xref: /linux/drivers/net/ethernet/cavium/liquidio/request_manager.c (revision b9b77222d4ff6b5bb8f5d87fca20de0910618bb9)
1 /**********************************************************************
2  * Author: Cavium, Inc.
3  *
4  * Contact: support@cavium.com
5  *          Please include "LiquidIO" in the subject.
6  *
7  * Copyright (c) 2003-2016 Cavium, Inc.
8  *
9  * This file is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License, Version 2, as
11  * published by the Free Software Foundation.
12  *
13  * This file is distributed in the hope that it will be useful, but
14  * AS-IS and WITHOUT ANY WARRANTY; without even the implied warranty
15  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE, TITLE, or
16  * NONINFRINGEMENT.  See the GNU General Public License for more
17  * details.
18  **********************************************************************/
19 #include <linux/pci.h>
20 #include <linux/netdevice.h>
21 #include <linux/vmalloc.h>
22 #include "liquidio_common.h"
23 #include "octeon_droq.h"
24 #include "octeon_iq.h"
25 #include "response_manager.h"
26 #include "octeon_device.h"
27 #include "octeon_main.h"
28 #include "octeon_network.h"
29 #include "cn66xx_device.h"
30 #include "cn23xx_pf_device.h"
31 #include "cn23xx_vf_device.h"
32 
33 struct iq_post_status {
34 	int status;
35 	int index;
36 };
37 
38 static void check_db_timeout(struct work_struct *work);
39 static void  __check_db_timeout(struct octeon_device *oct, u64 iq_no);
40 
41 static void (*reqtype_free_fn[MAX_OCTEON_DEVICES][REQTYPE_LAST + 1]) (void *);
42 
43 static inline int IQ_INSTR_MODE_64B(struct octeon_device *oct, int iq_no)
44 {
45 	struct octeon_instr_queue *iq =
46 	    (struct octeon_instr_queue *)oct->instr_queue[iq_no];
47 	return iq->iqcmd_64B;
48 }
49 
50 #define IQ_INSTR_MODE_32B(oct, iq_no)  (!IQ_INSTR_MODE_64B(oct, iq_no))
51 
52 /* Define this to return the request status comaptible to old code */
53 /*#define OCTEON_USE_OLD_REQ_STATUS*/
54 
55 /* Return 0 on success, 1 on failure */
56 int octeon_init_instr_queue(struct octeon_device *oct,
57 			    union oct_txpciq txpciq,
58 			    u32 num_descs)
59 {
60 	struct octeon_instr_queue *iq;
61 	struct octeon_iq_config *conf = NULL;
62 	u32 iq_no = (u32)txpciq.s.q_no;
63 	u32 q_size;
64 	struct cavium_wq *db_wq;
65 	int numa_node = dev_to_node(&oct->pci_dev->dev);
66 
67 	if (OCTEON_CN6XXX(oct))
68 		conf = &(CFG_GET_IQ_CFG(CHIP_CONF(oct, cn6xxx)));
69 	else if (OCTEON_CN23XX_PF(oct))
70 		conf = &(CFG_GET_IQ_CFG(CHIP_CONF(oct, cn23xx_pf)));
71 	else if (OCTEON_CN23XX_VF(oct))
72 		conf = &(CFG_GET_IQ_CFG(CHIP_CONF(oct, cn23xx_vf)));
73 
74 	if (!conf) {
75 		dev_err(&oct->pci_dev->dev, "Unsupported Chip %x\n",
76 			oct->chip_id);
77 		return 1;
78 	}
79 
80 	q_size = (u32)conf->instr_type * num_descs;
81 
82 	iq = oct->instr_queue[iq_no];
83 
84 	iq->oct_dev = oct;
85 
86 	iq->base_addr = lio_dma_alloc(oct, q_size, &iq->base_addr_dma);
87 	if (!iq->base_addr) {
88 		dev_err(&oct->pci_dev->dev, "Cannot allocate memory for instr queue %d\n",
89 			iq_no);
90 		return 1;
91 	}
92 
93 	iq->max_count = num_descs;
94 
95 	/* Initialize a list to holds requests that have been posted to Octeon
96 	 * but has yet to be fetched by octeon
97 	 */
98 	iq->request_list = vmalloc_node((sizeof(*iq->request_list) * num_descs),
99 					       numa_node);
100 	if (!iq->request_list)
101 		iq->request_list =
102 			vmalloc(array_size(num_descs,
103 					   sizeof(*iq->request_list)));
104 	if (!iq->request_list) {
105 		lio_dma_free(oct, q_size, iq->base_addr, iq->base_addr_dma);
106 		dev_err(&oct->pci_dev->dev, "Alloc failed for IQ[%d] nr free list\n",
107 			iq_no);
108 		return 1;
109 	}
110 
111 	memset(iq->request_list, 0, sizeof(*iq->request_list) * num_descs);
112 
113 	dev_dbg(&oct->pci_dev->dev, "IQ[%d]: base: %p basedma: %llx count: %d\n",
114 		iq_no, iq->base_addr, iq->base_addr_dma, iq->max_count);
115 
116 	iq->txpciq.u64 = txpciq.u64;
117 	iq->fill_threshold = (u32)conf->db_min;
118 	iq->fill_cnt = 0;
119 	iq->host_write_index = 0;
120 	iq->octeon_read_index = 0;
121 	iq->flush_index = 0;
122 	iq->last_db_time = 0;
123 	iq->do_auto_flush = 1;
124 	iq->db_timeout = (u32)conf->db_timeout;
125 	atomic_set(&iq->instr_pending, 0);
126 
127 	/* Initialize the spinlock for this instruction queue */
128 	spin_lock_init(&iq->lock);
129 	spin_lock_init(&iq->post_lock);
130 
131 	spin_lock_init(&iq->iq_flush_running_lock);
132 
133 	oct->io_qmask.iq |= BIT_ULL(iq_no);
134 
135 	/* Set the 32B/64B mode for each input queue */
136 	oct->io_qmask.iq64B |= ((conf->instr_type == 64) << iq_no);
137 	iq->iqcmd_64B = (conf->instr_type == 64);
138 
139 	oct->fn_list.setup_iq_regs(oct, iq_no);
140 
141 	oct->check_db_wq[iq_no].wq = alloc_workqueue("check_iq_db",
142 						     WQ_MEM_RECLAIM,
143 						     0);
144 	if (!oct->check_db_wq[iq_no].wq) {
145 		vfree(iq->request_list);
146 		iq->request_list = NULL;
147 		lio_dma_free(oct, q_size, iq->base_addr, iq->base_addr_dma);
148 		dev_err(&oct->pci_dev->dev, "check db wq create failed for iq %d\n",
149 			iq_no);
150 		return 1;
151 	}
152 
153 	db_wq = &oct->check_db_wq[iq_no];
154 
155 	INIT_DELAYED_WORK(&db_wq->wk.work, check_db_timeout);
156 	db_wq->wk.ctxptr = oct;
157 	db_wq->wk.ctxul = iq_no;
158 	queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(1));
159 
160 	return 0;
161 }
162 
163 int octeon_delete_instr_queue(struct octeon_device *oct, u32 iq_no)
164 {
165 	u64 desc_size = 0, q_size;
166 	struct octeon_instr_queue *iq = oct->instr_queue[iq_no];
167 
168 	cancel_delayed_work_sync(&oct->check_db_wq[iq_no].wk.work);
169 	destroy_workqueue(oct->check_db_wq[iq_no].wq);
170 
171 	if (OCTEON_CN6XXX(oct))
172 		desc_size =
173 		    CFG_GET_IQ_INSTR_TYPE(CHIP_CONF(oct, cn6xxx));
174 	else if (OCTEON_CN23XX_PF(oct))
175 		desc_size =
176 		    CFG_GET_IQ_INSTR_TYPE(CHIP_CONF(oct, cn23xx_pf));
177 	else if (OCTEON_CN23XX_VF(oct))
178 		desc_size =
179 		    CFG_GET_IQ_INSTR_TYPE(CHIP_CONF(oct, cn23xx_vf));
180 
181 	vfree(iq->request_list);
182 
183 	if (iq->base_addr) {
184 		q_size = iq->max_count * desc_size;
185 		lio_dma_free(oct, (u32)q_size, iq->base_addr,
186 			     iq->base_addr_dma);
187 		oct->io_qmask.iq &= ~(1ULL << iq_no);
188 		vfree(oct->instr_queue[iq_no]);
189 		oct->instr_queue[iq_no] = NULL;
190 		oct->num_iqs--;
191 		return 0;
192 	}
193 	return 1;
194 }
195 
196 /* Return 0 on success, 1 on failure */
197 int octeon_setup_iq(struct octeon_device *oct,
198 		    int ifidx,
199 		    int q_index,
200 		    union oct_txpciq txpciq,
201 		    u32 num_descs,
202 		    void *app_ctx)
203 {
204 	u32 iq_no = (u32)txpciq.s.q_no;
205 	int numa_node = dev_to_node(&oct->pci_dev->dev);
206 
207 	if (oct->instr_queue[iq_no]) {
208 		dev_dbg(&oct->pci_dev->dev, "IQ is in use. Cannot create the IQ: %d again\n",
209 			iq_no);
210 		oct->instr_queue[iq_no]->txpciq.u64 = txpciq.u64;
211 		oct->instr_queue[iq_no]->app_ctx = app_ctx;
212 		return 0;
213 	}
214 	oct->instr_queue[iq_no] =
215 	    vmalloc_node(sizeof(struct octeon_instr_queue), numa_node);
216 	if (!oct->instr_queue[iq_no])
217 		oct->instr_queue[iq_no] =
218 		    vmalloc(sizeof(struct octeon_instr_queue));
219 	if (!oct->instr_queue[iq_no])
220 		return 1;
221 
222 	memset(oct->instr_queue[iq_no], 0,
223 	       sizeof(struct octeon_instr_queue));
224 
225 	oct->instr_queue[iq_no]->q_index = q_index;
226 	oct->instr_queue[iq_no]->app_ctx = app_ctx;
227 	oct->instr_queue[iq_no]->ifidx = ifidx;
228 
229 	if (octeon_init_instr_queue(oct, txpciq, num_descs)) {
230 		vfree(oct->instr_queue[iq_no]);
231 		oct->instr_queue[iq_no] = NULL;
232 		return 1;
233 	}
234 
235 	oct->num_iqs++;
236 	if (oct->fn_list.enable_io_queues(oct))
237 		return 1;
238 
239 	return 0;
240 }
241 
242 int lio_wait_for_instr_fetch(struct octeon_device *oct)
243 {
244 	int i, retry = 1000, pending, instr_cnt = 0;
245 
246 	do {
247 		instr_cnt = 0;
248 
249 		for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
250 			if (!(oct->io_qmask.iq & BIT_ULL(i)))
251 				continue;
252 			pending =
253 			    atomic_read(&oct->instr_queue[i]->instr_pending);
254 			if (pending)
255 				__check_db_timeout(oct, i);
256 			instr_cnt += pending;
257 		}
258 
259 		if (instr_cnt == 0)
260 			break;
261 
262 		schedule_timeout_uninterruptible(1);
263 
264 	} while (retry-- && instr_cnt);
265 
266 	return instr_cnt;
267 }
268 
269 static inline void
270 ring_doorbell(struct octeon_device *oct, struct octeon_instr_queue *iq)
271 {
272 	if (atomic_read(&oct->status) == OCT_DEV_RUNNING) {
273 		writel(iq->fill_cnt, iq->doorbell_reg);
274 		/* make sure doorbell write goes through */
275 		mmiowb();
276 		iq->fill_cnt = 0;
277 		iq->last_db_time = jiffies;
278 		return;
279 	}
280 }
281 
282 void
283 octeon_ring_doorbell_locked(struct octeon_device *oct, u32 iq_no)
284 {
285 	struct octeon_instr_queue *iq;
286 
287 	iq = oct->instr_queue[iq_no];
288 	spin_lock(&iq->post_lock);
289 	if (iq->fill_cnt)
290 		ring_doorbell(oct, iq);
291 	spin_unlock(&iq->post_lock);
292 }
293 
294 static inline void __copy_cmd_into_iq(struct octeon_instr_queue *iq,
295 				      u8 *cmd)
296 {
297 	u8 *iqptr, cmdsize;
298 
299 	cmdsize = ((iq->iqcmd_64B) ? 64 : 32);
300 	iqptr = iq->base_addr + (cmdsize * iq->host_write_index);
301 
302 	memcpy(iqptr, cmd, cmdsize);
303 }
304 
305 static inline struct iq_post_status
306 __post_command2(struct octeon_instr_queue *iq, u8 *cmd)
307 {
308 	struct iq_post_status st;
309 
310 	st.status = IQ_SEND_OK;
311 
312 	/* This ensures that the read index does not wrap around to the same
313 	 * position if queue gets full before Octeon could fetch any instr.
314 	 */
315 	if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 1)) {
316 		st.status = IQ_SEND_FAILED;
317 		st.index = -1;
318 		return st;
319 	}
320 
321 	if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 2))
322 		st.status = IQ_SEND_STOP;
323 
324 	__copy_cmd_into_iq(iq, cmd);
325 
326 	/* "index" is returned, host_write_index is modified. */
327 	st.index = iq->host_write_index;
328 	iq->host_write_index = incr_index(iq->host_write_index, 1,
329 					  iq->max_count);
330 	iq->fill_cnt++;
331 
332 	/* Flush the command into memory. We need to be sure the data is in
333 	 * memory before indicating that the instruction is pending.
334 	 */
335 	wmb();
336 
337 	atomic_inc(&iq->instr_pending);
338 
339 	return st;
340 }
341 
342 int
343 octeon_register_reqtype_free_fn(struct octeon_device *oct, int reqtype,
344 				void (*fn)(void *))
345 {
346 	if (reqtype > REQTYPE_LAST) {
347 		dev_err(&oct->pci_dev->dev, "%s: Invalid reqtype: %d\n",
348 			__func__, reqtype);
349 		return -EINVAL;
350 	}
351 
352 	reqtype_free_fn[oct->octeon_id][reqtype] = fn;
353 
354 	return 0;
355 }
356 
357 static inline void
358 __add_to_request_list(struct octeon_instr_queue *iq,
359 		      int idx, void *buf, int reqtype)
360 {
361 	iq->request_list[idx].buf = buf;
362 	iq->request_list[idx].reqtype = reqtype;
363 }
364 
365 /* Can only run in process context */
366 int
367 lio_process_iq_request_list(struct octeon_device *oct,
368 			    struct octeon_instr_queue *iq, u32 napi_budget)
369 {
370 	struct cavium_wq *cwq = &oct->dma_comp_wq;
371 	int reqtype;
372 	void *buf;
373 	u32 old = iq->flush_index;
374 	u32 inst_count = 0;
375 	unsigned int pkts_compl = 0, bytes_compl = 0;
376 	struct octeon_soft_command *sc;
377 	struct octeon_instr_irh *irh;
378 	unsigned long flags;
379 
380 	while (old != iq->octeon_read_index) {
381 		reqtype = iq->request_list[old].reqtype;
382 		buf     = iq->request_list[old].buf;
383 
384 		if (reqtype == REQTYPE_NONE)
385 			goto skip_this;
386 
387 		octeon_update_tx_completion_counters(buf, reqtype, &pkts_compl,
388 						     &bytes_compl);
389 
390 		switch (reqtype) {
391 		case REQTYPE_NORESP_NET:
392 		case REQTYPE_NORESP_NET_SG:
393 		case REQTYPE_RESP_NET_SG:
394 			reqtype_free_fn[oct->octeon_id][reqtype](buf);
395 			break;
396 		case REQTYPE_RESP_NET:
397 		case REQTYPE_SOFT_COMMAND:
398 			sc = buf;
399 
400 			if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct))
401 				irh = (struct octeon_instr_irh *)
402 					&sc->cmd.cmd3.irh;
403 			else
404 				irh = (struct octeon_instr_irh *)
405 					&sc->cmd.cmd2.irh;
406 			if (irh->rflag) {
407 				/* We're expecting a response from Octeon.
408 				 * It's up to lio_process_ordered_list() to
409 				 * process  sc. Add sc to the ordered soft
410 				 * command response list because we expect
411 				 * a response from Octeon.
412 				 */
413 				spin_lock_irqsave
414 					(&oct->response_list
415 					 [OCTEON_ORDERED_SC_LIST].lock,
416 					 flags);
417 				atomic_inc(&oct->response_list
418 					[OCTEON_ORDERED_SC_LIST].
419 					pending_req_count);
420 				list_add_tail(&sc->node, &oct->response_list
421 					[OCTEON_ORDERED_SC_LIST].head);
422 				spin_unlock_irqrestore
423 					(&oct->response_list
424 					 [OCTEON_ORDERED_SC_LIST].lock,
425 					 flags);
426 			} else {
427 				if (sc->callback) {
428 					/* This callback must not sleep */
429 					sc->callback(oct, OCTEON_REQUEST_DONE,
430 						     sc->callback_arg);
431 				}
432 			}
433 			break;
434 		default:
435 			dev_err(&oct->pci_dev->dev,
436 				"%s Unknown reqtype: %d buf: %p at idx %d\n",
437 				__func__, reqtype, buf, old);
438 		}
439 
440 		iq->request_list[old].buf = NULL;
441 		iq->request_list[old].reqtype = 0;
442 
443  skip_this:
444 		inst_count++;
445 		old = incr_index(old, 1, iq->max_count);
446 
447 		if ((napi_budget) && (inst_count >= napi_budget))
448 			break;
449 	}
450 	if (bytes_compl)
451 		octeon_report_tx_completion_to_bql(iq->app_ctx, pkts_compl,
452 						   bytes_compl);
453 	iq->flush_index = old;
454 
455 	if (atomic_read(&oct->response_list
456 			[OCTEON_ORDERED_SC_LIST].pending_req_count))
457 		queue_delayed_work(cwq->wq, &cwq->wk.work, msecs_to_jiffies(1));
458 
459 	return inst_count;
460 }
461 
462 /* Can only be called from process context */
463 int
464 octeon_flush_iq(struct octeon_device *oct, struct octeon_instr_queue *iq,
465 		u32 napi_budget)
466 {
467 	u32 inst_processed = 0;
468 	u32 tot_inst_processed = 0;
469 	int tx_done = 1;
470 
471 	if (!spin_trylock(&iq->iq_flush_running_lock))
472 		return tx_done;
473 
474 	spin_lock_bh(&iq->lock);
475 
476 	iq->octeon_read_index = oct->fn_list.update_iq_read_idx(iq);
477 
478 	do {
479 		/* Process any outstanding IQ packets. */
480 		if (iq->flush_index == iq->octeon_read_index)
481 			break;
482 
483 		if (napi_budget)
484 			inst_processed =
485 				lio_process_iq_request_list(oct, iq,
486 							    napi_budget -
487 							    tot_inst_processed);
488 		else
489 			inst_processed =
490 				lio_process_iq_request_list(oct, iq, 0);
491 
492 		if (inst_processed) {
493 			atomic_sub(inst_processed, &iq->instr_pending);
494 			iq->stats.instr_processed += inst_processed;
495 		}
496 
497 		tot_inst_processed += inst_processed;
498 	} while (tot_inst_processed < napi_budget);
499 
500 	if (napi_budget && (tot_inst_processed >= napi_budget))
501 		tx_done = 0;
502 
503 	iq->last_db_time = jiffies;
504 
505 	spin_unlock_bh(&iq->lock);
506 
507 	spin_unlock(&iq->iq_flush_running_lock);
508 
509 	return tx_done;
510 }
511 
512 /* Process instruction queue after timeout.
513  * This routine gets called from a workqueue or when removing the module.
514  */
515 static void __check_db_timeout(struct octeon_device *oct, u64 iq_no)
516 {
517 	struct octeon_instr_queue *iq;
518 	u64 next_time;
519 
520 	if (!oct)
521 		return;
522 
523 	iq = oct->instr_queue[iq_no];
524 	if (!iq)
525 		return;
526 
527 	/* return immediately, if no work pending */
528 	if (!atomic_read(&iq->instr_pending))
529 		return;
530 	/* If jiffies - last_db_time < db_timeout do nothing  */
531 	next_time = iq->last_db_time + iq->db_timeout;
532 	if (!time_after(jiffies, (unsigned long)next_time))
533 		return;
534 	iq->last_db_time = jiffies;
535 
536 	/* Flush the instruction queue */
537 	octeon_flush_iq(oct, iq, 0);
538 
539 	lio_enable_irq(NULL, iq);
540 }
541 
542 /* Called by the Poll thread at regular intervals to check the instruction
543  * queue for commands to be posted and for commands that were fetched by Octeon.
544  */
545 static void check_db_timeout(struct work_struct *work)
546 {
547 	struct cavium_wk *wk = (struct cavium_wk *)work;
548 	struct octeon_device *oct = (struct octeon_device *)wk->ctxptr;
549 	u64 iq_no = wk->ctxul;
550 	struct cavium_wq *db_wq = &oct->check_db_wq[iq_no];
551 	u32 delay = 10;
552 
553 	__check_db_timeout(oct, iq_no);
554 	queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(delay));
555 }
556 
557 int
558 octeon_send_command(struct octeon_device *oct, u32 iq_no,
559 		    u32 force_db, void *cmd, void *buf,
560 		    u32 datasize, u32 reqtype)
561 {
562 	int xmit_stopped;
563 	struct iq_post_status st;
564 	struct octeon_instr_queue *iq = oct->instr_queue[iq_no];
565 
566 	/* Get the lock and prevent other tasks and tx interrupt handler from
567 	 * running.
568 	 */
569 	spin_lock_bh(&iq->post_lock);
570 
571 	st = __post_command2(iq, cmd);
572 
573 	if (st.status != IQ_SEND_FAILED) {
574 		xmit_stopped = octeon_report_sent_bytes_to_bql(buf, reqtype);
575 		__add_to_request_list(iq, st.index, buf, reqtype);
576 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, bytes_sent, datasize);
577 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_posted, 1);
578 
579 		if (iq->fill_cnt >= MAX_OCTEON_FILL_COUNT || force_db ||
580 		    xmit_stopped || st.status == IQ_SEND_STOP)
581 			ring_doorbell(oct, iq);
582 	} else {
583 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_dropped, 1);
584 	}
585 
586 	spin_unlock_bh(&iq->post_lock);
587 
588 	/* This is only done here to expedite packets being flushed
589 	 * for cases where there are no IQ completion interrupts.
590 	 */
591 
592 	return st.status;
593 }
594 
595 void
596 octeon_prepare_soft_command(struct octeon_device *oct,
597 			    struct octeon_soft_command *sc,
598 			    u8 opcode,
599 			    u8 subcode,
600 			    u32 irh_ossp,
601 			    u64 ossp0,
602 			    u64 ossp1)
603 {
604 	struct octeon_config *oct_cfg;
605 	struct octeon_instr_ih2 *ih2;
606 	struct octeon_instr_ih3 *ih3;
607 	struct octeon_instr_pki_ih3 *pki_ih3;
608 	struct octeon_instr_irh *irh;
609 	struct octeon_instr_rdp *rdp;
610 
611 	WARN_ON(opcode > 15);
612 	WARN_ON(subcode > 127);
613 
614 	oct_cfg = octeon_get_conf(oct);
615 
616 	if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
617 		ih3 = (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
618 
619 		ih3->pkind = oct->instr_queue[sc->iq_no]->txpciq.s.pkind;
620 
621 		pki_ih3 = (struct octeon_instr_pki_ih3 *)&sc->cmd.cmd3.pki_ih3;
622 
623 		pki_ih3->w           = 1;
624 		pki_ih3->raw         = 1;
625 		pki_ih3->utag        = 1;
626 		pki_ih3->uqpg        =
627 			oct->instr_queue[sc->iq_no]->txpciq.s.use_qpg;
628 		pki_ih3->utt         = 1;
629 		pki_ih3->tag     = LIO_CONTROL;
630 		pki_ih3->tagtype = ATOMIC_TAG;
631 		pki_ih3->qpg         =
632 			oct->instr_queue[sc->iq_no]->txpciq.s.ctrl_qpg;
633 
634 		pki_ih3->pm          = 0x7;
635 		pki_ih3->sl          = 8;
636 
637 		if (sc->datasize)
638 			ih3->dlengsz = sc->datasize;
639 
640 		irh            = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
641 		irh->opcode    = opcode;
642 		irh->subcode   = subcode;
643 
644 		/* opcode/subcode specific parameters (ossp) */
645 		irh->ossp       = irh_ossp;
646 		sc->cmd.cmd3.ossp[0] = ossp0;
647 		sc->cmd.cmd3.ossp[1] = ossp1;
648 
649 		if (sc->rdatasize) {
650 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd3.rdp;
651 			rdp->pcie_port = oct->pcie_port;
652 			rdp->rlen      = sc->rdatasize;
653 
654 			irh->rflag =  1;
655 			/*PKI IH3*/
656 			/* pki_ih3 irh+ossp[0]+ossp[1]+rdp+rptr = 48 bytes */
657 			ih3->fsz    = LIO_SOFTCMDRESP_IH3;
658 		} else {
659 			irh->rflag =  0;
660 			/*PKI IH3*/
661 			/* pki_h3 + irh + ossp[0] + ossp[1] = 32 bytes */
662 			ih3->fsz    = LIO_PCICMD_O3;
663 		}
664 
665 	} else {
666 		ih2          = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
667 		ih2->tagtype = ATOMIC_TAG;
668 		ih2->tag     = LIO_CONTROL;
669 		ih2->raw     = 1;
670 		ih2->grp     = CFG_GET_CTRL_Q_GRP(oct_cfg);
671 
672 		if (sc->datasize) {
673 			ih2->dlengsz = sc->datasize;
674 			ih2->rs = 1;
675 		}
676 
677 		irh            = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
678 		irh->opcode    = opcode;
679 		irh->subcode   = subcode;
680 
681 		/* opcode/subcode specific parameters (ossp) */
682 		irh->ossp       = irh_ossp;
683 		sc->cmd.cmd2.ossp[0] = ossp0;
684 		sc->cmd.cmd2.ossp[1] = ossp1;
685 
686 		if (sc->rdatasize) {
687 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd2.rdp;
688 			rdp->pcie_port = oct->pcie_port;
689 			rdp->rlen      = sc->rdatasize;
690 
691 			irh->rflag =  1;
692 			/* irh+ossp[0]+ossp[1]+rdp+rptr = 40 bytes */
693 			ih2->fsz   = LIO_SOFTCMDRESP_IH2;
694 		} else {
695 			irh->rflag =  0;
696 			/* irh + ossp[0] + ossp[1] = 24 bytes */
697 			ih2->fsz   = LIO_PCICMD_O2;
698 		}
699 	}
700 }
701 
702 int octeon_send_soft_command(struct octeon_device *oct,
703 			     struct octeon_soft_command *sc)
704 {
705 	struct octeon_instr_ih2 *ih2;
706 	struct octeon_instr_ih3 *ih3;
707 	struct octeon_instr_irh *irh;
708 	u32 len;
709 
710 	if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
711 		ih3 =  (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
712 		if (ih3->dlengsz) {
713 			WARN_ON(!sc->dmadptr);
714 			sc->cmd.cmd3.dptr = sc->dmadptr;
715 		}
716 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
717 		if (irh->rflag) {
718 			WARN_ON(!sc->dmarptr);
719 			WARN_ON(!sc->status_word);
720 			*sc->status_word = COMPLETION_WORD_INIT;
721 			sc->cmd.cmd3.rptr = sc->dmarptr;
722 		}
723 		len = (u32)ih3->dlengsz;
724 	} else {
725 		ih2 = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
726 		if (ih2->dlengsz) {
727 			WARN_ON(!sc->dmadptr);
728 			sc->cmd.cmd2.dptr = sc->dmadptr;
729 		}
730 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
731 		if (irh->rflag) {
732 			WARN_ON(!sc->dmarptr);
733 			WARN_ON(!sc->status_word);
734 			*sc->status_word = COMPLETION_WORD_INIT;
735 			sc->cmd.cmd2.rptr = sc->dmarptr;
736 		}
737 		len = (u32)ih2->dlengsz;
738 	}
739 
740 	if (sc->wait_time)
741 		sc->timeout = jiffies + sc->wait_time;
742 
743 	return (octeon_send_command(oct, sc->iq_no, 1, &sc->cmd, sc,
744 				    len, REQTYPE_SOFT_COMMAND));
745 }
746 
747 int octeon_setup_sc_buffer_pool(struct octeon_device *oct)
748 {
749 	int i;
750 	u64 dma_addr;
751 	struct octeon_soft_command *sc;
752 
753 	INIT_LIST_HEAD(&oct->sc_buf_pool.head);
754 	spin_lock_init(&oct->sc_buf_pool.lock);
755 	atomic_set(&oct->sc_buf_pool.alloc_buf_count, 0);
756 
757 	for (i = 0; i < MAX_SOFT_COMMAND_BUFFERS; i++) {
758 		sc = (struct octeon_soft_command *)
759 			lio_dma_alloc(oct,
760 				      SOFT_COMMAND_BUFFER_SIZE,
761 					  (dma_addr_t *)&dma_addr);
762 		if (!sc) {
763 			octeon_free_sc_buffer_pool(oct);
764 			return 1;
765 		}
766 
767 		sc->dma_addr = dma_addr;
768 		sc->size = SOFT_COMMAND_BUFFER_SIZE;
769 
770 		list_add_tail(&sc->node, &oct->sc_buf_pool.head);
771 	}
772 
773 	return 0;
774 }
775 
776 int octeon_free_sc_buffer_pool(struct octeon_device *oct)
777 {
778 	struct list_head *tmp, *tmp2;
779 	struct octeon_soft_command *sc;
780 
781 	spin_lock_bh(&oct->sc_buf_pool.lock);
782 
783 	list_for_each_safe(tmp, tmp2, &oct->sc_buf_pool.head) {
784 		list_del(tmp);
785 
786 		sc = (struct octeon_soft_command *)tmp;
787 
788 		lio_dma_free(oct, sc->size, sc, sc->dma_addr);
789 	}
790 
791 	INIT_LIST_HEAD(&oct->sc_buf_pool.head);
792 
793 	spin_unlock_bh(&oct->sc_buf_pool.lock);
794 
795 	return 0;
796 }
797 
798 struct octeon_soft_command *octeon_alloc_soft_command(struct octeon_device *oct,
799 						      u32 datasize,
800 						      u32 rdatasize,
801 						      u32 ctxsize)
802 {
803 	u64 dma_addr;
804 	u32 size;
805 	u32 offset = sizeof(struct octeon_soft_command);
806 	struct octeon_soft_command *sc = NULL;
807 	struct list_head *tmp;
808 
809 	WARN_ON((offset + datasize + rdatasize + ctxsize) >
810 	       SOFT_COMMAND_BUFFER_SIZE);
811 
812 	spin_lock_bh(&oct->sc_buf_pool.lock);
813 
814 	if (list_empty(&oct->sc_buf_pool.head)) {
815 		spin_unlock_bh(&oct->sc_buf_pool.lock);
816 		return NULL;
817 	}
818 
819 	list_for_each(tmp, &oct->sc_buf_pool.head)
820 		break;
821 
822 	list_del(tmp);
823 
824 	atomic_inc(&oct->sc_buf_pool.alloc_buf_count);
825 
826 	spin_unlock_bh(&oct->sc_buf_pool.lock);
827 
828 	sc = (struct octeon_soft_command *)tmp;
829 
830 	dma_addr = sc->dma_addr;
831 	size = sc->size;
832 
833 	memset(sc, 0, sc->size);
834 
835 	sc->dma_addr = dma_addr;
836 	sc->size = size;
837 
838 	if (ctxsize) {
839 		sc->ctxptr = (u8 *)sc + offset;
840 		sc->ctxsize = ctxsize;
841 	}
842 
843 	/* Start data at 128 byte boundary */
844 	offset = (offset + ctxsize + 127) & 0xffffff80;
845 
846 	if (datasize) {
847 		sc->virtdptr = (u8 *)sc + offset;
848 		sc->dmadptr = dma_addr + offset;
849 		sc->datasize = datasize;
850 	}
851 
852 	/* Start rdata at 128 byte boundary */
853 	offset = (offset + datasize + 127) & 0xffffff80;
854 
855 	if (rdatasize) {
856 		WARN_ON(rdatasize < 16);
857 		sc->virtrptr = (u8 *)sc + offset;
858 		sc->dmarptr = dma_addr + offset;
859 		sc->rdatasize = rdatasize;
860 		sc->status_word = (u64 *)((u8 *)(sc->virtrptr) + rdatasize - 8);
861 	}
862 
863 	return sc;
864 }
865 
866 void octeon_free_soft_command(struct octeon_device *oct,
867 			      struct octeon_soft_command *sc)
868 {
869 	spin_lock_bh(&oct->sc_buf_pool.lock);
870 
871 	list_add_tail(&sc->node, &oct->sc_buf_pool.head);
872 
873 	atomic_dec(&oct->sc_buf_pool.alloc_buf_count);
874 
875 	spin_unlock_bh(&oct->sc_buf_pool.lock);
876 }
877