source: libcfa/src/concurrency/locks.hfa@ a5a6a1a8

//
// Cforall Version 1.0.0 Copyright (C) 2021 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// locks.hfa -- PUBLIC
// Runtime locks that used with the runtime thread system.
//
// Author           : Colby Alexander Parsons
// Created On       : Thu Jan 21 19:46:50 2021
// Last Modified By :
// Last Modified On :
// Update Count     :
//

#pragma once

#include <stdbool.h>
#include <stdio.h>

#include "bits/weakso_locks.hfa"
#include "containers/queueLockFree.hfa"
#include "containers/list.hfa"

#include "limits.hfa"
#include "thread.hfa"

#include "time_t.hfa"
#include "time.hfa"

//-----------------------------------------------------------------------------
// Semaphores

// '0-nary' semaphore
// Similar to a counting semaphore except the value of one is never reached
// as a consequence, a V() that would bring the value to 1 *spins* until
// a P consumes it
struct Semaphore0nary {
        __spinlock_t lock; // needed to protect
        mpsc_queue(thread$) queue;
};

static inline bool P(Semaphore0nary & this, thread$ * thrd) {
        /* paranoid */ verify(!thrd`next);
        /* paranoid */ verify(!(&(*thrd)`next));

        push(this.queue, thrd);
        return true;
}

static inline bool P(Semaphore0nary & this) {
    thread$ * thrd = active_thread();
    P(this, thrd);
    park();
    return true;
}

static inline thread$ * V(Semaphore0nary & this, bool doUnpark = true) {
        thread$ * next;
        lock(this.lock __cfaabi_dbg_ctx2);
                for (;;) {
                        next = pop(this.queue);
                        if (next) break;
                        Pause();
                }
        unlock(this.lock);

        if (doUnpark) unpark(next);
        return next;
}

// Wrapper used on top of any sempahore to avoid potential locking
struct BinaryBenaphore {
        volatile ssize_t counter;
};

static inline {
        void ?{}(BinaryBenaphore & this) { this.counter = 0; }
        void ?{}(BinaryBenaphore & this, zero_t) { this.counter = 0; }
        void ?{}(BinaryBenaphore & this, one_t ) { this.counter = 1; }

        // returns true if no blocking needed
        bool P(BinaryBenaphore & this) {
                return __atomic_fetch_sub(&this.counter, 1, __ATOMIC_SEQ_CST) > 0;
        }

        bool tryP(BinaryBenaphore & this) {
                ssize_t c = this.counter;
                /* paranoid */ verify( c > MIN );
                return (c >= 1) && __atomic_compare_exchange_n(&this.counter, &c, c-1, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED);
        }

        // returns true if notify needed
        bool V(BinaryBenaphore & this) {
                ssize_t c = 0;
                for () {
                        /* paranoid */ verify( this.counter < MAX );
                        if (__atomic_compare_exchange_n(&this.counter, &c, c+1, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) {
                                if (c == 0) return true;
                                /* paranoid */ verify(c < 0);
                                return false;
                        } else {
                                if (c == 1) return true;
                                /* paranoid */ verify(c < 1);
                                Pause();
                        }
                }
        }
}

// Binary Semaphore based on the BinaryBenaphore on top of the 0-nary Semaphore
struct ThreadBenaphore {
        BinaryBenaphore ben;
        Semaphore0nary  sem;
};

static inline void ?{}(ThreadBenaphore & this) {}
static inline void ?{}(ThreadBenaphore & this, zero_t) { (this.ben){ 0 }; }
static inline void ?{}(ThreadBenaphore & this, one_t ) { (this.ben){ 1 }; }

static inline bool P(ThreadBenaphore & this)              { return P(this.ben) ? false : P(this.sem); }
static inline bool tryP(ThreadBenaphore & this)           { return tryP(this.ben); }
static inline bool P(ThreadBenaphore & this, bool wait)   { return wait ? P(this) : tryP(this); }

static inline thread$ * V(ThreadBenaphore & this, bool doUnpark = true) {
        if (V(this.ben)) return 0p;
        return V(this.sem, doUnpark);
}

//-----------------------------------------------------------------------------
// Semaphore
struct semaphore {
        __spinlock_t lock;
        int count;
        __queue_t(thread$) waiting;
};

void  ?{}(semaphore & this, int count = 1);
void ^?{}(semaphore & this);
bool   P (semaphore & this);
bool   V (semaphore & this);
bool   V (semaphore & this, unsigned count);
thread$ * V (semaphore & this, bool );

//----------
struct single_acquisition_lock {
        inline blocking_lock;
};

static inline void  ?{}( single_acquisition_lock & this ) {((blocking_lock &)this){ false, false };}
static inline void ^?{}( single_acquisition_lock & this ) {}
static inline void   lock     ( single_acquisition_lock & this ) { lock    ( (blocking_lock &)this ); }
static inline bool   try_lock ( single_acquisition_lock & this ) { return try_lock( (blocking_lock &)this ); }
static inline void   unlock   ( single_acquisition_lock & this ) { unlock  ( (blocking_lock &)this ); }
static inline size_t on_wait  ( single_acquisition_lock & this ) { return on_wait ( (blocking_lock &)this ); }
static inline void   on_wakeup( single_acquisition_lock & this, size_t v ) { on_wakeup ( (blocking_lock &)this, v ); }
static inline void   on_notify( single_acquisition_lock & this, struct thread$ * t ) { on_notify( (blocking_lock &)this, t ); }

//----------
struct owner_lock {
        inline blocking_lock;
};

static inline void  ?{}( owner_lock & this ) {((blocking_lock &)this){ true, true };}
static inline void ^?{}( owner_lock & this ) {}
static inline void   lock     ( owner_lock & this ) { lock    ( (blocking_lock &)this ); }
static inline bool   try_lock ( owner_lock & this ) { return try_lock( (blocking_lock &)this ); }
static inline void   unlock   ( owner_lock & this ) { unlock  ( (blocking_lock &)this ); }
static inline size_t on_wait  ( owner_lock & this ) { return on_wait ( (blocking_lock &)this ); }
static inline void   on_wakeup( owner_lock & this, size_t v ) { on_wakeup ( (blocking_lock &)this, v ); }
static inline void   on_notify( owner_lock & this, struct thread$ * t ) { on_notify( (blocking_lock &)this, t ); }

struct fast_lock {
        thread$ * volatile owner;
        ThreadBenaphore sem;
};

static inline void ?{}(fast_lock & this) { this.owner = 0p; }

static inline bool $try_lock(fast_lock & this, thread$ * thrd) {
    thread$ * exp = 0p;
    return __atomic_compare_exchange_n(&this.owner, &exp, thrd, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED);
}

static inline void lock( fast_lock & this ) __attribute__((artificial));
static inline void lock( fast_lock & this ) {
        thread$ * thrd = active_thread();
        /* paranoid */verify(thrd != this.owner);

        for (;;) {
                if ($try_lock(this, thrd)) return;
                P(this.sem);
        }
}

static inline bool try_lock( fast_lock & this ) __attribute__((artificial));
static inline bool try_lock ( fast_lock & this ) {
        thread$ * thrd = active_thread();
        /* paranoid */ verify(thrd != this.owner);
        return $try_lock(this, thrd);
}

static inline thread$ * unlock( fast_lock & this ) __attribute__((artificial));
static inline thread$ * unlock( fast_lock & this ) {
        /* paranoid */ verify(active_thread() == this.owner);

        // open 'owner' before unlocking anyone
        // so new and unlocked threads don't park incorrectly.
        // This may require additional fencing on ARM.
        this.owner = 0p;

        return V(this.sem);
}

static inline size_t on_wait( fast_lock & this ) { unlock(this); return 0; }
static inline void on_wakeup( fast_lock & this, size_t ) { lock(this); }
static inline void on_notify( fast_lock &, struct thread$ * t ) { unpark(t); }

struct mcs_node {
        mcs_node * volatile next;
        single_sem sem;
};

static inline void ?{}(mcs_node & this) { this.next = 0p; }

static inline mcs_node * volatile & ?`next ( mcs_node * node ) {
        return node->next;
}

struct mcs_lock {
        mcs_queue(mcs_node) queue;
};

static inline void lock(mcs_lock & l, mcs_node & n) {
        if(push(l.queue, &n))
                wait(n.sem);
}

static inline void unlock(mcs_lock & l, mcs_node & n) {
        mcs_node * next = advance(l.queue, &n);
        if(next) post(next->sem);
}

struct linear_backoff_then_block_lock {
        // Spin lock used for mutual exclusion
        __spinlock_t spinlock;

        // Current thread owning the lock
        struct thread$ * owner;

        // List of blocked threads
        dlist( thread$ ) blocked_threads;

        // Used for comparing and exchanging
        volatile size_t lock_value;

        // used for linear backoff spinning
        int spin_start;
        int spin_end;
        int spin_count;

        // after unsuccessful linear backoff yield this many times
        int yield_count;
};

static inline void  ?{}( linear_backoff_then_block_lock & this, int spin_start, int spin_end, int spin_count, int yield_count ) {
        this.spinlock{};
        this.blocked_threads{};
        this.lock_value = 0;
        this.spin_start = spin_start;
        this.spin_end = spin_end;
        this.spin_count = spin_count;
        this.yield_count = yield_count;
}
static inline void  ?{}( linear_backoff_then_block_lock & this ) { this{4, 1024, 16, 0}; }
static inline void ^?{}( linear_backoff_then_block_lock & this ) {}
static inline void ?{}( linear_backoff_then_block_lock & this, linear_backoff_then_block_lock this2 ) = void;
static inline void ?=?( linear_backoff_then_block_lock & this, linear_backoff_then_block_lock this2 ) = void;

static inline bool internal_try_lock(linear_backoff_then_block_lock & this, size_t & compare_val) with(this) {
        if (__atomic_compare_exchange_n(&lock_value, &compare_val, 1, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)) {
                owner = active_thread();
                return true;
        }
        return false;
}

static inline bool try_lock(linear_backoff_then_block_lock & this) { size_t compare_val = 0; return internal_try_lock(this, compare_val); }

static inline bool try_lock_contention(linear_backoff_then_block_lock & this) with(this) {
        if (__atomic_exchange_n(&lock_value, 2, __ATOMIC_ACQUIRE) == 0) {
                owner = active_thread();
                return true;
        }
        return false;
}

static inline bool block(linear_backoff_then_block_lock & this) with(this) {
        lock( spinlock __cfaabi_dbg_ctx2 );
        if (lock_value != 2) {
                unlock( spinlock );
                return true;
        }
        insert_last( blocked_threads, *active_thread() );
        unlock( spinlock );
        park( );
        return true;
}

static inline bool lock(linear_backoff_then_block_lock & this) with(this) {
        // if owner just return
        if (active_thread() == owner) return true;
        size_t compare_val = 0;
        int spin = spin_start;
        // linear backoff
        for( ;; ) {
                compare_val = 0;
                if (internal_try_lock(this, compare_val)) return true;
                if (2 == compare_val) break;
                for (int i = 0; i < spin; i++) Pause();
                if (spin >= spin_end) break;
                spin += spin;
        }

        // linear backoff bounded by spin_count
        spin = spin_start;
        int spin_counter = 0;
        int yield_counter = 0;
        for ( ;; ) {
                if(try_lock_contention(this)) return true;
                if(spin_counter < spin_count) {
                        for (int i = 0; i < spin; i++) Pause();
                        if (spin < spin_end) spin += spin;
                        else spin_counter++;
                } else if (yield_counter < yield_count) {
                        // after linear backoff yield yield_count times
                        yield_counter++;
                        yield();
                } else { break; }
        }

        // block until signalled
        while (block(this)) if(try_lock_contention(this)) return true;

        // this should never be reached as block(this) always returns true
        return false;
}

static inline bool lock_improved(linear_backoff_then_block_lock & this) with(this) {
        // if owner just return
        if (active_thread() == owner) return true;
        size_t compare_val = 0;
        int spin = spin_start;
        // linear backoff
        for( ;; ) {
                compare_val = 0;
                if (internal_try_lock(this, compare_val)) return true;
                if (2 == compare_val) break;
                for (int i = 0; i < spin; i++) Pause();
                if (spin >= spin_end) break;
                spin += spin;
        }

        // linear backoff bounded by spin_count
        spin = spin_start;
        int spin_counter = 0;
        int yield_counter = 0;
        for ( ;; ) {
                compare_val = 0;
                if(internal_try_lock(this, compare_val)) return true;
                if (2 == compare_val) break;
                if(spin_counter < spin_count) {
                        for (int i = 0; i < spin; i++) Pause();
                        if (spin < spin_end) spin += spin;
                        else spin_counter++;
                } else if (yield_counter < yield_count) {
                        // after linear backoff yield yield_count times
                        yield_counter++;
                        yield();
                } else { break; }
        }

        if(2 != compare_val && try_lock_contention(this)) return true;
        // block until signalled
        while (block(this)) if(try_lock_contention(this)) return true;

        // this should never be reached as block(this) always returns true
        return false;
}

static inline void unlock(linear_backoff_then_block_lock & this) with(this) {
        verify(lock_value > 0);
    owner = 0p;
    if (__atomic_exchange_n(&lock_value, 0, __ATOMIC_RELEASE) == 1) return;
        lock( spinlock __cfaabi_dbg_ctx2 );
        thread$ * t = &try_pop_front( blocked_threads );
        unlock( spinlock );
        unpark( t );
}

static inline void on_notify(linear_backoff_then_block_lock & this, struct thread$ * t ) { unpark(t); }
static inline size_t on_wait(linear_backoff_then_block_lock & this) { unlock(this); return 0; }
static inline void on_wakeup(linear_backoff_then_block_lock & this, size_t recursion ) { lock_improved(this); }

//-----------------------------------------------------------------------------
// is_blocking_lock
trait is_blocking_lock(L & | sized(L)) {
        // For synchronization locks to use when acquiring
        void on_notify( L &, struct thread$ * );

        // For synchronization locks to use when releasing
        size_t on_wait( L & );

        // to set recursion count after getting signalled;
        void on_wakeup( L &, size_t recursion );
};

//-----------------------------------------------------------------------------
// // info_thread
// // the info thread is a wrapper around a thread used
// // to store extra data for use in the condition variable
forall(L & | is_blocking_lock(L)) {
        struct info_thread;

        // // for use by sequence
        // info_thread(L) *& Back( info_thread(L) * this );
        // info_thread(L) *& Next( info_thread(L) * this );
}

//-----------------------------------------------------------------------------
// Synchronization Locks
forall(L & | is_blocking_lock(L)) {
        struct condition_variable {
                // Spin lock used for mutual exclusion
                __spinlock_t lock;

                // List of blocked threads
                dlist( info_thread(L) ) blocked_threads;

                // Count of current blocked threads
                int count;
        };


        void  ?{}( condition_variable(L) & this );
        void ^?{}( condition_variable(L) & this );

        bool notify_one( condition_variable(L) & this );
        bool notify_all( condition_variable(L) & this );

        uintptr_t front( condition_variable(L) & this );

        bool empty  ( condition_variable(L) & this );
        int  counter( condition_variable(L) & this );

        void wait( condition_variable(L) & this );
        void wait( condition_variable(L) & this, uintptr_t info );
        bool wait( condition_variable(L) & this, Duration duration );
        bool wait( condition_variable(L) & this, uintptr_t info, Duration duration );

        void wait( condition_variable(L) & this, L & l );
        void wait( condition_variable(L) & this, L & l, uintptr_t info );
        bool wait( condition_variable(L) & this, L & l, Duration duration );
        bool wait( condition_variable(L) & this, L & l, uintptr_t info, Duration duration );
}