2 * Copyright (c) 2005-2011 Atheros Communications Inc.
3 * Copyright (c) 2011-2013 Qualcomm Atheros, Inc.
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.
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.
25 #include <linux/log2.h>
27 /* slightly larger than one large A-MPDU */
28 #define HTT_RX_RING_SIZE_MIN 128
30 /* roughly 20 ms @ 1 Gbps of 1500B MSDUs */
31 #define HTT_RX_RING_SIZE_MAX 2048
33 #define HTT_RX_AVG_FRM_BYTES 1000
35 /* ms, very conservative */
36 #define HTT_RX_HOST_LATENCY_MAX_MS 20
38 /* ms, conservative */
39 #define HTT_RX_HOST_LATENCY_WORST_LIKELY_MS 10
41 /* when under memory pressure rx ring refill may fail and needs a retry */
42 #define HTT_RX_RING_REFILL_RETRY_MS 50
45 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb);
46 static void ath10k_htt_txrx_compl_task(unsigned long ptr);
48 static int ath10k_htt_rx_ring_size(struct ath10k_htt *htt)
53 * It is expected that the host CPU will typically be able to
54 * service the rx indication from one A-MPDU before the rx
55 * indication from the subsequent A-MPDU happens, roughly 1-2 ms
56 * later. However, the rx ring should be sized very conservatively,
57 * to accomodate the worst reasonable delay before the host CPU
58 * services a rx indication interrupt.
60 * The rx ring need not be kept full of empty buffers. In theory,
61 * the htt host SW can dynamically track the low-water mark in the
62 * rx ring, and dynamically adjust the level to which the rx ring
63 * is filled with empty buffers, to dynamically meet the desired
66 * In contrast, it's difficult to resize the rx ring itself, once
67 * it's in use. Thus, the ring itself should be sized very
68 * conservatively, while the degree to which the ring is filled
69 * with empty buffers should be sized moderately conservatively.
72 /* 1e6 bps/mbps / 1e3 ms per sec = 1000 */
74 htt->max_throughput_mbps +
76 (8 * HTT_RX_AVG_FRM_BYTES) * HTT_RX_HOST_LATENCY_MAX_MS;
78 if (size < HTT_RX_RING_SIZE_MIN)
79 size = HTT_RX_RING_SIZE_MIN;
81 if (size > HTT_RX_RING_SIZE_MAX)
82 size = HTT_RX_RING_SIZE_MAX;
84 size = roundup_pow_of_two(size);
89 static int ath10k_htt_rx_ring_fill_level(struct ath10k_htt *htt)
93 /* 1e6 bps/mbps / 1e3 ms per sec = 1000 */
95 htt->max_throughput_mbps *
97 (8 * HTT_RX_AVG_FRM_BYTES) * HTT_RX_HOST_LATENCY_WORST_LIKELY_MS;
100 * Make sure the fill level is at least 1 less than the ring size.
101 * Leaving 1 element empty allows the SW to easily distinguish
102 * between a full ring vs. an empty ring.
104 if (size >= htt->rx_ring.size)
105 size = htt->rx_ring.size - 1;
110 static void ath10k_htt_rx_ring_free(struct ath10k_htt *htt)
113 struct ath10k_skb_cb *cb;
116 for (i = 0; i < htt->rx_ring.fill_cnt; i++) {
117 skb = htt->rx_ring.netbufs_ring[i];
118 cb = ATH10K_SKB_CB(skb);
119 dma_unmap_single(htt->ar->dev, cb->paddr,
120 skb->len + skb_tailroom(skb),
122 dev_kfree_skb_any(skb);
125 htt->rx_ring.fill_cnt = 0;
128 static int __ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
130 struct htt_rx_desc *rx_desc;
135 idx = __le32_to_cpu(*(htt->rx_ring.alloc_idx.vaddr));
137 skb = dev_alloc_skb(HTT_RX_BUF_SIZE + HTT_RX_DESC_ALIGN);
143 if (!IS_ALIGNED((unsigned long)skb->data, HTT_RX_DESC_ALIGN))
145 PTR_ALIGN(skb->data, HTT_RX_DESC_ALIGN) -
148 /* Clear rx_desc attention word before posting to Rx ring */
149 rx_desc = (struct htt_rx_desc *)skb->data;
150 rx_desc->attention.flags = __cpu_to_le32(0);
152 paddr = dma_map_single(htt->ar->dev, skb->data,
153 skb->len + skb_tailroom(skb),
156 if (unlikely(dma_mapping_error(htt->ar->dev, paddr))) {
157 dev_kfree_skb_any(skb);
162 ATH10K_SKB_CB(skb)->paddr = paddr;
163 htt->rx_ring.netbufs_ring[idx] = skb;
164 htt->rx_ring.paddrs_ring[idx] = __cpu_to_le32(paddr);
165 htt->rx_ring.fill_cnt++;
169 idx &= htt->rx_ring.size_mask;
173 *(htt->rx_ring.alloc_idx.vaddr) = __cpu_to_le32(idx);
177 static int ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
179 lockdep_assert_held(&htt->rx_ring.lock);
180 return __ath10k_htt_rx_ring_fill_n(htt, num);
183 static void ath10k_htt_rx_msdu_buff_replenish(struct ath10k_htt *htt)
185 int ret, num_deficit, num_to_fill;
187 /* Refilling the whole RX ring buffer proves to be a bad idea. The
188 * reason is RX may take up significant amount of CPU cycles and starve
189 * other tasks, e.g. TX on an ethernet device while acting as a bridge
190 * with ath10k wlan interface. This ended up with very poor performance
191 * once CPU the host system was overwhelmed with RX on ath10k.
193 * By limiting the number of refills the replenishing occurs
194 * progressively. This in turns makes use of the fact tasklets are
195 * processed in FIFO order. This means actual RX processing can starve
196 * out refilling. If there's not enough buffers on RX ring FW will not
197 * report RX until it is refilled with enough buffers. This
198 * automatically balances load wrt to CPU power.
200 * This probably comes at a cost of lower maximum throughput but
201 * improves the avarage and stability. */
202 spin_lock_bh(&htt->rx_ring.lock);
203 num_deficit = htt->rx_ring.fill_level - htt->rx_ring.fill_cnt;
204 num_to_fill = min(ATH10K_HTT_MAX_NUM_REFILL, num_deficit);
205 num_deficit -= num_to_fill;
206 ret = ath10k_htt_rx_ring_fill_n(htt, num_to_fill);
207 if (ret == -ENOMEM) {
209 * Failed to fill it to the desired level -
210 * we'll start a timer and try again next time.
211 * As long as enough buffers are left in the ring for
212 * another A-MPDU rx, no special recovery is needed.
214 mod_timer(&htt->rx_ring.refill_retry_timer, jiffies +
215 msecs_to_jiffies(HTT_RX_RING_REFILL_RETRY_MS));
216 } else if (num_deficit > 0) {
217 tasklet_schedule(&htt->rx_replenish_task);
219 spin_unlock_bh(&htt->rx_ring.lock);
222 static void ath10k_htt_rx_ring_refill_retry(unsigned long arg)
224 struct ath10k_htt *htt = (struct ath10k_htt *)arg;
225 ath10k_htt_rx_msdu_buff_replenish(htt);
228 void ath10k_htt_rx_detach(struct ath10k_htt *htt)
230 int sw_rd_idx = htt->rx_ring.sw_rd_idx.msdu_payld;
232 del_timer_sync(&htt->rx_ring.refill_retry_timer);
233 tasklet_kill(&htt->rx_replenish_task);
234 tasklet_kill(&htt->txrx_compl_task);
236 skb_queue_purge(&htt->tx_compl_q);
237 skb_queue_purge(&htt->rx_compl_q);
239 while (sw_rd_idx != __le32_to_cpu(*(htt->rx_ring.alloc_idx.vaddr))) {
240 struct sk_buff *skb =
241 htt->rx_ring.netbufs_ring[sw_rd_idx];
242 struct ath10k_skb_cb *cb = ATH10K_SKB_CB(skb);
244 dma_unmap_single(htt->ar->dev, cb->paddr,
245 skb->len + skb_tailroom(skb),
247 dev_kfree_skb_any(htt->rx_ring.netbufs_ring[sw_rd_idx]);
249 sw_rd_idx &= htt->rx_ring.size_mask;
252 dma_free_coherent(htt->ar->dev,
254 sizeof(htt->rx_ring.paddrs_ring)),
255 htt->rx_ring.paddrs_ring,
256 htt->rx_ring.base_paddr);
258 dma_free_coherent(htt->ar->dev,
259 sizeof(*htt->rx_ring.alloc_idx.vaddr),
260 htt->rx_ring.alloc_idx.vaddr,
261 htt->rx_ring.alloc_idx.paddr);
263 kfree(htt->rx_ring.netbufs_ring);
266 static inline struct sk_buff *ath10k_htt_rx_netbuf_pop(struct ath10k_htt *htt)
269 struct sk_buff *msdu;
271 lockdep_assert_held(&htt->rx_ring.lock);
273 if (htt->rx_ring.fill_cnt == 0) {
274 ath10k_warn("tried to pop sk_buff from an empty rx ring\n");
278 idx = htt->rx_ring.sw_rd_idx.msdu_payld;
279 msdu = htt->rx_ring.netbufs_ring[idx];
282 idx &= htt->rx_ring.size_mask;
283 htt->rx_ring.sw_rd_idx.msdu_payld = idx;
284 htt->rx_ring.fill_cnt--;
289 static void ath10k_htt_rx_free_msdu_chain(struct sk_buff *skb)
291 struct sk_buff *next;
295 dev_kfree_skb_any(skb);
300 static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
301 u8 **fw_desc, int *fw_desc_len,
302 struct sk_buff **head_msdu,
303 struct sk_buff **tail_msdu)
305 int msdu_len, msdu_chaining = 0;
306 struct sk_buff *msdu;
307 struct htt_rx_desc *rx_desc;
309 lockdep_assert_held(&htt->rx_ring.lock);
311 if (htt->rx_confused) {
312 ath10k_warn("htt is confused. refusing rx\n");
316 msdu = *head_msdu = ath10k_htt_rx_netbuf_pop(htt);
318 int last_msdu, msdu_len_invalid, msdu_chained;
320 dma_unmap_single(htt->ar->dev,
321 ATH10K_SKB_CB(msdu)->paddr,
322 msdu->len + skb_tailroom(msdu),
325 ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt rx pop: ",
326 msdu->data, msdu->len + skb_tailroom(msdu));
328 rx_desc = (struct htt_rx_desc *)msdu->data;
330 /* FIXME: we must report msdu payload since this is what caller
332 skb_put(msdu, offsetof(struct htt_rx_desc, msdu_payload));
333 skb_pull(msdu, offsetof(struct htt_rx_desc, msdu_payload));
336 * Sanity check - confirm the HW is finished filling in the
338 * If the HW and SW are working correctly, then it's guaranteed
339 * that the HW's MAC DMA is done before this point in the SW.
340 * To prevent the case that we handle a stale Rx descriptor,
341 * just assert for now until we have a way to recover.
343 if (!(__le32_to_cpu(rx_desc->attention.flags)
344 & RX_ATTENTION_FLAGS_MSDU_DONE)) {
345 ath10k_htt_rx_free_msdu_chain(*head_msdu);
348 ath10k_err("htt rx stopped. cannot recover\n");
349 htt->rx_confused = true;
354 * Copy the FW rx descriptor for this MSDU from the rx
355 * indication message into the MSDU's netbuf. HL uses the
356 * same rx indication message definition as LL, and simply
357 * appends new info (fields from the HW rx desc, and the
358 * MSDU payload itself). So, the offset into the rx
359 * indication message only has to account for the standard
360 * offset of the per-MSDU FW rx desc info within the
361 * message, and how many bytes of the per-MSDU FW rx desc
362 * info have already been consumed. (And the endianness of
363 * the host, since for a big-endian host, the rx ind
364 * message contents, including the per-MSDU rx desc bytes,
365 * were byteswapped during upload.)
367 if (*fw_desc_len > 0) {
368 rx_desc->fw_desc.info0 = **fw_desc;
370 * The target is expected to only provide the basic
371 * per-MSDU rx descriptors. Just to be sure, verify
372 * that the target has not attached extension data
373 * (e.g. LRO flow ID).
376 /* or more, if there's extension data */
381 * When an oversized AMSDU happened, FW will lost
382 * some of MSDU status - in this case, the FW
383 * descriptors provided will be less than the
384 * actual MSDUs inside this MPDU. Mark the FW
385 * descriptors so that it will still deliver to
386 * upper stack, if no CRC error for this MPDU.
388 * FIX THIS - the FW descriptors are actually for
389 * MSDUs in the end of this A-MSDU instead of the
392 rx_desc->fw_desc.info0 = 0;
395 msdu_len_invalid = !!(__le32_to_cpu(rx_desc->attention.flags)
396 & (RX_ATTENTION_FLAGS_MPDU_LENGTH_ERR |
397 RX_ATTENTION_FLAGS_MSDU_LENGTH_ERR));
398 msdu_len = MS(__le32_to_cpu(rx_desc->msdu_start.info0),
399 RX_MSDU_START_INFO0_MSDU_LENGTH);
400 msdu_chained = rx_desc->frag_info.ring2_more_count;
401 msdu_chaining = msdu_chained;
403 if (msdu_len_invalid)
407 skb_put(msdu, min(msdu_len, HTT_RX_MSDU_SIZE));
408 msdu_len -= msdu->len;
410 /* FIXME: Do chained buffers include htt_rx_desc or not? */
411 while (msdu_chained--) {
412 struct sk_buff *next = ath10k_htt_rx_netbuf_pop(htt);
414 dma_unmap_single(htt->ar->dev,
415 ATH10K_SKB_CB(next)->paddr,
416 next->len + skb_tailroom(next),
419 ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL,
420 "htt rx chained: ", next->data,
421 next->len + skb_tailroom(next));
424 skb_put(next, min(msdu_len, HTT_RX_BUF_SIZE));
425 msdu_len -= next->len;
431 last_msdu = __le32_to_cpu(rx_desc->msdu_end.info0) &
432 RX_MSDU_END_INFO0_LAST_MSDU;
438 struct sk_buff *next = ath10k_htt_rx_netbuf_pop(htt);
446 * Don't refill the ring yet.
448 * First, the elements popped here are still in use - it is not
449 * safe to overwrite them until the matching call to
450 * mpdu_desc_list_next. Second, for efficiency it is preferable to
451 * refill the rx ring with 1 PPDU's worth of rx buffers (something
452 * like 32 x 3 buffers), rather than one MPDU's worth of rx buffers
453 * (something like 3 buffers). Consequently, we'll rely on the txrx
454 * SW to tell us when it is done pulling all the PPDU's rx buffers
455 * out of the rx ring, and then refill it just once.
458 return msdu_chaining;
461 static void ath10k_htt_rx_replenish_task(unsigned long ptr)
463 struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
464 ath10k_htt_rx_msdu_buff_replenish(htt);
467 int ath10k_htt_rx_attach(struct ath10k_htt *htt)
471 struct timer_list *timer = &htt->rx_ring.refill_retry_timer;
473 htt->rx_ring.size = ath10k_htt_rx_ring_size(htt);
474 if (!is_power_of_2(htt->rx_ring.size)) {
475 ath10k_warn("htt rx ring size is not power of 2\n");
479 htt->rx_ring.size_mask = htt->rx_ring.size - 1;
482 * Set the initial value for the level to which the rx ring
483 * should be filled, based on the max throughput and the
484 * worst likely latency for the host to fill the rx ring
485 * with new buffers. In theory, this fill level can be
486 * dynamically adjusted from the initial value set here, to
487 * reflect the actual host latency rather than a
488 * conservative assumption about the host latency.
490 htt->rx_ring.fill_level = ath10k_htt_rx_ring_fill_level(htt);
492 htt->rx_ring.netbufs_ring =
493 kmalloc(htt->rx_ring.size * sizeof(struct sk_buff *),
495 if (!htt->rx_ring.netbufs_ring)
498 vaddr = dma_alloc_coherent(htt->ar->dev,
499 (htt->rx_ring.size * sizeof(htt->rx_ring.paddrs_ring)),
504 htt->rx_ring.paddrs_ring = vaddr;
505 htt->rx_ring.base_paddr = paddr;
507 vaddr = dma_alloc_coherent(htt->ar->dev,
508 sizeof(*htt->rx_ring.alloc_idx.vaddr),
513 htt->rx_ring.alloc_idx.vaddr = vaddr;
514 htt->rx_ring.alloc_idx.paddr = paddr;
515 htt->rx_ring.sw_rd_idx.msdu_payld = 0;
516 *htt->rx_ring.alloc_idx.vaddr = 0;
518 /* Initialize the Rx refill retry timer */
519 setup_timer(timer, ath10k_htt_rx_ring_refill_retry, (unsigned long)htt);
521 spin_lock_init(&htt->rx_ring.lock);
523 htt->rx_ring.fill_cnt = 0;
524 if (__ath10k_htt_rx_ring_fill_n(htt, htt->rx_ring.fill_level))
527 tasklet_init(&htt->rx_replenish_task, ath10k_htt_rx_replenish_task,
530 skb_queue_head_init(&htt->tx_compl_q);
531 skb_queue_head_init(&htt->rx_compl_q);
533 tasklet_init(&htt->txrx_compl_task, ath10k_htt_txrx_compl_task,
536 ath10k_dbg(ATH10K_DBG_BOOT, "htt rx ring size %d fill_level %d\n",
537 htt->rx_ring.size, htt->rx_ring.fill_level);
541 ath10k_htt_rx_ring_free(htt);
542 dma_free_coherent(htt->ar->dev,
543 sizeof(*htt->rx_ring.alloc_idx.vaddr),
544 htt->rx_ring.alloc_idx.vaddr,
545 htt->rx_ring.alloc_idx.paddr);
547 dma_free_coherent(htt->ar->dev,
549 sizeof(htt->rx_ring.paddrs_ring)),
550 htt->rx_ring.paddrs_ring,
551 htt->rx_ring.base_paddr);
553 kfree(htt->rx_ring.netbufs_ring);
558 static int ath10k_htt_rx_crypto_param_len(enum htt_rx_mpdu_encrypt_type type)
561 case HTT_RX_MPDU_ENCRYPT_WEP40:
562 case HTT_RX_MPDU_ENCRYPT_WEP104:
564 case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
565 case HTT_RX_MPDU_ENCRYPT_WEP128: /* not tested */
566 case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
567 case HTT_RX_MPDU_ENCRYPT_WAPI: /* not tested */
568 case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
570 case HTT_RX_MPDU_ENCRYPT_NONE:
574 ath10k_warn("unknown encryption type %d\n", type);
578 static int ath10k_htt_rx_crypto_tail_len(enum htt_rx_mpdu_encrypt_type type)
581 case HTT_RX_MPDU_ENCRYPT_NONE:
582 case HTT_RX_MPDU_ENCRYPT_WEP40:
583 case HTT_RX_MPDU_ENCRYPT_WEP104:
584 case HTT_RX_MPDU_ENCRYPT_WEP128:
585 case HTT_RX_MPDU_ENCRYPT_WAPI:
587 case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
588 case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
590 case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
594 ath10k_warn("unknown encryption type %d\n", type);
598 /* Applies for first msdu in chain, before altering it. */
599 static struct ieee80211_hdr *ath10k_htt_rx_skb_get_hdr(struct sk_buff *skb)
601 struct htt_rx_desc *rxd;
602 enum rx_msdu_decap_format fmt;
604 rxd = (void *)skb->data - sizeof(*rxd);
605 fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
606 RX_MSDU_START_INFO1_DECAP_FORMAT);
608 if (fmt == RX_MSDU_DECAP_RAW)
609 return (void *)skb->data;
611 return (void *)skb->data - RX_HTT_HDR_STATUS_LEN;
614 /* This function only applies for first msdu in an msdu chain */
615 static bool ath10k_htt_rx_hdr_is_amsdu(struct ieee80211_hdr *hdr)
617 if (ieee80211_is_data_qos(hdr->frame_control)) {
618 u8 *qc = ieee80211_get_qos_ctl(hdr);
633 struct amsdu_subframe_hdr {
639 static int ath10k_htt_rx_nwifi_hdrlen(struct ieee80211_hdr *hdr)
641 /* nwifi header is padded to 4 bytes. this fixes 4addr rx */
642 return round_up(ieee80211_hdrlen(hdr->frame_control), 4);
645 static void ath10k_htt_rx_amsdu(struct ath10k_htt *htt,
646 struct htt_rx_info *info)
648 struct htt_rx_desc *rxd;
649 struct sk_buff *first;
650 struct sk_buff *skb = info->skb;
651 enum rx_msdu_decap_format fmt;
652 enum htt_rx_mpdu_encrypt_type enctype;
653 struct ieee80211_hdr *hdr;
654 u8 hdr_buf[64], addr[ETH_ALEN], *qos;
655 unsigned int hdr_len;
657 rxd = (void *)skb->data - sizeof(*rxd);
658 enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
659 RX_MPDU_START_INFO0_ENCRYPT_TYPE);
661 hdr = (struct ieee80211_hdr *)rxd->rx_hdr_status;
662 hdr_len = ieee80211_hdrlen(hdr->frame_control);
663 memcpy(hdr_buf, hdr, hdr_len);
664 hdr = (struct ieee80211_hdr *)hdr_buf;
671 rxd = (void *)skb->data - sizeof(*rxd);
672 fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
673 RX_MSDU_START_INFO1_DECAP_FORMAT);
674 decap_hdr = (void *)rxd->rx_hdr_status;
676 skb->ip_summed = ath10k_htt_rx_get_csum_state(skb);
678 /* First frame in an A-MSDU chain has more decapped data. */
680 len = round_up(ieee80211_hdrlen(hdr->frame_control), 4);
681 len += round_up(ath10k_htt_rx_crypto_param_len(enctype),
687 case RX_MSDU_DECAP_RAW:
688 /* remove trailing FCS */
689 skb_trim(skb, skb->len - FCS_LEN);
691 case RX_MSDU_DECAP_NATIVE_WIFI:
692 /* pull decapped header and copy DA */
693 hdr = (struct ieee80211_hdr *)skb->data;
694 hdr_len = ath10k_htt_rx_nwifi_hdrlen(hdr);
695 memcpy(addr, ieee80211_get_DA(hdr), ETH_ALEN);
696 skb_pull(skb, hdr_len);
698 /* push original 802.11 header */
699 hdr = (struct ieee80211_hdr *)hdr_buf;
700 hdr_len = ieee80211_hdrlen(hdr->frame_control);
701 memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
703 /* original A-MSDU header has the bit set but we're
704 * not including A-MSDU subframe header */
705 hdr = (struct ieee80211_hdr *)skb->data;
706 qos = ieee80211_get_qos_ctl(hdr);
707 qos[0] &= ~IEEE80211_QOS_CTL_A_MSDU_PRESENT;
709 /* original 802.11 header has a different DA */
710 memcpy(ieee80211_get_DA(hdr), addr, ETH_ALEN);
712 case RX_MSDU_DECAP_ETHERNET2_DIX:
713 /* strip ethernet header and insert decapped 802.11
714 * header, amsdu subframe header and rfc1042 header */
717 len += sizeof(struct rfc1042_hdr);
718 len += sizeof(struct amsdu_subframe_hdr);
720 skb_pull(skb, sizeof(struct ethhdr));
721 memcpy(skb_push(skb, len), decap_hdr, len);
722 memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
724 case RX_MSDU_DECAP_8023_SNAP_LLC:
725 /* insert decapped 802.11 header making a singly
727 memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
732 info->encrypt_type = enctype;
734 info->skb->next = NULL;
737 info->amsdu_more = true;
739 ath10k_process_rx(htt->ar, info);
742 /* FIXME: It might be nice to re-assemble the A-MSDU when there's a
743 * monitor interface active for sniffing purposes. */
746 static void ath10k_htt_rx_msdu(struct ath10k_htt *htt, struct htt_rx_info *info)
748 struct sk_buff *skb = info->skb;
749 struct htt_rx_desc *rxd;
750 struct ieee80211_hdr *hdr;
751 enum rx_msdu_decap_format fmt;
752 enum htt_rx_mpdu_encrypt_type enctype;
756 /* This shouldn't happen. If it does than it may be a FW bug. */
758 ath10k_warn("htt rx received chained non A-MSDU frame\n");
759 ath10k_htt_rx_free_msdu_chain(skb->next);
763 rxd = (void *)skb->data - sizeof(*rxd);
764 fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
765 RX_MSDU_START_INFO1_DECAP_FORMAT);
766 enctype = MS(__le32_to_cpu(rxd->mpdu_start.info0),
767 RX_MPDU_START_INFO0_ENCRYPT_TYPE);
768 hdr = (struct ieee80211_hdr *)rxd->rx_hdr_status;
769 hdr_len = ieee80211_hdrlen(hdr->frame_control);
771 skb->ip_summed = ath10k_htt_rx_get_csum_state(skb);
774 case RX_MSDU_DECAP_RAW:
775 /* remove trailing FCS */
776 skb_trim(skb, skb->len - FCS_LEN);
778 case RX_MSDU_DECAP_NATIVE_WIFI:
779 /* Pull decapped header */
780 hdr = (struct ieee80211_hdr *)skb->data;
781 hdr_len = ath10k_htt_rx_nwifi_hdrlen(hdr);
782 skb_pull(skb, hdr_len);
784 /* Push original header */
785 hdr = (struct ieee80211_hdr *)rxd->rx_hdr_status;
786 hdr_len = ieee80211_hdrlen(hdr->frame_control);
787 memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
789 case RX_MSDU_DECAP_ETHERNET2_DIX:
790 /* strip ethernet header and insert decapped 802.11 header and
794 rfc1042 += roundup(hdr_len, 4);
795 rfc1042 += roundup(ath10k_htt_rx_crypto_param_len(enctype), 4);
797 skb_pull(skb, sizeof(struct ethhdr));
798 memcpy(skb_push(skb, sizeof(struct rfc1042_hdr)),
799 rfc1042, sizeof(struct rfc1042_hdr));
800 memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
802 case RX_MSDU_DECAP_8023_SNAP_LLC:
803 /* remove A-MSDU subframe header and insert
804 * decapped 802.11 header. rfc1042 header is already there */
806 skb_pull(skb, sizeof(struct amsdu_subframe_hdr));
807 memcpy(skb_push(skb, hdr_len), hdr, hdr_len);
812 info->encrypt_type = enctype;
814 ath10k_process_rx(htt->ar, info);
817 static bool ath10k_htt_rx_has_decrypt_err(struct sk_buff *skb)
819 struct htt_rx_desc *rxd;
822 rxd = (void *)skb->data - sizeof(*rxd);
823 flags = __le32_to_cpu(rxd->attention.flags);
825 if (flags & RX_ATTENTION_FLAGS_DECRYPT_ERR)
831 static bool ath10k_htt_rx_has_fcs_err(struct sk_buff *skb)
833 struct htt_rx_desc *rxd;
836 rxd = (void *)skb->data - sizeof(*rxd);
837 flags = __le32_to_cpu(rxd->attention.flags);
839 if (flags & RX_ATTENTION_FLAGS_FCS_ERR)
845 static bool ath10k_htt_rx_has_mic_err(struct sk_buff *skb)
847 struct htt_rx_desc *rxd;
850 rxd = (void *)skb->data - sizeof(*rxd);
851 flags = __le32_to_cpu(rxd->attention.flags);
853 if (flags & RX_ATTENTION_FLAGS_TKIP_MIC_ERR)
859 static bool ath10k_htt_rx_is_mgmt(struct sk_buff *skb)
861 struct htt_rx_desc *rxd;
864 rxd = (void *)skb->data - sizeof(*rxd);
865 flags = __le32_to_cpu(rxd->attention.flags);
867 if (flags & RX_ATTENTION_FLAGS_MGMT_TYPE)
873 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb)
875 struct htt_rx_desc *rxd;
879 bool ip_csum_ok, tcpudp_csum_ok;
881 rxd = (void *)skb->data - sizeof(*rxd);
882 flags = __le32_to_cpu(rxd->attention.flags);
883 info = __le32_to_cpu(rxd->msdu_start.info1);
885 is_ip4 = !!(info & RX_MSDU_START_INFO1_IPV4_PROTO);
886 is_ip6 = !!(info & RX_MSDU_START_INFO1_IPV6_PROTO);
887 is_tcp = !!(info & RX_MSDU_START_INFO1_TCP_PROTO);
888 is_udp = !!(info & RX_MSDU_START_INFO1_UDP_PROTO);
889 ip_csum_ok = !(flags & RX_ATTENTION_FLAGS_IP_CHKSUM_FAIL);
890 tcpudp_csum_ok = !(flags & RX_ATTENTION_FLAGS_TCP_UDP_CHKSUM_FAIL);
892 if (!is_ip4 && !is_ip6)
893 return CHECKSUM_NONE;
894 if (!is_tcp && !is_udp)
895 return CHECKSUM_NONE;
897 return CHECKSUM_NONE;
899 return CHECKSUM_NONE;
901 return CHECKSUM_UNNECESSARY;
904 static int ath10k_unchain_msdu(struct sk_buff *msdu_head)
906 struct sk_buff *next = msdu_head->next;
907 struct sk_buff *to_free = next;
911 /* TODO: Might could optimize this by using
912 * skb_try_coalesce or similar method to
913 * decrease copying, or maybe get mac80211 to
914 * provide a way to just receive a list of
918 msdu_head->next = NULL;
920 /* Allocate total length all at once. */
922 total_len += next->len;
926 space = total_len - skb_tailroom(msdu_head);
928 (pskb_expand_head(msdu_head, 0, space, GFP_ATOMIC) < 0)) {
929 /* TODO: bump some rx-oom error stat */
930 /* put it back together so we can free the
931 * whole list at once.
933 msdu_head->next = to_free;
937 /* Walk list again, copying contents into
942 skb_copy_from_linear_data(next, skb_put(msdu_head, next->len),
947 /* If here, we have consolidated skb. Free the
948 * fragments and pass the main skb on up the
951 ath10k_htt_rx_free_msdu_chain(to_free);
955 static void ath10k_htt_rx_handler(struct ath10k_htt *htt,
956 struct htt_rx_indication *rx)
958 struct htt_rx_info info;
959 struct htt_rx_indication_mpdu_range *mpdu_ranges;
960 struct ieee80211_hdr *hdr;
966 lockdep_assert_held(&htt->rx_ring.lock);
968 memset(&info, 0, sizeof(info));
970 fw_desc_len = __le16_to_cpu(rx->prefix.fw_rx_desc_bytes);
971 fw_desc = (u8 *)&rx->fw_desc;
973 num_mpdu_ranges = MS(__le32_to_cpu(rx->hdr.info1),
974 HTT_RX_INDICATION_INFO1_NUM_MPDU_RANGES);
975 mpdu_ranges = htt_rx_ind_get_mpdu_ranges(rx);
977 ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt rx ind: ",
979 (sizeof(struct htt_rx_indication_mpdu_range) *
982 for (i = 0; i < num_mpdu_ranges; i++) {
983 info.status = mpdu_ranges[i].mpdu_range_status;
985 for (j = 0; j < mpdu_ranges[i].mpdu_count; j++) {
986 struct sk_buff *msdu_head, *msdu_tail;
987 enum htt_rx_mpdu_status status;
992 msdu_chaining = ath10k_htt_rx_amsdu_pop(htt,
999 ath10k_warn("htt rx no data!\n");
1003 if (msdu_head->len == 0) {
1004 ath10k_dbg(ATH10K_DBG_HTT,
1005 "htt rx dropping due to zero-len\n");
1006 ath10k_htt_rx_free_msdu_chain(msdu_head);
1010 if (ath10k_htt_rx_has_decrypt_err(msdu_head)) {
1011 ath10k_dbg(ATH10K_DBG_HTT,
1012 "htt rx dropping due to decrypt-err\n");
1013 ath10k_htt_rx_free_msdu_chain(msdu_head);
1017 status = info.status;
1019 /* Skip mgmt frames while we handle this in WMI */
1020 if (status == HTT_RX_IND_MPDU_STATUS_MGMT_CTRL ||
1021 ath10k_htt_rx_is_mgmt(msdu_head)) {
1022 ath10k_dbg(ATH10K_DBG_HTT, "htt rx mgmt ctrl\n");
1023 ath10k_htt_rx_free_msdu_chain(msdu_head);
1027 if (status != HTT_RX_IND_MPDU_STATUS_OK &&
1028 status != HTT_RX_IND_MPDU_STATUS_TKIP_MIC_ERR &&
1029 status != HTT_RX_IND_MPDU_STATUS_ERR_INV_PEER &&
1030 !htt->ar->monitor_enabled) {
1031 ath10k_dbg(ATH10K_DBG_HTT,
1032 "htt rx ignoring frame w/ status %d\n",
1034 ath10k_htt_rx_free_msdu_chain(msdu_head);
1038 if (test_bit(ATH10K_CAC_RUNNING, &htt->ar->dev_flags)) {
1039 ath10k_dbg(ATH10K_DBG_HTT,
1040 "htt rx CAC running\n");
1041 ath10k_htt_rx_free_msdu_chain(msdu_head);
1045 if (msdu_chaining &&
1046 (ath10k_unchain_msdu(msdu_head) < 0)) {
1047 ath10k_htt_rx_free_msdu_chain(msdu_head);
1051 info.skb = msdu_head;
1052 info.fcs_err = ath10k_htt_rx_has_fcs_err(msdu_head);
1053 info.mic_err = ath10k_htt_rx_has_mic_err(msdu_head);
1056 ath10k_dbg(ATH10K_DBG_HTT,
1057 "htt rx has FCS err\n");
1060 ath10k_dbg(ATH10K_DBG_HTT,
1061 "htt rx has MIC err\n");
1063 info.signal = ATH10K_DEFAULT_NOISE_FLOOR;
1064 info.signal += rx->ppdu.combined_rssi;
1066 info.rate.info0 = rx->ppdu.info0;
1067 info.rate.info1 = __le32_to_cpu(rx->ppdu.info1);
1068 info.rate.info2 = __le32_to_cpu(rx->ppdu.info2);
1069 info.tsf = __le32_to_cpu(rx->ppdu.tsf);
1071 hdr = ath10k_htt_rx_skb_get_hdr(msdu_head);
1073 if (ath10k_htt_rx_hdr_is_amsdu(hdr))
1074 ath10k_htt_rx_amsdu(htt, &info);
1076 ath10k_htt_rx_msdu(htt, &info);
1080 tasklet_schedule(&htt->rx_replenish_task);
1083 static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
1084 struct htt_rx_fragment_indication *frag)
1086 struct sk_buff *msdu_head, *msdu_tail;
1087 struct htt_rx_desc *rxd;
1088 enum rx_msdu_decap_format fmt;
1089 struct htt_rx_info info = {};
1090 struct ieee80211_hdr *hdr;
1095 int fw_desc_len, hdrlen, paramlen;
1098 fw_desc_len = __le16_to_cpu(frag->fw_rx_desc_bytes);
1099 fw_desc = (u8 *)frag->fw_msdu_rx_desc;
1104 spin_lock_bh(&htt->rx_ring.lock);
1105 msdu_chaining = ath10k_htt_rx_amsdu_pop(htt, &fw_desc, &fw_desc_len,
1106 &msdu_head, &msdu_tail);
1107 spin_unlock_bh(&htt->rx_ring.lock);
1109 ath10k_dbg(ATH10K_DBG_HTT_DUMP, "htt rx frag ahead\n");
1112 ath10k_warn("htt rx frag no data\n");
1116 if (msdu_chaining || msdu_head != msdu_tail) {
1117 ath10k_warn("aggregation with fragmentation?!\n");
1118 ath10k_htt_rx_free_msdu_chain(msdu_head);
1122 /* FIXME: implement signal strength */
1124 hdr = (struct ieee80211_hdr *)msdu_head->data;
1125 rxd = (void *)msdu_head->data - sizeof(*rxd);
1126 tkip_mic_err = !!(__le32_to_cpu(rxd->attention.flags) &
1127 RX_ATTENTION_FLAGS_TKIP_MIC_ERR);
1128 decrypt_err = !!(__le32_to_cpu(rxd->attention.flags) &
1129 RX_ATTENTION_FLAGS_DECRYPT_ERR);
1130 fmt = MS(__le32_to_cpu(rxd->msdu_start.info1),
1131 RX_MSDU_START_INFO1_DECAP_FORMAT);
1133 if (fmt != RX_MSDU_DECAP_RAW) {
1134 ath10k_warn("we dont support non-raw fragmented rx yet\n");
1135 dev_kfree_skb_any(msdu_head);
1139 info.skb = msdu_head;
1140 info.status = HTT_RX_IND_MPDU_STATUS_OK;
1141 info.encrypt_type = MS(__le32_to_cpu(rxd->mpdu_start.info0),
1142 RX_MPDU_START_INFO0_ENCRYPT_TYPE);
1143 info.skb->ip_summed = ath10k_htt_rx_get_csum_state(info.skb);
1146 ath10k_warn("tkip mic error\n");
1147 info.status = HTT_RX_IND_MPDU_STATUS_TKIP_MIC_ERR;
1151 ath10k_warn("decryption err in fragmented rx\n");
1152 dev_kfree_skb_any(info.skb);
1156 if (info.encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE) {
1157 hdrlen = ieee80211_hdrlen(hdr->frame_control);
1158 paramlen = ath10k_htt_rx_crypto_param_len(info.encrypt_type);
1160 /* It is more efficient to move the header than the payload */
1161 memmove((void *)info.skb->data + paramlen,
1162 (void *)info.skb->data,
1164 skb_pull(info.skb, paramlen);
1165 hdr = (struct ieee80211_hdr *)info.skb->data;
1168 /* remove trailing FCS */
1171 /* remove crypto trailer */
1172 trim += ath10k_htt_rx_crypto_tail_len(info.encrypt_type);
1174 /* last fragment of TKIP frags has MIC */
1175 if (!ieee80211_has_morefrags(hdr->frame_control) &&
1176 info.encrypt_type == HTT_RX_MPDU_ENCRYPT_TKIP_WPA)
1179 if (trim > info.skb->len) {
1180 ath10k_warn("htt rx fragment: trailer longer than the frame itself? drop\n");
1181 dev_kfree_skb_any(info.skb);
1185 skb_trim(info.skb, info.skb->len - trim);
1187 ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt rx frag mpdu: ",
1188 info.skb->data, info.skb->len);
1189 ath10k_process_rx(htt->ar, &info);
1192 if (fw_desc_len > 0) {
1193 ath10k_dbg(ATH10K_DBG_HTT,
1194 "expecting more fragmented rx in one indication %d\n",
1199 static void ath10k_htt_rx_frm_tx_compl(struct ath10k *ar,
1200 struct sk_buff *skb)
1202 struct ath10k_htt *htt = &ar->htt;
1203 struct htt_resp *resp = (struct htt_resp *)skb->data;
1204 struct htt_tx_done tx_done = {};
1205 int status = MS(resp->data_tx_completion.flags, HTT_DATA_TX_STATUS);
1209 lockdep_assert_held(&htt->tx_lock);
1212 case HTT_DATA_TX_STATUS_NO_ACK:
1213 tx_done.no_ack = true;
1215 case HTT_DATA_TX_STATUS_OK:
1217 case HTT_DATA_TX_STATUS_DISCARD:
1218 case HTT_DATA_TX_STATUS_POSTPONE:
1219 case HTT_DATA_TX_STATUS_DOWNLOAD_FAIL:
1220 tx_done.discard = true;
1223 ath10k_warn("unhandled tx completion status %d\n", status);
1224 tx_done.discard = true;
1228 ath10k_dbg(ATH10K_DBG_HTT, "htt tx completion num_msdus %d\n",
1229 resp->data_tx_completion.num_msdus);
1231 for (i = 0; i < resp->data_tx_completion.num_msdus; i++) {
1232 msdu_id = resp->data_tx_completion.msdus[i];
1233 tx_done.msdu_id = __le16_to_cpu(msdu_id);
1234 ath10k_txrx_tx_unref(htt, &tx_done);
1238 void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
1240 struct ath10k_htt *htt = &ar->htt;
1241 struct htt_resp *resp = (struct htt_resp *)skb->data;
1243 /* confirm alignment */
1244 if (!IS_ALIGNED((unsigned long)skb->data, 4))
1245 ath10k_warn("unaligned htt message, expect trouble\n");
1247 ath10k_dbg(ATH10K_DBG_HTT, "htt rx, msg_type: 0x%0X\n",
1248 resp->hdr.msg_type);
1249 switch (resp->hdr.msg_type) {
1250 case HTT_T2H_MSG_TYPE_VERSION_CONF: {
1251 htt->target_version_major = resp->ver_resp.major;
1252 htt->target_version_minor = resp->ver_resp.minor;
1253 complete(&htt->target_version_received);
1256 case HTT_T2H_MSG_TYPE_RX_IND:
1257 spin_lock_bh(&htt->rx_ring.lock);
1258 __skb_queue_tail(&htt->rx_compl_q, skb);
1259 spin_unlock_bh(&htt->rx_ring.lock);
1260 tasklet_schedule(&htt->txrx_compl_task);
1262 case HTT_T2H_MSG_TYPE_PEER_MAP: {
1263 struct htt_peer_map_event ev = {
1264 .vdev_id = resp->peer_map.vdev_id,
1265 .peer_id = __le16_to_cpu(resp->peer_map.peer_id),
1267 memcpy(ev.addr, resp->peer_map.addr, sizeof(ev.addr));
1268 ath10k_peer_map_event(htt, &ev);
1271 case HTT_T2H_MSG_TYPE_PEER_UNMAP: {
1272 struct htt_peer_unmap_event ev = {
1273 .peer_id = __le16_to_cpu(resp->peer_unmap.peer_id),
1275 ath10k_peer_unmap_event(htt, &ev);
1278 case HTT_T2H_MSG_TYPE_MGMT_TX_COMPLETION: {
1279 struct htt_tx_done tx_done = {};
1280 int status = __le32_to_cpu(resp->mgmt_tx_completion.status);
1283 __le32_to_cpu(resp->mgmt_tx_completion.desc_id);
1286 case HTT_MGMT_TX_STATUS_OK:
1288 case HTT_MGMT_TX_STATUS_RETRY:
1289 tx_done.no_ack = true;
1291 case HTT_MGMT_TX_STATUS_DROP:
1292 tx_done.discard = true;
1296 spin_lock_bh(&htt->tx_lock);
1297 ath10k_txrx_tx_unref(htt, &tx_done);
1298 spin_unlock_bh(&htt->tx_lock);
1301 case HTT_T2H_MSG_TYPE_TX_COMPL_IND:
1302 spin_lock_bh(&htt->tx_lock);
1303 __skb_queue_tail(&htt->tx_compl_q, skb);
1304 spin_unlock_bh(&htt->tx_lock);
1305 tasklet_schedule(&htt->txrx_compl_task);
1307 case HTT_T2H_MSG_TYPE_SEC_IND: {
1308 struct ath10k *ar = htt->ar;
1309 struct htt_security_indication *ev = &resp->security_indication;
1311 ath10k_dbg(ATH10K_DBG_HTT,
1312 "sec ind peer_id %d unicast %d type %d\n",
1313 __le16_to_cpu(ev->peer_id),
1314 !!(ev->flags & HTT_SECURITY_IS_UNICAST),
1315 MS(ev->flags, HTT_SECURITY_TYPE));
1316 complete(&ar->install_key_done);
1319 case HTT_T2H_MSG_TYPE_RX_FRAG_IND: {
1320 ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
1321 skb->data, skb->len);
1322 ath10k_htt_rx_frag_handler(htt, &resp->rx_frag_ind);
1325 case HTT_T2H_MSG_TYPE_TEST:
1328 case HTT_T2H_MSG_TYPE_STATS_CONF:
1329 trace_ath10k_htt_stats(skb->data, skb->len);
1331 case HTT_T2H_MSG_TYPE_TX_INSPECT_IND:
1332 case HTT_T2H_MSG_TYPE_RX_ADDBA:
1333 case HTT_T2H_MSG_TYPE_RX_DELBA:
1334 case HTT_T2H_MSG_TYPE_RX_FLUSH:
1336 ath10k_dbg(ATH10K_DBG_HTT, "htt event (%d) not handled\n",
1337 resp->hdr.msg_type);
1338 ath10k_dbg_dump(ATH10K_DBG_HTT_DUMP, NULL, "htt event: ",
1339 skb->data, skb->len);
1343 /* Free the indication buffer */
1344 dev_kfree_skb_any(skb);
1347 static void ath10k_htt_txrx_compl_task(unsigned long ptr)
1349 struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
1350 struct htt_resp *resp;
1351 struct sk_buff *skb;
1353 spin_lock_bh(&htt->tx_lock);
1354 while ((skb = __skb_dequeue(&htt->tx_compl_q))) {
1355 ath10k_htt_rx_frm_tx_compl(htt->ar, skb);
1356 dev_kfree_skb_any(skb);
1358 spin_unlock_bh(&htt->tx_lock);
1360 spin_lock_bh(&htt->rx_ring.lock);
1361 while ((skb = __skb_dequeue(&htt->rx_compl_q))) {
1362 resp = (struct htt_resp *)skb->data;
1363 ath10k_htt_rx_handler(htt, &resp->rx_ind);
1364 dev_kfree_skb_any(skb);
1366 spin_unlock_bh(&htt->rx_ring.lock);