xref: /linux/drivers/net/ethernet/cavium/liquidio/request_manager.c (revision ebf68996de0ab250c5d520eb2291ab65643e9a1e)
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 		iq->fill_cnt = 0;
282 		iq->last_db_time = jiffies;
283 		return;
284 	}
285 }
286 
287 void
288 octeon_ring_doorbell_locked(struct octeon_device *oct, u32 iq_no)
289 {
290 	struct octeon_instr_queue *iq;
291 
292 	iq = oct->instr_queue[iq_no];
293 	spin_lock(&iq->post_lock);
294 	if (iq->fill_cnt)
295 		ring_doorbell(oct, iq);
296 	spin_unlock(&iq->post_lock);
297 }
298 
299 static inline void __copy_cmd_into_iq(struct octeon_instr_queue *iq,
300 				      u8 *cmd)
301 {
302 	u8 *iqptr, cmdsize;
303 
304 	cmdsize = ((iq->iqcmd_64B) ? 64 : 32);
305 	iqptr = iq->base_addr + (cmdsize * iq->host_write_index);
306 
307 	memcpy(iqptr, cmd, cmdsize);
308 }
309 
310 static inline struct iq_post_status
311 __post_command2(struct octeon_instr_queue *iq, u8 *cmd)
312 {
313 	struct iq_post_status st;
314 
315 	st.status = IQ_SEND_OK;
316 
317 	/* This ensures that the read index does not wrap around to the same
318 	 * position if queue gets full before Octeon could fetch any instr.
319 	 */
320 	if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 1)) {
321 		st.status = IQ_SEND_FAILED;
322 		st.index = -1;
323 		return st;
324 	}
325 
326 	if (atomic_read(&iq->instr_pending) >= (s32)(iq->max_count - 2))
327 		st.status = IQ_SEND_STOP;
328 
329 	__copy_cmd_into_iq(iq, cmd);
330 
331 	/* "index" is returned, host_write_index is modified. */
332 	st.index = iq->host_write_index;
333 	iq->host_write_index = incr_index(iq->host_write_index, 1,
334 					  iq->max_count);
335 	iq->fill_cnt++;
336 
337 	/* Flush the command into memory. We need to be sure the data is in
338 	 * memory before indicating that the instruction is pending.
339 	 */
340 	wmb();
341 
342 	atomic_inc(&iq->instr_pending);
343 
344 	return st;
345 }
346 
347 int
348 octeon_register_reqtype_free_fn(struct octeon_device *oct, int reqtype,
349 				void (*fn)(void *))
350 {
351 	if (reqtype > REQTYPE_LAST) {
352 		dev_err(&oct->pci_dev->dev, "%s: Invalid reqtype: %d\n",
353 			__func__, reqtype);
354 		return -EINVAL;
355 	}
356 
357 	reqtype_free_fn[oct->octeon_id][reqtype] = fn;
358 
359 	return 0;
360 }
361 
362 static inline void
363 __add_to_request_list(struct octeon_instr_queue *iq,
364 		      int idx, void *buf, int reqtype)
365 {
366 	iq->request_list[idx].buf = buf;
367 	iq->request_list[idx].reqtype = reqtype;
368 }
369 
370 /* Can only run in process context */
371 int
372 lio_process_iq_request_list(struct octeon_device *oct,
373 			    struct octeon_instr_queue *iq, u32 napi_budget)
374 {
375 	struct cavium_wq *cwq = &oct->dma_comp_wq;
376 	int reqtype;
377 	void *buf;
378 	u32 old = iq->flush_index;
379 	u32 inst_count = 0;
380 	unsigned int pkts_compl = 0, bytes_compl = 0;
381 	struct octeon_soft_command *sc;
382 	unsigned long flags;
383 
384 	while (old != iq->octeon_read_index) {
385 		reqtype = iq->request_list[old].reqtype;
386 		buf     = iq->request_list[old].buf;
387 
388 		if (reqtype == REQTYPE_NONE)
389 			goto skip_this;
390 
391 		octeon_update_tx_completion_counters(buf, reqtype, &pkts_compl,
392 						     &bytes_compl);
393 
394 		switch (reqtype) {
395 		case REQTYPE_NORESP_NET:
396 		case REQTYPE_NORESP_NET_SG:
397 		case REQTYPE_RESP_NET_SG:
398 			reqtype_free_fn[oct->octeon_id][reqtype](buf);
399 			break;
400 		case REQTYPE_RESP_NET:
401 		case REQTYPE_SOFT_COMMAND:
402 			sc = buf;
403 			/* We're expecting a response from Octeon.
404 			 * It's up to lio_process_ordered_list() to
405 			 * process  sc. Add sc to the ordered soft
406 			 * command response list because we expect
407 			 * a response from Octeon.
408 			 */
409 			spin_lock_irqsave(&oct->response_list
410 					  [OCTEON_ORDERED_SC_LIST].lock, flags);
411 			atomic_inc(&oct->response_list
412 				   [OCTEON_ORDERED_SC_LIST].pending_req_count);
413 			list_add_tail(&sc->node, &oct->response_list
414 				[OCTEON_ORDERED_SC_LIST].head);
415 			spin_unlock_irqrestore(&oct->response_list
416 					       [OCTEON_ORDERED_SC_LIST].lock,
417 					       flags);
418 			break;
419 		default:
420 			dev_err(&oct->pci_dev->dev,
421 				"%s Unknown reqtype: %d buf: %p at idx %d\n",
422 				__func__, reqtype, buf, old);
423 		}
424 
425 		iq->request_list[old].buf = NULL;
426 		iq->request_list[old].reqtype = 0;
427 
428  skip_this:
429 		inst_count++;
430 		old = incr_index(old, 1, iq->max_count);
431 
432 		if ((napi_budget) && (inst_count >= napi_budget))
433 			break;
434 	}
435 	if (bytes_compl)
436 		octeon_report_tx_completion_to_bql(iq->app_ctx, pkts_compl,
437 						   bytes_compl);
438 	iq->flush_index = old;
439 
440 	if (atomic_read(&oct->response_list
441 			[OCTEON_ORDERED_SC_LIST].pending_req_count))
442 		queue_work(cwq->wq, &cwq->wk.work.work);
443 
444 	return inst_count;
445 }
446 
447 /* Can only be called from process context */
448 int
449 octeon_flush_iq(struct octeon_device *oct, struct octeon_instr_queue *iq,
450 		u32 napi_budget)
451 {
452 	u32 inst_processed = 0;
453 	u32 tot_inst_processed = 0;
454 	int tx_done = 1;
455 
456 	if (!spin_trylock(&iq->iq_flush_running_lock))
457 		return tx_done;
458 
459 	spin_lock_bh(&iq->lock);
460 
461 	iq->octeon_read_index = oct->fn_list.update_iq_read_idx(iq);
462 
463 	do {
464 		/* Process any outstanding IQ packets. */
465 		if (iq->flush_index == iq->octeon_read_index)
466 			break;
467 
468 		if (napi_budget)
469 			inst_processed =
470 				lio_process_iq_request_list(oct, iq,
471 							    napi_budget -
472 							    tot_inst_processed);
473 		else
474 			inst_processed =
475 				lio_process_iq_request_list(oct, iq, 0);
476 
477 		if (inst_processed) {
478 			iq->pkts_processed += inst_processed;
479 			atomic_sub(inst_processed, &iq->instr_pending);
480 			iq->stats.instr_processed += inst_processed;
481 		}
482 
483 		tot_inst_processed += inst_processed;
484 	} while (tot_inst_processed < napi_budget);
485 
486 	if (napi_budget && (tot_inst_processed >= napi_budget))
487 		tx_done = 0;
488 
489 	iq->last_db_time = jiffies;
490 
491 	spin_unlock_bh(&iq->lock);
492 
493 	spin_unlock(&iq->iq_flush_running_lock);
494 
495 	return tx_done;
496 }
497 
498 /* Process instruction queue after timeout.
499  * This routine gets called from a workqueue or when removing the module.
500  */
501 static void __check_db_timeout(struct octeon_device *oct, u64 iq_no)
502 {
503 	struct octeon_instr_queue *iq;
504 	u64 next_time;
505 
506 	if (!oct)
507 		return;
508 
509 	iq = oct->instr_queue[iq_no];
510 	if (!iq)
511 		return;
512 
513 	/* return immediately, if no work pending */
514 	if (!atomic_read(&iq->instr_pending))
515 		return;
516 	/* If jiffies - last_db_time < db_timeout do nothing  */
517 	next_time = iq->last_db_time + iq->db_timeout;
518 	if (!time_after(jiffies, (unsigned long)next_time))
519 		return;
520 	iq->last_db_time = jiffies;
521 
522 	/* Flush the instruction queue */
523 	octeon_flush_iq(oct, iq, 0);
524 
525 	lio_enable_irq(NULL, iq);
526 }
527 
528 /* Called by the Poll thread at regular intervals to check the instruction
529  * queue for commands to be posted and for commands that were fetched by Octeon.
530  */
531 static void check_db_timeout(struct work_struct *work)
532 {
533 	struct cavium_wk *wk = (struct cavium_wk *)work;
534 	struct octeon_device *oct = (struct octeon_device *)wk->ctxptr;
535 	u64 iq_no = wk->ctxul;
536 	struct cavium_wq *db_wq = &oct->check_db_wq[iq_no];
537 	u32 delay = 10;
538 
539 	__check_db_timeout(oct, iq_no);
540 	queue_delayed_work(db_wq->wq, &db_wq->wk.work, msecs_to_jiffies(delay));
541 }
542 
543 int
544 octeon_send_command(struct octeon_device *oct, u32 iq_no,
545 		    u32 force_db, void *cmd, void *buf,
546 		    u32 datasize, u32 reqtype)
547 {
548 	int xmit_stopped;
549 	struct iq_post_status st;
550 	struct octeon_instr_queue *iq = oct->instr_queue[iq_no];
551 
552 	/* Get the lock and prevent other tasks and tx interrupt handler from
553 	 * running.
554 	 */
555 	if (iq->allow_soft_cmds)
556 		spin_lock_bh(&iq->post_lock);
557 
558 	st = __post_command2(iq, cmd);
559 
560 	if (st.status != IQ_SEND_FAILED) {
561 		xmit_stopped = octeon_report_sent_bytes_to_bql(buf, reqtype);
562 		__add_to_request_list(iq, st.index, buf, reqtype);
563 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, bytes_sent, datasize);
564 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_posted, 1);
565 
566 		if (iq->fill_cnt >= MAX_OCTEON_FILL_COUNT || force_db ||
567 		    xmit_stopped || st.status == IQ_SEND_STOP)
568 			ring_doorbell(oct, iq);
569 	} else {
570 		INCR_INSTRQUEUE_PKT_COUNT(oct, iq_no, instr_dropped, 1);
571 	}
572 
573 	if (iq->allow_soft_cmds)
574 		spin_unlock_bh(&iq->post_lock);
575 
576 	/* This is only done here to expedite packets being flushed
577 	 * for cases where there are no IQ completion interrupts.
578 	 */
579 
580 	return st.status;
581 }
582 
583 void
584 octeon_prepare_soft_command(struct octeon_device *oct,
585 			    struct octeon_soft_command *sc,
586 			    u8 opcode,
587 			    u8 subcode,
588 			    u32 irh_ossp,
589 			    u64 ossp0,
590 			    u64 ossp1)
591 {
592 	struct octeon_config *oct_cfg;
593 	struct octeon_instr_ih2 *ih2;
594 	struct octeon_instr_ih3 *ih3;
595 	struct octeon_instr_pki_ih3 *pki_ih3;
596 	struct octeon_instr_irh *irh;
597 	struct octeon_instr_rdp *rdp;
598 
599 	WARN_ON(opcode > 15);
600 	WARN_ON(subcode > 127);
601 
602 	oct_cfg = octeon_get_conf(oct);
603 
604 	if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
605 		ih3 = (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
606 
607 		ih3->pkind = oct->instr_queue[sc->iq_no]->txpciq.s.pkind;
608 
609 		pki_ih3 = (struct octeon_instr_pki_ih3 *)&sc->cmd.cmd3.pki_ih3;
610 
611 		pki_ih3->w           = 1;
612 		pki_ih3->raw         = 1;
613 		pki_ih3->utag        = 1;
614 		pki_ih3->uqpg        =
615 			oct->instr_queue[sc->iq_no]->txpciq.s.use_qpg;
616 		pki_ih3->utt         = 1;
617 		pki_ih3->tag     = LIO_CONTROL;
618 		pki_ih3->tagtype = ATOMIC_TAG;
619 		pki_ih3->qpg         =
620 			oct->instr_queue[sc->iq_no]->txpciq.s.ctrl_qpg;
621 
622 		pki_ih3->pm          = 0x7;
623 		pki_ih3->sl          = 8;
624 
625 		if (sc->datasize)
626 			ih3->dlengsz = sc->datasize;
627 
628 		irh            = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
629 		irh->opcode    = opcode;
630 		irh->subcode   = subcode;
631 
632 		/* opcode/subcode specific parameters (ossp) */
633 		irh->ossp       = irh_ossp;
634 		sc->cmd.cmd3.ossp[0] = ossp0;
635 		sc->cmd.cmd3.ossp[1] = ossp1;
636 
637 		if (sc->rdatasize) {
638 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd3.rdp;
639 			rdp->pcie_port = oct->pcie_port;
640 			rdp->rlen      = sc->rdatasize;
641 
642 			irh->rflag =  1;
643 			/*PKI IH3*/
644 			/* pki_ih3 irh+ossp[0]+ossp[1]+rdp+rptr = 48 bytes */
645 			ih3->fsz    = LIO_SOFTCMDRESP_IH3;
646 		} else {
647 			irh->rflag =  0;
648 			/*PKI IH3*/
649 			/* pki_h3 + irh + ossp[0] + ossp[1] = 32 bytes */
650 			ih3->fsz    = LIO_PCICMD_O3;
651 		}
652 
653 	} else {
654 		ih2          = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
655 		ih2->tagtype = ATOMIC_TAG;
656 		ih2->tag     = LIO_CONTROL;
657 		ih2->raw     = 1;
658 		ih2->grp     = CFG_GET_CTRL_Q_GRP(oct_cfg);
659 
660 		if (sc->datasize) {
661 			ih2->dlengsz = sc->datasize;
662 			ih2->rs = 1;
663 		}
664 
665 		irh            = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
666 		irh->opcode    = opcode;
667 		irh->subcode   = subcode;
668 
669 		/* opcode/subcode specific parameters (ossp) */
670 		irh->ossp       = irh_ossp;
671 		sc->cmd.cmd2.ossp[0] = ossp0;
672 		sc->cmd.cmd2.ossp[1] = ossp1;
673 
674 		if (sc->rdatasize) {
675 			rdp = (struct octeon_instr_rdp *)&sc->cmd.cmd2.rdp;
676 			rdp->pcie_port = oct->pcie_port;
677 			rdp->rlen      = sc->rdatasize;
678 
679 			irh->rflag =  1;
680 			/* irh+ossp[0]+ossp[1]+rdp+rptr = 40 bytes */
681 			ih2->fsz   = LIO_SOFTCMDRESP_IH2;
682 		} else {
683 			irh->rflag =  0;
684 			/* irh + ossp[0] + ossp[1] = 24 bytes */
685 			ih2->fsz   = LIO_PCICMD_O2;
686 		}
687 	}
688 }
689 
690 int octeon_send_soft_command(struct octeon_device *oct,
691 			     struct octeon_soft_command *sc)
692 {
693 	struct octeon_instr_queue *iq;
694 	struct octeon_instr_ih2 *ih2;
695 	struct octeon_instr_ih3 *ih3;
696 	struct octeon_instr_irh *irh;
697 	u32 len;
698 
699 	iq = oct->instr_queue[sc->iq_no];
700 	if (!iq->allow_soft_cmds) {
701 		dev_err(&oct->pci_dev->dev, "Soft commands are not allowed on Queue %d\n",
702 			sc->iq_no);
703 		INCR_INSTRQUEUE_PKT_COUNT(oct, sc->iq_no, instr_dropped, 1);
704 		return IQ_SEND_FAILED;
705 	}
706 
707 	if (OCTEON_CN23XX_PF(oct) || OCTEON_CN23XX_VF(oct)) {
708 		ih3 =  (struct octeon_instr_ih3 *)&sc->cmd.cmd3.ih3;
709 		if (ih3->dlengsz) {
710 			WARN_ON(!sc->dmadptr);
711 			sc->cmd.cmd3.dptr = sc->dmadptr;
712 		}
713 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd3.irh;
714 		if (irh->rflag) {
715 			WARN_ON(!sc->dmarptr);
716 			WARN_ON(!sc->status_word);
717 			*sc->status_word = COMPLETION_WORD_INIT;
718 			sc->cmd.cmd3.rptr = sc->dmarptr;
719 		}
720 		len = (u32)ih3->dlengsz;
721 	} else {
722 		ih2 = (struct octeon_instr_ih2 *)&sc->cmd.cmd2.ih2;
723 		if (ih2->dlengsz) {
724 			WARN_ON(!sc->dmadptr);
725 			sc->cmd.cmd2.dptr = sc->dmadptr;
726 		}
727 		irh = (struct octeon_instr_irh *)&sc->cmd.cmd2.irh;
728 		if (irh->rflag) {
729 			WARN_ON(!sc->dmarptr);
730 			WARN_ON(!sc->status_word);
731 			*sc->status_word = COMPLETION_WORD_INIT;
732 			sc->cmd.cmd2.rptr = sc->dmarptr;
733 		}
734 		len = (u32)ih2->dlengsz;
735 	}
736 
737 	sc->expiry_time = jiffies + msecs_to_jiffies(LIO_SC_MAX_TMO_MS);
738 
739 	return (octeon_send_command(oct, sc->iq_no, 1, &sc->cmd, sc,
740 				    len, REQTYPE_SOFT_COMMAND));
741 }
742 
743 int octeon_setup_sc_buffer_pool(struct octeon_device *oct)
744 {
745 	int i;
746 	u64 dma_addr;
747 	struct octeon_soft_command *sc;
748 
749 	INIT_LIST_HEAD(&oct->sc_buf_pool.head);
750 	spin_lock_init(&oct->sc_buf_pool.lock);
751 	atomic_set(&oct->sc_buf_pool.alloc_buf_count, 0);
752 
753 	for (i = 0; i < MAX_SOFT_COMMAND_BUFFERS; i++) {
754 		sc = (struct octeon_soft_command *)
755 			lio_dma_alloc(oct,
756 				      SOFT_COMMAND_BUFFER_SIZE,
757 					  (dma_addr_t *)&dma_addr);
758 		if (!sc) {
759 			octeon_free_sc_buffer_pool(oct);
760 			return 1;
761 		}
762 
763 		sc->dma_addr = dma_addr;
764 		sc->size = SOFT_COMMAND_BUFFER_SIZE;
765 
766 		list_add_tail(&sc->node, &oct->sc_buf_pool.head);
767 	}
768 
769 	return 0;
770 }
771 
772 int octeon_free_sc_done_list(struct octeon_device *oct)
773 {
774 	struct octeon_response_list *done_sc_list, *zombie_sc_list;
775 	struct octeon_soft_command *sc;
776 	struct list_head *tmp, *tmp2;
777 	spinlock_t *sc_lists_lock; /* lock for response_list */
778 
779 	done_sc_list = &oct->response_list[OCTEON_DONE_SC_LIST];
780 	zombie_sc_list = &oct->response_list[OCTEON_ZOMBIE_SC_LIST];
781 
782 	if (!atomic_read(&done_sc_list->pending_req_count))
783 		return 0;
784 
785 	sc_lists_lock = &oct->response_list[OCTEON_ORDERED_SC_LIST].lock;
786 
787 	spin_lock_bh(sc_lists_lock);
788 
789 	list_for_each_safe(tmp, tmp2, &done_sc_list->head) {
790 		sc = list_entry(tmp, struct octeon_soft_command, node);
791 
792 		if (READ_ONCE(sc->caller_is_done)) {
793 			list_del(&sc->node);
794 			atomic_dec(&done_sc_list->pending_req_count);
795 
796 			if (*sc->status_word == COMPLETION_WORD_INIT) {
797 				/* timeout; move sc to zombie list */
798 				list_add_tail(&sc->node, &zombie_sc_list->head);
799 				atomic_inc(&zombie_sc_list->pending_req_count);
800 			} else {
801 				octeon_free_soft_command(oct, sc);
802 			}
803 		}
804 	}
805 
806 	spin_unlock_bh(sc_lists_lock);
807 
808 	return 0;
809 }
810 
811 int octeon_free_sc_zombie_list(struct octeon_device *oct)
812 {
813 	struct octeon_response_list *zombie_sc_list;
814 	struct octeon_soft_command *sc;
815 	struct list_head *tmp, *tmp2;
816 	spinlock_t *sc_lists_lock; /* lock for response_list */
817 
818 	zombie_sc_list = &oct->response_list[OCTEON_ZOMBIE_SC_LIST];
819 	sc_lists_lock = &oct->response_list[OCTEON_ORDERED_SC_LIST].lock;
820 
821 	spin_lock_bh(sc_lists_lock);
822 
823 	list_for_each_safe(tmp, tmp2, &zombie_sc_list->head) {
824 		list_del(tmp);
825 		atomic_dec(&zombie_sc_list->pending_req_count);
826 		sc = list_entry(tmp, struct octeon_soft_command, node);
827 		octeon_free_soft_command(oct, sc);
828 	}
829 
830 	spin_unlock_bh(sc_lists_lock);
831 
832 	return 0;
833 }
834 
835 int octeon_free_sc_buffer_pool(struct octeon_device *oct)
836 {
837 	struct list_head *tmp, *tmp2;
838 	struct octeon_soft_command *sc;
839 
840 	octeon_free_sc_zombie_list(oct);
841 
842 	spin_lock_bh(&oct->sc_buf_pool.lock);
843 
844 	list_for_each_safe(tmp, tmp2, &oct->sc_buf_pool.head) {
845 		list_del(tmp);
846 
847 		sc = (struct octeon_soft_command *)tmp;
848 
849 		lio_dma_free(oct, sc->size, sc, sc->dma_addr);
850 	}
851 
852 	INIT_LIST_HEAD(&oct->sc_buf_pool.head);
853 
854 	spin_unlock_bh(&oct->sc_buf_pool.lock);
855 
856 	return 0;
857 }
858 
859 struct octeon_soft_command *octeon_alloc_soft_command(struct octeon_device *oct,
860 						      u32 datasize,
861 						      u32 rdatasize,
862 						      u32 ctxsize)
863 {
864 	u64 dma_addr;
865 	u32 size;
866 	u32 offset = sizeof(struct octeon_soft_command);
867 	struct octeon_soft_command *sc = NULL;
868 	struct list_head *tmp;
869 
870 	if (!rdatasize)
871 		rdatasize = 16;
872 
873 	WARN_ON((offset + datasize + rdatasize + ctxsize) >
874 	       SOFT_COMMAND_BUFFER_SIZE);
875 
876 	spin_lock_bh(&oct->sc_buf_pool.lock);
877 
878 	if (list_empty(&oct->sc_buf_pool.head)) {
879 		spin_unlock_bh(&oct->sc_buf_pool.lock);
880 		return NULL;
881 	}
882 
883 	list_for_each(tmp, &oct->sc_buf_pool.head)
884 		break;
885 
886 	list_del(tmp);
887 
888 	atomic_inc(&oct->sc_buf_pool.alloc_buf_count);
889 
890 	spin_unlock_bh(&oct->sc_buf_pool.lock);
891 
892 	sc = (struct octeon_soft_command *)tmp;
893 
894 	dma_addr = sc->dma_addr;
895 	size = sc->size;
896 
897 	memset(sc, 0, sc->size);
898 
899 	sc->dma_addr = dma_addr;
900 	sc->size = size;
901 
902 	if (ctxsize) {
903 		sc->ctxptr = (u8 *)sc + offset;
904 		sc->ctxsize = ctxsize;
905 	}
906 
907 	/* Start data at 128 byte boundary */
908 	offset = (offset + ctxsize + 127) & 0xffffff80;
909 
910 	if (datasize) {
911 		sc->virtdptr = (u8 *)sc + offset;
912 		sc->dmadptr = dma_addr + offset;
913 		sc->datasize = datasize;
914 	}
915 
916 	/* Start rdata at 128 byte boundary */
917 	offset = (offset + datasize + 127) & 0xffffff80;
918 
919 	if (rdatasize) {
920 		WARN_ON(rdatasize < 16);
921 		sc->virtrptr = (u8 *)sc + offset;
922 		sc->dmarptr = dma_addr + offset;
923 		sc->rdatasize = rdatasize;
924 		sc->status_word = (u64 *)((u8 *)(sc->virtrptr) + rdatasize - 8);
925 	}
926 
927 	return sc;
928 }
929 
930 void octeon_free_soft_command(struct octeon_device *oct,
931 			      struct octeon_soft_command *sc)
932 {
933 	spin_lock_bh(&oct->sc_buf_pool.lock);
934 
935 	list_add_tail(&sc->node, &oct->sc_buf_pool.head);
936 
937 	atomic_dec(&oct->sc_buf_pool.alloc_buf_count);
938 
939 	spin_unlock_bh(&oct->sc_buf_pool.lock);
940 }
941