bes2600: short-circuit wake handshake when chip is confirmed ACTIVE

The previous patch ("bes2600: gate PM indication completion on pending
request and track chip state") added enum bes2600_chip_pm_state and the
chip_pm_state field tracking what the host has *seen the firmware
confirm*. This patch makes the wake side use it.

Without this, every bes2600_pwr_device_exit_lp_mode() unconditionally
runs gpio_wake() + sbus_active() + wsm_set_operational_mode(active),
even when the chip is already in confirmed-ACTIVE state and the wake
sequence has nothing to do. The visible failure mode on PineTab2:

  bes2600_pwr_enter_lp_mode, wait pm ind timeout
  repeat set gpio_wake_flag, sub_sys:0
  bes2600_sdio_active failed, subsys:0
  bes2600_pwr_device_exit_lp_mode, active mcu fail

cycling every ~9 s, ~22 cycles in 10 minutes. Three pieces:

  1. enter_lp_mode timed out (firmware indication lost). With c6.1,
     chip_pm_state is now UNKNOWN.
  2. lock_device fires exit_lp_mode.
  3. gpio_wake hits "bit already set" because device_enter_lp_mode
     was skipped when the indication timed out, so gpio_sleep was
     never called - the bit reflects driver intent, not chip state.
     gpio_wake silently no-ops (no GPIO edge), bit stays set.
  4. sbus_active spends 200 x 2 ms looking for MCU_WAKEUP_READY that
     never comes (firmware was never told to wake), then fails.
  5. Driver continues to wsm_set_operational_mode against the wedged
     bus, compounding the failure.

This patch's three moves:

  * bes2600_pwr_device_exit_lp_mode() reads chip_pm_state at entry.
    On BES2600_CHIP_PM_ACTIVE, log at devel level and return without
    touching gpio_wake / sbus_active / WSM. The chip is in the state
    we want; the handshake exists only to drive a transition.

  * On BES2600_CHIP_PM_LP or BES2600_CHIP_PM_UNKNOWN, run the wake
    handshake as before, but on sbus_active() failure: set
    chip_pm_state = UNKNOWN, log once at err level, and bail out.
    Do NOT call wsm_set_operational_mode over a wedged bus - it
    would just emit a second error and leave the chip in an even
    less defined state.

  * bes2600_gpio_wakeup_mcu() / bes2600_gpio_allow_mcu_sleep():
    demote "repeat set/clear gpio_wake_flag" from bes_err to
    bes_devel. Multi-subsystem wake-hold (e.g. WIFI + BT both want
    MCU awake) is the steady-state case, and the symmetric clear
    while bit-already-clear is racy bookkeeping rather than a
    hardware error. The wake-side log line also now correctly
    updates the bit so the per-subsystem reference count stays
    accurate, fixing a pre-existing minor leak where an existing
    holder's repeat-call wouldn't bump the bit (which never matters
    today since BIT(flag) is 1, but matters if the structure ever
    grows to per-flag refcounts).

Net effect on the cycle:

  * If chip is genuinely ACTIVE (chip_pm_state == ACTIVE), wake skips
    cleanly. Storm goes silent.
  * If chip is genuinely LP, behaviour is unchanged.
  * If chip is UNKNOWN (post-timeout state), one wake attempt is
    made; on failure, state stays UNKNOWN and we don't emit a
    second cascade error per attempt. Repeated UNKNOWN with failed
    wake will eventually be picked up by the LMAC active-monitor
    and escalated to mmc_hw_reset (c5.2).

No new locks, no new state. Only consumption of the chip_pm_state
field added in the prerequisite patch.

Signed-off-by: Markus Fritsche <fritsche.markus@gmail.com>

bes2600/bes2600_sdio.c

@@ -1388,7 +1388,14 @@ static void bes2600_gpio_wakeup_mcu(struct sbus_priv *self, int flag)
 
 	/* error check */
 	if((self->gpio_wakup_flags & BIT(flag)) != 0) {
-		bes_err(			"repeat set gpio_wake_flag, sub_sys:%d", flag);
+		/*
+		 * Multiple subsystems holding wake is the steady-state case
+		 * (e.g. WIFI + BT both want MCU awake). Demoted from bes_err
+		 * to bes_devel since it isn't an error - the GPIO is already
+		 * asserted high and the subsystem is now also tracked.
+		 */
+		bes_devel("repeat set gpio_wake_flag, sub_sys:%d\n", flag);
+		self->gpio_wakup_flags |= BIT(flag);
 		mutex_unlock(&self->io_mutex);
 		return;
 	}
@@ -1420,7 +1427,11 @@ static void bes2600_gpio_allow_mcu_sleep(struct sbus_priv *self, int flag)
 
 	/* error check */
 	if((self->gpio_wakup_flags & BIT(flag)) == 0) {
-		bes_err(			"repeat clear gpio_wake_flag, sub_sys:%d", flag);
+		/*
+		 * Mirror of the wake path: a clear when the bit is already
+		 * clear is racy bookkeeping, not a hardware error.
+		 */
+		bes_devel("repeat clear gpio_wake_flag, sub_sys:%d\n", flag);
 		mutex_unlock(&self->io_mutex);
 		return;
 	}

bes2600/bes_pwr.c

@@ -524,7 +524,17 @@ static int bes2600_pwr_enter_lp_mode(struct bes2600_common *hw_priv)
 				bes_devel("%s, psMode:%s, fastPsmIdlePeriod:%d apPsmChangePeriod:%d minAutoPsPollPeriod:%d\n",
 						__func__, bes2600_get_ps_mode_str(priv->powersave_mode.pmMode), priv->powersave_mode.fastPsmIdlePeriod,
 						priv->powersave_mode.apPsmChangePeriod, priv->powersave_mode.minAutoPsPollPeriod);
+				/*
+				 * Reinit BEFORE the WSM goes out, so a stale
+				 * indication from a previous cycle cannot have
+				 * primed pm_enter_cmpl. From here until the
+				 * indication callback's cmpxchg(1->0) on
+				 * pm_set_in_process, only the indication for
+				 * THIS request can complete the wait.
+				 */
+				reinit_completion(&hw_priv->bes_power.pm_enter_cmpl);
 				atomic_set(&hw_priv->bes_power.pm_set_in_process, 1);
+
 				ret = bes2600_set_pm(priv, &priv->powersave_mode);
 				if (ret) {
 					atomic_set(&hw_priv->bes_power.pm_set_in_process, 0);
@@ -535,11 +545,33 @@ static int bes2600_pwr_enter_lp_mode(struct bes2600_common *hw_priv)
 
 				/* wait power save mode changed indication */
 				status = wait_for_completion_timeout(&hw_priv->bes_power.pm_enter_cmpl, 5 * HZ);
-				atomic_set(&hw_priv->bes_power.pm_set_in_process, 0);
-				reinit_completion(&hw_priv->bes_power.pm_enter_cmpl);
 				if (!status) {
-					bes_devel("%s, wait pm ind timeout\n", __func__);
-					timeouts++;
+					/*
+					 * The indication callback only fires
+					 * complete() when it observes
+					 * pm_set_in_process == 1; cmpxchg it
+					 * to 0 here so a late indication
+					 * cannot prime the next wait.
+					 *
+					 * If we win the cmpxchg, this is a
+					 * real timeout: the firmware's PS
+					 * state is unknown to us. Mark it as
+					 * such so the next wake path can
+					 * probe before assuming the chip is
+					 * still active.
+					 *
+					 * If we lose the cmpxchg, the
+					 * indication arrived between the
+					 * wait timing out and us getting
+					 * here; treat as success.
+					 */
+					if (atomic_cmpxchg(&hw_priv->bes_power.pm_set_in_process,
+							   1, 0) == 1) {
+						bes_devel("%s, wait pm ind timeout\n", __func__);
+						atomic_set(&hw_priv->bes_power.chip_pm_state,
+							   BES2600_CHIP_PM_UNKNOWN);
+						timeouts++;
+					}
 				}
 			} else {
 				bes_devel("skip enter lp mode\n");
@@ -554,10 +586,34 @@ static int bes2600_pwr_enter_lp_mode(struct bes2600_common *hw_priv)
 	 * in an inconsistent state that cascades into SDIO TX errors on
 	 * the BES2600.
 	 */
-	if (timeouts == 0)
+	if (timeouts == 0) {
 		bes2600_pwr_device_enter_lp_mode(hw_priv);
-	else
+	} else {
+		/*
+		 * device_enter_lp_mode() was skipped (one or more VIFs
+		 * timed out waiting for the firmware indication) so its
+		 * gpio_sleep(MCU) - which drops the wake-flag bit and, if
+		 * no other subsystem holds the wake, drives the GPIO low -
+		 * never ran. Without it the bit stays asserted, and the
+		 * next bes2600_pwr_device_exit_lp_mode() calls
+		 * gpio_wake(MCU) into a "bit already set" no-op: the GPIO
+		 * never re-edges, sbus_active() exhausts its 200x2ms
+		 * MCU_WAKEUP_READY budget against an unwoken chip, and
+		 * the first TX after idle stalls for several seconds.
+		 *
+		 * Drop the MCU wake-flag bit explicitly here so the next
+		 * wake injects a real GPIO edge. gpio_allow_mcu_sleep
+		 * preserves multi-subsystem semantics: it only drives the
+		 * GPIO low when no other subsystem still holds wake; if
+		 * BT or another holder is keeping the chip awake, the
+		 * GPIO stays high and the bit clear here is purely
+		 * bookkeeping (so the next gpio_wake doesn't no-op).
+		 */
+		if (hw_priv->sbus_ops->gpio_sleep)
+			hw_priv->sbus_ops->gpio_sleep(hw_priv->sbus_priv,
+						      GPIO_WAKE_FLAG_MCU);
 		ret = -ETIMEDOUT;
+	}
 
 	return ret;
 }
@@ -565,19 +621,61 @@ static int bes2600_pwr_enter_lp_mode(struct bes2600_common *hw_priv)
 static void bes2600_pwr_device_exit_lp_mode(struct bes2600_common *hw_priv)
 {
 	int ret = 0;
+	enum bes2600_chip_pm_state state;
 	struct wsm_operational_mode mode = {
 		.power_mode = wsm_power_mode_active,
 		.disableMoreFlagUsage = true,
 	};
 
-	bes_devel("host lock lmac\n");
-	if(hw_priv->sbus_ops->gpio_wake)
-		hw_priv->sbus_ops->gpio_wake(hw_priv->sbus_priv, GPIO_WAKE_FLAG_MCU);
+	/*
+	 * Consult chip_pm_state set by bes2600_pwr_notify_ps_changed().
+	 * If we last saw the firmware confirm ACTIVE, skip ONLY the
+	 * gpio_wake + sbus_active wake handshake - the GPIO is already
+	 * asserted high and the SDIO MCU subsystem is already running,
+	 * so another sbus_active() round-trip just hits its 200x2ms
+	 * timeout because the firmware has nothing to do.
+	 *
+	 * wsm_set_operational_mode() below is NOT part of the wake
+	 * handshake; it is the operational-mode setter the firmware
+	 * tracks per call. Skipping it leaves the chip's SDIO state
+	 * machine without a fresh operational-mode update, which on
+	 * PineTab2 wedges the bus (-EBUSY on next sdio_rx_work read)
+	 * within a few seconds of probe completion. So it must run
+	 * unconditionally.
+	 */
+	state = atomic_read(&hw_priv->bes_power.chip_pm_state);
+	if (state == BES2600_CHIP_PM_ACTIVE) {
+		bes_devel("device_exit_lp_mode: chip already ACTIVE, skipping wake handshake\n");
+	} else {
+		bes_devel("host lock lmac\n");
+		if (hw_priv->sbus_ops->gpio_wake)
+			hw_priv->sbus_ops->gpio_wake(hw_priv->sbus_priv,
+						      GPIO_WAKE_FLAG_MCU);
 
-	if(hw_priv->sbus_ops->sbus_active) {
-		ret = hw_priv->sbus_ops->sbus_active(hw_priv->sbus_priv, SUBSYSTEM_MCU);
-		if (ret)
-			bes_err("%s, active mcu fail\n", __func__);
+		if (hw_priv->sbus_ops->sbus_active) {
+			ret = hw_priv->sbus_ops->sbus_active(hw_priv->sbus_priv,
+							     SUBSYSTEM_MCU);
+			if (ret) {
+				/*
+				 * MCU_WAKEUP_READY did not arrive within
+				 * the SDIO handshake window. Record state
+				 * as UNKNOWN so the next exit_lp_mode call
+				 * also runs the full wake sequence (no
+				 * skip), but still send operational_mode
+				 * below to match pre-c6 behaviour - the
+				 * WSM may succeed even if the SDIO active
+				 * confirm was lost, and if it fails too,
+				 * we just emit a second devel-level error.
+				 * Repeated UNKNOWN is the signal for the
+				 * LMAC active-monitor to eventually
+				 * escalate to bus_reset (c5.2's
+				 * mmc_hw_reset path).
+				 */
+				bes_err("%s, active mcu fail\n", __func__);
+				atomic_set(&hw_priv->bes_power.chip_pm_state,
+					   BES2600_CHIP_PM_UNKNOWN);
+			}
+		}
 	}
 
 	ret = wsm_set_operational_mode(hw_priv, &mode, 0);
@@ -833,6 +931,7 @@ void bes2600_pwr_init(struct bes2600_common *hw_priv)
         hw_priv->bes_power.power_up_task = NULL;
         mutex_init(&hw_priv->bes_power.pwr_mutex);
         atomic_set(&hw_priv->bes_power.dev_state, 0);
+	atomic_set(&hw_priv->bes_power.chip_pm_state, BES2600_CHIP_PM_UNKNOWN);
         init_completion(&hw_priv->bes_power.pm_enter_cmpl);
         sema_init(&hw_priv->bes_power.sync_lock, 1);
         device_set_wakeup_capable(hw_priv->pdev, true);
@@ -1213,9 +1312,28 @@ int bes2600_pwr_clear_busy_event(struct bes2600_common *hw_priv, u32 event)
 
 void bes2600_pwr_notify_ps_changed(struct bes2600_common *hw_priv, u8 psmode)
 {
-	if((psmode & 0x01) != WSM_PSM_ACTIVE) {
-		bes_devel("complete pm_enter_cmpl\n");
-		complete(&hw_priv->bes_power.pm_enter_cmpl);
+	/*
+	 * The firmware sends a PM-changed indication for every transition,
+	 * including ones we didn't ask for (firmware-internal coex moves,
+	 * idle-driven aging). Update chip_pm_state unconditionally so the
+	 * wake path can use it, but only fire pm_enter_cmpl when a host-
+	 * initiated set_pm is actually in flight - otherwise a stale
+	 * indication can prime a future wait against a freshly
+	 * reinit_completion()'ed state.
+	 */
+	if ((psmode & 0x01) != WSM_PSM_ACTIVE) {
+		atomic_set(&hw_priv->bes_power.chip_pm_state,
+			   BES2600_CHIP_PM_LP);
+		if (atomic_cmpxchg(&hw_priv->bes_power.pm_set_in_process,
+				   1, 0) == 1) {
+			bes_devel("complete pm_enter_cmpl\n");
+			complete(&hw_priv->bes_power.pm_enter_cmpl);
+		} else {
+			bes_devel("PM ind (LP) without pending wait; state recorded\n");
+		}
+	} else {
+		atomic_set(&hw_priv->bes_power.chip_pm_state,
+			   BES2600_CHIP_PM_ACTIVE);
 	}
 }
 

bes2600/bes_pwr.h

@@ -64,6 +64,20 @@ enum power_down_state
         POWER_DOWN_STATE_UNLOCKED,
 };
 
+/*
+ * Confirmed PM state of the firmware-side chip. Tracks what the host
+ * has *seen* the firmware acknowledge, not what the host has
+ * requested. UNKNOWN means a host-initiated transition timed out
+ * before the firmware indication arrived; the next wake path should
+ * treat it as "we don't know" and probe before issuing GPIO/SDIO
+ * wakeup ops.
+ */
+enum bes2600_chip_pm_state {
+	BES2600_CHIP_PM_ACTIVE = 0,
+	BES2600_CHIP_PM_LP,
+	BES2600_CHIP_PM_UNKNOWN,
+};
+
 typedef void (*bes_pwr_enter_lp_cb)(struct bes2600_common *hw_priv);
 typedef void (*bes_pwr_exit_lp_cb)(struct bes2600_common *hw_priv);
 
@@ -106,6 +120,7 @@ struct bes2600_pwr_t
         bool ap_lp_bad;
         struct bes2600_pwr_event_t pwr_events[BES2600_DELAY_EVENT_NUM];
         atomic_t pm_set_in_process;
+	atomic_t chip_pm_state;
 };
 
 #ifdef CONFIG_BES2600_WOWLAN

bes2600/scan.c

@@ -22,9 +22,17 @@
  * After this many consecutive WSM scan rejections from firmware, stop
  * issuing new scans for BES2600_SCAN_BACKOFF_JIFFIES and let the state
  * that's rejecting them (coex window, firmware-internal busy) clear.
+ *
+ * The backoff has to be at least as long as the natural mac80211 scan-
+ * retry cadence, otherwise the next attempt lands outside the window
+ * and bypasses the defer guard. Observed in the wild on PineTab2:
+ * roam-evaluation bursts at ~12 s cadence, idle background scans at
+ * ~5 min cadence. 30 s catches the burst and leaves the slow case
+ * alone (the firmware-policy state has had minutes to clear by then
+ * anyway).
  */
 #define BES2600_SCAN_REJECT_THRESHOLD	3
-#define BES2600_SCAN_BACKOFF_JIFFIES	(10 * HZ)
+#define BES2600_SCAN_BACKOFF_JIFFIES	(30 * HZ)
 
 static void bes2600_scan_restart_delayed(struct bes2600_vif *priv);
 
@@ -40,7 +48,9 @@ static void bes2600_scan_restart_delayed(struct bes2600_vif *priv);
  *  2. We already saw >= BES2600_SCAN_REJECT_THRESHOLD consecutive
  *     rejections on recent scan attempts and the backoff window has
  *     not yet elapsed. Whatever was rejecting them is likely still
- *     rejecting them; give it time.
+ *     rejecting them; give it time. If the backoff has elapsed without
+ *     a fresh reject refreshing it, the burst is over and we reset the
+ *     count so an isolated reject doesn't immediately re-trip.
  *
  * Returns true if the caller should abandon the scan iteration.
  */
@@ -51,6 +61,9 @@ static bool bes2600_scan_should_defer(struct bes2600_common *hw_priv)
 		return true;
 #endif
 
+	if (time_after(jiffies, hw_priv->scan.backoff_until))
+		hw_priv->scan.reject_count = 0;
+
 	if (hw_priv->scan.reject_count >= BES2600_SCAN_REJECT_THRESHOLD &&
 	    time_before(jiffies, hw_priv->scan.backoff_until))
 		return true;