[Copybara Auto Merge] Merge branch zuma into android14-gs-pixel-5.15

edgetpu: Add a debugfs node for thermal control
Bug: 266837631
edgetpu: Add missing pm error handling
gcip: use v*alloc without node
gcip: Add a new KCI code for thermal control
Bug: 266837631 (repeat)
gcip: Add a new KCI code for fw tracing level request
Bug: 262916889
gcip: fix up DMA fence macros
gcip: fix code style/formatting issues
edgetpu: fix firmware size check
Bug: 267206706
gcip: conditionally zero the non-contiguous region

Signed-off-by: Zuma copybara merger <zuma-automerger@google.com>
GitOrigin-RevId: 5bf0de6e2161bdd772fa8207f9a79ecbde14c54f
Change-Id: I433bbaae480764936f1ac5ba2230902d5f28321f

12 files changed, 150 insertions, 36 deletions

diff --git a/drivers/edgetpu/edgetpu-kci.c b/drivers/edgetpu/edgetpu-kci.c
index 194c9fb..ed9b9e2 100644
--- a/drivers/edgetpu/edgetpu-kci.c
+++ b/drivers/edgetpu/edgetpu-kci.c
@@ -636,6 +636,21 @@ int edgetpu_kci_block_bus_speed_control(struct edgetpu_dev *etdev, bool block)
 	return gcip_kci_send_cmd(etdev->etkci->kci, &cmd);
 }
 
+int edgetpu_kci_thermal_control(struct edgetpu_dev *etdev, bool enable)
+{
+	struct gcip_kci_command_element cmd = {
+		.code = GCIP_KCI_CODE_THERMAL_CONTROL,
+		.dma = {
+			.flags = (u32)enable,
+		},
+	};
+
+	if (!etdev->etkci)
+		return -ENODEV;
+
+	return gcip_kci_send_cmd(etdev->etkci->kci, &cmd);
+}
+
 int edgetpu_kci_resp_rkci_ack(struct edgetpu_dev *etdev, struct gcip_kci_response_element *rkci_cmd)
 {
 	struct gcip_kci_command_element cmd = {
diff --git a/drivers/edgetpu/edgetpu-kci.h b/drivers/edgetpu/edgetpu-kci.h
index 0368be6..6b75306 100644
--- a/drivers/edgetpu/edgetpu-kci.h
+++ b/drivers/edgetpu/edgetpu-kci.h
@@ -197,6 +197,14 @@ int edgetpu_kci_notify_throttling(struct edgetpu_dev *etdev, u32 level);
 int edgetpu_kci_block_bus_speed_control(struct edgetpu_dev *etdev, bool block);
 
 /*
+ * Request the firmware to enable or disable the thermal throttling.
+ *
+ * Returns KCI response code on success or < 0 on error (typically -ETIMEDOUT).
+ */
+
+int edgetpu_kci_thermal_control(struct edgetpu_dev *etdev, bool enable);
+
+/*
  * Send an ack to the FW after handling a reverse KCI request.
  *
  * The FW may wait for a response from the kernel for an RKCI request so a
diff --git a/drivers/edgetpu/edgetpu-mobile-platform.c b/drivers/edgetpu/edgetpu-mobile-platform.c
index 3c7e302..1af3bde 100644
--- a/drivers/edgetpu/edgetpu-mobile-platform.c
+++ b/drivers/edgetpu/edgetpu-mobile-platform.c
@@ -56,7 +56,7 @@ static int edgetpu_platform_setup_fw_region(struct edgetpu_mobile_platform_dev *
 	struct resource r;
 	struct device_node *np;
 	int err;
-	size_t region_map_size = EDGETPU_DEFAULT_FW_SIZE_MAX + EDGETPU_DEFAULT_REMAPPED_DATA_SIZE;
+	size_t region_map_size = EDGETPU_MAX_FW_LIMIT;
 
 	np = of_parse_phandle(dev->of_node, "memory-region", 0);
 	if (!np) {
@@ -93,7 +93,7 @@ static int edgetpu_platform_setup_fw_region(struct edgetpu_mobile_platform_dev *
 	}
 
 	etmdev->fw_region_paddr = r.start;
-	etmdev->fw_region_size = EDGETPU_DEFAULT_FW_SIZE_MAX;
+	etmdev->fw_region_size = EDGETPU_DEFAULT_FW_LIMIT;
 
 	etmdev->remapped_data_addr = EDGETPU_INSTRUCTION_REMAP_BASE + etmdev->fw_region_size;
 	etmdev->remapped_data_size = EDGETPU_DEFAULT_REMAPPED_DATA_SIZE;
diff --git a/drivers/edgetpu/edgetpu-thermal.h b/drivers/edgetpu/edgetpu-thermal.h
index b96d961..98bc4df 100644
--- a/drivers/edgetpu/edgetpu-thermal.h
+++ b/drivers/edgetpu/edgetpu-thermal.h
@@ -27,6 +27,7 @@ struct edgetpu_thermal {
 	unsigned int tpu_num_states;
 	struct edgetpu_dev *etdev;
 	bool thermal_suspended; /* TPU thermal suspended state */
+	bool enabled;
 };
 
 struct edgetpu_state_pwr {
diff --git a/drivers/edgetpu/gcip-kernel-driver/drivers/gcip/gcip-alloc-helper.c b/drivers/edgetpu/gcip-kernel-driver/drivers/gcip/gcip-alloc-helper.c
index 33c95e2..85af8e5 100644
--- a/drivers/edgetpu/gcip-kernel-driver/drivers/gcip/gcip-alloc-helper.c
+++ b/drivers/edgetpu/gcip-kernel-driver/drivers/gcip/gcip-alloc-helper.c
@@ -7,6 +7,7 @@
 
 #include <asm/page.h>
 #include <linux/device.h>
+#include <linux/gfp.h>
 #include <linux/mm_types.h>
 #include <linux/scatterlist.h>
 #include <linux/slab.h>
@@ -47,7 +48,10 @@ struct sg_table *gcip_alloc_noncontiguous(struct device *dev, size_t size, gfp_t
 
 	size = PAGE_ALIGN(size);
 	count = size >> PAGE_SHIFT;
-	mem = vzalloc_node(size, dev_to_node(dev));
+	if (gfp & __GFP_ZERO)
+		mem = vzalloc(size);
+	else
+		mem = vmalloc(size);
 	if (!mem) {
 		dev_err(dev, "GCIP noncontiguous alloc size=%#zx failed", size);
 		goto err_free_sh;
diff --git a/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-dma-fence.h b/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-dma-fence.h
index c0a7d68..1d4030a 100644
--- a/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-dma-fence.h
+++ b/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-dma-fence.h
@@ -17,8 +17,9 @@
 #define GCIP_FENCE_TIMELINE_NAME_LEN 128
 
 /* Used before accessing the list headed by mgr->fence_list_head. */
-#define GCIP_DMA_FENCE_LIST_LOCK(mgr, flags) spin_lock_irqsave(&mgr->fence_list_lock, flags)
-#define GCIP_DMA_FENCE_LIST_UNLOCK(mgr, flags) spin_unlock_irqrestore(&mgr->fence_list_lock, flags)
+#define GCIP_DMA_FENCE_LIST_LOCK(mgr, flags) spin_lock_irqsave(&(mgr)->fence_list_lock, flags)
+#define GCIP_DMA_FENCE_LIST_UNLOCK(mgr, flags)                                                     \
+	spin_unlock_irqrestore(&(mgr)->fence_list_lock, flags)
 
 /*
  * A macro to loop through all fences under a gcip_dma_fence_manager.
@@ -28,7 +29,7 @@
  * This macro must be wrapped by GCIP_DMA_FENCE_LIST_(UN)LOCK.
  */
 #define gcip_for_each_fence(mgr, gfence)                                                           \
-	list_for_each_entry(gfence, &mgr->fence_list_head, fence_list)
+	list_for_each_entry(gfence, &(mgr)->fence_list_head, fence_list)
 
 #define to_gcip_fence(fence) container_of(fence, struct gcip_dma_fence, fence)
 
diff --git a/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-image-config.h b/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-image-config.h
index 1d00ef9..bcc506f 100644
--- a/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-image-config.h
+++ b/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-image-config.h
@@ -92,27 +92,19 @@ struct gcip_image_config_parser {
 /* For decoding the size of ns_iommu_mappings. */
 static inline u32 gcip_ns_config_to_size(u32 cfg)
 {
-	u32 size;
-
 	if (cfg & GCIP_IMG_CFG_SIZE_MODE_BIT)
-		size = (cfg & GCIP_IMG_CFG_NS_SIZE_MASK) << PAGE_SHIFT;
-	else
-		size = (cfg & GCIP_IMG_CFG_NS_SIZE_MASK) << GCIP_IMG_CFG_MB_SHIFT;
+		return (cfg & GCIP_IMG_CFG_NS_SIZE_MASK) << PAGE_SHIFT;
 
-	return size;
+	return (cfg & GCIP_IMG_CFG_NS_SIZE_MASK) << GCIP_IMG_CFG_MB_SHIFT;
 }
 
 /* For decoding the size of iommu_mappings. */
 static inline u32 gcip_config_to_size(u32 cfg)
 {
-	u32 page_size;
-
 	if (cfg & GCIP_IMG_CFG_SIZE_MODE_BIT)
-		page_size = cfg & GCIP_IMG_CFG_SECURE_SIZE_MASK;
-	else
-		page_size = BIT(cfg & GCIP_IMG_CFG_SECURE_SIZE_MASK);
+		return (cfg & GCIP_IMG_CFG_SECURE_SIZE_MASK) << PAGE_SHIFT;
 
-	return page_size << PAGE_SHIFT;
+	return BIT(cfg & GCIP_IMG_CFG_SECURE_SIZE_MASK) << PAGE_SHIFT;
 }
 
 /*
diff --git a/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-kci.h b/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-kci.h
index 3649779..03cc078 100644
--- a/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-kci.h
+++ b/drivers/edgetpu/gcip-kernel-driver/include/gcip/gcip-kci.h
@@ -97,6 +97,8 @@ enum gcip_kci_code {
 	GCIP_KCI_CODE_RELEASE_VMBOX = 16,
 	GCIP_KCI_CODE_LINK_OFFLOAD_VMBOX = 17,
 	GCIP_KCI_CODE_UNLINK_OFFLOAD_VMBOX = 18,
+	GCIP_KCI_CODE_FIRMWARE_TRACING_LEVEL = 19,
+	GCIP_KCI_CODE_THERMAL_CONTROL = 20,
 
 	GCIP_KCI_CODE_RKCI_ACK = 256,
 };
@@ -109,11 +111,11 @@ enum gcip_kci_code {
  */
 enum gcip_reverse_kci_code {
 	GCIP_RKCI_CHIP_CODE_FIRST = 0,
-	GCIP_RKCI_PM_QOS_REQUEST = GCIP_RKCI_CHIP_CODE_FIRST + 1,
-	GCIP_RKCI_CHANGE_BTS_SCENARIO = GCIP_RKCI_CHIP_CODE_FIRST + 2,
-	GCIP_RKCI_PM_QOS_BTS_REQUEST = GCIP_RKCI_CHIP_CODE_FIRST + 3,
-	GCIP_RKCI_DSP_CORE_TELEMETRY_TRY_READ = GCIP_RKCI_CHIP_CODE_FIRST + 4,
-	GCIP_RKCI_CLIENT_FATAL_ERROR_NOTIFY = GCIP_RKCI_CHIP_CODE_FIRST + 5,
+	GCIP_RKCI_PM_QOS_REQUEST,
+	GCIP_RKCI_CHANGE_BTS_SCENARIO,
+	GCIP_RKCI_PM_QOS_BTS_REQUEST,
+	GCIP_RKCI_DSP_CORE_TELEMETRY_TRY_READ,
+	GCIP_RKCI_CLIENT_FATAL_ERROR_NOTIFY,
 	GCIP_RKCI_CHIP_CODE_LAST = 0x7FFF,
 	GCIP_RKCI_GENERIC_CODE_FIRST = 0x8000,
 	GCIP_RKCI_FIRMWARE_CRASH = GCIP_RKCI_GENERIC_CODE_FIRST + 0,
diff --git a/drivers/edgetpu/mobile-firmware.c b/drivers/edgetpu/mobile-firmware.c
index de6b4cd..e8f156f 100644
--- a/drivers/edgetpu/mobile-firmware.c
+++ b/drivers/edgetpu/mobile-firmware.c
@@ -99,11 +99,9 @@ static int mobile_firmware_alloc_buffer(
 		struct edgetpu_firmware_buffer *fw_buf)
 {
 	struct edgetpu_dev *etdev = et_fw->etdev;
-	struct edgetpu_mobile_platform_dev *etmdev = to_mobile_dev(etdev);
 
 	/* Allocate extra space the image header */
-	size_t buffer_size =
-		etmdev->fw_region_size + MOBILE_FW_HEADER_SIZE;
+	size_t buffer_size = EDGETPU_MAX_FW_LIMIT + MOBILE_FW_HEADER_SIZE;
 
 	fw_buf->vaddr = vmalloc(buffer_size);
 	if (!fw_buf->vaddr) {
@@ -351,7 +349,7 @@ static int mobile_firmware_setup_buffer(struct edgetpu_firmware *et_fw,
 	etmdev->fw_region_size =
 		image_config->remapped_data_start ?
 			image_config->remapped_data_start - image_config->firmware_base :
-			EDGETPU_DEFAULT_FW_SIZE_MAX;
+			EDGETPU_DEFAULT_FW_LIMIT;
 
 	image_vaddr = memremap(etmdev->fw_region_paddr, etmdev->fw_region_size, MEMREMAP_WC);
 	if (!image_vaddr) {
diff --git a/drivers/edgetpu/mobile-pm.c b/drivers/edgetpu/mobile-pm.c
index aaf6a2f..6ab65bf 100644
--- a/drivers/edgetpu/mobile-pm.c
+++ b/drivers/edgetpu/mobile-pm.c
@@ -354,9 +354,8 @@ static int mobile_pm_after_create(struct edgetpu_pm *etpm)
 
 	ret = pm_runtime_get_sync(dev);
 	if (ret) {
-		pm_runtime_put_noidle(dev);
 		dev_err(dev, "pm_runtime_get_sync returned %d\n", ret);
-		return ret;
+		goto err_pm_runtime_put;
 	}
 
 	mutex_init(&platform_pwr->policy_lock);
@@ -374,7 +373,21 @@ static int mobile_pm_after_create(struct edgetpu_pm *etpm)
 		debugfs_create_file("policy", 0660, platform_pwr->debugfs_dir, etdev,
 				    &fops_tpu_pwr_policy);
 	}
-	return edgetpu_soc_pm_init(etdev);
+
+	ret = edgetpu_soc_pm_init(etdev);
+	if (ret)
+		goto err_debugfs_remove;
+
+	return 0;
+
+err_debugfs_remove:
+	debugfs_remove_recursive(platform_pwr->debugfs_dir);
+
+err_pm_runtime_put:
+	pm_runtime_put_noidle(dev);
+	pm_runtime_disable(dev);
+
+	return ret;
 }
 
 static void mobile_pm_before_destroy(struct edgetpu_pm *etpm)
diff --git a/drivers/edgetpu/mobile-thermal.c b/drivers/edgetpu/mobile-thermal.c
index 4794053..a1bbda5 100644
--- a/drivers/edgetpu/mobile-thermal.c
+++ b/drivers/edgetpu/mobile-thermal.c
@@ -292,6 +292,70 @@ static ssize_t user_vote_store(struct device *dev, struct device_attribute *attr
 
 static DEVICE_ATTR_RW(user_vote);
 
+static int edgetpu_thermal_control_kci(struct edgetpu_dev *etdev, bool enable)
+{
+	int ret;
+
+	ret = edgetpu_kci_thermal_control(etdev, enable);
+	if (ret)
+		etdev_err_ratelimited(etdev, "Failed to %s the thermal, error:%d",
+				      enable ? "enable" : "disable", ret);
+
+	return ret;
+}
+
+static int edgetpu_thermal_control_kci_if_powered(struct edgetpu_dev *etdev, bool enable)
+{
+	int ret;
+
+	/* Use trylock since this function might be called during power up. */
+	if (!edgetpu_pm_get_if_powered(etdev->pm, true))
+		return -EAGAIN;
+
+	ret = edgetpu_thermal_control_kci(etdev, enable);
+
+	edgetpu_pm_put(etdev->pm);
+
+	return ret;
+}
+
+static int thermal_enable_get(void *data, u64 *val)
+{
+	struct edgetpu_thermal *thermal = (struct edgetpu_thermal *)data;
+
+	edgetpu_thermal_lock(thermal);
+	*val = thermal->enabled;
+	edgetpu_thermal_unlock(thermal);
+
+	return 0;
+}
+
+static int thermal_enable_set(void *data, u64 val)
+{
+	struct edgetpu_thermal *thermal = (struct edgetpu_thermal *)data;
+	int ret = 0;
+
+	edgetpu_thermal_lock(thermal);
+
+	if (thermal->enabled != (bool)val) {
+		ret = edgetpu_thermal_control_kci_if_powered(thermal->etdev, val);
+		/*
+		 * -EAGAIN means the fw is not powered and the value will be restored by
+		 * edgetpu_thermal_restore in next fw boot.
+		 */
+		if (!ret || ret == -EAGAIN) {
+			ret = 0;
+			thermal->enabled = val;
+		}
+	}
+
+	edgetpu_thermal_unlock(thermal);
+
+	return ret;
+}
+
+DEFINE_DEBUGFS_ATTRIBUTE(fops_thermal_enable, thermal_enable_get, thermal_enable_set, "%llu\n");
+
 static int tpu_thermal_cooling_register(struct edgetpu_thermal *thermal, char *type)
 {
 	struct device_node *cooling_node = NULL;
@@ -324,11 +388,18 @@ static int tpu_thermal_init(struct edgetpu_thermal *thermal, struct device *dev)
 	struct dentry *d;
 
 	thermal->dev = dev;
+	thermal->enabled = true;
+
 	d = debugfs_create_dir("cooling", edgetpu_fs_debugfs_dir());
 	/* don't let debugfs creation failure abort the init procedure */
-	if (IS_ERR_OR_NULL(d))
+	if (IS_ERR_OR_NULL(d)) {
 		dev_warn(dev, "failed to create debug fs for cooling");
-	thermal->cooling_root = d;
+		thermal->cooling_root = NULL;
+	} else {
+		thermal->cooling_root = d;
+		debugfs_create_file("enable", 0660, thermal->cooling_root, thermal,
+				    &fops_thermal_enable);
+	}
 
 	err = tpu_thermal_cooling_register(thermal, EDGETPU_COOLING_NAME);
 	if (err) {
@@ -446,7 +517,9 @@ int edgetpu_thermal_restore(struct edgetpu_dev *etdev)
 
 	edgetpu_thermal_lock(thermal);
 
-	if (edgetpu_thermal_is_suspended(thermal))
+	if (!thermal->enabled)
+		ret = edgetpu_thermal_control_kci(etdev, thermal->enabled);
+	else if (edgetpu_thermal_is_suspended(thermal))
 		ret = edgetpu_thermal_kci(etdev, TPU_OFF);
 	else if (thermal->cooling_state)
 		ret = edgetpu_thermal_kci(etdev, state_pwr_map[thermal->cooling_state].state);
diff --git a/drivers/edgetpu/rio/config.h b/drivers/edgetpu/rio/config.h
index 5b4f0f2..4434a2e 100644
--- a/drivers/edgetpu/rio/config.h
+++ b/drivers/edgetpu/rio/config.h
@@ -29,13 +29,20 @@
 /* Placeholder value */
 #define EDGETPU_TZ_MAILBOX_ID 31
 
-/* Default size of the area in remapped DRAM reserved for firmware code and internal data. */
-#define EDGETPU_DEFAULT_FW_SIZE_MAX 0x100000
+/* Default size limit of the area in remapped DRAM reserved for firmware code and internal data. */
+#define EDGETPU_DEFAULT_FW_LIMIT 0x100000
 
 /* Default size of remapped DRAM data region. */
 #define EDGETPU_DEFAULT_REMAPPED_DATA_SIZE 0x100000
 
 /*
+ * Maximum size limit of the area in remapped DRAM reserved for firmware code and internal data.
+ * The firmware image config may modify the split between code and data, but the total size of both
+ * must be respected.
+ */
+#define EDGETPU_MAX_FW_LIMIT (EDGETPU_DEFAULT_FW_LIMIT + EDGETPU_DEFAULT_REMAPPED_DATA_SIZE)
+
+/*
  * Instruction remap registers make carveout memory appear at address
  * 0x10000000 from the TPU CPU perspective
  */
@@ -46,7 +53,7 @@
  * Data in remapped DRAM starts after firmware code and internal data.
  */
 #define EDGETPU_DEFAULT_REMAPPED_DATA_ADDR                                                         \
-	(EDGETPU_INSTRUCTION_REMAP_BASE + EDGETPU_DEFAULT_FW_SIZE_MAX)
+	(EDGETPU_INSTRUCTION_REMAP_BASE + EDGETPU_DEFAULT_FW_LIMIT)
 
 /*
  * Size of memory for FW accessible debug dump segments