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: %pad 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 	iq->pkts_processed = 0;
127 
128 	/* Initialize the spinlock for this instruction queue */
129 	spin_lock_init(&iq->lock);
130 	if (iq_no == 0) {
131 		iq->allow_soft_cmds = true;
132 		spin_lock_init(&iq->post_lock);
133 	} else {
134 		iq->allow_soft_cmds = false;
135 	}
136 
137 	spin_lock_init(&iq->iq_flush_running_lock);
138 
139 	oct->io_qmask.iq |= BIT_ULL(iq_no);
140 
141 	/* Set the 32B/64B mode for each input queue */
142 	oct->io_qmask.iq64B |= ((conf->instr_type == 64) << iq_no);
143 	iq->iqcmd_64B = (conf->instr_type == 64);
144 
145 	oct->fn_list.setup_iq_regs(oct, iq_no);
146 
147 	oct->check_db_wq[iq_no].wq = alloc_workqueue("check_iq_db",
148 						     WQ_MEM_RECLAIM,
149 						     0);
150 	if (!oct->check_db_wq[iq_no].wq) {
151 		vfree(iq->request_list);
152 		iq->request_list = NULL;
153 		lio_dma_free(oct, q_size, iq->base_addr, iq->base_addr_dma);
154 		dev_err(&oct->pci_dev->dev, "check db wq create failed for iq %d\n",
155 			iq_no);
156 		return 1;
157 	}
158 
159 	db_wq = &oct->check_db_wq[iq_no];
160 
161 	INIT_DELAYED_WORK(&db_wq->wk.work, check_db_timeout);
162 	db_wq->wk.ctxptr = oct;
163 	db_wq->wk.ctxul = iq_no;
164 	queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(1));
165 
166 	return 0;
167 }
168 
169 int octeon_delete_instr_queue(struct octeon_device *oct, u32 iq_no)
170 {
171 	u64 desc_size = 0, q_size;
172 	struct octeon_instr_queue *iq = oct->instr_queue[iq_no];
173 
174 	cancel_delayed_work_sync(&oct->check_db_wq[iq_no].wk.work);
175 	destroy_workqueue(oct->check_db_wq[iq_no].wq);
176 
177 	if (OCTEON_CN6XXX(oct))
178 		desc_size =
179 		    CFG_GET_IQ_INSTR_TYPE(CHIP_CONF(oct, cn6xxx));
180 	else if (OCTEON_CN23XX_PF(oct))
181 		desc_size =
182 		    CFG_GET_IQ_INSTR_TYPE(CHIP_CONF(oct, cn23xx_pf));
183 	else if (OCTEON_CN23XX_VF(oct))
184 		desc_size =
185 		    CFG_GET_IQ_INSTR_TYPE(CHIP_CONF(oct, cn23xx_vf));
186 
187 	vfree(iq->request_list);
188 
189 	if (iq->base_addr) {
190 		q_size = iq->max_count * desc_size;
191 		lio_dma_free(oct, (u32)q_size, iq->base_addr,
192 			     iq->base_addr_dma);
193 		oct->io_qmask.iq &= ~(1ULL << iq_no);
194 		vfree(oct->instr_queue[iq_no]);
195 		oct->instr_queue[iq_no] = NULL;
196 		oct->num_iqs--;
197 		return 0;
198 	}
199 	return 1;
200 }
201 
202 /* Return 0 on success, 1 on failure */
203 int octeon_setup_iq(struct octeon_device *oct,
204 		    int ifidx,
205 		    int q_index,
206 		    union oct_txpciq txpciq,
207 		    u32 num_descs,
208 		    void *app_ctx)
209 {
210 	u32 iq_no = (u32)txpciq.s.q_no;
211 	int numa_node = dev_to_node(&oct->pci_dev->dev);
212 
213 	if (oct->instr_queue[iq_no]) {
214 		dev_dbg(&oct->pci_dev->dev, "IQ is in use. Cannot create the IQ: %d again\n",
215 			iq_no);
216 		oct->instr_queue[iq_no]->txpciq.u64 = txpciq.u64;
217 		oct->instr_queue[iq_no]->app_ctx = app_ctx;
218 		return 0;
219 	}
220 	oct->instr_queue[iq_no] =
221 	    vmalloc_node(sizeof(struct octeon_instr_queue), numa_node);
222 	if (!oct->instr_queue[iq_no])
223 		oct->instr_queue[iq_no] =
224 		    vmalloc(sizeof(struct octeon_instr_queue));
225 	if (!oct->instr_queue[iq_no])
226 		return 1;
227 
228 	memset(oct->instr_queue[iq_no], 0,
229 	       sizeof(struct octeon_instr_queue));
230 
231 	oct->instr_queue[iq_no]->q_index = q_index;
232 	oct->instr_queue[iq_no]->app_ctx = app_ctx;
233 	oct->instr_queue[iq_no]->ifidx = ifidx;
234 
235 	if (octeon_init_instr_queue(oct, txpciq, num_descs)) {
236 		vfree(oct->instr_queue[iq_no]);
237 		oct->instr_queue[iq_no] = NULL;
238 		return 1;
239 	}
240 
241 	oct->num_iqs++;
242 	if (oct->fn_list.enable_io_queues(oct))
243 		return 1;
244 
245 	return 0;
246 }
247 
248 int lio_wait_for_instr_fetch(struct octeon_device *oct)
249 {
250 	int i, retry = 1000, pending, instr_cnt = 0;
251 
252 	do {
253 		instr_cnt = 0;
254 
255 		for (i = 0; i < MAX_OCTEON_INSTR_QUEUES(oct); i++) {
256 			if (!(oct->io_qmask.iq & BIT_ULL(i)))
257 				continue;
258 			pending =
259 			    atomic_read(&oct->instr_queue[i]->instr_pending);
260 			if (pending)
261 				__check_db_timeout(oct, i);
262 			instr_cnt += pending;
263 		}
264 
265 		if (instr_cnt == 0)
266 			break;
267 
268 		schedule_timeout_uninterruptible(1);
269 
270 	} while (retry-- && instr_cnt);
271 
272 	return instr_cnt;
273 }
274 
275 static inline void
276 ring_doorbell(struct octeon_device *oct, struct octeon_instr_queue *iq)
277 {
278 	if (atomic_read(&oct->status) == OCT_DEV_RUNNING) {
279 		writel(iq->fill_cnt, iq->doorbell_reg);
280 		/* make sure doorbell write goes through */
281 		mmiowb();
282 		iq->fill_cnt = 0;
283 		iq->last_db_time = jiffies;
284 		return;
285 	}
286 }
287 
288 void
289 octeon_ring_doorbell_locked(struct octeon_device *oct, u32 iq_no)
290 {
291 	struct octeon_instr_queue *iq;
292 
293 	iq = oct->instr_queue[iq_no];
294 	spin_lock(&iq->post_lock);
295 	if (iq->fill_cnt)
296 		ring_doorbell(oct, iq);
297 	spin_unlock(&iq->post_lock);
298 }
299 
300 static inline void __copy_cmd_into_iq(struct octeon_instr_queue *iq,
301 				      u8 *cmd)
302 {
303 	u8 *iqptr, cmdsize;
304 
305 	cmdsize = ((iq->iqcmd_64B) ? 64 : 32);
306 	iqptr = iq->base_addr + (cmdsize * iq->host_write_index);
307 
308 	memcpy(iqptr, cmd, cmdsize);
309 }
310 
311 static inline struct iq_post_status
312 __post_command2(struct octeon_instr_queue *iq, u8 *cmd)
313 {
314 	struct iq_post_status st;
315 
316 	st.status = IQ_SEND_OK;
317 
318 	/* This ensures that the read index does not wrap around to the same
319 	 * position if queue gets full before Octeon could fetch any instr.
320 	 */
321 	if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 1)) {
322 		st.status = IQ_SEND_FAILED;
323 		st.index = -1;
324 		return st;
325 	}
326 
327 	if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 2))
328 		st.status = IQ_SEND_STOP;
329 
330 	__copy_cmd_into_iq(iq, cmd);
331 
332 	/* "index" is returned, host_write_index is modified. */
333 	st.index = iq->host_write_index;
334 	iq->host_write_index = incr_index(iq->host_write_index, 1,
335 					  iq->max_count);
336 	iq->fill_cnt++;
337 
338 	/* Flush the command into memory. We need to be sure the data is in
339 	 * memory before indicating that the instruction is pending.
340 	 */
341 	wmb();
342 
343 	atomic_inc(&iq->instr_pending);
344 
345 	return st;
346 }
347 
348 int
349 octeon_register_reqtype_free_fn(struct octeon_device *oct, int reqtype,
350 				void (*fn)(void *))
351 {
352 	if (reqtype > REQTYPE_LAST) {
353 		dev_err(&oct->pci_dev->dev, "%s: Invalid reqtype: %d\n",
354 			__func__, reqtype);
355 		return -EINVAL;
356 	}
357 
358 	reqtype_free_fn[oct->octeon_id][reqtype] = fn;
359 
360 	return 0;
361 }
362 
363 static inline void
364 __add_to_request_list(struct octeon_instr_queue *iq,
365 		      int idx, void *buf, int reqtype)
366 {
367 	iq->request_list[idx].buf = buf;
368 	iq->request_list[idx].reqtype = reqtype;
369 }
370 
371 /* Can only run in process context */
372 int
373 lio_process_iq_request_list(struct octeon_device *oct,
374 			    struct octeon_instr_queue *iq, u32 napi_budget)
375 {
376 	struct cavium_wq *cwq = &oct->dma_comp_wq;
377 	int reqtype;
378 	void *buf;
379 	u32 old = iq->flush_index;
380 	u32 inst_count = 0;
381 	unsigned int pkts_compl = 0, bytes_compl = 0;
382 	struct octeon_soft_command *sc;
383 	unsigned long flags;
384 
385 	while (old != iq->octeon_read_index) {
386 		reqtype = iq->request_list[old].reqtype;
387 		buf     = iq->request_list[old].buf;
388 
389 		if (reqtype == REQTYPE_NONE)
390 			goto skip_this;
391 
392 		octeon_update_tx_completion_counters(buf, reqtype, &pkts_compl,
393 						     &bytes_compl);
394 
395 		switch (reqtype) {
396 		case REQTYPE_NORESP_NET:
397 		case REQTYPE_NORESP_NET_SG:
398 		case REQTYPE_RESP_NET_SG:
399 			reqtype_free_fn[oct->octeon_id][reqtype](buf);
400 			break;
401 		case REQTYPE_RESP_NET:
402 		case REQTYPE_SOFT_COMMAND:
403 			sc = buf;
404 			/* We're expecting a response from Octeon.
405 			 * It's up to lio_process_ordered_list() to
406 			 * process  sc. Add sc to the ordered soft
407 			 * command response list because we expect
408 			 * a response from Octeon.
409 			 */
410 			spin_lock_irqsave(&oct->response_list
411 					  [OCTEON_ORDERED_SC_LIST].lock, flags);
412 			atomic_inc(&oct->response_list
413 				   [OCTEON_ORDERED_SC_LIST].pending_req_count);
414 			list_add_tail(&sc->node, &oct->response_list
415 				[OCTEON_ORDERED_SC_LIST].head);
416 			spin_unlock_irqrestore(&oct->response_list
417 					       [OCTEON_ORDERED_SC_LIST].lock,
418 					       flags);
419 			break;
420 		default:
421 			dev_err(&oct->pci_dev->dev,
422 				"%s Unknown reqtype: %d buf: %p at idx %d\n",
423 				__func__, reqtype, buf, old);
424 		}
425 
426 		iq->request_list[old].buf = NULL;
427 		iq->request_list[old].reqtype = 0;
428 
429  skip_this:
430 		inst_count++;
431 		old = incr_index(old, 1, iq->max_count);
432 
433 		if ((napi_budget) && (inst_count >= napi_budget))
434 			break;
435 	}
436 	if (bytes_compl)
437 		octeon_report_tx_completion_to_bql(iq->app_ctx, pkts_compl,
438 						   bytes_compl);
439 	iq->flush_index = old;
440 
441 	if (atomic_read(&oct->response_list
442 			[OCTEON_ORDERED_SC_LIST].pending_req_count))
443 		queue_work(cwq->wq, &cwq->wk.work.work);
444 
445 	return inst_count;
446 }
447 
448 /* Can only be called from process context */
449 int
450 octeon_flush_iq(struct octeon_device *oct, struct octeon_instr_queue *iq,
451 		u32 napi_budget)
452 {
453 	u32 inst_processed = 0;
454 	u32 tot_inst_processed = 0;
455 	int tx_done = 1;
456 
457 	if (!spin_trylock(&iq->iq_flush_running_lock))
458 		return tx_done;
459 
460 	spin_lock_bh(&iq->lock);
461 
462 	iq->octeon_read_index = oct->fn_list.update_iq_read_idx(iq);
463 
464 	do {
465 		/* Process any outstanding IQ packets. */
466 		if (iq->flush_index == iq->octeon_read_index)
467 			break;
468 
469 		if (napi_budget)
470 			inst_processed =
471 				lio_process_iq_request_list(oct, iq,
472 							    napi_budget -
473 							    tot_inst_processed);
474 		else
475 			inst_processed =
476 				lio_process_iq_request_list(oct, iq, 0);
477 
478 		if (inst_processed) {
479 			iq->pkts_processed += inst_processed;
480 			atomic_sub(inst_processed, &iq->instr_pending);
481 			iq->stats.instr_processed += inst_processed;
482 		}
483 
484 		tot_inst_processed += inst_processed;
485 	} while (tot_inst_processed < napi_budget);
486 
487 	if (napi_budget && (tot_inst_processed >= napi_budget))
488 		tx_done = 0;
489 
490 	iq->last_db_time = jiffies;
491 
492 	spin_unlock_bh(&iq->lock);
493 
494 	spin_unlock(&iq->iq_flush_running_lock);
495 
496 	return tx_done;
497 }
498 
499 /* Process instruction queue after timeout.
500  * This routine gets called from a workqueue or when removing the module.
501  */
502 static void __check_db_timeout(struct octeon_device *oct, u64 iq_no)
503 {
504 	struct octeon_instr_queue *iq;
505 	u64 next_time;
506 
507 	if (!oct)
508 		return;
509 
510 	iq = oct->instr_queue[iq_no];
511 	if (!iq)
512 		return;
513 
514 	/* return immediately, if no work pending */
515 	if (!atomic_read(&iq->instr_pending))
516 		return;
517 	/* If jiffies - last_db_time < db_timeout do nothing  */
518 	next_time = iq->last_db_time + iq->db_timeout;
519 	if (!time_after(jiffies, (unsigned long)next_time))
520 		return;
521 	iq->last_db_time = jiffies;
522 
523 	/* Flush the instruction queue */
524 	octeon_flush_iq(oct, iq, 0);
525 
526 	lio_enable_irq(NULL, iq);
527 }
528 
529 /* Called by the Poll thread at regular intervals to check the instruction
530  * queue for commands to be posted and for commands that were fetched by Octeon.
531  */
532 static void check_db_timeout(struct work_struct *work)
533 {
534 	struct cavium_wk *wk = (struct cavium_wk *)work;
535 	struct octeon_device *oct = (struct octeon_device *)wk->ctxptr;
536 	u64 iq_no = wk->ctxul;
537 	struct cavium_wq *db_wq = &oct->check_db_wq[iq_no];
538 	u32 delay = 10;
539 
540 	__check_db_timeout(oct, iq_no);
541 	queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(delay));
542 }
543 
544 int
545 octeon_send_command(struct octeon_device *oct, u32 iq_no,
546 		    u32 force_db, void *cmd, void *buf,
547 		    u32 datasize, u32 reqtype)
548 {
549 	int xmit_stopped;
550 	struct iq_post_status st;
551 	struct octeon_instr_queue *iq = oct->instr_queue[iq_no];
552 
553 	/* Get the lock and prevent other tasks and tx interrupt handler from
554 	 * running.
555 	 */
556 	if (iq->allow_soft_cmds)
557 		spin_lock_bh(&iq->post_lock);
558 
559 	st = __post_command2(iq, cmd);
560 
561 	if (st.status != IQ_SEND_FAILED) {
562 		xmit_stopped = octeon_report_sent_bytes_to_bql(buf, reqtype);
563 		__add_to_request_list(iq, st.index, buf, reqtype);
564 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, bytes_sent, datasize);
565 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_posted, 1);
566 
567 		if (iq->fill_cnt >= MAX_OCTEON_FILL_COUNT || force_db ||
568 		    xmit_stopped || st.status == IQ_SEND_STOP)
569 			ring_doorbell(oct, iq);
570 	} else {
571 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_dropped, 1);
572 	}
573 
574 	if (iq->allow_soft_cmds)
575 		spin_unlock_bh(&iq->post_lock);
576 
577 	/* This is only done here to expedite packets being flushed
578 	 * for cases where there are no IQ completion interrupts.
579 	 */
580 
581 	return st.status;
582 }
583 
584 void
585 octeon_prepare_soft_command(struct octeon_device *oct,
586 			    struct octeon_soft_command *sc,
587 			    u8 opcode,
588 			    u8 subcode,
589 			    u32 irh_ossp,
590 			    u64 ossp0,
591 			    u64 ossp1)
592 {
593 	struct octeon_config *oct_cfg;
594 	struct octeon_instr_ih2 *ih2;
595 	struct octeon_instr_ih3 *ih3;
596 	struct octeon_instr_pki_ih3 *pki_ih3;
597 	struct octeon_instr_irh *irh;
598 	struct octeon_instr_rdp *rdp;
599 
600 	WARN_ON(opcode > 15);
601 	WARN_ON(subcode > 127);
602 
603 	oct_cfg = octeon_get_conf(oct);
604 
605 	if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
606 		ih3 = (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
607 
608 		ih3->pkind = oct->instr_queue[sc->iq_no]->txpciq.s.pkind;
609 
610 		pki_ih3 = (struct octeon_instr_pki_ih3 *)&sc->cmd.cmd3.pki_ih3;
611 
612 		pki_ih3->w           = 1;
613 		pki_ih3->raw         = 1;
614 		pki_ih3->utag        = 1;
615 		pki_ih3->uqpg        =
616 			oct->instr_queue[sc->iq_no]->txpciq.s.use_qpg;
617 		pki_ih3->utt         = 1;
618 		pki_ih3->tag     = LIO_CONTROL;
619 		pki_ih3->tagtype = ATOMIC_TAG;
620 		pki_ih3->qpg         =
621 			oct->instr_queue[sc->iq_no]->txpciq.s.ctrl_qpg;
622 
623 		pki_ih3->pm          = 0x7;
624 		pki_ih3->sl          = 8;
625 
626 		if (sc->datasize)
627 			ih3->dlengsz = sc->datasize;
628 
629 		irh            = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
630 		irh->opcode    = opcode;
631 		irh->subcode   = subcode;
632 
633 		/* opcode/subcode specific parameters (ossp) */
634 		irh->ossp       = irh_ossp;
635 		sc->cmd.cmd3.ossp[0] = ossp0;
636 		sc->cmd.cmd3.ossp[1] = ossp1;
637 
638 		if (sc->rdatasize) {
639 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd3.rdp;
640 			rdp->pcie_port = oct->pcie_port;
641 			rdp->rlen      = sc->rdatasize;
642 
643 			irh->rflag =  1;
644 			/*PKI IH3*/
645 			/* pki_ih3 irh+ossp[0]+ossp[1]+rdp+rptr = 48 bytes */
646 			ih3->fsz    = LIO_SOFTCMDRESP_IH3;
647 		} else {
648 			irh->rflag =  0;
649 			/*PKI IH3*/
650 			/* pki_h3 + irh + ossp[0] + ossp[1] = 32 bytes */
651 			ih3->fsz    = LIO_PCICMD_O3;
652 		}
653 
654 	} else {
655 		ih2          = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
656 		ih2->tagtype = ATOMIC_TAG;
657 		ih2->tag     = LIO_CONTROL;
658 		ih2->raw     = 1;
659 		ih2->grp     = CFG_GET_CTRL_Q_GRP(oct_cfg);
660 
661 		if (sc->datasize) {
662 			ih2->dlengsz = sc->datasize;
663 			ih2->rs = 1;
664 		}
665 
666 		irh            = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
667 		irh->opcode    = opcode;
668 		irh->subcode   = subcode;
669 
670 		/* opcode/subcode specific parameters (ossp) */
671 		irh->ossp       = irh_ossp;
672 		sc->cmd.cmd2.ossp[0] = ossp0;
673 		sc->cmd.cmd2.ossp[1] = ossp1;
674 
675 		if (sc->rdatasize) {
676 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd2.rdp;
677 			rdp->pcie_port = oct->pcie_port;
678 			rdp->rlen      = sc->rdatasize;
679 
680 			irh->rflag =  1;
681 			/* irh+ossp[0]+ossp[1]+rdp+rptr = 40 bytes */
682 			ih2->fsz   = LIO_SOFTCMDRESP_IH2;
683 		} else {
684 			irh->rflag =  0;
685 			/* irh + ossp[0] + ossp[1] = 24 bytes */
686 			ih2->fsz   = LIO_PCICMD_O2;
687 		}
688 	}
689 }
690 
691 int octeon_send_soft_command(struct octeon_device *oct,
692 			     struct octeon_soft_command *sc)
693 {
694 	struct octeon_instr_queue *iq;
695 	struct octeon_instr_ih2 *ih2;
696 	struct octeon_instr_ih3 *ih3;
697 	struct octeon_instr_irh *irh;
698 	u32 len;
699 
700 	iq = oct->instr_queue[sc->iq_no];
701 	if (!iq->allow_soft_cmds) {
702 		dev_err(&oct->pci_dev->dev, "Soft commands are not allowed on Queue %d\n",
703 			sc->iq_no);
704 		INCR_INSTRQUEUE_PKT_COUNT(oct, sc->iq_no, instr_dropped, 1);
705 		return IQ_SEND_FAILED;
706 	}
707 
708 	if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
709 		ih3 =  (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
710 		if (ih3->dlengsz) {
711 			WARN_ON(!sc->dmadptr);
712 			sc->cmd.cmd3.dptr = sc->dmadptr;
713 		}
714 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
715 		if (irh->rflag) {
716 			WARN_ON(!sc->dmarptr);
717 			WARN_ON(!sc->status_word);
718 			*sc->status_word = COMPLETION_WORD_INIT;
719 			sc->cmd.cmd3.rptr = sc->dmarptr;
720 		}
721 		len = (u32)ih3->dlengsz;
722 	} else {
723 		ih2 = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
724 		if (ih2->dlengsz) {
725 			WARN_ON(!sc->dmadptr);
726 			sc->cmd.cmd2.dptr = sc->dmadptr;
727 		}
728 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
729 		if (irh->rflag) {
730 			WARN_ON(!sc->dmarptr);
731 			WARN_ON(!sc->status_word);
732 			*sc->status_word = COMPLETION_WORD_INIT;
733 			sc->cmd.cmd2.rptr = sc->dmarptr;
734 		}
735 		len = (u32)ih2->dlengsz;
736 	}
737 
738 	sc->expiry_time = jiffies + msecs_to_jiffies(LIO_SC_MAX_TMO_MS);
739 
740 	return (octeon_send_command(oct, sc->iq_no, 1, &sc->cmd, sc,
741 				    len, REQTYPE_SOFT_COMMAND));
742 }
743 
744 int octeon_setup_sc_buffer_pool(struct octeon_device *oct)
745 {
746 	int i;
747 	u64 dma_addr;
748 	struct octeon_soft_command *sc;
749 
750 	INIT_LIST_HEAD(&oct->sc_buf_pool.head);
751 	spin_lock_init(&oct->sc_buf_pool.lock);
752 	atomic_set(&oct->sc_buf_pool.alloc_buf_count, 0);
753 
754 	for (i = 0; i < MAX_SOFT_COMMAND_BUFFERS; i++) {
755 		sc = (struct octeon_soft_command *)
756 			lio_dma_alloc(oct,
757 				      SOFT_COMMAND_BUFFER_SIZE,
758 					  (dma_addr_t *)&dma_addr);
759 		if (!sc) {
760 			octeon_free_sc_buffer_pool(oct);
761 			return 1;
762 		}
763 
764 		sc->dma_addr = dma_addr;
765 		sc->size = SOFT_COMMAND_BUFFER_SIZE;
766 
767 		list_add_tail(&sc->node, &oct->sc_buf_pool.head);
768 	}
769 
770 	return 0;
771 }
772 
773 int octeon_free_sc_done_list(struct octeon_device *oct)
774 {
775 	struct octeon_response_list *done_sc_list, *zombie_sc_list;
776 	struct octeon_soft_command *sc;
777 	struct list_head *tmp, *tmp2;
778 	spinlock_t *sc_lists_lock; /* lock for response_list */
779 
780 	done_sc_list = &oct->response_list[OCTEON_DONE_SC_LIST];
781 	zombie_sc_list = &oct->response_list[OCTEON_ZOMBIE_SC_LIST];
782 
783 	if (!atomic_read(&done_sc_list->pending_req_count))
784 		return 0;
785 
786 	sc_lists_lock = &oct->response_list[OCTEON_ORDERED_SC_LIST].lock;
787 
788 	spin_lock_bh(sc_lists_lock);
789 
790 	list_for_each_safe(tmp, tmp2, &done_sc_list->head) {
791 		sc = list_entry(tmp, struct octeon_soft_command, node);
792 
793 		if (READ_ONCE(sc->caller_is_done)) {
794 			list_del(&sc->node);
795 			atomic_dec(&done_sc_list->pending_req_count);
796 
797 			if (*sc->status_word == COMPLETION_WORD_INIT) {
798 				/* timeout; move sc to zombie list */
799 				list_add_tail(&sc->node, &zombie_sc_list->head);
800 				atomic_inc(&zombie_sc_list->pending_req_count);
801 			} else {
802 				octeon_free_soft_command(oct, sc);
803 			}
804 		}
805 	}
806 
807 	spin_unlock_bh(sc_lists_lock);
808 
809 	return 0;
810 }
811 
812 int octeon_free_sc_zombie_list(struct octeon_device *oct)
813 {
814 	struct octeon_response_list *zombie_sc_list;
815 	struct octeon_soft_command *sc;
816 	struct list_head *tmp, *tmp2;
817 	spinlock_t *sc_lists_lock; /* lock for response_list */
818 
819 	zombie_sc_list = &oct->response_list[OCTEON_ZOMBIE_SC_LIST];
820 	sc_lists_lock = &oct->response_list[OCTEON_ORDERED_SC_LIST].lock;
821 
822 	spin_lock_bh(sc_lists_lock);
823 
824 	list_for_each_safe(tmp, tmp2, &zombie_sc_list->head) {
825 		list_del(tmp);
826 		atomic_dec(&zombie_sc_list->pending_req_count);
827 		sc = list_entry(tmp, struct octeon_soft_command, node);
828 		octeon_free_soft_command(oct, sc);
829 	}
830 
831 	spin_unlock_bh(sc_lists_lock);
832 
833 	return 0;
834 }
835 
836 int octeon_free_sc_buffer_pool(struct octeon_device *oct)
837 {
838 	struct list_head *tmp, *tmp2;
839 	struct octeon_soft_command *sc;
840 
841 	octeon_free_sc_zombie_list(oct);
842 
843 	spin_lock_bh(&oct->sc_buf_pool.lock);
844 
845 	list_for_each_safe(tmp, tmp2, &oct->sc_buf_pool.head) {
846 		list_del(tmp);
847 
848 		sc = (struct octeon_soft_command *)tmp;
849 
850 		lio_dma_free(oct, sc->size, sc, sc->dma_addr);
851 	}
852 
853 	INIT_LIST_HEAD(&oct->sc_buf_pool.head);
854 
855 	spin_unlock_bh(&oct->sc_buf_pool.lock);
856 
857 	return 0;
858 }
859 
860 struct octeon_soft_command *octeon_alloc_soft_command(struct octeon_device *oct,
861 						      u32 datasize,
862 						      u32 rdatasize,
863 						      u32 ctxsize)
864 {
865 	u64 dma_addr;
866 	u32 size;
867 	u32 offset = sizeof(struct octeon_soft_command);
868 	struct octeon_soft_command *sc = NULL;
869 	struct list_head *tmp;
870 
871 	if (!rdatasize)
872 		rdatasize = 16;
873 
874 	WARN_ON((offset + datasize + rdatasize + ctxsize) >
875 	       SOFT_COMMAND_BUFFER_SIZE);
876 
877 	spin_lock_bh(&oct->sc_buf_pool.lock);
878 
879 	if (list_empty(&oct->sc_buf_pool.head)) {
880 		spin_unlock_bh(&oct->sc_buf_pool.lock);
881 		return NULL;
882 	}
883 
884 	list_for_each(tmp, &oct->sc_buf_pool.head)
885 		break;
886 
887 	list_del(tmp);
888 
889 	atomic_inc(&oct->sc_buf_pool.alloc_buf_count);
890 
891 	spin_unlock_bh(&oct->sc_buf_pool.lock);
892 
893 	sc = (struct octeon_soft_command *)tmp;
894 
895 	dma_addr = sc->dma_addr;
896 	size = sc->size;
897 
898 	memset(sc, 0, sc->size);
899 
900 	sc->dma_addr = dma_addr;
901 	sc->size = size;
902 
903 	if (ctxsize) {
904 		sc->ctxptr = (u8 *)sc + offset;
905 		sc->ctxsize = ctxsize;
906 	}
907 
908 	/* Start data at 128 byte boundary */
909 	offset = (offset + ctxsize + 127) & 0xffffff80;
910 
911 	if (datasize) {
912 		sc->virtdptr = (u8 *)sc + offset;
913 		sc->dmadptr = dma_addr + offset;
914 		sc->datasize = datasize;
915 	}
916 
917 	/* Start rdata at 128 byte boundary */
918 	offset = (offset + datasize + 127) & 0xffffff80;
919 
920 	if (rdatasize) {
921 		WARN_ON(rdatasize < 16);
922 		sc->virtrptr = (u8 *)sc + offset;
923 		sc->dmarptr = dma_addr + offset;
924 		sc->rdatasize = rdatasize;
925 		sc->status_word = (u64 *)((u8 *)(sc->virtrptr) + rdatasize - 8);
926 	}
927 
928 	return sc;
929 }
930 
931 void octeon_free_soft_command(struct octeon_device *oct,
932 			      struct octeon_soft_command *sc)
933 {
934 	spin_lock_bh(&oct->sc_buf_pool.lock);
935 
936 	list_add_tail(&sc->node, &oct->sc_buf_pool.head);
937 
938 	atomic_dec(&oct->sc_buf_pool.alloc_buf_count);
939 
940 	spin_unlock_bh(&oct->sc_buf_pool.lock);
941 }
942