[Copybara Auto Merge] Merge branch zuma into android13-gs-pixel-5.15

edgetpu: Use thermal throttling KCI instead of set_policy
Bug: 258427511
edgetpu: Add a static debug pointer to driver state
Bug: 260768055

Signed-off-by: Zuma copybara merger <zuma-automerger@google.com>
GitOrigin-RevId: aaa7370d2eb6b4505495736b461a21736d449b59
Change-Id: Ib14b0b31209169197c2f507b82c55cf5f1e2825d

7 files changed, 45 insertions, 32 deletions

diff --git a/drivers/edgetpu/edgetpu-firmware.c b/drivers/edgetpu/edgetpu-firmware.c
index fc26c2f..4d5d853 100644
--- a/drivers/edgetpu/edgetpu-firmware.c
+++ b/drivers/edgetpu/edgetpu-firmware.c
@@ -176,6 +176,13 @@ static int edgetpu_firmware_handshake(struct edgetpu_firmware *et_fw)
 	ret = edgetpu_telemetry_kci(etdev);
 	if (ret)
 		etdev_warn(etdev, "telemetry KCI error: %d", ret);
+
+	if (etdev->thermal->cooling_state) {
+		ret = edgetpu_thermal_resume(etdev->dev);
+		if (ret)
+			etdev_warn(etdev, "Thermal resume error: %d", ret);
+	}
+
 	/* Set debug dump buffer in FW */
 	edgetpu_get_debug_dump(etdev, 0);
 	return 0;
diff --git a/drivers/edgetpu/edgetpu-mobile-platform.c b/drivers/edgetpu/edgetpu-mobile-platform.c
index 3ee70af..3c7e302 100644
--- a/drivers/edgetpu/edgetpu-mobile-platform.c
+++ b/drivers/edgetpu/edgetpu-mobile-platform.c
@@ -23,6 +23,8 @@
 #include "mobile-firmware.h"
 #include "mobile-pm.h"
 
+static struct edgetpu_dev *edgetpu_debug_pointer;
+
 static void set_telemetry_mem(struct edgetpu_mobile_platform_dev *etmdev,
 			      enum gcip_telemetry_type type, struct edgetpu_coherent_mem *mem)
 {
@@ -426,6 +428,8 @@ static int edgetpu_mobile_platform_probe(struct platform_device *pdev,
 	/* Turn the device off unless a client request is already received. */
 	edgetpu_pm_shutdown(etdev, false);
 
+	edgetpu_debug_pointer = etdev;
+
 	return 0;
 out_destroy_fw:
 	edgetpu_mobile_firmware_destroy(etdev);
@@ -460,5 +464,8 @@ static int edgetpu_mobile_platform_remove(struct platform_device *pdev)
 	edgetpu_pm_put(etdev->pm);
 	edgetpu_pm_shutdown(etdev, true);
 	edgetpu_mobile_pm_destroy(etdev);
+
+	edgetpu_debug_pointer = NULL;
+
 	return 0;
 }
diff --git a/drivers/edgetpu/edgetpu-soc.h b/drivers/edgetpu/edgetpu-soc.h
index 53d06d2..98b1f3d 100644
--- a/drivers/edgetpu/edgetpu-soc.h
+++ b/drivers/edgetpu/edgetpu-soc.h
@@ -32,9 +32,6 @@ long edgetpu_soc_pm_get_rate(struct edgetpu_dev *etdev, int flags);
 /* Power management set TPU clock rate */
 int edgetpu_soc_pm_set_rate(unsigned long rate);
 
-/* Set PM policy */
-int edgetpu_soc_pm_set_policy(u64 val);
-
 /* Power down */
 void edgetpu_soc_pm_power_down(struct edgetpu_dev *etdev);
 
diff --git a/drivers/edgetpu/edgetpu-thermal.h b/drivers/edgetpu/edgetpu-thermal.h
index dbd283f..bcfed04 100644
--- a/drivers/edgetpu/edgetpu-thermal.h
+++ b/drivers/edgetpu/edgetpu-thermal.h
@@ -56,6 +56,13 @@ int edgetpu_thermal_suspend(struct device *dev);
 int edgetpu_thermal_resume(struct device *dev);
 
 /*
+ * Sends the thermal throttling KCI if the device is powered.
+ *
+ * Returns the return value of KCI if the device is powered, otherwise 0.
+ */
+int edgetpu_thermal_kci_if_powered(struct edgetpu_dev *etdev, u32 state);
+
+/*
  * Holds thermal->lock.
  *
  * Does nothing if the thermal management is not supported.
diff --git a/drivers/edgetpu/mobile-pm.c b/drivers/edgetpu/mobile-pm.c
index 9a6a21e..aaf6a2f 100644
--- a/drivers/edgetpu/mobile-pm.c
+++ b/drivers/edgetpu/mobile-pm.c
@@ -18,6 +18,7 @@
 #include "edgetpu-mailbox.h"
 #include "edgetpu-mobile-platform.h"
 #include "edgetpu-pm.h"
+#include "edgetpu-thermal.h"
 #include "mobile-firmware.h"
 #include "mobile-pm.h"
 
@@ -148,7 +149,7 @@ static int mobile_pwr_policy_set(void *data, u64 val)
 	int ret;
 
 	mutex_lock(&platform_pwr->policy_lock);
-	ret = edgetpu_soc_pm_set_policy(val);
+	ret = edgetpu_thermal_kci_if_powered(etdev, val);
 
 	if (ret) {
 		dev_err(etmdev->edgetpu_dev.dev,
diff --git a/drivers/edgetpu/mobile-soc-gsx01.c b/drivers/edgetpu/mobile-soc-gsx01.c
index 104cf46..baefb33 100644
--- a/drivers/edgetpu/mobile-soc-gsx01.c
+++ b/drivers/edgetpu/mobile-soc-gsx01.c
@@ -358,11 +358,6 @@ int edgetpu_soc_pm_set_rate(unsigned long rate)
 	return -EOPNOTSUPP;
 }
 
-int edgetpu_soc_pm_set_policy(u64 val)
-{
-	return 0;
-}
-
 static int edgetpu_core_rate_get(void *data, u64 *val)
 {
 	struct edgetpu_dev *etdev = (typeof(etdev))data;
diff --git a/drivers/edgetpu/mobile-thermal.c b/drivers/edgetpu/mobile-thermal.c
index c8fb0a2..73b68d3 100644
--- a/drivers/edgetpu/mobile-thermal.c
+++ b/drivers/edgetpu/mobile-thermal.c
@@ -30,26 +30,6 @@
 #define OF_DATA_NUM_MAX (MAX_NUM_TPU_STATES * 2)
 static struct edgetpu_state_pwr state_pwr_map[MAX_NUM_TPU_STATES] = {0};
 
-/*
- * Sends the thermal throttling KCI if the device is powered.
- *
- * Returns the return value of KCI if the device is powered, otherwise 0.
- */
-static int edgetpu_thermal_kci_if_powered(struct edgetpu_dev *etdev, enum edgetpu_pwr_state state)
-{
-	int ret = 0;
-
-	if (edgetpu_pm_get_if_powered(etdev->pm)) {
-		ret = edgetpu_kci_notify_throttling(etdev, state);
-		if (ret)
-			etdev_err_ratelimited(etdev,
-					      "Failed to notify FW about power state %u, error:%d",
-					      state, ret);
-		edgetpu_pm_put(etdev->pm);
-	}
-	return ret;
-}
-
 static int edgetpu_get_max_state(struct thermal_cooling_device *cdev, unsigned long *state)
 {
 	struct edgetpu_thermal *thermal = cdev->devdata;
@@ -71,7 +51,7 @@ static int edgetpu_set_thermal_policy(struct edgetpu_dev *etdev, unsigned long p
 	if (edgetpu_is_powered(etdev))
 		edgetpu_kci_block_bus_speed_control(etdev, true);
 
-	ret = edgetpu_soc_pm_set_policy(pwr_state);
+	ret = edgetpu_thermal_kci_if_powered(etdev, pwr_state);
 
 	if (edgetpu_is_powered(etdev))
 		edgetpu_kci_block_bus_speed_control(etdev, false);
@@ -140,7 +120,7 @@ static int edgetpu_get_cur_state(struct thermal_cooling_device *cdev, unsigned l
 	dev_warn(cooling->dev, "Unknown cooling state: %lu, resetting\n", *state);
 	mutex_lock(&cooling->lock);
 
-	ret = edgetpu_soc_pm_set_policy(TPU_ACTIVE_NOM);
+	ret = edgetpu_thermal_kci_if_powered(cooling->etdev, TPU_ACTIVE_NOM);
 	if (ret) {
 		dev_err(cooling->dev, "error setting tpu policy: %d\n", ret);
 		mutex_unlock(&cooling->lock);
@@ -443,7 +423,11 @@ int edgetpu_thermal_resume(struct device *dev)
 	if (IS_ERR(cooling))
 		return PTR_ERR(cooling);
 	mutex_lock(&cooling->lock);
-	ret = edgetpu_thermal_kci_if_powered(etdev, state_pwr_map[0].state);
+
+	if (cooling->cooling_state >= cooling->tpu_num_states)
+		cooling->cooling_state = 0;
+
+	ret = edgetpu_thermal_kci_if_powered(etdev, state_pwr_map[cooling->cooling_state].state);
 	/*
 	 * Unlike edgetpu_thermal_suspend(), only set the device is resumed if the FW handled the
 	 * KCI request.
@@ -453,3 +437,18 @@ int edgetpu_thermal_resume(struct device *dev)
 	mutex_unlock(&cooling->lock);
 	return ret;
 }
+
+int edgetpu_thermal_kci_if_powered(struct edgetpu_dev *etdev, u32 state)
+{
+	int ret = 0;
+
+	if (edgetpu_pm_get_if_powered(etdev->pm)) {
+		ret = edgetpu_kci_notify_throttling(etdev, state);
+		if (ret)
+			etdev_err_ratelimited(etdev,
+					      "Failed to notify FW about power state %u, error:%d",
+					      state, ret);
+		edgetpu_pm_put(etdev->pm);
+	}
+	return ret;
+}