source: trunk/kernel/mm/ppm.c @ 623

/*
 * ppm.c - Per-cluster Physical Pages Manager implementation
 *
 * Authors  Ghassan Almaless (2008,2009,2010,2011,2012)
 *          Alain Greiner    (2016,2017,2018)
 *
 * 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_kernel_types.h>
#include <hal_special.h>
#include <printk.h>
#include <list.h>
#include <bits.h>
#include <page.h>
#include <dqdt.h>
#include <busylock.h>
#include <queuelock.h>
#include <thread.h>
#include <cluster.h>
#include <kmem.h>
#include <process.h>
#include <mapper.h>
#include <ppm.h>
#include <vfs.h>

////////////////////////////////////////////////////////////////////////////////////////
//     functions to  translate [ page <-> base <-> ppn ]
////////////////////////////////////////////////////////////////////////////////////////

////////////////////////////////////////////////
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 xptr_t ppm_page2base( xptr_t page_xp )
{
        ppm_t  * ppm      = &LOCAL_CLUSTER->ppm;

    cxy_t    page_cxy = GET_CXY( page_xp );
    page_t * page_ptr = GET_PTR( page_xp );

   void   * base_ptr = ppm->vaddr_base + 
                       ((page_ptr - ppm->pages_tbl)<<CONFIG_PPM_PAGE_SHIFT);

        return XPTR( page_cxy , base_ptr );

} // end ppm_page2base()

/////////////////////////////////////////////
inline xptr_t ppm_base2page( xptr_t base_xp )
{
        ppm_t  * ppm = &LOCAL_CLUSTER->ppm;

    cxy_t    base_cxy = GET_CXY( base_xp );
    void   * base_ptr = GET_PTR( base_xp );

        page_t * page_ptr = ppm->pages_tbl + 
                        ((base_ptr - ppm->vaddr_base)>>CONFIG_PPM_PAGE_SHIFT);

        return XPTR( base_cxy , page_ptr );

}  // end ppm_base2page()



///////////////////////////////////////////
inline ppn_t ppm_page2ppn( xptr_t page_xp )
{
        ppm_t  * ppm      = &LOCAL_CLUSTER->ppm;

    cxy_t    page_cxy = GET_CXY( page_xp );
    page_t * page_ptr = GET_PTR( page_xp );

    paddr_t  paddr    = PADDR( page_cxy , (page_ptr - ppm->pages_tbl)<<CONFIG_PPM_PAGE_SHIFT );

    return (ppn_t)(paddr >> CONFIG_PPM_PAGE_SHIFT);

}  // end hal_page2ppn()

///////////////////////////////////////
inline xptr_t ppm_ppn2page( ppn_t ppn )
{
        ppm_t   * ppm  = &LOCAL_CLUSTER->ppm;

    paddr_t  paddr = ((paddr_t)ppn) << CONFIG_PPM_PAGE_SHIFT;

    cxy_t    cxy   = CXY_FROM_PADDR( paddr );
    lpa_t    lpa   = LPA_FROM_PADDR( paddr );

    return XPTR( cxy , &ppm->pages_tbl[lpa>>CONFIG_PPM_PAGE_SHIFT] );

}  // end hal_ppn2page



///////////////////////////////////////
inline xptr_t ppm_ppn2base( ppn_t ppn )
{
        ppm_t  * ppm   = &LOCAL_CLUSTER->ppm;
   
    paddr_t  paddr = ((paddr_t)ppn) << CONFIG_PPM_PAGE_SHIFT;

    cxy_t    cxy   = CXY_FROM_PADDR( paddr );
    lpa_t    lpa   = LPA_FROM_PADDR( paddr );

        return XPTR( cxy , (void *)ppm->vaddr_base + lpa );

}  // end ppm_ppn2base()

///////////////////////////////////////////
inline ppn_t ppm_base2ppn( xptr_t base_xp )
{
        ppm_t  * ppm      = &LOCAL_CLUSTER->ppm;

    cxy_t    base_cxy = GET_CXY( base_xp );
    void   * base_ptr = GET_PTR( base_xp );

    paddr_t  paddr    = PADDR( base_cxy , (base_ptr - ppm->vaddr_base) );

    return (ppn_t)(paddr >> CONFIG_PPM_PAGE_SHIFT);

}  // end ppm_base2ppn()


////////////////////////////////////////////////////////////////////////////////////////
//     functions to  allocate / release  physical pages 
////////////////////////////////////////////////////////////////////////////////////////

///////////////////////////////////////////
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 (merged) page order

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

        assert( !page_is_flag( page , PG_FREE ) ,
    "page already released : ppn = %x\n" , ppm_page2ppn(XPTR(local_cxy,page)) );

        assert( !page_is_flag( page , PG_RESERVED ) ,
    "reserved page : ppn = %x\n" , ppm_page2ppn(XPTR(local_cxy,page)) );

        // 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] --;

                // 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] ++;

}  // end ppm_free_pages_nolock()

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

#if DEBUG_PPM_ALLOC_PAGES
thread_t * this = CURRENT_THREAD;
uint32_t cycle = (uint32_t)hal_get_cycles();
if( DEBUG_PPM_ALLOC_PAGES < cycle )
printk("\n[%s] thread[%x,%x] enter for %d page(s) / cycle %d\n",
__FUNCTION__, this->process->pid, this->trdid, 1<<order, cycle );
#endif

#if(DEBUG_PPM_ALLOC_PAGES & 0x1)
if( DEBUG_PPM_ALLOC_PAGES < cycle )
ppm_print("enter ppm_alloc_pages");
#endif

        ppm_t    * ppm = &LOCAL_CLUSTER->ppm;

// check order
assert( (order < CONFIG_PPM_MAX_ORDER) , "illegal order argument = %d\n" , order );

        page_t * block = NULL;  

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

#if DEBUG_PPM_ALLOC_PAGES
cycle = (uint32_t)hal_get_cycles();
if( DEBUG_PPM_ALLOC_PAGES < cycle )
printk("\n[%s] thread[%x,%x] cannot allocate %d page(s) / cycle %d\n",
__FUNCTION__, this->process->pid, this->trdid, 1<<order, cycle );
#endif

                return NULL;
        }

        // update free-lists after removing a block
        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] ++;
        }

        // update page descriptor
        page_clear_flag( block , PG_FREE );
        page_refcount_up( block );
        block->order = order;

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

    // update DQDT
    dqdt_increment_pages( order );

#if DEBUG_PPM_ALLOC_PAGES
cycle = (uint32_t)hal_get_cycles();
if( DEBUG_PPM_ALLOC_PAGES < cycle )
printk("\n[%s] thread[%x,%x] exit for %d page(s) / ppn = %x / cycle %d\n",
__FUNCTION__, this->process->pid, this->trdid, 
1<<order, ppm_page2ppn(XPTR( local_cxy , block )), cycle );
#endif

#if(DEBUG_PPM_ALLOC_PAGES & 0x1)
if( DEBUG_PPM_ALLOC_PAGES < cycle )
ppm_print("exit ppm_alloc_pages");
#endif

        return block;

}  // end ppm_alloc_pages()


////////////////////////////////////
void ppm_free_pages( page_t * page )
{
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;

#if DEBUG_PPM_FREE_PAGES
uint32_t cycle = (uint32_t)hal_get_cycles();
if( DEBUG_PPM_FREE_PAGES < cycle )
printk("\n[%s] thread[%x,%x] enter for %d page(s) / ppn %x / cycle %d\n",
__FUNCTION__, this->process->pid, this->trdid, 
1<<page->order, ppm_page2ppn(XPTR(local_cxy , page)), cycle );
#endif

#if(DEBUG_PPM_FREE_PAGES & 0x1)
if( DEBUG_PPM_FREE_PAGES < cycle )
ppm_print("enter ppm_free_pages");
#endif

        // get lock protecting free_pages[] array
        busylock_acquire( &ppm->free_lock );

        ppm_free_pages_nolock( page );

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

    // update DQDT
    dqdt_decrement_pages( page->order );

#if DEBUG_PPM_FREE_PAGES
cycle = (uint32_t)hal_get_cycles();
if( DEBUG_PPM_FREE_PAGES < cycle )
printk("\n[%s] thread[%x,%x] exit for %d page(s) / ppn %x / cycle %d\n",
__FUNCTION__, this->process->pid, this->trdid, 
1<<page->order, ppm_page2ppn(XPTR(local_cxy , page)), cycle );
#endif

#if(DEBUG_PPM_FREE_PAGES & 0x1)
if( DEBUG_PPM_FREE_PAGES < cycle )
ppm_print("exit ppm_free_pages");
#endif

}  // end ppm_free_pages()

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

    ppm_t * ppm = &LOCAL_CLUSTER->ppm;

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

        printk("\n***  PPM in cluster %x / %s / %d pages ***\n",
    local_cxy , string, ppm->pages_nr );

        for( order = 0 ; order < CONFIG_PPM_MAX_ORDER ; order++ )
        {
                printk("- order = %d / free_pages = %d\t: ",
                       order , ppm->free_pages_nr[order] );

                LIST_FOREACH( &ppm->free_pages_root[order] , iter )
                {
                        page = LIST_ELEMENT( iter , page_t , list );
                        printk("%x," , page - ppm->pages_tbl );
                }

                printk("\n");
        }

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

///////////////////////////////////////
error_t 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 );
                        if( page->order != order )  return -1;
                }
        }

        return 0;
}


//////////////////////////////////////////////////////////////////////////////////////
//     functions to handle  dirty physical pages
//////////////////////////////////////////////////////////////////////////////////////

//////////////////////////////////////////
bool_t ppm_page_do_dirty( xptr_t page_xp )
{
        bool_t done = false;

    // get page cluster and local pointer
    page_t * page_ptr = GET_PTR( page_xp );
    cxy_t    page_cxy = GET_CXY( page_xp );

    // get local pointer on PPM (same in all clusters)
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;

    // build extended pointers on page lock, page flags, and PPM dirty list lock
    xptr_t page_lock_xp  = XPTR( page_cxy , &page_ptr->lock  );    
    xptr_t page_flags_xp = XPTR( page_cxy , &page_ptr->flags );
    xptr_t dirty_lock_xp = XPTR( page_cxy , &ppm->dirty_lock );
           
        // lock the remote PPM dirty_list
        remote_queuelock_acquire( dirty_lock_xp );

    // lock the remote page
    remote_busylock_acquire( page_lock_xp );

    // get remote page flags
    uint32_t flags = hal_remote_l32( page_flags_xp );

        if( (flags & PG_DIRTY) == 0 )
        {
                // set dirty flag in page descriptor
        hal_remote_s32( page_flags_xp , flags | PG_DIRTY );

                // The PPM dirty list is a LOCAL list !!!
        // We must update 4 pointers to insert a new page in this list.
        // We can use the standard LIST API when the page is local,
        // but we cannot use the standard API if the page is remote...

        if( page_cxy == local_cxy )         // locally update the PPM dirty list
        {
            list_add_first( &ppm->dirty_root , &page_ptr->list );
        } 
        else                                // remotely update the PPM dirty list
        {
            // get local and remote pointers on "root" list entry 
            list_entry_t * root    = &ppm->dirty_root;
            xptr_t         root_xp = XPTR( page_cxy , root );

            // get local and remote pointers on "page" list entry 
            list_entry_t * list    = &page_ptr->list;
            xptr_t         list_xp = XPTR( page_cxy , list );

            // get local and remote pointers on first dirty page 
            list_entry_t * dirt    = hal_remote_lpt( XPTR( page_cxy, &root->next ) );
            xptr_t         dirt_xp = XPTR( page_cxy , dirt );

            // set root.next, list.next, list pred, curr.pred in remote cluster
            hal_remote_spt( root_xp                    , list );
            hal_remote_spt( list_xp                    , dirt );
            hal_remote_spt( list_xp + sizeof(intptr_t) , root );
            hal_remote_spt( dirt_xp + sizeof(intptr_t) , list );
        }

                done = true;
        }

    // unlock the remote page
    remote_busylock_release( page_lock_xp );

        // unlock the remote PPM dirty_list
        remote_queuelock_release( dirty_lock_xp );

        return done;

} // end ppm_page_do_dirty()

////////////////////////////////////////////
bool_t ppm_page_undo_dirty( xptr_t page_xp )
{
        bool_t done = false;

    // get page cluster and local pointer
    page_t * page_ptr = GET_PTR( page_xp );
    cxy_t    page_cxy = GET_CXY( page_xp );

    // get local pointer on PPM (same in all clusters)
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;

    // build extended pointers on page lock, page flags, and PPM dirty list lock
    xptr_t page_lock_xp  = XPTR( page_cxy , &page_ptr->lock  );
    xptr_t page_flags_xp = XPTR( page_cxy , &page_ptr->flags );
    xptr_t dirty_lock_xp = XPTR( page_cxy , &ppm->dirty_lock );
           
        // lock the remote PPM dirty_list
        remote_queuelock_acquire( XPTR( page_cxy , &ppm->dirty_lock ) );

    // lock the remote page
    remote_busylock_acquire( page_lock_xp );

    // get remote page flags
    uint32_t flags = hal_remote_l32( page_flags_xp );

        if( (flags & PG_DIRTY) )  // page is dirty
        {
                // reset dirty flag in page descriptor
        hal_remote_s32( page_flags_xp , flags & (~PG_DIRTY) );

                // The PPM dirty list is a LOCAL list !!!
        // We must update 4 pointers to remove a page from this list.
        // we can use the standard LIST API when the page is local,
        // but we cannot use the standard API if the page is remote...

        if( page_cxy == local_cxy )         // locally update the PPM dirty list
        {
            list_unlink( &page_ptr->list );
        } 
        else                                // remotely update the PPM dirty list
        {
            // get local and remote pointers on "page" list entry 
            list_entry_t * list    = &page_ptr->list;
            xptr_t         list_xp = XPTR( page_cxy , list );

            // get local and remote pointers on "next" page list entry 
            list_entry_t * next    = hal_remote_lpt( list_xp );
            xptr_t         next_xp = XPTR( page_cxy , next );

            // get local and remote pointers on "pred" page list entry 
            list_entry_t * pred    = hal_remote_lpt( list_xp + sizeof(intptr_t) );
            xptr_t         pred_xp = XPTR( page_cxy , pred );

            // set root.next, list.next, list pred, curr.pred in remote cluster
            hal_remote_spt( pred_xp                    , next );
            hal_remote_spt( list_xp                    , NULL );
            hal_remote_spt( list_xp + sizeof(intptr_t) , NULL );
            hal_remote_spt( next_xp + sizeof(intptr_t) , pred );
        }

                done = true;
        }

    // unlock the remote page
    remote_busylock_release( page_lock_xp );

        // unlock the remote PPM dirty_list
        remote_queuelock_release( dirty_lock_xp );

        return done;

}  // end ppm_page_undo_dirty()

/////////////////////////////////
void ppm_sync_dirty_pages( void )
{
        ppm_t * ppm = &LOCAL_CLUSTER->ppm;

    // get local pointer on PPM dirty_root
    list_entry_t * dirty_root = &ppm->dirty_root;

    // build extended pointer on PPM dirty_lock 
    xptr_t dirty_lock_xp = XPTR( local_cxy , &ppm->dirty_lock );

        // get the PPM dirty_list lock
        remote_queuelock_acquire( dirty_lock_xp );

        while( !list_is_empty( &ppm->dirty_root ) )
        {
                page_t * page = LIST_FIRST( dirty_root ,  page_t , list );
        xptr_t   page_xp = XPTR( local_cxy , page );

        // build extended pointer on page lock
        xptr_t page_lock_xp = XPTR( local_cxy , &page->lock );

                // get the page lock
                remote_busylock_acquire( page_lock_xp );

                // sync the page
                vfs_fs_move_page( page_xp , false );  // from mapper to device

                // release the page lock
                remote_busylock_release( page_lock_xp );
        }

        // release the PPM dirty_list lock
        remote_queuelock_release( dirty_lock_xp );

}  // end ppm_sync_dirty_pages()