xref: /linux/drivers/scsi/elx/efct/efct_hw.c (revision fd7d598270724cc787982ea48bbe17ad383a8b7f)
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 		memcpy(bmbx, cmd, SLI4_BMBX_SIZE);
1406 
1407 		if (sli_bmbx_command(&hw->sli) == 0) {
1408 			rc = 0;
1409 			memcpy(cmd, bmbx, SLI4_BMBX_SIZE);
1410 		}
1411 		mutex_unlock(&hw->bmbx_lock);
1412 	} else if (opts == EFCT_CMD_NOWAIT) {
1413 		struct efct_command_ctx	*ctx = NULL;
1414 
1415 		if (hw->state != EFCT_HW_STATE_ACTIVE) {
1416 			efc_log_err(hw->os, "Can't send command, HW state=%d\n",
1417 				    hw->state);
1418 			return -EIO;
1419 		}
1420 
1421 		ctx = mempool_alloc(hw->cmd_ctx_pool, GFP_ATOMIC);
1422 		if (!ctx)
1423 			return -ENOSPC;
1424 
1425 		memset(ctx, 0, sizeof(struct efct_command_ctx));
1426 
1427 		if (cb) {
1428 			ctx->cb = cb;
1429 			ctx->arg = arg;
1430 		}
1431 
1432 		memcpy(ctx->buf, cmd, SLI4_BMBX_SIZE);
1433 		ctx->ctx = hw;
1434 
1435 		spin_lock_irqsave(&hw->cmd_lock, flags);
1436 
1437 		/* Add to pending list */
1438 		INIT_LIST_HEAD(&ctx->list_entry);
1439 		list_add_tail(&ctx->list_entry, &hw->cmd_pending);
1440 
1441 		/* Submit as much of the pending list as we can */
1442 		rc = efct_hw_cmd_submit_pending(hw);
1443 
1444 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
1445 	}
1446 
1447 	return rc;
1448 }
1449 
1450 static int
1451 efct_hw_command_process(struct efct_hw *hw, int status, u8 *mqe,
1452 			size_t size)
1453 {
1454 	struct efct_command_ctx *ctx = NULL;
1455 	unsigned long flags = 0;
1456 
1457 	spin_lock_irqsave(&hw->cmd_lock, flags);
1458 	if (!list_empty(&hw->cmd_head)) {
1459 		ctx = list_first_entry(&hw->cmd_head,
1460 				       struct efct_command_ctx, list_entry);
1461 		list_del_init(&ctx->list_entry);
1462 	}
1463 	if (!ctx) {
1464 		efc_log_err(hw->os, "no command context\n");
1465 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
1466 		return -EIO;
1467 	}
1468 
1469 	hw->cmd_head_count--;
1470 
1471 	/* Post any pending requests */
1472 	efct_hw_cmd_submit_pending(hw);
1473 
1474 	spin_unlock_irqrestore(&hw->cmd_lock, flags);
1475 
1476 	if (ctx->cb) {
1477 		memcpy(ctx->buf, mqe, size);
1478 		ctx->cb(hw, status, ctx->buf, ctx->arg);
1479 	}
1480 
1481 	mempool_free(ctx, hw->cmd_ctx_pool);
1482 
1483 	return 0;
1484 }
1485 
1486 static int
1487 efct_hw_mq_process(struct efct_hw *hw,
1488 		   int status, struct sli4_queue *mq)
1489 {
1490 	u8 mqe[SLI4_BMBX_SIZE];
1491 	int rc;
1492 
1493 	rc = sli_mq_read(&hw->sli, mq, mqe);
1494 	if (!rc)
1495 		rc = efct_hw_command_process(hw, status, mqe, mq->size);
1496 
1497 	return rc;
1498 }
1499 
1500 static int
1501 efct_hw_command_cancel(struct efct_hw *hw)
1502 {
1503 	unsigned long flags = 0;
1504 	int rc = 0;
1505 
1506 	spin_lock_irqsave(&hw->cmd_lock, flags);
1507 
1508 	/*
1509 	 * Manually clean up remaining commands. Note: since this calls
1510 	 * efct_hw_command_process(), we'll also process the cmd_pending
1511 	 * list, so no need to manually clean that out.
1512 	 */
1513 	while (!list_empty(&hw->cmd_head)) {
1514 		u8		mqe[SLI4_BMBX_SIZE] = { 0 };
1515 		struct efct_command_ctx *ctx;
1516 
1517 		ctx = list_first_entry(&hw->cmd_head,
1518 				       struct efct_command_ctx, list_entry);
1519 
1520 		efc_log_debug(hw->os, "hung command %08x\n",
1521 			      !ctx ? U32_MAX : *((u32 *)ctx->buf));
1522 		spin_unlock_irqrestore(&hw->cmd_lock, flags);
1523 		rc = efct_hw_command_process(hw, -1, mqe, SLI4_BMBX_SIZE);
1524 		spin_lock_irqsave(&hw->cmd_lock, flags);
1525 	}
1526 
1527 	spin_unlock_irqrestore(&hw->cmd_lock, flags);
1528 
1529 	return rc;
1530 }
1531 
1532 static void
1533 efct_mbox_rsp_cb(struct efct_hw *hw, int status, u8 *mqe, void *arg)
1534 {
1535 	struct efct_mbox_rqst_ctx *ctx = arg;
1536 
1537 	if (ctx) {
1538 		if (ctx->callback)
1539 			(*ctx->callback)(hw->os->efcport, status, mqe,
1540 					 ctx->arg);
1541 
1542 		mempool_free(ctx, hw->mbox_rqst_pool);
1543 	}
1544 }
1545 
1546 int
1547 efct_issue_mbox_rqst(void *base, void *cmd, void *cb, void *arg)
1548 {
1549 	struct efct_mbox_rqst_ctx *ctx;
1550 	struct efct *efct = base;
1551 	struct efct_hw *hw = &efct->hw;
1552 	int rc;
1553 
1554 	/*
1555 	 * Allocate a callback context (which includes the mbox cmd buffer),
1556 	 * we need this to be persistent as the mbox cmd submission may be
1557 	 * queued and executed later execution.
1558 	 */
1559 	ctx = mempool_alloc(hw->mbox_rqst_pool, GFP_ATOMIC);
1560 	if (!ctx)
1561 		return -EIO;
1562 
1563 	ctx->callback = cb;
1564 	ctx->arg = arg;
1565 
1566 	rc = efct_hw_command(hw, cmd, EFCT_CMD_NOWAIT, efct_mbox_rsp_cb, ctx);
1567 	if (rc) {
1568 		efc_log_err(efct, "issue mbox rqst failure rc:%d\n", rc);
1569 		mempool_free(ctx, hw->mbox_rqst_pool);
1570 		return -EIO;
1571 	}
1572 
1573 	return 0;
1574 }
1575 
1576 static inline struct efct_hw_io *
1577 _efct_hw_io_alloc(struct efct_hw *hw)
1578 {
1579 	struct efct_hw_io *io = NULL;
1580 
1581 	if (!list_empty(&hw->io_free)) {
1582 		io = list_first_entry(&hw->io_free, struct efct_hw_io,
1583 				      list_entry);
1584 		list_del(&io->list_entry);
1585 	}
1586 	if (io) {
1587 		INIT_LIST_HEAD(&io->list_entry);
1588 		list_add_tail(&io->list_entry, &hw->io_inuse);
1589 		io->state = EFCT_HW_IO_STATE_INUSE;
1590 		io->abort_reqtag = U32_MAX;
1591 		io->wq = hw->wq_cpu_array[raw_smp_processor_id()];
1592 		if (!io->wq) {
1593 			efc_log_err(hw->os, "WQ not assigned for cpu:%d\n",
1594 				    raw_smp_processor_id());
1595 			io->wq = hw->hw_wq[0];
1596 		}
1597 		kref_init(&io->ref);
1598 		io->release = efct_hw_io_free_internal;
1599 	} else {
1600 		atomic_add(1, &hw->io_alloc_failed_count);
1601 	}
1602 
1603 	return io;
1604 }
1605 
1606 struct efct_hw_io *
1607 efct_hw_io_alloc(struct efct_hw *hw)
1608 {
1609 	struct efct_hw_io *io = NULL;
1610 	unsigned long flags = 0;
1611 
1612 	spin_lock_irqsave(&hw->io_lock, flags);
1613 	io = _efct_hw_io_alloc(hw);
1614 	spin_unlock_irqrestore(&hw->io_lock, flags);
1615 
1616 	return io;
1617 }
1618 
1619 static void
1620 efct_hw_io_free_move_correct_list(struct efct_hw *hw,
1621 				  struct efct_hw_io *io)
1622 {
1623 	/*
1624 	 * When an IO is freed, depending on the exchange busy flag,
1625 	 * move it to the correct list.
1626 	 */
1627 	if (io->xbusy) {
1628 		/*
1629 		 * add to wait_free list and wait for XRI_ABORTED CQEs to clean
1630 		 * up
1631 		 */
1632 		INIT_LIST_HEAD(&io->list_entry);
1633 		list_add_tail(&io->list_entry, &hw->io_wait_free);
1634 		io->state = EFCT_HW_IO_STATE_WAIT_FREE;
1635 	} else {
1636 		/* IO not busy, add to free list */
1637 		INIT_LIST_HEAD(&io->list_entry);
1638 		list_add_tail(&io->list_entry, &hw->io_free);
1639 		io->state = EFCT_HW_IO_STATE_FREE;
1640 	}
1641 }
1642 
1643 static inline void
1644 efct_hw_io_free_common(struct efct_hw *hw, struct efct_hw_io *io)
1645 {
1646 	/* initialize IO fields */
1647 	efct_hw_init_free_io(io);
1648 
1649 	/* Restore default SGL */
1650 	efct_hw_io_restore_sgl(hw, io);
1651 }
1652 
1653 void
1654 efct_hw_io_free_internal(struct kref *arg)
1655 {
1656 	unsigned long flags = 0;
1657 	struct efct_hw_io *io =	container_of(arg, struct efct_hw_io, ref);
1658 	struct efct_hw *hw = io->hw;
1659 
1660 	/* perform common cleanup */
1661 	efct_hw_io_free_common(hw, io);
1662 
1663 	spin_lock_irqsave(&hw->io_lock, flags);
1664 	/* remove from in-use list */
1665 	if (!list_empty(&io->list_entry) && !list_empty(&hw->io_inuse)) {
1666 		list_del_init(&io->list_entry);
1667 		efct_hw_io_free_move_correct_list(hw, io);
1668 	}
1669 	spin_unlock_irqrestore(&hw->io_lock, flags);
1670 }
1671 
1672 int
1673 efct_hw_io_free(struct efct_hw *hw, struct efct_hw_io *io)
1674 {
1675 	return kref_put(&io->ref, io->release);
1676 }
1677 
1678 struct efct_hw_io *
1679 efct_hw_io_lookup(struct efct_hw *hw, u32 xri)
1680 {
1681 	u32 ioindex;
1682 
1683 	ioindex = xri - hw->sli.ext[SLI4_RSRC_XRI].base[0];
1684 	return hw->io[ioindex];
1685 }
1686 
1687 int
1688 efct_hw_io_init_sges(struct efct_hw *hw, struct efct_hw_io *io,
1689 		     enum efct_hw_io_type type)
1690 {
1691 	struct sli4_sge	*data = NULL;
1692 	u32 i = 0;
1693 	u32 skips = 0;
1694 	u32 sge_flags = 0;
1695 
1696 	if (!io) {
1697 		efc_log_err(hw->os, "bad parameter hw=%p io=%p\n", hw, io);
1698 		return -EIO;
1699 	}
1700 
1701 	/* Clear / reset the scatter-gather list */
1702 	io->sgl = &io->def_sgl;
1703 	io->sgl_count = io->def_sgl_count;
1704 	io->first_data_sge = 0;
1705 
1706 	memset(io->sgl->virt, 0, 2 * sizeof(struct sli4_sge));
1707 	io->n_sge = 0;
1708 	io->sge_offset = 0;
1709 
1710 	io->type = type;
1711 
1712 	data = io->sgl->virt;
1713 
1714 	/*
1715 	 * Some IO types have underlying hardware requirements on the order
1716 	 * of SGEs. Process all special entries here.
1717 	 */
1718 	switch (type) {
1719 	case EFCT_HW_IO_TARGET_WRITE:
1720 
1721 		/* populate host resident XFER_RDY buffer */
1722 		sge_flags = le32_to_cpu(data->dw2_flags);
1723 		sge_flags &= (~SLI4_SGE_TYPE_MASK);
1724 		sge_flags |= (SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT);
1725 		data->buffer_address_high =
1726 			cpu_to_le32(upper_32_bits(io->xfer_rdy.phys));
1727 		data->buffer_address_low  =
1728 			cpu_to_le32(lower_32_bits(io->xfer_rdy.phys));
1729 		data->buffer_length = cpu_to_le32(io->xfer_rdy.size);
1730 		data->dw2_flags = cpu_to_le32(sge_flags);
1731 		data++;
1732 
1733 		skips = EFCT_TARGET_WRITE_SKIPS;
1734 
1735 		io->n_sge = 1;
1736 		break;
1737 	case EFCT_HW_IO_TARGET_READ:
1738 		/*
1739 		 * For FCP_TSEND64, the first 2 entries are SKIP SGE's
1740 		 */
1741 		skips = EFCT_TARGET_READ_SKIPS;
1742 		break;
1743 	case EFCT_HW_IO_TARGET_RSP:
1744 		/*
1745 		 * No skips, etc. for FCP_TRSP64
1746 		 */
1747 		break;
1748 	default:
1749 		efc_log_err(hw->os, "unsupported IO type %#x\n", type);
1750 		return -EIO;
1751 	}
1752 
1753 	/*
1754 	 * Write skip entries
1755 	 */
1756 	for (i = 0; i < skips; i++) {
1757 		sge_flags = le32_to_cpu(data->dw2_flags);
1758 		sge_flags &= (~SLI4_SGE_TYPE_MASK);
1759 		sge_flags |= (SLI4_SGE_TYPE_SKIP << SLI4_SGE_TYPE_SHIFT);
1760 		data->dw2_flags = cpu_to_le32(sge_flags);
1761 		data++;
1762 	}
1763 
1764 	io->n_sge += skips;
1765 
1766 	/*
1767 	 * Set last
1768 	 */
1769 	sge_flags = le32_to_cpu(data->dw2_flags);
1770 	sge_flags |= SLI4_SGE_LAST;
1771 	data->dw2_flags = cpu_to_le32(sge_flags);
1772 
1773 	return 0;
1774 }
1775 
1776 int
1777 efct_hw_io_add_sge(struct efct_hw *hw, struct efct_hw_io *io,
1778 		   uintptr_t addr, u32 length)
1779 {
1780 	struct sli4_sge	*data = NULL;
1781 	u32 sge_flags = 0;
1782 
1783 	if (!io || !addr || !length) {
1784 		efc_log_err(hw->os,
1785 			    "bad parameter hw=%p io=%p addr=%lx length=%u\n",
1786 			    hw, io, addr, length);
1787 		return -EIO;
1788 	}
1789 
1790 	if (length > hw->sli.sge_supported_length) {
1791 		efc_log_err(hw->os,
1792 			    "length of SGE %d bigger than allowed %d\n",
1793 			    length, hw->sli.sge_supported_length);
1794 		return -EIO;
1795 	}
1796 
1797 	data = io->sgl->virt;
1798 	data += io->n_sge;
1799 
1800 	sge_flags = le32_to_cpu(data->dw2_flags);
1801 	sge_flags &= ~SLI4_SGE_TYPE_MASK;
1802 	sge_flags |= SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT;
1803 	sge_flags &= ~SLI4_SGE_DATA_OFFSET_MASK;
1804 	sge_flags |= SLI4_SGE_DATA_OFFSET_MASK & io->sge_offset;
1805 
1806 	data->buffer_address_high = cpu_to_le32(upper_32_bits(addr));
1807 	data->buffer_address_low  = cpu_to_le32(lower_32_bits(addr));
1808 	data->buffer_length = cpu_to_le32(length);
1809 
1810 	/*
1811 	 * Always assume this is the last entry and mark as such.
1812 	 * If this is not the first entry unset the "last SGE"
1813 	 * indication for the previous entry
1814 	 */
1815 	sge_flags |= SLI4_SGE_LAST;
1816 	data->dw2_flags = cpu_to_le32(sge_flags);
1817 
1818 	if (io->n_sge) {
1819 		sge_flags = le32_to_cpu(data[-1].dw2_flags);
1820 		sge_flags &= ~SLI4_SGE_LAST;
1821 		data[-1].dw2_flags = cpu_to_le32(sge_flags);
1822 	}
1823 
1824 	/* Set first_data_bde if not previously set */
1825 	if (io->first_data_sge == 0)
1826 		io->first_data_sge = io->n_sge;
1827 
1828 	io->sge_offset += length;
1829 	io->n_sge++;
1830 
1831 	return 0;
1832 }
1833 
1834 void
1835 efct_hw_io_abort_all(struct efct_hw *hw)
1836 {
1837 	struct efct_hw_io *io_to_abort	= NULL;
1838 	struct efct_hw_io *next_io = NULL;
1839 
1840 	list_for_each_entry_safe(io_to_abort, next_io,
1841 				 &hw->io_inuse, list_entry) {
1842 		efct_hw_io_abort(hw, io_to_abort, true, NULL, NULL);
1843 	}
1844 }
1845 
1846 static void
1847 efct_hw_wq_process_abort(void *arg, u8 *cqe, int status)
1848 {
1849 	struct efct_hw_io *io = arg;
1850 	struct efct_hw *hw = io->hw;
1851 	u32 ext = 0;
1852 	u32 len = 0;
1853 	struct hw_wq_callback *wqcb;
1854 
1855 	/*
1856 	 * For IOs that were aborted internally, we may need to issue the
1857 	 * callback here depending on whether a XRI_ABORTED CQE is expected ot
1858 	 * not. If the status is Local Reject/No XRI, then
1859 	 * issue the callback now.
1860 	 */
1861 	ext = sli_fc_ext_status(&hw->sli, cqe);
1862 	if (status == SLI4_FC_WCQE_STATUS_LOCAL_REJECT &&
1863 	    ext == SLI4_FC_LOCAL_REJECT_NO_XRI && io->done) {
1864 		efct_hw_done_t done = io->done;
1865 
1866 		io->done = NULL;
1867 
1868 		/*
1869 		 * Use latched status as this is always saved for an internal
1870 		 * abort Note: We won't have both a done and abort_done
1871 		 * function, so don't worry about
1872 		 *       clobbering the len, status and ext fields.
1873 		 */
1874 		status = io->saved_status;
1875 		len = io->saved_len;
1876 		ext = io->saved_ext;
1877 		io->status_saved = false;
1878 		done(io, len, status, ext, io->arg);
1879 	}
1880 
1881 	if (io->abort_done) {
1882 		efct_hw_done_t done = io->abort_done;
1883 
1884 		io->abort_done = NULL;
1885 		done(io, len, status, ext, io->abort_arg);
1886 	}
1887 
1888 	/* clear abort bit to indicate abort is complete */
1889 	io->abort_in_progress = false;
1890 
1891 	/* Free the WQ callback */
1892 	if (io->abort_reqtag == U32_MAX) {
1893 		efc_log_err(hw->os, "HW IO already freed\n");
1894 		return;
1895 	}
1896 
1897 	wqcb = efct_hw_reqtag_get_instance(hw, io->abort_reqtag);
1898 	efct_hw_reqtag_free(hw, wqcb);
1899 
1900 	/*
1901 	 * Call efct_hw_io_free() because this releases the WQ reservation as
1902 	 * well as doing the refcount put. Don't duplicate the code here.
1903 	 */
1904 	(void)efct_hw_io_free(hw, io);
1905 }
1906 
1907 static void
1908 efct_hw_fill_abort_wqe(struct efct_hw *hw, struct efct_hw_wqe *wqe)
1909 {
1910 	struct sli4_abort_wqe *abort = (void *)wqe->wqebuf;
1911 
1912 	memset(abort, 0, hw->sli.wqe_size);
1913 
1914 	abort->criteria = SLI4_ABORT_CRITERIA_XRI_TAG;
1915 	abort->ia_ir_byte |= wqe->send_abts ? 0 : 1;
1916 
1917 	/* Suppress ABTS retries */
1918 	abort->ia_ir_byte |= SLI4_ABRT_WQE_IR;
1919 
1920 	abort->t_tag  = cpu_to_le32(wqe->id);
1921 	abort->command = SLI4_WQE_ABORT;
1922 	abort->request_tag = cpu_to_le16(wqe->abort_reqtag);
1923 
1924 	abort->dw10w0_flags = cpu_to_le16(SLI4_ABRT_WQE_QOSD);
1925 
1926 	abort->cq_id = cpu_to_le16(SLI4_CQ_DEFAULT);
1927 }
1928 
1929 int
1930 efct_hw_io_abort(struct efct_hw *hw, struct efct_hw_io *io_to_abort,
1931 		 bool send_abts, void *cb, void *arg)
1932 {
1933 	struct hw_wq_callback *wqcb;
1934 	unsigned long flags = 0;
1935 
1936 	if (!io_to_abort) {
1937 		efc_log_err(hw->os, "bad parameter hw=%p io=%p\n",
1938 			    hw, io_to_abort);
1939 		return -EIO;
1940 	}
1941 
1942 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
1943 		efc_log_err(hw->os, "cannot send IO abort, HW state=%d\n",
1944 			    hw->state);
1945 		return -EIO;
1946 	}
1947 
1948 	/* take a reference on IO being aborted */
1949 	if (kref_get_unless_zero(&io_to_abort->ref) == 0) {
1950 		/* command no longer active */
1951 		efc_log_debug(hw->os,
1952 			      "io not active xri=0x%x tag=0x%x\n",
1953 			      io_to_abort->indicator, io_to_abort->reqtag);
1954 		return -ENOENT;
1955 	}
1956 
1957 	/* Must have a valid WQ reference */
1958 	if (!io_to_abort->wq) {
1959 		efc_log_debug(hw->os, "io_to_abort xri=0x%x not active on WQ\n",
1960 			      io_to_abort->indicator);
1961 		/* efct_ref_get(): same function */
1962 		kref_put(&io_to_abort->ref, io_to_abort->release);
1963 		return -ENOENT;
1964 	}
1965 
1966 	/*
1967 	 * Validation checks complete; now check to see if already being
1968 	 * aborted, if not set the flag.
1969 	 */
1970 	if (cmpxchg(&io_to_abort->abort_in_progress, false, true)) {
1971 		/* efct_ref_get(): same function */
1972 		kref_put(&io_to_abort->ref, io_to_abort->release);
1973 		efc_log_debug(hw->os,
1974 			      "io already being aborted xri=0x%x tag=0x%x\n",
1975 			      io_to_abort->indicator, io_to_abort->reqtag);
1976 		return -EINPROGRESS;
1977 	}
1978 
1979 	/*
1980 	 * If we got here, the possibilities are:
1981 	 * - host owned xri
1982 	 *	- io_to_abort->wq_index != U32_MAX
1983 	 *		- submit ABORT_WQE to same WQ
1984 	 * - port owned xri:
1985 	 *	- rxri: io_to_abort->wq_index == U32_MAX
1986 	 *		- submit ABORT_WQE to any WQ
1987 	 *	- non-rxri
1988 	 *		- io_to_abort->index != U32_MAX
1989 	 *			- submit ABORT_WQE to same WQ
1990 	 *		- io_to_abort->index == U32_MAX
1991 	 *			- submit ABORT_WQE to any WQ
1992 	 */
1993 	io_to_abort->abort_done = cb;
1994 	io_to_abort->abort_arg  = arg;
1995 
1996 	/* Allocate a request tag for the abort portion of this IO */
1997 	wqcb = efct_hw_reqtag_alloc(hw, efct_hw_wq_process_abort, io_to_abort);
1998 	if (!wqcb) {
1999 		efc_log_err(hw->os, "can't allocate request tag\n");
2000 		return -ENOSPC;
2001 	}
2002 
2003 	io_to_abort->abort_reqtag = wqcb->instance_index;
2004 	io_to_abort->wqe.send_abts = send_abts;
2005 	io_to_abort->wqe.id = io_to_abort->indicator;
2006 	io_to_abort->wqe.abort_reqtag = io_to_abort->abort_reqtag;
2007 
2008 	/*
2009 	 * If the wqe is on the pending list, then set this wqe to be
2010 	 * aborted when the IO's wqe is removed from the list.
2011 	 */
2012 	if (io_to_abort->wq) {
2013 		spin_lock_irqsave(&io_to_abort->wq->queue->lock, flags);
2014 		if (io_to_abort->wqe.list_entry.next) {
2015 			io_to_abort->wqe.abort_wqe_submit_needed = true;
2016 			spin_unlock_irqrestore(&io_to_abort->wq->queue->lock,
2017 					       flags);
2018 			return 0;
2019 		}
2020 		spin_unlock_irqrestore(&io_to_abort->wq->queue->lock, flags);
2021 	}
2022 
2023 	efct_hw_fill_abort_wqe(hw, &io_to_abort->wqe);
2024 
2025 	/* ABORT_WQE does not actually utilize an XRI on the Port,
2026 	 * therefore, keep xbusy as-is to track the exchange's state,
2027 	 * not the ABORT_WQE's state
2028 	 */
2029 	if (efct_hw_wq_write(io_to_abort->wq, &io_to_abort->wqe)) {
2030 		io_to_abort->abort_in_progress = false;
2031 		/* efct_ref_get(): same function */
2032 		kref_put(&io_to_abort->ref, io_to_abort->release);
2033 		return -EIO;
2034 	}
2035 
2036 	return 0;
2037 }
2038 
2039 void
2040 efct_hw_reqtag_pool_free(struct efct_hw *hw)
2041 {
2042 	u32 i;
2043 	struct reqtag_pool *reqtag_pool = hw->wq_reqtag_pool;
2044 	struct hw_wq_callback *wqcb = NULL;
2045 
2046 	if (reqtag_pool) {
2047 		for (i = 0; i < U16_MAX; i++) {
2048 			wqcb = reqtag_pool->tags[i];
2049 			if (!wqcb)
2050 				continue;
2051 
2052 			kfree(wqcb);
2053 		}
2054 		kfree(reqtag_pool);
2055 		hw->wq_reqtag_pool = NULL;
2056 	}
2057 }
2058 
2059 struct reqtag_pool *
2060 efct_hw_reqtag_pool_alloc(struct efct_hw *hw)
2061 {
2062 	u32 i = 0;
2063 	struct reqtag_pool *reqtag_pool;
2064 	struct hw_wq_callback *wqcb;
2065 
2066 	reqtag_pool = kzalloc(sizeof(*reqtag_pool), GFP_KERNEL);
2067 	if (!reqtag_pool)
2068 		return NULL;
2069 
2070 	INIT_LIST_HEAD(&reqtag_pool->freelist);
2071 	/* initialize reqtag pool lock */
2072 	spin_lock_init(&reqtag_pool->lock);
2073 	for (i = 0; i < U16_MAX; i++) {
2074 		wqcb = kmalloc(sizeof(*wqcb), GFP_KERNEL);
2075 		if (!wqcb)
2076 			break;
2077 
2078 		reqtag_pool->tags[i] = wqcb;
2079 		wqcb->instance_index = i;
2080 		wqcb->callback = NULL;
2081 		wqcb->arg = NULL;
2082 		INIT_LIST_HEAD(&wqcb->list_entry);
2083 		list_add_tail(&wqcb->list_entry, &reqtag_pool->freelist);
2084 	}
2085 
2086 	return reqtag_pool;
2087 }
2088 
2089 struct hw_wq_callback *
2090 efct_hw_reqtag_alloc(struct efct_hw *hw,
2091 		     void (*callback)(void *arg, u8 *cqe, int status),
2092 		     void *arg)
2093 {
2094 	struct hw_wq_callback *wqcb = NULL;
2095 	struct reqtag_pool *reqtag_pool = hw->wq_reqtag_pool;
2096 	unsigned long flags = 0;
2097 
2098 	if (!callback)
2099 		return wqcb;
2100 
2101 	spin_lock_irqsave(&reqtag_pool->lock, flags);
2102 
2103 	if (!list_empty(&reqtag_pool->freelist)) {
2104 		wqcb = list_first_entry(&reqtag_pool->freelist,
2105 					struct hw_wq_callback, list_entry);
2106 	}
2107 
2108 	if (wqcb) {
2109 		list_del_init(&wqcb->list_entry);
2110 		spin_unlock_irqrestore(&reqtag_pool->lock, flags);
2111 		wqcb->callback = callback;
2112 		wqcb->arg = arg;
2113 	} else {
2114 		spin_unlock_irqrestore(&reqtag_pool->lock, flags);
2115 	}
2116 
2117 	return wqcb;
2118 }
2119 
2120 void
2121 efct_hw_reqtag_free(struct efct_hw *hw, struct hw_wq_callback *wqcb)
2122 {
2123 	unsigned long flags = 0;
2124 	struct reqtag_pool *reqtag_pool = hw->wq_reqtag_pool;
2125 
2126 	if (!wqcb->callback)
2127 		efc_log_err(hw->os, "WQCB is already freed\n");
2128 
2129 	spin_lock_irqsave(&reqtag_pool->lock, flags);
2130 	wqcb->callback = NULL;
2131 	wqcb->arg = NULL;
2132 	INIT_LIST_HEAD(&wqcb->list_entry);
2133 	list_add(&wqcb->list_entry, &hw->wq_reqtag_pool->freelist);
2134 	spin_unlock_irqrestore(&reqtag_pool->lock, flags);
2135 }
2136 
2137 struct hw_wq_callback *
2138 efct_hw_reqtag_get_instance(struct efct_hw *hw, u32 instance_index)
2139 {
2140 	struct hw_wq_callback *wqcb;
2141 
2142 	wqcb = hw->wq_reqtag_pool->tags[instance_index];
2143 	if (!wqcb)
2144 		efc_log_err(hw->os, "wqcb for instance %d is null\n",
2145 			    instance_index);
2146 
2147 	return wqcb;
2148 }
2149 
2150 int
2151 efct_hw_queue_hash_find(struct efct_queue_hash *hash, u16 id)
2152 {
2153 	int index = -1;
2154 	int i = id & (EFCT_HW_Q_HASH_SIZE - 1);
2155 
2156 	/*
2157 	 * Since the hash is always bigger than the maximum number of Qs, then
2158 	 * we never have to worry about an infinite loop. We will always find
2159 	 * an unused entry.
2160 	 */
2161 	do {
2162 		if (hash[i].in_use && hash[i].id == id)
2163 			index = hash[i].index;
2164 		else
2165 			i = (i + 1) & (EFCT_HW_Q_HASH_SIZE - 1);
2166 	} while (index == -1 && hash[i].in_use);
2167 
2168 	return index;
2169 }
2170 
2171 int
2172 efct_hw_process(struct efct_hw *hw, u32 vector,
2173 		u32 max_isr_time_msec)
2174 {
2175 	struct hw_eq *eq;
2176 
2177 	/*
2178 	 * The caller should disable interrupts if they wish to prevent us
2179 	 * from processing during a shutdown. The following states are defined:
2180 	 *   EFCT_HW_STATE_UNINITIALIZED - No queues allocated
2181 	 *   EFCT_HW_STATE_QUEUES_ALLOCATED - The state after a chip reset,
2182 	 *                                    queues are cleared.
2183 	 *   EFCT_HW_STATE_ACTIVE - Chip and queues are operational
2184 	 *   EFCT_HW_STATE_RESET_IN_PROGRESS - reset, we still want completions
2185 	 *   EFCT_HW_STATE_TEARDOWN_IN_PROGRESS - We still want mailbox
2186 	 *                                        completions.
2187 	 */
2188 	if (hw->state == EFCT_HW_STATE_UNINITIALIZED)
2189 		return 0;
2190 
2191 	/* Get pointer to struct hw_eq */
2192 	eq = hw->hw_eq[vector];
2193 	if (!eq)
2194 		return 0;
2195 
2196 	eq->use_count++;
2197 
2198 	return efct_hw_eq_process(hw, eq, max_isr_time_msec);
2199 }
2200 
2201 int
2202 efct_hw_eq_process(struct efct_hw *hw, struct hw_eq *eq,
2203 		   u32 max_isr_time_msec)
2204 {
2205 	u8 eqe[sizeof(struct sli4_eqe)] = { 0 };
2206 	u32 tcheck_count;
2207 	u64 tstart;
2208 	u64 telapsed;
2209 	bool done = false;
2210 
2211 	tcheck_count = EFCT_HW_TIMECHECK_ITERATIONS;
2212 	tstart = jiffies_to_msecs(jiffies);
2213 
2214 	while (!done && !sli_eq_read(&hw->sli, eq->queue, eqe)) {
2215 		u16 cq_id = 0;
2216 		int rc;
2217 
2218 		rc = sli_eq_parse(&hw->sli, eqe, &cq_id);
2219 		if (unlikely(rc)) {
2220 			if (rc == SLI4_EQE_STATUS_EQ_FULL) {
2221 				u32 i;
2222 
2223 				/*
2224 				 * Received a sentinel EQE indicating the
2225 				 * EQ is full. Process all CQs
2226 				 */
2227 				for (i = 0; i < hw->cq_count; i++)
2228 					efct_hw_cq_process(hw, hw->hw_cq[i]);
2229 				continue;
2230 			} else {
2231 				return rc;
2232 			}
2233 		} else {
2234 			int index;
2235 
2236 			index  = efct_hw_queue_hash_find(hw->cq_hash, cq_id);
2237 
2238 			if (likely(index >= 0))
2239 				efct_hw_cq_process(hw, hw->hw_cq[index]);
2240 			else
2241 				efc_log_err(hw->os, "bad CQ_ID %#06x\n", cq_id);
2242 		}
2243 
2244 		if (eq->queue->n_posted > eq->queue->posted_limit)
2245 			sli_queue_arm(&hw->sli, eq->queue, false);
2246 
2247 		if (tcheck_count && (--tcheck_count == 0)) {
2248 			tcheck_count = EFCT_HW_TIMECHECK_ITERATIONS;
2249 			telapsed = jiffies_to_msecs(jiffies) - tstart;
2250 			if (telapsed >= max_isr_time_msec)
2251 				done = true;
2252 		}
2253 	}
2254 	sli_queue_eq_arm(&hw->sli, eq->queue, true);
2255 
2256 	return 0;
2257 }
2258 
2259 static int
2260 _efct_hw_wq_write(struct hw_wq *wq, struct efct_hw_wqe *wqe)
2261 {
2262 	int queue_rc;
2263 
2264 	/* Every so often, set the wqec bit to generate comsummed completions */
2265 	if (wq->wqec_count)
2266 		wq->wqec_count--;
2267 
2268 	if (wq->wqec_count == 0) {
2269 		struct sli4_generic_wqe *genwqe = (void *)wqe->wqebuf;
2270 
2271 		genwqe->cmdtype_wqec_byte |= SLI4_GEN_WQE_WQEC;
2272 		wq->wqec_count = wq->wqec_set_count;
2273 	}
2274 
2275 	/* Decrement WQ free count */
2276 	wq->free_count--;
2277 
2278 	queue_rc = sli_wq_write(&wq->hw->sli, wq->queue, wqe->wqebuf);
2279 
2280 	return (queue_rc < 0) ? -EIO : 0;
2281 }
2282 
2283 static void
2284 hw_wq_submit_pending(struct hw_wq *wq, u32 update_free_count)
2285 {
2286 	struct efct_hw_wqe *wqe;
2287 	unsigned long flags = 0;
2288 
2289 	spin_lock_irqsave(&wq->queue->lock, flags);
2290 
2291 	/* Update free count with value passed in */
2292 	wq->free_count += update_free_count;
2293 
2294 	while ((wq->free_count > 0) && (!list_empty(&wq->pending_list))) {
2295 		wqe = list_first_entry(&wq->pending_list,
2296 				       struct efct_hw_wqe, list_entry);
2297 		list_del_init(&wqe->list_entry);
2298 		_efct_hw_wq_write(wq, wqe);
2299 
2300 		if (wqe->abort_wqe_submit_needed) {
2301 			wqe->abort_wqe_submit_needed = false;
2302 			efct_hw_fill_abort_wqe(wq->hw, wqe);
2303 			INIT_LIST_HEAD(&wqe->list_entry);
2304 			list_add_tail(&wqe->list_entry, &wq->pending_list);
2305 			wq->wq_pending_count++;
2306 		}
2307 	}
2308 
2309 	spin_unlock_irqrestore(&wq->queue->lock, flags);
2310 }
2311 
2312 void
2313 efct_hw_cq_process(struct efct_hw *hw, struct hw_cq *cq)
2314 {
2315 	u8 cqe[sizeof(struct sli4_mcqe)];
2316 	u16 rid = U16_MAX;
2317 	/* completion type */
2318 	enum sli4_qentry ctype;
2319 	u32 n_processed = 0;
2320 	u32 tstart, telapsed;
2321 
2322 	tstart = jiffies_to_msecs(jiffies);
2323 
2324 	while (!sli_cq_read(&hw->sli, cq->queue, cqe)) {
2325 		int status;
2326 
2327 		status = sli_cq_parse(&hw->sli, cq->queue, cqe, &ctype, &rid);
2328 		/*
2329 		 * The sign of status is significant. If status is:
2330 		 * == 0 : call completed correctly and
2331 		 * the CQE indicated success
2332 		 * > 0 : call completed correctly and
2333 		 * the CQE indicated an error
2334 		 * < 0 : call failed and no information is available about the
2335 		 * CQE
2336 		 */
2337 		if (status < 0) {
2338 			if (status == SLI4_MCQE_STATUS_NOT_COMPLETED)
2339 				/*
2340 				 * Notification that an entry was consumed,
2341 				 * but not completed
2342 				 */
2343 				continue;
2344 
2345 			break;
2346 		}
2347 
2348 		switch (ctype) {
2349 		case SLI4_QENTRY_ASYNC:
2350 			sli_cqe_async(&hw->sli, cqe);
2351 			break;
2352 		case SLI4_QENTRY_MQ:
2353 			/*
2354 			 * Process MQ entry. Note there is no way to determine
2355 			 * the MQ_ID from the completion entry.
2356 			 */
2357 			efct_hw_mq_process(hw, status, hw->mq);
2358 			break;
2359 		case SLI4_QENTRY_WQ:
2360 			efct_hw_wq_process(hw, cq, cqe, status, rid);
2361 			break;
2362 		case SLI4_QENTRY_WQ_RELEASE: {
2363 			u32 wq_id = rid;
2364 			int index;
2365 			struct hw_wq *wq = NULL;
2366 
2367 			index = efct_hw_queue_hash_find(hw->wq_hash, wq_id);
2368 
2369 			if (likely(index >= 0)) {
2370 				wq = hw->hw_wq[index];
2371 			} else {
2372 				efc_log_err(hw->os, "bad WQ_ID %#06x\n", wq_id);
2373 				break;
2374 			}
2375 			/* Submit any HW IOs that are on the WQ pending list */
2376 			hw_wq_submit_pending(wq, wq->wqec_set_count);
2377 
2378 			break;
2379 		}
2380 
2381 		case SLI4_QENTRY_RQ:
2382 			efct_hw_rqpair_process_rq(hw, cq, cqe);
2383 			break;
2384 		case SLI4_QENTRY_XABT: {
2385 			efct_hw_xabt_process(hw, cq, cqe, rid);
2386 			break;
2387 		}
2388 		default:
2389 			efc_log_debug(hw->os, "unhandled ctype=%#x rid=%#x\n",
2390 				      ctype, rid);
2391 			break;
2392 		}
2393 
2394 		n_processed++;
2395 		if (n_processed == cq->queue->proc_limit)
2396 			break;
2397 
2398 		if (cq->queue->n_posted >= cq->queue->posted_limit)
2399 			sli_queue_arm(&hw->sli, cq->queue, false);
2400 	}
2401 
2402 	sli_queue_arm(&hw->sli, cq->queue, true);
2403 
2404 	if (n_processed > cq->queue->max_num_processed)
2405 		cq->queue->max_num_processed = n_processed;
2406 	telapsed = jiffies_to_msecs(jiffies) - tstart;
2407 	if (telapsed > cq->queue->max_process_time)
2408 		cq->queue->max_process_time = telapsed;
2409 }
2410 
2411 void
2412 efct_hw_wq_process(struct efct_hw *hw, struct hw_cq *cq,
2413 		   u8 *cqe, int status, u16 rid)
2414 {
2415 	struct hw_wq_callback *wqcb;
2416 
2417 	if (rid == EFCT_HW_REQUE_XRI_REGTAG) {
2418 		if (status)
2419 			efc_log_err(hw->os, "reque xri failed, status = %d\n",
2420 				    status);
2421 		return;
2422 	}
2423 
2424 	wqcb = efct_hw_reqtag_get_instance(hw, rid);
2425 	if (!wqcb) {
2426 		efc_log_err(hw->os, "invalid request tag: x%x\n", rid);
2427 		return;
2428 	}
2429 
2430 	if (!wqcb->callback) {
2431 		efc_log_err(hw->os, "wqcb callback is NULL\n");
2432 		return;
2433 	}
2434 
2435 	(*wqcb->callback)(wqcb->arg, cqe, status);
2436 }
2437 
2438 void
2439 efct_hw_xabt_process(struct efct_hw *hw, struct hw_cq *cq,
2440 		     u8 *cqe, u16 rid)
2441 {
2442 	/* search IOs wait free list */
2443 	struct efct_hw_io *io = NULL;
2444 	unsigned long flags = 0;
2445 
2446 	io = efct_hw_io_lookup(hw, rid);
2447 	if (!io) {
2448 		/* IO lookup failure should never happen */
2449 		efc_log_err(hw->os, "xabt io lookup failed rid=%#x\n", rid);
2450 		return;
2451 	}
2452 
2453 	if (!io->xbusy)
2454 		efc_log_debug(hw->os, "xabt io not busy rid=%#x\n", rid);
2455 	else
2456 		/* mark IO as no longer busy */
2457 		io->xbusy = false;
2458 
2459 	/*
2460 	 * For IOs that were aborted internally, we need to issue any pending
2461 	 * callback here.
2462 	 */
2463 	if (io->done) {
2464 		efct_hw_done_t done = io->done;
2465 		void		*arg = io->arg;
2466 
2467 		/*
2468 		 * Use latched status as this is always saved for an internal
2469 		 * abort
2470 		 */
2471 		int status = io->saved_status;
2472 		u32 len = io->saved_len;
2473 		u32 ext = io->saved_ext;
2474 
2475 		io->done = NULL;
2476 		io->status_saved = false;
2477 
2478 		done(io, len, status, ext, arg);
2479 	}
2480 
2481 	spin_lock_irqsave(&hw->io_lock, flags);
2482 	if (io->state == EFCT_HW_IO_STATE_INUSE ||
2483 	    io->state == EFCT_HW_IO_STATE_WAIT_FREE) {
2484 		/* if on wait_free list, caller has already freed IO;
2485 		 * remove from wait_free list and add to free list.
2486 		 * if on in-use list, already marked as no longer busy;
2487 		 * just leave there and wait for caller to free.
2488 		 */
2489 		if (io->state == EFCT_HW_IO_STATE_WAIT_FREE) {
2490 			io->state = EFCT_HW_IO_STATE_FREE;
2491 			list_del_init(&io->list_entry);
2492 			efct_hw_io_free_move_correct_list(hw, io);
2493 		}
2494 	}
2495 	spin_unlock_irqrestore(&hw->io_lock, flags);
2496 }
2497 
2498 static int
2499 efct_hw_flush(struct efct_hw *hw)
2500 {
2501 	u32 i = 0;
2502 
2503 	/* Process any remaining completions */
2504 	for (i = 0; i < hw->eq_count; i++)
2505 		efct_hw_process(hw, i, ~0);
2506 
2507 	return 0;
2508 }
2509 
2510 int
2511 efct_hw_wq_write(struct hw_wq *wq, struct efct_hw_wqe *wqe)
2512 {
2513 	int rc = 0;
2514 	unsigned long flags = 0;
2515 
2516 	spin_lock_irqsave(&wq->queue->lock, flags);
2517 	if (list_empty(&wq->pending_list)) {
2518 		if (wq->free_count > 0) {
2519 			rc = _efct_hw_wq_write(wq, wqe);
2520 		} else {
2521 			INIT_LIST_HEAD(&wqe->list_entry);
2522 			list_add_tail(&wqe->list_entry, &wq->pending_list);
2523 			wq->wq_pending_count++;
2524 		}
2525 
2526 		spin_unlock_irqrestore(&wq->queue->lock, flags);
2527 		return rc;
2528 	}
2529 
2530 	INIT_LIST_HEAD(&wqe->list_entry);
2531 	list_add_tail(&wqe->list_entry, &wq->pending_list);
2532 	wq->wq_pending_count++;
2533 	while (wq->free_count > 0) {
2534 		wqe = list_first_entry(&wq->pending_list, struct efct_hw_wqe,
2535 				       list_entry);
2536 		if (!wqe)
2537 			break;
2538 
2539 		list_del_init(&wqe->list_entry);
2540 		rc = _efct_hw_wq_write(wq, wqe);
2541 		if (rc)
2542 			break;
2543 
2544 		if (wqe->abort_wqe_submit_needed) {
2545 			wqe->abort_wqe_submit_needed = false;
2546 			efct_hw_fill_abort_wqe(wq->hw, wqe);
2547 
2548 			INIT_LIST_HEAD(&wqe->list_entry);
2549 			list_add_tail(&wqe->list_entry, &wq->pending_list);
2550 			wq->wq_pending_count++;
2551 		}
2552 	}
2553 
2554 	spin_unlock_irqrestore(&wq->queue->lock, flags);
2555 
2556 	return rc;
2557 }
2558 
2559 int
2560 efct_efc_bls_send(struct efc *efc, u32 type, struct sli_bls_params *bls)
2561 {
2562 	struct efct *efct = efc->base;
2563 
2564 	return efct_hw_bls_send(efct, type, bls, NULL, NULL);
2565 }
2566 
2567 int
2568 efct_hw_bls_send(struct efct *efct, u32 type, struct sli_bls_params *bls_params,
2569 		 void *cb, void *arg)
2570 {
2571 	struct efct_hw *hw = &efct->hw;
2572 	struct efct_hw_io *hio;
2573 	struct sli_bls_payload bls;
2574 	int rc;
2575 
2576 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
2577 		efc_log_err(hw->os,
2578 			    "cannot send BLS, HW state=%d\n", hw->state);
2579 		return -EIO;
2580 	}
2581 
2582 	hio = efct_hw_io_alloc(hw);
2583 	if (!hio) {
2584 		efc_log_err(hw->os, "HIO allocation failed\n");
2585 		return -EIO;
2586 	}
2587 
2588 	hio->done = cb;
2589 	hio->arg  = arg;
2590 
2591 	bls_params->xri = hio->indicator;
2592 	bls_params->tag = hio->reqtag;
2593 
2594 	if (type == FC_RCTL_BA_ACC) {
2595 		hio->type = EFCT_HW_BLS_ACC;
2596 		bls.type = SLI4_SLI_BLS_ACC;
2597 		memcpy(&bls.u.acc, bls_params->payload, sizeof(bls.u.acc));
2598 	} else {
2599 		hio->type = EFCT_HW_BLS_RJT;
2600 		bls.type = SLI4_SLI_BLS_RJT;
2601 		memcpy(&bls.u.rjt, bls_params->payload, sizeof(bls.u.rjt));
2602 	}
2603 
2604 	bls.ox_id = cpu_to_le16(bls_params->ox_id);
2605 	bls.rx_id = cpu_to_le16(bls_params->rx_id);
2606 
2607 	if (sli_xmit_bls_rsp64_wqe(&hw->sli, hio->wqe.wqebuf,
2608 				   &bls, bls_params)) {
2609 		efc_log_err(hw->os, "XMIT_BLS_RSP64 WQE error\n");
2610 		return -EIO;
2611 	}
2612 
2613 	hio->xbusy = true;
2614 
2615 	/*
2616 	 * Add IO to active io wqe list before submitting, in case the
2617 	 * wcqe processing preempts this thread.
2618 	 */
2619 	hio->wq->use_count++;
2620 	rc = efct_hw_wq_write(hio->wq, &hio->wqe);
2621 	if (rc >= 0) {
2622 		/* non-negative return is success */
2623 		rc = 0;
2624 	} else {
2625 		/* failed to write wqe, remove from active wqe list */
2626 		efc_log_err(hw->os,
2627 			    "sli_queue_write failed: %d\n", rc);
2628 		hio->xbusy = false;
2629 	}
2630 
2631 	return rc;
2632 }
2633 
2634 static int
2635 efct_els_ssrs_send_cb(struct efct_hw_io *hio, u32 length, int status,
2636 		      u32 ext_status, void *arg)
2637 {
2638 	struct efc_disc_io *io = arg;
2639 
2640 	efc_disc_io_complete(io, length, status, ext_status);
2641 	return 0;
2642 }
2643 
2644 static inline void
2645 efct_fill_els_params(struct efc_disc_io *io, struct sli_els_params *params)
2646 {
2647 	u8 *cmd = io->req.virt;
2648 
2649 	params->cmd = *cmd;
2650 	params->s_id = io->s_id;
2651 	params->d_id = io->d_id;
2652 	params->ox_id = io->iparam.els.ox_id;
2653 	params->rpi = io->rpi;
2654 	params->vpi = io->vpi;
2655 	params->rpi_registered = io->rpi_registered;
2656 	params->xmit_len = io->xmit_len;
2657 	params->rsp_len = io->rsp_len;
2658 	params->timeout = io->iparam.els.timeout;
2659 }
2660 
2661 static inline void
2662 efct_fill_ct_params(struct efc_disc_io *io, struct sli_ct_params *params)
2663 {
2664 	params->r_ctl = io->iparam.ct.r_ctl;
2665 	params->type = io->iparam.ct.type;
2666 	params->df_ctl =  io->iparam.ct.df_ctl;
2667 	params->d_id = io->d_id;
2668 	params->ox_id = io->iparam.ct.ox_id;
2669 	params->rpi = io->rpi;
2670 	params->vpi = io->vpi;
2671 	params->rpi_registered = io->rpi_registered;
2672 	params->xmit_len = io->xmit_len;
2673 	params->rsp_len = io->rsp_len;
2674 	params->timeout = io->iparam.ct.timeout;
2675 }
2676 
2677 /**
2678  * efct_els_hw_srrs_send() - Send a single request and response cmd.
2679  * @efc: efc library structure
2680  * @io: Discovery IO used to hold els and ct cmd context.
2681  *
2682  * This routine supports communication sequences consisting of a single
2683  * request and single response between two endpoints. Examples include:
2684  *  - Sending an ELS request.
2685  *  - Sending an ELS response - To send an ELS response, the caller must provide
2686  * the OX_ID from the received request.
2687  *  - Sending a FC Common Transport (FC-CT) request - To send a FC-CT request,
2688  * the caller must provide the R_CTL, TYPE, and DF_CTL
2689  * values to place in the FC frame header.
2690  *
2691  * Return: Status of the request.
2692  */
2693 int
2694 efct_els_hw_srrs_send(struct efc *efc, struct efc_disc_io *io)
2695 {
2696 	struct efct *efct = efc->base;
2697 	struct efct_hw_io *hio;
2698 	struct efct_hw *hw = &efct->hw;
2699 	struct efc_dma *send = &io->req;
2700 	struct efc_dma *receive = &io->rsp;
2701 	struct sli4_sge	*sge = NULL;
2702 	int rc = 0;
2703 	u32 len = io->xmit_len;
2704 	u32 sge0_flags;
2705 	u32 sge1_flags;
2706 
2707 	hio = efct_hw_io_alloc(hw);
2708 	if (!hio) {
2709 		pr_err("HIO alloc failed\n");
2710 		return -EIO;
2711 	}
2712 
2713 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
2714 		efc_log_debug(hw->os,
2715 			      "cannot send SRRS, HW state=%d\n", hw->state);
2716 		return -EIO;
2717 	}
2718 
2719 	hio->done = efct_els_ssrs_send_cb;
2720 	hio->arg  = io;
2721 
2722 	sge = hio->sgl->virt;
2723 
2724 	/* clear both SGE */
2725 	memset(hio->sgl->virt, 0, 2 * sizeof(struct sli4_sge));
2726 
2727 	sge0_flags = le32_to_cpu(sge[0].dw2_flags);
2728 	sge1_flags = le32_to_cpu(sge[1].dw2_flags);
2729 	if (send->size) {
2730 		sge[0].buffer_address_high =
2731 			cpu_to_le32(upper_32_bits(send->phys));
2732 		sge[0].buffer_address_low  =
2733 			cpu_to_le32(lower_32_bits(send->phys));
2734 
2735 		sge0_flags |= (SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT);
2736 
2737 		sge[0].buffer_length = cpu_to_le32(len);
2738 	}
2739 
2740 	if (io->io_type == EFC_DISC_IO_ELS_REQ ||
2741 	    io->io_type == EFC_DISC_IO_CT_REQ) {
2742 		sge[1].buffer_address_high =
2743 			cpu_to_le32(upper_32_bits(receive->phys));
2744 		sge[1].buffer_address_low  =
2745 			cpu_to_le32(lower_32_bits(receive->phys));
2746 
2747 		sge1_flags |= (SLI4_SGE_TYPE_DATA << SLI4_SGE_TYPE_SHIFT);
2748 		sge1_flags |= SLI4_SGE_LAST;
2749 
2750 		sge[1].buffer_length = cpu_to_le32(receive->size);
2751 	} else {
2752 		sge0_flags |= SLI4_SGE_LAST;
2753 	}
2754 
2755 	sge[0].dw2_flags = cpu_to_le32(sge0_flags);
2756 	sge[1].dw2_flags = cpu_to_le32(sge1_flags);
2757 
2758 	switch (io->io_type) {
2759 	case EFC_DISC_IO_ELS_REQ: {
2760 		struct sli_els_params els_params;
2761 
2762 		hio->type = EFCT_HW_ELS_REQ;
2763 		efct_fill_els_params(io, &els_params);
2764 		els_params.xri = hio->indicator;
2765 		els_params.tag = hio->reqtag;
2766 
2767 		if (sli_els_request64_wqe(&hw->sli, hio->wqe.wqebuf, hio->sgl,
2768 					  &els_params)) {
2769 			efc_log_err(hw->os, "REQ WQE error\n");
2770 			rc = -EIO;
2771 		}
2772 		break;
2773 	}
2774 	case EFC_DISC_IO_ELS_RESP: {
2775 		struct sli_els_params els_params;
2776 
2777 		hio->type = EFCT_HW_ELS_RSP;
2778 		efct_fill_els_params(io, &els_params);
2779 		els_params.xri = hio->indicator;
2780 		els_params.tag = hio->reqtag;
2781 		if (sli_xmit_els_rsp64_wqe(&hw->sli, hio->wqe.wqebuf, send,
2782 					   &els_params)){
2783 			efc_log_err(hw->os, "RSP WQE error\n");
2784 			rc = -EIO;
2785 		}
2786 		break;
2787 	}
2788 	case EFC_DISC_IO_CT_REQ: {
2789 		struct sli_ct_params ct_params;
2790 
2791 		hio->type = EFCT_HW_FC_CT;
2792 		efct_fill_ct_params(io, &ct_params);
2793 		ct_params.xri = hio->indicator;
2794 		ct_params.tag = hio->reqtag;
2795 		if (sli_gen_request64_wqe(&hw->sli, hio->wqe.wqebuf, hio->sgl,
2796 					  &ct_params)){
2797 			efc_log_err(hw->os, "GEN WQE error\n");
2798 			rc = -EIO;
2799 		}
2800 		break;
2801 	}
2802 	case EFC_DISC_IO_CT_RESP: {
2803 		struct sli_ct_params ct_params;
2804 
2805 		hio->type = EFCT_HW_FC_CT_RSP;
2806 		efct_fill_ct_params(io, &ct_params);
2807 		ct_params.xri = hio->indicator;
2808 		ct_params.tag = hio->reqtag;
2809 		if (sli_xmit_sequence64_wqe(&hw->sli, hio->wqe.wqebuf, hio->sgl,
2810 					    &ct_params)){
2811 			efc_log_err(hw->os, "XMIT SEQ WQE error\n");
2812 			rc = -EIO;
2813 		}
2814 		break;
2815 	}
2816 	default:
2817 		efc_log_err(hw->os, "bad SRRS type %#x\n", io->io_type);
2818 		rc = -EIO;
2819 	}
2820 
2821 	if (rc == 0) {
2822 		hio->xbusy = true;
2823 
2824 		/*
2825 		 * Add IO to active io wqe list before submitting, in case the
2826 		 * wcqe processing preempts this thread.
2827 		 */
2828 		hio->wq->use_count++;
2829 		rc = efct_hw_wq_write(hio->wq, &hio->wqe);
2830 		if (rc >= 0) {
2831 			/* non-negative return is success */
2832 			rc = 0;
2833 		} else {
2834 			/* failed to write wqe, remove from active wqe list */
2835 			efc_log_err(hw->os,
2836 				    "sli_queue_write failed: %d\n", rc);
2837 			hio->xbusy = false;
2838 		}
2839 	}
2840 
2841 	return rc;
2842 }
2843 
2844 int
2845 efct_hw_io_send(struct efct_hw *hw, enum efct_hw_io_type type,
2846 		struct efct_hw_io *io, union efct_hw_io_param_u *iparam,
2847 		void *cb, void *arg)
2848 {
2849 	int rc = 0;
2850 	bool send_wqe = true;
2851 
2852 	if (!io) {
2853 		pr_err("bad parm hw=%p io=%p\n", hw, io);
2854 		return -EIO;
2855 	}
2856 
2857 	if (hw->state != EFCT_HW_STATE_ACTIVE) {
2858 		efc_log_err(hw->os, "cannot send IO, HW state=%d\n", hw->state);
2859 		return -EIO;
2860 	}
2861 
2862 	/*
2863 	 * Save state needed during later stages
2864 	 */
2865 	io->type  = type;
2866 	io->done  = cb;
2867 	io->arg   = arg;
2868 
2869 	/*
2870 	 * Format the work queue entry used to send the IO
2871 	 */
2872 	switch (type) {
2873 	case EFCT_HW_IO_TARGET_WRITE: {
2874 		u16 *flags = &iparam->fcp_tgt.flags;
2875 		struct fcp_txrdy *xfer = io->xfer_rdy.virt;
2876 
2877 		/*
2878 		 * Fill in the XFER_RDY for IF_TYPE 0 devices
2879 		 */
2880 		xfer->ft_data_ro = cpu_to_be32(iparam->fcp_tgt.offset);
2881 		xfer->ft_burst_len = cpu_to_be32(iparam->fcp_tgt.xmit_len);
2882 
2883 		if (io->xbusy)
2884 			*flags |= SLI4_IO_CONTINUATION;
2885 		else
2886 			*flags &= ~SLI4_IO_CONTINUATION;
2887 		iparam->fcp_tgt.xri = io->indicator;
2888 		iparam->fcp_tgt.tag = io->reqtag;
2889 
2890 		if (sli_fcp_treceive64_wqe(&hw->sli, io->wqe.wqebuf,
2891 					   &io->def_sgl, io->first_data_sge,
2892 					   SLI4_CQ_DEFAULT,
2893 					   0, 0, &iparam->fcp_tgt)) {
2894 			efc_log_err(hw->os, "TRECEIVE WQE error\n");
2895 			rc = -EIO;
2896 		}
2897 		break;
2898 	}
2899 	case EFCT_HW_IO_TARGET_READ: {
2900 		u16 *flags = &iparam->fcp_tgt.flags;
2901 
2902 		if (io->xbusy)
2903 			*flags |= SLI4_IO_CONTINUATION;
2904 		else
2905 			*flags &= ~SLI4_IO_CONTINUATION;
2906 
2907 		iparam->fcp_tgt.xri = io->indicator;
2908 		iparam->fcp_tgt.tag = io->reqtag;
2909 
2910 		if (sli_fcp_tsend64_wqe(&hw->sli, io->wqe.wqebuf,
2911 					&io->def_sgl, io->first_data_sge,
2912 					SLI4_CQ_DEFAULT,
2913 					0, 0, &iparam->fcp_tgt)) {
2914 			efc_log_err(hw->os, "TSEND WQE error\n");
2915 			rc = -EIO;
2916 		}
2917 		break;
2918 	}
2919 	case EFCT_HW_IO_TARGET_RSP: {
2920 		u16 *flags = &iparam->fcp_tgt.flags;
2921 
2922 		if (io->xbusy)
2923 			*flags |= SLI4_IO_CONTINUATION;
2924 		else
2925 			*flags &= ~SLI4_IO_CONTINUATION;
2926 
2927 		iparam->fcp_tgt.xri = io->indicator;
2928 		iparam->fcp_tgt.tag = io->reqtag;
2929 
2930 		if (sli_fcp_trsp64_wqe(&hw->sli, io->wqe.wqebuf,
2931 				       &io->def_sgl, SLI4_CQ_DEFAULT,
2932 				       0, &iparam->fcp_tgt)) {
2933 			efc_log_err(hw->os, "TRSP WQE error\n");
2934 			rc = -EIO;
2935 		}
2936 
2937 		break;
2938 	}
2939 	default:
2940 		efc_log_err(hw->os, "unsupported IO type %#x\n", type);
2941 		rc = -EIO;
2942 	}
2943 
2944 	if (send_wqe && rc == 0) {
2945 		io->xbusy = true;
2946 
2947 		/*
2948 		 * Add IO to active io wqe list before submitting, in case the
2949 		 * wcqe processing preempts this thread.
2950 		 */
2951 		hw->tcmd_wq_submit[io->wq->instance]++;
2952 		io->wq->use_count++;
2953 		rc = efct_hw_wq_write(io->wq, &io->wqe);
2954 		if (rc >= 0) {
2955 			/* non-negative return is success */
2956 			rc = 0;
2957 		} else {
2958 			/* failed to write wqe, remove from active wqe list */
2959 			efc_log_err(hw->os,
2960 				    "sli_queue_write failed: %d\n", rc);
2961 			io->xbusy = false;
2962 		}
2963 	}
2964 
2965 	return rc;
2966 }
2967 
2968 int
2969 efct_hw_send_frame(struct efct_hw *hw, struct fc_frame_header *hdr,
2970 		   u8 sof, u8 eof, struct efc_dma *payload,
2971 		   struct efct_hw_send_frame_context *ctx,
2972 		   void (*callback)(void *arg, u8 *cqe, int status),
2973 		   void *arg)
2974 {
2975 	int rc;
2976 	struct efct_hw_wqe *wqe;
2977 	u32 xri;
2978 	struct hw_wq *wq;
2979 
2980 	wqe = &ctx->wqe;
2981 
2982 	/* populate the callback object */
2983 	ctx->hw = hw;
2984 
2985 	/* Fetch and populate request tag */
2986 	ctx->wqcb = efct_hw_reqtag_alloc(hw, callback, arg);
2987 	if (!ctx->wqcb) {
2988 		efc_log_err(hw->os, "can't allocate request tag\n");
2989 		return -ENOSPC;
2990 	}
2991 
2992 	wq = hw->hw_wq[0];
2993 
2994 	/* Set XRI and RX_ID in the header based on which WQ, and which
2995 	 * send_frame_io we are using
2996 	 */
2997 	xri = wq->send_frame_io->indicator;
2998 
2999 	/* Build the send frame WQE */
3000 	rc = sli_send_frame_wqe(&hw->sli, wqe->wqebuf,
3001 				sof, eof, (u32 *)hdr, payload, payload->len,
3002 				EFCT_HW_SEND_FRAME_TIMEOUT, xri,
3003 				ctx->wqcb->instance_index);
3004 	if (rc) {
3005 		efc_log_err(hw->os, "sli_send_frame_wqe failed: %d\n", rc);
3006 		return -EIO;
3007 	}
3008 
3009 	/* Write to WQ */
3010 	rc = efct_hw_wq_write(wq, wqe);
3011 	if (rc) {
3012 		efc_log_err(hw->os, "efct_hw_wq_write failed: %d\n", rc);
3013 		return -EIO;
3014 	}
3015 
3016 	wq->use_count++;
3017 
3018 	return 0;
3019 }
3020 
3021 static int
3022 efct_hw_cb_link_stat(struct efct_hw *hw, int status,
3023 		     u8 *mqe, void  *arg)
3024 {
3025 	struct sli4_cmd_read_link_stats *mbox_rsp;
3026 	struct efct_hw_link_stat_cb_arg *cb_arg = arg;
3027 	struct efct_hw_link_stat_counts counts[EFCT_HW_LINK_STAT_MAX];
3028 	u32 num_counters, i;
3029 	u32 mbox_rsp_flags = 0;
3030 
3031 	mbox_rsp = (struct sli4_cmd_read_link_stats *)mqe;
3032 	mbox_rsp_flags = le32_to_cpu(mbox_rsp->dw1_flags);
3033 	num_counters = (mbox_rsp_flags & SLI4_READ_LNKSTAT_GEC) ? 20 : 13;
3034 	memset(counts, 0, sizeof(struct efct_hw_link_stat_counts) *
3035 				 EFCT_HW_LINK_STAT_MAX);
3036 
3037 	/* Fill overflow counts, mask starts from SLI4_READ_LNKSTAT_W02OF*/
3038 	for (i = 0; i < EFCT_HW_LINK_STAT_MAX; i++)
3039 		counts[i].overflow = (mbox_rsp_flags & (1 << (i + 2)));
3040 
3041 	counts[EFCT_HW_LINK_STAT_LINK_FAILURE_COUNT].counter =
3042 		 le32_to_cpu(mbox_rsp->linkfail_errcnt);
3043 	counts[EFCT_HW_LINK_STAT_LOSS_OF_SYNC_COUNT].counter =
3044 		 le32_to_cpu(mbox_rsp->losssync_errcnt);
3045 	counts[EFCT_HW_LINK_STAT_LOSS_OF_SIGNAL_COUNT].counter =
3046 		 le32_to_cpu(mbox_rsp->losssignal_errcnt);
3047 	counts[EFCT_HW_LINK_STAT_PRIMITIVE_SEQ_COUNT].counter =
3048 		 le32_to_cpu(mbox_rsp->primseq_errcnt);
3049 	counts[EFCT_HW_LINK_STAT_INVALID_XMIT_WORD_COUNT].counter =
3050 		 le32_to_cpu(mbox_rsp->inval_txword_errcnt);
3051 	counts[EFCT_HW_LINK_STAT_CRC_COUNT].counter =
3052 		le32_to_cpu(mbox_rsp->crc_errcnt);
3053 	counts[EFCT_HW_LINK_STAT_PRIMITIVE_SEQ_TIMEOUT_COUNT].counter =
3054 		le32_to_cpu(mbox_rsp->primseq_eventtimeout_cnt);
3055 	counts[EFCT_HW_LINK_STAT_ELASTIC_BUFFER_OVERRUN_COUNT].counter =
3056 		 le32_to_cpu(mbox_rsp->elastic_bufoverrun_errcnt);
3057 	counts[EFCT_HW_LINK_STAT_ARB_TIMEOUT_COUNT].counter =
3058 		 le32_to_cpu(mbox_rsp->arbit_fc_al_timeout_cnt);
3059 	counts[EFCT_HW_LINK_STAT_ADVERTISED_RCV_B2B_CREDIT].counter =
3060 		 le32_to_cpu(mbox_rsp->adv_rx_buftor_to_buf_credit);
3061 	counts[EFCT_HW_LINK_STAT_CURR_RCV_B2B_CREDIT].counter =
3062 		 le32_to_cpu(mbox_rsp->curr_rx_buf_to_buf_credit);
3063 	counts[EFCT_HW_LINK_STAT_ADVERTISED_XMIT_B2B_CREDIT].counter =
3064 		 le32_to_cpu(mbox_rsp->adv_tx_buf_to_buf_credit);
3065 	counts[EFCT_HW_LINK_STAT_CURR_XMIT_B2B_CREDIT].counter =
3066 		 le32_to_cpu(mbox_rsp->curr_tx_buf_to_buf_credit);
3067 	counts[EFCT_HW_LINK_STAT_RCV_EOFA_COUNT].counter =
3068 		 le32_to_cpu(mbox_rsp->rx_eofa_cnt);
3069 	counts[EFCT_HW_LINK_STAT_RCV_EOFDTI_COUNT].counter =
3070 		 le32_to_cpu(mbox_rsp->rx_eofdti_cnt);
3071 	counts[EFCT_HW_LINK_STAT_RCV_EOFNI_COUNT].counter =
3072 		 le32_to_cpu(mbox_rsp->rx_eofni_cnt);
3073 	counts[EFCT_HW_LINK_STAT_RCV_SOFF_COUNT].counter =
3074 		 le32_to_cpu(mbox_rsp->rx_soff_cnt);
3075 	counts[EFCT_HW_LINK_STAT_RCV_DROPPED_NO_AER_COUNT].counter =
3076 		 le32_to_cpu(mbox_rsp->rx_dropped_no_aer_cnt);
3077 	counts[EFCT_HW_LINK_STAT_RCV_DROPPED_NO_RPI_COUNT].counter =
3078 		 le32_to_cpu(mbox_rsp->rx_dropped_no_avail_rpi_rescnt);
3079 	counts[EFCT_HW_LINK_STAT_RCV_DROPPED_NO_XRI_COUNT].counter =
3080 		 le32_to_cpu(mbox_rsp->rx_dropped_no_avail_xri_rescnt);
3081 
3082 	if (cb_arg) {
3083 		if (cb_arg->cb) {
3084 			if (status == 0 && le16_to_cpu(mbox_rsp->hdr.status))
3085 				status = le16_to_cpu(mbox_rsp->hdr.status);
3086 			cb_arg->cb(status, num_counters, counts, cb_arg->arg);
3087 		}
3088 
3089 		kfree(cb_arg);
3090 	}
3091 
3092 	return 0;
3093 }
3094 
3095 int
3096 efct_hw_get_link_stats(struct efct_hw *hw, u8 req_ext_counters,
3097 		       u8 clear_overflow_flags, u8 clear_all_counters,
3098 		       void (*cb)(int status, u32 num_counters,
3099 				  struct efct_hw_link_stat_counts *counters,
3100 				  void *arg),
3101 		       void *arg)
3102 {
3103 	int rc = -EIO;
3104 	struct efct_hw_link_stat_cb_arg *cb_arg;
3105 	u8 mbxdata[SLI4_BMBX_SIZE];
3106 
3107 	cb_arg = kzalloc(sizeof(*cb_arg), GFP_ATOMIC);
3108 	if (!cb_arg)
3109 		return -ENOMEM;
3110 
3111 	cb_arg->cb = cb;
3112 	cb_arg->arg = arg;
3113 
3114 	/* Send the HW command */
3115 	if (!sli_cmd_read_link_stats(&hw->sli, mbxdata, req_ext_counters,
3116 				    clear_overflow_flags, clear_all_counters))
3117 		rc = efct_hw_command(hw, mbxdata, EFCT_CMD_NOWAIT,
3118 				     efct_hw_cb_link_stat, cb_arg);
3119 
3120 	if (rc)
3121 		kfree(cb_arg);
3122 
3123 	return rc;
3124 }
3125 
3126 static int
3127 efct_hw_cb_host_stat(struct efct_hw *hw, int status, u8 *mqe, void  *arg)
3128 {
3129 	struct sli4_cmd_read_status *mbox_rsp =
3130 					(struct sli4_cmd_read_status *)mqe;
3131 	struct efct_hw_host_stat_cb_arg *cb_arg = arg;
3132 	struct efct_hw_host_stat_counts counts[EFCT_HW_HOST_STAT_MAX];
3133 	u32 num_counters = EFCT_HW_HOST_STAT_MAX;
3134 
3135 	memset(counts, 0, sizeof(struct efct_hw_host_stat_counts) *
3136 	       EFCT_HW_HOST_STAT_MAX);
3137 
3138 	counts[EFCT_HW_HOST_STAT_TX_KBYTE_COUNT].counter =
3139 		 le32_to_cpu(mbox_rsp->trans_kbyte_cnt);
3140 	counts[EFCT_HW_HOST_STAT_RX_KBYTE_COUNT].counter =
3141 		 le32_to_cpu(mbox_rsp->recv_kbyte_cnt);
3142 	counts[EFCT_HW_HOST_STAT_TX_FRAME_COUNT].counter =
3143 		 le32_to_cpu(mbox_rsp->trans_frame_cnt);
3144 	counts[EFCT_HW_HOST_STAT_RX_FRAME_COUNT].counter =
3145 		 le32_to_cpu(mbox_rsp->recv_frame_cnt);
3146 	counts[EFCT_HW_HOST_STAT_TX_SEQ_COUNT].counter =
3147 		 le32_to_cpu(mbox_rsp->trans_seq_cnt);
3148 	counts[EFCT_HW_HOST_STAT_RX_SEQ_COUNT].counter =
3149 		 le32_to_cpu(mbox_rsp->recv_seq_cnt);
3150 	counts[EFCT_HW_HOST_STAT_TOTAL_EXCH_ORIG].counter =
3151 		 le32_to_cpu(mbox_rsp->tot_exchanges_orig);
3152 	counts[EFCT_HW_HOST_STAT_TOTAL_EXCH_RESP].counter =
3153 		 le32_to_cpu(mbox_rsp->tot_exchanges_resp);
3154 	counts[EFCT_HW_HOSY_STAT_RX_P_BSY_COUNT].counter =
3155 		 le32_to_cpu(mbox_rsp->recv_p_bsy_cnt);
3156 	counts[EFCT_HW_HOST_STAT_RX_F_BSY_COUNT].counter =
3157 		 le32_to_cpu(mbox_rsp->recv_f_bsy_cnt);
3158 	counts[EFCT_HW_HOST_STAT_DROP_FRM_DUE_TO_NO_RQ_BUF_COUNT].counter =
3159 		 le32_to_cpu(mbox_rsp->no_rq_buf_dropped_frames_cnt);
3160 	counts[EFCT_HW_HOST_STAT_EMPTY_RQ_TIMEOUT_COUNT].counter =
3161 		 le32_to_cpu(mbox_rsp->empty_rq_timeout_cnt);
3162 	counts[EFCT_HW_HOST_STAT_DROP_FRM_DUE_TO_NO_XRI_COUNT].counter =
3163 		 le32_to_cpu(mbox_rsp->no_xri_dropped_frames_cnt);
3164 	counts[EFCT_HW_HOST_STAT_EMPTY_XRI_POOL_COUNT].counter =
3165 		 le32_to_cpu(mbox_rsp->empty_xri_pool_cnt);
3166 
3167 	if (cb_arg) {
3168 		if (cb_arg->cb) {
3169 			if (status == 0 && le16_to_cpu(mbox_rsp->hdr.status))
3170 				status = le16_to_cpu(mbox_rsp->hdr.status);
3171 			cb_arg->cb(status, num_counters, counts, cb_arg->arg);
3172 		}
3173 
3174 		kfree(cb_arg);
3175 	}
3176 
3177 	return 0;
3178 }
3179 
3180 int
3181 efct_hw_get_host_stats(struct efct_hw *hw, u8 cc,
3182 		       void (*cb)(int status, u32 num_counters,
3183 				  struct efct_hw_host_stat_counts *counters,
3184 				  void *arg),
3185 		       void *arg)
3186 {
3187 	int rc = -EIO;
3188 	struct efct_hw_host_stat_cb_arg *cb_arg;
3189 	u8 mbxdata[SLI4_BMBX_SIZE];
3190 
3191 	cb_arg = kmalloc(sizeof(*cb_arg), GFP_ATOMIC);
3192 	if (!cb_arg)
3193 		return -ENOMEM;
3194 
3195 	cb_arg->cb = cb;
3196 	cb_arg->arg = arg;
3197 
3198 	 /* Send the HW command to get the host stats */
3199 	if (!sli_cmd_read_status(&hw->sli, mbxdata, cc))
3200 		rc = efct_hw_command(hw, mbxdata, EFCT_CMD_NOWAIT,
3201 				     efct_hw_cb_host_stat, cb_arg);
3202 
3203 	if (rc) {
3204 		efc_log_debug(hw->os, "READ_HOST_STATS failed\n");
3205 		kfree(cb_arg);
3206 	}
3207 
3208 	return rc;
3209 }
3210 
3211 struct efct_hw_async_call_ctx {
3212 	efct_hw_async_cb_t callback;
3213 	void *arg;
3214 	u8 cmd[SLI4_BMBX_SIZE];
3215 };
3216 
3217 static void
3218 efct_hw_async_cb(struct efct_hw *hw, int status, u8 *mqe, void *arg)
3219 {
3220 	struct efct_hw_async_call_ctx *ctx = arg;
3221 
3222 	if (ctx) {
3223 		if (ctx->callback)
3224 			(*ctx->callback)(hw, status, mqe, ctx->arg);
3225 
3226 		kfree(ctx);
3227 	}
3228 }
3229 
3230 int
3231 efct_hw_async_call(struct efct_hw *hw, efct_hw_async_cb_t callback, void *arg)
3232 {
3233 	struct efct_hw_async_call_ctx *ctx;
3234 	int rc;
3235 
3236 	/*
3237 	 * Allocate a callback context (which includes the mbox cmd buffer),
3238 	 * we need this to be persistent as the mbox cmd submission may be
3239 	 * queued and executed later execution.
3240 	 */
3241 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
3242 	if (!ctx)
3243 		return -ENOMEM;
3244 
3245 	ctx->callback = callback;
3246 	ctx->arg = arg;
3247 
3248 	/* Build and send a NOP mailbox command */
3249 	if (sli_cmd_common_nop(&hw->sli, ctx->cmd, 0)) {
3250 		efc_log_err(hw->os, "COMMON_NOP format failure\n");
3251 		kfree(ctx);
3252 		return -EIO;
3253 	}
3254 
3255 	rc = efct_hw_command(hw, ctx->cmd, EFCT_CMD_NOWAIT, efct_hw_async_cb,
3256 			     ctx);
3257 	if (rc) {
3258 		efc_log_err(hw->os, "COMMON_NOP command failure, rc=%d\n", rc);
3259 		kfree(ctx);
3260 		return -EIO;
3261 	}
3262 	return 0;
3263 }
3264 
3265 static int
3266 efct_hw_cb_fw_write(struct efct_hw *hw, int status, u8 *mqe, void  *arg)
3267 {
3268 	struct sli4_cmd_sli_config *mbox_rsp =
3269 					(struct sli4_cmd_sli_config *)mqe;
3270 	struct sli4_rsp_cmn_write_object *wr_obj_rsp;
3271 	struct efct_hw_fw_wr_cb_arg *cb_arg = arg;
3272 	u32 bytes_written;
3273 	u16 mbox_status;
3274 	u32 change_status;
3275 
3276 	wr_obj_rsp = (struct sli4_rsp_cmn_write_object *)
3277 		      &mbox_rsp->payload.embed;
3278 	bytes_written = le32_to_cpu(wr_obj_rsp->actual_write_length);
3279 	mbox_status = le16_to_cpu(mbox_rsp->hdr.status);
3280 	change_status = (le32_to_cpu(wr_obj_rsp->change_status_dword) &
3281 			 RSP_CHANGE_STATUS);
3282 
3283 	if (cb_arg) {
3284 		if (cb_arg->cb) {
3285 			if (!status && mbox_status)
3286 				status = mbox_status;
3287 			cb_arg->cb(status, bytes_written, change_status,
3288 				   cb_arg->arg);
3289 		}
3290 
3291 		kfree(cb_arg);
3292 	}
3293 
3294 	return 0;
3295 }
3296 
3297 int
3298 efct_hw_firmware_write(struct efct_hw *hw, struct efc_dma *dma, u32 size,
3299 		       u32 offset, int last,
3300 		       void (*cb)(int status, u32 bytes_written,
3301 				   u32 change_status, void *arg),
3302 		       void *arg)
3303 {
3304 	int rc = -EIO;
3305 	u8 mbxdata[SLI4_BMBX_SIZE];
3306 	struct efct_hw_fw_wr_cb_arg *cb_arg;
3307 	int noc = 0;
3308 
3309 	cb_arg = kzalloc(sizeof(*cb_arg), GFP_KERNEL);
3310 	if (!cb_arg)
3311 		return -ENOMEM;
3312 
3313 	cb_arg->cb = cb;
3314 	cb_arg->arg = arg;
3315 
3316 	/* Write a portion of a firmware image to the device */
3317 	if (!sli_cmd_common_write_object(&hw->sli, mbxdata,
3318 					 noc, last, size, offset, "/prg/",
3319 					 dma))
3320 		rc = efct_hw_command(hw, mbxdata, EFCT_CMD_NOWAIT,
3321 				     efct_hw_cb_fw_write, cb_arg);
3322 
3323 	if (rc != 0) {
3324 		efc_log_debug(hw->os, "COMMON_WRITE_OBJECT failed\n");
3325 		kfree(cb_arg);
3326 	}
3327 
3328 	return rc;
3329 }
3330 
3331 static int
3332 efct_hw_cb_port_control(struct efct_hw *hw, int status, u8 *mqe,
3333 			void  *arg)
3334 {
3335 	return 0;
3336 }
3337 
3338 int
3339 efct_hw_port_control(struct efct_hw *hw, enum efct_hw_port ctrl,
3340 		     uintptr_t value,
3341 		     void (*cb)(int status, uintptr_t value, void *arg),
3342 		     void *arg)
3343 {
3344 	int rc = -EIO;
3345 	u8 link[SLI4_BMBX_SIZE];
3346 	u32 speed = 0;
3347 	u8 reset_alpa = 0;
3348 
3349 	switch (ctrl) {
3350 	case EFCT_HW_PORT_INIT:
3351 		if (!sli_cmd_config_link(&hw->sli, link))
3352 			rc = efct_hw_command(hw, link, EFCT_CMD_NOWAIT,
3353 					     efct_hw_cb_port_control, NULL);
3354 
3355 		if (rc != 0) {
3356 			efc_log_err(hw->os, "CONFIG_LINK failed\n");
3357 			break;
3358 		}
3359 		speed = hw->config.speed;
3360 		reset_alpa = (u8)(value & 0xff);
3361 
3362 		rc = -EIO;
3363 		if (!sli_cmd_init_link(&hw->sli, link, speed, reset_alpa))
3364 			rc = efct_hw_command(hw, link, EFCT_CMD_NOWAIT,
3365 					     efct_hw_cb_port_control, NULL);
3366 		/* Free buffer on error, since no callback is coming */
3367 		if (rc)
3368 			efc_log_err(hw->os, "INIT_LINK failed\n");
3369 		break;
3370 
3371 	case EFCT_HW_PORT_SHUTDOWN:
3372 		if (!sli_cmd_down_link(&hw->sli, link))
3373 			rc = efct_hw_command(hw, link, EFCT_CMD_NOWAIT,
3374 					     efct_hw_cb_port_control, NULL);
3375 		/* Free buffer on error, since no callback is coming */
3376 		if (rc)
3377 			efc_log_err(hw->os, "DOWN_LINK failed\n");
3378 		break;
3379 
3380 	default:
3381 		efc_log_debug(hw->os, "unhandled control %#x\n", ctrl);
3382 		break;
3383 	}
3384 
3385 	return rc;
3386 }
3387 
3388 void
3389 efct_hw_teardown(struct efct_hw *hw)
3390 {
3391 	u32 i = 0;
3392 	u32 destroy_queues;
3393 	u32 free_memory;
3394 	struct efc_dma *dma;
3395 	struct efct *efct = hw->os;
3396 
3397 	destroy_queues = (hw->state == EFCT_HW_STATE_ACTIVE);
3398 	free_memory = (hw->state != EFCT_HW_STATE_UNINITIALIZED);
3399 
3400 	/* Cancel Sliport Healthcheck */
3401 	if (hw->sliport_healthcheck) {
3402 		hw->sliport_healthcheck = 0;
3403 		efct_hw_config_sli_port_health_check(hw, 0, 0);
3404 	}
3405 
3406 	if (hw->state != EFCT_HW_STATE_QUEUES_ALLOCATED) {
3407 		hw->state = EFCT_HW_STATE_TEARDOWN_IN_PROGRESS;
3408 
3409 		efct_hw_flush(hw);
3410 
3411 		if (list_empty(&hw->cmd_head))
3412 			efc_log_debug(hw->os,
3413 				      "All commands completed on MQ queue\n");
3414 		else
3415 			efc_log_debug(hw->os,
3416 				      "Some cmds still pending on MQ queue\n");
3417 
3418 		/* Cancel any remaining commands */
3419 		efct_hw_command_cancel(hw);
3420 	} else {
3421 		hw->state = EFCT_HW_STATE_TEARDOWN_IN_PROGRESS;
3422 	}
3423 
3424 	dma_free_coherent(&efct->pci->dev,
3425 			  hw->rnode_mem.size, hw->rnode_mem.virt,
3426 			  hw->rnode_mem.phys);
3427 	memset(&hw->rnode_mem, 0, sizeof(struct efc_dma));
3428 
3429 	if (hw->io) {
3430 		for (i = 0; i < hw->config.n_io; i++) {
3431 			if (hw->io[i] && hw->io[i]->sgl &&
3432 			    hw->io[i]->sgl->virt) {
3433 				dma_free_coherent(&efct->pci->dev,
3434 						  hw->io[i]->sgl->size,
3435 						  hw->io[i]->sgl->virt,
3436 						  hw->io[i]->sgl->phys);
3437 			}
3438 			kfree(hw->io[i]);
3439 			hw->io[i] = NULL;
3440 		}
3441 		kfree(hw->io);
3442 		hw->io = NULL;
3443 		kfree(hw->wqe_buffs);
3444 		hw->wqe_buffs = NULL;
3445 	}
3446 
3447 	dma = &hw->xfer_rdy;
3448 	dma_free_coherent(&efct->pci->dev,
3449 			  dma->size, dma->virt, dma->phys);
3450 	memset(dma, 0, sizeof(struct efc_dma));
3451 
3452 	dma = &hw->loop_map;
3453 	dma_free_coherent(&efct->pci->dev,
3454 			  dma->size, dma->virt, dma->phys);
3455 	memset(dma, 0, sizeof(struct efc_dma));
3456 
3457 	for (i = 0; i < hw->wq_count; i++)
3458 		sli_queue_free(&hw->sli, &hw->wq[i], destroy_queues,
3459 			       free_memory);
3460 
3461 	for (i = 0; i < hw->rq_count; i++)
3462 		sli_queue_free(&hw->sli, &hw->rq[i], destroy_queues,
3463 			       free_memory);
3464 
3465 	for (i = 0; i < hw->mq_count; i++)
3466 		sli_queue_free(&hw->sli, &hw->mq[i], destroy_queues,
3467 			       free_memory);
3468 
3469 	for (i = 0; i < hw->cq_count; i++)
3470 		sli_queue_free(&hw->sli, &hw->cq[i], destroy_queues,
3471 			       free_memory);
3472 
3473 	for (i = 0; i < hw->eq_count; i++)
3474 		sli_queue_free(&hw->sli, &hw->eq[i], destroy_queues,
3475 			       free_memory);
3476 
3477 	/* Free rq buffers */
3478 	efct_hw_rx_free(hw);
3479 
3480 	efct_hw_queue_teardown(hw);
3481 
3482 	kfree(hw->wq_cpu_array);
3483 
3484 	sli_teardown(&hw->sli);
3485 
3486 	/* record the fact that the queues are non-functional */
3487 	hw->state = EFCT_HW_STATE_UNINITIALIZED;
3488 
3489 	/* free sequence free pool */
3490 	kfree(hw->seq_pool);
3491 	hw->seq_pool = NULL;
3492 
3493 	/* free hw_wq_callback pool */
3494 	efct_hw_reqtag_pool_free(hw);
3495 
3496 	mempool_destroy(hw->cmd_ctx_pool);
3497 	mempool_destroy(hw->mbox_rqst_pool);
3498 
3499 	/* Mark HW setup as not having been called */
3500 	hw->hw_setup_called = false;
3501 }
3502 
3503 static int
3504 efct_hw_sli_reset(struct efct_hw *hw, enum efct_hw_reset reset,
3505 		  enum efct_hw_state prev_state)
3506 {
3507 	int rc = 0;
3508 
3509 	switch (reset) {
3510 	case EFCT_HW_RESET_FUNCTION:
3511 		efc_log_debug(hw->os, "issuing function level reset\n");
3512 		if (sli_reset(&hw->sli)) {
3513 			efc_log_err(hw->os, "sli_reset failed\n");
3514 			rc = -EIO;
3515 		}
3516 		break;
3517 	case EFCT_HW_RESET_FIRMWARE:
3518 		efc_log_debug(hw->os, "issuing firmware reset\n");
3519 		if (sli_fw_reset(&hw->sli)) {
3520 			efc_log_err(hw->os, "sli_soft_reset failed\n");
3521 			rc = -EIO;
3522 		}
3523 		/*
3524 		 * Because the FW reset leaves the FW in a non-running state,
3525 		 * follow that with a regular reset.
3526 		 */
3527 		efc_log_debug(hw->os, "issuing function level reset\n");
3528 		if (sli_reset(&hw->sli)) {
3529 			efc_log_err(hw->os, "sli_reset failed\n");
3530 			rc = -EIO;
3531 		}
3532 		break;
3533 	default:
3534 		efc_log_err(hw->os, "unknown type - no reset performed\n");
3535 		hw->state = prev_state;
3536 		rc = -EINVAL;
3537 		break;
3538 	}
3539 
3540 	return rc;
3541 }
3542 
3543 int
3544 efct_hw_reset(struct efct_hw *hw, enum efct_hw_reset reset)
3545 {
3546 	int rc = 0;
3547 	enum efct_hw_state prev_state = hw->state;
3548 
3549 	if (hw->state != EFCT_HW_STATE_ACTIVE)
3550 		efc_log_debug(hw->os,
3551 			      "HW state %d is not active\n", hw->state);
3552 
3553 	hw->state = EFCT_HW_STATE_RESET_IN_PROGRESS;
3554 
3555 	/*
3556 	 * If the prev_state is already reset/teardown in progress,
3557 	 * don't continue further
3558 	 */
3559 	if (prev_state == EFCT_HW_STATE_RESET_IN_PROGRESS ||
3560 	    prev_state == EFCT_HW_STATE_TEARDOWN_IN_PROGRESS)
3561 		return efct_hw_sli_reset(hw, reset, prev_state);
3562 
3563 	if (prev_state != EFCT_HW_STATE_UNINITIALIZED) {
3564 		efct_hw_flush(hw);
3565 
3566 		if (list_empty(&hw->cmd_head))
3567 			efc_log_debug(hw->os,
3568 				      "All commands completed on MQ queue\n");
3569 		else
3570 			efc_log_err(hw->os,
3571 				    "Some commands still pending on MQ queue\n");
3572 	}
3573 
3574 	/* Reset the chip */
3575 	rc = efct_hw_sli_reset(hw, reset, prev_state);
3576 	if (rc == -EINVAL)
3577 		return -EIO;
3578 
3579 	return rc;
3580 }
3581