Add GPU work period support for Xe driver

This patch implements the Android VSR requirement GPU work
period event for Intel GPUs

The requirement is implemented using a timer to give periodic
interrupts and a worker thread per app instance to accumulate
its run time on gpu in nanoseconds using the hw ctx timestamp
and emit the event.

Tracked-On: ALOS-1015
Signed-off-by: Aakash Sarkar <aakash.deep.sarkar@intel.com>

Author: Aakash-Sarkar

drivers/gpu/drm/xe/xe_device.c

@@ -8,6 +8,7 @@
 #include <linux/aperture.h>
 #include <linux/delay.h>
 #include <linux/fault-inject.h>
+#include <linux/jiffies.h>
 #include <linux/units.h>
 
 #include <drm/drm_atomic_helper.h>
@@ -65,9 +66,13 @@ static int xe_file_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct xe_device *xe = to_xe_device(dev);
 	struct xe_drm_client *client;
+	struct xe_user *user;
 	struct xe_file *xef;
 	int ret = -ENOMEM;
+	unsigned long flags;
+	unsigned int uid = 0;
 	struct task_struct *task = NULL;
+	const struct cred *cred = NULL;
 
 	xef = kzalloc(sizeof(*xef), GFP_KERNEL);
 	if (!xef)
@@ -94,11 +99,52 @@ static int xe_file_open(struct drm_device *dev, struct drm_file *file)
 
 	task = get_pid_task(rcu_access_pointer(file->pid), PIDTYPE_PID);
 	if (task) {
+		cred = get_task_cred(task);
+		if (cred) {
+			uid = cred->euid.val;
+			put_cred(cred);
+		}
 		xef->process_name = kstrdup(task->comm, GFP_KERNEL);
 		xef->pid = task->pid;
 		put_task_struct(task);
 	}
 
+	/*
+	 * Check if the calling process/uid has already been registered
+	 * by the xe driver during a previous file open call. If so then
+	 * take a reference to this xe file and add it to the list of xe
+	 * files belonging to the this process' xe user
+	 */
+	spin_lock_irqsave(&xe->work_period.lock, flags);
+	list_for_each_entry(user, &xe->work_period.user_list, entry) {
+		if (user->uid == uid) {
+			xef->user = xe_user_get(user);
+			goto filelist_add;
+		}
+	}
+	/* list_add could be ordered wrt to list_for_each_entry */
+	smp_mb();
+
+	/*
+	 * We couldn't find a xe user for this process. allocate a new
+	 * xe user structure and register it to the xe driver
+	 */
+	user = xe_user_alloc(uid);
+	if (!user) {
+		goto spin_unlk;
+	}
+
+	user->last_timestamp_ns = ktime_get_raw_ns();
+	list_add(&user->entry, &xe->work_period.user_list);
+	xef->user = user;
+
+filelist_add:
+	spin_lock(&user->filelist_lock);
+	list_add(&xef->user_link, &user->filelist);
+	spin_unlock(&user->filelist_lock);
+	user->xe = xe;
+spin_unlk:
+	spin_unlock_irqrestore(&xe->work_period.lock, flags);
 	return 0;
 }
 
@@ -113,6 +159,9 @@ static void xe_file_destroy(struct kref *ref)
 
 	xe_drm_client_put(xef->client);
 	kfree(xef->process_name);
+
+	list_del(&xef->user_link);
+	xe_user_put(xef->user);
 	kfree(xef);
 }
 
@@ -231,6 +280,24 @@ static long xe_drm_compat_ioctl(struct file *file, unsigned int cmd, unsigned lo
 #define xe_drm_compat_ioctl NULL
 #endif
 
+static void work_period_timer_fn(struct timer_list *timer)
+{
+	struct xe_device *xe = container_of(timer, typeof(*xe), work_period.timer);
+	struct xe_user *user;
+	unsigned long timeout = 0;
+
+	spin_lock(&xe->work_period.lock);
+	if (!list_empty(&xe->work_period.user_list)) {
+		list_for_each_entry(user, &xe->work_period.user_list, entry) {
+			queue_work(xe->work_period.wq, &user->work);
+		}
+	}
+	spin_unlock(&xe->work_period.lock);
+	timeout = jiffies + msecs_to_jiffies(500);
+
+	mod_timer(timer, timeout);
+}
+
 static const struct file_operations xe_driver_fops = {
 	.owner = THIS_MODULE,
 	.open = drm_open,
@@ -291,6 +358,11 @@ static void xe_device_destroy(struct drm_device *dev, void *dummy)
 	if (xe->destroy_wq)
 		destroy_workqueue(xe->destroy_wq);
 
+	if (xe->work_period.wq)
+		destroy_workqueue(xe->work_period.wq);
+
+	del_timer(&xe->work_period.timer);
+
 	ttm_device_fini(&xe->ttm);
 }
 
@@ -350,13 +422,22 @@ struct xe_device *xe_device_create(struct pci_dev *pdev,
 	INIT_LIST_HEAD(&xe->pinned.external_vram);
 	INIT_LIST_HEAD(&xe->pinned.evicted);
 
+	spin_lock_init(&xe->work_period.lock);
+	INIT_LIST_HEAD(&xe->work_period.user_list);
+	timer_setup(&xe->work_period.timer, work_period_timer_fn, 0);
+	xe->work_period.timer.expires = jiffies + msecs_to_jiffies(1000);
+	add_timer(&xe->work_period.timer);
+
+	xe->work_period.wq = alloc_workqueue("xe-work-period-wq", 0, 0);
+
 	xe->preempt_fence_wq = alloc_ordered_workqueue("xe-preempt-fence-wq",
 						       WQ_MEM_RECLAIM);
 	xe->ordered_wq = alloc_ordered_workqueue("xe-ordered-wq", 0);
 	xe->unordered_wq = alloc_workqueue("xe-unordered-wq", 0, 0);
 	xe->destroy_wq = alloc_workqueue("xe-destroy-wq", 0, 0);
 	if (!xe->ordered_wq || !xe->unordered_wq ||
-	    !xe->preempt_fence_wq || !xe->destroy_wq) {
+		!xe->preempt_fence_wq || !xe->destroy_wq ||
+		!xe->work_period.wq) {
 		/*
 		 * Cleanup done in xe_device_destroy via
 		 * drmm_add_action_or_reset register above

drivers/gpu/drm/xe/xe_device_types.h

@@ -405,6 +405,20 @@ struct xe_device {
 		struct list_head external_vram;
 	} pinned;
 
+	/** @work_period: gpu work period event */
+	struct {
+		/** @lock: lock protecting this structure */
+		spinlock_t lock;
+		/** @timer: timer to give periodic interrupts to emit the
+		 * gpu work period event
+		 */
+		struct timer_list timer;
+		/** @user_list: list of xe users using this xe device */
+		struct list_head user_list;
+		/** @wq: workqueue for work period event emitting work */
+		struct workqueue_struct *wq;
+	} work_period;
+
 	/** @ufence_wq: user fence wait queue */
 	wait_queue_head_t ufence_wq;
 
@@ -619,6 +633,9 @@ struct xe_file {
 	/** @run_ticks: hw engine class run time in ticks for this drm client */
 	u64 run_ticks[XE_ENGINE_CLASS_MAX];
 
+	/** @active_duration_ns: total run time in ns for this drm client */
+	u64 active_duration_ns;
+
 	/** @client: drm client */
 	struct xe_drm_client *client;
 
@@ -634,6 +651,15 @@ struct xe_file {
 	 */
 	pid_t pid;
 
+	/**
+	 * @user_link: entry into xe_user.filelist list
+	 */
+	struct list_head user_link;
+	/**
+	 * @user: pointer to the xe user struct that opened this xe file
+	 */
+	struct xe_user *user;
+
 	/** @refcount: ref count of this xe file */
 	struct kref refcount;
 };

drivers/gpu/drm/xe/xe_drm_client.c

@@ -21,6 +21,9 @@
 #include "xe_pm.h"
 #include "xe_trace.h"
 
+#define CREATE_TRACE_POINTS
+#include "xe_power_gpu_work_period_trace.h"
+
 /**
  * DOC: DRM Client usage stats
  *
@@ -395,3 +398,109 @@ void xe_drm_client_fdinfo(struct drm_printer *p, struct drm_file *file)
 	show_run_ticks(p, file);
 }
 #endif
+
+/**
+ * worker thread to emit gpu work period event for this xe user
+ * @work: work instance for this xe user
+ *
+ * Return: void
+ */
+static inline void work_period_worker(struct work_struct *work)
+{
+	struct xe_user *user = container_of(work, struct xe_user, work);
+	struct xe_device *xe = user->xe;
+	struct xe_file *xef;
+	struct xe_exec_queue *q;
+	u64 last_active_duration, last_timestamp;
+	u32 gpuid = 0, uid = user->uid;
+	u64 start_time, end_time, active_duration;
+	unsigned long i;
+
+	last_active_duration = user->active_duration_ns;
+	last_timestamp = user->last_timestamp_ns;
+
+	xe_pm_runtime_get(xe);
+
+	if (!list_empty(&user->filelist)) {
+		spin_lock(&user->filelist_lock);
+		list_for_each_entry(xef, &user->filelist, user_link) {
+
+			/*
+			 * Wait for any exec queue going away:
+			 * their cycles will get updated on context switch out,
+			 * so wait for that to happen
+			 */
+			wait_var_event(&xef->exec_queue.pending_removal,
+		    !atomic_read(&xef->exec_queue.pending_removal));
+
+			/* Accumulate all the exec queues from this xe file */
+			mutex_lock(&xef->exec_queue.lock);
+			xa_for_each(&xef->exec_queue.xa, i, q) {
+				xe_exec_queue_get(q);
+				mutex_unlock(&xef->exec_queue.lock);
+
+				xe_exec_queue_update_run_ticks(q);
+
+				mutex_lock(&xef->exec_queue.lock);
+				xe_exec_queue_put(q);
+			}
+			mutex_unlock(&xef->exec_queue.lock);
+			user->active_duration_ns += xef->active_duration_ns;
+		}
+		spin_unlock(&user->filelist_lock);
+	}
+
+	xe_pm_runtime_put(xe);
+
+	start_time = last_timestamp + 1;
+	end_time = ktime_get_raw_ns();
+	active_duration = user->active_duration_ns - last_active_duration;
+	/* emit the gpu work period event */
+	trace_gpu_work_period(gpuid, uid, start_time, end_time, active_duration);
+	user->last_timestamp_ns = end_time;
+}
+
+/**
+ * xe_user_alloc() - Allocate xe user
+ * @void: No arg
+ *
+ * Allocate xe user struct to track activity on the gpu
+ * by the application. Call this API whenever a new app
+ * has opened xe device.
+ *
+ * Return: pointer to user struct or NULL if can't allocate
+ */
+struct xe_user *xe_user_alloc(const unsigned int uid)
+{
+	struct xe_user *user;
+
+	user = kzalloc(sizeof(*user), GFP_KERNEL);
+	if (!user || !uid)
+		return NULL;
+
+	user->uid = uid;
+	kref_init(&user->kref);
+	spin_lock_init(&user->filelist_lock);
+	INIT_LIST_HEAD(&user->filelist);
+	INIT_LIST_HEAD(&user->entry);
+	INIT_WORK(&user->work, work_period_worker);
+	return user;
+}
+
+/**
+ * __xe_user_free() - Free user struct
+ * @kref: The reference
+ *
+ * destroy this xe user when the last xe file associated with
+ * this xe user is destroyed
+ *
+ * Return: void
+ */
+void __xe_user_free(struct kref *kref)
+{
+	struct xe_user *user =
+		container_of(kref, typeof(*user), kref);
+
+	list_del(&user->entry);
+	kfree(user);
+}

drivers/gpu/drm/xe/xe_drm_client.h

@@ -12,6 +12,7 @@
 #include <linux/rcupdate.h>
 #include <linux/sched.h>
 #include <linux/spinlock.h>
+#include <linux/workqueue.h>
 
 struct drm_file;
 struct drm_printer;
@@ -34,6 +35,50 @@ struct xe_drm_client {
 #endif
 };
 
+/**
+ * This is a per process/user id structure for a xe device
+ * client. It is allocated when a new process/app opens the
+ * xe device and destroyed when the last xe file for this
+ * process is destroyed
+ */
+struct xe_user {
+	struct kref kref;
+	struct xe_device *xe;
+	/**
+	 * @filelist_lock: lock protecting the filelist list
+	 */
+	spinlock_t filelist_lock;
+	/**
+	 * @filelist: list of xe files belonging to this process
+	 */
+	struct list_head filelist;
+	/**
+	 * @entry: entry into the xe.work_period.user_list list
+	 */
+	struct list_head entry;
+	/**
+	 * @work: work to emit the gpu work period event for this xe user
+	 */
+	struct work_struct work;
+	/**
+	 * @uid: user id for this process/app
+	 *
+	 * In android each app has its own user id. So we use uid to identify
+	 * an app that is using the gpu
+	 */
+	u32 uid;
+	/**
+	 * @active_duration_ns: sum total of xe_file.active_duration_ns for all
+	 * xe files belonging to this xe user
+	 */
+	u64 active_duration_ns;
+	/**
+	 * @last_timestamp_ns: timestamp in ns when we last emitted event for
+	 * this xe user
+	 */
+	u64 last_timestamp_ns;
+};
+
 	static inline struct xe_drm_client *
 xe_drm_client_get(struct xe_drm_client *client)
 {
@@ -67,4 +112,20 @@ static inline void xe_drm_client_remove_bo(struct xe_bo *bo)
 {
 }
 #endif
+
+struct xe_user *xe_user_alloc(const unsigned int uid);
+
+static inline struct xe_user *
+xe_user_get(struct xe_user *user)
+{
+	kref_get(&user->kref);
+	return user;
+}
+
+void __xe_user_free(struct kref *kref);
+
+static inline void xe_user_put(struct xe_user *user)
+{
+	kref_put(&user->kref, __xe_user_free);
+}
 #endif