This patch is not yet upstream...

From a5e660b4e294556822913627544f661e59b39716 Mon Sep 17 00:00:00 2001

From: Wey-Yi Guy <wey-yi.w.guy@intel.com>

Date: Mon, 1 Mar 2010 17:23:50 -0800

Subject: [PATCH 13/17] iwlwifi: Recover TX flow stall due to stuck queue

Monitors the internal TX queues periodically.  When a queue is stuck

for some unknown conditions causing the throughput to drop and the

transfer is stop, the driver will force firmware reload and bring the

system back to normal operational state.

The iwlwifi devices behave differently in this regard so this feature is

made part of the ops infrastructure so we can have more control on how to

monitor and recover from tx queue stall case per device.

Signed-off-by: Trieu 'Andrew' Nguyen <trieux.t.nguyen@intel.com>

Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-1000.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-1000.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-1000.c.orig	2010-03-22 15:33:38.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-1000.c	2010-03-22 15:48:54.000000000 -0400

@@ -135,6 +135,7 @@ static struct iwl_lib_ops iwl1000_lib = 

 		.temperature = iwl5000_temperature,

 		.set_ct_kill = iwl1000_set_ct_threshold,

 	 },

+	.recover_from_tx_stall = iwl_bg_monitor_recover,

 };

 static struct iwl_ops iwl1000_ops = {

@@ -163,5 +164,6 @@ struct iwl_cfg iwl1000_bgn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_EXT_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl3945-base.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl3945-base.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl3945-base.c.orig	2010-03-22 15:44:04.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl3945-base.c	2010-03-22 15:48:54.000000000 -0400

@@ -2453,6 +2453,13 @@ static void iwl3945_alive_start(struct i

 	/* After the ALIVE response, we can send commands to 3945 uCode */

 	set_bit(STATUS_ALIVE, &priv->status);

+	if (priv->cfg->ops->lib->recover_from_tx_stall) {

+		/* Enable timer to monitor the driver queues */

+		mod_timer(&priv->monitor_recover,

+			jiffies +

+			msecs_to_jiffies(priv->cfg->monitor_recover_period));

+	}

+

 	if (iwl_is_rfkill(priv))

 		return;

@@ -3730,6 +3737,13 @@ static void iwl3945_setup_deferred_work(

 	iwl3945_hw_setup_deferred_work(priv);

+	if (priv->cfg->ops->lib->recover_from_tx_stall) {

+		init_timer(&priv->monitor_recover);

+		priv->monitor_recover.data = (unsigned long)priv;

+		priv->monitor_recover.function =

+			priv->cfg->ops->lib->recover_from_tx_stall;

+	}

+

 	tasklet_init(&priv->irq_tasklet, (void (*)(unsigned long))

 		     iwl3945_irq_tasklet, (unsigned long)priv);

 }

@@ -3742,6 +3756,8 @@ static void iwl3945_cancel_deferred_work

 	cancel_delayed_work(&priv->scan_check);

 	cancel_delayed_work(&priv->alive_start);

 	cancel_work_sync(&priv->beacon_update);

+	if (priv->cfg->ops->lib->recover_from_tx_stall)

+		del_timer_sync(&priv->monitor_recover);

 }

 static struct attribute *iwl3945_sysfs_entries[] = {

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-3945.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-3945.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-3945.c.orig	2010-03-22 14:20:28.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-3945.c	2010-03-22 15:48:54.000000000 -0400

@@ -2897,6 +2897,7 @@ static struct iwl_cfg iwl3945_bg_cfg = {

 	.ht_greenfield_support = false,

 	.broken_powersave = true,

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 static struct iwl_cfg iwl3945_abg_cfg = {

@@ -2913,6 +2914,7 @@ static struct iwl_cfg iwl3945_abg_cfg = 

 	.ht_greenfield_support = false,

 	.broken_powersave = true,

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct pci_device_id iwl3945_hw_card_ids[] = {

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-4965.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-4965.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-4965.c.orig	2010-03-22 14:24:14.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-4965.c	2010-03-22 15:48:54.000000000 -0400

@@ -2364,6 +2364,7 @@ struct iwl_cfg iwl4965_agn_cfg = {

 	.ht_greenfield_support = false,

 	.broken_powersave = true,

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 /* Module firmware */

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-5000.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-5000.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-5000.c.orig	2010-03-22 14:27:05.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-5000.c	2010-03-22 15:48:54.000000000 -0400

@@ -1579,6 +1579,7 @@ struct iwl_lib_ops iwl5000_lib = {

 		.temperature = iwl5000_temperature,

 		.set_ct_kill = iwl5000_set_ct_threshold,

 	 },

+	.recover_from_tx_stall = iwl_bg_monitor_recover,

 };

 static struct iwl_lib_ops iwl5150_lib = {

@@ -1631,6 +1632,7 @@ static struct iwl_lib_ops iwl5150_lib = 

 		.temperature = iwl5150_temperature,

 		.set_ct_kill = iwl5150_set_ct_threshold,

 	 },

+	.recover_from_tx_stall = iwl_bg_monitor_recover,

 };

 struct iwl_ops iwl5000_ops = {

@@ -1673,6 +1675,7 @@ struct iwl_cfg iwl5300_agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl5100_bg_cfg = {

@@ -1691,6 +1694,7 @@ struct iwl_cfg iwl5100_bg_cfg = {

 	.need_pll_cfg = true,

 	.ht_greenfield_support = true,

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl5100_abg_cfg = {

@@ -1709,6 +1713,7 @@ struct iwl_cfg iwl5100_abg_cfg = {

 	.valid_rx_ant = ANT_AB,

 	.need_pll_cfg = true,

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl5100_agn_cfg = {

@@ -1728,6 +1733,7 @@ struct iwl_cfg iwl5100_agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl5350_agn_cfg = {

@@ -1747,6 +1753,7 @@ struct iwl_cfg iwl5350_agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl5150_agn_cfg = {

@@ -1766,6 +1773,7 @@ struct iwl_cfg iwl5150_agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_LONG_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 MODULE_FIRMWARE(IWL5000_MODULE_FIRMWARE(IWL5000_UCODE_API_MAX));

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-6000.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-6000.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-6000.c.orig	2010-03-22 14:28:04.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-6000.c	2010-03-22 15:51:12.000000000 -0400

@@ -137,6 +137,7 @@ static struct iwl_lib_ops iwl6000_lib = 

 		.temperature = iwl5000_temperature,

 		.set_ct_kill = iwl6000_set_ct_threshold,

 	 },

+	.recover_from_tx_stall = iwl_bg_monitor_recover,

 };

 static struct iwl_hcmd_utils_ops iwl6000_hcmd_utils = {

@@ -177,6 +178,7 @@ struct iwl_cfg iwl6000h_2agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 /*

@@ -202,6 +204,7 @@ struct iwl_cfg iwl6000i_2agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl6050_2agn_cfg = {

@@ -224,6 +227,7 @@ struct iwl_cfg iwl6050_2agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl6000_3agn_cfg = {

@@ -246,6 +250,7 @@ struct iwl_cfg iwl6000_3agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 struct iwl_cfg iwl6050_3agn_cfg = {

@@ -268,6 +273,7 @@ struct iwl_cfg iwl6050_3agn_cfg = {

 	.ht_greenfield_support = true,

 	.use_rts_for_ht = true, /* use rts/cts protection */

 	.plcp_delta_threshold = IWL_MAX_PLCP_ERR_THRESHOLD_DEF,

+	.monitor_recover_period = IWL_MONITORING_PERIOD,

 };

 MODULE_FIRMWARE(IWL6000_MODULE_FIRMWARE(IWL6000_UCODE_API_MAX));

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-agn.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-agn.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-agn.c.orig	2009-12-02 22:51:21.000000000 -0500

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-agn.c	2010-03-22 15:48:54.000000000 -0400

@@ -1755,6 +1755,13 @@ static void iwl_alive_start(struct iwl_p

 	/* After the ALIVE response, we can send host commands to the uCode */

 	set_bit(STATUS_ALIVE, &priv->status);

+	if (priv->cfg->ops->lib->recover_from_tx_stall) {

+		/* Enable timer to monitor the driver queues */

+		mod_timer(&priv->monitor_recover,

+			jiffies +

+			msecs_to_jiffies(priv->cfg->monitor_recover_period));

+	}

+

 	if (iwl_is_rfkill(priv))

 		return;

@@ -2829,6 +2836,13 @@ static void iwl_setup_deferred_work(stru

 	priv->statistics_periodic.data = (unsigned long)priv;

 	priv->statistics_periodic.function = iwl_bg_statistics_periodic;

+	if (priv->cfg->ops->lib->recover_from_tx_stall) {

+		init_timer(&priv->monitor_recover);

+		priv->monitor_recover.data = (unsigned long)priv;

+		priv->monitor_recover.function =

+			priv->cfg->ops->lib->recover_from_tx_stall;

+	}

+

 	if (!priv->cfg->use_isr_legacy)

 		tasklet_init(&priv->irq_tasklet, (void (*)(unsigned long))

 			iwl_irq_tasklet, (unsigned long)priv);

@@ -2847,6 +2861,8 @@ static void iwl_cancel_deferred_work(str

 	cancel_delayed_work(&priv->alive_start);

 	cancel_work_sync(&priv->beacon_update);

 	del_timer_sync(&priv->statistics_periodic);

+	if (priv->cfg->ops->lib->recover_from_tx_stall)

+		del_timer_sync(&priv->monitor_recover);

 }

 static struct attribute *iwl_sysfs_entries[] = {

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.c.orig	2010-03-22 15:40:48.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.c	2010-03-22 15:48:54.000000000 -0400

@@ -3107,6 +3107,99 @@ int iwl_force_reset(struct iwl_priv *pri

 	}

 	return 0;

 }

+EXPORT_SYMBOL(iwl_force_reset);

+

+/**

+ * iwl_bg_monitor_recover - Timer callback to check for stuck queue and recover

+ *

+ * During normal condition (no queue is stuck), the timer is continually set to

+ * execute every monitor_recover_period milliseconds after the last timer

+ * expired.  When the queue read_ptr is at the same place, the timer is

+ * shorten to 100mSecs.  This is

+ *      1) to reduce the chance that the read_ptr may wrap around (not stuck)

+ *      2) to detect the stuck queues quicker before the station and AP can

+ *      disassociate each other.

+ *

+ * This function monitors all the tx queues and recover from it if any

+ * of the queues are stuck.

+ * 1. It first check the cmd queue for stuck conditions.  If it is stuck,

+ *      it will recover by resetting the firmware and return.

+ * 2. Then, it checks for station association.  If it associates it will check

+ *      other queues.  If any queue is stuck, it will recover by resetting

+ *      the firmware.

+ * Note: It the number of times the queue read_ptr to be at the same place to

+ *      be MAX_REPEAT+1 in order to consider to be stuck.

+ */

+/*

+ * The maximum number of times the read pointer of the tx queue at the

+ * same place without considering to be stuck.

+ */

+#define MAX_REPEAT      (2)

+static int iwl_check_stuck_queue(struct iwl_priv *priv, int cnt)

+{

+	struct iwl_tx_queue *txq;

+	struct iwl_queue *q;

+

+	txq = &priv->txq[cnt];

+	q = &txq->q;

+	/* queue is empty, skip */

+	if (q->read_ptr != q->write_ptr) {

+		if (q->read_ptr == q->last_read_ptr) {

+			/* a queue has not been read from last time */

+			if (q->repeat_same_read_ptr > MAX_REPEAT) {

+				IWL_ERR(priv,

+					"queue %d stuck %d time. Fw reload.\n",

+					q->id, q->repeat_same_read_ptr);

+				q->repeat_same_read_ptr = 0;

+				iwl_force_reset(priv, IWL_FW_RESET);

+			} else {

+				q->repeat_same_read_ptr++;

+				IWL_DEBUG_RADIO(priv,

+						"queue %d, not read %d time\n",

+						q->id,

+						q->repeat_same_read_ptr);

+				mod_timer(&priv->monitor_recover, jiffies +

+					msecs_to_jiffies(IWL_ONE_HUNDRED_MSECS));

+			}

+			return 1;

+		} else {

+			q->last_read_ptr = q->read_ptr;

+			q->repeat_same_read_ptr = 0;

+		}

+	}

+	return 0;

+}

+

+void iwl_bg_monitor_recover(unsigned long data)

+{

+	struct iwl_priv *priv = (struct iwl_priv *)data;

+	int cnt;

+

+	if (test_bit(STATUS_EXIT_PENDING, &priv->status))

+		return;

+

+	/* monitor and check for stuck cmd queue */

+	if (iwl_check_stuck_queue(priv, IWL_CMD_QUEUE_NUM))

+		return;

+

+	/* monitor and check for other stuck queues */

+	if (iwl_is_associated(priv)) {

+		for (cnt = 0; cnt < priv->hw_params.max_txq_num; cnt++) {

+			/* skip as we already checked the command queue */

+			if (cnt == IWL_CMD_QUEUE_NUM)

+				continue;

+			if (iwl_check_stuck_queue(priv, cnt))

+				return;

+		}

+	}

+	/*

+	 * Reschedule the timer to occur in

+	 * priv->cfg->monitor_recover_period

+	 */

+	mod_timer(&priv->monitor_recover,

+		jiffies + msecs_to_jiffies(priv->cfg->monitor_recover_period));

+}

+EXPORT_SYMBOL(iwl_bg_monitor_recover);

 #ifdef CONFIG_PM

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.h.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.h

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.h.orig	2010-03-22 15:24:28.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-core.h	2010-03-22 15:48:54.000000000 -0400

@@ -183,6 +183,8 @@ struct iwl_lib_ops {

 	/* temperature */

 	struct iwl_temp_ops temp_ops;

+	/* recover from tx queue stall */

+	void (*recover_from_tx_stall)(unsigned long data);

 };

 struct iwl_ops {

@@ -260,6 +262,8 @@ struct iwl_cfg {

 	const bool broken_powersave;

 	bool use_rts_for_ht;

 	u8 plcp_delta_threshold;

+	/* timer period for monitor the driver queues */

+	u32 monitor_recover_period;

 };

 /***************************

@@ -543,6 +547,9 @@ static inline u16 iwl_pcie_link_ctl(stru

 	pci_read_config_word(priv->pci_dev, pos + PCI_EXP_LNKCTL, &pci_lnk_ctl);

 	return pci_lnk_ctl;

 }

+

+void iwl_bg_monitor_recover(unsigned long data);

+

 #ifdef CONFIG_PM

 int iwl_pci_suspend(struct pci_dev *pdev, pm_message_t state);

 int iwl_pci_resume(struct pci_dev *pdev);

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-dev.h.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-dev.h

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-dev.h.orig	2010-03-22 15:37:04.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-dev.h	2010-03-22 15:48:54.000000000 -0400

@@ -184,6 +184,10 @@ struct iwl_queue {

 	int n_bd;              /* number of BDs in this queue */

 	int write_ptr;       /* 1-st empty entry (index) host_w*/

 	int read_ptr;         /* last used entry (index) host_r*/

+	/* use for monitoring and recovering the stuck queue */

+	int last_read_ptr;      /* storing the last read_ptr */

+	/* number of time read_ptr and last_read_ptr are the same */

+	u8 repeat_same_read_ptr;

 	dma_addr_t dma_addr;   /* physical addr for BD's */

 	int n_window;	       /* safe queue window */

 	u32 id;

@@ -976,6 +980,11 @@ struct traffic_stats {

 #define IWL_DELAY_NEXT_FORCE_RF_RESET  (HZ*3)

 #define IWL_DELAY_NEXT_FORCE_FW_RELOAD (HZ*5)

+/* timer constants use to monitor and recover stuck tx queues in mSecs */

+#define IWL_MONITORING_PERIOD  (1000)

+#define IWL_ONE_HUNDRED_MSECS   (100)

+#define IWL_SIXTY_SECS          (60000)

+

 enum iwl_reset {

 	IWL_RF_RESET = 0,

 	IWL_FW_RESET,

@@ -1275,6 +1284,7 @@ struct iwl_priv {

 	u32 disable_tx_power_cal;

 	struct work_struct run_time_calib_work;

 	struct timer_list statistics_periodic;

+	struct timer_list monitor_recover;

 	bool hw_ready;

 	/*For 3945*/

 #define IWL_DEFAULT_TX_POWER 0x0F

diff -up linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-tx.c.orig linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-tx.c

--- linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-tx.c.orig	2010-03-22 11:07:02.000000000 -0400

+++ linux-2.6.32.noarch/drivers/net/wireless/iwlwifi/iwl-tx.c	2010-03-22 15:48:54.000000000 -0400

@@ -291,6 +291,8 @@ static int iwl_queue_init(struct iwl_pri

 		q->high_mark = 2;

 	q->write_ptr = q->read_ptr = 0;

+	q->last_read_ptr = 0;

+	q->repeat_same_read_ptr = 0;

 	return 0;

 }