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