The driver for the RTL8187L chips returns IEEE80211_TX_STAT_ACK for all
packets, even if the maximum number of retries was exhausted. In addition
it fails to setup max_rates in the ieee80211_hw struct, This behavior
may be responsible for the problems noted in Bug 14168. As the bug is very
old, testers have not been found, and I do not have the case where the
indicated signal is less than -70 dBm.
Signed-off-by: Larry Finger <[email protected]>
Cc: Stable <[email protected]>
---
V1 => V2: Protect against zero divide, and the case where the 16-bit retry
counter wraps to zero. The latter is unlikely, but tested for
completeness.
---
Index: wireless-testing/drivers/net/wireless/rtl818x/rtl8187/dev.c
===================================================================
--- wireless-testing.orig/drivers/net/wireless/rtl818x/rtl8187/dev.c
+++ wireless-testing/drivers/net/wireless/rtl818x/rtl8187/dev.c
@@ -869,23 +869,35 @@ static void rtl8187_work(struct work_str
/* The RTL8187 returns the retry count through register 0xFFFA. In
* addition, it appears to be a cumulative retry count, not the
* value for the current TX packet. When multiple TX entries are
- * queued, the retry count will be valid for the last one in the queue.
- * The "error" should not matter for purposes of rate setting. */
+ * waiting in the queue, the retry count will be the total for all.
+ * The "error" may matter for purposes of rate setting, but there is
+ * no other choice with this hardware.
+ */
struct rtl8187_priv *priv = container_of(work, struct rtl8187_priv,
work.work);
struct ieee80211_tx_info *info;
struct ieee80211_hw *dev = priv->dev;
static u16 retry;
u16 tmp;
+ u16 avg_retry;
+ int length;
mutex_lock(&priv->conf_mutex);
tmp = rtl818x_ioread16(priv, (__le16 *)0xFFFA);
+ length = skb_queue_len(&priv->b_tx_status.queue);
+ if (unlikely(!length))
+ length = 1;
+ if (unlikely(tmp < retry))
+ tmp = retry;
+ avg_retry = (tmp - retry) / length;
while (skb_queue_len(&priv->b_tx_status.queue) > 0) {
struct sk_buff *old_skb;
old_skb = skb_dequeue(&priv->b_tx_status.queue);
info = IEEE80211_SKB_CB(old_skb);
- info->status.rates[0].count = tmp - retry + 1;
+ info->status.rates[0].count = avg_retry + 1;
+ if (info->status.rates[0].count > RETRY_COUNT)
+ info->flags &= ~IEEE80211_TX_STAT_ACK;
ieee80211_tx_status_irqsafe(dev, old_skb);
}
retry = tmp;
@@ -931,8 +943,8 @@ static int rtl8187_start(struct ieee8021
rtl818x_iowrite32(priv, &priv->map->TX_CONF,
RTL818X_TX_CONF_HW_SEQNUM |
RTL818X_TX_CONF_DISREQQSIZE |
- (7 << 8 /* short retry limit */) |
- (7 << 0 /* long retry limit */) |
+ (RETRY_COUNT << 8 /* short retry limit */) |
+ (RETRY_COUNT << 0 /* long retry limit */) |
(7 << 21 /* MAX TX DMA */));
rtl8187_init_urbs(dev);
rtl8187b_init_status_urb(dev);
@@ -1376,6 +1388,9 @@ static int __devinit rtl8187_probe(struc
dev->flags = IEEE80211_HW_HOST_BROADCAST_PS_BUFFERING |
IEEE80211_HW_SIGNAL_DBM |
IEEE80211_HW_RX_INCLUDES_FCS;
+ /* Initialize rate-control variables */
+ dev->max_rates = 1;
+ dev->max_rate_tries = RETRY_COUNT;
eeprom.data = dev;
eeprom.register_read = rtl8187_eeprom_register_read;
Index: wireless-testing/drivers/net/wireless/rtl818x/rtl8187/rtl8187.h
===================================================================
--- wireless-testing.orig/drivers/net/wireless/rtl818x/rtl8187/rtl8187.h
+++ wireless-testing/drivers/net/wireless/rtl818x/rtl8187/rtl8187.h
@@ -35,6 +35,8 @@
#define RFKILL_MASK_8187_89_97 0x2
#define RFKILL_MASK_8198 0x4
+#define RETRY_COUNT 7
+
struct rtl8187_rx_info {
struct urb *urb;
struct ieee80211_hw *dev;
--- On Tue, 1/3/11, Larry Finger <[email protected]> wrote:
> The driver for the RTL8187L chips
> returns IEEE80211_TX_STAT_ACK for all
> packets, even if the maximum number of retries was
> exhausted. In addition
> it fails to setup max_rates in the ieee80211_hw struct,
> This behavior
> may be responsible for the problems noted in Bug 14168. As
> the bug is very
> old, testers have not been found, and I do not have the
> case where the
> indicated signal is less than -70 dBm.
>
> Signed-off-by: Larry Finger <[email protected]>
> Cc: Stable <[email protected]>
> ---
>
> V1 => V2: Protect against zero divide, and the case
> where the 16-bit retry
> ??? ? counter wraps to zero. The latter
> is unlikely, but tested for
> ??? ? completeness.
Acked-by: Hin-Tak Leung <[email protected]>
This may be style-thing - the meaning of some of the variables changes mid-routine. I think one alternative way (not necessarily better) is to unlock the mutex and return if length ==0, and something similiar to the register wrap-around?
I'm okay with this as is though.