Merge branch 'whitechapel' into android-gs-pixel-5.10

* android-gs-pixel-5.10:
  edgetpu: add error logs for group finalize failure
  edgetpu: abrolhos: fix update stats kci req on shutdown.
  edgetpu: abrolhos: initialize clock divider ratios
  edgetpu: add common sysfs watchdog_timeout_count attr
  edgetpu: remove secure IOMMU SMC
  edgetpu: add optional wrapper class for validating edgetpu fds

Signed-off-by: Nrithya Kanakasabapathy <nrithya@google.com>
Change-Id: Id4aff09f9fe65deaec60b670914e10e16af50d92

10 files changed, 126 insertions, 42 deletions

diff --git a/drivers/edgetpu/Kconfig b/drivers/edgetpu/Kconfig
index 3105089..fe40029 100644
--- a/drivers/edgetpu/Kconfig
+++ b/drivers/edgetpu/Kconfig
@@ -21,6 +21,24 @@ config ABROLHOS
 	  To compile this driver as a module, choose M here.  The module
 	  will be called "abrolhos".
 
+config EDGETPU_EXTERNAL_WRAPPER_CLASS
+        bool "EdgeTPU add external wrapper class"
+	depends on EDGETPU_FRAMEWORK
+	default n
+	help
+	  Selected if an external wrapper class is to be used.  This option is
+	  intended to be auto-selected by other config options that depend on
+	  external classes that wrap the EdgeTPU core driver and is not
+	  intended for interactive use.
+
+config EDGETPU_FPGA
+	bool "Build for EdgeTPU chip FPGA emulation"
+	depends on EDGETPU_FRAMEWORK
+	default n
+	help
+	  Say Y to build for HAPS/Palladium/etc. FPGA emulators, or N to build
+	  silicon with full number of tiles.
+
 config EDGETPU_TELEMETRY_TRACE
 	bool "Build EdgeTPU driver with firmware tracing support"
 	depends on EDGETPU_FRAMEWORK
diff --git a/drivers/edgetpu/abrolhos-platform.h b/drivers/edgetpu/abrolhos-platform.h
index 721f24d..980dc8e 100644
--- a/drivers/edgetpu/abrolhos-platform.h
+++ b/drivers/edgetpu/abrolhos-platform.h
@@ -11,6 +11,7 @@
 #include <linux/io.h>
 #include <linux/kernel.h>
 #include <linux/types.h>
+#include <soc/google/bcl.h>
 
 #include "abrolhos-debug-dump.h"
 #include "abrolhos-pm.h"
@@ -44,6 +45,7 @@ struct abrolhos_platform_dev {
 	struct edgetpu_coherent_mem log_mem;
 	struct edgetpu_coherent_mem trace_mem;
 	struct abrolhos_sscd_info sscd_info;
+	struct gs101_bcl_dev *bcl_dev;
 };
 
 #endif /* __ABROLHOS_PLATFORM_H__ */
diff --git a/drivers/edgetpu/abrolhos-pm.c b/drivers/edgetpu/abrolhos-pm.c
index aba134a..47fa7dc 100644
--- a/drivers/edgetpu/abrolhos-pm.c
+++ b/drivers/edgetpu/abrolhos-pm.c
@@ -10,7 +10,9 @@
 #include <linux/gsa/gsa_tpu.h>
 #include <linux/module.h>
 #include <linux/pm_runtime.h>
-#include <linux/soc/samsung/exynos-smc.h>
+#include <soc/google/bcl.h>
+#include <soc/google/bts.h>
+#include <soc/google/exynos_pm_qos.h>
 
 #include "abrolhos-platform.h"
 #include "abrolhos-pm.h"
@@ -22,13 +24,8 @@
 #include "edgetpu-pm.h"
 #include "edgetpu-telemetry.h"
 
-#include "soc/google/exynos_pm_qos.h"
-#include "soc/google/bts.h"
-
 #include "edgetpu-pm.c"
 
-#define TPU_SMC_ID			(0x15)
-
 /*
  * Encode INT/MIF values as a 16 bit pair in the 32-bit return value
  * (in units of MHz, to provide enough range)
@@ -98,12 +95,6 @@ static int abrolhos_pwr_state_set_locked(void *data, u64 val)
 			dev_err(dev, "pm_runtime_get_sync returned %d\n", ret);
 			return ret;
 		}
-		ret = exynos_smc(SMC_PROTECTION_SET, 0, TPU_SMC_ID,
-				 SMC_PROTECTION_ENABLE);
-		if (ret)
-			dev_warn(dev,
-				 "exynos_smc protection enable returned %d\n",
-				 ret);
 	}
 
 	ret = exynos_acpm_set_rate(TPU_ACPM_DOMAIN, (unsigned long)val);
@@ -114,13 +105,6 @@ static int abrolhos_pwr_state_set_locked(void *data, u64 val)
 	}
 
 	if (curr_state != TPU_OFF && val == TPU_OFF) {
-		ret = exynos_smc(SMC_PROTECTION_SET, 0, TPU_SMC_ID,
-				 SMC_PROTECTION_DISABLE);
-		if (ret)
-			dev_warn(dev,
-				 "exynos_smc protection disable returned %d\n",
-				 ret);
-
 		ret = pm_runtime_put_sync(dev);
 		if (ret) {
 			dev_err(dev, "%s: pm_runtime_put_sync returned %d\n",
@@ -522,8 +506,14 @@ static int abrolhos_power_up(struct edgetpu_pm *etpm)
 		etdev->state = ETDEV_STATE_GOOD; /* f/w handshake success */
 	mutex_unlock(&etdev->state_lock);
 
-	if (ret)
+	if (ret) {
 		abrolhos_power_down(etpm);
+	} else {
+		if (!edgetpu_pdev->bcl_dev)
+			edgetpu_pdev->bcl_dev = gs101_retrieve_bcl_handle();
+		if (edgetpu_pdev->bcl_dev)
+			gs101_init_tpu_ratio(edgetpu_pdev->bcl_dev);
+	}
 
 	return ret;
 }
@@ -611,7 +601,7 @@ static void abrolhos_power_down(struct edgetpu_pm *etpm)
 
 	if (etdev->kci && edgetpu_firmware_status_locked(etdev) == FW_VALID) {
 		/* Update usage stats before we power off fw. */
-		edgetpu_kci_update_usage(etdev);
+		edgetpu_kci_update_usage_locked(etdev);
 		abrolhos_pm_shutdown_firmware(edgetpu_pdev, etdev,
 					      edgetpu_pdev);
 		edgetpu_kci_cancel_work_queues(etdev->kci);
diff --git a/drivers/edgetpu/edgetpu-device-group.c b/drivers/edgetpu/edgetpu-device-group.c
index 22ace2b..5e2a9d0 100644
--- a/drivers/edgetpu/edgetpu-device-group.c
+++ b/drivers/edgetpu/edgetpu-device-group.c
@@ -216,18 +216,26 @@ edgetpu_device_group_kci_finalized(struct edgetpu_device_group *group)
 		 */
 		if (etdev->state != ETDEV_STATE_GOOD) {
 			ret = edgetpu_get_state_errno_locked(etdev);
+			etdev_err(group->etdev, "finalize dev %s state %u (%d)",
+				  etdev->dev_name, etdev->state, ret);
 			goto out_free;
 		}
 		ret = edgetpu_async_add_job(
 			ctx, &params[i],
 			(edgetpu_async_job_t)edgetpu_kci_join_group_worker);
-		if (ret)
+		if (ret) {
+			etdev_err(group->etdev,
+				  "finalize fw job for %s returns %d",
+				  etdev->dev_name, ret);
 			goto out_free;
+		}
 		atomic_inc(&etdev->job_count);
 	}
 	ret = edgetpu_async_wait(ctx);
-	if (ret)
+	if (ret) {
+		etdev_err(group->etdev, "finalize wait returns %d", ret);
 		goto out_free;
+	}
 	for_each_async_ret(ctx, val, i) {
 		if (val)
 			goto out_leave;
@@ -471,11 +479,21 @@ static bool edgetpu_group_check_contiguity(struct edgetpu_device_group *group)
 		mcp_n = mcp->total_num;
 	fr = group->etdev->mcp_die_index;
 	for (i = 1; i < group->n_clients; i++, fr = to) {
-		to = edgetpu_device_group_nth_etdev(group, i)->mcp_die_index;
+		struct edgetpu_dev *to_etdev =
+			edgetpu_device_group_nth_etdev(group, i);
+
+		to = to_etdev->mcp_die_index;
 		if (fr + 1 == to)
 			continue;
-		if (!mcp_n || (fr + 1) % mcp_n != to)
+		if (!mcp_n || (fr + 1) % mcp_n != to) {
+			struct edgetpu_dev *fr_etdev =
+				edgetpu_device_group_nth_etdev(group, i - 1);
+
+			etdev_err(group->etdev,
+				  "finalize group not contiguous at %s -> %s",
+				  fr_etdev->dev_name, to_etdev->dev_name);
 			return false;
+		}
 	}
 
 	return true;
@@ -749,13 +767,17 @@ int edgetpu_device_group_finalize(struct edgetpu_device_group *group)
 		goto err_unlock;
 
 	if (!edgetpu_device_group_is_waiting(group)) {
+		etdev_err(group->etdev, "finalize group is not waiting");
 		ret = -EINVAL;
 		goto err_unlock;
 	}
 
 	ret = group_alloc_members(group);
-	if (ret)
+	if (ret) {
+		etdev_err(group->etdev,
+			  "finalize alloc members returns %d", ret);
 		goto err_unlock;
+	}
 
 	/*
 	 * Check the contiguity here but not in edgetpu_clients_groupable()
@@ -780,20 +802,25 @@ int edgetpu_device_group_finalize(struct edgetpu_device_group *group)
 		ret = do_attach_mailbox_locked(group);
 		if (ret) {
 			etdev_err(group->etdev,
-				  "attach mailbox failed on finalization: %d",
-				  ret);
+				  "finalize attach mailbox failed: %d", ret);
 			goto err_release_members;
 		}
 	}
 #ifdef EDGETPU_HAS_P2P_MAILBOX
 	ret = edgetpu_p2p_mailbox_setup(group);
-	if (ret)
+	if (ret) {
+		etdev_err(group->etdev,
+			  "finalize p2p mailbox setup failed: %d", ret);
 		goto err_detach_mailbox;
+	}
 #endif
 
 	ret = edgetpu_group_setup_remote_dram(group);
-	if (ret)
+	if (ret) {
+		etdev_err(group->etdev,
+			  "finalize remote dram setup failed: %d", ret);
 		goto err_release_p2p;
+	}
 
 	edgetpu_usr_init_group(group);
 
diff --git a/drivers/edgetpu/edgetpu-firmware.c b/drivers/edgetpu/edgetpu-firmware.c
index a4ac9a5..673a51c 100644
--- a/drivers/edgetpu/edgetpu-firmware.c
+++ b/drivers/edgetpu/edgetpu-firmware.c
@@ -659,6 +659,7 @@ static void edgetpu_firmware_wdt_timeout_action(void *data)
 	struct edgetpu_dev *etdev = data;
 	struct edgetpu_firmware *et_fw = etdev->firmware;
 
+	etdev->watchdog_timeout_count++;
 	/* Don't attempt f/w restart if device is off. */
 	if (!edgetpu_is_powered(etdev))
 		return;
diff --git a/drivers/edgetpu/edgetpu-fs.c b/drivers/edgetpu/edgetpu-fs.c
index 6742d14..3966226 100644
--- a/drivers/edgetpu/edgetpu-fs.c
+++ b/drivers/edgetpu/edgetpu-fs.c
@@ -68,6 +68,8 @@ static struct dentry *edgetpu_debugfs_dir;
 
 static bool is_edgetpu_file(struct file *file)
 {
+	if (edgetpu_is_external_wrapper_class_file(file))
+		return true;
 	return file->f_op == &edgetpu_fops;
 }
 
@@ -911,8 +913,19 @@ static ssize_t firmware_crash_count_show(
 }
 static DEVICE_ATTR_RO(firmware_crash_count);
 
+static ssize_t watchdog_timeout_count_show(
+		struct device *dev, struct device_attribute *attr,
+		char *buf)
+{
+	struct edgetpu_dev *etdev = dev_get_drvdata(dev);
+
+	return scnprintf(buf, PAGE_SIZE, "%u\n", etdev->watchdog_timeout_count);
+}
+static DEVICE_ATTR_RO(watchdog_timeout_count);
+
 static struct attribute *edgetpu_dev_attrs[] = {
 	&dev_attr_firmware_crash_count.attr,
+	&dev_attr_watchdog_timeout_count.attr,
 	NULL,
 };
 
diff --git a/drivers/edgetpu/edgetpu-internal.h b/drivers/edgetpu/edgetpu-internal.h
index 12d47b8..1a8cc35 100644
--- a/drivers/edgetpu/edgetpu-internal.h
+++ b/drivers/edgetpu/edgetpu-internal.h
@@ -204,6 +204,7 @@ struct edgetpu_dev {
 
 	/* counts of error events */
 	uint firmware_crash_count;
+	uint watchdog_timeout_count;
 
 	struct edgetpu_coherent_mem debug_dump_mem;	/* debug dump memory */
 	/* debug dump handlers */
@@ -330,6 +331,15 @@ void edgetpu_free_coherent(struct edgetpu_dev *etdev,
 int edgetpu_open(struct edgetpu_dev *etdev, struct file *file);
 long edgetpu_ioctl(struct file *file, uint cmd, ulong arg);
 
+#if IS_ENABLED(CONFIG_EDGETPU_EXTERNAL_WRAPPER_CLASS)
+extern bool edgetpu_is_external_wrapper_class_file(struct file *file);
+#else
+static inline bool edgetpu_is_external_wrapper_class_file(struct file *file)
+{
+	return false;
+}
+#endif
+
 /* Handle firmware crash event */
 void edgetpu_handle_firmware_crash(struct edgetpu_dev *etdev,
 				   enum edgetpu_fw_crash_type crash_type);
diff --git a/drivers/edgetpu/edgetpu-kci.c b/drivers/edgetpu/edgetpu-kci.c
index 5d00c47..c8b5d86 100644
--- a/drivers/edgetpu/edgetpu-kci.c
+++ b/drivers/edgetpu/edgetpu-kci.c
@@ -842,6 +842,22 @@ enum edgetpu_fw_flavor edgetpu_kci_fw_info(struct edgetpu_kci *kci,
 
 int edgetpu_kci_update_usage(struct edgetpu_dev *etdev)
 {
+	int ret;
+
+	/* Quick return if device already powered down, else get PM ref. */
+	if (!edgetpu_is_powered(etdev))
+		return -EAGAIN;
+	ret = edgetpu_pm_get(etdev->pm);
+	if (ret)
+		return ret;
+	ret = edgetpu_kci_update_usage_locked(etdev);
+
+	edgetpu_pm_put(etdev->pm);
+	return ret;
+}
+
+int edgetpu_kci_update_usage_locked(struct edgetpu_dev *etdev)
+{
 #define EDGETPU_USAGE_BUFFER_SIZE	4096
 	struct edgetpu_command_element cmd = {
 		.code = KCI_CODE_GET_USAGE,
@@ -854,19 +870,13 @@ int edgetpu_kci_update_usage(struct edgetpu_dev *etdev)
 	struct edgetpu_kci_response_element resp;
 	int ret;
 
-	/* Quick return if device already powered down, else get PM ref. */
-	if (!edgetpu_is_powered(etdev))
-		return -EAGAIN;
-	ret = edgetpu_pm_get(etdev->pm);
-	if (ret)
-		return ret;
 	ret = edgetpu_iremap_alloc(etdev, EDGETPU_USAGE_BUFFER_SIZE, &mem,
 				   EDGETPU_CONTEXT_KCI);
 
 	if (ret) {
 		etdev_warn_once(etdev, "%s: failed to allocate usage buffer",
 				__func__);
-		goto out;
+		return ret;
 	}
 
 	cmd.dma.address = mem.tpu_addr;
@@ -883,8 +893,6 @@ int edgetpu_kci_update_usage(struct edgetpu_dev *etdev)
 
 	edgetpu_iremap_free(etdev, &mem, EDGETPU_CONTEXT_KCI);
 
-out:
-	edgetpu_pm_put(etdev->pm);
 	return ret;
 }
 
diff --git a/drivers/edgetpu/edgetpu-kci.h b/drivers/edgetpu/edgetpu-kci.h
index daeafc8..443c690 100644
--- a/drivers/edgetpu/edgetpu-kci.h
+++ b/drivers/edgetpu/edgetpu-kci.h
@@ -287,6 +287,16 @@ enum edgetpu_fw_flavor edgetpu_kci_fw_info(
 int edgetpu_kci_update_usage(struct edgetpu_dev *etdev);
 
 /*
+ * Works the same as edgetpu_kci_update_usage() except the caller of this
+ * function must guarantee the device stays powered up, typically by calling
+ * edgetpu_pm_get() or by calling this function from the power management
+ * functions themselves.
+ *
+ * Returns KCI response code on success or < 0 on error (typically -ETIMEDOUT).
+ */
+int edgetpu_kci_update_usage_locked(struct edgetpu_dev *etdev);
+
+/*
  * Sends the "Map Log Buffer" command and waits for remote response.
  *
  * Returns the code of response, or a negative errno on error.
diff --git a/drivers/edgetpu/edgetpu-mapping.h b/drivers/edgetpu/edgetpu-mapping.h
index 35906b0..6681427 100644
--- a/drivers/edgetpu/edgetpu-mapping.h
+++ b/drivers/edgetpu/edgetpu-mapping.h
@@ -141,13 +141,18 @@ void edgetpu_mappings_show(struct edgetpu_mapping_root *mappings,
 
 static inline int __dma_dir_to_iommu_prot(enum dma_data_direction dir)
 {
+	int prot = 0;
+
+#ifdef EDGETPU_IS_IO_COHERENT
+	prot = IOMMU_CACHE;
+#endif
 	switch (dir) {
 	case DMA_BIDIRECTIONAL:
-		return IOMMU_READ | IOMMU_WRITE;
+		return prot | IOMMU_READ | IOMMU_WRITE;
 	case DMA_TO_DEVICE:
-		return IOMMU_READ;
+		return prot | IOMMU_READ;
 	case DMA_FROM_DEVICE:
-		return IOMMU_WRITE;
+		return prot | IOMMU_WRITE;
 	default:
 		return 0;
 	}