ath10k: implement rx reorder support
authorMichal Kazior <michal.kazior@tieto.com>
Sat, 24 Jan 2015 10:14:48 +0000 (12:14 +0200)
committerKalle Valo <kvalo@qca.qualcomm.com>
Tue, 27 Jan 2015 13:55:58 +0000 (15:55 +0200)
New firmware and firmware (qca6174 hw3.0+ and fw
266+) are capable of full aggregation rx
reordering. If it's enabled then Rx is handled via
a new, separate htt event.

The rx ring behaviour is changed a little to
support the new rx scheme. These changes shouldn't
affect qca988x performance.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
drivers/net/wireless/ath/ath10k/core.c
drivers/net/wireless/ath/ath10k/core.h
drivers/net/wireless/ath/ath10k/htt.h
drivers/net/wireless/ath/ath10k/htt_rx.c
drivers/net/wireless/ath/ath10k/wmi-tlv.c

index a5465752a3ff19711f9a3527c7f14492d34a87d1..6860afbe68d97d6b304471d0f7f7f0ad9b2faaa9 100644 (file)
@@ -1061,6 +1061,18 @@ int ath10k_core_start(struct ath10k *ar, enum ath10k_firmware_mode mode)
                goto err_hif_stop;
        }
 
+       /* If firmware indicates Full Rx Reorder support it must be used in a
+        * slightly different manner. Let HTT code know.
+        */
+       ar->htt.rx_ring.in_ord_rx = !!(test_bit(WMI_SERVICE_RX_FULL_REORDER,
+                                               ar->wmi.svc_map));
+
+       status = ath10k_htt_rx_ring_refill(ar);
+       if (status) {
+               ath10k_err(ar, "failed to refill htt rx ring: %d\n", status);
+               goto err_hif_stop;
+       }
+
        /* we don't care about HTT in UTF mode */
        if (mode == ATH10K_FIRMWARE_MODE_NORMAL) {
                status = ath10k_htt_setup(&ar->htt);
index 739d9d69cf1cc75e047d620371ee343a4c02a477..774d8ceb405352c1e076ec7daade709b9e596a8b 100644 (file)
@@ -99,6 +99,7 @@ struct ath10k_skb_cb {
 
 struct ath10k_skb_rxcb {
        dma_addr_t paddr;
+       struct hlist_node hlist;
 };
 
 static inline struct ath10k_skb_cb *ATH10K_SKB_CB(struct sk_buff *skb)
@@ -114,6 +115,9 @@ static inline struct ath10k_skb_rxcb *ATH10K_SKB_RXCB(struct sk_buff *skb)
        return (struct ath10k_skb_rxcb *)skb->cb;
 }
 
+#define ATH10K_RXCB_SKB(rxcb) \
+               container_of((void *)rxcb, struct sk_buff, cb)
+
 static inline u32 host_interest_item_address(u32 item_offset)
 {
        return QCA988X_HOST_INTEREST_ADDRESS + item_offset;
index 8809b37e1912551ad5c946309b3f91f732b22655..d1f6eb287a10929702db563e9a28ac03190e915f 100644 (file)
@@ -21,6 +21,7 @@
 #include <linux/bug.h>
 #include <linux/interrupt.h>
 #include <linux/dmapool.h>
+#include <linux/hashtable.h>
 #include <net/mac80211.h>
 
 #include "htc.h"
@@ -286,7 +287,19 @@ enum htt_t2h_msg_type {
        HTT_T2H_MSG_TYPE_RC_UPDATE_IND          = 0xc,
        HTT_T2H_MSG_TYPE_TX_INSPECT_IND         = 0xd,
        HTT_T2H_MSG_TYPE_MGMT_TX_COMPLETION     = 0xe,
+       HTT_T2H_MSG_TYPE_TX_CREDIT_UPDATE_IND   = 0xf,
+       HTT_T2H_MSG_TYPE_RX_PN_IND              = 0x10,
+       HTT_T2H_MSG_TYPE_RX_OFFLOAD_DELIVER_IND = 0x11,
+       HTT_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND    = 0x12,
+       /* 0x13 reservd */
+       HTT_T2H_MSG_TYPE_WDI_IPA_OP_RESPONSE    = 0x14,
+
+       /* FIXME: Do not depend on this event id. Numbering of this event id is
+        * broken across different firmware revisions and HTT version fails to
+        * indicate this.
+        */
        HTT_T2H_MSG_TYPE_TEST,
+
        /* keep this last */
        HTT_T2H_NUM_MSGS
 };
@@ -655,6 +668,53 @@ struct htt_rx_fragment_indication {
 #define HTT_RX_FRAG_IND_INFO1_FLUSH_SEQ_NUM_END_MASK   0x00000FC0
 #define HTT_RX_FRAG_IND_INFO1_FLUSH_SEQ_NUM_END_LSB    6
 
+struct htt_rx_pn_ind {
+       __le16 peer_id;
+       u8 tid;
+       u8 seqno_start;
+       u8 seqno_end;
+       u8 pn_ie_count;
+       u8 reserved;
+       u8 pn_ies[0];
+} __packed;
+
+struct htt_rx_offload_msdu {
+       __le16 msdu_len;
+       __le16 peer_id;
+       u8 vdev_id;
+       u8 tid;
+       u8 fw_desc;
+       u8 payload[0];
+} __packed;
+
+struct htt_rx_offload_ind {
+       u8 reserved;
+       __le16 msdu_count;
+} __packed;
+
+struct htt_rx_in_ord_msdu_desc {
+       __le32 msdu_paddr;
+       __le16 msdu_len;
+       u8 fw_desc;
+       u8 reserved;
+} __packed;
+
+struct htt_rx_in_ord_ind {
+       u8 info;
+       __le16 peer_id;
+       u8 vdev_id;
+       u8 reserved;
+       __le16 msdu_count;
+       struct htt_rx_in_ord_msdu_desc msdu_descs[0];
+} __packed;
+
+#define HTT_RX_IN_ORD_IND_INFO_TID_MASK                0x0000001f
+#define HTT_RX_IN_ORD_IND_INFO_TID_LSB         0
+#define HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK    0x00000020
+#define HTT_RX_IN_ORD_IND_INFO_OFFLOAD_LSB     5
+#define HTT_RX_IN_ORD_IND_INFO_FRAG_MASK       0x00000040
+#define HTT_RX_IN_ORD_IND_INFO_FRAG_LSB                6
+
 /*
  * target -> host test message definition
  *
@@ -1150,6 +1210,9 @@ struct htt_resp {
                struct htt_rx_test rx_test;
                struct htt_pktlog_msg pktlog_msg;
                struct htt_stats_conf stats_conf;
+               struct htt_rx_pn_ind rx_pn_ind;
+               struct htt_rx_offload_ind rx_offload_ind;
+               struct htt_rx_in_ord_ind rx_in_ord_ind;
        };
 } __packed;
 
@@ -1197,6 +1260,20 @@ struct ath10k_htt {
                 * filled.
                 */
                struct sk_buff **netbufs_ring;
+
+               /* This is used only with firmware supporting IN_ORD_IND.
+                *
+                * With Full Rx Reorder the HTT Rx Ring is more of a temporary
+                * buffer ring from which buffer addresses are copied by the
+                * firmware to MAC Rx ring. Firmware then delivers IN_ORD_IND
+                * pointing to specific (re-ordered) buffers.
+                *
+                * FIXME: With kernel generic hashing functions there's a lot
+                * of hash collisions for sk_buffs.
+                */
+               bool in_ord_rx;
+               DECLARE_HASHTABLE(skb_table, 4);
+
                /*
                 * Ring of buffer addresses -
                 * This ring holds the "physical" device address of the
@@ -1270,6 +1347,7 @@ struct ath10k_htt {
        struct tasklet_struct txrx_compl_task;
        struct sk_buff_head tx_compl_q;
        struct sk_buff_head rx_compl_q;
+       struct sk_buff_head rx_in_ord_compl_q;
 
        /* rx_status template */
        struct ieee80211_rx_status rx_status;
@@ -1333,6 +1411,7 @@ int ath10k_htt_tx_alloc(struct ath10k_htt *htt);
 void ath10k_htt_tx_free(struct ath10k_htt *htt);
 
 int ath10k_htt_rx_alloc(struct ath10k_htt *htt);
+int ath10k_htt_rx_ring_refill(struct ath10k *ar);
 void ath10k_htt_rx_free(struct ath10k_htt *htt);
 
 void ath10k_htt_htc_tx_complete(struct ath10k *ar, struct sk_buff *skb);
index 2e55ed7241aef8d3c5023e9ee1b0add36a3a71b8..661785fb99061f4b8468b1411a96308e48c8396d 100644 (file)
@@ -25,8 +25,8 @@
 
 #include <linux/log2.h>
 
-#define HTT_RX_RING_SIZE 1024
-#define HTT_RX_RING_FILL_LEVEL 1000
+#define HTT_RX_RING_SIZE HTT_RX_RING_SIZE_MAX
+#define HTT_RX_RING_FILL_LEVEL (((HTT_RX_RING_SIZE) / 2) - 1)
 
 /* when under memory pressure rx ring refill may fail and needs a retry */
 #define HTT_RX_RING_REFILL_RETRY_MS 50
 static int ath10k_htt_rx_get_csum_state(struct sk_buff *skb);
 static void ath10k_htt_txrx_compl_task(unsigned long ptr);
 
+static struct sk_buff *
+ath10k_htt_rx_find_skb_paddr(struct ath10k *ar, u32 paddr)
+{
+       struct ath10k_skb_rxcb *rxcb;
+
+       hash_for_each_possible(ar->htt.rx_ring.skb_table, rxcb, hlist, paddr)
+               if (rxcb->paddr == paddr)
+                       return ATH10K_RXCB_SKB(rxcb);
+
+       WARN_ON_ONCE(1);
+       return NULL;
+}
+
 static void ath10k_htt_rx_ring_free(struct ath10k_htt *htt)
 {
        struct sk_buff *skb;
-       struct ath10k_skb_rxcb *cb;
+       struct ath10k_skb_rxcb *rxcb;
+       struct hlist_node *n;
        int i;
 
-       for (i = 0; i < htt->rx_ring.fill_cnt; i++) {
-               skb = htt->rx_ring.netbufs_ring[i];
-               cb = ATH10K_SKB_RXCB(skb);
-               dma_unmap_single(htt->ar->dev, cb->paddr,
-                                skb->len + skb_tailroom(skb),
-                                DMA_FROM_DEVICE);
-               dev_kfree_skb_any(skb);
+       if (htt->rx_ring.in_ord_rx) {
+               hash_for_each_safe(htt->rx_ring.skb_table, i, n, rxcb, hlist) {
+                       skb = ATH10K_RXCB_SKB(rxcb);
+                       dma_unmap_single(htt->ar->dev, rxcb->paddr,
+                                        skb->len + skb_tailroom(skb),
+                                        DMA_FROM_DEVICE);
+                       hash_del(&rxcb->hlist);
+                       dev_kfree_skb_any(skb);
+               }
+       } else {
+               for (i = 0; i < htt->rx_ring.size; i++) {
+                       skb = htt->rx_ring.netbufs_ring[i];
+                       if (!skb)
+                               continue;
+
+                       rxcb = ATH10K_SKB_RXCB(skb);
+                       dma_unmap_single(htt->ar->dev, rxcb->paddr,
+                                        skb->len + skb_tailroom(skb),
+                                        DMA_FROM_DEVICE);
+                       dev_kfree_skb_any(skb);
+               }
        }
 
        htt->rx_ring.fill_cnt = 0;
+       hash_init(htt->rx_ring.skb_table);
+       memset(htt->rx_ring.netbufs_ring, 0,
+              htt->rx_ring.size * sizeof(htt->rx_ring.netbufs_ring[0]));
 }
 
 static int __ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
 {
        struct htt_rx_desc *rx_desc;
+       struct ath10k_skb_rxcb *rxcb;
        struct sk_buff *skb;
        dma_addr_t paddr;
        int ret = 0, idx;
 
+       /* The Full Rx Reorder firmware has no way of telling the host
+        * implicitly when it copied HTT Rx Ring buffers to MAC Rx Ring.
+        * To keep things simple make sure ring is always half empty. This
+        * guarantees there'll be no replenishment overruns possible.
+        */
+       BUILD_BUG_ON(HTT_RX_RING_FILL_LEVEL >= HTT_RX_RING_SIZE / 2);
+
        idx = __le32_to_cpu(*htt->rx_ring.alloc_idx.vaddr);
        while (num > 0) {
                skb = dev_alloc_skb(HTT_RX_BUF_SIZE + HTT_RX_DESC_ALIGN);
@@ -86,11 +125,18 @@ static int __ath10k_htt_rx_ring_fill_n(struct ath10k_htt *htt, int num)
                        goto fail;
                }
 
-               ATH10K_SKB_RXCB(skb)->paddr = paddr;
+               rxcb = ATH10K_SKB_RXCB(skb);
+               rxcb->paddr = paddr;
                htt->rx_ring.netbufs_ring[idx] = skb;
                htt->rx_ring.paddrs_ring[idx] = __cpu_to_le32(paddr);
                htt->rx_ring.fill_cnt++;
 
+               if (htt->rx_ring.in_ord_rx) {
+                       hash_add(htt->rx_ring.skb_table,
+                                &ATH10K_SKB_RXCB(skb)->hlist,
+                                (u32)paddr);
+               }
+
                num--;
                idx++;
                idx &= htt->rx_ring.size_mask;
@@ -158,22 +204,20 @@ static void ath10k_htt_rx_ring_refill_retry(unsigned long arg)
        ath10k_htt_rx_msdu_buff_replenish(htt);
 }
 
-static void ath10k_htt_rx_ring_clean_up(struct ath10k_htt *htt)
+int ath10k_htt_rx_ring_refill(struct ath10k *ar)
 {
-       struct sk_buff *skb;
-       int i;
+       struct ath10k_htt *htt = &ar->htt;
+       int ret;
 
-       for (i = 0; i < htt->rx_ring.size; i++) {
-               skb = htt->rx_ring.netbufs_ring[i];
-               if (!skb)
-                       continue;
+       spin_lock_bh(&htt->rx_ring.lock);
+       ret = ath10k_htt_rx_ring_fill_n(htt, (htt->rx_ring.fill_level -
+                                             htt->rx_ring.fill_cnt));
+       spin_unlock_bh(&htt->rx_ring.lock);
 
-               dma_unmap_single(htt->ar->dev, ATH10K_SKB_RXCB(skb)->paddr,
-                                skb->len + skb_tailroom(skb),
-                                DMA_FROM_DEVICE);
-               dev_kfree_skb_any(skb);
-               htt->rx_ring.netbufs_ring[i] = NULL;
-       }
+       if (ret)
+               ath10k_htt_rx_ring_free(htt);
+
+       return ret;
 }
 
 void ath10k_htt_rx_free(struct ath10k_htt *htt)
@@ -184,8 +228,9 @@ void ath10k_htt_rx_free(struct ath10k_htt *htt)
 
        skb_queue_purge(&htt->tx_compl_q);
        skb_queue_purge(&htt->rx_compl_q);
+       skb_queue_purge(&htt->rx_in_ord_compl_q);
 
-       ath10k_htt_rx_ring_clean_up(htt);
+       ath10k_htt_rx_ring_free(htt);
 
        dma_free_coherent(htt->ar->dev,
                          (htt->rx_ring.size *
@@ -217,6 +262,7 @@ static inline struct sk_buff *ath10k_htt_rx_netbuf_pop(struct ath10k_htt *htt)
        idx = htt->rx_ring.sw_rd_idx.msdu_payld;
        msdu = htt->rx_ring.netbufs_ring[idx];
        htt->rx_ring.netbufs_ring[idx] = NULL;
+       htt->rx_ring.paddrs_ring[idx] = 0;
 
        idx++;
        idx &= htt->rx_ring.size_mask;
@@ -384,6 +430,82 @@ static void ath10k_htt_rx_replenish_task(unsigned long ptr)
        ath10k_htt_rx_msdu_buff_replenish(htt);
 }
 
+static struct sk_buff *ath10k_htt_rx_pop_paddr(struct ath10k_htt *htt,
+                                              u32 paddr)
+{
+       struct ath10k *ar = htt->ar;
+       struct ath10k_skb_rxcb *rxcb;
+       struct sk_buff *msdu;
+
+       lockdep_assert_held(&htt->rx_ring.lock);
+
+       msdu = ath10k_htt_rx_find_skb_paddr(ar, paddr);
+       if (!msdu)
+               return NULL;
+
+       rxcb = ATH10K_SKB_RXCB(msdu);
+       hash_del(&rxcb->hlist);
+       htt->rx_ring.fill_cnt--;
+
+       dma_unmap_single(htt->ar->dev, rxcb->paddr,
+                        msdu->len + skb_tailroom(msdu),
+                        DMA_FROM_DEVICE);
+       ath10k_dbg_dump(ar, ATH10K_DBG_HTT_DUMP, NULL, "htt rx netbuf pop: ",
+                       msdu->data, msdu->len + skb_tailroom(msdu));
+
+       return msdu;
+}
+
+static int ath10k_htt_rx_pop_paddr_list(struct ath10k_htt *htt,
+                                       struct htt_rx_in_ord_ind *ev,
+                                       struct sk_buff_head *list)
+{
+       struct ath10k *ar = htt->ar;
+       struct htt_rx_in_ord_msdu_desc *msdu_desc = ev->msdu_descs;
+       struct htt_rx_desc *rxd;
+       struct sk_buff *msdu;
+       int msdu_count;
+       bool is_offload;
+       u32 paddr;
+
+       lockdep_assert_held(&htt->rx_ring.lock);
+
+       msdu_count = __le16_to_cpu(ev->msdu_count);
+       is_offload = !!(ev->info & HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK);
+
+       while (msdu_count--) {
+               paddr = __le32_to_cpu(msdu_desc->msdu_paddr);
+
+               msdu = ath10k_htt_rx_pop_paddr(htt, paddr);
+               if (!msdu) {
+                       __skb_queue_purge(list);
+                       return -ENOENT;
+               }
+
+               __skb_queue_tail(list, msdu);
+
+               if (!is_offload) {
+                       rxd = (void *)msdu->data;
+
+                       trace_ath10k_htt_rx_desc(ar, rxd, sizeof(*rxd));
+
+                       skb_put(msdu, sizeof(*rxd));
+                       skb_pull(msdu, sizeof(*rxd));
+                       skb_put(msdu, __le16_to_cpu(msdu_desc->msdu_len));
+
+                       if (!(__le32_to_cpu(rxd->attention.flags) &
+                             RX_ATTENTION_FLAGS_MSDU_DONE)) {
+                               ath10k_warn(htt->ar, "tried to pop an incomplete frame, oops!\n");
+                               return -EIO;
+                       }
+               }
+
+               msdu_desc++;
+       }
+
+       return 0;
+}
+
 int ath10k_htt_rx_alloc(struct ath10k_htt *htt)
 {
        struct ath10k *ar = htt->ar;
@@ -429,7 +551,7 @@ int ath10k_htt_rx_alloc(struct ath10k_htt *htt)
 
        htt->rx_ring.alloc_idx.vaddr = vaddr;
        htt->rx_ring.alloc_idx.paddr = paddr;
-       htt->rx_ring.sw_rd_idx.msdu_payld = 0;
+       htt->rx_ring.sw_rd_idx.msdu_payld = htt->rx_ring.size_mask;
        *htt->rx_ring.alloc_idx.vaddr = 0;
 
        /* Initialize the Rx refill retry timer */
@@ -438,14 +560,15 @@ int ath10k_htt_rx_alloc(struct ath10k_htt *htt)
        spin_lock_init(&htt->rx_ring.lock);
 
        htt->rx_ring.fill_cnt = 0;
-       if (__ath10k_htt_rx_ring_fill_n(htt, htt->rx_ring.fill_level))
-               goto err_fill_ring;
+       htt->rx_ring.sw_rd_idx.msdu_payld = 0;
+       hash_init(htt->rx_ring.skb_table);
 
        tasklet_init(&htt->rx_replenish_task, ath10k_htt_rx_replenish_task,
                     (unsigned long)htt);
 
        skb_queue_head_init(&htt->tx_compl_q);
        skb_queue_head_init(&htt->rx_compl_q);
+       skb_queue_head_init(&htt->rx_in_ord_compl_q);
 
        tasklet_init(&htt->txrx_compl_task, ath10k_htt_txrx_compl_task,
                     (unsigned long)htt);
@@ -454,12 +577,6 @@ int ath10k_htt_rx_alloc(struct ath10k_htt *htt)
                   htt->rx_ring.size, htt->rx_ring.fill_level);
        return 0;
 
-err_fill_ring:
-       ath10k_htt_rx_ring_free(htt);
-       dma_free_coherent(htt->ar->dev,
-                         sizeof(*htt->rx_ring.alloc_idx.vaddr),
-                         htt->rx_ring.alloc_idx.vaddr,
-                         htt->rx_ring.alloc_idx.paddr);
 err_dma_idx:
        dma_free_coherent(htt->ar->dev,
                          (htt->rx_ring.size *
@@ -1583,6 +1700,194 @@ static void ath10k_htt_rx_delba(struct ath10k *ar, struct htt_resp *resp)
        spin_unlock_bh(&ar->data_lock);
 }
 
+static int ath10k_htt_rx_extract_amsdu(struct sk_buff_head *list,
+                                      struct sk_buff_head *amsdu)
+{
+       struct sk_buff *msdu;
+       struct htt_rx_desc *rxd;
+
+       if (skb_queue_empty(list))
+               return -ENOBUFS;
+
+       if (WARN_ON(!skb_queue_empty(amsdu)))
+               return -EINVAL;
+
+       while ((msdu = __skb_dequeue(list))) {
+               __skb_queue_tail(amsdu, msdu);
+
+               rxd = (void *)msdu->data - sizeof(*rxd);
+               if (rxd->msdu_end.info0 &
+                   __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))
+                       break;
+       }
+
+       msdu = skb_peek_tail(amsdu);
+       rxd = (void *)msdu->data - sizeof(*rxd);
+       if (!(rxd->msdu_end.info0 &
+             __cpu_to_le32(RX_MSDU_END_INFO0_LAST_MSDU))) {
+               skb_queue_splice_init(amsdu, list);
+               return -EAGAIN;
+       }
+
+       return 0;
+}
+
+static void ath10k_htt_rx_h_rx_offload_prot(struct ieee80211_rx_status *status,
+                                           struct sk_buff *skb)
+{
+       struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+
+       if (!ieee80211_has_protected(hdr->frame_control))
+               return;
+
+       /* Offloaded frames are already decrypted but firmware insists they are
+        * protected in the 802.11 header. Strip the flag.  Otherwise mac80211
+        * will drop the frame.
+        */
+
+       hdr->frame_control &= ~__cpu_to_le16(IEEE80211_FCTL_PROTECTED);
+       status->flag |= RX_FLAG_DECRYPTED |
+                       RX_FLAG_IV_STRIPPED |
+                       RX_FLAG_MMIC_STRIPPED;
+}
+
+static void ath10k_htt_rx_h_rx_offload(struct ath10k *ar,
+                                      struct sk_buff_head *list)
+{
+       struct ath10k_htt *htt = &ar->htt;
+       struct ieee80211_rx_status *status = &htt->rx_status;
+       struct htt_rx_offload_msdu *rx;
+       struct sk_buff *msdu;
+       size_t offset;
+
+       while ((msdu = __skb_dequeue(list))) {
+               /* Offloaded frames don't have Rx descriptor. Instead they have
+                * a short meta information header.
+                */
+
+               rx = (void *)msdu->data;
+
+               skb_put(msdu, sizeof(*rx));
+               skb_pull(msdu, sizeof(*rx));
+
+               if (skb_tailroom(msdu) < __le16_to_cpu(rx->msdu_len)) {
+                       ath10k_warn(ar, "dropping frame: offloaded rx msdu is too long!\n");
+                       dev_kfree_skb_any(msdu);
+                       continue;
+               }
+
+               skb_put(msdu, __le16_to_cpu(rx->msdu_len));
+
+               /* Offloaded rx header length isn't multiple of 2 nor 4 so the
+                * actual payload is unaligned. Align the frame.  Otherwise
+                * mac80211 complains.  This shouldn't reduce performance much
+                * because these offloaded frames are rare.
+                */
+               offset = 4 - ((unsigned long)msdu->data & 3);
+               skb_put(msdu, offset);
+               memmove(msdu->data + offset, msdu->data, msdu->len);
+               skb_pull(msdu, offset);
+
+               /* FIXME: The frame is NWifi. Re-construct QoS Control
+                * if possible later.
+                */
+
+               memset(status, 0, sizeof(*status));
+               status->flag |= RX_FLAG_NO_SIGNAL_VAL;
+
+               ath10k_htt_rx_h_rx_offload_prot(status, msdu);
+               ath10k_htt_rx_h_channel(ar, status);
+               ath10k_process_rx(ar, status, msdu);
+       }
+}
+
+static void ath10k_htt_rx_in_ord_ind(struct ath10k *ar, struct sk_buff *skb)
+{
+       struct ath10k_htt *htt = &ar->htt;
+       struct htt_resp *resp = (void *)skb->data;
+       struct ieee80211_rx_status *status = &htt->rx_status;
+       struct sk_buff_head list;
+       struct sk_buff_head amsdu;
+       u16 peer_id;
+       u16 msdu_count;
+       u8 vdev_id;
+       u8 tid;
+       bool offload;
+       bool frag;
+       int ret;
+
+       lockdep_assert_held(&htt->rx_ring.lock);
+
+       if (htt->rx_confused)
+               return;
+
+       skb_pull(skb, sizeof(resp->hdr));
+       skb_pull(skb, sizeof(resp->rx_in_ord_ind));
+
+       peer_id = __le16_to_cpu(resp->rx_in_ord_ind.peer_id);
+       msdu_count = __le16_to_cpu(resp->rx_in_ord_ind.msdu_count);
+       vdev_id = resp->rx_in_ord_ind.vdev_id;
+       tid = SM(resp->rx_in_ord_ind.info, HTT_RX_IN_ORD_IND_INFO_TID);
+       offload = !!(resp->rx_in_ord_ind.info &
+                       HTT_RX_IN_ORD_IND_INFO_OFFLOAD_MASK);
+       frag = !!(resp->rx_in_ord_ind.info & HTT_RX_IN_ORD_IND_INFO_FRAG_MASK);
+
+       ath10k_dbg(ar, ATH10K_DBG_HTT,
+                  "htt rx in ord vdev %i peer %i tid %i offload %i frag %i msdu count %i\n",
+                  vdev_id, peer_id, tid, offload, frag, msdu_count);
+
+       if (skb->len < msdu_count * sizeof(*resp->rx_in_ord_ind.msdu_descs)) {
+               ath10k_warn(ar, "dropping invalid in order rx indication\n");
+               return;
+       }
+
+       /* The event can deliver more than 1 A-MSDU. Each A-MSDU is later
+        * extracted and processed.
+        */
+       __skb_queue_head_init(&list);
+       ret = ath10k_htt_rx_pop_paddr_list(htt, &resp->rx_in_ord_ind, &list);
+       if (ret < 0) {
+               ath10k_warn(ar, "failed to pop paddr list: %d\n", ret);
+               htt->rx_confused = true;
+               return;
+       }
+
+       /* Offloaded frames are very different and need to be handled
+        * separately.
+        */
+       if (offload)
+               ath10k_htt_rx_h_rx_offload(ar, &list);
+
+       while (!skb_queue_empty(&list)) {
+               __skb_queue_head_init(&amsdu);
+               ret = ath10k_htt_rx_extract_amsdu(&list, &amsdu);
+               switch (ret) {
+               case 0:
+                       /* Note: The in-order indication may report interleaved
+                        * frames from different PPDUs meaning reported rx rate
+                        * to mac80211 isn't accurate/reliable. It's still
+                        * better to report something than nothing though. This
+                        * should still give an idea about rx rate to the user.
+                        */
+                       ath10k_htt_rx_h_ppdu(ar, &amsdu, status);
+                       ath10k_htt_rx_h_filter(ar, &amsdu, status);
+                       ath10k_htt_rx_h_mpdu(ar, &amsdu, status);
+                       ath10k_htt_rx_h_deliver(ar, &amsdu, status);
+                       break;
+               case -EAGAIN:
+                       /* fall through */
+               default:
+                       /* Should not happen. */
+                       ath10k_warn(ar, "failed to extract amsdu: %d\n", ret);
+                       htt->rx_confused = true;
+                       __skb_queue_purge(&list);
+                       return;
+               }
+       }
+
+       tasklet_schedule(&htt->rx_replenish_task);
+}
+
 void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
 {
        struct ath10k_htt *htt = &ar->htt;
@@ -1705,6 +2010,20 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
                 */
                break;
        }
+       case HTT_T2H_MSG_TYPE_RX_IN_ORD_PADDR_IND: {
+               spin_lock_bh(&htt->rx_ring.lock);
+               __skb_queue_tail(&htt->rx_in_ord_compl_q, skb);
+               spin_unlock_bh(&htt->rx_ring.lock);
+               tasklet_schedule(&htt->txrx_compl_task);
+               return;
+       }
+       case HTT_T2H_MSG_TYPE_TX_CREDIT_UPDATE_IND:
+               /* FIXME: This WMI-TLV event is overlapping with 10.2
+                * CHAN_CHANGE - both being 0xF. Neither is being used in
+                * practice so no immediate action is necessary. Nevertheless
+                * HTT may need an abstraction layer like WMI has one day.
+                */
+               break;
        default:
                ath10k_warn(ar, "htt event (%d) not handled\n",
                            resp->hdr.msg_type);
@@ -1720,6 +2039,7 @@ void ath10k_htt_t2h_msg_handler(struct ath10k *ar, struct sk_buff *skb)
 static void ath10k_htt_txrx_compl_task(unsigned long ptr)
 {
        struct ath10k_htt *htt = (struct ath10k_htt *)ptr;
+       struct ath10k *ar = htt->ar;
        struct htt_resp *resp;
        struct sk_buff *skb;
 
@@ -1736,5 +2056,10 @@ static void ath10k_htt_txrx_compl_task(unsigned long ptr)
                ath10k_htt_rx_handler(htt, &resp->rx_ind);
                dev_kfree_skb_any(skb);
        }
+
+       while ((skb = __skb_dequeue(&htt->rx_in_ord_compl_q))) {
+               ath10k_htt_rx_in_ord_ind(ar, skb);
+               dev_kfree_skb_any(skb);
+       }
        spin_unlock_bh(&htt->rx_ring.lock);
 }
index a6c634a585ddb3f0682c9fdbefb3b5c3a82ec386..35f519123752c2d74ac41dcb8e007cca9e24cc54 100644 (file)
@@ -1052,8 +1052,15 @@ static struct sk_buff *ath10k_wmi_tlv_op_gen_init(struct ath10k *ar)
 
        cfg->num_vdevs = __cpu_to_le32(TARGET_TLV_NUM_VDEVS);
        cfg->num_peers = __cpu_to_le32(TARGET_TLV_NUM_PEERS);
-       cfg->num_offload_peers = __cpu_to_le32(0);
-       cfg->num_offload_reorder_bufs = __cpu_to_le32(0);
+
+       if (test_bit(WMI_SERVICE_RX_FULL_REORDER, ar->wmi.svc_map)) {
+               cfg->num_offload_peers = __cpu_to_le32(3);
+               cfg->num_offload_reorder_bufs = __cpu_to_le32(3);
+       } else {
+               cfg->num_offload_peers = __cpu_to_le32(0);
+               cfg->num_offload_reorder_bufs = __cpu_to_le32(0);
+       }
+
        cfg->num_peer_keys = __cpu_to_le32(2);
        cfg->num_tids = __cpu_to_le32(TARGET_TLV_NUM_TIDS);
        cfg->ast_skid_limit = __cpu_to_le32(0x10);