source: trunk/kernel/mm/ppm.c @ 8

/*
 * ppm.c - Per-cluster Physical Pages Manager implementation
 *
 * Authors  Ghassan Almaless (2008,2009,2010,2011,2012)
 *          Alain Greiner    (2016)
 *
 * 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 <almos_config.h>
#include <hal_types.h>
#include <hal_special.h>
#include <printk.h>
#include <list.h>
#include <bits.h>
#include <page.h>
#include <spinlock.h>
#include <thread.h>
#include <cluster.h>
#include <kmem.h>
#include <process.h>
#include <dqdt.h>
#include <ppm.h>



////////////////////////////////////////////////
inline bool_t ppm_page_is_valid( page_t * page )
{
    ppm_t    * ppm  = &LOCAL_CLUSTER->ppm;
        uint32_t   pgnr = (uint32_t)( page - ppm->pages_tbl );
        return (pgnr <= ppm->pages_nr);
}

////////////////////////////////////////////
inline void * ppm_page2base( page_t * page )
{
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;
        return (void*)((page - ppm->pages_tbl) << CONFIG_PPM_PAGE_SHIFT);
}

////////////////////////////////////////////
inline page_t * ppm_base2page( void * base ) 
{
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;
        return (ppm->pages_tbl + (((uint32_t)base ) >> CONFIG_PPM_PAGE_SHIFT));
}

//////////////////////////////////////////
inline ppn_t ppm_page2ppn( page_t * page )
{
        ppm_t  * ppm  = &LOCAL_CLUSTER->ppm;
        return (ppn_t)( page - ppm->pages_tbl );
}

/////////////////////////////////////////
inline page_t * ppm_ppn2page( ppn_t ppn )
{
        ppm_t  * ppm  = &LOCAL_CLUSTER->ppm;
        return &ppm->pages_tbl[ppn];
}

///////////////////////////////////////
inline void * ppm_ppn2base( ppn_t ppn )
{
        return (void*)( ppn << CONFIG_PPM_PAGE_SHIFT );
}

////////////////////////////////////////
inline ppn_t ppm_base2ppn( void * base )
{
        return (ppn_t)( (uint32_t)base >> CONFIG_PPM_PAGE_SHIFT );
}

//////////////////////////////////////////////////
static void ppm_free_pages_nolock( page_t * page )
{
        page_t   * buddy;            // searched buddy page descriptor
        uint32_t   buddy_index;      // buddy page index
        page_t   * current;          // current (merged) page descriptor
        uint32_t   current_index;    // current (merged) page index
        uint32_t   current_order;    // current (merget) page order

    ppm_t    * ppm         = &LOCAL_CLUSTER->ppm;
    page_t   * pages_tbl   = ppm->pages_tbl;

    // update released page descriptor flags
        page_set_flag( page , PG_FREE );

    // search the buddy page descriptor
    // - merge with current page descriptor if found
    // - exit to release the current page descriptor if not found 
    current       = page , 
    current_index = (uint32_t)(page - ppm->pages_tbl);
        for( current_order = page->order ; 
         current_order < CONFIG_PPM_MAX_ORDER ;
         current_order++ )
    {
                buddy_index = current_index ^ (1 << current_order);
                buddy       = pages_tbl + buddy_index;
    
                if( !page_is_flag( buddy , PG_FREE ) || (buddy->order != current_order) ) break;

        // remove buddy from free list 
                list_unlink( &buddy->list );
                ppm->free_pages_nr[current_order] --;
        ppm->total_free_pages -= (1 << current_order);
    
        // merge buddy with current
                buddy->order = 0;
                current_index &= buddy_index;
        }
  
    // update merged page descriptor order
        current        = pages_tbl + current_index;
        current->order = current_order;

    // insert current in free list
        list_add_first( &ppm->free_pages_root[current_order] , &current->list );
        ppm->free_pages_nr[current_order] ++;
    ppm->total_free_pages += (1 << current_order);

}  // end ppm_free_pages_nolock()

//////////////////////////////
void ppm_init( ppm_t    * ppm,
               uint32_t   pages_nr,        // total pages number
               uint32_t   pages_offset )   // occupied pages
{
        uint32_t   i;

    // set signature
        ppm->signature = PPM_SIGNATURE;

    // initialize lock protecting the free_pages[] array
        spinlock_init( &ppm->free_lock );

    // initialize free_pages[] array as empty
        ppm->total_free_pages = 0;
        for( i = 0 ; i < CONFIG_PPM_MAX_ORDER ; i++ )
        {
                list_root_init( &ppm->free_pages_root[i] );
                ppm->free_pages_nr[i] = 0;
        }

#if( CONFIG_PPM_DEBUG )
ppm_print( ppm , "after reset" );
#endif
  
    // initialize dirty_list as empty
    list_root_init( &ppm->dirty_root );

    // initialize pointer on page descriptors array 
        ppm->pages_tbl = (page_t*)( pages_offset << CONFIG_PPM_PAGE_SHIFT );

    // compute size of pages descriptor array rounded to an integer number of pages
    uint32_t bytes = ARROUND_UP( pages_nr * sizeof(page_t), CONFIG_PPM_PAGE_SIZE );

    // compute number of pages required to store page descriptor array
        uint32_t pages_array  = bytes >> CONFIG_PPM_PAGE_SHIFT;

    // compute total number of reserved pages (kernel code & pages_tbl[])
        uint32_t reserved_pages = pages_offset + pages_array;

        // set pages numbers 
        ppm->pages_nr      = pages_nr;
    ppm->pages_offset  = reserved_pages;

    // initialises all page descriptors in pages_tbl[]
        for( i = 0 ; i < pages_nr ; i++ )
    { 
        page_init( &ppm->pages_tbl[i] );

        // TODO optimisation : make only a partial init [AG]
        // complete the initialisation when page is allocated [AG]
        // ppm->pages_tbl[i].flags = 0;
    } 

    // - set PG_RESERVED flag for reserved pages (kernel code & pages_tbl[]) 
    // - release all other pages to populate the free lists
        for( i = 0 ; i < reserved_pages ; i++) 
    {
        page_set_flag( &ppm->pages_tbl[i] , PG_RESERVED );
    }
        for( i = reserved_pages ; i < pages_nr ; i++ )
        {
            ppm_free_pages_nolock( &ppm->pages_tbl[i] );

        // TODO optimisation : decompose this enormous set of small pages
        // to a set big pages with various order values
        }

    // check consistency
    ppm_assert_order( ppm );

    kinit_dmsg("\n[INFO] %s : done in cluster %x at cycle %d\n",
               __FUNCTION__ , local_cxy , hal_time_stamp() );

#if( CONFIG_PPM_DEBUG )
ppm_print( ppm , "after init" );
#endif
    
} // end ppm_init()

////////////////////////////////////////////
page_t * ppm_alloc_pages( uint32_t   order )
{
    uint32_t   current_order;
        page_t   * remaining_block;
        uint32_t   current_size;

    ppm_t    * ppm = &LOCAL_CLUSTER->ppm;

        assert( (ppm->signature == PPM_SIGNATURE) , __FUNCTION__ , "PPM non initialised" );

        assert( (order < CONFIG_PPM_MAX_ORDER) , __FUNCTION__ , "illegal order argument" );

        page_t * block = NULL;

    ppm_dmsg("\n[INFO] %s : enters / order = %d\n",
             __FUNCTION__ , order );

#if( CONFIG_PPM_DEBUG )
ppm_print( ppm , "before allocation" );
#endif

    // take lock protecting free lists
        spinlock_lock( &ppm->free_lock );

    // find a free block equal or larger to requested size
        for( current_order = order ; current_order < CONFIG_PPM_MAX_ORDER ; current_order ++ )
        {
                if( !list_is_empty( &ppm->free_pages_root[current_order] ) )
                {
                        block = LIST_FIRST( &ppm->free_pages_root[current_order] , page_t , list );
                        list_unlink( &block->list );
                        break;
                }
        }

        if( block == NULL ) // return failure
        {
        // release lock protecting free lists
            spinlock_unlock( &ppm->free_lock );

        return NULL;
    }
  
    // update free-lists after removing a block 
        ppm->total_free_pages -= (1 << current_order);
        ppm->free_pages_nr[current_order] --;  
        current_size = (1 << current_order);

    // split the removed block in smaller sub-blocks if required
    // and update the free-lists accordingly
        while( current_order > order )
        {
                current_order --;
                current_size >>= 1;
    
                remaining_block = block + current_size;
                remaining_block->order = current_order;

                list_add_first( &ppm->free_pages_root[current_order] , &remaining_block->list );
                ppm->free_pages_nr[current_order] ++;
        ppm->total_free_pages += (1 << current_order);
        }
  
    // update page descriptor
    page_clear_flag( block , PG_FREE );
        page_refcount_up( block );
        block->order = order;

    // release lock protecting free lists
        spinlock_unlock( &ppm->free_lock );
  
    ppm_dmsg("\n[INFO] %s : base = %x / order = %d\n",
             __FUNCTION__ , (uint32_t)ppm_page2base( block ) , order );

#if CONFIG_PPM_DEBUG
ppm_print( ppm , "after allocation" );
#endif

        return block;
}  // end pmm_alloc-pages()


////////////////////////////////////
void ppm_free_pages( page_t * page )
{
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;
  
    // get lock protecting free_pages[] array
        spinlock_lock( &ppm->free_lock );

        ppm_free_pages_nolock( page ); 

    // release lock protecting free_pages[] array
        spinlock_unlock( &ppm->free_lock );
}

////////////////////////////
void ppm_print( ppm_t * ppm,
                char  * string )
{
        uint32_t       order;
        list_entry_t * iter;
        page_t       * page;

    // get lock protecting free lists
        spinlock_lock( &ppm->free_lock );

        printk("\n***  PPM state in cluster %x %s : pages = %d / offset = %d / free = %d ***\n",
               local_cxy , string , ppm->pages_nr , ppm->pages_offset , ppm->total_free_pages ); 
        
        for( order = 0 ; order < CONFIG_PPM_MAX_ORDER ; order++ )
        {
                printk("- order = %d / free_pages = %d  [",
               order , ppm->free_pages_nr[order] );
                
                LIST_FOREACH( &ppm->free_pages_root[order] , iter )
                {
                        page = LIST_ELEMENT( iter , page_t , list );
                        printk("%d," , page - ppm->pages_tbl );
                }
    
                printk("]\n", NULL );
        }

    // release lock protecting free lists
        spinlock_unlock( &ppm->free_lock );

}  // end ppm_print()
 
//////////////////////////u/////////
void ppm_assert_order( ppm_t * ppm )
{
        uint32_t       order;
        list_entry_t * iter;
        page_t       * page;
        
        for(order=0; order < CONFIG_PPM_MAX_ORDER; order++)
        {
                if( list_is_empty( &ppm->free_pages_root[order] ) ) continue;
                
                LIST_FOREACH( &ppm->free_pages_root[order] , iter )
                {
                        page = LIST_ELEMENT( iter , page_t , list );

                        assert( (page->order == order) , __FUNCTION__ , "PPM inconsistency" );
                }
        }

}  // end ppm_assert_order()