xref: /openbmc/linux/drivers/net/wireless/ath/ath10k/ce.c (revision 48c926cd)
1 /*
2  * Copyright (c) 2005-2011 Atheros Communications Inc.
3  * Copyright (c) 2011-2013 Qualcomm Atheros, Inc.
4  *
5  * Permission to use, copy, modify, and/or distribute this software for any
6  * purpose with or without fee is hereby granted, provided that the above
7  * copyright notice and this permission notice appear in all copies.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
16  */
17 
18 #include "hif.h"
19 #include "ce.h"
20 #include "debug.h"
21 
22 /*
23  * Support for Copy Engine hardware, which is mainly used for
24  * communication between Host and Target over a PCIe interconnect.
25  */
26 
27 /*
28  * A single CopyEngine (CE) comprises two "rings":
29  *   a source ring
30  *   a destination ring
31  *
32  * Each ring consists of a number of descriptors which specify
33  * an address, length, and meta-data.
34  *
35  * Typically, one side of the PCIe/AHB/SNOC interconnect (Host or Target)
36  * controls one ring and the other side controls the other ring.
37  * The source side chooses when to initiate a transfer and it
38  * chooses what to send (buffer address, length). The destination
39  * side keeps a supply of "anonymous receive buffers" available and
40  * it handles incoming data as it arrives (when the destination
41  * receives an interrupt).
42  *
43  * The sender may send a simple buffer (address/length) or it may
44  * send a small list of buffers.  When a small list is sent, hardware
45  * "gathers" these and they end up in a single destination buffer
46  * with a single interrupt.
47  *
48  * There are several "contexts" managed by this layer -- more, it
49  * may seem -- than should be needed. These are provided mainly for
50  * maximum flexibility and especially to facilitate a simpler HIF
51  * implementation. There are per-CopyEngine recv, send, and watermark
52  * contexts. These are supplied by the caller when a recv, send,
53  * or watermark handler is established and they are echoed back to
54  * the caller when the respective callbacks are invoked. There is
55  * also a per-transfer context supplied by the caller when a buffer
56  * (or sendlist) is sent and when a buffer is enqueued for recv.
57  * These per-transfer contexts are echoed back to the caller when
58  * the buffer is sent/received.
59  */
60 
61 static inline unsigned int
62 ath10k_set_ring_byte(unsigned int offset,
63 		     struct ath10k_hw_ce_regs_addr_map *addr_map)
64 {
65 	return ((offset << addr_map->lsb) & addr_map->mask);
66 }
67 
68 static inline unsigned int
69 ath10k_get_ring_byte(unsigned int offset,
70 		     struct ath10k_hw_ce_regs_addr_map *addr_map)
71 {
72 	return ((offset & addr_map->mask) >> (addr_map->lsb));
73 }
74 
75 static inline u32 ath10k_ce_read32(struct ath10k *ar, u32 offset)
76 {
77 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
78 
79 	return ce->bus_ops->read32(ar, offset);
80 }
81 
82 static inline void ath10k_ce_write32(struct ath10k *ar, u32 offset, u32 value)
83 {
84 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
85 
86 	ce->bus_ops->write32(ar, offset, value);
87 }
88 
89 static inline void ath10k_ce_dest_ring_write_index_set(struct ath10k *ar,
90 						       u32 ce_ctrl_addr,
91 						       unsigned int n)
92 {
93 	ath10k_ce_write32(ar, ce_ctrl_addr +
94 			  ar->hw_ce_regs->dst_wr_index_addr, n);
95 }
96 
97 static inline u32 ath10k_ce_dest_ring_write_index_get(struct ath10k *ar,
98 						      u32 ce_ctrl_addr)
99 {
100 	return ath10k_ce_read32(ar, ce_ctrl_addr +
101 				ar->hw_ce_regs->dst_wr_index_addr);
102 }
103 
104 static inline void ath10k_ce_src_ring_write_index_set(struct ath10k *ar,
105 						      u32 ce_ctrl_addr,
106 						      unsigned int n)
107 {
108 	ath10k_ce_write32(ar, ce_ctrl_addr +
109 			  ar->hw_ce_regs->sr_wr_index_addr, n);
110 }
111 
112 static inline u32 ath10k_ce_src_ring_write_index_get(struct ath10k *ar,
113 						     u32 ce_ctrl_addr)
114 {
115 	return ath10k_ce_read32(ar, ce_ctrl_addr +
116 				ar->hw_ce_regs->sr_wr_index_addr);
117 }
118 
119 static inline u32 ath10k_ce_src_ring_read_index_get(struct ath10k *ar,
120 						    u32 ce_ctrl_addr)
121 {
122 	return ath10k_ce_read32(ar, ce_ctrl_addr +
123 				ar->hw_ce_regs->current_srri_addr);
124 }
125 
126 static inline void ath10k_ce_src_ring_base_addr_set(struct ath10k *ar,
127 						    u32 ce_ctrl_addr,
128 						    unsigned int addr)
129 {
130 	ath10k_ce_write32(ar, ce_ctrl_addr +
131 			  ar->hw_ce_regs->sr_base_addr, addr);
132 }
133 
134 static inline void ath10k_ce_src_ring_size_set(struct ath10k *ar,
135 					       u32 ce_ctrl_addr,
136 					       unsigned int n)
137 {
138 	ath10k_ce_write32(ar, ce_ctrl_addr +
139 			  ar->hw_ce_regs->sr_size_addr, n);
140 }
141 
142 static inline void ath10k_ce_src_ring_dmax_set(struct ath10k *ar,
143 					       u32 ce_ctrl_addr,
144 					       unsigned int n)
145 {
146 	struct ath10k_hw_ce_ctrl1 *ctrl_regs = ar->hw_ce_regs->ctrl1_regs;
147 
148 	u32 ctrl1_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
149 					  ctrl_regs->addr);
150 
151 	ath10k_ce_write32(ar, ce_ctrl_addr + ctrl_regs->addr,
152 			  (ctrl1_addr &  ~(ctrl_regs->dmax->mask)) |
153 			  ath10k_set_ring_byte(n, ctrl_regs->dmax));
154 }
155 
156 static inline void ath10k_ce_src_ring_byte_swap_set(struct ath10k *ar,
157 						    u32 ce_ctrl_addr,
158 						    unsigned int n)
159 {
160 	struct ath10k_hw_ce_ctrl1 *ctrl_regs = ar->hw_ce_regs->ctrl1_regs;
161 
162 	u32 ctrl1_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
163 					  ctrl_regs->addr);
164 
165 	ath10k_ce_write32(ar, ce_ctrl_addr + ctrl_regs->addr,
166 			  (ctrl1_addr & ~(ctrl_regs->src_ring->mask)) |
167 			  ath10k_set_ring_byte(n, ctrl_regs->src_ring));
168 }
169 
170 static inline void ath10k_ce_dest_ring_byte_swap_set(struct ath10k *ar,
171 						     u32 ce_ctrl_addr,
172 						     unsigned int n)
173 {
174 	struct ath10k_hw_ce_ctrl1 *ctrl_regs = ar->hw_ce_regs->ctrl1_regs;
175 
176 	u32 ctrl1_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
177 					  ctrl_regs->addr);
178 
179 	ath10k_ce_write32(ar, ce_ctrl_addr + ctrl_regs->addr,
180 			  (ctrl1_addr & ~(ctrl_regs->dst_ring->mask)) |
181 			  ath10k_set_ring_byte(n, ctrl_regs->dst_ring));
182 }
183 
184 static inline u32 ath10k_ce_dest_ring_read_index_get(struct ath10k *ar,
185 						     u32 ce_ctrl_addr)
186 {
187 	return ath10k_ce_read32(ar, ce_ctrl_addr +
188 				ar->hw_ce_regs->current_drri_addr);
189 }
190 
191 static inline void ath10k_ce_dest_ring_base_addr_set(struct ath10k *ar,
192 						     u32 ce_ctrl_addr,
193 						     u32 addr)
194 {
195 	ath10k_ce_write32(ar, ce_ctrl_addr +
196 			  ar->hw_ce_regs->dr_base_addr, addr);
197 }
198 
199 static inline void ath10k_ce_dest_ring_size_set(struct ath10k *ar,
200 						u32 ce_ctrl_addr,
201 						unsigned int n)
202 {
203 	ath10k_ce_write32(ar, ce_ctrl_addr +
204 			  ar->hw_ce_regs->dr_size_addr, n);
205 }
206 
207 static inline void ath10k_ce_src_ring_highmark_set(struct ath10k *ar,
208 						   u32 ce_ctrl_addr,
209 						   unsigned int n)
210 {
211 	struct ath10k_hw_ce_dst_src_wm_regs *srcr_wm = ar->hw_ce_regs->wm_srcr;
212 	u32 addr = ath10k_ce_read32(ar, ce_ctrl_addr + srcr_wm->addr);
213 
214 	ath10k_ce_write32(ar, ce_ctrl_addr + srcr_wm->addr,
215 			  (addr & ~(srcr_wm->wm_high->mask)) |
216 			  (ath10k_set_ring_byte(n, srcr_wm->wm_high)));
217 }
218 
219 static inline void ath10k_ce_src_ring_lowmark_set(struct ath10k *ar,
220 						  u32 ce_ctrl_addr,
221 						  unsigned int n)
222 {
223 	struct ath10k_hw_ce_dst_src_wm_regs *srcr_wm = ar->hw_ce_regs->wm_srcr;
224 	u32 addr = ath10k_ce_read32(ar, ce_ctrl_addr + srcr_wm->addr);
225 
226 	ath10k_ce_write32(ar, ce_ctrl_addr + srcr_wm->addr,
227 			  (addr & ~(srcr_wm->wm_low->mask)) |
228 			  (ath10k_set_ring_byte(n, srcr_wm->wm_low)));
229 }
230 
231 static inline void ath10k_ce_dest_ring_highmark_set(struct ath10k *ar,
232 						    u32 ce_ctrl_addr,
233 						    unsigned int n)
234 {
235 	struct ath10k_hw_ce_dst_src_wm_regs *dstr_wm = ar->hw_ce_regs->wm_dstr;
236 	u32 addr = ath10k_ce_read32(ar, ce_ctrl_addr + dstr_wm->addr);
237 
238 	ath10k_ce_write32(ar, ce_ctrl_addr + dstr_wm->addr,
239 			  (addr & ~(dstr_wm->wm_high->mask)) |
240 			  (ath10k_set_ring_byte(n, dstr_wm->wm_high)));
241 }
242 
243 static inline void ath10k_ce_dest_ring_lowmark_set(struct ath10k *ar,
244 						   u32 ce_ctrl_addr,
245 						   unsigned int n)
246 {
247 	struct ath10k_hw_ce_dst_src_wm_regs *dstr_wm = ar->hw_ce_regs->wm_dstr;
248 	u32 addr = ath10k_ce_read32(ar, ce_ctrl_addr + dstr_wm->addr);
249 
250 	ath10k_ce_write32(ar, ce_ctrl_addr + dstr_wm->addr,
251 			  (addr & ~(dstr_wm->wm_low->mask)) |
252 			  (ath10k_set_ring_byte(n, dstr_wm->wm_low)));
253 }
254 
255 static inline void ath10k_ce_copy_complete_inter_enable(struct ath10k *ar,
256 							u32 ce_ctrl_addr)
257 {
258 	struct ath10k_hw_ce_host_ie *host_ie = ar->hw_ce_regs->host_ie;
259 
260 	u32 host_ie_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
261 					    ar->hw_ce_regs->host_ie_addr);
262 
263 	ath10k_ce_write32(ar, ce_ctrl_addr + ar->hw_ce_regs->host_ie_addr,
264 			  host_ie_addr | host_ie->copy_complete->mask);
265 }
266 
267 static inline void ath10k_ce_copy_complete_intr_disable(struct ath10k *ar,
268 							u32 ce_ctrl_addr)
269 {
270 	struct ath10k_hw_ce_host_ie *host_ie = ar->hw_ce_regs->host_ie;
271 
272 	u32 host_ie_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
273 					    ar->hw_ce_regs->host_ie_addr);
274 
275 	ath10k_ce_write32(ar, ce_ctrl_addr + ar->hw_ce_regs->host_ie_addr,
276 			  host_ie_addr & ~(host_ie->copy_complete->mask));
277 }
278 
279 static inline void ath10k_ce_watermark_intr_disable(struct ath10k *ar,
280 						    u32 ce_ctrl_addr)
281 {
282 	struct ath10k_hw_ce_host_wm_regs *wm_regs = ar->hw_ce_regs->wm_regs;
283 
284 	u32 host_ie_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
285 					    ar->hw_ce_regs->host_ie_addr);
286 
287 	ath10k_ce_write32(ar, ce_ctrl_addr + ar->hw_ce_regs->host_ie_addr,
288 			  host_ie_addr & ~(wm_regs->wm_mask));
289 }
290 
291 static inline void ath10k_ce_error_intr_enable(struct ath10k *ar,
292 					       u32 ce_ctrl_addr)
293 {
294 	struct ath10k_hw_ce_misc_regs *misc_regs = ar->hw_ce_regs->misc_regs;
295 
296 	u32 misc_ie_addr = ath10k_ce_read32(ar, ce_ctrl_addr +
297 					    ar->hw_ce_regs->misc_ie_addr);
298 
299 	ath10k_ce_write32(ar,
300 			  ce_ctrl_addr + ar->hw_ce_regs->misc_ie_addr,
301 			  misc_ie_addr | misc_regs->err_mask);
302 }
303 
304 static inline void ath10k_ce_error_intr_disable(struct ath10k *ar,
305 						u32 ce_ctrl_addr)
306 {
307 	struct ath10k_hw_ce_misc_regs *misc_regs = ar->hw_ce_regs->misc_regs;
308 
309 	u32 misc_ie_addr = ath10k_ce_read32(ar,
310 			ce_ctrl_addr + ar->hw_ce_regs->misc_ie_addr);
311 
312 	ath10k_ce_write32(ar,
313 			  ce_ctrl_addr + ar->hw_ce_regs->misc_ie_addr,
314 			  misc_ie_addr & ~(misc_regs->err_mask));
315 }
316 
317 static inline void ath10k_ce_engine_int_status_clear(struct ath10k *ar,
318 						     u32 ce_ctrl_addr,
319 						     unsigned int mask)
320 {
321 	struct ath10k_hw_ce_host_wm_regs *wm_regs = ar->hw_ce_regs->wm_regs;
322 
323 	ath10k_ce_write32(ar, ce_ctrl_addr + wm_regs->addr, mask);
324 }
325 
326 /*
327  * Guts of ath10k_ce_send.
328  * The caller takes responsibility for any needed locking.
329  */
330 int ath10k_ce_send_nolock(struct ath10k_ce_pipe *ce_state,
331 			  void *per_transfer_context,
332 			  u32 buffer,
333 			  unsigned int nbytes,
334 			  unsigned int transfer_id,
335 			  unsigned int flags)
336 {
337 	struct ath10k *ar = ce_state->ar;
338 	struct ath10k_ce_ring *src_ring = ce_state->src_ring;
339 	struct ce_desc *desc, sdesc;
340 	unsigned int nentries_mask = src_ring->nentries_mask;
341 	unsigned int sw_index = src_ring->sw_index;
342 	unsigned int write_index = src_ring->write_index;
343 	u32 ctrl_addr = ce_state->ctrl_addr;
344 	u32 desc_flags = 0;
345 	int ret = 0;
346 
347 	if (nbytes > ce_state->src_sz_max)
348 		ath10k_warn(ar, "%s: send more we can (nbytes: %d, max: %d)\n",
349 			    __func__, nbytes, ce_state->src_sz_max);
350 
351 	if (unlikely(CE_RING_DELTA(nentries_mask,
352 				   write_index, sw_index - 1) <= 0)) {
353 		ret = -ENOSR;
354 		goto exit;
355 	}
356 
357 	desc = CE_SRC_RING_TO_DESC(src_ring->base_addr_owner_space,
358 				   write_index);
359 
360 	desc_flags |= SM(transfer_id, CE_DESC_FLAGS_META_DATA);
361 
362 	if (flags & CE_SEND_FLAG_GATHER)
363 		desc_flags |= CE_DESC_FLAGS_GATHER;
364 	if (flags & CE_SEND_FLAG_BYTE_SWAP)
365 		desc_flags |= CE_DESC_FLAGS_BYTE_SWAP;
366 
367 	sdesc.addr   = __cpu_to_le32(buffer);
368 	sdesc.nbytes = __cpu_to_le16(nbytes);
369 	sdesc.flags  = __cpu_to_le16(desc_flags);
370 
371 	*desc = sdesc;
372 
373 	src_ring->per_transfer_context[write_index] = per_transfer_context;
374 
375 	/* Update Source Ring Write Index */
376 	write_index = CE_RING_IDX_INCR(nentries_mask, write_index);
377 
378 	/* WORKAROUND */
379 	if (!(flags & CE_SEND_FLAG_GATHER))
380 		ath10k_ce_src_ring_write_index_set(ar, ctrl_addr, write_index);
381 
382 	src_ring->write_index = write_index;
383 exit:
384 	return ret;
385 }
386 
387 void __ath10k_ce_send_revert(struct ath10k_ce_pipe *pipe)
388 {
389 	struct ath10k *ar = pipe->ar;
390 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
391 	struct ath10k_ce_ring *src_ring = pipe->src_ring;
392 	u32 ctrl_addr = pipe->ctrl_addr;
393 
394 	lockdep_assert_held(&ce->ce_lock);
395 
396 	/*
397 	 * This function must be called only if there is an incomplete
398 	 * scatter-gather transfer (before index register is updated)
399 	 * that needs to be cleaned up.
400 	 */
401 	if (WARN_ON_ONCE(src_ring->write_index == src_ring->sw_index))
402 		return;
403 
404 	if (WARN_ON_ONCE(src_ring->write_index ==
405 			 ath10k_ce_src_ring_write_index_get(ar, ctrl_addr)))
406 		return;
407 
408 	src_ring->write_index--;
409 	src_ring->write_index &= src_ring->nentries_mask;
410 
411 	src_ring->per_transfer_context[src_ring->write_index] = NULL;
412 }
413 
414 int ath10k_ce_send(struct ath10k_ce_pipe *ce_state,
415 		   void *per_transfer_context,
416 		   u32 buffer,
417 		   unsigned int nbytes,
418 		   unsigned int transfer_id,
419 		   unsigned int flags)
420 {
421 	struct ath10k *ar = ce_state->ar;
422 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
423 	int ret;
424 
425 	spin_lock_bh(&ce->ce_lock);
426 	ret = ath10k_ce_send_nolock(ce_state, per_transfer_context,
427 				    buffer, nbytes, transfer_id, flags);
428 	spin_unlock_bh(&ce->ce_lock);
429 
430 	return ret;
431 }
432 
433 int ath10k_ce_num_free_src_entries(struct ath10k_ce_pipe *pipe)
434 {
435 	struct ath10k *ar = pipe->ar;
436 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
437 	int delta;
438 
439 	spin_lock_bh(&ce->ce_lock);
440 	delta = CE_RING_DELTA(pipe->src_ring->nentries_mask,
441 			      pipe->src_ring->write_index,
442 			      pipe->src_ring->sw_index - 1);
443 	spin_unlock_bh(&ce->ce_lock);
444 
445 	return delta;
446 }
447 
448 int __ath10k_ce_rx_num_free_bufs(struct ath10k_ce_pipe *pipe)
449 {
450 	struct ath10k *ar = pipe->ar;
451 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
452 	struct ath10k_ce_ring *dest_ring = pipe->dest_ring;
453 	unsigned int nentries_mask = dest_ring->nentries_mask;
454 	unsigned int write_index = dest_ring->write_index;
455 	unsigned int sw_index = dest_ring->sw_index;
456 
457 	lockdep_assert_held(&ce->ce_lock);
458 
459 	return CE_RING_DELTA(nentries_mask, write_index, sw_index - 1);
460 }
461 
462 int __ath10k_ce_rx_post_buf(struct ath10k_ce_pipe *pipe, void *ctx, u32 paddr)
463 {
464 	struct ath10k *ar = pipe->ar;
465 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
466 	struct ath10k_ce_ring *dest_ring = pipe->dest_ring;
467 	unsigned int nentries_mask = dest_ring->nentries_mask;
468 	unsigned int write_index = dest_ring->write_index;
469 	unsigned int sw_index = dest_ring->sw_index;
470 	struct ce_desc *base = dest_ring->base_addr_owner_space;
471 	struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, write_index);
472 	u32 ctrl_addr = pipe->ctrl_addr;
473 
474 	lockdep_assert_held(&ce->ce_lock);
475 
476 	if ((pipe->id != 5) &&
477 	    CE_RING_DELTA(nentries_mask, write_index, sw_index - 1) == 0)
478 		return -ENOSPC;
479 
480 	desc->addr = __cpu_to_le32(paddr);
481 	desc->nbytes = 0;
482 
483 	dest_ring->per_transfer_context[write_index] = ctx;
484 	write_index = CE_RING_IDX_INCR(nentries_mask, write_index);
485 	ath10k_ce_dest_ring_write_index_set(ar, ctrl_addr, write_index);
486 	dest_ring->write_index = write_index;
487 
488 	return 0;
489 }
490 
491 void ath10k_ce_rx_update_write_idx(struct ath10k_ce_pipe *pipe, u32 nentries)
492 {
493 	struct ath10k *ar = pipe->ar;
494 	struct ath10k_ce_ring *dest_ring = pipe->dest_ring;
495 	unsigned int nentries_mask = dest_ring->nentries_mask;
496 	unsigned int write_index = dest_ring->write_index;
497 	u32 ctrl_addr = pipe->ctrl_addr;
498 	u32 cur_write_idx = ath10k_ce_dest_ring_write_index_get(ar, ctrl_addr);
499 
500 	/* Prevent CE ring stuck issue that will occur when ring is full.
501 	 * Make sure that write index is 1 less than read index.
502 	 */
503 	if ((cur_write_idx + nentries)  == dest_ring->sw_index)
504 		nentries -= 1;
505 
506 	write_index = CE_RING_IDX_ADD(nentries_mask, write_index, nentries);
507 	ath10k_ce_dest_ring_write_index_set(ar, ctrl_addr, write_index);
508 	dest_ring->write_index = write_index;
509 }
510 
511 int ath10k_ce_rx_post_buf(struct ath10k_ce_pipe *pipe, void *ctx, u32 paddr)
512 {
513 	struct ath10k *ar = pipe->ar;
514 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
515 	int ret;
516 
517 	spin_lock_bh(&ce->ce_lock);
518 	ret = __ath10k_ce_rx_post_buf(pipe, ctx, paddr);
519 	spin_unlock_bh(&ce->ce_lock);
520 
521 	return ret;
522 }
523 
524 /*
525  * Guts of ath10k_ce_completed_recv_next.
526  * The caller takes responsibility for any necessary locking.
527  */
528 int ath10k_ce_completed_recv_next_nolock(struct ath10k_ce_pipe *ce_state,
529 					 void **per_transfer_contextp,
530 					 unsigned int *nbytesp)
531 {
532 	struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
533 	unsigned int nentries_mask = dest_ring->nentries_mask;
534 	unsigned int sw_index = dest_ring->sw_index;
535 
536 	struct ce_desc *base = dest_ring->base_addr_owner_space;
537 	struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, sw_index);
538 	struct ce_desc sdesc;
539 	u16 nbytes;
540 
541 	/* Copy in one go for performance reasons */
542 	sdesc = *desc;
543 
544 	nbytes = __le16_to_cpu(sdesc.nbytes);
545 	if (nbytes == 0) {
546 		/*
547 		 * This closes a relatively unusual race where the Host
548 		 * sees the updated DRRI before the update to the
549 		 * corresponding descriptor has completed. We treat this
550 		 * as a descriptor that is not yet done.
551 		 */
552 		return -EIO;
553 	}
554 
555 	desc->nbytes = 0;
556 
557 	/* Return data from completed destination descriptor */
558 	*nbytesp = nbytes;
559 
560 	if (per_transfer_contextp)
561 		*per_transfer_contextp =
562 			dest_ring->per_transfer_context[sw_index];
563 
564 	/* Copy engine 5 (HTT Rx) will reuse the same transfer context.
565 	 * So update transfer context all CEs except CE5.
566 	 */
567 	if (ce_state->id != 5)
568 		dest_ring->per_transfer_context[sw_index] = NULL;
569 
570 	/* Update sw_index */
571 	sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
572 	dest_ring->sw_index = sw_index;
573 
574 	return 0;
575 }
576 
577 int ath10k_ce_completed_recv_next(struct ath10k_ce_pipe *ce_state,
578 				  void **per_transfer_contextp,
579 				  unsigned int *nbytesp)
580 {
581 	struct ath10k *ar = ce_state->ar;
582 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
583 	int ret;
584 
585 	spin_lock_bh(&ce->ce_lock);
586 	ret = ath10k_ce_completed_recv_next_nolock(ce_state,
587 						   per_transfer_contextp,
588 						   nbytesp);
589 	spin_unlock_bh(&ce->ce_lock);
590 
591 	return ret;
592 }
593 
594 int ath10k_ce_revoke_recv_next(struct ath10k_ce_pipe *ce_state,
595 			       void **per_transfer_contextp,
596 			       u32 *bufferp)
597 {
598 	struct ath10k_ce_ring *dest_ring;
599 	unsigned int nentries_mask;
600 	unsigned int sw_index;
601 	unsigned int write_index;
602 	int ret;
603 	struct ath10k *ar;
604 	struct ath10k_ce *ce;
605 
606 	dest_ring = ce_state->dest_ring;
607 
608 	if (!dest_ring)
609 		return -EIO;
610 
611 	ar = ce_state->ar;
612 	ce = ath10k_ce_priv(ar);
613 
614 	spin_lock_bh(&ce->ce_lock);
615 
616 	nentries_mask = dest_ring->nentries_mask;
617 	sw_index = dest_ring->sw_index;
618 	write_index = dest_ring->write_index;
619 	if (write_index != sw_index) {
620 		struct ce_desc *base = dest_ring->base_addr_owner_space;
621 		struct ce_desc *desc = CE_DEST_RING_TO_DESC(base, sw_index);
622 
623 		/* Return data from completed destination descriptor */
624 		*bufferp = __le32_to_cpu(desc->addr);
625 
626 		if (per_transfer_contextp)
627 			*per_transfer_contextp =
628 				dest_ring->per_transfer_context[sw_index];
629 
630 		/* sanity */
631 		dest_ring->per_transfer_context[sw_index] = NULL;
632 		desc->nbytes = 0;
633 
634 		/* Update sw_index */
635 		sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
636 		dest_ring->sw_index = sw_index;
637 		ret = 0;
638 	} else {
639 		ret = -EIO;
640 	}
641 
642 	spin_unlock_bh(&ce->ce_lock);
643 
644 	return ret;
645 }
646 
647 /*
648  * Guts of ath10k_ce_completed_send_next.
649  * The caller takes responsibility for any necessary locking.
650  */
651 int ath10k_ce_completed_send_next_nolock(struct ath10k_ce_pipe *ce_state,
652 					 void **per_transfer_contextp)
653 {
654 	struct ath10k_ce_ring *src_ring = ce_state->src_ring;
655 	u32 ctrl_addr = ce_state->ctrl_addr;
656 	struct ath10k *ar = ce_state->ar;
657 	unsigned int nentries_mask = src_ring->nentries_mask;
658 	unsigned int sw_index = src_ring->sw_index;
659 	unsigned int read_index;
660 	struct ce_desc *desc;
661 
662 	if (src_ring->hw_index == sw_index) {
663 		/*
664 		 * The SW completion index has caught up with the cached
665 		 * version of the HW completion index.
666 		 * Update the cached HW completion index to see whether
667 		 * the SW has really caught up to the HW, or if the cached
668 		 * value of the HW index has become stale.
669 		 */
670 
671 		read_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
672 		if (read_index == 0xffffffff)
673 			return -ENODEV;
674 
675 		read_index &= nentries_mask;
676 		src_ring->hw_index = read_index;
677 	}
678 
679 	read_index = src_ring->hw_index;
680 
681 	if (read_index == sw_index)
682 		return -EIO;
683 
684 	if (per_transfer_contextp)
685 		*per_transfer_contextp =
686 			src_ring->per_transfer_context[sw_index];
687 
688 	/* sanity */
689 	src_ring->per_transfer_context[sw_index] = NULL;
690 	desc = CE_SRC_RING_TO_DESC(src_ring->base_addr_owner_space,
691 				   sw_index);
692 	desc->nbytes = 0;
693 
694 	/* Update sw_index */
695 	sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
696 	src_ring->sw_index = sw_index;
697 
698 	return 0;
699 }
700 
701 /* NB: Modeled after ath10k_ce_completed_send_next */
702 int ath10k_ce_cancel_send_next(struct ath10k_ce_pipe *ce_state,
703 			       void **per_transfer_contextp,
704 			       u32 *bufferp,
705 			       unsigned int *nbytesp,
706 			       unsigned int *transfer_idp)
707 {
708 	struct ath10k_ce_ring *src_ring;
709 	unsigned int nentries_mask;
710 	unsigned int sw_index;
711 	unsigned int write_index;
712 	int ret;
713 	struct ath10k *ar;
714 	struct ath10k_ce *ce;
715 
716 	src_ring = ce_state->src_ring;
717 
718 	if (!src_ring)
719 		return -EIO;
720 
721 	ar = ce_state->ar;
722 	ce = ath10k_ce_priv(ar);
723 
724 	spin_lock_bh(&ce->ce_lock);
725 
726 	nentries_mask = src_ring->nentries_mask;
727 	sw_index = src_ring->sw_index;
728 	write_index = src_ring->write_index;
729 
730 	if (write_index != sw_index) {
731 		struct ce_desc *base = src_ring->base_addr_owner_space;
732 		struct ce_desc *desc = CE_SRC_RING_TO_DESC(base, sw_index);
733 
734 		/* Return data from completed source descriptor */
735 		*bufferp = __le32_to_cpu(desc->addr);
736 		*nbytesp = __le16_to_cpu(desc->nbytes);
737 		*transfer_idp = MS(__le16_to_cpu(desc->flags),
738 						CE_DESC_FLAGS_META_DATA);
739 
740 		if (per_transfer_contextp)
741 			*per_transfer_contextp =
742 				src_ring->per_transfer_context[sw_index];
743 
744 		/* sanity */
745 		src_ring->per_transfer_context[sw_index] = NULL;
746 
747 		/* Update sw_index */
748 		sw_index = CE_RING_IDX_INCR(nentries_mask, sw_index);
749 		src_ring->sw_index = sw_index;
750 		ret = 0;
751 	} else {
752 		ret = -EIO;
753 	}
754 
755 	spin_unlock_bh(&ce->ce_lock);
756 
757 	return ret;
758 }
759 
760 int ath10k_ce_completed_send_next(struct ath10k_ce_pipe *ce_state,
761 				  void **per_transfer_contextp)
762 {
763 	struct ath10k *ar = ce_state->ar;
764 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
765 	int ret;
766 
767 	spin_lock_bh(&ce->ce_lock);
768 	ret = ath10k_ce_completed_send_next_nolock(ce_state,
769 						   per_transfer_contextp);
770 	spin_unlock_bh(&ce->ce_lock);
771 
772 	return ret;
773 }
774 
775 /*
776  * Guts of interrupt handler for per-engine interrupts on a particular CE.
777  *
778  * Invokes registered callbacks for recv_complete,
779  * send_complete, and watermarks.
780  */
781 void ath10k_ce_per_engine_service(struct ath10k *ar, unsigned int ce_id)
782 {
783 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
784 	struct ath10k_ce_pipe *ce_state = &ce->ce_states[ce_id];
785 	struct ath10k_hw_ce_host_wm_regs *wm_regs = ar->hw_ce_regs->wm_regs;
786 	u32 ctrl_addr = ce_state->ctrl_addr;
787 
788 	spin_lock_bh(&ce->ce_lock);
789 
790 	/* Clear the copy-complete interrupts that will be handled here. */
791 	ath10k_ce_engine_int_status_clear(ar, ctrl_addr,
792 					  wm_regs->cc_mask);
793 
794 	spin_unlock_bh(&ce->ce_lock);
795 
796 	if (ce_state->recv_cb)
797 		ce_state->recv_cb(ce_state);
798 
799 	if (ce_state->send_cb)
800 		ce_state->send_cb(ce_state);
801 
802 	spin_lock_bh(&ce->ce_lock);
803 
804 	/*
805 	 * Misc CE interrupts are not being handled, but still need
806 	 * to be cleared.
807 	 */
808 	ath10k_ce_engine_int_status_clear(ar, ctrl_addr, wm_regs->wm_mask);
809 
810 	spin_unlock_bh(&ce->ce_lock);
811 }
812 
813 /*
814  * Handler for per-engine interrupts on ALL active CEs.
815  * This is used in cases where the system is sharing a
816  * single interrput for all CEs
817  */
818 
819 void ath10k_ce_per_engine_service_any(struct ath10k *ar)
820 {
821 	int ce_id;
822 	u32 intr_summary;
823 
824 	intr_summary = ath10k_ce_interrupt_summary(ar);
825 
826 	for (ce_id = 0; intr_summary && (ce_id < CE_COUNT); ce_id++) {
827 		if (intr_summary & (1 << ce_id))
828 			intr_summary &= ~(1 << ce_id);
829 		else
830 			/* no intr pending on this CE */
831 			continue;
832 
833 		ath10k_ce_per_engine_service(ar, ce_id);
834 	}
835 }
836 
837 /*
838  * Adjust interrupts for the copy complete handler.
839  * If it's needed for either send or recv, then unmask
840  * this interrupt; otherwise, mask it.
841  *
842  * Called with ce_lock held.
843  */
844 static void ath10k_ce_per_engine_handler_adjust(struct ath10k_ce_pipe *ce_state)
845 {
846 	u32 ctrl_addr = ce_state->ctrl_addr;
847 	struct ath10k *ar = ce_state->ar;
848 	bool disable_copy_compl_intr = ce_state->attr_flags & CE_ATTR_DIS_INTR;
849 
850 	if ((!disable_copy_compl_intr) &&
851 	    (ce_state->send_cb || ce_state->recv_cb))
852 		ath10k_ce_copy_complete_inter_enable(ar, ctrl_addr);
853 	else
854 		ath10k_ce_copy_complete_intr_disable(ar, ctrl_addr);
855 
856 	ath10k_ce_watermark_intr_disable(ar, ctrl_addr);
857 }
858 
859 int ath10k_ce_disable_interrupts(struct ath10k *ar)
860 {
861 	int ce_id;
862 
863 	for (ce_id = 0; ce_id < CE_COUNT; ce_id++) {
864 		u32 ctrl_addr = ath10k_ce_base_address(ar, ce_id);
865 
866 		ath10k_ce_copy_complete_intr_disable(ar, ctrl_addr);
867 		ath10k_ce_error_intr_disable(ar, ctrl_addr);
868 		ath10k_ce_watermark_intr_disable(ar, ctrl_addr);
869 	}
870 
871 	return 0;
872 }
873 
874 void ath10k_ce_enable_interrupts(struct ath10k *ar)
875 {
876 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
877 	int ce_id;
878 	struct ath10k_ce_pipe *ce_state;
879 
880 	/* Skip the last copy engine, CE7 the diagnostic window, as that
881 	 * uses polling and isn't initialized for interrupts.
882 	 */
883 	for (ce_id = 0; ce_id < CE_COUNT - 1; ce_id++) {
884 		ce_state  = &ce->ce_states[ce_id];
885 		ath10k_ce_per_engine_handler_adjust(ce_state);
886 	}
887 }
888 
889 static int ath10k_ce_init_src_ring(struct ath10k *ar,
890 				   unsigned int ce_id,
891 				   const struct ce_attr *attr)
892 {
893 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
894 	struct ath10k_ce_pipe *ce_state = &ce->ce_states[ce_id];
895 	struct ath10k_ce_ring *src_ring = ce_state->src_ring;
896 	u32 nentries, ctrl_addr = ath10k_ce_base_address(ar, ce_id);
897 
898 	nentries = roundup_pow_of_two(attr->src_nentries);
899 
900 	memset(src_ring->base_addr_owner_space, 0,
901 	       nentries * sizeof(struct ce_desc));
902 
903 	src_ring->sw_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
904 	src_ring->sw_index &= src_ring->nentries_mask;
905 	src_ring->hw_index = src_ring->sw_index;
906 
907 	src_ring->write_index =
908 		ath10k_ce_src_ring_write_index_get(ar, ctrl_addr);
909 	src_ring->write_index &= src_ring->nentries_mask;
910 
911 	ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
912 					 src_ring->base_addr_ce_space);
913 	ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
914 	ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
915 	ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
916 	ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
917 	ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
918 
919 	ath10k_dbg(ar, ATH10K_DBG_BOOT,
920 		   "boot init ce src ring id %d entries %d base_addr %pK\n",
921 		   ce_id, nentries, src_ring->base_addr_owner_space);
922 
923 	return 0;
924 }
925 
926 static int ath10k_ce_init_dest_ring(struct ath10k *ar,
927 				    unsigned int ce_id,
928 				    const struct ce_attr *attr)
929 {
930 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
931 	struct ath10k_ce_pipe *ce_state = &ce->ce_states[ce_id];
932 	struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
933 	u32 nentries, ctrl_addr = ath10k_ce_base_address(ar, ce_id);
934 
935 	nentries = roundup_pow_of_two(attr->dest_nentries);
936 
937 	memset(dest_ring->base_addr_owner_space, 0,
938 	       nentries * sizeof(struct ce_desc));
939 
940 	dest_ring->sw_index = ath10k_ce_dest_ring_read_index_get(ar, ctrl_addr);
941 	dest_ring->sw_index &= dest_ring->nentries_mask;
942 	dest_ring->write_index =
943 		ath10k_ce_dest_ring_write_index_get(ar, ctrl_addr);
944 	dest_ring->write_index &= dest_ring->nentries_mask;
945 
946 	ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr,
947 					  dest_ring->base_addr_ce_space);
948 	ath10k_ce_dest_ring_size_set(ar, ctrl_addr, nentries);
949 	ath10k_ce_dest_ring_byte_swap_set(ar, ctrl_addr, 0);
950 	ath10k_ce_dest_ring_lowmark_set(ar, ctrl_addr, 0);
951 	ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, nentries);
952 
953 	ath10k_dbg(ar, ATH10K_DBG_BOOT,
954 		   "boot ce dest ring id %d entries %d base_addr %pK\n",
955 		   ce_id, nentries, dest_ring->base_addr_owner_space);
956 
957 	return 0;
958 }
959 
960 static struct ath10k_ce_ring *
961 ath10k_ce_alloc_src_ring(struct ath10k *ar, unsigned int ce_id,
962 			 const struct ce_attr *attr)
963 {
964 	struct ath10k_ce_ring *src_ring;
965 	u32 nentries = attr->src_nentries;
966 	dma_addr_t base_addr;
967 
968 	nentries = roundup_pow_of_two(nentries);
969 
970 	src_ring = kzalloc(sizeof(*src_ring) +
971 			   (nentries *
972 			    sizeof(*src_ring->per_transfer_context)),
973 			   GFP_KERNEL);
974 	if (src_ring == NULL)
975 		return ERR_PTR(-ENOMEM);
976 
977 	src_ring->nentries = nentries;
978 	src_ring->nentries_mask = nentries - 1;
979 
980 	/*
981 	 * Legacy platforms that do not support cache
982 	 * coherent DMA are unsupported
983 	 */
984 	src_ring->base_addr_owner_space_unaligned =
985 		dma_alloc_coherent(ar->dev,
986 				   (nentries * sizeof(struct ce_desc) +
987 				    CE_DESC_RING_ALIGN),
988 				   &base_addr, GFP_KERNEL);
989 	if (!src_ring->base_addr_owner_space_unaligned) {
990 		kfree(src_ring);
991 		return ERR_PTR(-ENOMEM);
992 	}
993 
994 	src_ring->base_addr_ce_space_unaligned = base_addr;
995 
996 	src_ring->base_addr_owner_space = PTR_ALIGN(
997 			src_ring->base_addr_owner_space_unaligned,
998 			CE_DESC_RING_ALIGN);
999 	src_ring->base_addr_ce_space = ALIGN(
1000 			src_ring->base_addr_ce_space_unaligned,
1001 			CE_DESC_RING_ALIGN);
1002 
1003 	return src_ring;
1004 }
1005 
1006 static struct ath10k_ce_ring *
1007 ath10k_ce_alloc_dest_ring(struct ath10k *ar, unsigned int ce_id,
1008 			  const struct ce_attr *attr)
1009 {
1010 	struct ath10k_ce_ring *dest_ring;
1011 	u32 nentries;
1012 	dma_addr_t base_addr;
1013 
1014 	nentries = roundup_pow_of_two(attr->dest_nentries);
1015 
1016 	dest_ring = kzalloc(sizeof(*dest_ring) +
1017 			    (nentries *
1018 			     sizeof(*dest_ring->per_transfer_context)),
1019 			    GFP_KERNEL);
1020 	if (dest_ring == NULL)
1021 		return ERR_PTR(-ENOMEM);
1022 
1023 	dest_ring->nentries = nentries;
1024 	dest_ring->nentries_mask = nentries - 1;
1025 
1026 	/*
1027 	 * Legacy platforms that do not support cache
1028 	 * coherent DMA are unsupported
1029 	 */
1030 	dest_ring->base_addr_owner_space_unaligned =
1031 		dma_zalloc_coherent(ar->dev,
1032 				    (nentries * sizeof(struct ce_desc) +
1033 				     CE_DESC_RING_ALIGN),
1034 				    &base_addr, GFP_KERNEL);
1035 	if (!dest_ring->base_addr_owner_space_unaligned) {
1036 		kfree(dest_ring);
1037 		return ERR_PTR(-ENOMEM);
1038 	}
1039 
1040 	dest_ring->base_addr_ce_space_unaligned = base_addr;
1041 
1042 	dest_ring->base_addr_owner_space = PTR_ALIGN(
1043 			dest_ring->base_addr_owner_space_unaligned,
1044 			CE_DESC_RING_ALIGN);
1045 	dest_ring->base_addr_ce_space = ALIGN(
1046 			dest_ring->base_addr_ce_space_unaligned,
1047 			CE_DESC_RING_ALIGN);
1048 
1049 	return dest_ring;
1050 }
1051 
1052 /*
1053  * Initialize a Copy Engine based on caller-supplied attributes.
1054  * This may be called once to initialize both source and destination
1055  * rings or it may be called twice for separate source and destination
1056  * initialization. It may be that only one side or the other is
1057  * initialized by software/firmware.
1058  */
1059 int ath10k_ce_init_pipe(struct ath10k *ar, unsigned int ce_id,
1060 			const struct ce_attr *attr)
1061 {
1062 	int ret;
1063 
1064 	if (attr->src_nentries) {
1065 		ret = ath10k_ce_init_src_ring(ar, ce_id, attr);
1066 		if (ret) {
1067 			ath10k_err(ar, "Failed to initialize CE src ring for ID: %d (%d)\n",
1068 				   ce_id, ret);
1069 			return ret;
1070 		}
1071 	}
1072 
1073 	if (attr->dest_nentries) {
1074 		ret = ath10k_ce_init_dest_ring(ar, ce_id, attr);
1075 		if (ret) {
1076 			ath10k_err(ar, "Failed to initialize CE dest ring for ID: %d (%d)\n",
1077 				   ce_id, ret);
1078 			return ret;
1079 		}
1080 	}
1081 
1082 	return 0;
1083 }
1084 
1085 static void ath10k_ce_deinit_src_ring(struct ath10k *ar, unsigned int ce_id)
1086 {
1087 	u32 ctrl_addr = ath10k_ce_base_address(ar, ce_id);
1088 
1089 	ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr, 0);
1090 	ath10k_ce_src_ring_size_set(ar, ctrl_addr, 0);
1091 	ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, 0);
1092 	ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, 0);
1093 }
1094 
1095 static void ath10k_ce_deinit_dest_ring(struct ath10k *ar, unsigned int ce_id)
1096 {
1097 	u32 ctrl_addr = ath10k_ce_base_address(ar, ce_id);
1098 
1099 	ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr, 0);
1100 	ath10k_ce_dest_ring_size_set(ar, ctrl_addr, 0);
1101 	ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, 0);
1102 }
1103 
1104 void ath10k_ce_deinit_pipe(struct ath10k *ar, unsigned int ce_id)
1105 {
1106 	ath10k_ce_deinit_src_ring(ar, ce_id);
1107 	ath10k_ce_deinit_dest_ring(ar, ce_id);
1108 }
1109 
1110 int ath10k_ce_alloc_pipe(struct ath10k *ar, int ce_id,
1111 			 const struct ce_attr *attr)
1112 {
1113 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
1114 	struct ath10k_ce_pipe *ce_state = &ce->ce_states[ce_id];
1115 	int ret;
1116 
1117 	/*
1118 	 * Make sure there's enough CE ringbuffer entries for HTT TX to avoid
1119 	 * additional TX locking checks.
1120 	 *
1121 	 * For the lack of a better place do the check here.
1122 	 */
1123 	BUILD_BUG_ON(2 * TARGET_NUM_MSDU_DESC >
1124 		     (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1125 	BUILD_BUG_ON(2 * TARGET_10_4_NUM_MSDU_DESC_PFC >
1126 		     (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1127 	BUILD_BUG_ON(2 * TARGET_TLV_NUM_MSDU_DESC >
1128 		     (CE_HTT_H2T_MSG_SRC_NENTRIES - 1));
1129 
1130 	ce_state->ar = ar;
1131 	ce_state->id = ce_id;
1132 	ce_state->ctrl_addr = ath10k_ce_base_address(ar, ce_id);
1133 	ce_state->attr_flags = attr->flags;
1134 	ce_state->src_sz_max = attr->src_sz_max;
1135 
1136 	if (attr->src_nentries)
1137 		ce_state->send_cb = attr->send_cb;
1138 
1139 	if (attr->dest_nentries)
1140 		ce_state->recv_cb = attr->recv_cb;
1141 
1142 	if (attr->src_nentries) {
1143 		ce_state->src_ring = ath10k_ce_alloc_src_ring(ar, ce_id, attr);
1144 		if (IS_ERR(ce_state->src_ring)) {
1145 			ret = PTR_ERR(ce_state->src_ring);
1146 			ath10k_err(ar, "failed to allocate copy engine source ring %d: %d\n",
1147 				   ce_id, ret);
1148 			ce_state->src_ring = NULL;
1149 			return ret;
1150 		}
1151 	}
1152 
1153 	if (attr->dest_nentries) {
1154 		ce_state->dest_ring = ath10k_ce_alloc_dest_ring(ar, ce_id,
1155 								attr);
1156 		if (IS_ERR(ce_state->dest_ring)) {
1157 			ret = PTR_ERR(ce_state->dest_ring);
1158 			ath10k_err(ar, "failed to allocate copy engine destination ring %d: %d\n",
1159 				   ce_id, ret);
1160 			ce_state->dest_ring = NULL;
1161 			return ret;
1162 		}
1163 	}
1164 
1165 	return 0;
1166 }
1167 
1168 void ath10k_ce_free_pipe(struct ath10k *ar, int ce_id)
1169 {
1170 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
1171 	struct ath10k_ce_pipe *ce_state = &ce->ce_states[ce_id];
1172 
1173 	if (ce_state->src_ring) {
1174 		dma_free_coherent(ar->dev,
1175 				  (ce_state->src_ring->nentries *
1176 				   sizeof(struct ce_desc) +
1177 				   CE_DESC_RING_ALIGN),
1178 				  ce_state->src_ring->base_addr_owner_space,
1179 				  ce_state->src_ring->base_addr_ce_space);
1180 		kfree(ce_state->src_ring);
1181 	}
1182 
1183 	if (ce_state->dest_ring) {
1184 		dma_free_coherent(ar->dev,
1185 				  (ce_state->dest_ring->nentries *
1186 				   sizeof(struct ce_desc) +
1187 				   CE_DESC_RING_ALIGN),
1188 				  ce_state->dest_ring->base_addr_owner_space,
1189 				  ce_state->dest_ring->base_addr_ce_space);
1190 		kfree(ce_state->dest_ring);
1191 	}
1192 
1193 	ce_state->src_ring = NULL;
1194 	ce_state->dest_ring = NULL;
1195 }
1196 
1197 void ath10k_ce_dump_registers(struct ath10k *ar,
1198 			      struct ath10k_fw_crash_data *crash_data)
1199 {
1200 	struct ath10k_ce *ce = ath10k_ce_priv(ar);
1201 	struct ath10k_ce_crash_data ce_data;
1202 	u32 addr, id;
1203 
1204 	lockdep_assert_held(&ar->data_lock);
1205 
1206 	ath10k_err(ar, "Copy Engine register dump:\n");
1207 
1208 	spin_lock_bh(&ce->ce_lock);
1209 	for (id = 0; id < CE_COUNT; id++) {
1210 		addr = ath10k_ce_base_address(ar, id);
1211 		ce_data.base_addr = cpu_to_le32(addr);
1212 
1213 		ce_data.src_wr_idx =
1214 			cpu_to_le32(ath10k_ce_src_ring_write_index_get(ar, addr));
1215 		ce_data.src_r_idx =
1216 			cpu_to_le32(ath10k_ce_src_ring_read_index_get(ar, addr));
1217 		ce_data.dst_wr_idx =
1218 			cpu_to_le32(ath10k_ce_dest_ring_write_index_get(ar, addr));
1219 		ce_data.dst_r_idx =
1220 			cpu_to_le32(ath10k_ce_dest_ring_read_index_get(ar, addr));
1221 
1222 		if (crash_data)
1223 			crash_data->ce_crash_data[id] = ce_data;
1224 
1225 		ath10k_err(ar, "[%02d]: 0x%08x %3u %3u %3u %3u", id,
1226 			   le32_to_cpu(ce_data.base_addr),
1227 			   le32_to_cpu(ce_data.src_wr_idx),
1228 			   le32_to_cpu(ce_data.src_r_idx),
1229 			   le32_to_cpu(ce_data.dst_wr_idx),
1230 			   le32_to_cpu(ce_data.dst_r_idx));
1231 	}
1232 
1233 	spin_unlock_bh(&ce->ce_lock);
1234 }
1235