source: trunk/kernel/kern/cluster.c @ 409

/*
 * cluster.c - Cluster-Manager related operations
 *
 * Author  Ghassan Almaless (2008,2009,2010,2011,2012)
 *         Mohamed Lamine Karaoui (2015)
 *         Alain Greiner (2016,2017)
 *
 * Copyright (c) UPMC Sorbonne Universites
 *
 * This file is part of ALMOS-MKH..
 *
 * ALMOS-MKH. is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2.0 of the License.
 *
 * ALMOS-MKH. is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with ALMOS-MKH.; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <kernel_config.h>
#include <hal_types.h>
#include <hal_atomic.h>
#include <hal_special.h>
#include <hal_ppm.h>
#include <remote_fifo.h>
#include <printk.h>
#include <errno.h>
#include <spinlock.h>
#include <core.h>
#include <scheduler.h>
#include <list.h>
#include <cluster.h>
#include <boot_info.h>
#include <bits.h>
#include <ppm.h>
#include <thread.h>
#include <kmem.h>
#include <process.h>
#include <dqdt.h>

/////////////////////////////////////////////////////////////////////////////////////
// Extern global variables
/////////////////////////////////////////////////////////////////////////////////////

extern process_t process_zero;     // allocated in kernel_init.c file


/////////////////////////////////////////////////
error_t cluster_init( struct boot_info_s * info )
{
    error_t     error;
    lpid_t      lpid;     // local process_index
    lid_t       lid;      // local core index

        cluster_t * cluster = LOCAL_CLUSTER;

    // initialize cluster global parameters
        cluster->paddr_width     = info->paddr_width;
        cluster->x_width         = info->x_width;
        cluster->y_width         = info->y_width;
        cluster->x_size          = info->x_size;
        cluster->y_size          = info->y_size;
        cluster->io_cxy          = info->io_cxy;

    // initialize cluster local parameters
        cluster->cores_nr        = info->cores_nr;

    // initialize the lock protecting the embedded kcm allocator
        spinlock_init( &cluster->kcm_lock );

cluster_dmsg("\n[DBG] %s for cluster %x enters\n",
__FUNCTION__ , local_cxy );

    // initialises DQDT
    cluster->dqdt_root_level = dqdt_init( info->x_size,
                                          info->y_size,
                                          info->y_width );
    cluster->threads_var = 0;
    cluster->pages_var   = 0;

    // initialises embedded PPM
        error = hal_ppm_init( info );

    if( error )
    {
        printk("\n[ERROR] in %s : cannot initialize PPM in cluster %x\n",
               __FUNCTION__ , local_cxy );
        return ENOMEM;
    }

cluster_dmsg("\n[DBG] %s : PPM initialized in cluster %x at cycle %d\n",
__FUNCTION__ , local_cxy , hal_get_cycles() );

    // initialises embedded KHM
        khm_init( &cluster->khm );

    cluster_dmsg("\n[DBG] %s : KHM initialized in cluster %x at cycle %d\n",
                 __FUNCTION__ , local_cxy , hal_get_cycles() );

    // initialises embedded KCM
        kcm_init( &cluster->kcm , KMEM_KCM );

    cluster_dmsg("\n[DBG] %s : KCM initialized in cluster %x at cycle %d\n",
                 __FUNCTION__ , local_cxy , hal_get_cycles() );

    // initialises all cores descriptors 
        for( lid = 0 ; lid < cluster->cores_nr; lid++ )
        {
                core_init( &cluster->core_tbl[lid],    // target core descriptor
                       lid,                        // local core index
                       info->core[lid].gid );      // gid from boot_info_t
        }

cluster_dmsg("\n[DBG] %s : cores initialized in cluster %x at cycle %d\n",
__FUNCTION__ , local_cxy , hal_get_cycles() );

    // initialises RPC fifo
        local_fifo_init( &cluster->rpc_fifo );
    cluster->rpc_threads = 0;

cluster_dmsg("\n[DBG] %s : RPC fifo inialized in cluster %x at cycle %d\n",
__FUNCTION__ , local_cxy , hal_get_cycles() );

    // initialise pref_tbl[] in process manager
        spinlock_init( &cluster->pmgr.pref_lock );
    cluster->pmgr.pref_nr = 0;
    cluster->pmgr.pref_tbl[0] = XPTR( local_cxy , &process_zero );
    for( lpid = 1 ; lpid < CONFIG_MAX_PROCESS_PER_CLUSTER ; lpid++ )
    {
        cluster->pmgr.pref_tbl[lpid] = XPTR_NULL;
    }

    // initialise local_list in process manager
        remote_spinlock_init( XPTR( local_cxy , &cluster->pmgr.local_lock ) );
    xlist_root_init( XPTR( local_cxy , &cluster->pmgr.local_root ) );
    cluster->pmgr.local_nr = 0;

    // initialise copies_lists in process manager
    for( lpid = 0 ; lpid < CONFIG_MAX_PROCESS_PER_CLUSTER ; lpid++ )
    {
            remote_spinlock_init( XPTR( local_cxy , &cluster->pmgr.copies_lock[lpid] ) );
        cluster->pmgr.copies_nr[lpid] = 0;
        xlist_root_init( XPTR( local_cxy , &cluster->pmgr.copies_root[lpid] ) );
    }

cluster_dmsg("\n[DBG] %s Process Manager initialized in cluster %x at cycle %d\n",
__FUNCTION__ , local_cxy , hal_get_cycles() );

    hal_fence();

        return 0;
} // end cluster_init()

////////////////////////////////////////
bool_t cluster_is_undefined( cxy_t cxy )
{
    cluster_t * cluster = LOCAL_CLUSTER;

    uint32_t y_width = cluster->y_width;

    uint32_t x = cxy >> y_width;
    uint32_t y = cxy & ((1<<y_width)-1);

    if( x >= cluster->x_size ) return true;
    if( y >= cluster->y_size ) return true;

    return false;
}

////////////////////////////////////////////////////////////////////////////////////
//  Cores related functions
////////////////////////////////////////////////////////////////////////////////////

/////////////////////////////////
lid_t cluster_select_local_core()
{
    uint32_t min = 100;
    lid_t    sel = 0;
    lid_t    lid;

    cluster_t * cluster = LOCAL_CLUSTER;

    for( lid = 0 ; lid < cluster->cores_nr ; lid++ )
    {
        if( cluster->core_tbl[lid].usage < min )
        {
            min = cluster->core_tbl[lid].usage;
            sel = lid;
        }
    }
    return sel;
}

////////////////////////////////////////////////////////////////////////////////////
//  Process management related functions
////////////////////////////////////////////////////////////////////////////////////

//////////////////////////////////////////////////////////
xptr_t cluster_get_reference_process_from_pid( pid_t pid )
{
    xptr_t ref_xp;   // extended pointer on reference process descriptor

    cluster_t * cluster = LOCAL_CLUSTER;

    // get owner cluster and lpid
    cxy_t  owner_cxy = CXY_FROM_PID( pid );
    lpid_t lpid      = LPID_FROM_PID( pid );

    // Check valid PID
    if( lpid >= CONFIG_MAX_PROCESS_PER_CLUSTER )  return XPTR_NULL;

    if( local_cxy == owner_cxy )   // local cluster is owner cluster
    {
        ref_xp = cluster->pmgr.pref_tbl[lpid];
    }
    else                              // use a remote_lwd to access owner cluster
    {
        ref_xp = (xptr_t)hal_remote_lwd( XPTR( owner_cxy , &cluster->pmgr.pref_tbl[lpid] ) );
    }

    return ref_xp;
}

////////////////////////////////////////////////
error_t cluster_pid_alloc( xptr_t    process_xp,
                           pid_t   * pid )
{
    error_t     error;
    lpid_t      lpid;
    bool_t      found;

    pmgr_t    * pm         = &LOCAL_CLUSTER->pmgr;

    // get the process manager lock
    spinlock_lock( &pm->pref_lock );

    // search an empty slot
    found = false;
    for( lpid = 0 ; lpid < CONFIG_MAX_PROCESS_PER_CLUSTER ; lpid++ )
    {
        if( pm->pref_tbl[lpid] == XPTR_NULL )
        {
            found = true;
            break;
        }
    }

    if( found )
    {
        // register process in pref_tbl[]
        pm->pref_tbl[lpid] = process_xp;
        pm->pref_nr++;

        // returns pid
        *pid = PID( local_cxy , lpid );

        error = 0;
    }
    else
    {
        error = EAGAIN;
    }

    // release the processs_manager lock
    spinlock_unlock( &pm->pref_lock );

    return error;

} // end cluster_pid_alloc()

/////////////////////////////////////
void cluster_pid_release( pid_t pid )
{
    cxy_t  owner_cxy  = CXY_FROM_PID( pid );
    lpid_t lpid       = LPID_FROM_PID( pid );

    pmgr_t  * pm = &LOCAL_CLUSTER->pmgr;

    // check pid argument
    assert( (lpid < CONFIG_MAX_PROCESS_PER_CLUSTER) && (owner_cxy == local_cxy) ,
    __FUNCTION__ , "illegal PID" );

    // check number of copies
    assert( (pm->copies_nr[lpid] == 0) ,
    __FUNCTION__ , "number of copies must be 0" );  

    // get the process manager lock
    spinlock_lock( &pm->pref_lock );

    // remove process from pref_tbl[]
    pm->pref_tbl[lpid] = XPTR_NULL;
    pm->pref_nr--;

    // release the processs_manager lock
    spinlock_unlock( &pm->pref_lock );

} // end cluster_pid_release()

///////////////////////////////////////////////////////////
process_t * cluster_get_local_process_from_pid( pid_t pid )
{
    xptr_t         process_xp;
    process_t    * process_ptr;
    xptr_t         root_xp;
    xptr_t         iter_xp;
    bool_t         found;

    found   = false;
    root_xp = XPTR( local_cxy , &LOCAL_CLUSTER->pmgr.local_root );

    XLIST_FOREACH( root_xp , iter_xp )
    {
        process_xp  = XLIST_ELEMENT( iter_xp , process_t , local_list );
        process_ptr = (process_t *)GET_PTR( process_xp );
        if( process_ptr->pid == pid )
        {
            found = true;
            break;
        }
    }

    if (found ) return process_ptr;
    else        return NULL;

}  // end cluster_get_local_process_from_pid()

//////////////////////////////////////////////////////
void cluster_process_local_link( process_t * process )
{
    uint32_t irq_state;
    pmgr_t * pm = &LOCAL_CLUSTER->pmgr;

    // get lock protecting the process manager local list
    remote_spinlock_lock_busy( XPTR( local_cxy , &pm->local_lock ) , & irq_state );

    xlist_add_first( XPTR( local_cxy , &pm->local_root ),
                     XPTR( local_cxy , &process->local_list ) );
    pm->local_nr++;

    // release lock protecting the process manager local list
    remote_spinlock_unlock_busy( XPTR( local_cxy , &pm->local_lock ) , irq_state );
}

////////////////////////////////////////////////////////
void cluster_process_local_unlink( process_t * process )
{
    uint32_t irq_state;
    pmgr_t * pm = &LOCAL_CLUSTER->pmgr;

    // get lock protecting the process manager local list
    remote_spinlock_lock_busy( XPTR( local_cxy , &pm->local_lock ) , &irq_state );

    xlist_unlink( XPTR( local_cxy , &process->local_list ) );
    pm->local_nr--;

    // release lock protecting the process manager local list
    remote_spinlock_unlock_busy( XPTR( local_cxy , &pm->local_lock ) , irq_state );
}

///////////////////////////////////////////////////////
void cluster_process_copies_link( process_t * process )
{
    uint32_t irq_state;
    pmgr_t * pm = &LOCAL_CLUSTER->pmgr;

    // get owner cluster identifier CXY and process LPID
    pid_t    pid        = process->pid;
    cxy_t    owner_cxy  = CXY_FROM_PID( pid );
    lpid_t   lpid       = LPID_FROM_PID( pid );

    // get extended pointer on lock protecting copies_list[lpid]
    xptr_t copies_lock  = XPTR( owner_cxy , &pm->copies_lock[lpid] );

    // get extended pointer on the copies_list[lpid] root
    xptr_t copies_root  = XPTR( owner_cxy , &pm->copies_root[lpid] );

    // get extended pointer on the local copies_list entry
    xptr_t copies_entry = XPTR( local_cxy , &process->copies_list );

    // get lock protecting copies_list[lpid]
    remote_spinlock_lock_busy( copies_lock , &irq_state );

    xlist_add_first( copies_root , copies_entry );
    hal_remote_atomic_add( XPTR( owner_cxy , &pm->copies_nr[lpid] ) , 1 );

    // release lock protecting copies_list[lpid]
    remote_spinlock_unlock_busy( copies_lock , irq_state );
}

/////////////////////////////////////////////////////////
void cluster_process_copies_unlink( process_t * process )
{
    uint32_t irq_state;
    pmgr_t * pm = &LOCAL_CLUSTER->pmgr;

    // get owner cluster identifier CXY and process LPID
    pid_t    pid        = process->pid;
    cxy_t    owner_cxy  = CXY_FROM_PID( pid );
    lpid_t   lpid       = LPID_FROM_PID( pid );

    // get extended pointer on lock protecting copies_list[lpid]
    xptr_t copies_lock  = hal_remote_lwd( XPTR( owner_cxy , &pm->copies_lock[lpid] ) );

    // get extended pointer on the local copies_list entry
    xptr_t copies_entry = XPTR( local_cxy , &process->copies_list );

    // get lock protecting copies_list[lpid]
    remote_spinlock_lock_busy( copies_lock , &irq_state );

    xlist_unlink( copies_entry );
    hal_remote_atomic_add( XPTR( owner_cxy , &pm->copies_nr[lpid] ) , -1 );

    // release lock protecting copies_list[lpid]
    remote_spinlock_unlock_busy( copies_lock , irq_state );
}

////////////////////////////////////////////////////////////////////////////////////////
// TODO Il me semble que la seule chose que fait ce kernel thread à chaque réveil
// est de mettre à jour la DQDT, et de se rendormir... A-t-on besoin d'un thread ? [AG]
//////////////////////////////////////////////////////////////////////////////////////////

#if 0
void * cluster_manager_thread( void * arg )
{
        register struct dqdt_cluster_s * root;
        register struct cluster_s      * root_home;

        register uint32_t                tm_start;
        register uint32_t                tm_end;
        register uint32_t                cpu_id;
        struct cluster_s               * cluster;
        struct thread_s                * this;
        struct event_s                   event;
        struct alarm_info_s              info;
        register uint32_t                cntr;
        register bool_t                  isRootMgr;
        register uint32_t                period;

        cpu_enable_all_irq( NULL );

        cluster   = arg;
        this      = CURRENT_THREAD;
        cpu_id    = cpu_get_id();
        root      = dqdt_root;
        root_home = dqdt_root->home;
        isRootMgr = (cluster == root_home) ? true : false;
        cntr      = 0;
        period    = (isRootMgr) ?
                CONFIG_DQDT_ROOTMGR_PERIOD * MSEC_PER_TICK :
                CONFIG_DQDT_MGR_PERIOD * MSEC_PER_TICK;

        event_set_senderId(&event, this);
        event_set_priority(&event, E_CHR);
        event_set_handler(&event, &manager_alarm_event_handler);

        info.event = &event;
        thread_preempt_disable(CURRENT_THREAD);

    // infinite loop
        while(1)
        {
                tm_start = cpu_time_stamp();
                dqdt_update();
                tm_end   = cpu_time_stamp();

                if(isRootMgr)
                {
                        if((cntr % 10) == 0)
                        {
                                printk(INFO, "INFO: cpu %d, DQDT update ended [ %u - %u ]\n",
                                       cpu_id,
                                       tm_end,
                                       tm_end - tm_start);

                                dqdt_print_summary(root);
                        }
                }

                alarm_wait( &info , period );
                sched_sleep(this);
                cntr ++;
        }

        return NULL;
} // end cluster_manager_thread()

//////////////////////////////////////////
EVENT_HANDLER(manager_alarm_event_handler)
{
        struct thread_s *manager;

        manager = event_get_senderId(event);

        thread_preempt_disable(CURRENT_THREAD);

        //printk(INFO, "%s: cpu %d [%u]\n", __FUNCTION__, cpu_get_id(), cpu_time_stamp());

        sched_wakeup(manager);

        thread_preempt_enable(CURRENT_THREAD);

        return 0;
}

///////////////////////////////////////////////
EVENT_HANDLER(cluster_key_create_event_handler)
{
        struct cluster_s *cluster;
        struct thread_s *sender;
        ckey_t *ckey;
        uint32_t key;

        sender  = event_get_senderId(event);
        ckey    = event_get_argument(event);
        cluster = current_cluster;
        key     = cluster->next_key;

        while((key < CLUSTER_TOTAL_KEYS_NR) && (cluster->keys_tbl[key] != NULL))
                key ++;

        if(key < CLUSTER_TOTAL_KEYS_NR)
        {
                ckey->val = key;
                cluster->keys_tbl[key] = (void *) 0x1; // Reserved
                cluster->next_key = key;
                event_set_error(event, 0);
        }
        else
                event_set_error(event, ENOSPC);

        sched_wakeup(sender);
        return 0;
}

///////////////////////////////////////////////
EVENT_HANDLER(cluster_key_delete_event_handler)
{
        struct cluster_s *cluster;
        struct thread_s *sender;
        ckey_t *ckey;
        uint32_t key;

        sender  = event_get_senderId(event);
        ckey    = event_get_argument(event);
        cluster = current_cluster;
        key     = ckey->val;

        if(key < cluster->next_key)
                cluster->next_key = key;

        cluster->keys_tbl[key] = NULL;
        event_set_error(event, 0);

        sched_wakeup(sender);
        return 0;
}

#define _CKEY_CREATE  0x0
#define _CKEY_DELETE  0x1

error_t cluster_do_key_op(ckey_t *key, uint32_t op)
{
        struct event_s event;
        struct thread_s *this;
        struct cluster_s *cluster;
        struct cpu_s *cpu;

        this = CURRENT_THREAD;

        event_set_priority(&event, E_FUNC);
        event_set_senderId(&event, this);
        event_set_argument(&event, key);

        if(op == _CKEY_CREATE)
                event_set_handler(&event, cluster_key_create_event_handler);
        else
                event_set_handler(&event, cluster_key_delete_event_handler);

        cluster = current_cluster;
        cpu     = cluster->bscluster->bscpu;
        event_send(&event, &cpu->re_listner);

        sched_sleep(this);

        return event_get_error(&event);
}

error_t cluster_key_create(ckey_t *key)
{
        return cluster_do_key_op(key, _CKEY_CREATE);
}

error_t cluster_key_delete(ckey_t *key)
{
        return cluster_do_key_op(key, _CKEY_DELETE);
}

void* cluster_getspecific(ckey_t *key)
{
        struct cluster_s *cluster;

        cluster = current_cluster;
        return cluster->keys_tbl[key->val];
}

void  cluster_setspecific(ckey_t *key, void *val)
{
        struct cluster_s *cluster;

        cluster = current_cluster;
        cluster->keys_tbl[key->val] = val;
}
#endif