1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Copyright (C) 2020-2023 Intel Corporation
4  */
5 #include <net/tso.h>
6 #include <linux/tcp.h>
7 
8 #include "iwl-debug.h"
9 #include "iwl-io.h"
10 #include "fw/api/commands.h"
11 #include "fw/api/tx.h"
12 #include "fw/api/datapath.h"
13 #include "fw/api/debug.h"
14 #include "queue/tx.h"
15 #include "iwl-fh.h"
16 #include "iwl-scd.h"
17 #include <linux/dmapool.h>
18 
19 /*
20  * iwl_txq_update_byte_tbl - Set up entry in Tx byte-count array
21  */
iwl_pcie_gen2_update_byte_tbl(struct iwl_trans * trans,struct iwl_txq * txq,u16 byte_cnt,int num_tbs)22 static void iwl_pcie_gen2_update_byte_tbl(struct iwl_trans *trans,
23 					  struct iwl_txq *txq, u16 byte_cnt,
24 					  int num_tbs)
25 {
26 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
27 	u8 filled_tfd_size, num_fetch_chunks;
28 	u16 len = byte_cnt;
29 	__le16 bc_ent;
30 
31 	if (WARN(idx >= txq->n_window, "%d >= %d\n", idx, txq->n_window))
32 		return;
33 
34 	filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) +
35 			  num_tbs * sizeof(struct iwl_tfh_tb);
36 	/*
37 	 * filled_tfd_size contains the number of filled bytes in the TFD.
38 	 * Dividing it by 64 will give the number of chunks to fetch
39 	 * to SRAM- 0 for one chunk, 1 for 2 and so on.
40 	 * If, for example, TFD contains only 3 TBs then 32 bytes
41 	 * of the TFD are used, and only one chunk of 64 bytes should
42 	 * be fetched
43 	 */
44 	num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1;
45 
46 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
47 		struct iwl_gen3_bc_tbl_entry *scd_bc_tbl_gen3 = txq->bc_tbl.addr;
48 
49 		/* Starting from AX210, the HW expects bytes */
50 		WARN_ON(trans->txqs.bc_table_dword);
51 		WARN_ON(len > 0x3FFF);
52 		bc_ent = cpu_to_le16(len | (num_fetch_chunks << 14));
53 		scd_bc_tbl_gen3[idx].tfd_offset = bc_ent;
54 	} else {
55 		struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr;
56 
57 		/* Before AX210, the HW expects DW */
58 		WARN_ON(!trans->txqs.bc_table_dword);
59 		len = DIV_ROUND_UP(len, 4);
60 		WARN_ON(len > 0xFFF);
61 		bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12));
62 		scd_bc_tbl->tfd_offset[idx] = bc_ent;
63 	}
64 }
65 
66 /*
67  * iwl_txq_inc_wr_ptr - Send new write index to hardware
68  */
iwl_txq_inc_wr_ptr(struct iwl_trans * trans,struct iwl_txq * txq)69 void iwl_txq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_txq *txq)
70 {
71 	lockdep_assert_held(&txq->lock);
72 
73 	IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr);
74 
75 	/*
76 	 * if not in power-save mode, uCode will never sleep when we're
77 	 * trying to tx (during RFKILL, we're not trying to tx).
78 	 */
79 	iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16));
80 }
81 
iwl_txq_gen2_get_num_tbs(struct iwl_trans * trans,struct iwl_tfh_tfd * tfd)82 static u8 iwl_txq_gen2_get_num_tbs(struct iwl_trans *trans,
83 				   struct iwl_tfh_tfd *tfd)
84 {
85 	return le16_to_cpu(tfd->num_tbs) & 0x1f;
86 }
87 
iwl_txq_gen2_set_tb(struct iwl_trans * trans,struct iwl_tfh_tfd * tfd,dma_addr_t addr,u16 len)88 int iwl_txq_gen2_set_tb(struct iwl_trans *trans, struct iwl_tfh_tfd *tfd,
89 			dma_addr_t addr, u16 len)
90 {
91 	int idx = iwl_txq_gen2_get_num_tbs(trans, tfd);
92 	struct iwl_tfh_tb *tb;
93 
94 	/* Only WARN here so we know about the issue, but we mess up our
95 	 * unmap path because not every place currently checks for errors
96 	 * returned from this function - it can only return an error if
97 	 * there's no more space, and so when we know there is enough we
98 	 * don't always check ...
99 	 */
100 	WARN(iwl_txq_crosses_4g_boundary(addr, len),
101 	     "possible DMA problem with iova:0x%llx, len:%d\n",
102 	     (unsigned long long)addr, len);
103 
104 	if (WARN_ON(idx >= IWL_TFH_NUM_TBS))
105 		return -EINVAL;
106 	tb = &tfd->tbs[idx];
107 
108 	/* Each TFD can point to a maximum max_tbs Tx buffers */
109 	if (le16_to_cpu(tfd->num_tbs) >= trans->txqs.tfd.max_tbs) {
110 		IWL_ERR(trans, "Error can not send more than %d chunks\n",
111 			trans->txqs.tfd.max_tbs);
112 		return -EINVAL;
113 	}
114 
115 	put_unaligned_le64(addr, &tb->addr);
116 	tb->tb_len = cpu_to_le16(len);
117 
118 	tfd->num_tbs = cpu_to_le16(idx + 1);
119 
120 	return idx;
121 }
122 
iwl_txq_set_tfd_invalid_gen2(struct iwl_trans * trans,struct iwl_tfh_tfd * tfd)123 static void iwl_txq_set_tfd_invalid_gen2(struct iwl_trans *trans,
124 					 struct iwl_tfh_tfd *tfd)
125 {
126 	tfd->num_tbs = 0;
127 
128 	iwl_txq_gen2_set_tb(trans, tfd, trans->invalid_tx_cmd.dma,
129 			    trans->invalid_tx_cmd.size);
130 }
131 
iwl_txq_gen2_tfd_unmap(struct iwl_trans * trans,struct iwl_cmd_meta * meta,struct iwl_tfh_tfd * tfd)132 void iwl_txq_gen2_tfd_unmap(struct iwl_trans *trans, struct iwl_cmd_meta *meta,
133 			    struct iwl_tfh_tfd *tfd)
134 {
135 	int i, num_tbs;
136 
137 	/* Sanity check on number of chunks */
138 	num_tbs = iwl_txq_gen2_get_num_tbs(trans, tfd);
139 
140 	if (num_tbs > trans->txqs.tfd.max_tbs) {
141 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
142 		return;
143 	}
144 
145 	/* first TB is never freed - it's the bidirectional DMA data */
146 	for (i = 1; i < num_tbs; i++) {
147 		if (meta->tbs & BIT(i))
148 			dma_unmap_page(trans->dev,
149 				       le64_to_cpu(tfd->tbs[i].addr),
150 				       le16_to_cpu(tfd->tbs[i].tb_len),
151 				       DMA_TO_DEVICE);
152 		else
153 			dma_unmap_single(trans->dev,
154 					 le64_to_cpu(tfd->tbs[i].addr),
155 					 le16_to_cpu(tfd->tbs[i].tb_len),
156 					 DMA_TO_DEVICE);
157 	}
158 
159 	iwl_txq_set_tfd_invalid_gen2(trans, tfd);
160 }
161 
iwl_txq_gen2_free_tfd(struct iwl_trans * trans,struct iwl_txq * txq)162 void iwl_txq_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
163 {
164 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
165 	 * idx is bounded by n_window
166 	 */
167 	int idx = iwl_txq_get_cmd_index(txq, txq->read_ptr);
168 	struct sk_buff *skb;
169 
170 	lockdep_assert_held(&txq->lock);
171 
172 	if (!txq->entries)
173 		return;
174 
175 	iwl_txq_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
176 			       iwl_txq_get_tfd(trans, txq, idx));
177 
178 	skb = txq->entries[idx].skb;
179 
180 	/* Can be called from irqs-disabled context
181 	 * If skb is not NULL, it means that the whole queue is being
182 	 * freed and that the queue is not empty - free the skb
183 	 */
184 	if (skb) {
185 		iwl_op_mode_free_skb(trans->op_mode, skb);
186 		txq->entries[idx].skb = NULL;
187 	}
188 }
189 
get_workaround_page(struct iwl_trans * trans,struct sk_buff * skb)190 static struct page *get_workaround_page(struct iwl_trans *trans,
191 					struct sk_buff *skb)
192 {
193 	struct page **page_ptr;
194 	struct page *ret;
195 
196 	page_ptr = (void *)((u8 *)skb->cb + trans->txqs.page_offs);
197 
198 	ret = alloc_page(GFP_ATOMIC);
199 	if (!ret)
200 		return NULL;
201 
202 	/* set the chaining pointer to the previous page if there */
203 	*(void **)((u8 *)page_address(ret) + PAGE_SIZE - sizeof(void *)) = *page_ptr;
204 	*page_ptr = ret;
205 
206 	return ret;
207 }
208 
209 /*
210  * Add a TB and if needed apply the FH HW bug workaround;
211  * meta != NULL indicates that it's a page mapping and we
212  * need to dma_unmap_page() and set the meta->tbs bit in
213  * this case.
214  */
iwl_txq_gen2_set_tb_with_wa(struct iwl_trans * trans,struct sk_buff * skb,struct iwl_tfh_tfd * tfd,dma_addr_t phys,void * virt,u16 len,struct iwl_cmd_meta * meta)215 static int iwl_txq_gen2_set_tb_with_wa(struct iwl_trans *trans,
216 				       struct sk_buff *skb,
217 				       struct iwl_tfh_tfd *tfd,
218 				       dma_addr_t phys, void *virt,
219 				       u16 len, struct iwl_cmd_meta *meta)
220 {
221 	dma_addr_t oldphys = phys;
222 	struct page *page;
223 	int ret;
224 
225 	if (unlikely(dma_mapping_error(trans->dev, phys)))
226 		return -ENOMEM;
227 
228 	if (likely(!iwl_txq_crosses_4g_boundary(phys, len))) {
229 		ret = iwl_txq_gen2_set_tb(trans, tfd, phys, len);
230 
231 		if (ret < 0)
232 			goto unmap;
233 
234 		if (meta)
235 			meta->tbs |= BIT(ret);
236 
237 		ret = 0;
238 		goto trace;
239 	}
240 
241 	/*
242 	 * Work around a hardware bug. If (as expressed in the
243 	 * condition above) the TB ends on a 32-bit boundary,
244 	 * then the next TB may be accessed with the wrong
245 	 * address.
246 	 * To work around it, copy the data elsewhere and make
247 	 * a new mapping for it so the device will not fail.
248 	 */
249 
250 	if (WARN_ON(len > PAGE_SIZE - sizeof(void *))) {
251 		ret = -ENOBUFS;
252 		goto unmap;
253 	}
254 
255 	page = get_workaround_page(trans, skb);
256 	if (!page) {
257 		ret = -ENOMEM;
258 		goto unmap;
259 	}
260 
261 	memcpy(page_address(page), virt, len);
262 
263 	phys = dma_map_single(trans->dev, page_address(page), len,
264 			      DMA_TO_DEVICE);
265 	if (unlikely(dma_mapping_error(trans->dev, phys)))
266 		return -ENOMEM;
267 	ret = iwl_txq_gen2_set_tb(trans, tfd, phys, len);
268 	if (ret < 0) {
269 		/* unmap the new allocation as single */
270 		oldphys = phys;
271 		meta = NULL;
272 		goto unmap;
273 	}
274 	IWL_WARN(trans,
275 		 "TB bug workaround: copied %d bytes from 0x%llx to 0x%llx\n",
276 		 len, (unsigned long long)oldphys, (unsigned long long)phys);
277 
278 	ret = 0;
279 unmap:
280 	if (meta)
281 		dma_unmap_page(trans->dev, oldphys, len, DMA_TO_DEVICE);
282 	else
283 		dma_unmap_single(trans->dev, oldphys, len, DMA_TO_DEVICE);
284 trace:
285 	trace_iwlwifi_dev_tx_tb(trans->dev, skb, virt, phys, len);
286 
287 	return ret;
288 }
289 
290 #ifdef CONFIG_INET
get_page_hdr(struct iwl_trans * trans,size_t len,struct sk_buff * skb)291 struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len,
292 				      struct sk_buff *skb)
293 {
294 	struct iwl_tso_hdr_page *p = this_cpu_ptr(trans->txqs.tso_hdr_page);
295 	struct page **page_ptr;
296 
297 	page_ptr = (void *)((u8 *)skb->cb + trans->txqs.page_offs);
298 
299 	if (WARN_ON(*page_ptr))
300 		return NULL;
301 
302 	if (!p->page)
303 		goto alloc;
304 
305 	/*
306 	 * Check if there's enough room on this page
307 	 *
308 	 * Note that we put a page chaining pointer *last* in the
309 	 * page - we need it somewhere, and if it's there then we
310 	 * avoid DMA mapping the last bits of the page which may
311 	 * trigger the 32-bit boundary hardware bug.
312 	 *
313 	 * (see also get_workaround_page() in tx-gen2.c)
314 	 */
315 	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE -
316 			   sizeof(void *))
317 		goto out;
318 
319 	/* We don't have enough room on this page, get a new one. */
320 	__free_page(p->page);
321 
322 alloc:
323 	p->page = alloc_page(GFP_ATOMIC);
324 	if (!p->page)
325 		return NULL;
326 	p->pos = page_address(p->page);
327 	/* set the chaining pointer to NULL */
328 	*(void **)((u8 *)page_address(p->page) + PAGE_SIZE - sizeof(void *)) = NULL;
329 out:
330 	*page_ptr = p->page;
331 	get_page(p->page);
332 	return p;
333 }
334 #endif
335 
iwl_txq_gen2_build_amsdu(struct iwl_trans * trans,struct sk_buff * skb,struct iwl_tfh_tfd * tfd,int start_len,u8 hdr_len,struct iwl_device_tx_cmd * dev_cmd)336 static int iwl_txq_gen2_build_amsdu(struct iwl_trans *trans,
337 				    struct sk_buff *skb,
338 				    struct iwl_tfh_tfd *tfd, int start_len,
339 				    u8 hdr_len,
340 				    struct iwl_device_tx_cmd *dev_cmd)
341 {
342 #ifdef CONFIG_INET
343 	struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload;
344 	struct ieee80211_hdr *hdr = (void *)skb->data;
345 	unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room;
346 	unsigned int mss = skb_shinfo(skb)->gso_size;
347 	u16 length, amsdu_pad;
348 	u8 *start_hdr;
349 	struct iwl_tso_hdr_page *hdr_page;
350 	struct tso_t tso;
351 
352 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
353 			     &dev_cmd->hdr, start_len, 0);
354 
355 	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
356 	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
357 	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len;
358 	amsdu_pad = 0;
359 
360 	/* total amount of header we may need for this A-MSDU */
361 	hdr_room = DIV_ROUND_UP(total_len, mss) *
362 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr));
363 
364 	/* Our device supports 9 segments at most, it will fit in 1 page */
365 	hdr_page = get_page_hdr(trans, hdr_room, skb);
366 	if (!hdr_page)
367 		return -ENOMEM;
368 
369 	start_hdr = hdr_page->pos;
370 
371 	/*
372 	 * Pull the ieee80211 header to be able to use TSO core,
373 	 * we will restore it for the tx_status flow.
374 	 */
375 	skb_pull(skb, hdr_len);
376 
377 	/*
378 	 * Remove the length of all the headers that we don't actually
379 	 * have in the MPDU by themselves, but that we duplicate into
380 	 * all the different MSDUs inside the A-MSDU.
381 	 */
382 	le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen);
383 
384 	tso_start(skb, &tso);
385 
386 	while (total_len) {
387 		/* this is the data left for this subframe */
388 		unsigned int data_left = min_t(unsigned int, mss, total_len);
389 		unsigned int tb_len;
390 		dma_addr_t tb_phys;
391 		u8 *subf_hdrs_start = hdr_page->pos;
392 
393 		total_len -= data_left;
394 
395 		memset(hdr_page->pos, 0, amsdu_pad);
396 		hdr_page->pos += amsdu_pad;
397 		amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen +
398 				  data_left)) & 0x3;
399 		ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr));
400 		hdr_page->pos += ETH_ALEN;
401 		ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr));
402 		hdr_page->pos += ETH_ALEN;
403 
404 		length = snap_ip_tcp_hdrlen + data_left;
405 		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
406 		hdr_page->pos += sizeof(length);
407 
408 		/*
409 		 * This will copy the SNAP as well which will be considered
410 		 * as MAC header.
411 		 */
412 		tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len);
413 
414 		hdr_page->pos += snap_ip_tcp_hdrlen;
415 
416 		tb_len = hdr_page->pos - start_hdr;
417 		tb_phys = dma_map_single(trans->dev, start_hdr,
418 					 tb_len, DMA_TO_DEVICE);
419 		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
420 			goto out_err;
421 		/*
422 		 * No need for _with_wa, this is from the TSO page and
423 		 * we leave some space at the end of it so can't hit
424 		 * the buggy scenario.
425 		 */
426 		iwl_txq_gen2_set_tb(trans, tfd, tb_phys, tb_len);
427 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
428 					tb_phys, tb_len);
429 		/* add this subframe's headers' length to the tx_cmd */
430 		le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start);
431 
432 		/* prepare the start_hdr for the next subframe */
433 		start_hdr = hdr_page->pos;
434 
435 		/* put the payload */
436 		while (data_left) {
437 			int ret;
438 
439 			tb_len = min_t(unsigned int, tso.size, data_left);
440 			tb_phys = dma_map_single(trans->dev, tso.data,
441 						 tb_len, DMA_TO_DEVICE);
442 			ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd,
443 							  tb_phys, tso.data,
444 							  tb_len, NULL);
445 			if (ret)
446 				goto out_err;
447 
448 			data_left -= tb_len;
449 			tso_build_data(skb, &tso, tb_len);
450 		}
451 	}
452 
453 	/* re -add the WiFi header */
454 	skb_push(skb, hdr_len);
455 
456 	return 0;
457 
458 out_err:
459 #endif
460 	return -EINVAL;
461 }
462 
463 static struct
iwl_txq_gen2_build_tx_amsdu(struct iwl_trans * trans,struct iwl_txq * txq,struct iwl_device_tx_cmd * dev_cmd,struct sk_buff * skb,struct iwl_cmd_meta * out_meta,int hdr_len,int tx_cmd_len)464 iwl_tfh_tfd *iwl_txq_gen2_build_tx_amsdu(struct iwl_trans *trans,
465 					 struct iwl_txq *txq,
466 					 struct iwl_device_tx_cmd *dev_cmd,
467 					 struct sk_buff *skb,
468 					 struct iwl_cmd_meta *out_meta,
469 					 int hdr_len,
470 					 int tx_cmd_len)
471 {
472 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
473 	struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx);
474 	dma_addr_t tb_phys;
475 	int len;
476 	void *tb1_addr;
477 
478 	tb_phys = iwl_txq_get_first_tb_dma(txq, idx);
479 
480 	/*
481 	 * No need for _with_wa, the first TB allocation is aligned up
482 	 * to a 64-byte boundary and thus can't be at the end or cross
483 	 * a page boundary (much less a 2^32 boundary).
484 	 */
485 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
486 
487 	/*
488 	 * The second TB (tb1) points to the remainder of the TX command
489 	 * and the 802.11 header - dword aligned size
490 	 * (This calculation modifies the TX command, so do it before the
491 	 * setup of the first TB)
492 	 */
493 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
494 	      IWL_FIRST_TB_SIZE;
495 
496 	/* do not align A-MSDU to dword as the subframe header aligns it */
497 
498 	/* map the data for TB1 */
499 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
500 	tb_phys = dma_map_single(trans->dev, tb1_addr, len, DMA_TO_DEVICE);
501 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
502 		goto out_err;
503 	/*
504 	 * No need for _with_wa(), we ensure (via alignment) that the data
505 	 * here can never cross or end at a page boundary.
506 	 */
507 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, len);
508 
509 	if (iwl_txq_gen2_build_amsdu(trans, skb, tfd, len + IWL_FIRST_TB_SIZE,
510 				     hdr_len, dev_cmd))
511 		goto out_err;
512 
513 	/* building the A-MSDU might have changed this data, memcpy it now */
514 	memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE);
515 	return tfd;
516 
517 out_err:
518 	iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd);
519 	return NULL;
520 }
521 
iwl_txq_gen2_tx_add_frags(struct iwl_trans * trans,struct sk_buff * skb,struct iwl_tfh_tfd * tfd,struct iwl_cmd_meta * out_meta)522 static int iwl_txq_gen2_tx_add_frags(struct iwl_trans *trans,
523 				     struct sk_buff *skb,
524 				     struct iwl_tfh_tfd *tfd,
525 				     struct iwl_cmd_meta *out_meta)
526 {
527 	int i;
528 
529 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
530 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
531 		dma_addr_t tb_phys;
532 		unsigned int fragsz = skb_frag_size(frag);
533 		int ret;
534 
535 		if (!fragsz)
536 			continue;
537 
538 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
539 					   fragsz, DMA_TO_DEVICE);
540 		ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
541 						  skb_frag_address(frag),
542 						  fragsz, out_meta);
543 		if (ret)
544 			return ret;
545 	}
546 
547 	return 0;
548 }
549 
550 static struct
iwl_txq_gen2_build_tx(struct iwl_trans * trans,struct iwl_txq * txq,struct iwl_device_tx_cmd * dev_cmd,struct sk_buff * skb,struct iwl_cmd_meta * out_meta,int hdr_len,int tx_cmd_len,bool pad)551 iwl_tfh_tfd *iwl_txq_gen2_build_tx(struct iwl_trans *trans,
552 				   struct iwl_txq *txq,
553 				   struct iwl_device_tx_cmd *dev_cmd,
554 				   struct sk_buff *skb,
555 				   struct iwl_cmd_meta *out_meta,
556 				   int hdr_len,
557 				   int tx_cmd_len,
558 				   bool pad)
559 {
560 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
561 	struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx);
562 	dma_addr_t tb_phys;
563 	int len, tb1_len, tb2_len;
564 	void *tb1_addr;
565 	struct sk_buff *frag;
566 
567 	tb_phys = iwl_txq_get_first_tb_dma(txq, idx);
568 
569 	/* The first TB points to bi-directional DMA data */
570 	memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE);
571 
572 	/*
573 	 * No need for _with_wa, the first TB allocation is aligned up
574 	 * to a 64-byte boundary and thus can't be at the end or cross
575 	 * a page boundary (much less a 2^32 boundary).
576 	 */
577 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
578 
579 	/*
580 	 * The second TB (tb1) points to the remainder of the TX command
581 	 * and the 802.11 header - dword aligned size
582 	 * (This calculation modifies the TX command, so do it before the
583 	 * setup of the first TB)
584 	 */
585 	len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len -
586 	      IWL_FIRST_TB_SIZE;
587 
588 	if (pad)
589 		tb1_len = ALIGN(len, 4);
590 	else
591 		tb1_len = len;
592 
593 	/* map the data for TB1 */
594 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
595 	tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE);
596 	if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
597 		goto out_err;
598 	/*
599 	 * No need for _with_wa(), we ensure (via alignment) that the data
600 	 * here can never cross or end at a page boundary.
601 	 */
602 	iwl_txq_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
603 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr,
604 			     IWL_FIRST_TB_SIZE + tb1_len, hdr_len);
605 
606 	/* set up TFD's third entry to point to remainder of skb's head */
607 	tb2_len = skb_headlen(skb) - hdr_len;
608 
609 	if (tb2_len > 0) {
610 		int ret;
611 
612 		tb_phys = dma_map_single(trans->dev, skb->data + hdr_len,
613 					 tb2_len, DMA_TO_DEVICE);
614 		ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
615 						  skb->data + hdr_len, tb2_len,
616 						  NULL);
617 		if (ret)
618 			goto out_err;
619 	}
620 
621 	if (iwl_txq_gen2_tx_add_frags(trans, skb, tfd, out_meta))
622 		goto out_err;
623 
624 	skb_walk_frags(skb, frag) {
625 		int ret;
626 
627 		tb_phys = dma_map_single(trans->dev, frag->data,
628 					 skb_headlen(frag), DMA_TO_DEVICE);
629 		ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
630 						  frag->data,
631 						  skb_headlen(frag), NULL);
632 		if (ret)
633 			goto out_err;
634 		if (iwl_txq_gen2_tx_add_frags(trans, frag, tfd, out_meta))
635 			goto out_err;
636 	}
637 
638 	return tfd;
639 
640 out_err:
641 	iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd);
642 	return NULL;
643 }
644 
645 static
iwl_txq_gen2_build_tfd(struct iwl_trans * trans,struct iwl_txq * txq,struct iwl_device_tx_cmd * dev_cmd,struct sk_buff * skb,struct iwl_cmd_meta * out_meta)646 struct iwl_tfh_tfd *iwl_txq_gen2_build_tfd(struct iwl_trans *trans,
647 					   struct iwl_txq *txq,
648 					   struct iwl_device_tx_cmd *dev_cmd,
649 					   struct sk_buff *skb,
650 					   struct iwl_cmd_meta *out_meta)
651 {
652 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
653 	int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
654 	struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx);
655 	int len, hdr_len;
656 	bool amsdu;
657 
658 	/* There must be data left over for TB1 or this code must be changed */
659 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE);
660 	BUILD_BUG_ON(sizeof(struct iwl_cmd_header) +
661 		     offsetofend(struct iwl_tx_cmd_gen2, dram_info) >
662 		     IWL_FIRST_TB_SIZE);
663 	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen3) < IWL_FIRST_TB_SIZE);
664 	BUILD_BUG_ON(sizeof(struct iwl_cmd_header) +
665 		     offsetofend(struct iwl_tx_cmd_gen3, dram_info) >
666 		     IWL_FIRST_TB_SIZE);
667 
668 	memset(tfd, 0, sizeof(*tfd));
669 
670 	if (trans->trans_cfg->device_family < IWL_DEVICE_FAMILY_AX210)
671 		len = sizeof(struct iwl_tx_cmd_gen2);
672 	else
673 		len = sizeof(struct iwl_tx_cmd_gen3);
674 
675 	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
676 			(*ieee80211_get_qos_ctl(hdr) &
677 			 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
678 
679 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
680 
681 	/*
682 	 * Only build A-MSDUs here if doing so by GSO, otherwise it may be
683 	 * an A-MSDU for other reasons, e.g. NAN or an A-MSDU having been
684 	 * built in the higher layers already.
685 	 */
686 	if (amsdu && skb_shinfo(skb)->gso_size)
687 		return iwl_txq_gen2_build_tx_amsdu(trans, txq, dev_cmd, skb,
688 						    out_meta, hdr_len, len);
689 	return iwl_txq_gen2_build_tx(trans, txq, dev_cmd, skb, out_meta,
690 				      hdr_len, len, !amsdu);
691 }
692 
iwl_txq_space(struct iwl_trans * trans,const struct iwl_txq * q)693 int iwl_txq_space(struct iwl_trans *trans, const struct iwl_txq *q)
694 {
695 	unsigned int max;
696 	unsigned int used;
697 
698 	/*
699 	 * To avoid ambiguity between empty and completely full queues, there
700 	 * should always be less than max_tfd_queue_size elements in the queue.
701 	 * If q->n_window is smaller than max_tfd_queue_size, there is no need
702 	 * to reserve any queue entries for this purpose.
703 	 */
704 	if (q->n_window < trans->trans_cfg->base_params->max_tfd_queue_size)
705 		max = q->n_window;
706 	else
707 		max = trans->trans_cfg->base_params->max_tfd_queue_size - 1;
708 
709 	/*
710 	 * max_tfd_queue_size is a power of 2, so the following is equivalent to
711 	 * modulo by max_tfd_queue_size and is well defined.
712 	 */
713 	used = (q->write_ptr - q->read_ptr) &
714 		(trans->trans_cfg->base_params->max_tfd_queue_size - 1);
715 
716 	if (WARN_ON(used > max))
717 		return 0;
718 
719 	return max - used;
720 }
721 
iwl_txq_gen2_tx(struct iwl_trans * trans,struct sk_buff * skb,struct iwl_device_tx_cmd * dev_cmd,int txq_id)722 int iwl_txq_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
723 		    struct iwl_device_tx_cmd *dev_cmd, int txq_id)
724 {
725 	struct iwl_cmd_meta *out_meta;
726 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
727 	u16 cmd_len;
728 	int idx;
729 	void *tfd;
730 
731 	if (WARN_ONCE(txq_id >= IWL_MAX_TVQM_QUEUES,
732 		      "queue %d out of range", txq_id))
733 		return -EINVAL;
734 
735 	if (WARN_ONCE(!test_bit(txq_id, trans->txqs.queue_used),
736 		      "TX on unused queue %d\n", txq_id))
737 		return -EINVAL;
738 
739 	if (skb_is_nonlinear(skb) &&
740 	    skb_shinfo(skb)->nr_frags > IWL_TRANS_MAX_FRAGS(trans) &&
741 	    __skb_linearize(skb))
742 		return -ENOMEM;
743 
744 	spin_lock(&txq->lock);
745 
746 	if (iwl_txq_space(trans, txq) < txq->high_mark) {
747 		iwl_txq_stop(trans, txq);
748 
749 		/* don't put the packet on the ring, if there is no room */
750 		if (unlikely(iwl_txq_space(trans, txq) < 3)) {
751 			struct iwl_device_tx_cmd **dev_cmd_ptr;
752 
753 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
754 					       trans->txqs.dev_cmd_offs);
755 
756 			*dev_cmd_ptr = dev_cmd;
757 			__skb_queue_tail(&txq->overflow_q, skb);
758 			spin_unlock(&txq->lock);
759 			return 0;
760 		}
761 	}
762 
763 	idx = iwl_txq_get_cmd_index(txq, txq->write_ptr);
764 
765 	/* Set up driver data for this TFD */
766 	txq->entries[idx].skb = skb;
767 	txq->entries[idx].cmd = dev_cmd;
768 
769 	dev_cmd->hdr.sequence =
770 		cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) |
771 			    INDEX_TO_SEQ(idx)));
772 
773 	/* Set up first empty entry in queue's array of Tx/cmd buffers */
774 	out_meta = &txq->entries[idx].meta;
775 	out_meta->flags = 0;
776 
777 	tfd = iwl_txq_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta);
778 	if (!tfd) {
779 		spin_unlock(&txq->lock);
780 		return -1;
781 	}
782 
783 	if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) {
784 		struct iwl_tx_cmd_gen3 *tx_cmd_gen3 =
785 			(void *)dev_cmd->payload;
786 
787 		cmd_len = le16_to_cpu(tx_cmd_gen3->len);
788 	} else {
789 		struct iwl_tx_cmd_gen2 *tx_cmd_gen2 =
790 			(void *)dev_cmd->payload;
791 
792 		cmd_len = le16_to_cpu(tx_cmd_gen2->len);
793 	}
794 
795 	/* Set up entry for this TFD in Tx byte-count array */
796 	iwl_pcie_gen2_update_byte_tbl(trans, txq, cmd_len,
797 				      iwl_txq_gen2_get_num_tbs(trans, tfd));
798 
799 	/* start timer if queue currently empty */
800 	if (txq->read_ptr == txq->write_ptr && txq->wd_timeout)
801 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
802 
803 	/* Tell device the write index *just past* this latest filled TFD */
804 	txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr);
805 	iwl_txq_inc_wr_ptr(trans, txq);
806 	/*
807 	 * At this point the frame is "transmitted" successfully
808 	 * and we will get a TX status notification eventually.
809 	 */
810 	spin_unlock(&txq->lock);
811 	return 0;
812 }
813 
814 /*************** HOST COMMAND QUEUE FUNCTIONS   *****/
815 
816 /*
817  * iwl_txq_gen2_unmap -  Unmap any remaining DMA mappings and free skb's
818  */
iwl_txq_gen2_unmap(struct iwl_trans * trans,int txq_id)819 void iwl_txq_gen2_unmap(struct iwl_trans *trans, int txq_id)
820 {
821 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
822 
823 	spin_lock_bh(&txq->lock);
824 	while (txq->write_ptr != txq->read_ptr) {
825 		IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n",
826 				   txq_id, txq->read_ptr);
827 
828 		if (txq_id != trans->txqs.cmd.q_id) {
829 			int idx = iwl_txq_get_cmd_index(txq, txq->read_ptr);
830 			struct sk_buff *skb = txq->entries[idx].skb;
831 
832 			if (!WARN_ON_ONCE(!skb))
833 				iwl_txq_free_tso_page(trans, skb);
834 		}
835 		iwl_txq_gen2_free_tfd(trans, txq);
836 		txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr);
837 	}
838 
839 	while (!skb_queue_empty(&txq->overflow_q)) {
840 		struct sk_buff *skb = __skb_dequeue(&txq->overflow_q);
841 
842 		iwl_op_mode_free_skb(trans->op_mode, skb);
843 	}
844 
845 	spin_unlock_bh(&txq->lock);
846 
847 	/* just in case - this queue may have been stopped */
848 	iwl_wake_queue(trans, txq);
849 }
850 
iwl_txq_gen2_free_memory(struct iwl_trans * trans,struct iwl_txq * txq)851 static void iwl_txq_gen2_free_memory(struct iwl_trans *trans,
852 				     struct iwl_txq *txq)
853 {
854 	struct device *dev = trans->dev;
855 
856 	/* De-alloc circular buffer of TFDs */
857 	if (txq->tfds) {
858 		dma_free_coherent(dev,
859 				  trans->txqs.tfd.size * txq->n_window,
860 				  txq->tfds, txq->dma_addr);
861 		dma_free_coherent(dev,
862 				  sizeof(*txq->first_tb_bufs) * txq->n_window,
863 				  txq->first_tb_bufs, txq->first_tb_dma);
864 	}
865 
866 	kfree(txq->entries);
867 	if (txq->bc_tbl.addr)
868 		dma_pool_free(trans->txqs.bc_pool,
869 			      txq->bc_tbl.addr, txq->bc_tbl.dma);
870 	kfree(txq);
871 }
872 
873 /*
874  * iwl_pcie_txq_free - Deallocate DMA queue.
875  * @txq: Transmit queue to deallocate.
876  *
877  * Empty queue by removing and destroying all BD's.
878  * Free all buffers.
879  * 0-fill, but do not free "txq" descriptor structure.
880  */
iwl_txq_gen2_free(struct iwl_trans * trans,int txq_id)881 static void iwl_txq_gen2_free(struct iwl_trans *trans, int txq_id)
882 {
883 	struct iwl_txq *txq;
884 	int i;
885 
886 	if (WARN_ONCE(txq_id >= IWL_MAX_TVQM_QUEUES,
887 		      "queue %d out of range", txq_id))
888 		return;
889 
890 	txq = trans->txqs.txq[txq_id];
891 
892 	if (WARN_ON(!txq))
893 		return;
894 
895 	iwl_txq_gen2_unmap(trans, txq_id);
896 
897 	/* De-alloc array of command/tx buffers */
898 	if (txq_id == trans->txqs.cmd.q_id)
899 		for (i = 0; i < txq->n_window; i++) {
900 			kfree_sensitive(txq->entries[i].cmd);
901 			kfree_sensitive(txq->entries[i].free_buf);
902 		}
903 	del_timer_sync(&txq->stuck_timer);
904 
905 	iwl_txq_gen2_free_memory(trans, txq);
906 
907 	trans->txqs.txq[txq_id] = NULL;
908 
909 	clear_bit(txq_id, trans->txqs.queue_used);
910 }
911 
912 /*
913  * iwl_queue_init - Initialize queue's high/low-water and read/write indexes
914  */
iwl_queue_init(struct iwl_txq * q,int slots_num)915 static int iwl_queue_init(struct iwl_txq *q, int slots_num)
916 {
917 	q->n_window = slots_num;
918 
919 	/* slots_num must be power-of-two size, otherwise
920 	 * iwl_txq_get_cmd_index is broken. */
921 	if (WARN_ON(!is_power_of_2(slots_num)))
922 		return -EINVAL;
923 
924 	q->low_mark = q->n_window / 4;
925 	if (q->low_mark < 4)
926 		q->low_mark = 4;
927 
928 	q->high_mark = q->n_window / 8;
929 	if (q->high_mark < 2)
930 		q->high_mark = 2;
931 
932 	q->write_ptr = 0;
933 	q->read_ptr = 0;
934 
935 	return 0;
936 }
937 
iwl_txq_init(struct iwl_trans * trans,struct iwl_txq * txq,int slots_num,bool cmd_queue)938 int iwl_txq_init(struct iwl_trans *trans, struct iwl_txq *txq, int slots_num,
939 		 bool cmd_queue)
940 {
941 	int ret;
942 	u32 tfd_queue_max_size =
943 		trans->trans_cfg->base_params->max_tfd_queue_size;
944 
945 	txq->need_update = false;
946 
947 	/* max_tfd_queue_size must be power-of-two size, otherwise
948 	 * iwl_txq_inc_wrap and iwl_txq_dec_wrap are broken. */
949 	if (WARN_ONCE(tfd_queue_max_size & (tfd_queue_max_size - 1),
950 		      "Max tfd queue size must be a power of two, but is %d",
951 		      tfd_queue_max_size))
952 		return -EINVAL;
953 
954 	/* Initialize queue's high/low-water marks, and head/tail indexes */
955 	ret = iwl_queue_init(txq, slots_num);
956 	if (ret)
957 		return ret;
958 
959 	spin_lock_init(&txq->lock);
960 
961 	if (cmd_queue) {
962 		static struct lock_class_key iwl_txq_cmd_queue_lock_class;
963 
964 		lockdep_set_class(&txq->lock, &iwl_txq_cmd_queue_lock_class);
965 	}
966 
967 	__skb_queue_head_init(&txq->overflow_q);
968 
969 	return 0;
970 }
971 
iwl_txq_free_tso_page(struct iwl_trans * trans,struct sk_buff * skb)972 void iwl_txq_free_tso_page(struct iwl_trans *trans, struct sk_buff *skb)
973 {
974 	struct page **page_ptr;
975 	struct page *next;
976 
977 	page_ptr = (void *)((u8 *)skb->cb + trans->txqs.page_offs);
978 	next = *page_ptr;
979 	*page_ptr = NULL;
980 
981 	while (next) {
982 		struct page *tmp = next;
983 
984 		next = *(void **)((u8 *)page_address(next) + PAGE_SIZE -
985 				  sizeof(void *));
986 		__free_page(tmp);
987 	}
988 }
989 
iwl_txq_log_scd_error(struct iwl_trans * trans,struct iwl_txq * txq)990 void iwl_txq_log_scd_error(struct iwl_trans *trans, struct iwl_txq *txq)
991 {
992 	u32 txq_id = txq->id;
993 	u32 status;
994 	bool active;
995 	u8 fifo;
996 
997 	if (trans->trans_cfg->gen2) {
998 		IWL_ERR(trans, "Queue %d is stuck %d %d\n", txq_id,
999 			txq->read_ptr, txq->write_ptr);
1000 		/* TODO: access new SCD registers and dump them */
1001 		return;
1002 	}
1003 
1004 	status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id));
1005 	fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7;
1006 	active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE));
1007 
1008 	IWL_ERR(trans,
1009 		"Queue %d is %sactive on fifo %d and stuck for %u ms. SW [%d, %d] HW [%d, %d] FH TRB=0x0%x\n",
1010 		txq_id, active ? "" : "in", fifo,
1011 		jiffies_to_msecs(txq->wd_timeout),
1012 		txq->read_ptr, txq->write_ptr,
1013 		iwl_read_prph(trans, SCD_QUEUE_RDPTR(txq_id)) &
1014 			(trans->trans_cfg->base_params->max_tfd_queue_size - 1),
1015 			iwl_read_prph(trans, SCD_QUEUE_WRPTR(txq_id)) &
1016 			(trans->trans_cfg->base_params->max_tfd_queue_size - 1),
1017 			iwl_read_direct32(trans, FH_TX_TRB_REG(fifo)));
1018 }
1019 
iwl_txq_stuck_timer(struct timer_list * t)1020 static void iwl_txq_stuck_timer(struct timer_list *t)
1021 {
1022 	struct iwl_txq *txq = from_timer(txq, t, stuck_timer);
1023 	struct iwl_trans *trans = txq->trans;
1024 
1025 	spin_lock(&txq->lock);
1026 	/* check if triggered erroneously */
1027 	if (txq->read_ptr == txq->write_ptr) {
1028 		spin_unlock(&txq->lock);
1029 		return;
1030 	}
1031 	spin_unlock(&txq->lock);
1032 
1033 	iwl_txq_log_scd_error(trans, txq);
1034 
1035 	iwl_force_nmi(trans);
1036 }
1037 
iwl_txq_set_tfd_invalid_gen1(struct iwl_trans * trans,struct iwl_tfd * tfd)1038 static void iwl_txq_set_tfd_invalid_gen1(struct iwl_trans *trans,
1039 					 struct iwl_tfd *tfd)
1040 {
1041 	tfd->num_tbs = 0;
1042 
1043 	iwl_pcie_gen1_tfd_set_tb(trans, tfd, 0, trans->invalid_tx_cmd.dma,
1044 				 trans->invalid_tx_cmd.size);
1045 }
1046 
iwl_txq_alloc(struct iwl_trans * trans,struct iwl_txq * txq,int slots_num,bool cmd_queue)1047 int iwl_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq, int slots_num,
1048 		  bool cmd_queue)
1049 {
1050 	size_t num_entries = trans->trans_cfg->gen2 ?
1051 		slots_num : trans->trans_cfg->base_params->max_tfd_queue_size;
1052 	size_t tfd_sz;
1053 	size_t tb0_buf_sz;
1054 	int i;
1055 
1056 	if (WARN_ONCE(slots_num <= 0, "Invalid slots num:%d\n", slots_num))
1057 		return -EINVAL;
1058 
1059 	if (WARN_ON(txq->entries || txq->tfds))
1060 		return -EINVAL;
1061 
1062 	tfd_sz = trans->txqs.tfd.size * num_entries;
1063 
1064 	timer_setup(&txq->stuck_timer, iwl_txq_stuck_timer, 0);
1065 	txq->trans = trans;
1066 
1067 	txq->n_window = slots_num;
1068 
1069 	txq->entries = kcalloc(slots_num,
1070 			       sizeof(struct iwl_pcie_txq_entry),
1071 			       GFP_KERNEL);
1072 
1073 	if (!txq->entries)
1074 		goto error;
1075 
1076 	if (cmd_queue)
1077 		for (i = 0; i < slots_num; i++) {
1078 			txq->entries[i].cmd =
1079 				kmalloc(sizeof(struct iwl_device_cmd),
1080 					GFP_KERNEL);
1081 			if (!txq->entries[i].cmd)
1082 				goto error;
1083 		}
1084 
1085 	/* Circular buffer of transmit frame descriptors (TFDs),
1086 	 * shared with device */
1087 	txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz,
1088 				       &txq->dma_addr, GFP_KERNEL);
1089 	if (!txq->tfds)
1090 		goto error;
1091 
1092 	BUILD_BUG_ON(sizeof(*txq->first_tb_bufs) != IWL_FIRST_TB_SIZE_ALIGN);
1093 
1094 	tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num;
1095 
1096 	txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz,
1097 						&txq->first_tb_dma,
1098 						GFP_KERNEL);
1099 	if (!txq->first_tb_bufs)
1100 		goto err_free_tfds;
1101 
1102 	for (i = 0; i < num_entries; i++) {
1103 		void *tfd = iwl_txq_get_tfd(trans, txq, i);
1104 
1105 		if (trans->trans_cfg->gen2)
1106 			iwl_txq_set_tfd_invalid_gen2(trans, tfd);
1107 		else
1108 			iwl_txq_set_tfd_invalid_gen1(trans, tfd);
1109 	}
1110 
1111 	return 0;
1112 err_free_tfds:
1113 	dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr);
1114 	txq->tfds = NULL;
1115 error:
1116 	if (txq->entries && cmd_queue)
1117 		for (i = 0; i < slots_num; i++)
1118 			kfree(txq->entries[i].cmd);
1119 	kfree(txq->entries);
1120 	txq->entries = NULL;
1121 
1122 	return -ENOMEM;
1123 }
1124 
1125 static struct iwl_txq *
iwl_txq_dyn_alloc_dma(struct iwl_trans * trans,int size,unsigned int timeout)1126 iwl_txq_dyn_alloc_dma(struct iwl_trans *trans, int size, unsigned int timeout)
1127 {
1128 	size_t bc_tbl_size, bc_tbl_entries;
1129 	struct iwl_txq *txq;
1130 	int ret;
1131 
1132 	WARN_ON(!trans->txqs.bc_tbl_size);
1133 
1134 	bc_tbl_size = trans->txqs.bc_tbl_size;
1135 	bc_tbl_entries = bc_tbl_size / sizeof(u16);
1136 
1137 	if (WARN_ON(size > bc_tbl_entries))
1138 		return ERR_PTR(-EINVAL);
1139 
1140 	txq = kzalloc(sizeof(*txq), GFP_KERNEL);
1141 	if (!txq)
1142 		return ERR_PTR(-ENOMEM);
1143 
1144 	txq->bc_tbl.addr = dma_pool_alloc(trans->txqs.bc_pool, GFP_KERNEL,
1145 					  &txq->bc_tbl.dma);
1146 	if (!txq->bc_tbl.addr) {
1147 		IWL_ERR(trans, "Scheduler BC Table allocation failed\n");
1148 		kfree(txq);
1149 		return ERR_PTR(-ENOMEM);
1150 	}
1151 
1152 	ret = iwl_txq_alloc(trans, txq, size, false);
1153 	if (ret) {
1154 		IWL_ERR(trans, "Tx queue alloc failed\n");
1155 		goto error;
1156 	}
1157 	ret = iwl_txq_init(trans, txq, size, false);
1158 	if (ret) {
1159 		IWL_ERR(trans, "Tx queue init failed\n");
1160 		goto error;
1161 	}
1162 
1163 	txq->wd_timeout = msecs_to_jiffies(timeout);
1164 
1165 	return txq;
1166 
1167 error:
1168 	iwl_txq_gen2_free_memory(trans, txq);
1169 	return ERR_PTR(ret);
1170 }
1171 
iwl_txq_alloc_response(struct iwl_trans * trans,struct iwl_txq * txq,struct iwl_host_cmd * hcmd)1172 static int iwl_txq_alloc_response(struct iwl_trans *trans, struct iwl_txq *txq,
1173 				  struct iwl_host_cmd *hcmd)
1174 {
1175 	struct iwl_tx_queue_cfg_rsp *rsp;
1176 	int ret, qid;
1177 	u32 wr_ptr;
1178 
1179 	if (WARN_ON(iwl_rx_packet_payload_len(hcmd->resp_pkt) !=
1180 		    sizeof(*rsp))) {
1181 		ret = -EINVAL;
1182 		goto error_free_resp;
1183 	}
1184 
1185 	rsp = (void *)hcmd->resp_pkt->data;
1186 	qid = le16_to_cpu(rsp->queue_number);
1187 	wr_ptr = le16_to_cpu(rsp->write_pointer);
1188 
1189 	if (qid >= ARRAY_SIZE(trans->txqs.txq)) {
1190 		WARN_ONCE(1, "queue index %d unsupported", qid);
1191 		ret = -EIO;
1192 		goto error_free_resp;
1193 	}
1194 
1195 	if (test_and_set_bit(qid, trans->txqs.queue_used)) {
1196 		WARN_ONCE(1, "queue %d already used", qid);
1197 		ret = -EIO;
1198 		goto error_free_resp;
1199 	}
1200 
1201 	if (WARN_ONCE(trans->txqs.txq[qid],
1202 		      "queue %d already allocated\n", qid)) {
1203 		ret = -EIO;
1204 		goto error_free_resp;
1205 	}
1206 
1207 	txq->id = qid;
1208 	trans->txqs.txq[qid] = txq;
1209 	wr_ptr &= (trans->trans_cfg->base_params->max_tfd_queue_size - 1);
1210 
1211 	/* Place first TFD at index corresponding to start sequence number */
1212 	txq->read_ptr = wr_ptr;
1213 	txq->write_ptr = wr_ptr;
1214 
1215 	IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid);
1216 
1217 	iwl_free_resp(hcmd);
1218 	return qid;
1219 
1220 error_free_resp:
1221 	iwl_free_resp(hcmd);
1222 	iwl_txq_gen2_free_memory(trans, txq);
1223 	return ret;
1224 }
1225 
iwl_txq_dyn_alloc(struct iwl_trans * trans,u32 flags,u32 sta_mask,u8 tid,int size,unsigned int timeout)1226 int iwl_txq_dyn_alloc(struct iwl_trans *trans, u32 flags, u32 sta_mask,
1227 		      u8 tid, int size, unsigned int timeout)
1228 {
1229 	struct iwl_txq *txq;
1230 	union {
1231 		struct iwl_tx_queue_cfg_cmd old;
1232 		struct iwl_scd_queue_cfg_cmd new;
1233 	} cmd;
1234 	struct iwl_host_cmd hcmd = {
1235 		.flags = CMD_WANT_SKB,
1236 	};
1237 	int ret;
1238 
1239 	if (trans->trans_cfg->device_family == IWL_DEVICE_FAMILY_BZ &&
1240 	    trans->hw_rev_step == SILICON_A_STEP)
1241 		size = 4096;
1242 
1243 	txq = iwl_txq_dyn_alloc_dma(trans, size, timeout);
1244 	if (IS_ERR(txq))
1245 		return PTR_ERR(txq);
1246 
1247 	if (trans->txqs.queue_alloc_cmd_ver == 0) {
1248 		memset(&cmd.old, 0, sizeof(cmd.old));
1249 		cmd.old.tfdq_addr = cpu_to_le64(txq->dma_addr);
1250 		cmd.old.byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma);
1251 		cmd.old.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size));
1252 		cmd.old.flags = cpu_to_le16(flags | TX_QUEUE_CFG_ENABLE_QUEUE);
1253 		cmd.old.tid = tid;
1254 
1255 		if (hweight32(sta_mask) != 1) {
1256 			ret = -EINVAL;
1257 			goto error;
1258 		}
1259 		cmd.old.sta_id = ffs(sta_mask) - 1;
1260 
1261 		hcmd.id = SCD_QUEUE_CFG;
1262 		hcmd.len[0] = sizeof(cmd.old);
1263 		hcmd.data[0] = &cmd.old;
1264 	} else if (trans->txqs.queue_alloc_cmd_ver == 3) {
1265 		memset(&cmd.new, 0, sizeof(cmd.new));
1266 		cmd.new.operation = cpu_to_le32(IWL_SCD_QUEUE_ADD);
1267 		cmd.new.u.add.tfdq_dram_addr = cpu_to_le64(txq->dma_addr);
1268 		cmd.new.u.add.bc_dram_addr = cpu_to_le64(txq->bc_tbl.dma);
1269 		cmd.new.u.add.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size));
1270 		cmd.new.u.add.flags = cpu_to_le32(flags);
1271 		cmd.new.u.add.sta_mask = cpu_to_le32(sta_mask);
1272 		cmd.new.u.add.tid = tid;
1273 
1274 		hcmd.id = WIDE_ID(DATA_PATH_GROUP, SCD_QUEUE_CONFIG_CMD);
1275 		hcmd.len[0] = sizeof(cmd.new);
1276 		hcmd.data[0] = &cmd.new;
1277 	} else {
1278 		ret = -EOPNOTSUPP;
1279 		goto error;
1280 	}
1281 
1282 	ret = iwl_trans_send_cmd(trans, &hcmd);
1283 	if (ret)
1284 		goto error;
1285 
1286 	return iwl_txq_alloc_response(trans, txq, &hcmd);
1287 
1288 error:
1289 	iwl_txq_gen2_free_memory(trans, txq);
1290 	return ret;
1291 }
1292 
iwl_txq_dyn_free(struct iwl_trans * trans,int queue)1293 void iwl_txq_dyn_free(struct iwl_trans *trans, int queue)
1294 {
1295 	if (WARN(queue >= IWL_MAX_TVQM_QUEUES,
1296 		 "queue %d out of range", queue))
1297 		return;
1298 
1299 	/*
1300 	 * Upon HW Rfkill - we stop the device, and then stop the queues
1301 	 * in the op_mode. Just for the sake of the simplicity of the op_mode,
1302 	 * allow the op_mode to call txq_disable after it already called
1303 	 * stop_device.
1304 	 */
1305 	if (!test_and_clear_bit(queue, trans->txqs.queue_used)) {
1306 		WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status),
1307 			  "queue %d not used", queue);
1308 		return;
1309 	}
1310 
1311 	iwl_txq_gen2_free(trans, queue);
1312 
1313 	IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue);
1314 }
1315 
iwl_txq_gen2_tx_free(struct iwl_trans * trans)1316 void iwl_txq_gen2_tx_free(struct iwl_trans *trans)
1317 {
1318 	int i;
1319 
1320 	memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used));
1321 
1322 	/* Free all TX queues */
1323 	for (i = 0; i < ARRAY_SIZE(trans->txqs.txq); i++) {
1324 		if (!trans->txqs.txq[i])
1325 			continue;
1326 
1327 		iwl_txq_gen2_free(trans, i);
1328 	}
1329 }
1330 
iwl_txq_gen2_init(struct iwl_trans * trans,int txq_id,int queue_size)1331 int iwl_txq_gen2_init(struct iwl_trans *trans, int txq_id, int queue_size)
1332 {
1333 	struct iwl_txq *queue;
1334 	int ret;
1335 
1336 	/* alloc and init the tx queue */
1337 	if (!trans->txqs.txq[txq_id]) {
1338 		queue = kzalloc(sizeof(*queue), GFP_KERNEL);
1339 		if (!queue) {
1340 			IWL_ERR(trans, "Not enough memory for tx queue\n");
1341 			return -ENOMEM;
1342 		}
1343 		trans->txqs.txq[txq_id] = queue;
1344 		ret = iwl_txq_alloc(trans, queue, queue_size, true);
1345 		if (ret) {
1346 			IWL_ERR(trans, "Tx %d queue init failed\n", txq_id);
1347 			goto error;
1348 		}
1349 	} else {
1350 		queue = trans->txqs.txq[txq_id];
1351 	}
1352 
1353 	ret = iwl_txq_init(trans, queue, queue_size,
1354 			   (txq_id == trans->txqs.cmd.q_id));
1355 	if (ret) {
1356 		IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id);
1357 		goto error;
1358 	}
1359 	trans->txqs.txq[txq_id]->id = txq_id;
1360 	set_bit(txq_id, trans->txqs.queue_used);
1361 
1362 	return 0;
1363 
1364 error:
1365 	iwl_txq_gen2_tx_free(trans);
1366 	return ret;
1367 }
1368 
iwl_txq_gen1_tfd_tb_get_addr(struct iwl_trans * trans,struct iwl_tfd * tfd,u8 idx)1369 static inline dma_addr_t iwl_txq_gen1_tfd_tb_get_addr(struct iwl_trans *trans,
1370 						      struct iwl_tfd *tfd, u8 idx)
1371 {
1372 	struct iwl_tfd_tb *tb = &tfd->tbs[idx];
1373 	dma_addr_t addr;
1374 	dma_addr_t hi_len;
1375 
1376 	addr = get_unaligned_le32(&tb->lo);
1377 
1378 	if (sizeof(dma_addr_t) <= sizeof(u32))
1379 		return addr;
1380 
1381 	hi_len = le16_to_cpu(tb->hi_n_len) & 0xF;
1382 
1383 	/*
1384 	 * shift by 16 twice to avoid warnings on 32-bit
1385 	 * (where this code never runs anyway due to the
1386 	 * if statement above)
1387 	 */
1388 	return addr | ((hi_len << 16) << 16);
1389 }
1390 
iwl_txq_gen1_tfd_unmap(struct iwl_trans * trans,struct iwl_cmd_meta * meta,struct iwl_txq * txq,int index)1391 void iwl_txq_gen1_tfd_unmap(struct iwl_trans *trans,
1392 			    struct iwl_cmd_meta *meta,
1393 			    struct iwl_txq *txq, int index)
1394 {
1395 	int i, num_tbs;
1396 	struct iwl_tfd *tfd = iwl_txq_get_tfd(trans, txq, index);
1397 
1398 	/* Sanity check on number of chunks */
1399 	num_tbs = iwl_txq_gen1_tfd_get_num_tbs(trans, tfd);
1400 
1401 	if (num_tbs > trans->txqs.tfd.max_tbs) {
1402 		IWL_ERR(trans, "Too many chunks: %i\n", num_tbs);
1403 		/* @todo issue fatal error, it is quite serious situation */
1404 		return;
1405 	}
1406 
1407 	/* first TB is never freed - it's the bidirectional DMA data */
1408 
1409 	for (i = 1; i < num_tbs; i++) {
1410 		if (meta->tbs & BIT(i))
1411 			dma_unmap_page(trans->dev,
1412 				       iwl_txq_gen1_tfd_tb_get_addr(trans,
1413 								    tfd, i),
1414 				       iwl_txq_gen1_tfd_tb_get_len(trans,
1415 								   tfd, i),
1416 				       DMA_TO_DEVICE);
1417 		else
1418 			dma_unmap_single(trans->dev,
1419 					 iwl_txq_gen1_tfd_tb_get_addr(trans,
1420 								      tfd, i),
1421 					 iwl_txq_gen1_tfd_tb_get_len(trans,
1422 								     tfd, i),
1423 					 DMA_TO_DEVICE);
1424 	}
1425 
1426 	meta->tbs = 0;
1427 
1428 	iwl_txq_set_tfd_invalid_gen1(trans, tfd);
1429 }
1430 
1431 #define IWL_TX_CRC_SIZE 4
1432 #define IWL_TX_DELIMITER_SIZE 4
1433 
1434 /*
1435  * iwl_txq_gen1_update_byte_cnt_tbl - Set up entry in Tx byte-count array
1436  */
iwl_txq_gen1_update_byte_cnt_tbl(struct iwl_trans * trans,struct iwl_txq * txq,u16 byte_cnt,int num_tbs)1437 void iwl_txq_gen1_update_byte_cnt_tbl(struct iwl_trans *trans,
1438 				      struct iwl_txq *txq, u16 byte_cnt,
1439 				      int num_tbs)
1440 {
1441 	struct iwlagn_scd_bc_tbl *scd_bc_tbl;
1442 	int write_ptr = txq->write_ptr;
1443 	int txq_id = txq->id;
1444 	u8 sec_ctl = 0;
1445 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
1446 	__le16 bc_ent;
1447 	struct iwl_device_tx_cmd *dev_cmd = txq->entries[txq->write_ptr].cmd;
1448 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
1449 	u8 sta_id = tx_cmd->sta_id;
1450 
1451 	scd_bc_tbl = trans->txqs.scd_bc_tbls.addr;
1452 
1453 	sec_ctl = tx_cmd->sec_ctl;
1454 
1455 	switch (sec_ctl & TX_CMD_SEC_MSK) {
1456 	case TX_CMD_SEC_CCM:
1457 		len += IEEE80211_CCMP_MIC_LEN;
1458 		break;
1459 	case TX_CMD_SEC_TKIP:
1460 		len += IEEE80211_TKIP_ICV_LEN;
1461 		break;
1462 	case TX_CMD_SEC_WEP:
1463 		len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN;
1464 		break;
1465 	}
1466 	if (trans->txqs.bc_table_dword)
1467 		len = DIV_ROUND_UP(len, 4);
1468 
1469 	if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX))
1470 		return;
1471 
1472 	bc_ent = cpu_to_le16(len | (sta_id << 12));
1473 
1474 	scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent;
1475 
1476 	if (write_ptr < TFD_QUEUE_SIZE_BC_DUP)
1477 		scd_bc_tbl[txq_id].tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] =
1478 			bc_ent;
1479 }
1480 
iwl_txq_gen1_inval_byte_cnt_tbl(struct iwl_trans * trans,struct iwl_txq * txq)1481 void iwl_txq_gen1_inval_byte_cnt_tbl(struct iwl_trans *trans,
1482 				     struct iwl_txq *txq)
1483 {
1484 	struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans->txqs.scd_bc_tbls.addr;
1485 	int txq_id = txq->id;
1486 	int read_ptr = txq->read_ptr;
1487 	u8 sta_id = 0;
1488 	__le16 bc_ent;
1489 	struct iwl_device_tx_cmd *dev_cmd = txq->entries[read_ptr].cmd;
1490 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
1491 
1492 	WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX);
1493 
1494 	if (txq_id != trans->txqs.cmd.q_id)
1495 		sta_id = tx_cmd->sta_id;
1496 
1497 	bc_ent = cpu_to_le16(1 | (sta_id << 12));
1498 
1499 	scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent;
1500 
1501 	if (read_ptr < TFD_QUEUE_SIZE_BC_DUP)
1502 		scd_bc_tbl[txq_id].tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] =
1503 			bc_ent;
1504 }
1505 
1506 /*
1507  * iwl_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr]
1508  * @trans - transport private data
1509  * @txq - tx queue
1510  * @dma_dir - the direction of the DMA mapping
1511  *
1512  * Does NOT advance any TFD circular buffer read/write indexes
1513  * Does NOT free the TFD itself (which is within circular buffer)
1514  */
iwl_txq_free_tfd(struct iwl_trans * trans,struct iwl_txq * txq)1515 void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq)
1516 {
1517 	/* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and
1518 	 * idx is bounded by n_window
1519 	 */
1520 	int rd_ptr = txq->read_ptr;
1521 	int idx = iwl_txq_get_cmd_index(txq, rd_ptr);
1522 	struct sk_buff *skb;
1523 
1524 	lockdep_assert_held(&txq->lock);
1525 
1526 	if (!txq->entries)
1527 		return;
1528 
1529 	/* We have only q->n_window txq->entries, but we use
1530 	 * TFD_QUEUE_SIZE_MAX tfds
1531 	 */
1532 	if (trans->trans_cfg->gen2)
1533 		iwl_txq_gen2_tfd_unmap(trans, &txq->entries[idx].meta,
1534 				       iwl_txq_get_tfd(trans, txq, rd_ptr));
1535 	else
1536 		iwl_txq_gen1_tfd_unmap(trans, &txq->entries[idx].meta,
1537 				       txq, rd_ptr);
1538 
1539 	/* free SKB */
1540 	skb = txq->entries[idx].skb;
1541 
1542 	/* Can be called from irqs-disabled context
1543 	 * If skb is not NULL, it means that the whole queue is being
1544 	 * freed and that the queue is not empty - free the skb
1545 	 */
1546 	if (skb) {
1547 		iwl_op_mode_free_skb(trans->op_mode, skb);
1548 		txq->entries[idx].skb = NULL;
1549 	}
1550 }
1551 
iwl_txq_progress(struct iwl_txq * txq)1552 void iwl_txq_progress(struct iwl_txq *txq)
1553 {
1554 	lockdep_assert_held(&txq->lock);
1555 
1556 	if (!txq->wd_timeout)
1557 		return;
1558 
1559 	/*
1560 	 * station is asleep and we send data - that must
1561 	 * be uAPSD or PS-Poll. Don't rearm the timer.
1562 	 */
1563 	if (txq->frozen)
1564 		return;
1565 
1566 	/*
1567 	 * if empty delete timer, otherwise move timer forward
1568 	 * since we're making progress on this queue
1569 	 */
1570 	if (txq->read_ptr == txq->write_ptr)
1571 		del_timer(&txq->stuck_timer);
1572 	else
1573 		mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout);
1574 }
1575 
1576 /* Frees buffers until index _not_ inclusive */
iwl_txq_reclaim(struct iwl_trans * trans,int txq_id,int ssn,struct sk_buff_head * skbs,bool is_flush)1577 void iwl_txq_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
1578 		     struct sk_buff_head *skbs, bool is_flush)
1579 {
1580 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
1581 	int tfd_num, read_ptr, last_to_free;
1582 
1583 	/* This function is not meant to release cmd queue*/
1584 	if (WARN_ON(txq_id == trans->txqs.cmd.q_id))
1585 		return;
1586 
1587 	if (WARN_ON(!txq))
1588 		return;
1589 
1590 	tfd_num = iwl_txq_get_cmd_index(txq, ssn);
1591 
1592 	spin_lock_bh(&txq->lock);
1593 	read_ptr = iwl_txq_get_cmd_index(txq, txq->read_ptr);
1594 
1595 	if (!test_bit(txq_id, trans->txqs.queue_used)) {
1596 		IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n",
1597 				    txq_id, ssn);
1598 		goto out;
1599 	}
1600 
1601 	if (read_ptr == tfd_num)
1602 		goto out;
1603 
1604 	IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d -> %d (%d)\n",
1605 			   txq_id, txq->read_ptr, tfd_num, ssn);
1606 
1607 	/*Since we free until index _not_ inclusive, the one before index is
1608 	 * the last we will free. This one must be used */
1609 	last_to_free = iwl_txq_dec_wrap(trans, tfd_num);
1610 
1611 	if (!iwl_txq_used(txq, last_to_free)) {
1612 		IWL_ERR(trans,
1613 			"%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n",
1614 			__func__, txq_id, last_to_free,
1615 			trans->trans_cfg->base_params->max_tfd_queue_size,
1616 			txq->write_ptr, txq->read_ptr);
1617 
1618 		iwl_op_mode_time_point(trans->op_mode,
1619 				       IWL_FW_INI_TIME_POINT_FAKE_TX,
1620 				       NULL);
1621 		goto out;
1622 	}
1623 
1624 	if (WARN_ON(!skb_queue_empty(skbs)))
1625 		goto out;
1626 
1627 	for (;
1628 	     read_ptr != tfd_num;
1629 	     txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr),
1630 	     read_ptr = iwl_txq_get_cmd_index(txq, txq->read_ptr)) {
1631 		struct sk_buff *skb = txq->entries[read_ptr].skb;
1632 
1633 		if (WARN_ON_ONCE(!skb))
1634 			continue;
1635 
1636 		iwl_txq_free_tso_page(trans, skb);
1637 
1638 		__skb_queue_tail(skbs, skb);
1639 
1640 		txq->entries[read_ptr].skb = NULL;
1641 
1642 		if (!trans->trans_cfg->gen2)
1643 			iwl_txq_gen1_inval_byte_cnt_tbl(trans, txq);
1644 
1645 		iwl_txq_free_tfd(trans, txq);
1646 	}
1647 
1648 	iwl_txq_progress(txq);
1649 
1650 	if (iwl_txq_space(trans, txq) > txq->low_mark &&
1651 	    test_bit(txq_id, trans->txqs.queue_stopped)) {
1652 		struct sk_buff_head overflow_skbs;
1653 		struct sk_buff *skb;
1654 
1655 		__skb_queue_head_init(&overflow_skbs);
1656 		skb_queue_splice_init(&txq->overflow_q,
1657 				      is_flush ? skbs : &overflow_skbs);
1658 
1659 		/*
1660 		 * We are going to transmit from the overflow queue.
1661 		 * Remember this state so that wait_for_txq_empty will know we
1662 		 * are adding more packets to the TFD queue. It cannot rely on
1663 		 * the state of &txq->overflow_q, as we just emptied it, but
1664 		 * haven't TXed the content yet.
1665 		 */
1666 		txq->overflow_tx = true;
1667 
1668 		/*
1669 		 * This is tricky: we are in reclaim path which is non
1670 		 * re-entrant, so noone will try to take the access the
1671 		 * txq data from that path. We stopped tx, so we can't
1672 		 * have tx as well. Bottom line, we can unlock and re-lock
1673 		 * later.
1674 		 */
1675 		spin_unlock_bh(&txq->lock);
1676 
1677 		while ((skb = __skb_dequeue(&overflow_skbs))) {
1678 			struct iwl_device_tx_cmd *dev_cmd_ptr;
1679 
1680 			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
1681 						 trans->txqs.dev_cmd_offs);
1682 
1683 			/*
1684 			 * Note that we can very well be overflowing again.
1685 			 * In that case, iwl_txq_space will be small again
1686 			 * and we won't wake mac80211's queue.
1687 			 */
1688 			iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id);
1689 		}
1690 
1691 		if (iwl_txq_space(trans, txq) > txq->low_mark)
1692 			iwl_wake_queue(trans, txq);
1693 
1694 		spin_lock_bh(&txq->lock);
1695 		txq->overflow_tx = false;
1696 	}
1697 
1698 out:
1699 	spin_unlock_bh(&txq->lock);
1700 }
1701 
1702 /* Set wr_ptr of specific device and txq  */
iwl_txq_set_q_ptrs(struct iwl_trans * trans,int txq_id,int ptr)1703 void iwl_txq_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr)
1704 {
1705 	struct iwl_txq *txq = trans->txqs.txq[txq_id];
1706 
1707 	spin_lock_bh(&txq->lock);
1708 
1709 	txq->write_ptr = ptr;
1710 	txq->read_ptr = txq->write_ptr;
1711 
1712 	spin_unlock_bh(&txq->lock);
1713 }
1714 
iwl_trans_txq_freeze_timer(struct iwl_trans * trans,unsigned long txqs,bool freeze)1715 void iwl_trans_txq_freeze_timer(struct iwl_trans *trans, unsigned long txqs,
1716 				bool freeze)
1717 {
1718 	int queue;
1719 
1720 	for_each_set_bit(queue, &txqs, BITS_PER_LONG) {
1721 		struct iwl_txq *txq = trans->txqs.txq[queue];
1722 		unsigned long now;
1723 
1724 		spin_lock_bh(&txq->lock);
1725 
1726 		now = jiffies;
1727 
1728 		if (txq->frozen == freeze)
1729 			goto next_queue;
1730 
1731 		IWL_DEBUG_TX_QUEUES(trans, "%s TXQ %d\n",
1732 				    freeze ? "Freezing" : "Waking", queue);
1733 
1734 		txq->frozen = freeze;
1735 
1736 		if (txq->read_ptr == txq->write_ptr)
1737 			goto next_queue;
1738 
1739 		if (freeze) {
1740 			if (unlikely(time_after(now,
1741 						txq->stuck_timer.expires))) {
1742 				/*
1743 				 * The timer should have fired, maybe it is
1744 				 * spinning right now on the lock.
1745 				 */
1746 				goto next_queue;
1747 			}
1748 			/* remember how long until the timer fires */
1749 			txq->frozen_expiry_remainder =
1750 				txq->stuck_timer.expires - now;
1751 			del_timer(&txq->stuck_timer);
1752 			goto next_queue;
1753 		}
1754 
1755 		/*
1756 		 * Wake a non-empty queue -> arm timer with the
1757 		 * remainder before it froze
1758 		 */
1759 		mod_timer(&txq->stuck_timer,
1760 			  now + txq->frozen_expiry_remainder);
1761 
1762 next_queue:
1763 		spin_unlock_bh(&txq->lock);
1764 	}
1765 }
1766 
1767 #define HOST_COMPLETE_TIMEOUT	(2 * HZ)
1768 
iwl_trans_txq_send_hcmd_sync(struct iwl_trans * trans,struct iwl_host_cmd * cmd)1769 static int iwl_trans_txq_send_hcmd_sync(struct iwl_trans *trans,
1770 					struct iwl_host_cmd *cmd)
1771 {
1772 	const char *cmd_str = iwl_get_cmd_string(trans, cmd->id);
1773 	struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id];
1774 	int cmd_idx;
1775 	int ret;
1776 
1777 	IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str);
1778 
1779 	if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE,
1780 				  &trans->status),
1781 		 "Command %s: a command is already active!\n", cmd_str))
1782 		return -EIO;
1783 
1784 	IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str);
1785 
1786 	cmd_idx = trans->ops->send_cmd(trans, cmd);
1787 	if (cmd_idx < 0) {
1788 		ret = cmd_idx;
1789 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1790 		IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n",
1791 			cmd_str, ret);
1792 		return ret;
1793 	}
1794 
1795 	ret = wait_event_timeout(trans->wait_command_queue,
1796 				 !test_bit(STATUS_SYNC_HCMD_ACTIVE,
1797 					   &trans->status),
1798 				 HOST_COMPLETE_TIMEOUT);
1799 	if (!ret) {
1800 		IWL_ERR(trans, "Error sending %s: time out after %dms.\n",
1801 			cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT));
1802 
1803 		IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n",
1804 			txq->read_ptr, txq->write_ptr);
1805 
1806 		clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status);
1807 		IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n",
1808 			       cmd_str);
1809 		ret = -ETIMEDOUT;
1810 
1811 		iwl_trans_sync_nmi(trans);
1812 		goto cancel;
1813 	}
1814 
1815 	if (test_bit(STATUS_FW_ERROR, &trans->status)) {
1816 		if (!test_and_clear_bit(STATUS_SUPPRESS_CMD_ERROR_ONCE,
1817 					&trans->status)) {
1818 			IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str);
1819 			dump_stack();
1820 		}
1821 		ret = -EIO;
1822 		goto cancel;
1823 	}
1824 
1825 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1826 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1827 		IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n");
1828 		ret = -ERFKILL;
1829 		goto cancel;
1830 	}
1831 
1832 	if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) {
1833 		IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str);
1834 		ret = -EIO;
1835 		goto cancel;
1836 	}
1837 
1838 	return 0;
1839 
1840 cancel:
1841 	if (cmd->flags & CMD_WANT_SKB) {
1842 		/*
1843 		 * Cancel the CMD_WANT_SKB flag for the cmd in the
1844 		 * TX cmd queue. Otherwise in case the cmd comes
1845 		 * in later, it will possibly set an invalid
1846 		 * address (cmd->meta.source).
1847 		 */
1848 		txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB;
1849 	}
1850 
1851 	if (cmd->resp_pkt) {
1852 		iwl_free_resp(cmd);
1853 		cmd->resp_pkt = NULL;
1854 	}
1855 
1856 	return ret;
1857 }
1858 
iwl_trans_txq_send_hcmd(struct iwl_trans * trans,struct iwl_host_cmd * cmd)1859 int iwl_trans_txq_send_hcmd(struct iwl_trans *trans,
1860 			    struct iwl_host_cmd *cmd)
1861 {
1862 	/* Make sure the NIC is still alive in the bus */
1863 	if (test_bit(STATUS_TRANS_DEAD, &trans->status))
1864 		return -ENODEV;
1865 
1866 	if (!(cmd->flags & CMD_SEND_IN_RFKILL) &&
1867 	    test_bit(STATUS_RFKILL_OPMODE, &trans->status)) {
1868 		IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n",
1869 				  cmd->id);
1870 		return -ERFKILL;
1871 	}
1872 
1873 	if (unlikely(trans->system_pm_mode == IWL_PLAT_PM_MODE_D3 &&
1874 		     !(cmd->flags & CMD_SEND_IN_D3))) {
1875 		IWL_DEBUG_WOWLAN(trans, "Dropping CMD 0x%x: D3\n", cmd->id);
1876 		return -EHOSTDOWN;
1877 	}
1878 
1879 	if (cmd->flags & CMD_ASYNC) {
1880 		int ret;
1881 
1882 		/* An asynchronous command can not expect an SKB to be set. */
1883 		if (WARN_ON(cmd->flags & CMD_WANT_SKB))
1884 			return -EINVAL;
1885 
1886 		ret = trans->ops->send_cmd(trans, cmd);
1887 		if (ret < 0) {
1888 			IWL_ERR(trans,
1889 				"Error sending %s: enqueue_hcmd failed: %d\n",
1890 				iwl_get_cmd_string(trans, cmd->id), ret);
1891 			return ret;
1892 		}
1893 		return 0;
1894 	}
1895 
1896 	return iwl_trans_txq_send_hcmd_sync(trans, cmd);
1897 }
1898 
1899