//
// Cforall Version 1.0.0 Copyright (C) 2016 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// kernel --
//
// Author           : Thierry Delisle
// Created On       : Tue Jan 17 12:27:26 2017
// Last Modified By : Peter A. Buhr
// Last Modified On : Wed Dec  4 07:54:51 2019
// Update Count     : 18
//

#pragma once

#include <stdbool.h>

#include "invoke.h"
#include "time_t.hfa"
#include "coroutine.hfa"

extern "C" {
#include <pthread.h>
#include <semaphore.h>
}

//-----------------------------------------------------------------------------
// Locks
struct semaphore {
	__spinlock_t lock;
	int count;
	__queue_t(thread_desc) waiting;
};

void  ?{}(semaphore & this, int count = 1);
void ^?{}(semaphore & this);
void   P (semaphore & this);
void   V (semaphore & this);


//-----------------------------------------------------------------------------
// Processor
extern struct cluster * mainCluster;

enum FinishOpCode { No_Action, Release, Schedule, Release_Schedule, Release_Multi, Release_Multi_Schedule, Callback };

typedef void (*__finish_callback_fptr_t)(void);

//TODO use union, many of these fields are mutually exclusive (i.e. MULTI vs NOMULTI)
struct FinishAction {
	FinishOpCode action_code;
	/*
	// Union of possible actions
	union {
		// Option 1 : locks and threads
		struct {
			// 1 thread or N thread
			union {
				thread_desc * thrd;
				struct {
					thread_desc ** thrds;
					unsigned short thrd_count;
				};
			};
			// 1 lock or N lock
			union {
				__spinlock_t * lock;
				struct {
					__spinlock_t ** locks;
					unsigned short lock_count;
				};
			};
		};
		// Option 2 : action pointer
		__finish_callback_fptr_t callback;
	};
	/*/
	thread_desc * thrd;
	thread_desc ** thrds;
	unsigned short thrd_count;
	__spinlock_t * lock;
	__spinlock_t ** locks;
	unsigned short lock_count;
	__finish_callback_fptr_t callback;
	//*/
};
static inline void ?{}(FinishAction & this) {
	this.action_code = No_Action;
	this.thrd = 0p;
	this.lock = 0p;
}
static inline void ^?{}(FinishAction &) {}

// Processor
coroutine processorCtx_t {
	struct processor * proc;
};

// Wrapper around kernel threads
struct processor {
	// Main state
	// Coroutine ctx who does keeps the state of the processor
	struct processorCtx_t runner;

	// Cluster from which to get threads
	struct cluster * cltr;
	unsigned int id;

	// Name of the processor
	const char * name;

	// Handle to pthreads
	pthread_t kernel_thread;

	// RunThread data
	// Action to do after a thread is ran
	struct FinishAction finish;

	// Preemption data
	// Node which is added in the discrete event simulaiton
	struct alarm_node_t * preemption_alarm;

	// If true, a preemption was triggered in an unsafe region, the processor must preempt as soon as possible
	bool pending_preemption;

	// Idle lock
	__bin_sem_t idleLock;

	// Termination
	// Set to true to notify the processor should terminate
	volatile bool do_terminate;

	// Termination synchronisation
	semaphore terminated;

	// pthread Stack
	void * stack;

	// Link lists fields
	struct __dbg_node_proc {
		struct processor * next;
		struct processor * prev;
	} node;

#ifdef __CFA_DEBUG__
	// Last function to enable preemption on this processor
	const char * last_enable;
#endif
};

void  ?{}(processor & this, const char * name, struct cluster & cltr);
void ^?{}(processor & this);

static inline void  ?{}(processor & this)                    { this{ "Anonymous Processor", *mainCluster}; }
static inline void  ?{}(processor & this, struct cluster & cltr)    { this{ "Anonymous Processor", cltr}; }
static inline void  ?{}(processor & this, const char * name) { this{name, *mainCluster }; }

static inline [processor *&, processor *& ] __get( processor & this ) {
	return this.node.[next, prev];
}


//-----------------------------------------------------------------------------
// Cluster Tools
struct __processor_id;

// Reader-Writer lock protecting the ready-queue
struct __clusterRWLock_t {
	// total cachelines allocated
	unsigned int max;

	// cachelines currently in use
	volatile unsigned int alloc;

	// cachelines ready to itereate over
	// (!= to alloc when thread is in second half of doregister)
	volatile unsigned int ready;

	// writer lock
	volatile bool lock;

	// data pointer
	__processor_id * data;
};

void  ?{}(__clusterRWLock_t & this);
void ^?{}(__clusterRWLock_t & this);

// Underlying sub quues of the ready queue
struct __attribute__((aligned(128))) __intrusive_ready_queue_t {
	// spin lock protecting the queue
	volatile bool lock;
	unsigned int last_id;
	unsigned int count;

	// anchor for the head and the tail of the queue
	struct __sentinel_t {
		// Link lists fields
		// instrusive link field for threads
		// must be exactly as in thread_desc
		__thread_desc_link link;
	} before, after;

	// Optional statistic counters
	#if !defined(__CFA_NO_SCHED_STATS__)
		struct __attribute__((aligned(64))) {
			// difference between number of push and pops
			ssize_t diff;

			// total number of pushes and pops
			size_t  push;
			size_t  pop ;
		} stat;
	#endif
};

void  ?{}(__intrusive_ready_queue_t & this);
void ^?{}(__intrusive_ready_queue_t & this);

typedef unsigned long long __cfa_readyQ_mask_t;

// enum {
// 	__cfa_ready_queue_mask_size = (64 - sizeof(size_t)) / sizeof(size_t),
// 	__cfa_max_ready_queues = __cfa_ready_queue_mask_size * 8 * sizeof(size_t)
// };

#define __cfa_readyQ_mask_size ((64 - sizeof(size_t)) / sizeof(__cfa_readyQ_mask_t))
#define __cfa_max_readyQs (__cfa_readyQ_mask_size * 8 * sizeof(__cfa_readyQ_mask_t))

//TODO adjust cache size to ARCHITECTURE
struct __attribute__((aligned(128))) __ready_queue_t {
	struct {
		volatile size_t count;
		volatile __cfa_readyQ_mask_t mask[ __cfa_readyQ_mask_size ];
	} empty;

	struct __attribute__((aligned(64))) {
		__intrusive_ready_queue_t * volatile data;
		volatile size_t count;
	} list;

	#if !defined(__CFA_NO_STATISTICS__)
		__attribute__((aligned(64))) struct {
			struct {
				struct {
					volatile size_t attempt;
					volatile size_t success;
				} push;
				struct {
					volatile size_t maskrds;
					volatile size_t attempt;
					volatile size_t success;
				} pop;
			} pick;
			struct {
				volatile size_t value;
				volatile size_t count;
			} full;
		} global_stats;

	#endif
};

void  ?{}(__ready_queue_t & this);
void ^?{}(__ready_queue_t & this);

//-----------------------------------------------------------------------------
// Cluster
struct cluster {
	// Ready queue locks
	__clusterRWLock_t ready_lock;

	// Ready queue for threads
	__ready_queue_t ready_queue;

	// Name of the cluster
	const char * name;

	// Preemption rate on this cluster
	Duration preemption_rate;

	// List of processors
	__spinlock_t proc_list_lock;
	__dllist_t(struct processor) idles;

	// List of threads
	__spinlock_t thread_list_lock;
	__dllist_t(struct thread_desc) threads;
	unsigned int nthreads;

	// Link lists fields
	struct __dbg_node_cltr {
		cluster * next;
		cluster * prev;
	} node;
};
extern Duration default_preemption();

void ?{} (cluster & this, const char * name, Duration preemption_rate);
void ^?{}(cluster & this);

static inline void ?{} (cluster & this)                           { this{"Anonymous Cluster", default_preemption()}; }
static inline void ?{} (cluster & this, Duration preemption_rate) { this{"Anonymous Cluster", preemption_rate}; }
static inline void ?{} (cluster & this, const char * name)        { this{name, default_preemption()}; }

static inline [cluster *&, cluster *& ] __get( cluster & this ) {
	return this.node.[next, prev];
}

static inline struct processor * active_processor() { return TL_GET( this_processor ); } // UNSAFE
static inline struct cluster   * active_cluster  () { return TL_GET( this_processor )->cltr; }

// Local Variables: //
// mode: c //
// tab-width: 4 //
// End: //