source: libcfa/src/concurrency/kernel/cluster.cfa @ 9d5eacb

//
// Cforall Version 1.0.0 Copyright (C) 2022 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// cluster.cfa -- file that includes helpers for subsystem that need cluster wide support
//
// Author           : Thierry Delisle
// Created On       : Fri Mar 11 12:39:24 2022
// Last Modified By :
// Last Modified On :
// Update Count     :
//

#define __cforall_thread__

#include "bits/defs.hfa"
#include "device/cpu.hfa"
#include "kernel/cluster.hfa"
#include "kernel/private.hfa"

#include "stdlib.hfa"
#include "limits.hfa"
#include "math.hfa"

#include "ready_subqueue.hfa"
#include "io/types.hfa"

#include <errno.h>
#include <unistd.h>

extern "C" {
        #include <sys/syscall.h>  // __NR_xxx
}

// No overriden function, no environment variable, no define
// fall back to a magic number
#ifndef __CFA_MAX_PROCESSORS__
        #define __CFA_MAX_PROCESSORS__ 1024
#endif

#if !defined(__CFA_NO_STATISTICS__)
        #define __STATS(...) __VA_ARGS__
#else
        #define __STATS(...)
#endif

// returns the maximum number of processors the RWLock support
__attribute__((weak)) unsigned __max_processors() libcfa_public {
        const char * max_cores_s = getenv("CFA_MAX_PROCESSORS");
        if(!max_cores_s) {
                __cfadbg_print_nolock(ready_queue, "No CFA_MAX_PROCESSORS in ENV\n");
                return __CFA_MAX_PROCESSORS__;
        }

        char * endptr = 0p;
        long int max_cores_l = strtol(max_cores_s, &endptr, 10);
        if(max_cores_l < 1 || max_cores_l > 65535) {
                __cfadbg_print_nolock(ready_queue, "CFA_MAX_PROCESSORS out of range : %ld\n", max_cores_l);
                return __CFA_MAX_PROCESSORS__;
        }
        if('\0' != *endptr) {
                __cfadbg_print_nolock(ready_queue, "CFA_MAX_PROCESSORS not a decimal number : %s\n", max_cores_s);
                return __CFA_MAX_PROCESSORS__;
        }

        return max_cores_l;
}

//=======================================================================
// Cluster wide reader-writer lock
//=======================================================================
void  ?{}(__scheduler_RWLock_t & this) {
        this.lock.max   = __max_processors();
        this.lock.alloc = 0;
        this.lock.ready = 0;
        this.lock.data  = alloc(this.lock.max);
        this.lock.write_lock  = false;

        /*paranoid*/ verify(__atomic_is_lock_free(sizeof(this.lock.alloc), &this.lock.alloc));
        /*paranoid*/ verify(__atomic_is_lock_free(sizeof(this.lock.ready), &this.lock.ready));

}
void ^?{}(__scheduler_RWLock_t & this) {
        free(this.lock.data);
}


//=======================================================================
// Lock-Free registering/unregistering of threads
unsigned register_proc_id( void ) with(__scheduler_lock.lock) {
        bool * handle = (bool *)&kernelTLS().sched_lock;

        // Step - 1 : check if there is already space in the data
        uint_fast32_t s = ready;

        // Check among all the ready
        for(uint_fast32_t i = 0; i < s; i++) {
                bool * volatile * cell = (bool * volatile *)&data[i]; // Cforall is bugged and the double volatiles causes problems
                /* paranoid */ verify( handle != *cell );

                bool * null = 0p; // Re-write every loop since compare thrashes it
                if( __atomic_load_n(cell, (int)__ATOMIC_RELAXED) == null
                        && __atomic_compare_exchange_n( cell, &null, handle, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) {
                        /* paranoid */ verify(i < ready);
                        /* paranoid */ verify( (kernelTLS().sched_id = i, true) );
                        return i;
                }
        }

        if(max <= alloc) abort("Trying to create more than %ud processors", __scheduler_lock.lock.max);

        // Step - 2 : F&A to get a new spot in the array.
        uint_fast32_t n = __atomic_fetch_add(&alloc, 1, __ATOMIC_SEQ_CST);
        if(max <= n) abort("Trying to create more than %ud processors", __scheduler_lock.lock.max);

        // Step - 3 : Mark space as used and then publish it.
        data[n] = handle;
        while() {
                unsigned copy = n;
                if( __atomic_load_n(&ready, __ATOMIC_RELAXED) == n
                        && __atomic_compare_exchange_n(&ready, &copy, n + 1, true, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
                        break;
                Pause();
        }

        // Return new spot.
        /* paranoid */ verify(n < ready);
        /* paranoid */ verify( (kernelTLS().sched_id = n, true) );
        return n;
}

void unregister_proc_id( unsigned id ) with(__scheduler_lock.lock) {
        /* paranoid */ verify(id < ready);
        /* paranoid */ verify(id == kernelTLS().sched_id);
        /* paranoid */ verify(data[id] == &kernelTLS().sched_lock);

        bool * volatile * cell = (bool * volatile *)&data[id]; // Cforall is bugged and the double volatiles causes problems

        __atomic_store_n(cell, 0p, __ATOMIC_RELEASE);
}

//-----------------------------------------------------------------------
// Writer side : acquire when changing the ready queue, e.g. adding more
//  queues or removing them.
uint_fast32_t ready_mutate_lock( void ) with(__scheduler_lock.lock) {
        /* paranoid */ verify( ! __preemption_enabled() );

        // Step 1 : lock global lock
        // It is needed to avoid processors that register mid Critical-Section
        //   to simply lock their own lock and enter.
        __atomic_acquire( &write_lock );

        // Make sure we won't deadlock ourself
        // Checking before acquiring the writer lock isn't safe
        // because someone else could have locked us.
        /* paranoid */ verify( ! kernelTLS().sched_lock );

        // Step 2 : lock per-proc lock
        // Processors that are currently being registered aren't counted
        //   but can't be in read_lock or in the critical section.
        // All other processors are counted
        uint_fast32_t s = ready;
        for(uint_fast32_t i = 0; i < s; i++) {
                volatile bool * llock = data[i];
                if(llock) __atomic_acquire( llock );
        }

        /* paranoid */ verify( ! __preemption_enabled() );
        return s;
}

void ready_mutate_unlock( uint_fast32_t last_s ) with(__scheduler_lock.lock) {
        /* paranoid */ verify( ! __preemption_enabled() );

        // Step 1 : release local locks
        // This must be done while the global lock is held to avoid
        //   threads that where created mid critical section
        //   to race to lock their local locks and have the writer
        //   immidiately unlock them
        // Alternative solution : return s in write_lock and pass it to write_unlock
        for(uint_fast32_t i = 0; i < last_s; i++) {
                volatile bool * llock = data[i];
                if(llock) __atomic_store_n(llock, (bool)false, __ATOMIC_RELEASE);
        }

        // Step 2 : release global lock
        /*paranoid*/ assert(true == write_lock);
        __atomic_store_n(&write_lock, (bool)false, __ATOMIC_RELEASE);

        /* paranoid */ verify( ! __preemption_enabled() );
}

//=======================================================================
// Cluster growth
static const unsigned __readyq_single_shard = 2;

void  ?{}(__timestamp_t & this) { this.t.tv = 0; this.t.ma = 0; }
void ^?{}(__timestamp_t &) {}

//-----------------------------------------------------------------------
// Check that all the intrusive queues in the data structure are still consistent
static void check_readyQ( cluster * cltr ) with (cltr->sched) {
        #if defined(__CFA_WITH_VERIFY__)
                {
                        const unsigned lanes_count = readyQ.count;
                        for( idx ; lanes_count ) {
                                __intrusive_lane_t & sl = readyQ.data[idx];
                                assert(!readyQ.data[idx].l.lock);

                                        if(is_empty(sl)) {
                                                assert( sl.l.anchor.next == 0p );
                                                assert( sl.l.anchor.ts   == MAX );
                                                assert( mock_head(sl)  == sl.l.prev );
                                        } else {
                                                assert( sl.l.anchor.next != 0p );
                                                assert( sl.l.anchor.ts   != MAX );
                                                assert( mock_head(sl)  != sl.l.prev );
                                        }
                        }
                }
        #endif
}

// Call this function of the intrusive list was moved using memcpy
// fixes the list so that the pointers back to anchors aren't left dangling
static inline void fix(__intrusive_lane_t & ll) {
        if(is_empty(ll)) {
                verify(ll.l.anchor.next == 0p);
                ll.l.prev = mock_head(ll);
        }
}

static void assign_list(unsigned & valrq, unsigned & valio, dlist(struct processor) & list, unsigned count) {
        struct processor * it = &list`first;
        for(unsigned i = 0; i < count; i++) {
                /* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count);
                it->rdq.id = valrq;
                it->rdq.target = UINT_MAX;
                valrq += __shard_factor.readyq;
                #if defined(CFA_HAVE_LINUX_IO_URING_H)
                        it->io.ctx->cq.id = valio;
                        it->io.target = UINT_MAX;
                        valio += __shard_factor.io;
                #endif
                it = &(*it)`next;
        }
}

static void reassign_cltr_id(struct cluster * cltr) {
        unsigned prefrq = 0;
        unsigned prefio = 0;
        assign_list(prefrq, prefio, cltr->procs.actives, cltr->procs.total - cltr->procs.idle);
        assign_list(prefrq, prefio, cltr->procs.idles  , cltr->procs.idle );
}

#if defined(CFA_HAVE_LINUX_IO_URING_H)
        static void assign_io(io_context$ ** data, size_t count, dlist(struct processor) & list) {
                struct processor * it = &list`first;
                while(it) {
                        /* paranoid */ verifyf( it, "Unexpected null iterator\n");
                        /* paranoid */ verifyf( it->io.ctx->cq.id < count, "Processor %p has id %u above count %zu\n", it, it->rdq.id, count);
                        data[it->io.ctx->cq.id] = it->io.ctx;
                        it = &(*it)`next;
                }
        }

        static void reassign_cltr_io(struct cluster * cltr) {
                assign_io(cltr->sched.io.data, cltr->sched.io.count, cltr->procs.actives);
                assign_io(cltr->sched.io.data, cltr->sched.io.count, cltr->procs.idles  );
        }
#else
        static void reassign_cltr_io(struct cluster *) {}
#endif

static void fix_times( __timestamp_t * volatile & tscs, unsigned count ) {
        tscs = alloc(count, tscs`realloc);
        for(i; count) {
                tscs[i].t.tv = rdtscl();
                tscs[i].t.ma = 0;
        }
}

// Grow the ready queue
void ready_queue_grow(struct cluster * cltr) {
        int target = cltr->procs.total;

        /* paranoid */ verify( ready_mutate_islocked() );
        __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue\n");

        // Make sure that everything is consistent
        /* paranoid */ check_readyQ( cltr );


        // Find new count
        // Make sure we always have atleast 1 list
        size_t ocount = cltr->sched.readyQ.count;
        size_t ncount = max(target * __shard_factor.readyq, __readyq_single_shard);

        // Do we have to do anything?
        if( ocount != ncount ) {

                // grow the ready queue
                with( cltr->sched ) {

                        // Allocate new array (uses realloc and memcpies the data)
                        readyQ.data = alloc( ncount, readyQ.data`realloc );

                        // Fix the moved data
                        for( idx; ocount ) {
                                fix(readyQ.data[idx]);
                        }

                        // Construct new data
                        for( idx; ocount ~ ncount) {
                                (readyQ.data[idx]){};
                        }

                        // Update original count
                        readyQ.count = ncount;
                }


                fix_times(cltr->sched.readyQ.tscs, cltr->sched.readyQ.count);
        }

        // Fix the io times
        cltr->sched.io.count = target * __shard_factor.io;
        fix_times(cltr->sched.io.tscs, cltr->sched.io.count);

        // realloc the caches
        cltr->sched.caches = alloc( target, cltr->sched.caches`realloc );

        // reassign the clusters.
        reassign_cltr_id(cltr);

        cltr->sched.io.data = alloc( cltr->sched.io.count, cltr->sched.io.data`realloc );
        reassign_cltr_io(cltr);

        // Make sure that everything is consistent
        /* paranoid */ check_readyQ( cltr );
//      /* paranoid */ verify( (target == 0) == (cltr->sched.caches == 0p) );

        __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue done\n");

        /* paranoid */ verify( ready_mutate_islocked() );
}

// Shrink the ready queue
void ready_queue_shrink(struct cluster * cltr) {
        /* paranoid */ verify( ready_mutate_islocked() );
        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue\n");

        // Make sure that everything is consistent
        /* paranoid */ check_readyQ( cltr );

        int target = cltr->procs.total;

        with( cltr->sched ) {
                // Remember old count
                size_t ocount = readyQ.count;

                // Find new count
                // Make sure we always have atleast 1 list
                size_t ncount = max(target * __shard_factor.readyq, __readyq_single_shard);
                /* paranoid */ verifyf( ocount >= ncount, "Error in shrinking size calculation, %zu >= %zu", ocount, ncount );
                /* paranoid */ verifyf( ncount == target * __shard_factor.readyq || ncount == __readyq_single_shard,
                /* paranoid */          "Error in shrinking size calculation, expected %u or %u, got %zu", target * __shard_factor.readyq, __readyq_single_shard, ncount );

                readyQ.count = ncount;

                // for printing count the number of displaced threads
                #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
                        __attribute__((unused)) size_t displaced = 0;
                #endif

                // redistribute old data
                for( idx; ncount ~ ocount) {
                        // Lock is not strictly needed but makes checking invariants much easier
                        __attribute__((unused)) bool locked = __atomic_try_acquire(&readyQ.data[idx].l.lock);
                        verify(locked);

                        // As long as we can pop from this lane to push the threads somewhere else in the queue
                        while(!is_empty(readyQ.data[idx])) {
                                struct thread$ * thrd;
                                unsigned long long _;
                                [thrd, _] = pop(readyQ.data[idx]);

                                push(cltr, thrd, true);

                                // for printing count the number of displaced threads
                                #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
                                        displaced++;
                                #endif
                        }

                        // Unlock the lane
                        __atomic_unlock(&readyQ.data[idx].l.lock);

                        // TODO print the queue statistics here

                        ^(readyQ.data[idx]){};
                }

                __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue displaced %zu threads\n", displaced);

                // Allocate new array (uses realloc and memcpies the data)
                readyQ.data = alloc( ncount, readyQ.data`realloc );

                // Fix the moved data
                for( idx; ncount ) {
                        fix(readyQ.data[idx]);
                }

                fix_times(readyQ.tscs, ncount);
        }

        cltr->sched.caches = alloc( target, cltr->sched.caches`realloc );

        // Fix the io times
        cltr->sched.io.count = target * __shard_factor.io;
        fix_times(cltr->sched.io.tscs, cltr->sched.io.count);

        reassign_cltr_id(cltr);

        cltr->sched.io.data = alloc( cltr->sched.io.count, cltr->sched.io.data`realloc );
        reassign_cltr_io(cltr);

        // Make sure that everything is consistent
//      /* paranoid */ verify( (target == 0) == (cltr->sched.caches == 0p) );
        /* paranoid */ check_readyQ( cltr );

        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue done\n");
        /* paranoid */ verify( ready_mutate_islocked() );
}

void ready_queue_close(struct cluster * cltr) {
        free( cltr->sched.readyQ.data );
        free( cltr->sched.readyQ.tscs );
        cltr->sched.readyQ.data = 0p;
        cltr->sched.readyQ.tscs = 0p;
        cltr->sched.readyQ.count = 0;

        free( cltr->sched.io.tscs );
        free( cltr->sched.caches );
}

#define nested_offsetof(type, field) ((off_t)(&(((type*)0)-> field)))

// Ctor
void ?{}( __intrusive_lane_t & this ) {
        this.l.lock = false;
        this.l.prev = mock_head(this);
        this.l.anchor.next = 0p;
        this.l.anchor.ts   = MAX;
        #if !defined(__CFA_NO_STATISTICS__)
                this.l.cnt  = 0;
        #endif

        // We add a boat-load of assertions here because the anchor code is very fragile
        /* paranoid */ _Static_assert( offsetof( thread$, rdy_link ) == nested_offsetof(__intrusive_lane_t, l.anchor) );
        /* paranoid */ verify( offsetof( thread$, rdy_link ) == nested_offsetof(__intrusive_lane_t, l.anchor) );
        /* paranoid */ verify( ((uintptr_t)( mock_head(this) ) + offsetof( thread$, rdy_link )) == (uintptr_t)(&this.l.anchor) );
        /* paranoid */ verify( &mock_head(this)->rdy_link.next == &this.l.anchor.next );
        /* paranoid */ verify( &mock_head(this)->rdy_link.ts   == &this.l.anchor.ts   );
        /* paranoid */ verify( mock_head(this)->rdy_link.next == 0p );
        /* paranoid */ verify( mock_head(this)->rdy_link.ts   == MAX );
        /* paranoid */ verify( mock_head(this) == this.l.prev );
        /* paranoid */ verify( __alignof__(__intrusive_lane_t) == 64 );
        /* paranoid */ verify( __alignof__(this) == 64 );
        /* paranoid */ verifyf( ((intptr_t)(&this) % 64) == 0, "Expected address to be aligned %p %% 64 == %zd", &this, ((intptr_t)(&this) % 64) );
}

#undef nested_offsetof

// Dtor is trivial
void ^?{}( __intrusive_lane_t & this ) {
        // Make sure the list is empty
        /* paranoid */ verify( this.l.anchor.next == 0p );
        /* paranoid */ verify( this.l.anchor.ts   == MAX );
        /* paranoid */ verify( mock_head(this)    == this.l.prev );
}