//
// 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_private.hfa --
//
// Author           : Thierry Delisle
// Created On       : Mon Feb 13 12:27:26 2017
// Last Modified By : Peter A. Buhr
// Last Modified On : Thu Mar 29 14:06:40 2018
// Update Count     : 3
//

#pragma once

#include "kernel.hfa"
#include "thread.hfa"

#include "alarm.hfa"


//-----------------------------------------------------------------------------
// Scheduler

extern "C" {
	void disable_interrupts() OPTIONAL_THREAD;
	void enable_interrupts_noPoll();
	void enable_interrupts( __cfaabi_dbg_ctx_param );
}

void ScheduleThread( thread_desc * );
static inline void WakeThread( thread_desc * thrd ) {
	if( !thrd ) return;

	verify(thrd->state == Inactive);

	disable_interrupts();
	ScheduleThread( thrd );
	enable_interrupts( __cfaabi_dbg_ctx );
}
thread_desc * nextThread(cluster * this);

//Block current thread and release/wake-up the following resources
void BlockInternal(void);
void BlockInternal(__spinlock_t * lock);
void BlockInternal(thread_desc * thrd);
void BlockInternal(__spinlock_t * lock, thread_desc * thrd);
void BlockInternal(__spinlock_t * locks [], unsigned short count);
void BlockInternal(__spinlock_t * locks [], unsigned short count, thread_desc * thrds [], unsigned short thrd_count);
void BlockInternal(__finish_callback_fptr_t callback);
void LeaveThread(__spinlock_t * lock, thread_desc * thrd);

//-----------------------------------------------------------------------------
// Processor
void main(processorCtx_t *);

static inline void wake_fast(processor * this) {
	__cfaabi_dbg_print_safe("Kernel : Waking up processor %p\n", this);
	post( this->idleLock );
}

static inline void wake(processor * this) {
	disable_interrupts();
	wake_fast(this);
	enable_interrupts( __cfaabi_dbg_ctx );
}

struct event_kernel_t {
	alarm_list_t alarms;
	__spinlock_t lock;
};

extern event_kernel_t * event_kernel;

struct __cfa_kernel_preemption_state_t {
	bool enabled;
	bool in_progress;
	unsigned short disable_count;
};

extern volatile thread_local __cfa_kernel_preemption_state_t preemption_state __attribute__ ((tls_model ( "initial-exec" )));

//-----------------------------------------------------------------------------
// Threads
extern "C" {
      forall(dtype T | is_thread(T))
      void CtxInvokeThread(T * this);
}

extern void ThreadCtxSwitch(coroutine_desc * src, coroutine_desc * dst);

__cfaabi_dbg_debug_do(
	extern void __cfaabi_dbg_thread_register  ( thread_desc * thrd );
	extern void __cfaabi_dbg_thread_unregister( thread_desc * thrd );
)

//-----------------------------------------------------------------------------
// Utils
#define KERNEL_STORAGE(T,X) __attribute((aligned(__alignof__(T)))) static char storage_##X[sizeof(T)]

static inline uint32_t tls_rand() {
	kernelTLS.rand_seed ^= kernelTLS.rand_seed << 6;
	kernelTLS.rand_seed ^= kernelTLS.rand_seed >> 21;
	kernelTLS.rand_seed ^= kernelTLS.rand_seed << 7;
	return kernelTLS.rand_seed;
}


void doregister( struct cluster & cltr );
void unregister( struct cluster & cltr );

void doregister( struct cluster * cltr, struct thread_desc & thrd );
void unregister( struct cluster * cltr, struct thread_desc & thrd );

//=======================================================================
// Cluster lock API
//=======================================================================
struct __attribute__((aligned(64))) __processor_id {
	processor * volatile handle;
	volatile bool lock;
};

// Lock-Free registering/unregistering of threads
// Register a processor to a given cluster and get its unique id in return
unsigned doregister( struct cluster * cltr, struct processor * proc );

// Unregister a processor from a given cluster using its id, getting back the original pointer
void     unregister( struct cluster * cltr, struct processor * proc );

//=======================================================================
// Reader-writer lock implementation
// Concurrent with doregister/unregister,
//    i.e., threads can be added at any point during or between the entry/exit
static inline void __atomic_acquire(volatile bool * ll) {
	while( __builtin_expect(__atomic_exchange_n(ll, (bool)true, __ATOMIC_SEQ_CST), false) ) {
		while(__atomic_load_n(ll, (int)__ATOMIC_RELAXED))
			asm volatile("pause");
	}
	/* paranoid */ verify(*ll);
}

static inline bool __atomic_try_acquire(volatile bool * ll) {
	return !__atomic_exchange_n(ll, (bool)true, __ATOMIC_SEQ_CST);
}

static inline void __atomic_unlock(volatile bool * ll) {
	/* paranoid */ verify(*ll);
	__atomic_store_n(ll, (bool)false, __ATOMIC_RELEASE);
}

//-----------------------------------------------------------------------
// Reader side : acquire when using the ready queue to schedule but not
//  creating/destroying queues
static inline void ready_schedule_lock( struct cluster * cltr, struct processor * proc) with(cltr->ready_lock) {
	unsigned iproc = proc->id;
	/*paranoid*/ verify(data[iproc].handle == proc);
	/*paranoid*/ verify(iproc < ready);

	// Step 1 : make sure no writer are in the middle of the critical section
	while(__atomic_load_n(&lock, (int)__ATOMIC_RELAXED))
		asm volatile("pause");

	// Fence needed because we don't want to start trying to acquire the lock
	// before we read a false.
	// Not needed on x86
	// std::atomic_thread_fence(std::memory_order_seq_cst);

	// Step 2 : acquire our local lock
	__atomic_acquire( &data[iproc].lock );
	/*paranoid*/ verify(data[iproc].lock);
}

static inline void ready_schedule_unlock( struct cluster * cltr, struct processor * proc) with(cltr->ready_lock) {
	unsigned iproc = proc->id;
	/*paranoid*/ verify(data[iproc].handle == proc);
	/*paranoid*/ verify(iproc < ready);
	/*paranoid*/ verify(data[iproc].lock);
	__atomic_store_n(&data[iproc].lock, false, __ATOMIC_RELEASE);
}

//-----------------------------------------------------------------------
// Writer side : acquire when changing the ready queue, e.g. adding more
//  queues or removing them.
uint_fast32_t ready_mutate_lock( struct cluster & cltr );

void ready_mutate_unlock( struct cluster & cltr, uint_fast32_t );

//=======================================================================
// Ready-Queue API

__attribute__((hot)) bool push(struct cluster * cltr, struct thread_desc * thrd);
__attribute__((hot)) thread_desc * pop(struct cluster * cltr);
void ready_queue_grow  (struct cluster * cltr);
void ready_queue_shrink(struct cluster * cltr);

#if !defined(__CFA_NO_STATISTICS__)
void stats_tls_tally(struct cluster * cltr);
#else
static inline void stats_tls_tally(struct cluster * cltr) {}
#endif

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