diff options
author | Christian Lamparter <chunkeey@web.de> | 2009-06-06 03:07:23 (GMT) |
---|---|---|
committer | John W. Linville <linville@tuxdriver.com> | 2009-06-10 17:27:54 (GMT) |
commit | 9b9c5aaeedfda256ed77094303e2a7242c3290da (patch) | |
tree | 07dc845cc308d9a4b721bf59605217bc884f01b1 | |
parent | 2543a0c4c0fde46f9f206cec1e1cf951a2a63a66 (diff) | |
download | linux-9b9c5aaeedfda256ed77094303e2a7242c3290da.tar.xz |
ar9170: xmit code revamp
This patch is a back-port from aggregation testing code.
In the past, we didn't limit the amount of active tx urbs.
However, ar9170 only has a limited buffer reserved for
pending data frames.
This wasn't much of a problem with the slower 802.11b/g.
We simply stopped the full queue and moved on to something different
in the mean time. But - as you guessed it - this simple approach
stands in way for a decent aggregation implementation.
Signed-off-by: Christian Lamparter <chunkeey@web.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
-rw-r--r-- | drivers/net/wireless/ath/ar9170/ar9170.h | 34 | ||||
-rw-r--r-- | drivers/net/wireless/ath/ar9170/hw.h | 3 | ||||
-rw-r--r-- | drivers/net/wireless/ath/ar9170/main.c | 644 | ||||
-rw-r--r-- | drivers/net/wireless/ath/ar9170/usb.c | 122 | ||||
-rw-r--r-- | drivers/net/wireless/ath/ar9170/usb.h | 7 |
5 files changed, 550 insertions, 260 deletions
diff --git a/drivers/net/wireless/ath/ar9170/ar9170.h b/drivers/net/wireless/ath/ar9170/ar9170.h index c7cba66..bb97981 100644 --- a/drivers/net/wireless/ath/ar9170/ar9170.h +++ b/drivers/net/wireless/ath/ar9170/ar9170.h @@ -109,6 +109,11 @@ struct ar9170_rxstream_mpdu_merge { bool has_plcp; }; +#define AR9170_QUEUE_TIMEOUT 64 +#define AR9170_TX_TIMEOUT 8 +#define AR9170_JANITOR_DELAY 128 +#define AR9170_TX_INVALID_RATE 0xffffffff + struct ar9170 { struct ieee80211_hw *hw; struct mutex mutex; @@ -117,10 +122,11 @@ struct ar9170 { int (*open)(struct ar9170 *); void (*stop)(struct ar9170 *); - int (*tx)(struct ar9170 *, struct sk_buff *, bool, unsigned int); + int (*tx)(struct ar9170 *, struct sk_buff *); int (*exec_cmd)(struct ar9170 *, enum ar9170_cmd, u32 , void *, u32 , void *); void (*callback_cmd)(struct ar9170 *, u32 , void *); + int (*flush)(struct ar9170 *); /* interface mode settings */ struct ieee80211_vif *vif; @@ -177,10 +183,10 @@ struct ar9170 { struct ar9170_eeprom eeprom; struct ath_regulatory regulatory; - /* global tx status for unregistered Stations. */ - struct sk_buff_head global_tx_status; - struct sk_buff_head global_tx_status_waste; - struct delayed_work tx_status_janitor; + /* tx queues - as seen by hw - */ + struct sk_buff_head tx_pending[__AR9170_NUM_TXQ]; + struct sk_buff_head tx_status[__AR9170_NUM_TXQ]; + struct delayed_work tx_janitor; /* rxstream mpdu merge */ struct ar9170_rxstream_mpdu_merge rx_mpdu; @@ -189,11 +195,19 @@ struct ar9170 { }; struct ar9170_sta_info { - struct sk_buff_head tx_status[__AR9170_NUM_TXQ]; }; -#define IS_STARTED(a) (a->state >= AR9170_STARTED) -#define IS_ACCEPTING_CMD(a) (a->state >= AR9170_IDLE) +#define AR9170_TX_FLAG_WAIT_FOR_ACK BIT(0) +#define AR9170_TX_FLAG_NO_ACK BIT(1) +#define AR9170_TX_FLAG_BLOCK_ACK BIT(2) + +struct ar9170_tx_info { + unsigned long timeout; + unsigned int flags; +}; + +#define IS_STARTED(a) (((struct ar9170 *)a)->state >= AR9170_STARTED) +#define IS_ACCEPTING_CMD(a) (((struct ar9170 *)a)->state >= AR9170_IDLE) #define AR9170_FILTER_CHANGED_MODE BIT(0) #define AR9170_FILTER_CHANGED_MULTICAST BIT(1) @@ -204,9 +218,9 @@ void *ar9170_alloc(size_t priv_size); int ar9170_register(struct ar9170 *ar, struct device *pdev); void ar9170_rx(struct ar9170 *ar, struct sk_buff *skb); void ar9170_unregister(struct ar9170 *ar); -void ar9170_handle_tx_status(struct ar9170 *ar, struct sk_buff *skb, - bool update_statistics, u16 tx_status); +void ar9170_tx_callback(struct ar9170 *ar, struct sk_buff *skb); void ar9170_handle_command_response(struct ar9170 *ar, void *buf, u32 len); +int ar9170_nag_limiter(struct ar9170 *ar); /* MAC */ int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb); diff --git a/drivers/net/wireless/ath/ar9170/hw.h b/drivers/net/wireless/ath/ar9170/hw.h index 3c8004f..6cbfb2f 100644 --- a/drivers/net/wireless/ath/ar9170/hw.h +++ b/drivers/net/wireless/ath/ar9170/hw.h @@ -420,4 +420,7 @@ enum ar9170_txq { __AR9170_NUM_TXQ, }; +#define AR9170_TXQ_DEPTH 32 +#define AR9170_TX_MAX_PENDING 128 + #endif /* __AR9170_HW_H */ diff --git a/drivers/net/wireless/ath/ar9170/main.c b/drivers/net/wireless/ath/ar9170/main.c index de57aa9..9d38cf6 100644 --- a/drivers/net/wireless/ath/ar9170/main.c +++ b/drivers/net/wireless/ath/ar9170/main.c @@ -173,59 +173,122 @@ static struct ieee80211_supported_band ar9170_band_5GHz = { .ht_cap = AR9170_HT_CAP, }; -#ifdef AR9170_QUEUE_DEBUG -/* - * In case some wants works with AR9170's crazy tx_status queueing techniques. - * He might need this rather useful probing function. - * - * NOTE: caller must hold the queue's spinlock! - */ +static void ar9170_tx(struct ar9170 *ar); +#ifdef AR9170_QUEUE_DEBUG static void ar9170_print_txheader(struct ar9170 *ar, struct sk_buff *skb) { struct ar9170_tx_control *txc = (void *) skb->data; - struct ieee80211_hdr *hdr = (void *)txc->frame_data; + struct ieee80211_tx_info *txinfo = IEEE80211_SKB_CB(skb); + struct ar9170_tx_info *arinfo = (void *) txinfo->rate_driver_data; + struct ieee80211_hdr *hdr = (void *) txc->frame_data; - printk(KERN_DEBUG "%s: => FRAME [skb:%p, queue:%d, DA:[%pM] " - "mac_control:%04x, phy_control:%08x]\n", + printk(KERN_DEBUG "%s: => FRAME [skb:%p, q:%d, DA:[%pM] flags:%x " + "mac_ctrl:%04x, phy_ctrl:%08x, timeout:[%d ms]]\n", wiphy_name(ar->hw->wiphy), skb, skb_get_queue_mapping(skb), - ieee80211_get_DA(hdr), le16_to_cpu(txc->mac_control), - le32_to_cpu(txc->phy_control)); + ieee80211_get_DA(hdr), arinfo->flags, + le16_to_cpu(txc->mac_control), le32_to_cpu(txc->phy_control), + jiffies_to_msecs(arinfo->timeout - jiffies)); } -static void ar9170_dump_station_tx_status_queue(struct ar9170 *ar, - struct sk_buff_head *queue) +static void __ar9170_dump_txqueue(struct ar9170 *ar, + struct sk_buff_head *queue) { struct sk_buff *skb; int i = 0; printk(KERN_DEBUG "---[ cut here ]---\n"); - printk(KERN_DEBUG "%s: %d entries in tx_status queue.\n", + printk(KERN_DEBUG "%s: %d entries in queue.\n", wiphy_name(ar->hw->wiphy), skb_queue_len(queue)); skb_queue_walk(queue, skb) { - struct ar9170_tx_control *txc = (void *) skb->data; - struct ieee80211_hdr *hdr = (void *)txc->frame_data; - - printk(KERN_DEBUG "index:%d => \n", i); + printk(KERN_DEBUG "index:%d => \n", i++); ar9170_print_txheader(ar, skb); } + if (i != skb_queue_len(queue)) + printk(KERN_DEBUG "WARNING: queue frame counter " + "mismatch %d != %d\n", skb_queue_len(queue), i); printk(KERN_DEBUG "---[ end ]---\n"); } -#endif /* AR9170_QUEUE_DEBUG */ -void ar9170_handle_tx_status(struct ar9170 *ar, struct sk_buff *skb, - bool valid_status, u16 tx_status) +static void ar9170_dump_txqueue(struct ar9170 *ar, + struct sk_buff_head *queue) +{ + unsigned long flags; + + spin_lock_irqsave(&queue->lock, flags); + __ar9170_dump_txqueue(ar, queue); + spin_unlock_irqrestore(&queue->lock, flags); +} + +static void __ar9170_dump_txstats(struct ar9170 *ar) +{ + int i; + + printk(KERN_DEBUG "%s: QoS queue stats\n", + wiphy_name(ar->hw->wiphy)); + + for (i = 0; i < __AR9170_NUM_TXQ; i++) + printk(KERN_DEBUG "%s: queue:%d limit:%d len:%d waitack:%d\n", + wiphy_name(ar->hw->wiphy), i, ar->tx_stats[i].limit, + ar->tx_stats[i].len, skb_queue_len(&ar->tx_status[i])); +} + +static void ar9170_dump_txstats(struct ar9170 *ar) { - struct ieee80211_tx_info *txinfo; - unsigned int retries = 0, queue = skb_get_queue_mapping(skb); unsigned long flags; spin_lock_irqsave(&ar->tx_stats_lock, flags); - ar->tx_stats[queue].len--; - if (ieee80211_queue_stopped(ar->hw, queue)) - ieee80211_wake_queue(ar->hw, queue); + __ar9170_dump_txstats(ar); spin_unlock_irqrestore(&ar->tx_stats_lock, flags); +} +#endif /* AR9170_QUEUE_DEBUG */ + +/* caller must guarantee exclusive access for _bin_ queue. */ +static void ar9170_recycle_expired(struct ar9170 *ar, + struct sk_buff_head *queue, + struct sk_buff_head *bin) +{ + struct sk_buff *skb, *old = NULL; + unsigned long flags; + + spin_lock_irqsave(&queue->lock, flags); + while ((skb = skb_peek(queue))) { + struct ieee80211_tx_info *txinfo; + struct ar9170_tx_info *arinfo; + + txinfo = IEEE80211_SKB_CB(skb); + arinfo = (void *) txinfo->rate_driver_data; + + if (time_is_before_jiffies(arinfo->timeout)) { +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: [%ld > %ld] frame expired => " + "recycle \n", wiphy_name(ar->hw->wiphy), + jiffies, arinfo->timeout); + ar9170_print_txheader(ar, skb); +#endif /* AR9170_QUEUE_DEBUG */ + __skb_unlink(skb, queue); + __skb_queue_tail(bin, skb); + } else { + break; + } + + if (unlikely(old == skb)) { + /* bail out - queue is shot. */ + + WARN_ON(1); + break; + } + old = skb; + } + spin_unlock_irqrestore(&queue->lock, flags); +} + +static void ar9170_tx_status(struct ar9170 *ar, struct sk_buff *skb, + u16 tx_status) +{ + struct ieee80211_tx_info *txinfo; + unsigned int retries = 0; txinfo = IEEE80211_SKB_CB(skb); ieee80211_tx_info_clear_status(txinfo); @@ -247,45 +310,61 @@ void ar9170_handle_tx_status(struct ar9170 *ar, struct sk_buff *skb, break; } - if (valid_status) - txinfo->status.rates[0].count = retries + 1; - + txinfo->status.rates[0].count = retries + 1; skb_pull(skb, sizeof(struct ar9170_tx_control)); ieee80211_tx_status_irqsafe(ar->hw, skb); } -static struct sk_buff *ar9170_find_skb_in_queue(struct ar9170 *ar, - const u8 *mac, - const u32 queue, - struct sk_buff_head *q) +void ar9170_tx_callback(struct ar9170 *ar, struct sk_buff *skb) { + struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); + struct ar9170_tx_info *arinfo = (void *) info->rate_driver_data; + unsigned int queue = skb_get_queue_mapping(skb); unsigned long flags; - struct sk_buff *skb; - spin_lock_irqsave(&q->lock, flags); - skb_queue_walk(q, skb) { - struct ar9170_tx_control *txc = (void *) skb->data; - struct ieee80211_hdr *hdr = (void *) txc->frame_data; - u32 txc_queue = (le32_to_cpu(txc->phy_control) & - AR9170_TX_PHY_QOS_MASK) >> - AR9170_TX_PHY_QOS_SHIFT; + spin_lock_irqsave(&ar->tx_stats_lock, flags); + ar->tx_stats[queue].len--; - if ((queue != txc_queue) || - (compare_ether_addr(ieee80211_get_DA(hdr), mac))) - continue; + if (skb_queue_empty(&ar->tx_pending[queue])) { +#ifdef AR9170_QUEUE_STOP_DEBUG + printk(KERN_DEBUG "%s: wake queue %d\n", + wiphy_name(ar->hw->wiphy), queue); + __ar9170_dump_txstats(ar); +#endif /* AR9170_QUEUE_STOP_DEBUG */ + ieee80211_wake_queue(ar->hw, queue); + } + spin_unlock_irqrestore(&ar->tx_stats_lock, flags); - __skb_unlink(skb, q); - spin_unlock_irqrestore(&q->lock, flags); - return skb; + if (arinfo->flags & AR9170_TX_FLAG_BLOCK_ACK) { + dev_kfree_skb_any(skb); + } else if (arinfo->flags & AR9170_TX_FLAG_WAIT_FOR_ACK) { + arinfo->timeout = jiffies + + msecs_to_jiffies(AR9170_TX_TIMEOUT); + + skb_queue_tail(&ar->tx_status[queue], skb); + } else if (arinfo->flags & AR9170_TX_FLAG_NO_ACK) { + ar9170_tx_status(ar, skb, AR9170_TX_STATUS_FAILED); + } else { +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: unsupported frame flags!\n", + wiphy_name(ar->hw->wiphy)); + ar9170_print_txheader(ar, skb); +#endif /* AR9170_QUEUE_DEBUG */ + dev_kfree_skb_any(skb); + } + + if (!ar->tx_stats[queue].len && + !skb_queue_empty(&ar->tx_pending[queue])) { + ar9170_tx(ar); } - spin_unlock_irqrestore(&q->lock, flags); - return NULL; } -static struct sk_buff *ar9170_find_queued_skb(struct ar9170 *ar, const u8 *mac, - const u32 queue) +static struct sk_buff *ar9170_get_queued_skb(struct ar9170 *ar, + const u8 *mac, + struct sk_buff_head *queue, + const u32 rate) { - struct ieee80211_sta *sta; + unsigned long flags; struct sk_buff *skb; /* @@ -296,78 +375,91 @@ static struct sk_buff *ar9170_find_queued_skb(struct ar9170 *ar, const u8 *mac, * the firmware provided (-> destination MAC, and phy_control) - * and hope that we picked the right one... */ - rcu_read_lock(); - sta = ieee80211_find_sta(ar->hw, mac); - - if (likely(sta)) { - struct ar9170_sta_info *sta_priv = (void *) sta->drv_priv; - skb = skb_dequeue(&sta_priv->tx_status[queue]); - rcu_read_unlock(); - if (likely(skb)) - return skb; - } else - rcu_read_unlock(); - - /* scan the waste queue for candidates */ - skb = ar9170_find_skb_in_queue(ar, mac, queue, - &ar->global_tx_status_waste); - if (!skb) { - /* so it still _must_ be in the global list. */ - skb = ar9170_find_skb_in_queue(ar, mac, queue, - &ar->global_tx_status); - } + spin_lock_irqsave(&queue->lock, flags); + skb_queue_walk(queue, skb) { + struct ar9170_tx_control *txc = (void *) skb->data; + struct ieee80211_hdr *hdr = (void *) txc->frame_data; + u32 r; + + if (mac && compare_ether_addr(ieee80211_get_DA(hdr), mac)) { #ifdef AR9170_QUEUE_DEBUG - if (unlikely((!skb) && net_ratelimit())) { - printk(KERN_ERR "%s: ESS:[%pM] does not have any " - "outstanding frames in this queue (%d).\n", - wiphy_name(ar->hw->wiphy), mac, queue); + printk(KERN_DEBUG "%s: skip frame => DA %pM != %pM\n", + wiphy_name(ar->hw->wiphy), mac, + ieee80211_get_DA(hdr)); + ar9170_print_txheader(ar, skb); +#endif /* AR9170_QUEUE_DEBUG */ + continue; + } + + r = (le32_to_cpu(txc->phy_control) & AR9170_TX_PHY_MCS_MASK) >> + AR9170_TX_PHY_MCS_SHIFT; + + if ((rate != AR9170_TX_INVALID_RATE) && (r != rate)) { +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: skip frame => rate %d != %d\n", + wiphy_name(ar->hw->wiphy), rate, r); + ar9170_print_txheader(ar, skb); +#endif /* AR9170_QUEUE_DEBUG */ + continue; + } + + __skb_unlink(skb, queue); + spin_unlock_irqrestore(&queue->lock, flags); + return skb; } + +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_ERR "%s: ESS:[%pM] does not have any " + "outstanding frames in queue.\n", + wiphy_name(ar->hw->wiphy), mac); + __ar9170_dump_txqueue(ar, queue); #endif /* AR9170_QUEUE_DEBUG */ - return skb; + spin_unlock_irqrestore(&queue->lock, flags); + + return NULL; } /* - * This worker tries to keep the global tx_status queue empty. - * So we can guarantee that incoming tx_status reports for - * unregistered stations are always synced with the actual - * frame - which we think - belongs to. + * This worker tries to keeps an maintain tx_status queues. + * So we can guarantee that incoming tx_status reports are + * actually for a pending frame. */ -static void ar9170_tx_status_janitor(struct work_struct *work) +static void ar9170_tx_janitor(struct work_struct *work) { struct ar9170 *ar = container_of(work, struct ar9170, - tx_status_janitor.work); - struct sk_buff *skb; + tx_janitor.work); + struct sk_buff_head waste; + unsigned int i; + bool resched = false; if (unlikely(!IS_STARTED(ar))) return ; - /* recycle the garbage back to mac80211... one by one. */ - while ((skb = skb_dequeue(&ar->global_tx_status_waste))) { + skb_queue_head_init(&waste); + + for (i = 0; i < __AR9170_NUM_TXQ; i++) { #ifdef AR9170_QUEUE_DEBUG - printk(KERN_DEBUG "%s: dispose queued frame =>\n", - wiphy_name(ar->hw->wiphy)); - ar9170_print_txheader(ar, skb); + printk(KERN_DEBUG "%s: garbage collector scans queue:%d\n", + wiphy_name(ar->hw->wiphy), i); + ar9170_dump_txqueue(ar, &ar->tx_pending[i]); + ar9170_dump_txqueue(ar, &ar->tx_status[i]); #endif /* AR9170_QUEUE_DEBUG */ - ar9170_handle_tx_status(ar, skb, false, - AR9170_TX_STATUS_FAILED); - } - while ((skb = skb_dequeue(&ar->global_tx_status))) { -#ifdef AR9170_QUEUE_DEBUG - printk(KERN_DEBUG "%s: moving frame into waste queue =>\n", - wiphy_name(ar->hw->wiphy)); + ar9170_recycle_expired(ar, &ar->tx_status[i], &waste); + ar9170_recycle_expired(ar, &ar->tx_pending[i], &waste); + skb_queue_purge(&waste); - ar9170_print_txheader(ar, skb); -#endif /* AR9170_QUEUE_DEBUG */ - skb_queue_tail(&ar->global_tx_status_waste, skb); + if (!skb_queue_empty(&ar->tx_status[i]) || + !skb_queue_empty(&ar->tx_pending[i])) + resched = true; } - /* recall the janitor in 100ms - if there's garbage in the can. */ - if (skb_queue_len(&ar->global_tx_status_waste) > 0) - queue_delayed_work(ar->hw->workqueue, &ar->tx_status_janitor, - msecs_to_jiffies(100)); + if (resched) + queue_delayed_work(ar->hw->workqueue, + &ar->tx_janitor, + msecs_to_jiffies(AR9170_JANITOR_DELAY)); } void ar9170_handle_command_response(struct ar9170 *ar, void *buf, u32 len) @@ -394,15 +486,21 @@ void ar9170_handle_command_response(struct ar9170 *ar, void *buf, u32 len) */ struct sk_buff *skb; - u32 queue = (le32_to_cpu(cmd->tx_status.rate) & - AR9170_TX_PHY_QOS_MASK) >> AR9170_TX_PHY_QOS_SHIFT; + u32 phy = le32_to_cpu(cmd->tx_status.rate); + u32 q = (phy & AR9170_TX_PHY_QOS_MASK) >> + AR9170_TX_PHY_QOS_SHIFT; +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: recv tx_status for %pM, p:%08x, q:%d\n", + wiphy_name(ar->hw->wiphy), cmd->tx_status.dst, phy, q); +#endif /* AR9170_QUEUE_DEBUG */ - skb = ar9170_find_queued_skb(ar, cmd->tx_status.dst, queue); + skb = ar9170_get_queued_skb(ar, cmd->tx_status.dst, + &ar->tx_status[q], + AR9170_TX_INVALID_RATE); if (unlikely(!skb)) return ; - ar9170_handle_tx_status(ar, skb, true, - le16_to_cpu(cmd->tx_status.status)); + ar9170_tx_status(ar, skb, le16_to_cpu(cmd->tx_status.status)); break; } @@ -487,7 +585,7 @@ static void ar9170_rx_reset_rx_mpdu(struct ar9170 *ar) ar->rx_mpdu.has_plcp = false; } -static int ar9170_nag_limiter(struct ar9170 *ar) +int ar9170_nag_limiter(struct ar9170 *ar) { bool print_message; @@ -988,8 +1086,8 @@ static int ar9170_op_start(struct ieee80211_hw *hw) /* reinitialize queues statistics */ memset(&ar->tx_stats, 0, sizeof(ar->tx_stats)); - for (i = 0; i < ARRAY_SIZE(ar->tx_stats); i++) - ar->tx_stats[i].limit = 8; + for (i = 0; i < __AR9170_NUM_TXQ; i++) + ar->tx_stats[i].limit = AR9170_TXQ_DEPTH; /* reset QoS defaults */ AR9170_FILL_QUEUE(ar->edcf[0], 3, 15, 1023, 0); /* BEST EFFORT*/ @@ -1035,18 +1133,17 @@ out: static void ar9170_op_stop(struct ieee80211_hw *hw) { struct ar9170 *ar = hw->priv; + unsigned int i; if (IS_STARTED(ar)) ar->state = AR9170_IDLE; flush_workqueue(ar->hw->workqueue); - cancel_delayed_work_sync(&ar->tx_status_janitor); + cancel_delayed_work_sync(&ar->tx_janitor); cancel_work_sync(&ar->filter_config_work); cancel_work_sync(&ar->beacon_work); mutex_lock(&ar->mutex); - skb_queue_purge(&ar->global_tx_status_waste); - skb_queue_purge(&ar->global_tx_status); if (IS_ACCEPTING_CMD(ar)) { ar9170_set_leds_state(ar, 0); @@ -1056,51 +1153,32 @@ static void ar9170_op_stop(struct ieee80211_hw *hw) ar->stop(ar); } + for (i = 0; i < __AR9170_NUM_TXQ; i++) { + skb_queue_purge(&ar->tx_pending[i]); + skb_queue_purge(&ar->tx_status[i]); + } mutex_unlock(&ar->mutex); } -int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb) +static int ar9170_tx_prepare(struct ar9170 *ar, struct sk_buff *skb) { - struct ar9170 *ar = hw->priv; struct ieee80211_hdr *hdr; struct ar9170_tx_control *txc; struct ieee80211_tx_info *info; - struct ieee80211_rate *rate = NULL; struct ieee80211_tx_rate *txrate; + struct ar9170_tx_info *arinfo; unsigned int queue = skb_get_queue_mapping(skb); - unsigned long flags = 0; - struct ar9170_sta_info *sta_info = NULL; - u32 power, chains; u16 keytype = 0; u16 len, icv = 0; - int err; - bool tx_status; - if (unlikely(!IS_STARTED(ar))) - goto err_free; + BUILD_BUG_ON(sizeof(*arinfo) > sizeof(info->rate_driver_data)); hdr = (void *)skb->data; info = IEEE80211_SKB_CB(skb); len = skb->len; - spin_lock_irqsave(&ar->tx_stats_lock, flags); - if (ar->tx_stats[queue].limit < ar->tx_stats[queue].len) { - spin_unlock_irqrestore(&ar->tx_stats_lock, flags); - return NETDEV_TX_OK; - } - - ar->tx_stats[queue].len++; - ar->tx_stats[queue].count++; - if (ar->tx_stats[queue].limit == ar->tx_stats[queue].len) - ieee80211_stop_queue(hw, queue); - - spin_unlock_irqrestore(&ar->tx_stats_lock, flags); - txc = (void *)skb_push(skb, sizeof(*txc)); - tx_status = (((info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE) != 0) || - ((info->flags & IEEE80211_TX_CTL_REQ_TX_STATUS) != 0)); - if (info->control.hw_key) { icv = info->control.hw_key->icv_len; @@ -1116,7 +1194,7 @@ int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb) break; default: WARN_ON(1); - goto err_dequeue; + goto err_out; } } @@ -1133,16 +1211,65 @@ int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb) if (info->flags & IEEE80211_TX_CTL_NO_ACK) txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_NO_ACK); - if (info->flags & IEEE80211_TX_CTL_AMPDU) - txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_AGGR); - txrate = &info->control.rates[0]; - if (txrate->flags & IEEE80211_TX_RC_USE_CTS_PROTECT) txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_PROT_CTS); else if (txrate->flags & IEEE80211_TX_RC_USE_RTS_CTS) txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_PROT_RTS); + arinfo = (void *)info->rate_driver_data; + arinfo->timeout = jiffies + msecs_to_jiffies(AR9170_QUEUE_TIMEOUT); + + if (!(info->flags & IEEE80211_TX_CTL_NO_ACK) && + (is_valid_ether_addr(ieee80211_get_DA(hdr)))) { + if (info->flags & IEEE80211_TX_CTL_AMPDU) { + if (unlikely(!info->control.sta)) + goto err_out; + + txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_AGGR); + arinfo->flags = AR9170_TX_FLAG_BLOCK_ACK; + goto out; + } + + txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_RATE_PROBE); + /* + * WARNING: + * Putting the QoS queue bits into an unexplored territory is + * certainly not elegant. + * + * In my defense: This idea provides a reasonable way to + * smuggle valuable information to the tx_status callback. + * Also, the idea behind this bit-abuse came straight from + * the original driver code. + */ + + txc->phy_control |= + cpu_to_le32(queue << AR9170_TX_PHY_QOS_SHIFT); + arinfo->flags = AR9170_TX_FLAG_WAIT_FOR_ACK; + } else { + arinfo->flags = AR9170_TX_FLAG_NO_ACK; + } + +out: + return 0; + +err_out: + skb_pull(skb, sizeof(*txc)); + return -EINVAL; +} + +static void ar9170_tx_prepare_phy(struct ar9170 *ar, struct sk_buff *skb) +{ + struct ar9170_tx_control *txc; + struct ieee80211_tx_info *info; + struct ieee80211_rate *rate = NULL; + struct ieee80211_tx_rate *txrate; + u32 power, chains; + + txc = (void *) skb->data; + info = IEEE80211_SKB_CB(skb); + txrate = &info->control.rates[0]; + if (txrate->flags & IEEE80211_TX_RC_GREEN_FIELD) txc->phy_control |= cpu_to_le32(AR9170_TX_PHY_GREENFIELD); @@ -1162,9 +1289,12 @@ int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb) u32 r = txrate->idx; u8 *txpower; + /* heavy clip control */ + txc->phy_control |= cpu_to_le32((r & 0x7) << 7); + r <<= AR9170_TX_PHY_MCS_SHIFT; - if (WARN_ON(r & ~AR9170_TX_PHY_MCS_MASK)) - goto err_dequeue; + BUG_ON(r & ~AR9170_TX_PHY_MCS_MASK); + txc->phy_control |= cpu_to_le32(r & AR9170_TX_PHY_MCS_MASK); txc->phy_control |= cpu_to_le32(AR9170_TX_PHY_MOD_HT); @@ -1226,53 +1356,154 @@ int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb) chains = AR9170_TX_PHY_TXCHAIN_1; } txc->phy_control |= cpu_to_le32(chains << AR9170_TX_PHY_TXCHAIN_SHIFT); +} - if (tx_status) { - txc->mac_control |= cpu_to_le16(AR9170_TX_MAC_RATE_PROBE); - /* - * WARNING: - * Putting the QoS queue bits into an unexplored territory is - * certainly not elegant. - * - * In my defense: This idea provides a reasonable way to - * smuggle valuable information to the tx_status callback. - * Also, the idea behind this bit-abuse came straight from - * the original driver code. - */ +static void ar9170_tx(struct ar9170 *ar) +{ + struct sk_buff *skb; + unsigned long flags; + struct ieee80211_tx_info *info; + struct ar9170_tx_info *arinfo; + unsigned int i, frames, frames_failed, remaining_space; + int err; + bool schedule_garbagecollector = false; - txc->phy_control |= - cpu_to_le32(queue << AR9170_TX_PHY_QOS_SHIFT); + BUILD_BUG_ON(sizeof(*arinfo) > sizeof(info->rate_driver_data)); - if (info->control.sta) { - sta_info = (void *) info->control.sta->drv_priv; - skb_queue_tail(&sta_info->tx_status[queue], skb); - } else { - skb_queue_tail(&ar->global_tx_status, skb); + if (unlikely(!IS_STARTED(ar))) + return ; + + remaining_space = AR9170_TX_MAX_PENDING; + + for (i = 0; i < __AR9170_NUM_TXQ; i++) { + spin_lock_irqsave(&ar->tx_stats_lock, flags); + if (ar->tx_stats[i].len >= ar->tx_stats[i].limit) { +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: queue %d full\n", + wiphy_name(ar->hw->wiphy), i); + + __ar9170_dump_txstats(ar); + printk(KERN_DEBUG "stuck frames: ===> \n"); + ar9170_dump_txqueue(ar, &ar->tx_pending[i]); + ar9170_dump_txqueue(ar, &ar->tx_status[i]); +#endif /* AR9170_QUEUE_DEBUG */ + ieee80211_stop_queue(ar->hw, i); + spin_unlock_irqrestore(&ar->tx_stats_lock, flags); + continue; + } - queue_delayed_work(ar->hw->workqueue, - &ar->tx_status_janitor, - msecs_to_jiffies(100)); + frames = min(ar->tx_stats[i].limit - ar->tx_stats[i].len, + skb_queue_len(&ar->tx_pending[i])); + + if (remaining_space < frames) { +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: tx quota reached queue:%d, " + "remaining slots:%d, needed:%d\n", + wiphy_name(ar->hw->wiphy), i, remaining_space, + frames); + + ar9170_dump_txstats(ar); +#endif /* AR9170_QUEUE_DEBUG */ + frames = remaining_space; + } + + ar->tx_stats[i].len += frames; + ar->tx_stats[i].count += frames; + spin_unlock_irqrestore(&ar->tx_stats_lock, flags); + + if (!frames) + continue; + + frames_failed = 0; + while (frames) { + skb = skb_dequeue(&ar->tx_pending[i]); + if (unlikely(!skb)) { + frames_failed += frames; + frames = 0; + break; + } + + info = IEEE80211_SKB_CB(skb); + arinfo = (void *) info->rate_driver_data; + + /* TODO: cancel stuck frames */ + arinfo->timeout = jiffies + + msecs_to_jiffies(AR9170_TX_TIMEOUT); + +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: send frame q:%d =>\n", + wiphy_name(ar->hw->wiphy), i); + ar9170_print_txheader(ar, skb); +#endif /* AR9170_QUEUE_DEBUG */ + + err = ar->tx(ar, skb); + if (unlikely(err)) { + frames_failed++; + dev_kfree_skb_any(skb); + } else { + remaining_space--; + schedule_garbagecollector = true; + } + + frames--; + } + +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: ar9170_tx report for queue %d\n", + wiphy_name(ar->hw->wiphy), i); + + printk(KERN_DEBUG "%s: unprocessed pending frames left:\n", + wiphy_name(ar->hw->wiphy)); + ar9170_dump_txqueue(ar, &ar->tx_pending[i]); +#endif /* AR9170_QUEUE_DEBUG */ + + if (unlikely(frames_failed)) { +#ifdef AR9170_QUEUE_DEBUG + printk(KERN_DEBUG "%s: frames failed =>\n", + wiphy_name(ar->hw->wiphy), frames_failed); +#endif /* AR9170_QUEUE_DEBUG */ + + spin_lock_irqsave(&ar->tx_stats_lock, flags); + ar->tx_stats[i].len -= frames_failed; + ar->tx_stats[i].count -= frames_failed; + ieee80211_wake_queue(ar->hw, i); + spin_unlock_irqrestore(&ar->tx_stats_lock, flags); } } - err = ar->tx(ar, skb, tx_status, 0); - if (unlikely(tx_status && err)) { - if (info->control.sta) - skb_unlink(skb, &sta_info->tx_status[queue]); - else - skb_unlink(skb, &ar->global_tx_status); + if (schedule_garbagecollector) + queue_delayed_work(ar->hw->workqueue, + &ar->tx_janitor, + msecs_to_jiffies(AR9170_JANITOR_DELAY)); +} + +int ar9170_op_tx(struct ieee80211_hw *hw, struct sk_buff *skb) +{ + struct ar9170 *ar = hw->priv; + struct ieee80211_tx_info *info; + + if (unlikely(!IS_STARTED(ar))) + goto err_free; + + if (unlikely(ar9170_tx_prepare(ar, skb))) + goto err_free; + + info = IEEE80211_SKB_CB(skb); + if (info->flags & IEEE80211_TX_CTL_AMPDU) { + /* drop frame, we do not allow TX A-MPDU aggregation yet. */ + goto err_free; + } else { + unsigned int queue = skb_get_queue_mapping(skb); + + ar9170_tx_prepare_phy(ar, skb); + skb_queue_tail(&ar->tx_pending[queue], skb); } + ar9170_tx(ar); return NETDEV_TX_OK; -err_dequeue: - spin_lock_irqsave(&ar->tx_stats_lock, flags); - ar->tx_stats[queue].len--; - ar->tx_stats[queue].count--; - spin_unlock_irqrestore(&ar->tx_stats_lock, flags); - err_free: - dev_kfree_skb(skb); + dev_kfree_skb_any(skb); return NETDEV_TX_OK; } @@ -1698,43 +1929,6 @@ static void ar9170_sta_notify(struct ieee80211_hw *hw, enum sta_notify_cmd cmd, struct ieee80211_sta *sta) { - struct ar9170 *ar = hw->priv; - struct ar9170_sta_info *info = (void *) sta->drv_priv; - struct sk_buff *skb; - unsigned int i; - - switch (cmd) { - case STA_NOTIFY_ADD: - for (i = 0; i < ar->hw->queues; i++) - skb_queue_head_init(&info->tx_status[i]); - break; - - case STA_NOTIFY_REMOVE: - - /* - * transfer all outstanding frames that need a tx_status - * reports to the global tx_status queue - */ - - for (i = 0; i < ar->hw->queues; i++) { - while ((skb = skb_dequeue(&info->tx_status[i]))) { -#ifdef AR9170_QUEUE_DEBUG - printk(KERN_DEBUG "%s: queueing frame in " - "global tx_status queue =>\n", - wiphy_name(ar->hw->wiphy)); - - ar9170_print_txheader(ar, skb); -#endif /* AR9170_QUEUE_DEBUG */ - skb_queue_tail(&ar->global_tx_status, skb); - } - } - queue_delayed_work(ar->hw->workqueue, &ar->tx_status_janitor, - msecs_to_jiffies(100)); - break; - - default: - break; - } } static int ar9170_get_stats(struct ieee80211_hw *hw, @@ -1773,7 +1967,7 @@ static int ar9170_conf_tx(struct ieee80211_hw *hw, u16 queue, int ret; mutex_lock(&ar->mutex); - if ((param) && !(queue > ar->hw->queues)) { + if ((param) && !(queue > __AR9170_NUM_TXQ)) { memcpy(&ar->edcf[ar9170_qos_hwmap[queue]], param, sizeof(*param)); @@ -1849,12 +2043,14 @@ void *ar9170_alloc(size_t priv_size) mutex_init(&ar->mutex); spin_lock_init(&ar->cmdlock); spin_lock_init(&ar->tx_stats_lock); - skb_queue_head_init(&ar->global_tx_status); - skb_queue_head_init(&ar->global_tx_status_waste); + for (i = 0; i < __AR9170_NUM_TXQ; i++) { + skb_queue_head_init(&ar->tx_status[i]); + skb_queue_head_init(&ar->tx_pending[i]); + } ar9170_rx_reset_rx_mpdu(ar); INIT_WORK(&ar->filter_config_work, ar9170_set_filters); INIT_WORK(&ar->beacon_work, ar9170_new_beacon); - INIT_DELAYED_WORK(&ar->tx_status_janitor, ar9170_tx_status_janitor); + INIT_DELAYED_WORK(&ar->tx_janitor, ar9170_tx_janitor); /* all hw supports 2.4 GHz, so set channel to 1 by default */ ar->channel = &ar9170_2ghz_chantable[0]; diff --git a/drivers/net/wireless/ath/ar9170/usb.c b/drivers/net/wireless/ath/ar9170/usb.c index f752698..754b1f8 100644 --- a/drivers/net/wireless/ath/ar9170/usb.c +++ b/drivers/net/wireless/ath/ar9170/usb.c @@ -96,7 +96,49 @@ static struct usb_device_id ar9170_usb_ids[] = { }; MODULE_DEVICE_TABLE(usb, ar9170_usb_ids); -static void ar9170_usb_tx_urb_complete_free(struct urb *urb) +static void ar9170_usb_submit_urb(struct ar9170_usb *aru) +{ + struct urb *urb; + unsigned long flags; + int err; + + if (unlikely(!IS_STARTED(&aru->common))) + return ; + + spin_lock_irqsave(&aru->tx_urb_lock, flags); + if (aru->tx_submitted_urbs >= AR9170_NUM_TX_URBS) { + spin_unlock_irqrestore(&aru->tx_urb_lock, flags); + return ; + } + aru->tx_submitted_urbs++; + + urb = usb_get_from_anchor(&aru->tx_pending); + if (!urb) { + aru->tx_submitted_urbs--; + spin_unlock_irqrestore(&aru->tx_urb_lock, flags); + + return ; + } + spin_unlock_irqrestore(&aru->tx_urb_lock, flags); + + aru->tx_pending_urbs--; + usb_anchor_urb(urb, &aru->tx_submitted); + + err = usb_submit_urb(urb, GFP_ATOMIC); + if (unlikely(err)) { + if (ar9170_nag_limiter(&aru->common)) + dev_err(&aru->udev->dev, "submit_urb failed (%d).\n", + err); + + usb_unanchor_urb(urb); + aru->tx_submitted_urbs--; + ar9170_tx_callback(&aru->common, urb->context); + } + + usb_free_urb(urb); +} + +static void ar9170_usb_tx_urb_complete_frame(struct urb *urb) { struct sk_buff *skb = urb->context; struct ar9170_usb *aru = (struct ar9170_usb *) @@ -107,8 +149,11 @@ static void ar9170_usb_tx_urb_complete_free(struct urb *urb) return ; } - ar9170_handle_tx_status(&aru->common, skb, false, - AR9170_TX_STATUS_COMPLETE); + aru->tx_submitted_urbs--; + + ar9170_tx_callback(&aru->common, skb); + + ar9170_usb_submit_urb(aru); } static void ar9170_usb_tx_urb_complete(struct urb *urb) @@ -290,21 +335,47 @@ err_out: return err; } -static void ar9170_usb_cancel_urbs(struct ar9170_usb *aru) +static int ar9170_usb_flush(struct ar9170 *ar) { - int ret; + struct ar9170_usb *aru = (void *) ar; + struct urb *urb; + int ret, err = 0; - aru->common.state = AR9170_UNKNOWN_STATE; + if (IS_STARTED(ar)) + aru->common.state = AR9170_IDLE; - usb_unlink_anchored_urbs(&aru->tx_submitted); + usb_wait_anchor_empty_timeout(&aru->tx_pending, + msecs_to_jiffies(800)); + while ((urb = usb_get_from_anchor(&aru->tx_pending))) { + ar9170_tx_callback(&aru->common, (void *) urb->context); + usb_free_urb(urb); + } - /* give the LED OFF command and the deauth frame a chance to air. */ + /* lets wait a while until the tx - queues are dried out */ ret = usb_wait_anchor_empty_timeout(&aru->tx_submitted, msecs_to_jiffies(100)); if (ret == 0) - dev_err(&aru->udev->dev, "kill pending tx urbs.\n"); - usb_poison_anchored_urbs(&aru->tx_submitted); + err = -ETIMEDOUT; + + usb_kill_anchored_urbs(&aru->tx_submitted); + + if (IS_ACCEPTING_CMD(ar)) + aru->common.state = AR9170_STARTED; + + return err; +} + +static void ar9170_usb_cancel_urbs(struct ar9170_usb *aru) +{ + int err; + aru->common.state = AR9170_UNKNOWN_STATE; + + err = ar9170_usb_flush(&aru->common); + if (err) + dev_err(&aru->udev->dev, "stuck tx urbs!\n"); + + usb_poison_anchored_urbs(&aru->tx_submitted); usb_poison_anchored_urbs(&aru->rx_submitted); } @@ -388,12 +459,10 @@ err_free: return err; } -static int ar9170_usb_tx(struct ar9170 *ar, struct sk_buff *skb, - bool txstatus_needed, unsigned int extra_len) +static int ar9170_usb_tx(struct ar9170 *ar, struct sk_buff *skb) { struct ar9170_usb *aru = (struct ar9170_usb *) ar; struct urb *urb; - int err; if (unlikely(!IS_STARTED(ar))) { /* Seriously, what were you drink... err... thinking!? */ @@ -406,18 +475,17 @@ static int ar9170_usb_tx(struct ar9170 *ar, struct sk_buff *skb, usb_fill_bulk_urb(urb, aru->udev, usb_sndbulkpipe(aru->udev, AR9170_EP_TX), - skb->data, skb->len + extra_len, (txstatus_needed ? - ar9170_usb_tx_urb_complete : - ar9170_usb_tx_urb_complete_free), skb); + skb->data, skb->len, + ar9170_usb_tx_urb_complete_frame, skb); urb->transfer_flags |= URB_ZERO_PACKET; - usb_anchor_urb(urb, &aru->tx_submitted); - err = usb_submit_urb(urb, GFP_ATOMIC); - if (unlikely(err)) - usb_unanchor_urb(urb); + usb_anchor_urb(urb, &aru->tx_pending); + aru->tx_pending_urbs++; usb_free_urb(urb); - return err; + + ar9170_usb_submit_urb(aru); + return 0; } static void ar9170_usb_callback_cmd(struct ar9170 *ar, u32 len , void *buffer) @@ -617,10 +685,8 @@ static void ar9170_usb_stop(struct ar9170 *ar) if (IS_ACCEPTING_CMD(ar)) aru->common.state = AR9170_STOPPED; - /* lets wait a while until the tx - queues are dried out */ - ret = usb_wait_anchor_empty_timeout(&aru->tx_submitted, - msecs_to_jiffies(1000)); - if (ret == 0) + ret = ar9170_usb_flush(ar); + if (ret) dev_err(&aru->udev->dev, "kill pending tx urbs.\n"); usb_poison_anchored_urbs(&aru->tx_submitted); @@ -716,10 +782,16 @@ static int ar9170_usb_probe(struct usb_interface *intf, SET_IEEE80211_DEV(ar->hw, &udev->dev); init_usb_anchor(&aru->rx_submitted); + init_usb_anchor(&aru->tx_pending); init_usb_anchor(&aru->tx_submitted); init_completion(&aru->cmd_wait); + spin_lock_init(&aru->tx_urb_lock); + + aru->tx_pending_urbs = 0; + aru->tx_submitted_urbs = 0; aru->common.stop = ar9170_usb_stop; + aru->common.flush = ar9170_usb_flush; aru->common.open = ar9170_usb_open; aru->common.tx = ar9170_usb_tx; aru->common.exec_cmd = ar9170_usb_exec_cmd; diff --git a/drivers/net/wireless/ath/ar9170/usb.h b/drivers/net/wireless/ath/ar9170/usb.h index 69f4bce..d098f4d 100644 --- a/drivers/net/wireless/ath/ar9170/usb.h +++ b/drivers/net/wireless/ath/ar9170/usb.h @@ -51,6 +51,7 @@ #include "ar9170.h" #define AR9170_NUM_RX_URBS 16 +#define AR9170_NUM_TX_URBS 8 struct firmware; @@ -60,11 +61,15 @@ struct ar9170_usb { struct usb_interface *intf; struct usb_anchor rx_submitted; + struct usb_anchor tx_pending; struct usb_anchor tx_submitted; bool req_one_stage_fw; - spinlock_t cmdlock; + spinlock_t tx_urb_lock; + unsigned int tx_submitted_urbs; + unsigned int tx_pending_urbs; + struct completion cmd_wait; int readlen; u8 *readbuf; |