Track freed FW blocks during Tx. If no blocks were freed during a
predefined timeout, initiate a HW recovery. This helps in situations
when the FW watchdog fails.
Don't trigger recovery during activities that can temporarily stop
Tx. This includes:
- scanning
- buffering packets for sleeping stations (AP role)
- ROC on any role
Signed-off-by: Arik Nemtsov <[email protected]>
Signed-off-by: Eliad Peller <[email protected]>
---
drivers/net/wireless/wl12xx/cmd.c | 8 +++
drivers/net/wireless/wl12xx/conf.h | 3 +
drivers/net/wireless/wl12xx/main.c | 110 ++++++++++++++++++++++++++++++++++
drivers/net/wireless/wl12xx/scan.c | 6 ++
drivers/net/wireless/wl12xx/tx.c | 4 +
drivers/net/wireless/wl12xx/tx.h | 1 +
drivers/net/wireless/wl12xx/wl12xx.h | 3 +
7 files changed, 135 insertions(+), 0 deletions(-)
diff --git a/drivers/net/wireless/wl12xx/cmd.c b/drivers/net/wireless/wl12xx/cmd.c
index ae1f3d7..b0f377d 100644
--- a/drivers/net/wireless/wl12xx/cmd.c
+++ b/drivers/net/wireless/wl12xx/cmd.c
@@ -1818,6 +1818,14 @@ int wl12xx_croc(struct wl1271 *wl, u8 role_id)
goto out;
__clear_bit(role_id, wl->roc_map);
+
+ /*
+ * Rearm the tx watchdog when removing the last ROC. This prevents
+ * recoveries due to just finished ROCs - when Tx hasn't yet had
+ * a chance to get out.
+ */
+ if (find_first_bit(wl->roc_map, WL12XX_MAX_ROLES) >= WL12XX_MAX_ROLES)
+ wl12xx_rearm_tx_watchdog_locked(wl);
out:
return ret;
}
diff --git a/drivers/net/wireless/wl12xx/conf.h b/drivers/net/wireless/wl12xx/conf.h
index cc50faa..3e581e1 100644
--- a/drivers/net/wireless/wl12xx/conf.h
+++ b/drivers/net/wireless/wl12xx/conf.h
@@ -690,6 +690,9 @@ struct conf_tx_settings {
*/
u8 tmpl_short_retry_limit;
u8 tmpl_long_retry_limit;
+
+ /* Time in ms for Tx watchdog timer to expire */
+ u32 tx_watchdog_timeout;
};
enum {
diff --git a/drivers/net/wireless/wl12xx/main.c b/drivers/net/wireless/wl12xx/main.c
index 418b2b8..a9b9fce 100644
--- a/drivers/net/wireless/wl12xx/main.c
+++ b/drivers/net/wireless/wl12xx/main.c
@@ -217,6 +217,7 @@ static struct conf_drv_settings default_conf = {
.basic_rate_5 = CONF_HW_BIT_RATE_6MBPS,
.tmpl_short_retry_limit = 10,
.tmpl_long_retry_limit = 10,
+ .tx_watchdog_timeout = 5000,
},
.conn = {
.wake_up_event = CONF_WAKE_UP_EVENT_DTIM,
@@ -623,6 +624,80 @@ static void wl1271_rx_streaming_timer(unsigned long data)
ieee80211_queue_work(wl->hw, &wlvif->rx_streaming_disable_work);
}
+/* wl->mutex must be taken */
+void wl12xx_rearm_tx_watchdog_locked(struct wl1271 *wl)
+{
+ /* if the watchdog is not armed, don't do anything */
+ if (wl->tx_allocated_blocks == 0)
+ return;
+
+ cancel_delayed_work(&wl->tx_watchdog_work);
+ ieee80211_queue_delayed_work(wl->hw, &wl->tx_watchdog_work,
+ msecs_to_jiffies(wl->conf.tx.tx_watchdog_timeout));
+}
+
+static void wl12xx_tx_watchdog_work(struct work_struct *work)
+{
+ struct delayed_work *dwork;
+ struct wl1271 *wl;
+
+ dwork = container_of(work, struct delayed_work, work);
+ wl = container_of(dwork, struct wl1271, tx_watchdog_work);
+
+ mutex_lock(&wl->mutex);
+
+ if (unlikely(wl->state == WL1271_STATE_OFF))
+ goto out;
+
+ /* Tx went out in the meantime - everything is ok */
+ if (unlikely(wl->tx_allocated_blocks == 0))
+ goto out;
+
+ /*
+ * if a ROC is in progress, we might not have any Tx for a long
+ * time (e.g. pending Tx on the non-ROC channels)
+ */
+ if (find_first_bit(wl->roc_map, WL12XX_MAX_ROLES) < WL12XX_MAX_ROLES) {
+ wl1271_debug(DEBUG_TX, "No Tx (in FW) for %d ms due to ROC",
+ wl->conf.tx.tx_watchdog_timeout);
+ wl12xx_rearm_tx_watchdog_locked(wl);
+ goto out;
+ }
+
+ /*
+ * if a scan is in progress, we might not have any Tx for a long
+ * time
+ */
+ if (wl->scan.state != WL1271_SCAN_STATE_IDLE) {
+ wl1271_debug(DEBUG_TX, "No Tx (in FW) for %d ms due to scan",
+ wl->conf.tx.tx_watchdog_timeout);
+ wl12xx_rearm_tx_watchdog_locked(wl);
+ goto out;
+ }
+
+ /*
+ * AP might cache a frame for a long time for a sleeping station,
+ * so rearm the timer if there's an AP interface with stations. If
+ * Tx is genuinely stuck we will most hopefully discover it when all
+ * stations are removed due to inactivity.
+ */
+ if (wl->active_sta_count) {
+ wl1271_debug(DEBUG_TX, "No Tx (in FW) for %d ms. AP has "
+ " %d stations",
+ wl->conf.tx.tx_watchdog_timeout,
+ wl->active_sta_count);
+ wl12xx_rearm_tx_watchdog_locked(wl);
+ goto out;
+ }
+
+ wl1271_error("Tx stuck (in FW) for %d ms. Starting recovery",
+ wl->conf.tx.tx_watchdog_timeout);
+ wl12xx_queue_recovery_work(wl);
+
+out:
+ mutex_unlock(&wl->mutex);
+}
+
static void wl1271_conf_init(struct wl1271 *wl)
{
@@ -815,6 +890,18 @@ static void wl12xx_fw_status(struct wl1271 *wl,
wl->tx_allocated_blocks -= freed_blocks;
+ /*
+ * If the FW freed some blocks:
+ * If we still have allocated blocks - re-arm the timer, Tx is
+ * not stuck. Otherwise, cancel the timer (no Tx currently).
+ */
+ if (freed_blocks) {
+ if (wl->tx_allocated_blocks)
+ wl12xx_rearm_tx_watchdog_locked(wl);
+ else
+ cancel_delayed_work(&wl->tx_watchdog_work);
+ }
+
avail = le32_to_cpu(status->tx_total) - wl->tx_allocated_blocks;
/*
@@ -1487,6 +1574,7 @@ int wl1271_plt_stop(struct wl1271 *wl)
cancel_work_sync(&wl->netstack_work);
cancel_work_sync(&wl->recovery_work);
cancel_delayed_work_sync(&wl->elp_work);
+ cancel_delayed_work_sync(&wl->tx_watchdog_work);
mutex_lock(&wl->mutex);
wl1271_power_off(wl);
@@ -1865,6 +1953,7 @@ static void wl1271_op_stop(struct ieee80211_hw *hw)
cancel_work_sync(&wl->netstack_work);
cancel_work_sync(&wl->tx_work);
cancel_delayed_work_sync(&wl->elp_work);
+ cancel_delayed_work_sync(&wl->tx_watchdog_work);
/* let's notify MAC80211 about the remaining pending TX frames */
wl12xx_tx_reset(wl, true);
@@ -2298,6 +2387,12 @@ static void __wl1271_op_remove_interface(struct wl1271 *wl,
if (wl->scan.state != WL1271_SCAN_STATE_IDLE &&
wl->scan_vif == vif) {
+ /*
+ * Rearm the tx watchdog just before idling scan. This
+ * prevents just-finished scans from triggering the watchdog
+ */
+ wl12xx_rearm_tx_watchdog_locked(wl);
+
wl->scan.state = WL1271_SCAN_STATE_IDLE;
memset(wl->scan.scanned_ch, 0, sizeof(wl->scan.scanned_ch));
wl->scan_vif = NULL;
@@ -3223,6 +3318,13 @@ static void wl1271_op_cancel_hw_scan(struct ieee80211_hw *hw,
if (ret < 0)
goto out_sleep;
}
+
+ /*
+ * Rearm the tx watchdog just before idling scan. This
+ * prevents just-finished scans from triggering the watchdog
+ */
+ wl12xx_rearm_tx_watchdog_locked(wl);
+
wl->scan.state = WL1271_SCAN_STATE_IDLE;
memset(wl->scan.scanned_ch, 0, sizeof(wl->scan.scanned_ch));
wl->scan_vif = NULL;
@@ -4234,6 +4336,13 @@ void wl1271_free_sta(struct wl1271 *wl, struct wl12xx_vif *wlvif, u8 hlid)
__clear_bit(hlid, (unsigned long *)&wl->ap_fw_ps_map);
wl12xx_free_link(wl, wlvif, &hlid);
wl->active_sta_count--;
+
+ /*
+ * rearm the tx watchdog when the last STA is freed - give the FW a
+ * chance to return STA-buffered packets before complaining.
+ */
+ if (wl->active_sta_count == 0)
+ wl12xx_rearm_tx_watchdog_locked(wl);
}
static int wl1271_op_sta_add(struct ieee80211_hw *hw,
@@ -5271,6 +5380,7 @@ static struct ieee80211_hw *wl1271_alloc_hw(void)
INIT_WORK(&wl->tx_work, wl1271_tx_work);
INIT_WORK(&wl->recovery_work, wl1271_recovery_work);
INIT_DELAYED_WORK(&wl->scan_complete_work, wl1271_scan_complete_work);
+ INIT_DELAYED_WORK(&wl->tx_watchdog_work, wl12xx_tx_watchdog_work);
wl->freezable_wq = create_freezable_workqueue("wl12xx_wq");
if (!wl->freezable_wq) {
diff --git a/drivers/net/wireless/wl12xx/scan.c b/drivers/net/wireless/wl12xx/scan.c
index e43a6b2..fcba055 100644
--- a/drivers/net/wireless/wl12xx/scan.c
+++ b/drivers/net/wireless/wl12xx/scan.c
@@ -55,6 +55,12 @@ void wl1271_scan_complete_work(struct work_struct *work)
vif = wl->scan_vif;
wlvif = wl12xx_vif_to_data(vif);
+ /*
+ * Rearm the tx watchdog just before idling scan. This
+ * prevents just-finished scans from triggering the watchdog
+ */
+ wl12xx_rearm_tx_watchdog_locked(wl);
+
wl->scan.state = WL1271_SCAN_STATE_IDLE;
memset(wl->scan.scanned_ch, 0, sizeof(wl->scan.scanned_ch));
wl->scan.req = NULL;
diff --git a/drivers/net/wireless/wl12xx/tx.c b/drivers/net/wireless/wl12xx/tx.c
index 8f78fdd..43ae491 100644
--- a/drivers/net/wireless/wl12xx/tx.c
+++ b/drivers/net/wireless/wl12xx/tx.c
@@ -226,6 +226,10 @@ static int wl1271_tx_allocate(struct wl1271 *wl, struct wl12xx_vif *wlvif,
wl->tx_blocks_available -= total_blocks;
wl->tx_allocated_blocks += total_blocks;
+ /* If the FW was empty before, arm the Tx watchdog */
+ if (wl->tx_allocated_blocks == total_blocks)
+ wl12xx_rearm_tx_watchdog_locked(wl);
+
ac = wl1271_tx_get_queue(skb_get_queue_mapping(skb));
wl->tx_allocated_pkts[ac]++;
diff --git a/drivers/net/wireless/wl12xx/tx.h b/drivers/net/wireless/wl12xx/tx.h
index e3977b5..5cf8c32 100644
--- a/drivers/net/wireless/wl12xx/tx.h
+++ b/drivers/net/wireless/wl12xx/tx.h
@@ -227,5 +227,6 @@ void wl12xx_rearm_rx_streaming(struct wl1271 *wl, unsigned long *active_hlids);
/* from main.c */
void wl1271_free_sta(struct wl1271 *wl, struct wl12xx_vif *wlvif, u8 hlid);
+void wl12xx_rearm_tx_watchdog_locked(struct wl1271 *wl);
#endif
diff --git a/drivers/net/wireless/wl12xx/wl12xx.h b/drivers/net/wireless/wl12xx/wl12xx.h
index 9035241..7f58454 100644
--- a/drivers/net/wireless/wl12xx/wl12xx.h
+++ b/drivers/net/wireless/wl12xx/wl12xx.h
@@ -495,6 +495,9 @@ struct wl1271 {
/* last wlvif we transmitted from */
struct wl12xx_vif *last_wlvif;
+
+ /* work to fire when Tx is stuck */
+ struct delayed_work tx_watchdog_work;
};
struct wl1271_station {
--
1.7.5.4
On Sat, 2012-03-03 at 22:18 +0200, Arik Nemtsov wrote:
> Track freed FW blocks during Tx. If no blocks were freed during a
> predefined timeout, initiate a HW recovery. This helps in situations
> when the FW watchdog fails.
>
> Don't trigger recovery during activities that can temporarily stop
> Tx. This includes:
> - scanning
> - buffering packets for sleeping stations (AP role)
> - ROC on any role
>
> Signed-off-by: Arik Nemtsov <[email protected]>
> Signed-off-by: Eliad Peller <[email protected]>
> ---
Applied and pushed to wl12xx. Thanks!
--
Cheers,
Luca.