xref: /freebsd/sys/dev/liquidio/base/lio_request_manager.c (revision 2e3507c25e42292b45a5482e116d278f5515d04d)
1 /*
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2017 Cavium, Inc.. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Cavium, Inc. nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER(S) OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "lio_bsd.h"
35 #include "lio_common.h"
36 #include "lio_droq.h"
37 #include "lio_iq.h"
38 #include "lio_response_manager.h"
39 #include "lio_device.h"
40 #include "lio_main.h"
41 #include "lio_network.h"
42 #include "cn23xx_pf_device.h"
43 #include "lio_rxtx.h"
44 
45 struct lio_iq_post_status {
46 	int	status;
47 	int	index;
48 };
49 
50 static void	lio_check_db_timeout(void *arg, int pending);
51 static void	__lio_check_db_timeout(struct octeon_device *oct,
52 				       uint64_t iq_no);
53 
54 /* Return 0 on success, 1 on failure */
55 int
56 lio_init_instr_queue(struct octeon_device *oct, union octeon_txpciq txpciq,
57 		     uint32_t num_descs)
58 {
59 	struct lio_instr_queue	*iq;
60 	struct lio_iq_config	*conf = NULL;
61 	struct lio_tq		*db_tq;
62 	struct lio_request_list	*request_buf;
63 	bus_size_t		max_size;
64 	uint32_t		iq_no = (uint32_t)txpciq.s.q_no;
65 	uint32_t		q_size;
66 	int			error, i;
67 
68 	if (LIO_CN23XX_PF(oct))
69 		conf = &(LIO_GET_IQ_CFG(LIO_CHIP_CONF(oct, cn23xx_pf)));
70 	if (conf == NULL) {
71 		lio_dev_err(oct, "Unsupported Chip %x\n", oct->chip_id);
72 		return (1);
73 	}
74 
75 	q_size = (uint32_t)conf->instr_type * num_descs;
76 	iq = oct->instr_queue[iq_no];
77 	iq->oct_dev = oct;
78 
79 	max_size = LIO_CN23XX_PKI_MAX_FRAME_SIZE * num_descs;
80 
81 	error = bus_dma_tag_create(bus_get_dma_tag(oct->device),	/* parent */
82 				   1, 0,				/* alignment, bounds */
83 				   BUS_SPACE_MAXADDR,			/* lowaddr */
84 				   BUS_SPACE_MAXADDR,			/* highaddr */
85 				   NULL, NULL,				/* filter, filterarg */
86 				   max_size,				/* maxsize */
87 				   LIO_MAX_SG,				/* nsegments */
88 				   PAGE_SIZE,				/* maxsegsize */
89 				   0,					/* flags */
90 				   NULL,				/* lockfunc */
91 				   NULL,				/* lockfuncarg */
92 				   &iq->txtag);
93 	if (error) {
94 		lio_dev_err(oct, "Cannot allocate memory for instr queue %d\n",
95 			    iq_no);
96 		return (1);
97 	}
98 
99 	iq->base_addr = lio_dma_alloc(q_size, (vm_paddr_t *)&iq->base_addr_dma);
100 	if (!iq->base_addr) {
101 		lio_dev_err(oct, "Cannot allocate memory for instr queue %d\n",
102 			    iq_no);
103 		return (1);
104 	}
105 
106 	iq->max_count = num_descs;
107 
108 	/*
109 	 * Initialize a list to holds requests that have been posted to
110 	 * Octeon but has yet to be fetched by octeon
111 	 */
112 	iq->request_list = malloc(sizeof(*iq->request_list) * num_descs,
113 				  M_DEVBUF, M_NOWAIT | M_ZERO);
114 	if (iq->request_list == NULL) {
115 		lio_dev_err(oct, "Alloc failed for IQ[%d] nr free list\n",
116 			    iq_no);
117 		return (1);
118 	}
119 
120 	lio_dev_dbg(oct, "IQ[%d]: base: %p basedma: %llx count: %d\n",
121 		    iq_no, iq->base_addr, LIO_CAST64(iq->base_addr_dma),
122 		    iq->max_count);
123 
124 	/* Create the descriptor buffer dma maps */
125 	request_buf = iq->request_list;
126 	for (i = 0; i < num_descs; i++, request_buf++) {
127 		error = bus_dmamap_create(iq->txtag, 0, &request_buf->map);
128 		if (error) {
129 			lio_dev_err(oct, "Unable to create TX DMA map\n");
130 			return (1);
131 		}
132 	}
133 
134 	iq->txpciq.txpciq64 = txpciq.txpciq64;
135 	iq->fill_cnt = 0;
136 	iq->host_write_index = 0;
137 	iq->octeon_read_index = 0;
138 	iq->flush_index = 0;
139 	iq->last_db_time = 0;
140 	iq->db_timeout = (uint32_t)conf->db_timeout;
141 	atomic_store_rel_int(&iq->instr_pending, 0);
142 
143 	/* Initialize the lock for this instruction queue */
144 	mtx_init(&iq->lock, "Tx_lock", NULL, MTX_DEF);
145 	mtx_init(&iq->post_lock, "iq_post_lock", NULL, MTX_DEF);
146 	mtx_init(&iq->enq_lock, "enq_lock", NULL, MTX_DEF);
147 
148 	mtx_init(&iq->iq_flush_running_lock, "iq_flush_running_lock", NULL,
149 		 MTX_DEF);
150 
151 	oct->io_qmask.iq |= BIT_ULL(iq_no);
152 
153 	/* Set the 32B/64B mode for each input queue */
154 	oct->io_qmask.iq64B |= ((conf->instr_type == 64) << iq_no);
155 	iq->iqcmd_64B = (conf->instr_type == 64);
156 
157 	oct->fn_list.setup_iq_regs(oct, iq_no);
158 
159 	db_tq = &oct->check_db_tq[iq_no];
160 	db_tq->tq = taskqueue_create("lio_check_db_timeout", M_WAITOK,
161 				     taskqueue_thread_enqueue, &db_tq->tq);
162 	if (db_tq->tq == NULL) {
163 		lio_dev_err(oct, "check db wq create failed for iq %d\n",
164 			    iq_no);
165 		return (1);
166 	}
167 
168 	TIMEOUT_TASK_INIT(db_tq->tq, &db_tq->work, 0, lio_check_db_timeout,
169 			  (void *)db_tq);
170 	db_tq->ctxul = iq_no;
171 	db_tq->ctxptr = oct;
172 
173 	taskqueue_start_threads(&db_tq->tq, 1, PI_NET,
174 				"lio%d_check_db_timeout:%d",
175 				oct->octeon_id, iq_no);
176 	taskqueue_enqueue_timeout(db_tq->tq, &db_tq->work, 1);
177 
178 	/* Allocate a buf ring */
179 	oct->instr_queue[iq_no]->br =
180 		buf_ring_alloc(LIO_BR_SIZE, M_DEVBUF, M_WAITOK,
181 			       &oct->instr_queue[iq_no]->enq_lock);
182 	if (oct->instr_queue[iq_no]->br == NULL) {
183 		lio_dev_err(oct, "Critical Failure setting up buf ring\n");
184 		return (1);
185 	}
186 
187 	return (0);
188 }
189 
190 int
191 lio_delete_instr_queue(struct octeon_device *oct, uint32_t iq_no)
192 {
193 	struct lio_instr_queue		*iq = oct->instr_queue[iq_no];
194 	struct lio_request_list		*request_buf;
195 	struct lio_mbuf_free_info	*finfo;
196 	uint64_t			desc_size = 0, q_size;
197 	int				i;
198 
199 	lio_dev_dbg(oct, "%s[%d]\n", __func__, iq_no);
200 
201 	if (oct->check_db_tq[iq_no].tq != NULL) {
202 		while (taskqueue_cancel_timeout(oct->check_db_tq[iq_no].tq,
203 						&oct->check_db_tq[iq_no].work,
204 						NULL))
205 			taskqueue_drain_timeout(oct->check_db_tq[iq_no].tq,
206 						&oct->check_db_tq[iq_no].work);
207 		taskqueue_free(oct->check_db_tq[iq_no].tq);
208 		oct->check_db_tq[iq_no].tq = NULL;
209 	}
210 
211 	if (LIO_CN23XX_PF(oct))
212 		desc_size =
213 		    LIO_GET_IQ_INSTR_TYPE_CFG(LIO_CHIP_CONF(oct, cn23xx_pf));
214 
215 	request_buf = iq->request_list;
216 	for (i = 0; i < iq->max_count; i++, request_buf++) {
217 		if ((request_buf->reqtype == LIO_REQTYPE_NORESP_NET) ||
218 		    (request_buf->reqtype == LIO_REQTYPE_NORESP_NET_SG)) {
219 			if (request_buf->buf != NULL) {
220 				finfo = request_buf->buf;
221 				bus_dmamap_sync(iq->txtag, request_buf->map,
222 						BUS_DMASYNC_POSTWRITE);
223 				bus_dmamap_unload(iq->txtag,
224 						  request_buf->map);
225 				m_freem(finfo->mb);
226 				request_buf->buf = NULL;
227 				if (request_buf->map != NULL) {
228 					bus_dmamap_destroy(iq->txtag,
229 							   request_buf->map);
230 					request_buf->map = NULL;
231 				}
232 			} else if (request_buf->map != NULL) {
233 				bus_dmamap_unload(iq->txtag, request_buf->map);
234 				bus_dmamap_destroy(iq->txtag, request_buf->map);
235 				request_buf->map = NULL;
236 			}
237 		}
238 	}
239 
240 	if (iq->br != NULL) {
241 		buf_ring_free(iq->br, M_DEVBUF);
242 		iq->br = NULL;
243 	}
244 
245 	if (iq->request_list != NULL) {
246 		free(iq->request_list, M_DEVBUF);
247 		iq->request_list = NULL;
248 	}
249 
250 	if (iq->txtag != NULL) {
251 		bus_dma_tag_destroy(iq->txtag);
252 		iq->txtag = NULL;
253 	}
254 
255 	if (iq->base_addr) {
256 		q_size = iq->max_count * desc_size;
257 		lio_dma_free((uint32_t)q_size, iq->base_addr);
258 
259 		oct->io_qmask.iq &= ~(1ULL << iq_no);
260 		bzero(oct->instr_queue[iq_no], sizeof(struct lio_instr_queue));
261 		oct->num_iqs--;
262 
263 		return (0);
264 	}
265 
266 	return (1);
267 }
268 
269 /* Return 0 on success, 1 on failure */
270 int
271 lio_setup_iq(struct octeon_device *oct, int ifidx, int q_index,
272 	     union octeon_txpciq txpciq, uint32_t num_descs)
273 {
274 	uint32_t	iq_no = (uint32_t)txpciq.s.q_no;
275 
276 	if (oct->instr_queue[iq_no]->oct_dev != NULL) {
277 		lio_dev_dbg(oct, "IQ is in use. Cannot create the IQ: %d again\n",
278 			    iq_no);
279 		oct->instr_queue[iq_no]->txpciq.txpciq64 = txpciq.txpciq64;
280 		return (0);
281 	}
282 
283 	oct->instr_queue[iq_no]->q_index = q_index;
284 	oct->instr_queue[iq_no]->ifidx = ifidx;
285 
286 	if (lio_init_instr_queue(oct, txpciq, num_descs)) {
287 		lio_delete_instr_queue(oct, iq_no);
288 		return (1);
289 	}
290 
291 	oct->num_iqs++;
292 	if (oct->fn_list.enable_io_queues(oct))
293 		return (1);
294 
295 	return (0);
296 }
297 
298 int
299 lio_wait_for_instr_fetch(struct octeon_device *oct)
300 {
301 	int	i, retry = 1000, pending, instr_cnt = 0;
302 
303 	do {
304 		instr_cnt = 0;
305 
306 		for (i = 0; i < LIO_MAX_INSTR_QUEUES(oct); i++) {
307 			if (!(oct->io_qmask.iq & BIT_ULL(i)))
308 				continue;
309 			pending = atomic_load_acq_int(
310 					&oct->instr_queue[i]->instr_pending);
311 			if (pending)
312 				__lio_check_db_timeout(oct, i);
313 			instr_cnt += pending;
314 		}
315 
316 		if (instr_cnt == 0)
317 			break;
318 
319 		lio_sleep_timeout(1);
320 
321 	} while (retry-- && instr_cnt);
322 
323 	return (instr_cnt);
324 }
325 
326 static inline void
327 lio_ring_doorbell(struct octeon_device *oct, struct lio_instr_queue *iq)
328 {
329 
330 	if (atomic_load_acq_int(&oct->status) == LIO_DEV_RUNNING) {
331 		lio_write_csr32(oct, iq->doorbell_reg, iq->fill_cnt);
332 		/* make sure doorbell write goes through */
333 		__compiler_membar();
334 		iq->fill_cnt = 0;
335 		iq->last_db_time = ticks;
336 		return;
337 	}
338 }
339 
340 static inline void
341 __lio_copy_cmd_into_iq(struct lio_instr_queue *iq, uint8_t *cmd)
342 {
343 	uint8_t	*iqptr, cmdsize;
344 
345 	cmdsize = ((iq->iqcmd_64B) ? 64 : 32);
346 	iqptr = iq->base_addr + (cmdsize * iq->host_write_index);
347 
348 	memcpy(iqptr, cmd, cmdsize);
349 }
350 
351 static inline struct lio_iq_post_status
352 __lio_post_command2(struct lio_instr_queue *iq, uint8_t *cmd)
353 {
354 	struct lio_iq_post_status	st;
355 
356 	st.status = LIO_IQ_SEND_OK;
357 
358 	/*
359 	 * This ensures that the read index does not wrap around to the same
360 	 * position if queue gets full before Octeon could fetch any instr.
361 	 */
362 	if (atomic_load_acq_int(&iq->instr_pending) >=
363 	    (int32_t)(iq->max_count - 1)) {
364 		st.status = LIO_IQ_SEND_FAILED;
365 		st.index = -1;
366 		return (st);
367 	}
368 
369 	if (atomic_load_acq_int(&iq->instr_pending) >=
370 	    (int32_t)(iq->max_count - 2))
371 		st.status = LIO_IQ_SEND_STOP;
372 
373 	__lio_copy_cmd_into_iq(iq, cmd);
374 
375 	/* "index" is returned, host_write_index is modified. */
376 	st.index = iq->host_write_index;
377 	iq->host_write_index = lio_incr_index(iq->host_write_index, 1,
378 					      iq->max_count);
379 	iq->fill_cnt++;
380 
381 	/*
382 	 * Flush the command into memory. We need to be sure the data is in
383 	 * memory before indicating that the instruction is pending.
384 	 */
385 	wmb();
386 
387 	atomic_add_int(&iq->instr_pending, 1);
388 
389 	return (st);
390 }
391 
392 static inline void
393 __lio_add_to_request_list(struct lio_instr_queue *iq, int idx, void *buf,
394 			  int reqtype)
395 {
396 
397 	iq->request_list[idx].buf = buf;
398 	iq->request_list[idx].reqtype = reqtype;
399 }
400 
401 /* Can only run in process context */
402 int
403 lio_process_iq_request_list(struct octeon_device *oct,
404 			    struct lio_instr_queue *iq, uint32_t budget)
405 {
406 	struct lio_soft_command		*sc;
407 	struct octeon_instr_irh		*irh = NULL;
408 	void				*buf;
409 	uint32_t			inst_count = 0;
410 	uint32_t			old = iq->flush_index;
411 	int				reqtype;
412 
413 	while (old != iq->octeon_read_index) {
414 		reqtype = iq->request_list[old].reqtype;
415 		buf = iq->request_list[old].buf;
416 
417 		if (reqtype == LIO_REQTYPE_NONE)
418 			goto skip_this;
419 
420 		switch (reqtype) {
421 		case LIO_REQTYPE_NORESP_NET:
422 			lio_free_mbuf(iq, buf);
423 			break;
424 		case LIO_REQTYPE_NORESP_NET_SG:
425 			lio_free_sgmbuf(iq, buf);
426 			break;
427 		case LIO_REQTYPE_RESP_NET:
428 		case LIO_REQTYPE_SOFT_COMMAND:
429 			sc = buf;
430 			if (LIO_CN23XX_PF(oct))
431 				irh = (struct octeon_instr_irh *)
432 					&sc->cmd.cmd3.irh;
433 			if (irh->rflag) {
434 				/*
435 				 * We're expecting a response from Octeon.
436 				 * It's up to lio_process_ordered_list() to
437 				 * process  sc. Add sc to the ordered soft
438 				 * command response list because we expect
439 				 * a response from Octeon.
440 				 */
441 				mtx_lock(&oct->response_list
442 					 [LIO_ORDERED_SC_LIST].lock);
443 				atomic_add_int(&oct->response_list
444 					       [LIO_ORDERED_SC_LIST].
445 					       pending_req_count, 1);
446 				STAILQ_INSERT_TAIL(&oct->response_list
447 						   [LIO_ORDERED_SC_LIST].
448 						   head, &sc->node, entries);
449 				mtx_unlock(&oct->response_list
450 					   [LIO_ORDERED_SC_LIST].lock);
451 			} else {
452 				if (sc->callback != NULL) {
453 					/* This callback must not sleep */
454 					sc->callback(oct, LIO_REQUEST_DONE,
455 						     sc->callback_arg);
456 				}
457 			}
458 
459 			break;
460 		default:
461 			lio_dev_err(oct, "%s Unknown reqtype: %d buf: %p at idx %d\n",
462 				    __func__, reqtype, buf, old);
463 		}
464 
465 		iq->request_list[old].buf = NULL;
466 		iq->request_list[old].reqtype = 0;
467 
468 skip_this:
469 		inst_count++;
470 		old = lio_incr_index(old, 1, iq->max_count);
471 
472 		if ((budget) && (inst_count >= budget))
473 			break;
474 	}
475 
476 	iq->flush_index = old;
477 
478 	return (inst_count);
479 }
480 
481 /* Can only be called from process context */
482 int
483 lio_flush_iq(struct octeon_device *oct, struct lio_instr_queue *iq,
484 	     uint32_t budget)
485 {
486 	uint32_t	inst_processed = 0;
487 	uint32_t	tot_inst_processed = 0;
488 	int		tx_done = 1;
489 
490 	if (!mtx_trylock(&iq->iq_flush_running_lock))
491 		return (tx_done);
492 
493 	mtx_lock(&iq->lock);
494 
495 	iq->octeon_read_index = oct->fn_list.update_iq_read_idx(iq);
496 
497 	do {
498 		/* Process any outstanding IQ packets. */
499 		if (iq->flush_index == iq->octeon_read_index)
500 			break;
501 
502 		if (budget)
503 			inst_processed =
504 				lio_process_iq_request_list(oct, iq,
505 							    budget -
506 							    tot_inst_processed);
507 		else
508 			inst_processed =
509 				lio_process_iq_request_list(oct, iq, 0);
510 
511 		if (inst_processed) {
512 			atomic_subtract_int(&iq->instr_pending, inst_processed);
513 			iq->stats.instr_processed += inst_processed;
514 		}
515 		tot_inst_processed += inst_processed;
516 		inst_processed = 0;
517 
518 	} while (tot_inst_processed < budget);
519 
520 	if (budget && (tot_inst_processed >= budget))
521 		tx_done = 0;
522 
523 	iq->last_db_time = ticks;
524 
525 	mtx_unlock(&iq->lock);
526 
527 	mtx_unlock(&iq->iq_flush_running_lock);
528 
529 	return (tx_done);
530 }
531 
532 /*
533  * Process instruction queue after timeout.
534  * This routine gets called from a taskqueue or when removing the module.
535  */
536 static void
537 __lio_check_db_timeout(struct octeon_device *oct, uint64_t iq_no)
538 {
539 	struct lio_instr_queue	*iq;
540 	uint64_t		next_time;
541 
542 	if (oct == NULL)
543 		return;
544 
545 	iq = oct->instr_queue[iq_no];
546 	if (iq == NULL)
547 		return;
548 
549 	if (atomic_load_acq_int(&iq->instr_pending)) {
550 		/* If ticks - last_db_time < db_timeout do nothing  */
551 		next_time = iq->last_db_time + lio_ms_to_ticks(iq->db_timeout);
552 		if (!lio_check_timeout(ticks, next_time))
553 			return;
554 
555 		iq->last_db_time = ticks;
556 
557 		/* Flush the instruction queue */
558 		lio_flush_iq(oct, iq, 0);
559 
560 		lio_enable_irq(NULL, iq);
561 	}
562 
563 	if (oct->props.ifp != NULL && iq->br != NULL) {
564 		if (mtx_trylock(&iq->enq_lock)) {
565 			if (!drbr_empty(oct->props.ifp, iq->br))
566 				lio_mq_start_locked(oct->props.ifp, iq);
567 
568 			mtx_unlock(&iq->enq_lock);
569 		}
570 	}
571 }
572 
573 /*
574  * Called by the Poll thread at regular intervals to check the instruction
575  * queue for commands to be posted and for commands that were fetched by Octeon.
576  */
577 static void
578 lio_check_db_timeout(void *arg, int pending)
579 {
580 	struct lio_tq		*db_tq = (struct lio_tq *)arg;
581 	struct octeon_device	*oct = db_tq->ctxptr;
582 	uint64_t		iq_no = db_tq->ctxul;
583 	uint32_t		delay = 10;
584 
585 	__lio_check_db_timeout(oct, iq_no);
586 	taskqueue_enqueue_timeout(db_tq->tq, &db_tq->work,
587 				  lio_ms_to_ticks(delay));
588 }
589 
590 int
591 lio_send_command(struct octeon_device *oct, uint32_t iq_no,
592 		 uint32_t force_db, void *cmd, void *buf,
593 		 uint32_t datasize, uint32_t reqtype)
594 {
595 	struct lio_iq_post_status	st;
596 	struct lio_instr_queue		*iq = oct->instr_queue[iq_no];
597 
598 	/*
599 	 * Get the lock and prevent other tasks and tx interrupt handler
600 	 * from running.
601 	 */
602 	mtx_lock(&iq->post_lock);
603 
604 	st = __lio_post_command2(iq, cmd);
605 
606 	if (st.status != LIO_IQ_SEND_FAILED) {
607 		__lio_add_to_request_list(iq, st.index, buf, reqtype);
608 		LIO_INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, bytes_sent, datasize);
609 		LIO_INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_posted, 1);
610 
611 		if (force_db || (st.status == LIO_IQ_SEND_STOP))
612 			lio_ring_doorbell(oct, iq);
613 	} else {
614 		LIO_INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_dropped, 1);
615 	}
616 
617 	mtx_unlock(&iq->post_lock);
618 
619 	/*
620 	 * This is only done here to expedite packets being flushed for
621 	 * cases where there are no IQ completion interrupts.
622 	 */
623 
624 	return (st.status);
625 }
626 
627 void
628 lio_prepare_soft_command(struct octeon_device *oct, struct lio_soft_command *sc,
629 			 uint8_t opcode, uint8_t subcode, uint32_t irh_ossp,
630 			 uint64_t ossp0, uint64_t ossp1)
631 {
632 	struct octeon_instr_ih3		*ih3;
633 	struct octeon_instr_pki_ih3	*pki_ih3;
634 	struct octeon_instr_irh		*irh;
635 	struct octeon_instr_rdp		*rdp;
636 
637 	KASSERT(opcode <= 15, ("%s, %d, opcode > 15", __func__, __LINE__));
638 	KASSERT(subcode <= 127, ("%s, %d, opcode > 127", __func__, __LINE__));
639 
640 	if (LIO_CN23XX_PF(oct)) {
641 		ih3 = (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
642 
643 		ih3->pkind = oct->instr_queue[sc->iq_no]->txpciq.s.pkind;
644 
645 		pki_ih3 = (struct octeon_instr_pki_ih3 *)&sc->cmd.cmd3.pki_ih3;
646 
647 		pki_ih3->w = 1;
648 		pki_ih3->raw = 1;
649 		pki_ih3->utag = 1;
650 		pki_ih3->uqpg = oct->instr_queue[sc->iq_no]->txpciq.s.use_qpg;
651 		pki_ih3->utt = 1;
652 		pki_ih3->tag = LIO_CONTROL;
653 		pki_ih3->tagtype = LIO_ATOMIC_TAG;
654 		pki_ih3->qpg = oct->instr_queue[sc->iq_no]->txpciq.s.qpg;
655 		pki_ih3->pm = 0x7;
656 		pki_ih3->sl = 8;
657 
658 		if (sc->datasize)
659 			ih3->dlengsz = sc->datasize;
660 
661 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
662 		irh->opcode = opcode;
663 		irh->subcode = subcode;
664 
665 		/* opcode/subcode specific parameters (ossp) */
666 		irh->ossp = irh_ossp;
667 		sc->cmd.cmd3.ossp[0] = ossp0;
668 		sc->cmd.cmd3.ossp[1] = ossp1;
669 
670 		if (sc->rdatasize) {
671 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd3.rdp;
672 			rdp->pcie_port = oct->pcie_port;
673 			rdp->rlen = sc->rdatasize;
674 
675 			irh->rflag = 1;
676 			/* PKI IH3 */
677 			/* pki_ih3 irh+ossp[0]+ossp[1]+rdp+rptr = 48 bytes */
678 			ih3->fsz = LIO_SOFTCMDRESP_IH3;
679 		} else {
680 			irh->rflag = 0;
681 			/* PKI IH3 */
682 			/* pki_h3 + irh + ossp[0] + ossp[1] = 32 bytes */
683 			ih3->fsz = LIO_PCICMD_O3;
684 		}
685 	}
686 }
687 
688 int
689 lio_send_soft_command(struct octeon_device *oct, struct lio_soft_command *sc)
690 {
691 	struct octeon_instr_ih3	*ih3;
692 	struct octeon_instr_irh	*irh;
693 	uint32_t		len = 0;
694 
695 	if (LIO_CN23XX_PF(oct)) {
696 		ih3 = (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
697 		if (ih3->dlengsz) {
698 			KASSERT(sc->dmadptr, ("%s, %d, sc->dmadptr is NULL",
699 					      __func__, __LINE__));
700 			sc->cmd.cmd3.dptr = sc->dmadptr;
701 		}
702 
703 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
704 		if (irh->rflag) {
705 			KASSERT(sc->dmarptr, ("%s, %d, sc->dmarptr is NULL",
706 					      __func__, __LINE__));
707 			KASSERT(sc->status_word, ("%s, %d, sc->status_word is NULL",
708 						  __func__, __LINE__));
709 			*sc->status_word = COMPLETION_WORD_INIT;
710 			sc->cmd.cmd3.rptr = sc->dmarptr;
711 		}
712 		len = (uint32_t)ih3->dlengsz;
713 	}
714 	if (sc->wait_time)
715 		sc->timeout = ticks + lio_ms_to_ticks(sc->wait_time);
716 
717 	return (lio_send_command(oct, sc->iq_no, 1, &sc->cmd, sc,
718 				 len, LIO_REQTYPE_SOFT_COMMAND));
719 }
720 
721 int
722 lio_setup_sc_buffer_pool(struct octeon_device *oct)
723 {
724 	struct lio_soft_command	*sc;
725 	uint64_t		dma_addr;
726 	int			i;
727 
728 	STAILQ_INIT(&oct->sc_buf_pool.head);
729 	mtx_init(&oct->sc_buf_pool.lock, "sc_pool_lock", NULL, MTX_DEF);
730 	atomic_store_rel_int(&oct->sc_buf_pool.alloc_buf_count, 0);
731 
732 	for (i = 0; i < LIO_MAX_SOFT_COMMAND_BUFFERS; i++) {
733 		sc = (struct lio_soft_command *)
734 			lio_dma_alloc(LIO_SOFT_COMMAND_BUFFER_SIZE, (vm_paddr_t *)&dma_addr);
735 		if (sc == NULL) {
736 			lio_free_sc_buffer_pool(oct);
737 			return (1);
738 		}
739 
740 		sc->dma_addr = dma_addr;
741 		sc->size = LIO_SOFT_COMMAND_BUFFER_SIZE;
742 
743 		STAILQ_INSERT_TAIL(&oct->sc_buf_pool.head, &sc->node, entries);
744 	}
745 
746 	return (0);
747 }
748 
749 int
750 lio_free_sc_buffer_pool(struct octeon_device *oct)
751 {
752 	struct lio_stailq_node	*tmp, *tmp2;
753 	struct lio_soft_command	*sc;
754 
755 	mtx_lock(&oct->sc_buf_pool.lock);
756 
757 	STAILQ_FOREACH_SAFE(tmp, &oct->sc_buf_pool.head, entries, tmp2) {
758 		sc = LIO_STAILQ_FIRST_ENTRY(&oct->sc_buf_pool.head,
759 					    struct lio_soft_command, node);
760 
761 		STAILQ_REMOVE_HEAD(&oct->sc_buf_pool.head, entries);
762 
763 		lio_dma_free(sc->size, sc);
764 	}
765 
766 	STAILQ_INIT(&oct->sc_buf_pool.head);
767 
768 	mtx_unlock(&oct->sc_buf_pool.lock);
769 
770 	return (0);
771 }
772 
773 struct lio_soft_command *
774 lio_alloc_soft_command(struct octeon_device *oct, uint32_t datasize,
775 		       uint32_t rdatasize, uint32_t ctxsize)
776 {
777 	struct lio_soft_command	*sc = NULL;
778 	struct lio_stailq_node	*tmp;
779 	uint64_t		dma_addr;
780 	uint32_t		size;
781 	uint32_t		offset = sizeof(struct lio_soft_command);
782 
783 	KASSERT((offset + datasize + rdatasize + ctxsize) <=
784 		LIO_SOFT_COMMAND_BUFFER_SIZE,
785 		("%s, %d, offset + datasize + rdatasize + ctxsize > LIO_SOFT_COMMAND_BUFFER_SIZE",
786 		 __func__, __LINE__));
787 
788 	mtx_lock(&oct->sc_buf_pool.lock);
789 
790 	if (STAILQ_EMPTY(&oct->sc_buf_pool.head)) {
791 		mtx_unlock(&oct->sc_buf_pool.lock);
792 		return (NULL);
793 	}
794 	tmp = STAILQ_LAST(&oct->sc_buf_pool.head, lio_stailq_node, entries);
795 
796 	STAILQ_REMOVE(&oct->sc_buf_pool.head, tmp, lio_stailq_node, entries);
797 
798 	atomic_add_int(&oct->sc_buf_pool.alloc_buf_count, 1);
799 
800 	mtx_unlock(&oct->sc_buf_pool.lock);
801 
802 	sc = (struct lio_soft_command *)tmp;
803 
804 	dma_addr = sc->dma_addr;
805 	size = sc->size;
806 
807 	bzero(sc, sc->size);
808 
809 	sc->dma_addr = dma_addr;
810 	sc->size = size;
811 
812 	if (ctxsize) {
813 		sc->ctxptr = (uint8_t *)sc + offset;
814 		sc->ctxsize = ctxsize;
815 	}
816 
817 	/* Start data at 128 byte boundary */
818 	offset = (offset + ctxsize + 127) & 0xffffff80;
819 
820 	if (datasize) {
821 		sc->virtdptr = (uint8_t *)sc + offset;
822 		sc->dmadptr = dma_addr + offset;
823 		sc->datasize = datasize;
824 	}
825 	/* Start rdata at 128 byte boundary */
826 	offset = (offset + datasize + 127) & 0xffffff80;
827 
828 	if (rdatasize) {
829 		KASSERT(rdatasize >= 16, ("%s, %d, rdatasize < 16", __func__,
830 					  __LINE__));
831 		sc->virtrptr = (uint8_t *)sc + offset;
832 		sc->dmarptr = dma_addr + offset;
833 		sc->rdatasize = rdatasize;
834 		sc->status_word = (uint64_t *)((uint8_t *)(sc->virtrptr) +
835 					       rdatasize - 8);
836 	}
837 	return (sc);
838 }
839 
840 void
841 lio_free_soft_command(struct octeon_device *oct,
842 		      struct lio_soft_command *sc)
843 {
844 
845 	mtx_lock(&oct->sc_buf_pool.lock);
846 
847 	STAILQ_INSERT_TAIL(&oct->sc_buf_pool.head, &sc->node, entries);
848 
849 	atomic_subtract_int(&oct->sc_buf_pool.alloc_buf_count, 1);
850 
851 	mtx_unlock(&oct->sc_buf_pool.lock);
852 }
853