqe->list_entry);
2541 		rc = _efct_hw_wq_write(wq, wqe);
2542 		if (rc)
2543 			break;
2544 
2545 		if (wqe->abort_wqe_submit_needed) {
2546 			wqe->abort_wqe_submit_needed = false;
2547 			efct_hw_fill_abort_wqe(wq->hw, wqe);
2548 
2549 			INIT_LIST_HEAD(&wqe->list_entry);
2550 			list_add_tail(&wqe->list_entry, &wq->pending_list);
2551 			wq->wq_pending_count++;
2552 		}
2553 	}
2554 
2555 	spin_unlock_irqrestore(&wq->queue->lock, flags);
2556 
2557 	return rc;
2558 }
2559 
2560 int
2561 efct_efc_bls_send(struct efc *efc, u32 type, struct sli_bls_params *bls)
2562 {
2563 	struct efct *efct = efc->base;
2564 
2565 	return efct_hw_bls_send(efct, type, bls, NULL, NULL);
2566 }
2567 
2568 int
2569 efct_hw_bls_send(struct efct *efct, u32 type, struct sli_bls_params *bls_params,
2570 		 void *cb, void *arg)
2571 {
2572 	struct efct_hw *hw = &efct->hw;
2573 	struct efct_hw_io *hio;
2574 	struct sli_bls_payload bls;
2575 	int rc;
2576 
2577 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
2578 		efc_log_err(hw->os,
2579 			    "cannot send BLS, HW state=%d\n", hw->state);
2580 		return -EIO;
2581 	}
2582 
2583 	hio = efct_hw_io_alloc(hw);
2584 	if (!hio) {
2585 		efc_log_err(hw->os, "HIO allocation failed\n");
2586 		return -EIO;
2587 	}
2588 
2589 	hio->done = cb;
2590 	hio->arg  = arg;
2591 
2592 	bls_params->xri = hio->indicator;
2593 	bls_params->tag = hio->reqtag;
2594 
2595 	if (type == FC_RCTL_BA_ACC) {
2596 		hio->type = EFCT_HW_BLS_ACC;
2597 		bls.type = SLI4_SLI_BLS_ACC;
2598 		memcpy(&bls.u.acc, bls_params->payload, sizeof(bls.u.acc));
2599 	} else {
2600 		hio->type = EFCT_HW_BLS_RJT;
2601 		bls.type = SLI4_SLI_BLS_RJT;
2602 		memcpy(&bls.u.rjt, bls_params->payload, sizeof(bls.u.rjt));
2603 	}
2604 
2605 	bls.ox_id = cpu_to_le16(bls_params->ox_id);
2606 	bls.rx_id = cpu_to_le16(bls_params->rx_id);
2607 
2608 	if (sli_xmit_bls_rsp64_wqe(&hw->sli, hio->wqe.wqebuf,
2609 				   &bls, bls_params)) {
2610 		efc_log_err(hw->os, "XMIT_BLS_RSP64 WQE error\n");
2611 		return -EIO;
2612 	}
2613 
2614 	hio->xbusy = true;
2615 
2616 	/*
2617 	 * Add IO to active io wqe list before submitting, in case the
2618 	 * wcqe processing preempts this thread.
2619 	 */
2620 	hio->wq->use_count++;
2621 	rc = efct_hw_wq_write(hio->wq, &hio->wqe);
2622 	if (rc >= 0) {
2623 		/* non-negative return is success */
2624 		rc = 0;
2625 	} else {
2626 		/* failed to write wqe, remove from active wqe list */
2627 		efc_log_err(hw->os,
2628 			    "sli_queue_write failed: %d\n", rc);
2629 		hio->xbusy = false;
2630 	}
2631 
2632 	return rc;
2633 }
2634 
2635 static int
2636 efct_els_ssrs_send_cb(struct efct_hw_io *hio, u32 length, int status,
2637 		      u32 ext_status, void *arg)
2638 {
2639 	struct efc_disc_io *io = arg;
2640 
2641 	efc_disc_io_complete(io, length, status, ext_status);
2642 	return 0;
2643 }
2644 
2645 static inline void
2646 efct_fill_els_params(struct efc_disc_io *io, struct sli_els_params *params)
2647 {
2648 	u8 *cmd = io->req.virt;
2649 
2650 	params->cmd = *cmd;
2651 	params->s_id = io->s_id;
2652 	params->d_id = io->d_id;
2653 	params->ox_id = io->iparam.els.ox_id;
2654 	params->rpi = io->rpi;
2655 	params->vpi = io->vpi;
2656 	params->rpi_registered = io->rpi_registered;
2657 	params->xmit_len = io->xmit_len;
2658 	params->rsp_len = io->rsp_len;
2659 	params->timeout = io->iparam.els.timeout;
2660 }
2661 
2662 static inline void
2663 efct_fill_ct_params(struct efc_disc_io *io, struct sli_ct_params *params)
2664 {
2665 	params->r_ctl = io->iparam.ct.r_ctl;
2666 	params->type = io->iparam.ct.type;
2667 	params->df_ctl =  io->iparam.ct.df_ctl;
2668 	params->d_id = io->d_id;
2669 	params->ox_id = io->iparam.ct.ox_id;
2670 	params->rpi = io->rpi;
2671 	params->vpi = io->vpi;
2672 	params->rpi_registered = io->rpi_registered;
2673 	params->xmit_len = io->xmit_len;
2674 	params->rsp_len = io->rsp_len;
2675 	params->timeout = io->iparam.ct.timeout;
2676 }
2677 
2678 /**
2679  * efct_els_hw_srrs_send() - Send a single request and response cmd.
2680  * @efc: efc library structure
2681  * @io: Discovery IO used to hold els and ct cmd context.
2682  *
2683  * This routine supports communication sequences consisting of a single
2684  * request and single response between two endpoints. Examples include:
2685  *  - Sending an ELS request.
2686  *  - Sending an ELS response - To send an ELS response, the caller must provide
2687  * the OX_ID from the received request.
2688  *  - Sending a FC Common Transport (FC-CT) request - To send a FC-CT request,
2689  * the caller must provide the R_CTL, TYPE, and DF_CTL
2690  * values to place in the FC frame header.
2691  *
2692  * Return: Status of the request.
2693  */
2694 int
2695 efct_els_hw_srrs_send(struct efc *efc, struct efc_disc_io *io)
2696 {
2697 	struct efct *efct = efc->base;
2698 	struct efct_hw_io *hio;
2699 	struct efct_hw *hw = &efct->hw;
2700 	struct efc_dma *send = &io->req;
2701 	struct efc_dma *receive = &io->rsp;
2702 	struct sli4_sge	*sge = NULL;
2703 	int rc = 0;
2704 	u32 len = io->xmit_len;
2705 	u32 sge0_flags;
2706 	u32 sge1_flags;
2707 
2708 	hio = efct_hw_io_alloc(hw);
2709 	if (!hio) {
2710 		pr_err("HIO alloc failed\n");
2711 		return -EIO;
2712 	}
2713 
2714 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
2715 		efc_log_debug(hw->os,
2716 			      "cannot send SRRS, HW state=%d\n", hw->state);
2717 		return -EIO;
2718 	}
2719 
2720 	hio->done = efct_els_ssrs_send_cb;
2721 	hio->arg  = io;
2722 
2723 	sge = hio->sgl->virt;
2724 
2725 	/* clear both SGE */
2726 	memset(hio->sgl->virt, 0, 2 * sizeof(struct sli4_sge));
2727 
2728 	sge0_flags = le32_to_cpu(sge[0].dw2_flags);
2729 	sge1_flags = le32_to_cpu(sge[1].dw2_flags);
2730 	if (send->size) {
2731 		sge[0].buffer_address_high =
2732 			cpu_to_le32(upper_32_bits(send->phys));
2733 		sge[0].buffer_address_low  =
2734 			cpu_to_le32(lower_32_bits(send->phys));
2735 
2736 		sge0_flags |= (SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT);
2737 
2738 		sge[0].buffer_length = cpu_to_le32(len);
2739 	}
2740 
2741 	if (io->io_type == EFC_DISC_IO_ELS_REQ ||
2742 	    io->io_type == EFC_DISC_IO_CT_REQ) {
2743 		sge[1].buffer_address_high =
2744 			cpu_to_le32(upper_32_bits(receive->phys));
2745 		sge[1].buffer_address_low  =
2746 			cpu_to_le32(lower_32_bits(receive->phys));
2747 
2748 		sge1_flags |= (SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT);
2749 		sge1_flags |= SLI4_SGE_LAST;
2750 
2751 		sge[1].buffer_length = cpu_to_le32(receive->size);
2752 	} else {
2753 		sge0_flags |= SLI4_SGE_LAST;
2754 	}
2755 
2756 	sge[0].dw2_flags = cpu_to_le32(sge0_flags);
2757 	sge[1].dw2_flags = cpu_to_le32(sge1_flags);
2758 
2759 	switch (io->io_type) {
2760 	case EFC_DISC_IO_ELS_REQ: {
2761 		struct sli_els_params els_params;
2762 
2763 		hio->type = EFCT_HW_ELS_REQ;
2764 		efct_fill_els_params(io, &els_params);
2765 		els_params.xri = hio->indicator;
2766 		els_params.tag = hio->reqtag;
2767 
2768 		if (sli_els_request64_wqe(&hw->sli, hio->wqe.wqebuf, hio->sgl,
2769 					  &els_params)) {
2770 			efc_log_err(hw->os, "REQ WQE error\n");
2771 			rc = -EIO;
2772 		}
2773 		break;
2774 	}
2775 	case EFC_DISC_IO_ELS_RESP: {
2776 		struct sli_els_params els_params;
2777 
2778 		hio->type = EFCT_HW_ELS_RSP;
2779 		efct_fill_els_params(io, &els_params);
2780 		els_params.xri = hio->indicator;
2781 		els_params.tag = hio->reqtag;
2782 		if (sli_xmit_els_rsp64_wqe(&hw->sli, hio->wqe.wqebuf, send,
2783 					   &els_params)){
2784 			efc_log_err(hw->os, "RSP WQE error\n");
2785 			rc = -EIO;
2786 		}
2787 		break;
2788 	}
2789 	case EFC_DISC_IO_CT_REQ: {
2790 		struct sli_ct_params ct_params;
2791 
2792 		hio->type = EFCT_HW_FC_CT;
2793 		efct_fill_ct_params(io, &ct_params);
2794 		ct_params.xri = hio->indicator;
2795 		ct_params.tag = hio->reqtag;
2796 		if (sli_gen_request64_wqe(&hw->sli, hio->wqe.wqebuf, hio->sgl,
2797 					  &ct_params)){
2798 			efc_log_err(hw->os, "GEN WQE error\n");
2799 			rc = -EIO;
2800 		}
2801 		break;
2802 	}
2803 	case EFC_DISC_IO_CT_RESP: {
2804 		struct sli_ct_params ct_params;
2805 
2806 		hio->type = EFCT_HW_FC_CT_RSP;
2807 		efct_fill_ct_params(io, &ct_params);
2808 		ct_params.xri = hio->indicator;
2809 		ct_params.tag = hio->reqtag;
2810 		if (sli_xmit_sequence64_wqe(&hw->sli, hio->wqe.wqebuf, hio->sgl,
2811 					    &ct_params)){
2812 			efc_log_err(hw->os, "XMIT SEQ WQE error\n");
2813 			rc = -EIO;
2814 		}
2815 		break;
2816 	}
2817 	default:
2818 		efc_log_err(hw->os, "bad SRRS type %#x\n", io->io_type);
2819 		rc = -EIO;
2820 	}
2821 
2822 	if (rc == 0) {
2823 		hio->xbusy = true;
2824 
2825 		/*
2826 		 * Add IO to active io wqe list before submitting, in case the
2827 		 * wcqe processing preempts this thread.
2828 		 */
2829 		hio->wq->use_count++;
2830 		rc = efct_hw_wq_write(hio->wq, &hio->wqe);
2831 		if (rc >= 0) {
2832 			/* non-negative return is success */
2833 			rc = 0;
2834 		} else {
2835 			/* failed to write wqe, remove from active wqe list */
2836 			efc_log_err(hw->os,
2837 				    "sli_queue_write failed: %d\n", rc);
2838 			hio->xbusy = false;
2839 		}
2840 	}
2841 
2842 	return rc;
2843 }
2844 
2845 int
2846 efct_hw_io_send(struct efct_hw *hw, enum efct_hw_io_type type,
2847 		struct efct_hw_io *io, union efct_hw_io_param_u *iparam,
2848 		void *cb, void *arg)
2849 {
2850 	int rc = 0;
2851 	bool send_wqe = true;
2852 
2853 	if (!io) {
2854 		pr_err("bad parm hw=%p io=%p\n", hw, io);
2855 		return -EIO;
2856 	}
2857 
2858 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
2859 		efc_log_err(hw->os, "cannot send IO, HW state=%d\n", hw->state);
2860 		return -EIO;
2861 	}
2862 
2863 	/*
2864 	 * Save state needed during later stages
2865 	 */
2866 	io->type  = type;
2867 	io->done  = cb;
2868 	io->arg   = arg;
2869 
2870 	/*
2871 	 * Format the work queue entry used to send the IO
2872 	 */
2873 	switch (type) {
2874 	case EFCT_HW_IO_TARGET_WRITE: {
2875 		u16 *flags = &iparam->fcp_tgt.flags;
2876 		struct fcp_txrdy *xfer = io->xfer_rdy.virt;
2877 
2878 		/*
2879 		 * Fill in the XFER_RDY for IF_TYPE 0 devices
2880 		 */
2881 		xfer->ft_data_ro = cpu_to_be32(iparam->fcp_tgt.offset);
2882 		xfer->ft_burst_len = cpu_to_be32(iparam->fcp_tgt.xmit_len);
2883 
2884 		if (io->xbusy)
2885 			*flags |= SLI4_IO_CONTINUATION;
2886 		else
2887 			*flags &= ~SLI4_IO_CONTINUATION;
2888 		iparam->fcp_tgt.xri = io->indicator;
2889 		iparam->fcp_tgt.tag = io->reqtag;
2890 
2891 		if (sli_fcp_treceive64_wqe(&hw->sli, io->wqe.wqebuf,
2892 					   &io->def_sgl, io->first_data_sge,
2893 					   SLI4_CQ_DEFAULT,
2894 					   0, 0, &iparam->fcp_tgt)) {
2895 			efc_log_err(hw->os, "TRECEIVE WQE error\n");
2896 			rc = -EIO;
2897 		}
2898 		break;
2899 	}
2900 	case EFCT_HW_IO_TARGET_READ: {
2901 		u16 *flags = &iparam->fcp_tgt.flags;
2902 
2903 		if (io->xbusy)
2904 			*flags |= SLI4_IO_CONTINUATION;
2905 		else
2906 			*flags &= ~SLI4_IO_CONTINUATION;
2907 
2908 		iparam->fcp_tgt.xri = io->indicator;
2909 		iparam->fcp_tgt.tag = io->reqtag;
2910 
2911 		if (sli_fcp_tsend64_wqe(&hw->sli, io->wqe.wqebuf,
2912 					&io->def_sgl, io->first_data_sge,
2913 					SLI4_CQ_DEFAULT,
2914 					0, 0, &iparam->fcp_tgt)) {
2915 			efc_log_err(hw->os, "TSEND WQE error\n");
2916 			rc = -EIO;
2917 		}
2918 		break;
2919 	}
2920 	case EFCT_HW_IO_TARGET_RSP: {
2921 		u16 *flags = &iparam->fcp_tgt.flags;
2922 
2923 		if (io->xbusy)
2924 			*flags |= SLI4_IO_CONTINUATION;
2925 		else
2926 			*flags &= ~SLI4_IO_CONTINUATION;
2927 
2928 		iparam->fcp_tgt.xri = io->indicator;
2929 		iparam->fcp_tgt.tag = io->reqtag;
2930 
2931 		if (sli_fcp_trsp64_wqe(&hw->sli, io->wqe.wqebuf,
2932 				       &io->def_sgl, SLI4_CQ_DEFAULT,
2933 				       0, &iparam->fcp_tgt)) {
2934 			efc_log_err(hw->os, "TRSP WQE error\n");
2935 			rc = -EIO;
2936 		}
2937 
2938 		break;
2939 	}
2940 	default:
2941 		efc_log_err(hw->os, "unsupported IO type %#x\n", type);
2942 		rc = -EIO;
2943 	}
2944 
2945 	if (send_wqe && rc == 0) {
2946 		io->xbusy = true;
2947 
2948 		/*
2949 		 * Add IO to active io wqe list before submitting, in case the
2950 		 * wcqe processing preempts this thread.
2951 		 */
2952 		hw->tcmd_wq_submit[io->wq->instance]++;
2953 		io->wq->use_count++;
2954 		rc = efct_hw_wq_write(io->wq, &io->wqe);
2955 		if (rc >= 0) {
2956 			/* non-negative return is success */
2957 			rc = 0;
2958 		} else {
2959 			/* failed to write wqe, remove from active wqe list */
2960 			efc_log_err(hw->os,
2961 				    "sli_queue_write failed: %d\n", rc);
2962 			io->xbusy = false;
2963 		}
2964 	}
2965 
2966 	return rc;
2967 }
2968 
2969 int
2970 efct_hw_send_frame(struct efct_hw *hw, struct fc_frame_header *hdr,
2971 		   u8 sof, u8 eof, struct efc_dma *payload,
2972 		   struct efct_hw_send_frame_context *ctx,
2973 		   void (*callback)(void *arg, u8 *cqe, int status),
2974 		   void *arg)
2975 {
2976 	int rc;
2977 	struct efct_hw_wqe *wqe;
2978 	u32 xri;
2979 	struct hw_wq *wq;
2980 
2981 	wqe = &ctx->wqe;
2982 
2983 	/* populate the callback object */
2984 	ctx->hw = hw;
2985 
2986 	/* Fetch and populate request tag */
2987 	ctx->wqcb = efct_hw_reqtag_alloc(hw, callback, arg);
2988 	if (!ctx->wqcb) {
2989 		efc_log_err(hw->os, "can't allocate request tag\n");
2990 		return -ENOSPC;
2991 	}
2992 
2993 	wq = hw->hw_wq[0];
2994 
2995 	/* Set XRI and RX_ID in the header based on which WQ, and which
2996 	 * send_frame_io we are using
2997 	 */
2998 	xri = wq->send_frame_io->indicator;
2999 
3000 	/* Build the send frame WQE */
3001 	rc = sli_send_frame_wqe(&hw->sli, wqe->wqebuf,
3002 				sof, eof, (u32 *)hdr, payload, payload->len,
3003 				EFCT_HW_SEND_FRAME_TIMEOUT, xri,
3004 				ctx->wqcb->instance_index);
3005 	if (rc) {
3006 		efc_log_err(hw->os, "sli_send_frame_wqe failed: %d\n", rc);
3007 		return -EIO;
3008 	}
3009 
3010 	/* Write to WQ */
3011 	rc = efct_hw_wq_write(wq, wqe);
3012 	if (rc) {
3013 		efc_log_err(hw->os, "efct_hw_wq_write failed: %d\n", rc);
3014 		return -EIO;
3015 	}
3016 
3017 	wq->use_count++;
3018 
3019 	return 0;
3020 }
3021 
3022 static int
3023 efct_hw_cb_link_stat(struct efct_hw *hw, int status,
3024 		     u8 *mqe, void  *arg)
3025 {
3026 	struct sli4_cmd_read_link_stats *mbox_rsp;
3027 	struct efct_hw_link_stat_cb_arg *cb_arg = arg;
3028 	struct efct_hw_link_stat_counts counts[EFCT_HW_LINK_STAT_MAX];
3029 	u32 num_counters, i;
3030 	u32 mbox_rsp_flags = 0;
3031 
3032 	mbox_rsp = (struct sli4_cmd_read_link_stats *)mqe;
3033 	mbox_rsp_flags = le32_to_cpu(mbox_rsp->dw1_flags);
3034 	num_counters = (mbox_rsp_flags & SLI4_READ_LNKSTAT_GEC) ? 20 : 13;
3035 	memset(counts, 0, sizeof(struct efct_hw_link_stat_counts) *
3036 				 EFCT_HW_LINK_STAT_MAX);
3037 
3038 	/* Fill overflow counts, mask starts from SLI4_READ_LNKSTAT_W02OF*/
3039 	for (i = 0; i < EFCT_HW_LINK_STAT_MAX; i++)
3040 		counts[i].overflow = (mbox_rsp_flags & (1 << (i + 2)));
3041 
3042 	counts[EFCT_HW_LINK_STAT_LINK_FAILURE_COUNT].counter =
3043 		 le32_to_cpu(mbox_rsp->linkfail_errcnt);
3044 	counts[EFCT_HW_LINK_STAT_LOSS_OF_SYNC_COUNT].counter =
3045 		 le32_to_cpu(mbox_rsp->losssync_errcnt);
3046 	counts[EFCT_HW_LINK_STAT_LOSS_OF_SIGNAL_COUNT].counter =
3047 		 le32_to_cpu(mbox_rsp->losssignal_errcnt);
3048 	counts[EFCT_HW_LINK_STAT_PRIMITIVE_SEQ_COUNT].counter =
3049 		 le32_to_cpu(mbox_rsp->primseq_errcnt);
3050 	counts[EFCT_HW_LINK_STAT_INVALID_XMIT_WORD_COUNT].counter =
3051 		 le32_to_cpu(mbox_rsp->inval_txword_errcnt);
3052 	counts[EFCT_HW_LINK_STAT_CRC_COUNT].counter =
3053 		le32_to_cpu(mbox_rsp->crc_errcnt);
3054 	counts[EFCT_HW_LINK_STAT_PRIMITIVE_SEQ_TIMEOUT_COUNT].counter =
3055 		le32_to_cpu(mbox_rsp->primseq_eventtimeout_cnt);
3056 	counts[EFCT_HW_LINK_STAT_ELASTIC_BUFFER_OVERRUN_COUNT].counter =
3057 		 le32_to_cpu(mbox_rsp->elastic_bufoverrun_errcnt);
3058 	counts[EFCT_HW_LINK_STAT_ARB_TIMEOUT_COUNT].counter =
3059 		 le32_to_cpu(mbox_rsp->arbit_fc_al_timeout_cnt);
3060 	counts[EFCT_HW_LINK_STAT_ADVERTISED_RCV_B2B_CREDIT].counter =
3061 		 le32_to_cpu(mbox_rsp->adv_rx_buftor_to_buf_credit);
3062 	counts[EFCT_HW_LINK_STAT_CURR_RCV_B2B_CREDIT].counter =
3063 		 le32_to_cpu(mbox_rsp->curr_rx_buf_to_buf_credit);
3064 	counts[EFCT_HW_LINK_STAT_ADVERTISED_XMIT_B2B_CREDIT].counter =
3065 		 le32_to_cpu(mbox_rsp->adv_tx_buf_to_buf_credit);
3066 	counts[EFCT_HW_LINK_STAT_CURR_XMIT_B2B_CREDIT].counter =
3067 		 le32_to_cpu(mbox_rsp->curr_tx_buf_to_buf_credit);
3068 	counts[EFCT_HW_LINK_STAT_RCV_EOFA_COUNT].counter =
3069 		 le32_to_cpu(mbox_rsp->rx_eofa_cnt);
3070 	counts[EFCT_HW_LINK_STAT_RCV_EOFDTI_COUNT].counter =
3071 		 le32_to_cpu(mbox_rsp->rx_eofdti_cnt);
3072 	counts[EFCT_HW_LINK_STAT_RCV_EOFNI_COUNT].counter =
3073 		 le32_to_cpu(mbox_rsp->rx_eofni_cnt);
3074 	counts[EFCT_HW_LINK_STAT_RCV_SOFF_COUNT].counter =
3075 		 le32_to_cpu(mbox_rsp->rx_soff_cnt);
3076 	counts[EFCT_HW_LINK_STAT_RCV_DROPPED_NO_AER_COUNT].counter =
3077 		 le32_to_cpu(mbox_rsp->rx_dropped_no_aer_cnt);
3078 	counts[EFCT_HW_LINK_STAT_RCV_DROPPED_NO_RPI_COUNT].counter =
3079 		 le32_to_cpu(mbox_rsp->rx_dropped_no_avail_rpi_rescnt);
3080 	counts[EFCT_HW_LINK_STAT_RCV_DROPPED_NO_XRI_COUNT].counter =
3081 		 le32_to_cpu(mbox_rsp->rx_dropped_no_avail_xri_rescnt);
3082 
3083 	if (cb_arg) {
3084 		if (cb_arg->cb) {
3085 			if (status == 0 && le16_to_cpu(mbox_rsp->hdr.status))
3086 				status = le16_to_cpu(mbox_rsp->hdr.status);
3087 			cb_arg->cb(status, num_counters, counts, cb_arg->arg);
3088 		}
3089 
3090 		kfree(cb_arg);
3091 	}
3092 
3093 	return 0;
3094 }
3095 
3096 int
3097 efct_hw_get_link_stats(struct efct_hw *hw, u8 req_ext_counters,
3098 		       u8 clear_overflow_flags, u8 clear_all_counters,
3099 		       void (*cb)(int status, u32 num_counters,
3100 				  struct efct_hw_link_stat_counts *counters,
3101 				  void *arg),
3102 		       void *arg)
3103 {
3104 	int rc = -EIO;
3105 	struct efct_hw_link_stat_cb_arg *cb_arg;
3106 	u8 mbxdata[SLI4_BMBX_SIZE];
3107 
3108 	cb_arg = kzalloc(sizeof(*cb_arg), GFP_ATOMIC);
3109 	if (!cb_arg)
3110 		return -ENOMEM;
3111 
3112 	cb_arg->cb = cb;
3113 	cb_arg->arg = arg;
3114 
3115 	/* Send the HW command */
3116 	if (!sli_cmd_read_link_stats(&hw->sli, mbxdata, req_ext_counters,
3117 				    clear_overflow_flags, clear_all_counters))
3118 		rc = efct_hw_command(hw, mbxdata, EFCT_CMD_NOWAIT,
3119 				     efct_hw_cb_link_stat, cb_arg);
3120 
3121 	if (rc)
3122 		kfree(cb_arg);
3123 
3124 	return rc;
3125 }
3126 
3127 static int
3128 efct_hw_cb_host_stat(struct efct_hw *hw, int status, u8 *mqe, void  *arg)
3129 {
3130 	struct sli4_cmd_read_status *mbox_rsp =
3131 					(struct sli4_cmd_read_status *)mqe;
3132 	struct efct_hw_host_stat_cb_arg *cb_arg = arg;
3133 	struct efct_hw_host_stat_counts counts[EFCT_HW_HOST_STAT_MAX];
3134 	u32 num_counters = EFCT_HW_HOST_STAT_MAX;
3135 
3136 	memset(counts, 0, sizeof(struct efct_hw_host_stat_counts) *
3137 	       EFCT_HW_HOST_STAT_MAX);
3138 
3139 	counts[EFCT_HW_HOST_STAT_TX_KBYTE_COUNT].counter =
3140 		 le32_to_cpu(mbox_rsp->trans_kbyte_cnt);
3141 	counts[EFCT_HW_HOST_STAT_RX_KBYTE_COUNT].counter =
3142 		 le32_to_cpu(mbox_rsp->recv_kbyte_cnt);
3143 	counts[EFCT_HW_HOST_STAT_TX_FRAME_COUNT].counter =
3144 		 le32_to_cpu(mbox_rsp->trans_frame_cnt);
3145 	counts[EFCT_HW_HOST_STAT_RX_FRAME_COUNT].counter =
3146 		 le32_to_cpu(mbox_rsp->recv_frame_cnt);
3147 	counts[EFCT_HW_HOST_STAT_TX_SEQ_COUNT].counter =
3148 		 le32_to_cpu(mbox_rsp->trans_seq_cnt);
3149 	counts[EFCT_HW_HOST_STAT_RX_SEQ_COUNT].counter =
3150 		 le32_to_cpu(mbox_rsp->recv_seq_cnt);
3151 	counts[EFCT_HW_HOST_STAT_TOTAL_EXCH_ORIG].counter =
3152 		 le32_to_cpu(mbox_rsp->tot_exchanges_orig);
3153 	counts[EFCT_HW_HOST_STAT_TOTAL_EXCH_RESP].counter =
3154 		 le32_to_cpu(mbox_rsp->tot_exchanges_resp);
3155 	counts[EFCT_HW_HOSY_STAT_RX_P_BSY_COUNT].counter =
3156 		 le32_to_cpu(mbox_rsp->recv_p_bsy_cnt);
3157 	counts[EFCT_HW_HOST_STAT_RX_F_BSY_COUNT].counter =
3158 		 le32_to_cpu(mbox_rsp->recv_f_bsy_cnt);
3159 	counts[EFCT_HW_HOST_STAT_DROP_FRM_DUE_TO_NO_RQ_BUF_COUNT].counter =
3160 		 le32_to_cpu(mbox_rsp->no_rq_buf_dropped_frames_cnt);
3161 	counts[EFCT_HW_HOST_STAT_EMPTY_RQ_TIMEOUT_COUNT].counter =
3162 		 le32_to_cpu(mbox_rsp->empty_rq_timeout_cnt);
3163 	counts[EFCT_HW_HOST_STAT_DROP_FRM_DUE_TO_NO_XRI_COUNT].counter =
3164 		 le32_to_cpu(mbox_rsp->no_xri_dropped_frames_cnt);
3165 	counts[EFCT_HW_HOST_STAT_EMPTY_XRI_POOL_COUNT].counter =
3166 		 le32_to_cpu(mbox_rsp->empty_xri_pool_cnt);
3167 
3168 	if (cb_arg) {
3169 		if (cb_arg->cb) {
3170 			if (status == 0 && le16_to_cpu(mbox_rsp->hdr.status))
3171 				status = le16_to_cpu(mbox_rsp->hdr.status);
3172 			cb_arg->cb(status, num_counters, counts, cb_arg->arg);
3173 		}
3174 
3175 		kfree(cb_arg);
3176 	}
3177 
3178 	return 0;
3179 }
3180 
3181 int
3182 efct_hw_get_host_stats(struct efct_hw *hw, u8 cc,
3183 		       void (*cb)(int status, u32 num_counters,
3184 				  struct efct_hw_host_stat_counts *counters,
3185 				  void *arg),
3186 		       void *arg)
3187 {
3188 	int rc = -EIO;
3189 	struct efct_hw_host_stat_cb_arg *cb_arg;
3190 	u8 mbxdata[SLI4_BMBX_SIZE];
3191 
3192 	cb_arg = kmalloc(sizeof(*cb_arg), GFP_ATOMIC);
3193 	if (!cb_arg)
3194 		return -ENOMEM;
3195 
3196 	cb_arg->cb = cb;
3197 	cb_arg->arg = arg;
3198 
3199 	 /* Send the HW command to get the host stats */
3200 	if (!sli_cmd_read_status(&hw->sli, mbxdata, cc))
3201 		rc = efct_hw_command(hw, mbxdata, EFCT_CMD_NOWAIT,
3202 				     efct_hw_cb_host_stat, cb_arg);
3203 
3204 	if (rc) {
3205 		efc_log_debug(hw->os, "READ_HOST_STATS failed\n");
3206 		kfree(cb_arg);
3207 	}
3208 
3209 	return rc;
3210 }
3211 
3212 struct efct_hw_async_call_ctx {
3213 	efct_hw_async_cb_t callback;
3214 	void *arg;
3215 	u8 cmd[SLI4_BMBX_SIZE];
3216 };
3217 
3218 static void
3219 efct_hw_async_cb(struct efct_hw *hw, int status, u8 *mqe, void *arg)
3220 {
3221 	struct efct_hw_async_call_ctx *ctx = arg;
3222 
3223 	if (ctx) {
3224 		if (ctx->callback)
3225 			(*ctx->callback)(hw, status, mqe, ctx->arg);
3226 
3227 		kfree(ctx);
3228 	}
3229 }
3230 
3231 int
3232 efct_hw_async_call(struct efct_hw *hw, efct_hw_async_cb_t callback, void *arg)
3233 {
3234 	struct efct_hw_async_call_ctx *ctx;
3235 	int rc;
3236 
3237 	/*
3238 	 * Allocate a callback context (which includes the mbox cmd buffer),
3239 	 * we need this to be persistent as the mbox cmd submission may be
3240 	 * queued and executed later execution.
3241 	 */
3242 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
3243 	if (!ctx)
3244 		return -ENOMEM;
3245 
3246 	ctx->callback = callback;
3247 	ctx->arg = arg;
3248 
3249 	/* Build and send a NOP mailbox command */
3250 	if (sli_cmd_common_nop(&hw->sli, ctx->cmd, 0)) {
3251 		efc_log_err(hw->os, "COMMON_NOP format failure\n");
3252 		kfree(ctx);
3253 		return -EIO;
3254 	}
3255 
3256 	rc = efct_hw_command(hw, ctx->cmd, EFCT_CMD_NOWAIT, efct_hw_async_cb,
3257 			     ctx);
3258 	if (rc) {
3259 		efc_log_err(hw->os, "COMMON_NOP command failure, rc=%d\n", rc);
3260 		kfree(ctx);
3261 		return -EIO;
3262 	}
3263 	return 0;
3264 }
3265 
3266 static int
3267 efct_hw_cb_fw_write(struct efct_hw *hw, int status, u8 *mqe, void  *arg)
3268 {
3269 	struct sli4_cmd_sli_config *mbox_rsp =
3270 					(struct sli4_cmd_sli_config *)mqe;
3271 	struct sli4_rsp_cmn_write_object *wr_obj_rsp;
3272 	struct efct_hw_fw_wr_cb_arg *cb_arg = arg;
3273 	u32 bytes_written;
3274 	u16 mbox_status;
3275 	u32 change_status;
3276 
3277 	wr_obj_rsp = (struct sli4_rsp_cmn_write_object *)
3278 		      &mbox_rsp->payload.embed;
3279 	bytes_written = le32_to_cpu(wr_obj_rsp->actual_write_length);
3280 	mbox_status = le16_to_cpu(mbox_rsp->hdr.status);
3281 	change_status = (le32_to_cpu(wr_obj_rsp->change_status_dword) &
3282 			 RSP_CHANGE_STATUS);
3283 
3284 	if (cb_arg) {
3285 		if (cb_arg->cb) {
3286 			if (!status && mbox_status)
3287 				status = mbox_status;
3288 			cb_arg->cb(status, bytes_written, change_status,
3289 				   cb_arg->arg);
3290 		}
3291 
3292 		kfree(cb_arg);
3293 	}
3294 
3295 	return 0;
3296 }
3297 
3298 int
3299 efct_hw_firmware_write(struct efct_hw *hw, struct efc_dma *dma, u32 size,
3300 		       u32 offset, int last,
3301 		       void (*cb)(int status, u32 bytes_written,
3302 				   u32 change_status, void *arg),
3303 		       void *arg)
3304 {
3305 	int rc = -EIO;
3306 	u8 mbxdata[SLI4_BMBX_SIZE];
3307 	struct efct_hw_fw_wr_cb_arg *cb_arg;
3308 	int noc = 0;
3309 
3310 	cb_arg = kzalloc(sizeof(*cb_arg), GFP_KERNEL);
3311 	if (!cb_arg)
3312 		return -ENOMEM;
3313 
3314 	cb_arg->cb = cb;
3315 	cb_arg->arg = arg;
3316 
3317 	/* Write a portion of a firmware image to the device */
3318 	if (!sli_cmd_common_write_object(&hw->sli, mbxdata,
3319 					 noc, last, size, offset, "/prg/",
3320 					 dma))
3321 		rc = efct_hw_command(hw, mbxdata, EFCT_CMD_NOWAIT,
3322 				     efct_hw_cb_fw_write, cb_arg);
3323 
3324 	if (rc != 0) {
3325 		efc_log_debug(hw->os, "COMMON_WRITE_OBJECT failed\n");
3326 		kfree(cb_arg);
3327 	}
3328 
3329 	return rc;
3330 }
3331 
3332 static int
3333 efct_hw_cb_port_control(struct efct_hw *hw, int status, u8 *mqe,
3334 			void  *arg)
3335 {
3336 	return 0;
3337 }
3338 
3339 int
3340 efct_hw_port_control(struct efct_hw *hw, enum efct_hw_port ctrl,
3341 		     uintptr_t value,
3342 		     void (*cb)(int status, uintptr_t value, void *arg),
3343 		     void *arg)
3344 {
3345 	int rc = -EIO;
3346 	u8 link[SLI4_BMBX_SIZE];
3347 	u32 speed = 0;
3348 	u8 reset_alpa = 0;
3349 
3350 	switch (ctrl) {
3351 	case EFCT_HW_PORT_INIT:
3352 		if (!sli_cmd_config_link(&hw->sli, link))
3353 			rc = efct_hw_command(hw, link, EFCT_CMD_NOWAIT,
3354 					     efct_hw_cb_port_control, NULL);
3355 
3356 		if (rc != 0) {
3357 			efc_log_err(hw->os, "CONFIG_LINK failed\n");
3358 			break;
3359 		}
3360 		speed = hw->config.speed;
3361 		reset_alpa = (u8)(value & 0xff);
3362 
3363 		rc = -EIO;
3364 		if (!sli_cmd_init_link(&hw->sli, link, speed, reset_alpa))
3365 			rc = efct_hw_command(hw, link, EFCT_CMD_NOWAIT,
3366 					     efct_hw_cb_port_control, NULL);
3367 		/* Free buffer on error, since no callback is coming */
3368 		if (rc)
3369 			efc_log_err(hw->os, "INIT_LINK failed\n");
3370 		break;
3371 
3372 	case EFCT_HW_PORT_SHUTDOWN:
3373 		if (!sli_cmd_down_link(&hw->sli, link))
3374 			rc = efct_hw_command(hw, link, EFCT_CMD_NOWAIT,
3375 					     efct_hw_cb_port_control, NULL);
3376 		/* Free buffer on error, since no callback is coming */
3377 		if (rc)
3378 			efc_log_err(hw->os, "DOWN_LINK failed\n");
3379 		break;
3380 
3381 	default:
3382 		efc_log_debug(hw->os, "unhandled control %#x\n", ctrl);
3383 		break;
3384 	}
3385 
3386 	return rc;
3387 }
3388 
3389 void
3390 efct_hw_teardown(struct efct_hw *hw)
3391 {
3392 	u32 i = 0;
3393 	u32 destroy_queues;
3394 	u32 free_memory;
3395 	struct efc_dma *dma;
3396 	struct efct *efct = hw->os;
3397 
3398 	destroy_queues = (hw->state == EFCT_HW_STATE_ACTIVE);
3399 	free_memory = (hw->state != EFCT_HW_STATE_UNINITIALIZED);
3400 
3401 	/* Cancel Sliport Healthcheck */
3402 	if (hw->sliport_healthcheck) {
3403 		hw->sliport_healthcheck = 0;
3404 		efct_hw_config_sli_port_health_check(hw, 0, 0);
3405 	}
3406 
3407 	if (hw->state != EFCT_HW_STATE_QUEUES_ALLOCATED) {
3408 		hw->state = EFCT_HW_STATE_TEARDOWN_IN_PROGRESS;
3409 
3410 		efct_hw_flush(hw);
3411 
3412 		if (list_empty(&hw->cmd_head))
3413 			efc_log_debug(hw->os,
3414 				      "All commands completed on MQ queue\n");
3415 		else
3416 			efc_log_debug(hw->os,
3417 				      "Some cmds still pending on MQ queue\n");
3418 
3419 		/* Cancel any remaining commands */
3420 		efct_hw_command_cancel(hw);
3421 	} else {
3422 		hw->state = EFCT_HW_STATE_TEARDOWN_IN_PROGRESS;
3423 	}
3424 
3425 	dma_free_coherent(&efct->pci->dev,
3426 			  hw->rnode_mem.size, hw->rnode_mem.virt,
3427 			  hw->rnode_mem.phys);
3428 	memset(&hw->rnode_mem, 0, sizeof(struct efc_dma));
3429 
3430 	if (hw->io) {
3431 		for (i = 0; i < hw->config.n_io; i++) {
3432 			if (hw->io[i] && hw->io[i]->sgl &&
3433 			    hw->io[i]->sgl->virt) {
3434 				dma_free_coherent(&efct->pci->dev,
3435 						  hw->io[i]->sgl->size,
3436 						  hw->io[i]->sgl->virt,
3437 						  hw->io[i]->sgl->phys);
3438 			}
3439 			kfree(hw->io[i]);
3440 			hw->io[i] = NULL;
3441 		}
3442 		kfree(hw->io);
3443 		hw->io = NULL;
3444 		kfree(hw->wqe_buffs);
3445 		hw->wqe_buffs = NULL;
3446 	}
3447 
3448 	dma = &hw->xfer_rdy;
3449 	dma_free_coherent(&efct->pci->dev,
3450 			  dma->size, dma->virt, dma->phys);
3451 	memset(dma, 0, sizeof(struct efc_dma));
3452 
3453 	dma = &hw->loop_map;
3454 	dma_free_coherent(&efct->pci->dev,
3455 			  dma->size, dma->virt, dma->phys);
3456 	memset(dma, 0, sizeof(struct efc_dma));
3457 
3458 	for (i = 0; i < hw->wq_count; i++)
3459 		sli_queue_free(&hw->sli, &hw->wq[i], destroy_queues,
3460 			       free_memory);
3461 
3462 	for (i = 0; i < hw->rq_count; i++)
3463 		sli_queue_free(&hw->sli, &hw->rq[i], destroy_queues,
3464 			       free_memory);
3465 
3466 	for (i = 0; i < hw->mq_count; i++)
3467 		sli_queue_free(&hw->sli, &hw->mq[i], destroy_queues,
3468 			       free_memory);
3469 
3470 	for (i = 0; i < hw->cq_count; i++)
3471 		sli_queue_free(&hw->sli, &hw->cq[i], destroy_queues,
3472 			       free_memory);
3473 
3474 	for (i = 0; i < hw->eq_count; i++)
3475 		sli_queue_free(&hw->sli, &hw->eq[i], destroy_queues,
3476 			       free_memory);
3477 
3478 	/* Free rq buffers */
3479 	efct_hw_rx_free(hw);
3480 
3481 	efct_hw_queue_teardown(hw);
3482 
3483 	kfree(hw->wq_cpu_array);
3484 
3485 	sli_teardown(&hw->sli);
3486 
3487 	/* record the fact that the queues are non-functional */
3488 	hw->state = EFCT_HW_STATE_UNINITIALIZED;
3489 
3490 	/* free sequence free pool */
3491 	kfree(hw->seq_pool);
3492 	hw->seq_pool = NULL;
3493 
3494 	/* free hw_wq_callback pool */
3495 	efct_hw_reqtag_pool_free(hw);
3496 
3497 	mempool_destroy(hw->cmd_ctx_pool);
3498 	mempool_destroy(hw->mbox_rqst_pool);
3499 
3500 	/* Mark HW setup as not having been called */
3501 	hw->hw_setup_called = false;
3502 }
3503 
3504 static int
3505 efct_hw_sli_reset(struct efct_hw *hw, enum efct_hw_reset reset,
3506 		  enum efct_hw_state prev_state)
3507 {
3508 	int rc = 0;
3509 
3510 	switch (reset) {
3511 	case EFCT_HW_RESET_FUNCTION:
3512 		efc_log_debug(hw->os, "issuing function level reset\n");
3513 		if (sli_reset(&hw->sli)) {
3514 			efc_log_err(hw->os, "sli_reset failed\n");
3515 			rc = -EIO;
3516 		}
3517 		break;
3518 	case EFCT_HW_RESET_FIRMWARE:
3519 		efc_log_debug(hw->os, "issuing firmware reset\n");
3520 		if (sli_fw_reset(&hw->sli)) {
3521 			efc_log_err(hw->os, "sli_soft_reset failed\n");
3522 			rc = -EIO;
3523 		}
3524 		/*
3525 		 * Because the FW reset leaves the FW in a non-running state,
3526 		 * follow that with a regular reset.
3527 		 */
3528 		efc_log_debug(hw->os, "issuing function level reset\n");
3529 		if (sli_reset(&hw->sli)) {
3530 			efc_log_err(hw->os, "sli_reset failed\n");
3531 			rc = -EIO;
3532 		}
3533 		break;
3534 	default:
3535 		efc_log_err(hw->os, "unknown type - no reset performed\n");
3536 		hw->state = prev_state;
3537 		rc = -EINVAL;
3538 		break;
3539 	}
3540 
3541 	return rc;
3542 }
3543 
3544 int
3545 efct_hw_reset(struct efct_hw *hw, enum efct_hw_reset reset)
3546 {
3547 	int rc = 0;
3548 	enum efct_hw_state prev_state = hw->state;
3549 
3550 	if (hw->state != EFCT_HW_STATE_ACTIVE)
3551 		efc_log_debug(hw->os,
3552 			      "HW state %d is not active\n", hw->state);
3553 
3554 	hw->state = EFCT_HW_STATE_RESET_IN_PROGRESS;
3555 
3556 	/*
3557 	 * If the prev_state is already reset/teardown in progress,
3558 	 * don't continue further
3559 	 */
3560 	if (prev_state == EFCT_HW_STATE_RESET_IN_PROGRESS ||
3561 	    prev_state == EFCT_HW_STATE_TEARDOWN_IN_PROGRESS)
3562 		return efct_hw_sli_reset(hw, reset, prev_state);
3563 
3564 	if (prev_state != EFCT_HW_STATE_UNINITIALIZED) {
3565 		efct_hw_flush(hw);
3566 
3567 		if (list_empty(&hw->cmd_head))
3568 			efc_log_debug(hw->os,
3569 				      "All commands completed on MQ queue\n");
3570 		else
3571 			efc_log_err(hw->os,
3572 				    "Some commands still pending on MQ queue\n");
3573 	}
3574 
3575 	/* Reset the chip */
3576 	rc = efct_hw_sli_reset(hw, reset, prev_state);
3577 	if (rc == -EINVAL)
3578 		return -EIO;
3579 
3580 	return rc;
3581 }
3582