xref: /freebsd/sys/dev/ocs_fc/ocs_xport.c (revision 7cc42f6d)
1 /*-
2  * Copyright (c) 2017 Broadcom. All rights reserved.
3  * The term "Broadcom" refers to Broadcom Limited and/or its subsidiaries.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions are met:
7  *
8  * 1. Redistributions of source code must retain the above copyright notice,
9  *    this list of conditions and the following disclaimer.
10  *
11  * 2. Redistributions in binary form must reproduce the above copyright notice,
12  *    this list of conditions and the following disclaimer in the documentation
13  *    and/or other materials provided with the distribution.
14  *
15  * 3. Neither the name of the copyright holder nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
23  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  *
31  * $FreeBSD$
32  */
33 
34 /**
35  * @file
36  * FC transport API
37  *
38  */
39 
40 #include "ocs.h"
41 #include "ocs_device.h"
42 
43 static void ocs_xport_link_stats_cb(int32_t status, uint32_t num_counters, ocs_hw_link_stat_counts_t *counters, void *arg);
44 static void ocs_xport_host_stats_cb(int32_t status, uint32_t num_counters, ocs_hw_host_stat_counts_t *counters, void *arg);
45 /**
46  * @brief Post node event callback argument.
47  */
48 typedef struct {
49 	ocs_sem_t sem;
50 	ocs_node_t *node;
51 	ocs_sm_event_t evt;
52 	void *context;
53 } ocs_xport_post_node_event_t;
54 
55 /**
56  * @brief Allocate a transport object.
57  *
58  * @par Description
59  * A transport object is allocated, and associated with a device instance.
60  *
61  * @param ocs Pointer to device instance.
62  *
63  * @return Returns the pointer to the allocated transport object, or NULL if failed.
64  */
65 ocs_xport_t *
66 ocs_xport_alloc(ocs_t *ocs)
67 {
68 	ocs_xport_t *xport;
69 
70 	ocs_assert(ocs, NULL);
71 	xport = ocs_malloc(ocs, sizeof(*xport), OCS_M_ZERO);
72 	if (xport != NULL) {
73 		xport->ocs = ocs;
74 	}
75 	return xport;
76 }
77 
78 /**
79  * @brief Create the RQ threads and the circular buffers used to pass sequences.
80  *
81  * @par Description
82  * Creates the circular buffers and the servicing threads for RQ processing.
83  *
84  * @param xport Pointer to transport object
85  *
86  * @return Returns 0 on success, or a non-zero value on failure.
87  */
88 static void
89 ocs_xport_rq_threads_teardown(ocs_xport_t *xport)
90 {
91 	ocs_t *ocs = xport->ocs;
92 	uint32_t i;
93 
94 	if (xport->num_rq_threads == 0 ||
95 	    xport->rq_thread_info == NULL) {
96 		return;
97 	}
98 
99 	/* Abort any threads */
100 	for (i = 0; i < xport->num_rq_threads; i++) {
101 		if (xport->rq_thread_info[i].thread_started) {
102 			ocs_thread_terminate(&xport->rq_thread_info[i].thread);
103 			/* wait for the thread to exit */
104 			ocs_log_debug(ocs, "wait for thread %d to exit\n", i);
105 			while (xport->rq_thread_info[i].thread_started) {
106 				ocs_udelay(10000);
107 			}
108 			ocs_log_debug(ocs, "thread %d to exited\n", i);
109 		}
110 		if (xport->rq_thread_info[i].seq_cbuf != NULL) {
111 			ocs_cbuf_free(xport->rq_thread_info[i].seq_cbuf);
112 			xport->rq_thread_info[i].seq_cbuf = NULL;
113 		}
114 	}
115 }
116 
117 /**
118  * @brief Create the RQ threads and the circular buffers used to pass sequences.
119  *
120  * @par Description
121  * Creates the circular buffers and the servicing threads for RQ processing.
122  *
123  * @param xport Pointer to transport object.
124  * @param num_rq_threads Number of RQ processing threads that the
125  * driver creates.
126  *
127  * @return Returns 0 on success, or a non-zero value on failure.
128  */
129 static int32_t
130 ocs_xport_rq_threads_create(ocs_xport_t *xport, uint32_t num_rq_threads)
131 {
132 	ocs_t *ocs = xport->ocs;
133 	int32_t rc = 0;
134 	uint32_t i;
135 
136 	xport->num_rq_threads = num_rq_threads;
137 	ocs_log_debug(ocs, "number of RQ threads %d\n", num_rq_threads);
138 	if (num_rq_threads == 0) {
139 		return 0;
140 	}
141 
142 	/* Allocate the space for the thread objects */
143 	xport->rq_thread_info = ocs_malloc(ocs, sizeof(ocs_xport_rq_thread_info_t) * num_rq_threads, OCS_M_ZERO);
144 	if (xport->rq_thread_info == NULL) {
145 		ocs_log_err(ocs, "memory allocation failure\n");
146 		return -1;
147 	}
148 
149 	/* Create the circular buffers and threads. */
150 	for (i = 0; i < num_rq_threads; i++) {
151 		xport->rq_thread_info[i].ocs = ocs;
152 		xport->rq_thread_info[i].seq_cbuf = ocs_cbuf_alloc(ocs, OCS_HW_RQ_NUM_HDR);
153 		if (xport->rq_thread_info[i].seq_cbuf == NULL) {
154 			goto ocs_xport_rq_threads_create_error;
155 		}
156 
157 		ocs_snprintf(xport->rq_thread_info[i].thread_name,
158 			     sizeof(xport->rq_thread_info[i].thread_name),
159 			     "ocs_unsol_rq:%d:%d", ocs->instance_index, i);
160 		rc = ocs_thread_create(ocs, &xport->rq_thread_info[i].thread, ocs_unsol_rq_thread,
161 				       xport->rq_thread_info[i].thread_name,
162 				       &xport->rq_thread_info[i], OCS_THREAD_RUN);
163 		if (rc) {
164 			ocs_log_err(ocs, "ocs_thread_create failed: %d\n", rc);
165 			goto ocs_xport_rq_threads_create_error;
166 		}
167 		xport->rq_thread_info[i].thread_started = TRUE;
168 	}
169 	return 0;
170 
171 ocs_xport_rq_threads_create_error:
172 	ocs_xport_rq_threads_teardown(xport);
173 	return -1;
174 }
175 
176 /**
177  * @brief Do as much allocation as possible, but do not initialization the device.
178  *
179  * @par Description
180  * Performs the functions required to get a device ready to run.
181  *
182  * @param xport Pointer to transport object.
183  *
184  * @return Returns 0 on success, or a non-zero value on failure.
185  */
186 int32_t
187 ocs_xport_attach(ocs_xport_t *xport)
188 {
189 	ocs_t *ocs = xport->ocs;
190 	int32_t rc;
191 	uint32_t max_sgl;
192 	uint32_t n_sgl;
193 	uint32_t i;
194 	uint32_t value;
195 	uint32_t max_remote_nodes;
196 
197 	/* booleans used for cleanup if initialization fails */
198 	uint8_t io_pool_created = FALSE;
199 	uint8_t node_pool_created = FALSE;
200 	uint8_t rq_threads_created = FALSE;
201 
202 	ocs_list_init(&ocs->domain_list, ocs_domain_t, link);
203 
204 	for (i = 0; i < SLI4_MAX_FCFI; i++) {
205 		xport->fcfi[i].hold_frames = 1;
206 		ocs_lock_init(ocs, &xport->fcfi[i].pend_frames_lock, "xport pend_frames[%d]", i);
207 		ocs_list_init(&xport->fcfi[i].pend_frames, ocs_hw_sequence_t, link);
208 	}
209 
210 	rc = ocs_hw_set_ptr(&ocs->hw, OCS_HW_WAR_VERSION, ocs->hw_war_version);
211 	if (rc) {
212 		ocs_log_test(ocs, "can't set OCS_HW_WAR_VERSION\n");
213 		return -1;
214 	}
215 
216 	rc = ocs_hw_setup(&ocs->hw, ocs, SLI4_PORT_TYPE_FC);
217 	if (rc) {
218 		ocs_log_err(ocs, "%s: Can't setup hardware\n", ocs->desc);
219 		return -1;
220 	} else if (ocs->ctrlmask & OCS_CTRLMASK_CRASH_RESET) {
221 		ocs_log_debug(ocs, "stopping after ocs_hw_setup\n");
222 		return -1;
223 	}
224 
225 	ocs_hw_set(&ocs->hw, OCS_HW_BOUNCE, ocs->hw_bounce);
226 	ocs_log_debug(ocs, "HW bounce: %d\n", ocs->hw_bounce);
227 
228 	ocs_hw_set(&ocs->hw, OCS_HW_RQ_SELECTION_POLICY, ocs->rq_selection_policy);
229 	ocs_hw_set(&ocs->hw, OCS_HW_RR_QUANTA, ocs->rr_quanta);
230 	ocs_hw_get(&ocs->hw, OCS_HW_RQ_SELECTION_POLICY, &value);
231 	ocs_log_debug(ocs, "RQ Selection Policy: %d\n", value);
232 
233 	ocs_hw_set_ptr(&ocs->hw, OCS_HW_FILTER_DEF, (void*) ocs->filter_def);
234 
235 	ocs_hw_get(&ocs->hw, OCS_HW_MAX_SGL, &max_sgl);
236 	max_sgl -= SLI4_SGE_MAX_RESERVED;
237 	n_sgl = MIN(OCS_FC_MAX_SGL, max_sgl);
238 
239 	/* EVT: For chained SGL testing */
240 	if (ocs->ctrlmask & OCS_CTRLMASK_TEST_CHAINED_SGLS) {
241 		n_sgl = 4;
242 	}
243 
244 	/* Note: number of SGLs must be set for ocs_node_create_pool */
245 	if (ocs_hw_set(&ocs->hw, OCS_HW_N_SGL, n_sgl) != OCS_HW_RTN_SUCCESS) {
246 		ocs_log_err(ocs, "%s: Can't set number of SGLs\n", ocs->desc);
247 		return -1;
248 	} else {
249 		ocs_log_debug(ocs, "%s: Configured for %d SGLs\n", ocs->desc, n_sgl);
250 	}
251 
252 	ocs_hw_get(&ocs->hw, OCS_HW_MAX_NODES, &max_remote_nodes);
253 
254 	if (!ocs->max_remote_nodes)
255 		ocs->max_remote_nodes = max_remote_nodes;
256 
257 	rc = ocs_node_create_pool(ocs, ocs->max_remote_nodes);
258 	if (rc) {
259 		ocs_log_err(ocs, "Can't allocate node pool\n");
260 		goto ocs_xport_attach_cleanup;
261 	} else {
262 		node_pool_created = TRUE;
263 	}
264 
265 	/* EVT: if testing chained SGLs allocate OCS_FC_MAX_SGL SGE's in the IO */
266 	xport->io_pool = ocs_io_pool_create(ocs, ocs->num_scsi_ios,
267 		(ocs->ctrlmask & OCS_CTRLMASK_TEST_CHAINED_SGLS) ? OCS_FC_MAX_SGL : n_sgl);
268 	if (xport->io_pool == NULL) {
269 		ocs_log_err(ocs, "Can't allocate IO pool\n");
270 		goto ocs_xport_attach_cleanup;
271 	} else {
272 		io_pool_created = TRUE;
273 	}
274 
275 	/*
276 	 * setup the RQ processing threads
277 	 */
278 	if (ocs_xport_rq_threads_create(xport, ocs->rq_threads) != 0) {
279 		ocs_log_err(ocs, "failure creating RQ threads\n");
280 		goto ocs_xport_attach_cleanup;
281 	}
282 	rq_threads_created = TRUE;
283 
284 	return 0;
285 
286 ocs_xport_attach_cleanup:
287 	if (io_pool_created) {
288 		ocs_io_pool_free(xport->io_pool);
289 	}
290 
291 	if (node_pool_created) {
292 		ocs_node_free_pool(ocs);
293 	}
294 
295 	return -1;
296 }
297 
298 /**
299  * @brief Determines how to setup auto Xfer ready.
300  *
301  * @par Description
302  * @param xport Pointer to transport object.
303  *
304  * @return Returns 0 on success or a non-zero value on failure.
305  */
306 static int32_t
307 ocs_xport_initialize_auto_xfer_ready(ocs_xport_t *xport)
308 {
309 	ocs_t *ocs = xport->ocs;
310 	uint32_t auto_xfer_rdy;
311 	char prop_buf[32];
312 	uint32_t ramdisc_blocksize = 512;
313 	uint8_t p_type = 0;
314 
315 	ocs_hw_get(&ocs->hw, OCS_HW_AUTO_XFER_RDY_CAPABLE, &auto_xfer_rdy);
316 	if (!auto_xfer_rdy) {
317 		ocs->auto_xfer_rdy_size = 0;
318 		ocs_log_test(ocs, "Cannot enable auto xfer rdy for this port\n");
319 		return 0;
320 	}
321 
322 	if (ocs_hw_set(&ocs->hw, OCS_HW_AUTO_XFER_RDY_SIZE, ocs->auto_xfer_rdy_size)) {
323 		ocs_log_test(ocs, "%s: Can't set auto xfer rdy mode\n", ocs->desc);
324 		return -1;
325 	}
326 
327 	/*
328 	 * Determine if we are doing protection in the backend. We are looking
329 	 * at the modules parameters here. The backend cannot allow a format
330 	 * command to change the protection mode when using this feature,
331 	 * otherwise the firmware will not do the proper thing.
332 	 */
333 	if (ocs_get_property("p_type", prop_buf, sizeof(prop_buf)) == 0) {
334 		p_type = ocs_strtoul(prop_buf, 0, 0);
335 	}
336 	if (ocs_get_property("ramdisc_blocksize", prop_buf, sizeof(prop_buf)) == 0) {
337 		ramdisc_blocksize = ocs_strtoul(prop_buf, 0, 0);
338 	}
339 	if (ocs_get_property("external_dif", prop_buf, sizeof(prop_buf)) == 0) {
340 		if(ocs_strlen(prop_buf)) {
341 			if (p_type == 0) {
342 				p_type = 1;
343 			}
344 		}
345 	}
346 
347 	if (p_type != 0) {
348 		if (ocs_hw_set(&ocs->hw, OCS_HW_AUTO_XFER_RDY_T10_ENABLE, TRUE)) {
349 			ocs_log_test(ocs, "%s: Can't set auto xfer rdy mode\n", ocs->desc);
350 			return -1;
351 		}
352 		if (ocs_hw_set(&ocs->hw, OCS_HW_AUTO_XFER_RDY_BLK_SIZE, ramdisc_blocksize)) {
353 			ocs_log_test(ocs, "%s: Can't set auto xfer rdy blk size\n", ocs->desc);
354 			return -1;
355 		}
356 		if (ocs_hw_set(&ocs->hw, OCS_HW_AUTO_XFER_RDY_P_TYPE, p_type)) {
357 			ocs_log_test(ocs, "%s: Can't set auto xfer rdy mode\n", ocs->desc);
358 			return -1;
359 		}
360 		if (ocs_hw_set(&ocs->hw, OCS_HW_AUTO_XFER_RDY_REF_TAG_IS_LBA, TRUE)) {
361 			ocs_log_test(ocs, "%s: Can't set auto xfer rdy ref tag\n", ocs->desc);
362 			return -1;
363 		}
364 		if (ocs_hw_set(&ocs->hw, OCS_HW_AUTO_XFER_RDY_APP_TAG_VALID, FALSE)) {
365 			ocs_log_test(ocs, "%s: Can't set auto xfer rdy app tag valid\n", ocs->desc);
366 			return -1;
367 		}
368 	}
369 	ocs_log_debug(ocs, "Auto xfer rdy is enabled, p_type=%d, blksize=%d\n",
370 		p_type, ramdisc_blocksize);
371 	return 0;
372 }
373 
374 /**
375  * @brief Initializes the device.
376  *
377  * @par Description
378  * Performs the functions required to make a device functional.
379  *
380  * @param xport Pointer to transport object.
381  *
382  * @return Returns 0 on success, or a non-zero value on failure.
383  */
384 int32_t
385 ocs_xport_initialize(ocs_xport_t *xport)
386 {
387 	ocs_t *ocs = xport->ocs;
388 	int32_t rc;
389 	uint32_t i;
390 	uint32_t max_hw_io;
391 	uint32_t max_sgl;
392 	uint32_t hlm;
393 	uint32_t rq_limit;
394 	uint32_t dif_capable;
395 	uint8_t dif_separate = 0;
396 	char prop_buf[32];
397 
398 	/* booleans used for cleanup if initialization fails */
399 	uint8_t ini_device_set = FALSE;
400 	uint8_t tgt_device_set = FALSE;
401 	uint8_t hw_initialized = FALSE;
402 
403 	ocs_hw_get(&ocs->hw, OCS_HW_MAX_IO, &max_hw_io);
404 	if (ocs_hw_set(&ocs->hw, OCS_HW_N_IO, max_hw_io) != OCS_HW_RTN_SUCCESS) {
405 		ocs_log_err(ocs, "%s: Can't set number of IOs\n", ocs->desc);
406 		return -1;
407 	}
408 
409 	ocs_hw_get(&ocs->hw, OCS_HW_MAX_SGL, &max_sgl);
410 	max_sgl -= SLI4_SGE_MAX_RESERVED;
411 
412 	if (ocs->enable_hlm) {
413 		ocs_hw_get(&ocs->hw, OCS_HW_HIGH_LOGIN_MODE, &hlm);
414 		if (!hlm) {
415 			ocs->enable_hlm = FALSE;
416 			ocs_log_err(ocs, "Cannot enable high login mode for this port\n");
417 		} else {
418                         ocs_log_debug(ocs, "High login mode is enabled\n");
419 			if (ocs_hw_set(&ocs->hw, OCS_HW_HIGH_LOGIN_MODE, TRUE)) {
420 				ocs_log_err(ocs, "%s: Can't set high login mode\n", ocs->desc);
421 				return -1;
422 			}
423 		}
424 	}
425 
426 	/* validate the auto xfer_rdy size */
427 	if (ocs->auto_xfer_rdy_size > 0 &&
428 	    (ocs->auto_xfer_rdy_size < 2048 ||
429 	     ocs->auto_xfer_rdy_size > 65536)) {
430 		ocs_log_err(ocs, "Auto XFER_RDY size is out of range (2K-64K)\n");
431 		return -1;
432 	}
433 
434 	ocs_hw_get(&ocs->hw, OCS_HW_MAX_IO, &max_hw_io);
435 
436 	if (ocs->auto_xfer_rdy_size > 0) {
437 		if (ocs_xport_initialize_auto_xfer_ready(xport)) {
438 			ocs_log_err(ocs, "%s: Failed auto xfer ready setup\n", ocs->desc);
439 			return -1;
440 		}
441 		if (ocs->esoc){
442 			ocs_hw_set(&ocs->hw, OCS_ESOC, TRUE);
443 		}
444 	}
445 
446 	if (ocs->explicit_buffer_list) {
447 		/* Are pre-registered SGL's required? */
448 		ocs_hw_get(&ocs->hw, OCS_HW_PREREGISTER_SGL, &i);
449 		if (i == TRUE) {
450 			ocs_log_err(ocs, "Explicit Buffer List not supported on this device, not enabled\n");
451 		} else {
452 			ocs_hw_set(&ocs->hw, OCS_HW_PREREGISTER_SGL, FALSE);
453 		}
454 	}
455 
456 	if (ocs_hw_set(&ocs->hw, OCS_HW_TOPOLOGY, ocs->topology) != OCS_HW_RTN_SUCCESS) {
457 		ocs_log_err(ocs, "%s: Can't set the toplogy\n", ocs->desc);
458 		return -1;
459 	}
460 	ocs_hw_set(&ocs->hw, OCS_HW_RQ_DEFAULT_BUFFER_SIZE, OCS_FC_RQ_SIZE_DEFAULT);
461 
462 	if (ocs_hw_set(&ocs->hw, OCS_HW_LINK_SPEED, ocs->speed) != OCS_HW_RTN_SUCCESS) {
463 		ocs_log_err(ocs, "%s: Can't set the link speed\n", ocs->desc);
464 		return -1;
465 	}
466 
467 	if (ocs_hw_set(&ocs->hw, OCS_HW_ETH_LICENSE, ocs->ethernet_license) != OCS_HW_RTN_SUCCESS) {
468 		ocs_log_err(ocs, "%s: Can't set the ethernet license\n", ocs->desc);
469 		return -1;
470 	}
471 
472 	/* currently only lancer support setting the CRC seed value */
473 	if (ocs->hw.sli.asic_type == SLI4_ASIC_TYPE_LANCER) {
474 		if (ocs_hw_set(&ocs->hw, OCS_HW_DIF_SEED, OCS_FC_DIF_SEED) != OCS_HW_RTN_SUCCESS) {
475 			ocs_log_err(ocs, "%s: Can't set the DIF seed\n", ocs->desc);
476 			return -1;
477 		}
478 	}
479 
480 	/* Set the Dif mode */
481 	if (0 == ocs_hw_get(&ocs->hw, OCS_HW_DIF_CAPABLE, &dif_capable)) {
482 		if (dif_capable) {
483 			if (ocs_get_property("dif_separate", prop_buf, sizeof(prop_buf)) == 0) {
484 				dif_separate = ocs_strtoul(prop_buf, 0, 0);
485 			}
486 
487 			if ((rc = ocs_hw_set(&ocs->hw, OCS_HW_DIF_MODE,
488 			      (dif_separate == 0 ? OCS_HW_DIF_MODE_INLINE : OCS_HW_DIF_MODE_SEPARATE)))) {
489 				ocs_log_err(ocs, "Requested DIF MODE not supported\n");
490 			}
491 		}
492 	}
493 
494 	if (ocs->target_io_timer_sec) {
495 		ocs_log_debug(ocs, "setting target io timer=%d\n", ocs->target_io_timer_sec);
496 		ocs_hw_set(&ocs->hw, OCS_HW_EMULATE_TARGET_WQE_TIMEOUT, TRUE);
497 	}
498 
499 	ocs_hw_callback(&ocs->hw, OCS_HW_CB_DOMAIN, ocs_domain_cb, ocs);
500 	ocs_hw_callback(&ocs->hw, OCS_HW_CB_REMOTE_NODE, ocs_remote_node_cb, ocs);
501 	ocs_hw_callback(&ocs->hw, OCS_HW_CB_UNSOLICITED, ocs_unsolicited_cb, ocs);
502 	ocs_hw_callback(&ocs->hw, OCS_HW_CB_PORT, ocs_port_cb, ocs);
503 
504 	ocs->fw_version = (const char*) ocs_hw_get_ptr(&ocs->hw, OCS_HW_FW_REV);
505 
506 	/* Initialize vport list */
507 	ocs_list_init(&xport->vport_list, ocs_vport_spec_t, link);
508 	ocs_lock_init(ocs, &xport->io_pending_lock, "io_pending_lock[%d]", ocs->instance_index);
509 	ocs_list_init(&xport->io_pending_list, ocs_io_t, io_pending_link);
510 	ocs_atomic_init(&xport->io_active_count, 0);
511 	ocs_atomic_init(&xport->io_pending_count, 0);
512 	ocs_atomic_init(&xport->io_total_free, 0);
513 	ocs_atomic_init(&xport->io_total_pending, 0);
514 	ocs_atomic_init(&xport->io_alloc_failed_count, 0);
515 	ocs_atomic_init(&xport->io_pending_recursing, 0);
516 	ocs_lock_init(ocs, &ocs->hw.watchdog_lock, " Watchdog Lock[%d]", ocs_instance(ocs));
517 	rc = ocs_hw_init(&ocs->hw);
518 	if (rc) {
519 		ocs_log_err(ocs, "ocs_hw_init failure\n");
520 		goto ocs_xport_init_cleanup;
521 	} else {
522 		hw_initialized = TRUE;
523 	}
524 
525 	rq_limit = max_hw_io/2;
526 	if (ocs_hw_set(&ocs->hw, OCS_HW_RQ_PROCESS_LIMIT, rq_limit) != OCS_HW_RTN_SUCCESS) {
527 		ocs_log_err(ocs, "%s: Can't set the RQ process limit\n", ocs->desc);
528 	}
529 
530 	if (ocs->config_tgt) {
531 		rc = ocs_scsi_tgt_new_device(ocs);
532 		if (rc) {
533 			ocs_log_err(ocs, "failed to initialize target\n");
534 			goto ocs_xport_init_cleanup;
535 		} else {
536 			tgt_device_set = TRUE;
537 		}
538 	}
539 
540 	if (ocs->enable_ini) {
541 		rc = ocs_scsi_ini_new_device(ocs);
542 		if (rc) {
543 			ocs_log_err(ocs, "failed to initialize initiator\n");
544 			goto ocs_xport_init_cleanup;
545 		} else {
546 			ini_device_set = TRUE;
547 		}
548 	}
549 
550 	/* Add vports */
551 	if (ocs->num_vports != 0) {
552 		uint32_t max_vports;
553 		ocs_hw_get(&ocs->hw, OCS_HW_MAX_VPORTS, &max_vports);
554 
555 		if (ocs->num_vports < max_vports) {
556 			ocs_log_debug(ocs, "Provisioning %d vports\n", ocs->num_vports);
557 			for (i = 0; i < ocs->num_vports; i++) {
558 				ocs_vport_create_spec(ocs, 0, 0, UINT32_MAX, ocs->enable_ini, ocs->enable_tgt, NULL, NULL);
559 			}
560 		} else {
561 			ocs_log_err(ocs, "failed to create vports. num_vports range should be (1-%d) \n", max_vports-1);
562 			goto ocs_xport_init_cleanup;
563 		}
564 	}
565 
566 	return 0;
567 
568 ocs_xport_init_cleanup:
569 	if (ini_device_set) {
570 		ocs_scsi_ini_del_device(ocs);
571 	}
572 
573 	if (tgt_device_set) {
574 		ocs_scsi_tgt_del_device(ocs);
575 	}
576 
577 	if (hw_initialized) {
578 		/* ocs_hw_teardown can only execute after ocs_hw_init */
579 		ocs_hw_teardown(&ocs->hw);
580 	}
581 
582 	return -1;
583 }
584 
585 /**
586  * @brief Detaches the transport from the device.
587  *
588  * @par Description
589  * Performs the functions required to shut down a device.
590  *
591  * @param xport Pointer to transport object.
592  *
593  * @return Returns 0 on success or a non-zero value on failure.
594  */
595 int32_t
596 ocs_xport_detach(ocs_xport_t *xport)
597 {
598 	ocs_t *ocs = xport->ocs;
599 
600 	/* free resources associated with target-server and initiator-client */
601 	if (ocs->config_tgt)
602 		ocs_scsi_tgt_del_device(ocs);
603 
604 	if (ocs->enable_ini) {
605 		ocs_scsi_ini_del_device(ocs);
606 
607 		/*Shutdown FC Statistics timer*/
608 		if (ocs_timer_pending(&ocs->xport->stats_timer))
609 			ocs_del_timer(&ocs->xport->stats_timer);
610 	}
611 
612 	ocs_hw_teardown(&ocs->hw);
613 
614 	return 0;
615 }
616 
617 /**
618  * @brief domain list empty callback
619  *
620  * @par Description
621  * Function is invoked when the device domain list goes empty. By convention
622  * @c arg points to an ocs_sem_t instance, that is incremented.
623  *
624  * @param ocs Pointer to device object.
625  * @param arg Pointer to semaphore instance.
626  *
627  * @return None.
628  */
629 
630 static void
631 ocs_xport_domain_list_empty_cb(ocs_t *ocs, void *arg)
632 {
633 	ocs_sem_t *sem = arg;
634 
635 	ocs_assert(ocs);
636 	ocs_assert(sem);
637 
638 	ocs_sem_v(sem);
639 }
640 
641 /**
642  * @brief post node event callback
643  *
644  * @par Description
645  * This function is called from the mailbox completion interrupt context to post an
646  * event to a node object. By doing this in the interrupt context, it has
647  * the benefit of only posting events in the interrupt context, deferring the need to
648  * create a per event node lock.
649  *
650  * @param hw Pointer to HW structure.
651  * @param status Completion status for mailbox command.
652  * @param mqe Mailbox queue completion entry.
653  * @param arg Callback argument.
654  *
655  * @return Returns 0 on success, a negative error code value on failure.
656  */
657 
658 static int32_t
659 ocs_xport_post_node_event_cb(ocs_hw_t *hw, int32_t status, uint8_t *mqe, void *arg)
660 {
661 	ocs_xport_post_node_event_t *payload = arg;
662 
663 	if (payload != NULL) {
664 		ocs_node_post_event(payload->node, payload->evt, payload->context);
665 		ocs_sem_v(&payload->sem);
666 	}
667 
668         return 0;
669 }
670 
671 /**
672  * @brief Initiate force free.
673  *
674  * @par Description
675  * Perform force free of OCS.
676  *
677  * @param xport Pointer to transport object.
678  *
679  * @return None.
680  */
681 
682 static void
683 ocs_xport_force_free(ocs_xport_t *xport)
684 {
685 	ocs_t *ocs = xport->ocs;
686 	ocs_domain_t *domain;
687 	ocs_domain_t *next;
688 
689 	ocs_log_debug(ocs, "reset required, do force shutdown\n");
690 	ocs_device_lock(ocs);
691 		ocs_list_foreach_safe(&ocs->domain_list, domain, next) {
692 			ocs_domain_force_free(domain);
693 		}
694 	ocs_device_unlock(ocs);
695 }
696 
697 /**
698  * @brief Perform transport attach function.
699  *
700  * @par Description
701  * Perform the attach function, which for the FC transport makes a HW call
702  * to bring up the link.
703  *
704  * @param xport pointer to transport object.
705  * @param cmd command to execute.
706  *
707  * ocs_xport_control(ocs_xport_t *xport, OCS_XPORT_PORT_ONLINE)
708  * ocs_xport_control(ocs_xport_t *xport, OCS_XPORT_PORT_OFFLINE)
709  * ocs_xport_control(ocs_xport_t *xport, OCS_XPORT_PORT_SHUTDOWN)
710  * ocs_xport_control(ocs_xport_t *xport, OCS_XPORT_POST_NODE_EVENT, ocs_node_t *node, ocs_sm_event_t, void *context)
711  *
712  * @return Returns 0 on success, or a negative error code value on failure.
713  */
714 
715 int32_t
716 ocs_xport_control(ocs_xport_t *xport, ocs_xport_ctrl_e cmd, ...)
717 {
718 	uint32_t rc = 0;
719 	ocs_t *ocs = NULL;
720 	va_list argp;
721 
722 	ocs_assert(xport, -1);
723 	ocs_assert(xport->ocs, -1);
724 	ocs = xport->ocs;
725 
726 	switch (cmd) {
727 	case OCS_XPORT_PORT_ONLINE: {
728 		/* Bring the port on-line */
729 		rc = ocs_hw_port_control(&ocs->hw, OCS_HW_PORT_INIT, 0, NULL, NULL);
730 		if (rc) {
731 			ocs_log_err(ocs, "%s: Can't init port\n", ocs->desc);
732 		} else {
733 			xport->configured_link_state = cmd;
734 		}
735 		break;
736 	}
737 	case OCS_XPORT_PORT_OFFLINE: {
738 		if (ocs_hw_port_control(&ocs->hw, OCS_HW_PORT_SHUTDOWN, 0, NULL, NULL)) {
739 			ocs_log_err(ocs, "port shutdown failed\n");
740 		} else {
741 			xport->configured_link_state = cmd;
742 		}
743 		break;
744 	}
745 
746 	case OCS_XPORT_SHUTDOWN: {
747 		ocs_sem_t sem;
748 		uint32_t reset_required;
749 
750 		/* if a PHYSDEV reset was performed (e.g. hw dump), will affect
751 		 * all PCI functions; orderly shutdown won't work, just force free
752 		 */
753 		/* TODO: need to poll this regularly... */
754 		if (ocs_hw_get(&ocs->hw, OCS_HW_RESET_REQUIRED, &reset_required) != OCS_HW_RTN_SUCCESS) {
755 			reset_required = 0;
756 		}
757 
758 		if (reset_required) {
759 			ocs_log_debug(ocs, "reset required, do force shutdown\n");
760 			ocs_xport_force_free(xport);
761 			break;
762 		}
763 		ocs_sem_init(&sem, 0, "domain_list_sem");
764 		ocs_register_domain_list_empty_cb(ocs, ocs_xport_domain_list_empty_cb, &sem);
765 
766 		if (ocs_hw_port_control(&ocs->hw, OCS_HW_PORT_SHUTDOWN, 0, NULL, NULL)) {
767 			ocs_log_debug(ocs, "port shutdown failed, do force shutdown\n");
768 			ocs_xport_force_free(xport);
769 		} else {
770 			ocs_log_debug(ocs, "Waiting %d seconds for domain shutdown.\n", (OCS_FC_DOMAIN_SHUTDOWN_TIMEOUT_USEC/1000000));
771 
772 			rc = ocs_sem_p(&sem, OCS_FC_DOMAIN_SHUTDOWN_TIMEOUT_USEC);
773 			if (rc) {
774 				ocs_log_debug(ocs, "Note: Domain shutdown timed out\n");
775 				ocs_xport_force_free(xport);
776 			}
777 		}
778 
779 		ocs_register_domain_list_empty_cb(ocs, NULL, NULL);
780 
781 		/* Free up any saved virtual ports */
782 		ocs_vport_del_all(ocs);
783 		break;
784 	}
785 
786 	/*
787 	 * POST_NODE_EVENT:  post an event to a node object
788 	 *
789 	 * This transport function is used to post an event to a node object. It does
790 	 * this by submitting a NOP mailbox command to defer execution to the
791 	 * interrupt context (thereby enforcing the serialized execution of event posting
792 	 * to the node state machine instances)
793 	 *
794 	 * A counting semaphore is used to make the call synchronous (we wait until
795 	 * the callback increments the semaphore before returning (or times out)
796 	 */
797 	case OCS_XPORT_POST_NODE_EVENT: {
798 		ocs_node_t *node;
799 		ocs_sm_event_t evt;
800 		void *context;
801 		ocs_xport_post_node_event_t payload;
802 		ocs_t *ocs;
803 		ocs_hw_t *hw;
804 
805 		/* Retrieve arguments */
806 		va_start(argp, cmd);
807 		node = va_arg(argp, ocs_node_t*);
808 		evt = va_arg(argp, ocs_sm_event_t);
809 		context = va_arg(argp, void *);
810 		va_end(argp);
811 
812 		ocs_assert(node, -1);
813 		ocs_assert(node->ocs, -1);
814 
815 		ocs = node->ocs;
816 		hw = &ocs->hw;
817 
818 		/* if node's state machine is disabled, don't bother continuing */
819 		if (!node->sm.current_state) {
820 			ocs_log_test(ocs, "node %p state machine disabled\n", node);
821 			return -1;
822 		}
823 
824 		/* Setup payload */
825 		ocs_memset(&payload, 0, sizeof(payload));
826 		ocs_sem_init(&payload.sem, 0, "xport_post_node_Event");
827 		payload.node = node;
828 		payload.evt = evt;
829 		payload.context = context;
830 
831 		if (ocs_hw_async_call(hw, ocs_xport_post_node_event_cb, &payload)) {
832 			ocs_log_test(ocs, "ocs_hw_async_call failed\n");
833 			rc = -1;
834 			break;
835 		}
836 
837 		/* Wait for completion */
838 		if (ocs_sem_p(&payload.sem, OCS_SEM_FOREVER)) {
839 			ocs_log_test(ocs, "POST_NODE_EVENT: sem wait failed\n");
840 			rc = -1;
841 		}
842 
843 		break;
844 	}
845 	/*
846 	 * Set wwnn for the port.  This will be used instead of the default provided by FW.
847 	 */
848 	case OCS_XPORT_WWNN_SET: {
849 		uint64_t wwnn;
850 
851 		/* Retrieve arguments */
852 		va_start(argp, cmd);
853 		wwnn = va_arg(argp, uint64_t);
854 		va_end(argp);
855 
856 		ocs_log_debug(ocs, " WWNN %016" PRIx64 "\n", wwnn);
857 		xport->req_wwnn = wwnn;
858 
859 		break;
860 	}
861 	/*
862 	 * Set wwpn for the port.  This will be used instead of the default provided by FW.
863 	 */
864 	case OCS_XPORT_WWPN_SET: {
865 		uint64_t wwpn;
866 
867 		/* Retrieve arguments */
868 		va_start(argp, cmd);
869 		wwpn = va_arg(argp, uint64_t);
870 		va_end(argp);
871 
872 		ocs_log_debug(ocs, " WWPN %016" PRIx64 "\n", wwpn);
873 		xport->req_wwpn = wwpn;
874 
875 		break;
876 	}
877 
878 	default:
879 		break;
880 	}
881 	return rc;
882 }
883 
884 /**
885  * @brief Return status on a link.
886  *
887  * @par Description
888  * Returns status information about a link.
889  *
890  * @param xport Pointer to transport object.
891  * @param cmd Command to execute.
892  * @param result Pointer to result value.
893  *
894  * ocs_xport_status(ocs_xport_t *xport, OCS_XPORT_PORT_STATUS)
895  * ocs_xport_status(ocs_xport_t *xport, OCS_XPORT_LINK_SPEED, ocs_xport_stats_t *result)
896  *	return link speed in MB/sec
897  * ocs_xport_status(ocs_xport_t *xport, OCS_XPORT_IS_SUPPORTED_LINK_SPEED, ocs_xport_stats_t *result)
898  *	[in] *result is speed to check in MB/s
899  *	returns 1 if supported, 0 if not
900  * ocs_xport_status(ocs_xport_t *xport, OCS_XPORT_LINK_STATISTICS, ocs_xport_stats_t *result)
901  *	return link/host port stats
902  * ocs_xport_status(ocs_xport_t *xport, OCS_XPORT_LINK_STAT_RESET, ocs_xport_stats_t *result)
903  *	resets link/host stats
904  *
905  *
906  * @return Returns 0 on success, or a negative error code value on failure.
907  */
908 
909 int32_t
910 ocs_xport_status(ocs_xport_t *xport, ocs_xport_status_e cmd, ocs_xport_stats_t *result)
911 {
912 	uint32_t rc = 0;
913 	ocs_t *ocs = NULL;
914 	ocs_xport_stats_t value;
915 	ocs_hw_rtn_e hw_rc;
916 
917 	ocs_assert(xport, -1);
918 	ocs_assert(xport->ocs, -1);
919 
920 	ocs = xport->ocs;
921 
922 	switch (cmd) {
923 	case OCS_XPORT_CONFIG_PORT_STATUS:
924 		ocs_assert(result, -1);
925 		if (xport->configured_link_state == 0) {
926 			/* Initial state is offline. configured_link_state is    */
927 			/* set to online explicitly when port is brought online. */
928 			xport->configured_link_state = OCS_XPORT_PORT_OFFLINE;
929 		}
930 		result->value = xport->configured_link_state;
931 		break;
932 
933 	case OCS_XPORT_PORT_STATUS:
934 		ocs_assert(result, -1);
935 		/* Determine port status based on link speed. */
936 		hw_rc = ocs_hw_get(&(ocs->hw), OCS_HW_LINK_SPEED, &value.value);
937 		if (hw_rc == OCS_HW_RTN_SUCCESS) {
938 			if (value.value == 0) {
939 				result->value = 0;
940 			} else {
941 				result->value = 1;
942 			}
943 			rc = 0;
944 		} else {
945 			rc = -1;
946 		}
947 		break;
948 
949 	case OCS_XPORT_LINK_SPEED: {
950 		uint32_t speed;
951 
952 		ocs_assert(result, -1);
953 		result->value = 0;
954 
955 		rc = ocs_hw_get(&ocs->hw, OCS_HW_LINK_SPEED, &speed);
956 		if (rc == 0) {
957 			result->value = speed;
958 		}
959 		break;
960 	}
961 
962 	case OCS_XPORT_IS_SUPPORTED_LINK_SPEED: {
963 		uint32_t speed;
964 		uint32_t link_module_type;
965 
966 		ocs_assert(result, -1);
967 		speed = result->value;
968 
969 		rc = ocs_hw_get(&ocs->hw, OCS_HW_LINK_MODULE_TYPE, &link_module_type);
970 		if (rc == 0) {
971 			switch(speed) {
972 			case 1000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_1GB) != 0; break;
973 			case 2000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_2GB) != 0; break;
974 			case 4000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_4GB) != 0; break;
975 			case 8000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_8GB) != 0; break;
976 			case 10000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_10GB) != 0; break;
977 			case 16000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_16GB) != 0; break;
978 			case 32000:	rc = (link_module_type & OCS_HW_LINK_MODULE_TYPE_32GB) != 0; break;
979 			default:	rc = 0; break;
980 			}
981 		} else {
982 			rc = 0;
983 		}
984 		break;
985 	}
986 	case OCS_XPORT_LINK_STATISTICS:
987 		ocs_device_lock(ocs);
988 			ocs_memcpy((void *)result, &ocs->xport->fc_xport_stats, sizeof(ocs_xport_stats_t));
989 		ocs_device_unlock(ocs);
990 		break;
991 	case OCS_XPORT_LINK_STAT_RESET: {
992 		/* Create a semaphore to synchronize the stat reset process. */
993 		ocs_sem_init(&(result->stats.semaphore), 0, "fc_stats_reset");
994 
995 		/* First reset the link stats */
996 		if ((rc = ocs_hw_get_link_stats(&ocs->hw, 0, 1, 1, ocs_xport_link_stats_cb, result)) != 0) {
997 			ocs_log_err(ocs, "%s: Failed to reset link statistics\n", __func__);
998 			break;
999 		}
1000 
1001 		/* Wait for semaphore to be signaled when the command completes */
1002 		/* TODO:  Should there be a timeout on this?  If so, how long? */
1003 		if (ocs_sem_p(&(result->stats.semaphore), OCS_SEM_FOREVER) != 0) {
1004 			/* Undefined failure */
1005 			ocs_log_test(ocs, "ocs_sem_p failed\n");
1006 			rc = -ENXIO;
1007 			break;
1008 		}
1009 
1010 		/* Next reset the host stats */
1011 		if ((rc = ocs_hw_get_host_stats(&ocs->hw, 1,  ocs_xport_host_stats_cb, result)) != 0) {
1012 			ocs_log_err(ocs, "%s: Failed to reset host statistics\n", __func__);
1013 			break;
1014 		}
1015 
1016 		/* Wait for semaphore to be signaled when the command completes */
1017 		if (ocs_sem_p(&(result->stats.semaphore), OCS_SEM_FOREVER) != 0) {
1018 			/* Undefined failure */
1019 			ocs_log_test(ocs, "ocs_sem_p failed\n");
1020 			rc = -ENXIO;
1021 			break;
1022 		}
1023 		break;
1024 	}
1025 	case OCS_XPORT_IS_QUIESCED:
1026 		ocs_device_lock(ocs);
1027 			result->value = ocs_list_empty(&ocs->domain_list);
1028 		ocs_device_unlock(ocs);
1029 		break;
1030 	default:
1031 		rc = -1;
1032 		break;
1033 	}
1034 
1035 	return rc;
1036 
1037 }
1038 
1039 static void
1040 ocs_xport_link_stats_cb(int32_t status, uint32_t num_counters, ocs_hw_link_stat_counts_t *counters, void *arg)
1041 {
1042         ocs_xport_stats_t *result = arg;
1043 
1044         result->stats.link_stats.link_failure_error_count = counters[OCS_HW_LINK_STAT_LINK_FAILURE_COUNT].counter;
1045         result->stats.link_stats.loss_of_sync_error_count = counters[OCS_HW_LINK_STAT_LOSS_OF_SYNC_COUNT].counter;
1046         result->stats.link_stats.primitive_sequence_error_count = counters[OCS_HW_LINK_STAT_PRIMITIVE_SEQ_COUNT].counter;
1047         result->stats.link_stats.invalid_transmission_word_error_count = counters[OCS_HW_LINK_STAT_INVALID_XMIT_WORD_COUNT].counter;
1048         result->stats.link_stats.crc_error_count = counters[OCS_HW_LINK_STAT_CRC_COUNT].counter;
1049 
1050         ocs_sem_v(&(result->stats.semaphore));
1051 }
1052 
1053 static void
1054 ocs_xport_host_stats_cb(int32_t status, uint32_t num_counters, ocs_hw_host_stat_counts_t *counters, void *arg)
1055 {
1056         ocs_xport_stats_t *result = arg;
1057 
1058         result->stats.host_stats.transmit_kbyte_count = counters[OCS_HW_HOST_STAT_TX_KBYTE_COUNT].counter;
1059         result->stats.host_stats.receive_kbyte_count = counters[OCS_HW_HOST_STAT_RX_KBYTE_COUNT].counter;
1060         result->stats.host_stats.transmit_frame_count = counters[OCS_HW_HOST_STAT_TX_FRAME_COUNT].counter;
1061         result->stats.host_stats.receive_frame_count = counters[OCS_HW_HOST_STAT_RX_FRAME_COUNT].counter;
1062 
1063         ocs_sem_v(&(result->stats.semaphore));
1064 }
1065 
1066 /**
1067  * @brief Free a transport object.
1068  *
1069  * @par Description
1070  * The transport object is freed.
1071  *
1072  * @param xport Pointer to transport object.
1073  *
1074  * @return None.
1075  */
1076 
1077 void
1078 ocs_xport_free(ocs_xport_t *xport)
1079 {
1080 	ocs_t *ocs;
1081 	uint32_t i;
1082 
1083 	if (xport) {
1084 		ocs = xport->ocs;
1085 		ocs_io_pool_free(xport->io_pool);
1086 		ocs_node_free_pool(ocs);
1087 		if(mtx_initialized(&xport->io_pending_lock.lock))
1088 			ocs_lock_free(&xport->io_pending_lock);
1089 
1090 		for (i = 0; i < SLI4_MAX_FCFI; i++) {
1091 			ocs_lock_free(&xport->fcfi[i].pend_frames_lock);
1092 		}
1093 
1094 		ocs_xport_rq_threads_teardown(xport);
1095 
1096 		ocs_free(ocs, xport, sizeof(*xport));
1097 	}
1098 }
1099