edgetpu: sync fixes from darwinn-internal

Roll-up the following commits suggested for including in pixel 5.10:

90c2dbb04 edgetpu: add sysfs attr clients to dump client and wakelock state
465a02519 edgetpu: PM log clients holding TPU wakelocks at suspend reject time
6614f7487 edgetpu: add list of clients per device
bcbadfe1c edgetpu: fix kernel paging error in edgetpu_mmu_attach_domain

Bug: 193484549
Bug: 193591701
Signed-off-by: Claire Chang <tientzu@google.com>
Signed-off-by: Todd Poynor <toddpoynor@google.com>
Change-Id: I19eb15187af66d35360349929738c9de6f5cc05b

5 files changed, 93 insertions, 18 deletions

diff --git a/drivers/edgetpu/edgetpu-core.c b/drivers/edgetpu/edgetpu-core.c
index 7701a71..bf2ad2d 100644
--- a/drivers/edgetpu/edgetpu-core.c
+++ b/drivers/edgetpu/edgetpu-core.c
@@ -371,6 +371,8 @@ int edgetpu_device_add(struct edgetpu_dev *etdev,
 	INIT_LIST_HEAD(&etdev->groups);
 	etdev->n_groups = 0;
 	etdev->group_join_lockout = false;
+	mutex_init(&etdev->clients_lock);
+	INIT_LIST_HEAD(&etdev->clients);
 	etdev->vcid_pool = (1u << EDGETPU_NUM_VCIDS) - 1;
 	mutex_init(&etdev->state_lock);
 	etdev->state = ETDEV_STATE_NOFW;
@@ -457,13 +459,19 @@ void edgetpu_device_remove(struct edgetpu_dev *etdev)
 struct edgetpu_client *edgetpu_client_add(struct edgetpu_dev *etdev)
 {
 	struct edgetpu_client *client;
+	struct edgetpu_list_device_client *l = kmalloc(sizeof(*l), GFP_KERNEL);
 
+	if (!l)
+		return ERR_PTR(-ENOMEM);
 	client = kzalloc(sizeof(*client), GFP_KERNEL);
-	if (!client)
+	if (!client) {
+		kfree(l);
 		return ERR_PTR(-ENOMEM);
+	}
 	client->wakelock = edgetpu_wakelock_alloc(etdev);
 	if (!client->wakelock) {
 		kfree(client);
+		kfree(l);
 		return ERR_PTR(-ENOMEM);
 	}
 
@@ -474,6 +482,10 @@ struct edgetpu_client *edgetpu_client_add(struct edgetpu_dev *etdev)
 	/* equivalent to edgetpu_client_get() */
 	refcount_set(&client->count, 1);
 	client->perdie_events = 0;
+	mutex_lock(&etdev->clients_lock);
+	l->client = client;
+	list_add_tail(&l->list, &etdev->clients);
+	mutex_unlock(&etdev->clients_lock);
 	return client;
 }
 
@@ -494,14 +506,27 @@ void edgetpu_client_put(struct edgetpu_client *client)
 void edgetpu_client_remove(struct edgetpu_client *client)
 {
 	struct edgetpu_dev *etdev;
+	struct edgetpu_list_device_client *lc;
 
 	if (IS_ERR_OR_NULL(client))
 		return;
 	etdev = client->etdev;
+	mutex_lock(&etdev->clients_lock);
+	/* remove the client from the device list */
+	for_each_list_device_client(etdev, lc) {
+		if (lc->client == client) {
+			list_del(&lc->list);
+			kfree(lc);
+			break;
+		}
+	}
+	mutex_unlock(&etdev->clients_lock);
 	/*
 	 * A quick check without holding client->group_lock.
 	 *
-	 * If client doesn't belong to a group then we are fine to not proceed.
+	 * If client doesn't belong to a group then we are fine to not remove
+	 * from groups.
+	 *
 	 * If there is a race that the client belongs to a group but is removing
 	 * by another process - this will be detected by the check with holding
 	 * client->group_lock later.
diff --git a/drivers/edgetpu/edgetpu-device-group.c b/drivers/edgetpu/edgetpu-device-group.c
index 6172b2c..9e07a03 100644
--- a/drivers/edgetpu/edgetpu-device-group.c
+++ b/drivers/edgetpu/edgetpu-device-group.c
@@ -704,18 +704,11 @@ edgetpu_device_group_alloc(struct edgetpu_client *client,
 	group->mbox_attr = *attr;
 	if (attr->priority & EDGETPU_PRIORITY_DETACHABLE)
 		group->mailbox_detachable = true;
-	/* adds @client as the first entry */
-	ret = edgetpu_device_group_add(group, client);
-	if (ret) {
-		etdev_dbg(group->etdev, "%s: group %u add failed ret=%d",
-			  __func__, group->workload_id, ret);
-		goto error_put_group;
-	}
 
 	etdomain = edgetpu_mmu_alloc_domain(group->etdev);
 	if (!etdomain) {
 		ret = -ENOMEM;
-		goto error_leave_group;
+		goto error_put_group;
 	}
 	group->etdomain = etdomain;
 	if (etdomain->token != EDGETPU_DOMAIN_TOKEN_END)
@@ -723,10 +716,18 @@ edgetpu_device_group_alloc(struct edgetpu_client *client,
 			EDGETPU_CONTEXT_DOMAIN_TOKEN | etdomain->token;
 	else
 		group->context_id = EDGETPU_CONTEXT_INVALID;
+
+	/* adds @client as the first entry */
+	ret = edgetpu_device_group_add(group, client);
+	if (ret) {
+		etdev_dbg(group->etdev, "%s: group %u add failed ret=%d",
+			  __func__, group->workload_id, ret);
+		goto error_free_mmu_domain;
+	}
 	return group;
 
-error_leave_group:
-	edgetpu_device_group_leave(client);
+error_free_mmu_domain:
+	edgetpu_mmu_free_domain(group->etdev, group->etdomain);
 error_put_group:
 	edgetpu_device_group_put(group);
 error:
diff --git a/drivers/edgetpu/edgetpu-fs.c b/drivers/edgetpu/edgetpu-fs.c
index 1b6e039..c16c964 100644
--- a/drivers/edgetpu/edgetpu-fs.c
+++ b/drivers/edgetpu/edgetpu-fs.c
@@ -1003,9 +1003,32 @@ static ssize_t watchdog_timeout_count_show(
 }
 static DEVICE_ATTR_RO(watchdog_timeout_count);
 
+static ssize_t clients_show(
+		struct device *dev, struct device_attribute *attr,
+		char *buf)
+{
+	struct edgetpu_dev *etdev = dev_get_drvdata(dev);
+	struct edgetpu_list_device_client *lc;
+	ssize_t ret = 0;
+
+	mutex_lock(&etdev->clients_lock);
+	for_each_list_device_client(etdev, lc) {
+		ret += scnprintf(buf, PAGE_SIZE - ret,
+				 "pid %d tgid %d wakelock %d\n",
+				 lc->client->pid, lc->client->tgid,
+				 NO_WAKELOCK(lc->client->wakelock) ?
+				 0 : lc->client->wakelock->req_count);
+		buf += ret;
+	}
+	mutex_unlock(&etdev->clients_lock);
+	return ret;
+}
+static DEVICE_ATTR_RO(clients);
+
 static struct attribute *edgetpu_dev_attrs[] = {
 	&dev_attr_firmware_crash_count.attr,
 	&dev_attr_watchdog_timeout_count.attr,
+	&dev_attr_clients.attr,
 	NULL,
 };
 
diff --git a/drivers/edgetpu/edgetpu-internal.h b/drivers/edgetpu/edgetpu-internal.h
index 23e0c12..4c1d0a4 100644
--- a/drivers/edgetpu/edgetpu-internal.h
+++ b/drivers/edgetpu/edgetpu-internal.h
@@ -131,6 +131,16 @@ struct edgetpu_client {
 	u64 perdie_events;
 };
 
+/* edgetpu_dev#clients list entry. */
+struct edgetpu_list_device_client {
+	struct list_head list;
+	struct edgetpu_client *client;
+};
+
+/* Macro to loop through etdev->clients (hold clients_lock prior). */
+#define for_each_list_device_client(etdev, c)                                  \
+	list_for_each_entry(c, &etdev->clients, list)
+
 struct edgetpu_mapping;
 struct edgetpu_mailbox_manager;
 struct edgetpu_kci;
@@ -179,6 +189,8 @@ struct edgetpu_dev {
 
 	/* end of fields protected by @groups_lock */
 
+	struct mutex clients_lock; /* protects clients */
+	struct list_head clients;
 	void *mmu_cookie;	   /* mmu driver private data */
 	void *dram_cookie;	   /* on-device DRAM private data */
 	struct edgetpu_mailbox_manager *mailbox_manager;
diff --git a/drivers/edgetpu/edgetpu-pm.c b/drivers/edgetpu/edgetpu-pm.c
index df1c179..872149f 100644
--- a/drivers/edgetpu/edgetpu-pm.c
+++ b/drivers/edgetpu/edgetpu-pm.c
@@ -15,6 +15,7 @@
 #include "edgetpu-mailbox.h"
 #include "edgetpu-pm.h"
 #include "edgetpu-sw-watchdog.h"
+#include "edgetpu-wakelock.h"
 
 #if IS_ENABLED(CONFIG_EDGETPU_TEST)
 #include "unittests/factory/fake-edgetpu-firmware.h"
@@ -310,15 +311,28 @@ void edgetpu_pchannel_power_up(struct edgetpu_dev *etdev)
 int edgetpu_pm_suspend(struct edgetpu_dev *etdev)
 {
 	struct edgetpu_pm *etpm = etdev->pm;
+	struct edgetpu_list_device_client *lc;
 
-	if (etpm && etpm->p->power_up_count) {
-		etdev_warn_ratelimited(
-			etdev, "cannot suspend with power up count = %d\n",
-			etpm->p->power_up_count);
+	if (!etpm || !etpm->p->power_up_count)
+		return 0;
+
+	etdev_warn_ratelimited(
+		etdev, "cannot suspend with power up count = %d\n",
+		etpm->p->power_up_count);
+
+	if (!mutex_trylock(&etdev->clients_lock))
 		return -EAGAIN;
+	for_each_list_device_client(etdev, lc) {
+		if (NO_WAKELOCK(lc->client->wakelock) ||
+		    !lc->client->wakelock->req_count)
+			continue;
+		etdev_warn_ratelimited(etdev, "pid %d tgid %d count %d\n",
+				       lc->client->pid,
+				       lc->client->tgid,
+				       lc->client->wakelock->req_count);
 	}
-
-	return 0;
+	mutex_unlock(&etdev->clients_lock);
+	return -EAGAIN;
 }
 
 int edgetpu_pm_resume(struct edgetpu_dev *etdev)