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

* whitechapel:
  edgetpu: check file op on group join
  edgetpu: restart firmware on unrecoverable fault notification
  edgetpu: remove firmware crash extra_info field
  edgetpu: fix size and offset on host buffer map
  edgetpu: Create common f/w interface for mobile chipsets
  edgetpu: remove unbalanced unlock in group alloc
  edgetpu: abrolhos fix uninit power lock
  edgetpu: fixup KCI header file comments on response element
  edgetpu: raise log severity of MMU setup
  edgetpu: abort probe when MMU setup failed

Signed-off-by: Nrithya Kanakasabapathy <nrithya@google.com>
Change-Id: I4d968724c8b73a86b607085090289b2845ccfe18

11 files changed, 122 insertions, 41 deletions

diff --git a/drivers/edgetpu/abrolhos-firmware.c b/drivers/edgetpu/abrolhos-firmware.c
index 2c9b87e..4170c49 100644
--- a/drivers/edgetpu/abrolhos-firmware.c
+++ b/drivers/edgetpu/abrolhos-firmware.c
@@ -9,13 +9,13 @@
 #include <linux/gsa/gsa_tpu.h>
 #include <linux/slab.h>
 
-#include "abrolhos-firmware.h"
 #include "abrolhos-platform.h"
 #include "edgetpu-config.h"
 #include "edgetpu-firmware.h"
 #include "edgetpu-internal.h"
 #include "edgetpu-kci.h"
 #include "edgetpu-mailbox.h"
+#include "mobile-firmware.h"
 
 static int abrolhos_firmware_alloc_buffer(
 		struct edgetpu_firmware *et_fw,
@@ -26,7 +26,7 @@ static int abrolhos_firmware_alloc_buffer(
 
 	/* Allocate extra space the image header */
 	size_t buffer_size =
-		edgetpu_pdev->fw_region_size + ABROLHOS_FW_HEADER_SIZE;
+		edgetpu_pdev->fw_region_size + MOBILE_FW_HEADER_SIZE;
 
 	fw_buf->vaddr = kzalloc(buffer_size, GFP_KERNEL);
 	if (!fw_buf->vaddr) {
@@ -70,14 +70,14 @@ static int abrolhos_firmware_prepare_run(struct edgetpu_firmware *et_fw,
 	struct edgetpu_dev *etdev = et_fw->etdev;
 	struct abrolhos_platform_dev *edgetpu_pdev = to_abrolhos_dev(etdev);
 	void *image_vaddr, *header_vaddr;
-	struct abrolhos_image_config *image_config;
+	struct mobile_image_config *image_config;
 	phys_addr_t image_start, image_end, carveout_start, carveout_end;
 	dma_addr_t header_dma_addr;
 	int ret, tpu_state;
 
-	if (fw_buf->used_size < ABROLHOS_FW_HEADER_SIZE) {
+	if (fw_buf->used_size < MOBILE_FW_HEADER_SIZE) {
 		etdev_err(etdev, "Invalid buffer size: %zu < %d\n",
-			  fw_buf->used_size, ABROLHOS_FW_HEADER_SIZE);
+			  fw_buf->used_size, MOBILE_FW_HEADER_SIZE);
 		return -EINVAL;
 	}
 
@@ -109,12 +109,12 @@ static int abrolhos_firmware_prepare_run(struct edgetpu_firmware *et_fw,
 	}
 
 	/* Skip the header */
-	memcpy(image_vaddr, fw_buf->vaddr + ABROLHOS_FW_HEADER_SIZE,
-	       fw_buf->used_size - ABROLHOS_FW_HEADER_SIZE);
+	memcpy(image_vaddr, fw_buf->vaddr + MOBILE_FW_HEADER_SIZE,
+	       fw_buf->used_size - MOBILE_FW_HEADER_SIZE);
 
 	/* Allocate coherent memory for the image header */
 	header_vaddr = dma_alloc_coherent(edgetpu_pdev->gsa_dev,
-					  ABROLHOS_FW_HEADER_SIZE,
+					  MOBILE_FW_HEADER_SIZE,
 					  &header_dma_addr, GFP_KERNEL);
 	if (!header_vaddr) {
 		etdev_err(etdev,
@@ -123,7 +123,7 @@ static int abrolhos_firmware_prepare_run(struct edgetpu_firmware *et_fw,
 		goto out_unmap;
 	}
 
-	memcpy(header_vaddr, fw_buf->vaddr, ABROLHOS_FW_HEADER_SIZE);
+	memcpy(header_vaddr, fw_buf->vaddr, MOBILE_FW_HEADER_SIZE);
 	etdev_dbg(etdev,
 		  "Requesting GSA image load. meta = %llX payload = %llX",
 		  header_dma_addr, (u64)edgetpu_pdev->fw_region_paddr);
@@ -137,7 +137,7 @@ static int abrolhos_firmware_prepare_run(struct edgetpu_firmware *et_fw,
 	}
 
 	/* fetch the firmware versions */
-	image_config = fw_buf->vaddr + ABROLHOS_IMAGE_CONFIG_OFFSET;
+	image_config = fw_buf->vaddr + MOBILE_IMAGE_CONFIG_OFFSET;
 	memcpy(&etdev->fw_version, &image_config->firmware_versions,
 	       sizeof(etdev->fw_version));
 
@@ -173,7 +173,7 @@ static int abrolhos_firmware_prepare_run(struct edgetpu_firmware *et_fw,
 	}
 
 out_free_gsa:
-	dma_free_coherent(edgetpu_pdev->gsa_dev, ABROLHOS_FW_HEADER_SIZE,
+	dma_free_coherent(edgetpu_pdev->gsa_dev, MOBILE_FW_HEADER_SIZE,
 			  header_vaddr, header_dma_addr);
 out_unmap:
 	memunmap(image_vaddr);
@@ -188,12 +188,12 @@ static const struct edgetpu_firmware_handlers abrolhos_firmware_handlers = {
 	.prepare_run = abrolhos_firmware_prepare_run,
 };
 
-int abrolhos_edgetpu_firmware_create(struct edgetpu_dev *etdev)
+int mobile_edgetpu_firmware_create(struct edgetpu_dev *etdev)
 {
 	return edgetpu_firmware_create(etdev, &abrolhos_firmware_handlers);
 }
 
-void abrolhos_edgetpu_firmware_destroy(struct edgetpu_dev *etdev)
+void mobile_edgetpu_firmware_destroy(struct edgetpu_dev *etdev)
 {
 	edgetpu_firmware_destroy(etdev);
 }
diff --git a/drivers/edgetpu/abrolhos-platform.c b/drivers/edgetpu/abrolhos-platform.c
index e5e9a94..cc7ea13 100644
--- a/drivers/edgetpu/abrolhos-platform.c
+++ b/drivers/edgetpu/abrolhos-platform.c
@@ -20,7 +20,6 @@
 #include <linux/slab.h>
 #include <linux/types.h>
 
-#include "abrolhos-firmware.h"
 #include "abrolhos-platform.h"
 #include "abrolhos-pm.h"
 #include "edgetpu-config.h"
@@ -30,6 +29,7 @@
 #include "edgetpu-mmu.h"
 #include "edgetpu-pm.h"
 #include "edgetpu-telemetry.h"
+#include "mobile-firmware.h"
 
 static const struct of_device_id edgetpu_of_match[] = {
 	{ .compatible = "google,darwinn", },
@@ -172,14 +172,15 @@ static void edgetpu_platform_cleanup_fw_region(
 	etpdev->shared_mem_vaddr = NULL;
 }
 
-void edgetpu_setup_mmu(struct edgetpu_dev *etdev)
+int edgetpu_setup_mmu(struct edgetpu_dev *etdev)
 {
 	int ret;
 
 	/* No MMU info to pass to attach, IOMMU API will handle. */
 	ret = edgetpu_mmu_attach(etdev, NULL);
 	if (ret)
-		dev_warn(etdev->dev, "failed to attach IOMMU: %d\n", ret);
+		dev_err(etdev->dev, "failed to attach IOMMU: %d\n", ret);
+	return ret;
 }
 
 static int abrolhos_parse_ssmt(struct abrolhos_platform_dev *etpdev)
@@ -305,7 +306,7 @@ static int edgetpu_platform_probe(struct platform_device *pdev)
 	if (ret)
 		goto out_remove_device;
 
-	ret = abrolhos_edgetpu_firmware_create(&edgetpu_pdev->edgetpu_dev);
+	ret = mobile_edgetpu_firmware_create(&edgetpu_pdev->edgetpu_dev);
 	if (ret) {
 		dev_err(dev,
 			"%s initialize firmware downloader failed: %d\n",
@@ -346,7 +347,7 @@ static int edgetpu_platform_remove(struct platform_device *pdev)
 	struct edgetpu_dev *etdev = platform_get_drvdata(pdev);
 	struct abrolhos_platform_dev *edgetpu_pdev = to_abrolhos_dev(etdev);
 
-	abrolhos_edgetpu_firmware_destroy(etdev);
+	mobile_edgetpu_firmware_destroy(etdev);
 	if (edgetpu_pdev->irq >= 0)
 		edgetpu_unregister_irq(etdev, edgetpu_pdev->irq);
 
diff --git a/drivers/edgetpu/abrolhos-pm.c b/drivers/edgetpu/abrolhos-pm.c
index 07f0369..aba134a 100644
--- a/drivers/edgetpu/abrolhos-pm.c
+++ b/drivers/edgetpu/abrolhos-pm.c
@@ -634,13 +634,12 @@ static int abrolhos_pm_after_create(struct edgetpu_pm *etpm)
 	if (ret)
 		return ret;
 
+	mutex_init(&edgetpu_pdev->platform_pwr.policy_lock);
+	mutex_init(&edgetpu_pdev->platform_pwr.state_lock);
 	ret = abrolhos_pwr_state_set(etdev,
 				     abrolhos_get_initial_pwr_state(dev));
 	if (ret)
 		return ret;
-
-	mutex_init(&edgetpu_pdev->platform_pwr.policy_lock);
-	mutex_init(&edgetpu_pdev->platform_pwr.state_lock);
 	abrolhos_pwr_debugfs_dir =
 		debugfs_create_dir("power", edgetpu_fs_debugfs_dir());
 	if (!abrolhos_pwr_debugfs_dir) {
diff --git a/drivers/edgetpu/abrolhos-thermal.c b/drivers/edgetpu/abrolhos-thermal.c
index cbb9abe..27429a9 100644
--- a/drivers/edgetpu/abrolhos-thermal.c
+++ b/drivers/edgetpu/abrolhos-thermal.c
@@ -16,7 +16,6 @@
 #include <linux/version.h>
 #include <linux/of.h>
 
-#include "abrolhos-firmware.h"
 #include "abrolhos-platform.h"
 #include "abrolhos-pm.h"
 #include "edgetpu-config.h"
diff --git a/drivers/edgetpu/edgetpu-core.c b/drivers/edgetpu/edgetpu-core.c
index 891cde6..33ca204 100644
--- a/drivers/edgetpu/edgetpu-core.c
+++ b/drivers/edgetpu/edgetpu-core.c
@@ -31,6 +31,7 @@
 #include "edgetpu-mailbox.h"
 #include "edgetpu-mcp.h"
 #include "edgetpu-mmu.h"
+#include "edgetpu-sw-watchdog.h"
 #include "edgetpu-telemetry.h"
 #include "edgetpu-usage-stats.h"
 #include "edgetpu-wakelock.h"
@@ -256,7 +257,9 @@ int edgetpu_device_add(struct edgetpu_dev *etdev,
 			etdev->dev_name, ret);
 		goto remove_dev;
 	}
-	edgetpu_setup_mmu(etdev);
+	ret = edgetpu_setup_mmu(etdev);
+	if (ret)
+		goto remove_dev;
 
 	edgetpu_usage_stats_init(etdev);
 
@@ -443,12 +446,16 @@ void edgetpu_free_coherent(struct edgetpu_dev *etdev,
 	mem->vaddr = NULL;
 }
 
-void edgetpu_handle_firmware_crash(struct edgetpu_dev *etdev, u16 crash_type,
-				   u32 extra_info)
+void edgetpu_handle_firmware_crash(struct edgetpu_dev *etdev,
+				   enum edgetpu_fw_crash_type crash_type)
 {
-	etdev_err(etdev, "firmware crashed: %u 0x%x", crash_type, extra_info);
+	etdev_err(etdev, "firmware crashed: %u", crash_type);
 	etdev->firmware_crash_count++;
 	edgetpu_fatal_error_notify(etdev);
+
+	if (crash_type == EDGETPU_FW_CRASH_UNRECOV_FAULT)
+		/* Restart firmware without chip reset */
+		edgetpu_watchdog_bite(etdev, false);
 }
 
 int __init edgetpu_init(void)
diff --git a/drivers/edgetpu/edgetpu-device-group.c b/drivers/edgetpu/edgetpu-device-group.c
index 94129e1..22ace2b 100644
--- a/drivers/edgetpu/edgetpu-device-group.c
+++ b/drivers/edgetpu/edgetpu-device-group.c
@@ -664,8 +664,6 @@ edgetpu_device_group_alloc(struct edgetpu_client *client,
 			EDGETPU_CONTEXT_DOMAIN_TOKEN | etdomain->token;
 	else
 		group->context_id = EDGETPU_CONTEXT_INVALID;
-
-	mutex_unlock(&client->etdev->state_lock);
 	return group;
 
 error_leave_group:
@@ -1073,9 +1071,14 @@ static struct page **edgetpu_pin_user_pages(struct edgetpu_device_group *group,
 	int i;
 	int ret;
 
-	num_pages = size / PAGE_SIZE;
+	if (size == 0)
+		return ERR_PTR(-EINVAL);
 	offset = host_addr & (PAGE_SIZE - 1);
-	if (offset)
+	/* overflow check */
+	if (unlikely(size + offset < size))
+		return ERR_PTR(-ENOMEM);
+	num_pages = (size + offset) / PAGE_SIZE;
+	if ((size + offset) % PAGE_SIZE)
 		num_pages++;
 
 	etdev_dbg(etdev, "%s: hostaddr=0x%llx pages=%u dir=%x", __func__,
diff --git a/drivers/edgetpu/edgetpu-fs.c b/drivers/edgetpu/edgetpu-fs.c
index 9612c97..6742d14 100644
--- a/drivers/edgetpu/edgetpu-fs.c
+++ b/drivers/edgetpu/edgetpu-fs.c
@@ -66,6 +66,11 @@ static struct dentry *edgetpu_debugfs_dir;
 		}                                                              \
 	} while (0)
 
+static bool is_edgetpu_file(struct file *file)
+{
+	return file->f_op == &edgetpu_fops;
+}
+
 int edgetpu_open(struct edgetpu_dev *etdev, struct file *file)
 {
 	struct edgetpu_client *client;
@@ -254,6 +259,10 @@ static int edgetpu_ioctl_join_group(struct edgetpu_client *client,
 		ret = -EBADF;
 		goto out;
 	}
+	if (!is_edgetpu_file(file)) {
+		ret = -EINVAL;
+		goto out;
+	}
 	leader = file->private_data;
 	if (!leader) {
 		ret = -EINVAL;
@@ -697,9 +706,6 @@ long edgetpu_ioctl(struct file *file, uint cmd, ulong arg)
 
 static long edgetpu_fs_ioctl(struct file *file, uint cmd, ulong arg)
 {
-	if (file->f_op != &edgetpu_fops)
-		return -ENOTTY;
-
 	return edgetpu_ioctl(file, cmd, arg);
 }
 
diff --git a/drivers/edgetpu/edgetpu-internal.h b/drivers/edgetpu/edgetpu-internal.h
index 9d4f10d..12d47b8 100644
--- a/drivers/edgetpu/edgetpu-internal.h
+++ b/drivers/edgetpu/edgetpu-internal.h
@@ -211,6 +211,15 @@ struct edgetpu_dev {
 	struct work_struct debug_dump_work;
 };
 
+/* Firmware crash_type codes */
+enum edgetpu_fw_crash_type {
+	EDGETPU_FW_CRASH_ASSERT = 0,
+	EDGETPU_FW_CRASH_DATA_ABORT = 1,
+	EDGETPU_FW_CRASH_PREFETCH_ABORT = 2,
+	EDGETPU_FW_CRASH_UNDEF_EXCEPT = 3,
+	EDGETPU_FW_CRASH_UNRECOV_FAULT = 4,
+};
+
 extern const struct file_operations edgetpu_fops;
 
 /* Status regs dump. */
@@ -322,8 +331,8 @@ int edgetpu_open(struct edgetpu_dev *etdev, struct file *file);
 long edgetpu_ioctl(struct file *file, uint cmd, ulong arg);
 
 /* Handle firmware crash event */
-void edgetpu_handle_firmware_crash(struct edgetpu_dev *etdev, u16 crash_type,
-				   u32 extra_info);
+void edgetpu_handle_firmware_crash(struct edgetpu_dev *etdev,
+				   enum edgetpu_fw_crash_type crash_type);
 
 /* Bus (Platform/PCI) <-> Core API */
 
@@ -353,8 +362,12 @@ void edgetpu_chip_exit(struct edgetpu_dev *etdev);
 /* IRQ handler */
 irqreturn_t edgetpu_chip_irq_handler(int irq, void *arg);
 
-/* Called from core to chip layer when MMU is needed during device init. */
-void edgetpu_setup_mmu(struct edgetpu_dev *etdev);
+/*
+ * Called from core to chip layer when MMU is needed during device init.
+ *
+ * Returns 0 on success, otherwise -errno.
+ */
+int edgetpu_setup_mmu(struct edgetpu_dev *etdev);
 
 /* Read TPU timestamp */
 u64 edgetpu_chip_tpu_timestamp(struct edgetpu_dev *etdev);
diff --git a/drivers/edgetpu/edgetpu-kci.c b/drivers/edgetpu/edgetpu-kci.c
index d48d799..5d00c47 100644
--- a/drivers/edgetpu/edgetpu-kci.c
+++ b/drivers/edgetpu/edgetpu-kci.c
@@ -90,8 +90,8 @@ edgetpu_reverse_kci_consume_response(struct edgetpu_dev *etdev,
 
 	switch (resp->code) {
 	case RKCI_FIRMWARE_CRASH:
-		edgetpu_handle_firmware_crash(etdev, resp->status,
-					      resp->retval);
+		edgetpu_handle_firmware_crash(
+		      etdev, (enum edgetpu_fw_crash_type)resp->retval);
 		break;
 	default:
 		etdev_warn(etdev, "%s: Unrecognized KCI request: 0x%x\n",
diff --git a/drivers/edgetpu/edgetpu-kci.h b/drivers/edgetpu/edgetpu-kci.h
index 98a61df..daeafc8 100644
--- a/drivers/edgetpu/edgetpu-kci.h
+++ b/drivers/edgetpu/edgetpu-kci.h
@@ -70,10 +70,9 @@ struct edgetpu_kci_response_element {
 	u64 seq;
 	u16 code;
 	/*
-	 * Reserved on firmware side for KCI response - they can't touch this.
+	 * Reserved for host use - firmware can't touch this.
 	 * If a value is written here it will be discarded and overwritten
 	 * during response processing.
-	 * For a Reverse KCI command firmware does set this as value1.
 	 */
 	u16 status;
 	/*
diff --git a/drivers/edgetpu/mobile-firmware.h b/drivers/edgetpu/mobile-firmware.h
new file mode 100644
index 0000000..e0c8dd8
--- /dev/null
+++ b/drivers/edgetpu/mobile-firmware.h
@@ -0,0 +1,54 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Implements utilities for firmware management of mobile chipsets.
+ *
+ * Copyright (C) 2021 Google, Inc.
+ */
+#ifndef __MOBILE_FIRMWARE_H__
+#define __MOBILE_FIRMWARE_H__
+
+#include <linux/sizes.h>
+
+#include "edgetpu-internal.h"
+#include "edgetpu.h"
+
+/* mobile FW header size */
+#define MOBILE_FW_HEADER_SIZE SZ_4K
+/* The offset to the signed firmware header. */
+#define MOBILE_HEADER_OFFSET 0x400
+/* The offset to image configuration. */
+#define MOBILE_IMAGE_CONFIG_OFFSET (MOBILE_HEADER_OFFSET + 0x160)
+
+/*
+ * The image configuration attached to the signed firmware.
+ */
+struct mobile_image_config {
+	__u32 carveout_base;
+	__u32 firmware_base;
+	__u32 firmware_size;
+	struct edgetpu_fw_version firmware_versions;
+} __packed;
+
+/*
+ * Mobile firmware header.
+ */
+struct mobile_image_header {
+	char sig[512];
+	char pub[512];
+	int Magic;
+	int Generation;
+	int RollbackInfo;
+	int Length;
+	char Flags[16];
+	char BodyHash[32];
+	char ChipId[32];
+	char AuthConfig[256];
+	struct mobile_image_config ImageConfig;
+};
+
+int mobile_edgetpu_firmware_create(struct edgetpu_dev *etdev);
+void mobile_edgetpu_firmware_destroy(struct edgetpu_dev *etdev);
+
+int mobile_edgetpu_firmware_run_default(struct edgetpu_dev *etdev);
+
+#endif /* __MOBILE_FIRMWARE_H__ */