source: trunk/kernel/kern/dqdt.c @ 676

Last change on this file since 676 was 669, checked in by alain, 4 years ago

1) Introduce up to 4 command lines arguments in the KSH "load" command.
These arguments are transfered to the user process through the
argc/argv mechanism, using the user space "args" vseg.

2) Introduce the named and anonymous "pipes", for inter-process communication
through the pipe() and mkfifo() syscalls.

3) Introduce the "chat" application to validate the two above mechanisms.

4) Improve printk() and assert() fonctions in printk.c.
File size: 24.3 KB
Line 
1/*
2 * dqdt.c - Distributed Quaternary Decision Tree implementation.
3 *
4 * Author : Alain Greiner (2016,2017,2018,2019)
5 *
6 * Copyright (c)  UPMC Sorbonne Universites
7 *
8 * This file is part of ALMOS-MKH.
9 *
10 * ALMOS-MKH is free software; you can redistribute it and/or modify it
11 * under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; version 2.0 of the License.
13 *
14 * ALMOS-MKH is distributed in the hope that it will be useful, but
15 * WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
21 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24#include <kernel_config.h>
25#include <hal_kernel_types.h>
26#include <hal_special.h>
27#include <hal_macros.h>
28#include <hal_atomic.h>
29#include <hal_remote.h>
30#include <thread.h>
31#include <printk.h>
32#include <chdev.h>
33#include <cluster.h>
34#include <bits.h>
35#include <dqdt.h>
36
37
38///////////////////////////////////////////////////////////////////////////////////////////
39//      Extern variables
40///////////////////////////////////////////////////////////////////////////////////////////
41
42extern chdev_directory_t  chdev_dir;  // defined in chdev.h / allocated in kernel_init.c
43
44///////////////////////////////////////////////////////////////////////////////////////////
45// This static recursive function traverse the DQDT quad-tree from root to bottom.
46///////////////////////////////////////////////////////////////////////////////////////////
47static void dqdt_recursive_print( xptr_t  node_xp )
48{
49        uint32_t x;
50        uint32_t y;
51    dqdt_node_t node;
52
53    // get node local copy
54    hal_remote_memcpy( XPTR( local_cxy , &node ), node_xp , sizeof(dqdt_node_t) );
55
56    // display node content
57        nolock_printk("- [%d,%x] : threads %x / pages %x / clusters %d / cores %d / parent_cxy %x\n",
58                  node.level, GET_CXY( node_xp ),
59                  node.threads, node.pages,
60                  node.clusters, node.cores,
61                  GET_CXY( node.parent ) );
62
63    // recursive call on children if node is not terminal
64    if ( node.level > 0 )
65    {
66        for ( x = 0 ; x < 2 ; x++ )
67        {
68            for ( y = 0 ; y < 2 ; y++ )
69            {
70                xptr_t iter_xp = node.children[x][y];
71                if ( iter_xp != XPTR_NULL ) dqdt_recursive_print( iter_xp );
72            }
73        }
74    }
75}
76
77/////////////////////////
78void dqdt_display( void )
79{
80    // get extended pointer on DQDT root node
81        cluster_t * cluster = &cluster_manager;
82    xptr_t      root_xp = cluster->dqdt_root_xp;
83
84    // get pointers on TXT0 chdev
85    xptr_t    txt0_xp  = chdev_dir.txt_tx[0];
86    cxy_t     txt0_cxy = GET_CXY( txt0_xp );
87    chdev_t * txt0_ptr = GET_PTR( txt0_xp );
88
89    // get extended pointer on remote TXT0 lock
90    xptr_t  lock_xp = XPTR( txt0_cxy , &txt0_ptr->wait_lock );
91
92    // get TXT0 lock
93    remote_busylock_acquire( lock_xp );
94
95    // print header
96    nolock_printk("\n***** DQDT state\n\n");
97
98    // call recursive function
99    dqdt_recursive_print( root_xp );
100
101    // release TXT0 lock
102    remote_busylock_release( lock_xp );
103}
104
105///////////////////////////////////////////////////////////////////////////////////////
106// This static function initializes recursively, from top to bottom, the quad-tree
107// infrastructure. The DQDT nodes are allocated as global variables in each local
108// cluster manager. At each level in the quad-tree, this function initializes the
109// node identified by the <cxy> and <level> arguments, selects in each child
110// macro-cluster the precise cluster where will be placed the subtree root node,
111// and call recursively itself to initialize the child node in the selected cluster.
112///////////////////////////////////////////////////////////////////////////////////////
113// @ node cxy  : cluster containing the node to initialize
114// @ level     : level of node to be initialised
115// @ parent_xp : extended pointer on the parent node
116///////////////////////////////////////////////////////////////////////////////////////
117static void dqdt_recursive_build( cxy_t    node_cxy,
118                                  uint32_t level,
119                                  xptr_t   parent_xp )
120{
121    assert( __FUNCTION__, (level <= 5) , "illegal DQDT level\n" );
122 
123    uint32_t node_x;         // node X coordinate
124    uint32_t node_y;         // node Y coordinate
125    uint32_t mask;           // to compute associated macro-cluster coordinates
126    uint32_t node_base_x;    // associated macro_cluster X coordinate
127    uint32_t node_base_y;    // associated macro_cluster y coordinate
128    uint32_t half;           // associated macro-cluster half size
129    uint32_t cores;          // number of cores in macro cluster
130    uint32_t clusters;       // number of clusters in macro cluster
131
132    // get node cluster coordinates
133    node_x = HAL_X_FROM_CXY( node_cxy );
134    node_y = HAL_Y_FROM_CXY( node_cxy );
135       
136    // get macro-cluster mask and half-size
137    mask   = (1 << level) - 1;
138    half   = (level > 0) ? (1 << (level - 1)) : 0;
139
140    // get macro-cluster coordinates
141    node_base_x = node_x & ~mask;
142    node_base_y = node_y & ~mask;
143
144    // get pointer on local cluster manager
145    cluster_t * cluster = LOCAL_CLUSTER;
146
147    // build local and extended pointer on node to be initialized
148    dqdt_node_t * node_ptr = &cluster->dqdt_tbl[level];
149    xptr_t        node_xp  = XPTR( node_cxy , node_ptr );
150
151#if DEBUG_DQDT_INIT
152printk("\n[%s] thread[%x,%x] : cxy(%d,%d) / level %d / mask %x / half %d / ptr %x\n",
153__FUNCTION__, node_x, node_y, level, mask, half, node_ptr );
154#endif
155 
156    // make remote node default initialisation
157    hal_remote_memset( node_xp , 0 , sizeof( dqdt_node_t ) );
158
159    // initialize <parent> field
160    hal_remote_s64( XPTR( node_cxy , &node_ptr->parent ) , parent_xp );
161
162    // initialize <level> field
163    hal_remote_s32( XPTR( node_cxy , &node_ptr->level ) , level );
164
165    // recursive initialisation
166    if( level == 0 )                      // terminal case : cluster
167    {
168        // initialize <clusters> field in node
169        hal_remote_s32( XPTR( node_cxy , &node_ptr->clusters ) , 1 );
170
171        // initialize <cores> field in node
172        cores = hal_remote_l32( XPTR ( node_cxy , &cluster->cores_nr ) );
173        hal_remote_s32( XPTR( node_cxy , &node_ptr->cores ) , cores );
174    }
175    else                                  // non terminal : macro-cluster
176    {
177        bool_t        found;
178        uint32_t      x;
179        uint32_t      y;
180        cxy_t         child_cxy;
181        xptr_t        child_xp;
182        dqdt_node_t * child_ptr =  &cluster->dqdt_tbl[level-1];
183
184        // search an active cluster in child[0][0] macro-cluster
185        found = false; 
186        for( x = node_base_x ; 
187        (x < (node_base_x + half)) && (found == false) ; x++ )
188        {
189            for( y = node_base_y ; 
190            (y < (node_base_y + half)) && (found == false) ; y++ )
191            {
192                child_cxy = HAL_CXY_FROM_XY( x , y );
193
194                if( cluster_is_active( child_cxy ) )
195                {
196                    // initialize recursively selected child[0][0] node
197                    dqdt_recursive_build( child_cxy , level-1 , node_xp );
198
199                    // build extended pointer on child[0][0] node
200                    child_xp = XPTR( child_cxy , child_ptr );
201
202                    // update <cores> field in node
203                    cores = hal_remote_l32( XPTR ( child_cxy , &child_ptr->cores ) );
204                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->cores ) , cores );
205
206                    // update <clusters> field in node
207                    clusters = hal_remote_l32( XPTR ( child_cxy , &child_ptr->clusters ) );
208                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->clusters ) , clusters );
209
210                    // update <child[0][0]> field in node
211                    hal_remote_s64( XPTR( node_cxy , &node_ptr->children[0][0] ), child_xp );
212
213                    // udate <arity> field in node
214                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->arity ) , 1 );
215   
216                    // exit loops
217                    found = true;
218                }
219            }
220        }
221
222        // search an active cluster in child[0][1] macro-cluster
223        found = false; 
224        for( x = node_base_x ; 
225        (x < (node_base_x + half)) && (found == false) ; x++ )
226        {
227            for( y = (node_base_y + half) ; 
228            (y < (node_base_y + (half<<1))) && (found == false) ; y++ )
229            {
230                child_cxy = HAL_CXY_FROM_XY( x , y );
231
232                if( cluster_is_active( child_cxy ) )
233                {
234                    // initialize recursively selected child[0][1] node
235                    dqdt_recursive_build( child_cxy , level-1 , node_xp );
236
237                    // build extended pointer on child[0][1] node
238                    child_xp = XPTR( child_cxy , child_ptr );
239
240                    // update <cores> field in node
241                    cores = hal_remote_l32( XPTR ( child_cxy , &child_ptr->cores ) );
242                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->cores ) , cores );
243
244                    // update <clusters> field in node
245                    clusters = hal_remote_l32( XPTR ( child_cxy , &child_ptr->clusters ) );
246                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->clusters ) , clusters );
247
248                    // update <child[0][1]> field in node
249                    hal_remote_s64( XPTR( node_cxy , &node_ptr->children[0][1] ), child_xp );
250
251                    // udate <arity> field in node
252                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->arity ) , 1 );
253   
254                    // exit loops
255                    found = true;
256                }
257            }
258        }
259
260        // search an active cluster in child[1][0] macro-cluster
261        found = false; 
262        for( x = (node_base_x +half) ; 
263        (x < (node_base_x + (half<<1))) && (found == false) ; x++ )
264        {
265            for( y = node_base_y ; 
266            (y < (node_base_y + half)) && (found == false) ; y++ )
267            {
268                child_cxy = HAL_CXY_FROM_XY( x , y );
269
270                if( cluster_is_active( child_cxy ) )
271                {
272                    // initialize recursively selected child[1][0] node
273                    dqdt_recursive_build( child_cxy , level-1 , node_xp );
274
275                    // build extended pointer on child[1][0] node
276                    child_xp = XPTR( child_cxy , child_ptr );
277
278                    // update <cores> field in node
279                    cores = hal_remote_l32( XPTR ( child_cxy , &child_ptr->cores ) );
280                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->cores ) , cores );
281
282                    // update <clusters> field in node
283                    clusters = hal_remote_l32( XPTR ( child_cxy , &child_ptr->clusters ) );
284                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->clusters ) , clusters );
285
286                    // update <child[1][0]> field in node
287                    hal_remote_s64( XPTR( node_cxy , &node_ptr->children[1][0] ), child_xp );
288
289                    // udate <arity> field in node
290                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->arity ) , 1 );
291   
292                    // exit loops
293                    found = true;
294                }
295            }
296        }
297
298        // search an active cluster in child[1][1] macro-cluster
299        found = false; 
300        for( x = (node_base_x + half) ; 
301        (x < (node_base_x + (half<<1))) && (found == false) ; x++ )
302        {
303            for( y = (node_base_y + half) ; 
304            (y < (node_base_y + (half<<1))) && (found == false) ; y++ )
305            {
306                child_cxy = HAL_CXY_FROM_XY( x , y );
307
308                if( cluster_is_active( child_cxy ) )
309                {
310                    // initialize recursively selected child[1][1] node
311                    dqdt_recursive_build( child_cxy , level-1 , node_xp );
312
313                    // build extended pointer on child[1][1] node
314                    child_xp = XPTR( child_cxy , child_ptr );
315
316                    // update <cores> field in node
317                    cores = hal_remote_l32( XPTR ( child_cxy , &child_ptr->cores ) );
318                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->cores ) , cores );
319
320                    // update <clusters> field in node
321                    clusters = hal_remote_l32( XPTR ( child_cxy , &child_ptr->clusters ) );
322                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->clusters ) , clusters );
323
324                    // update <child[1][1]> field in node
325                    hal_remote_s64( XPTR( node_cxy , &node_ptr->children[1][1] ), child_xp );
326
327                    // udate <arity> field in node
328                    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->arity ) , 1 );
329   
330                    // exit loops
331                    found = true;
332                }
333            }
334        }
335    }
336}  // end dqdt_recursive_build()
337
338//////////////////////
339void dqdt_init( void )
340{
341    // get x_size & y_size
342    cluster_t * cluster = LOCAL_CLUSTER;
343    uint32_t    x_size  = cluster->x_size;
344    uint32_t    y_size  = cluster->y_size;
345
346    assert( __FUNCTION__, ((x_size <= 16) && (y_size <= 16)) , "illegal mesh size\n");
347 
348    // compute level_max
349    uint32_t  x_size_ext = POW2_ROUNDUP( x_size );
350    uint32_t  y_size_ext = POW2_ROUNDUP( y_size );
351    uint32_t  size_ext   = MAX( x_size_ext , y_size_ext );
352    uint32_t  level_max  = bits_log2( size_ext );
353
354    // all CP0s register the DQDT root in local cluster manager
355    cluster->dqdt_root_xp = XPTR( 0 , &cluster->dqdt_tbl[level_max] );
356
357    // only CP0 in cluster 0 build the DQDT
358    if( local_cxy == 0 )
359    {
360
361#if DEBUG_DQDT_INIT
362thread_t * this = CURRENT_THREAD;
363printk("\n[%s] thread[%x,%x] enters : x_size = %d / y_size = %d / level_max = %d\n",
364__FUNCTION__, this->process->pid, this->trdid, x_size, y_size, level_max );
365#endif
366   
367    // only CP0 in cluster 0 call the recursive function to build the quad-tree
368    if (local_cxy == 0) dqdt_recursive_build( local_cxy , level_max , XPTR_NULL );
369
370#if DEBUG_DQDT_INIT
371dqdt_display();
372#endif
373
374    }
375}  // end dqdt_init()
376
377
378///////////////////////////////////////////////////////////////////////////
379// This recursive function is called by both the dqdt_increment_pages()
380// and by the dqdt_decrement_pages() functions.
381// It traverses the quad tree from clusters to root.
382///////////////////////////////////////////////////////////////////////////
383// @ node_xp    : extended pointer on current node
384// @ increment  : number of pages variation
385///////////////////////////////////////////////////////////////////////////
386static void dqdt_propagate_pages( xptr_t  node_xp,
387                                  int32_t increment )
388{
389    // get current node cluster identifier and local pointer
390    cxy_t         node_cxy = GET_CXY( node_xp );
391    dqdt_node_t * node_ptr = GET_PTR( node_xp );
392
393    // update current node pages number
394    hal_remote_atomic_add( XPTR( node_cxy , &node_ptr->pages ) , increment );
395
396    // get extended pointer on parent node
397    xptr_t parent_xp = (xptr_t)hal_remote_l64( XPTR( node_cxy , &node_ptr->parent ) );
398
399    // propagate if required
400    if ( parent_xp != XPTR_NULL ) dqdt_propagate_pages( parent_xp, increment );
401}
402
403////////////////////////////////////////
404void dqdt_increment_pages( cxy_t    cxy,
405                           uint32_t order )
406{
407    // get local pointer on node[0] (same in all clusters)
408    dqdt_node_t * node_ptr = &LOCAL_CLUSTER->dqdt_tbl[0];
409
410    // update DQDT node[0] in remote cluster cxy
411    hal_remote_atomic_add( XPTR( cxy , &node_ptr->pages ) , (1 << order) );
412
413    // get extended pointer on parent node in remote cluster cxy
414    xptr_t parent_xp = hal_remote_l64( XPTR( cxy , &node_ptr->parent ) );
415
416     // propagate to DQDT upper levels
417    if( parent_xp != XPTR_NULL ) dqdt_propagate_pages( parent_xp , (1 << order) );
418
419#if DEBUG_DQDT_UPDATE_PAGES
420uint32_t cycle = hal_get_cycles();
421if( cycle > DEBUG_DQDT_UPDATE_PAGES )
422printk("\n[DBG] %s : thread %x in process %x / %x pages in cluster %x / cycle %d\n",
423__FUNCTION__, CURRENT_THREAD->trdid, CURRENT_THREAD->process->pid,
424hal_remote_l32( XPTR( cxy , &node_ptr->pages ), cxy, cycle );
425#endif
426
427}
428
429////////////////////////////////////////
430void dqdt_decrement_pages( cxy_t    cxy,
431                           uint32_t order )
432{
433    // get local pointer on node[0] (same in all clusters)
434    dqdt_node_t * node_ptr = &LOCAL_CLUSTER->dqdt_tbl[0];
435
436    // update DQDT node[0] in remote cluster cxy
437    hal_remote_atomic_add( XPTR( cxy , &node_ptr->pages ) , -(1 << order) );
438
439    // get extended pointer on parent node in remote cluster cxy
440    xptr_t parent_xp = hal_remote_l64( XPTR( cxy , &node_ptr->parent ) );
441
442     // propagate to DQDT upper levels
443    if( parent_xp != XPTR_NULL ) dqdt_propagate_pages( parent_xp , -(1 << order) );
444
445#if DEBUG_DQDT_UPDATE_PAGES
446uint32_t cycle = hal_get_cycles();
447if( cycle > DEBUG_DQDT_UPDATE_PAGES )
448printk("\n[DBG] %s : thread %x in process %x / %x pages in cluster %x / cycle %d\n",
449__FUNCTION__, CURRENT_THREAD->trdid, CURRENT_THREAD->process->pid,
450hal_remote_l32( XPTR( cxy , &node_ptr->pages ), cxy, cycle );
451#endif
452
453}
454
455
456
457///////////////////////////////////////////////////////////////////////////
458// This recursive function is called by both the dqdt_increment_threads()
459// and by the dqdt_decrement_threads functions.
460// It traverses the quad tree from clusters to root.
461///////////////////////////////////////////////////////////////////////////
462// @ node       : extended pointer on current node
463// @ increment  : number of pages variation
464///////////////////////////////////////////////////////////////////////////
465static void dqdt_propagate_threads( xptr_t  node,
466                                    int32_t increment )
467{
468    // get current node cluster identifier and local pointer
469    cxy_t         cxy = GET_CXY( node );
470    dqdt_node_t * ptr = GET_PTR( node );
471
472    // update current node threads number
473    hal_remote_atomic_add( XPTR( cxy , &ptr->threads ) , increment );
474
475    // get extended pointer on parent node
476    xptr_t parent = (xptr_t)hal_remote_l64( XPTR( cxy , &ptr->parent ) );
477
478    // propagate if required
479    if ( parent != XPTR_NULL ) dqdt_propagate_threads( parent, increment );
480}
481
482///////////////////////////////////
483void dqdt_increment_threads( void )
484{
485        cluster_t   * cluster = LOCAL_CLUSTER;
486    dqdt_node_t * node    = &cluster->dqdt_tbl[0];
487
488    // update DQDT node level 0
489    hal_atomic_add( &node->threads , 1 );
490
491    // propagate to DQDT upper levels
492    if( node->parent != XPTR_NULL ) dqdt_propagate_threads( node->parent , 1 );
493
494#if DEBUG_DQDT_UPDATE_THREADS
495uint32_t cycle = hal_get_cycles();
496if( cycle > DEBUG_DQDT_UPDATE_THREADS )
497printk("\n[DBG] %s : thread %x in process %x / %d threads in cluster %x / cycle %d\n",
498__FUNCTION__, CURRENT_THREAD->trdid, CURRENT_THREAD->process->pid, 
499node->threads, local_cxy, cycle );
500#endif
501
502}
503
504///////////////////////////////////
505void dqdt_decrement_threads( void )
506{
507        cluster_t   * cluster = LOCAL_CLUSTER;
508    dqdt_node_t * node    = &cluster->dqdt_tbl[0];
509
510    // update DQDT node level 0
511    hal_atomic_add( &node->threads , -1 );
512
513    // propagate to DQDT upper levels
514    if( node->parent != XPTR_NULL ) dqdt_propagate_threads( node->parent , -1 );
515
516#if DEBUG_DQDT_UPDATE_THREADS
517uint32_t cycle = hal_get_cycles();
518if( cycle > DEBUG_DQDT_UPDATE_THREADS )
519printk("\n[DBG] %s : thread %x in process %x / %d threads in cluster %x / cycle %d\n",
520__FUNCTION__, CURRENT_THREAD->trdid, CURRENT_THREAD->process->pid, 
521node->threads, local_cxy, cycle );
522#endif
523
524}
525
526///////////////////////////////////
527xptr_t dqdt_get_root( cxy_t    cxy,
528                      uint32_t level )
529{
530    xptr_t        node_xp;
531    cxy_t         node_cxy;
532    dqdt_node_t * node_ptr;
533    uint32_t      current_level;
534
535assert( __FUNCTION__, (level <= 5) , "illegal DQDT level\n" );
536
537#if DEBUG_DQDT_GET_ROOT
538thread_t * this = CURRENT_THREAD;
539printk("\n[%s] thread[%x,%x] enters / cxy %x / level %d\n",
540__FUNCTION__, this->process->pid, this->trdid, cxy, level );
541#endif
542
543    // check macro-cluster
544    if( cluster_is_active( cxy ) )
545    {   
546        // initialise node_xp and current_level
547        node_xp       = XPTR( cxy , &LOCAL_CLUSTER->dqdt_tbl[0] );
548        current_level = 0;
549
550        // traverse the quad-tree from bottom to root
551        while( current_level < level )
552        {
553            node_cxy = GET_CXY( node_xp );
554            node_ptr = GET_PTR( node_xp );
555
556            node_xp = hal_remote_l64( XPTR( node_cxy , &node_ptr->parent ) );
557            current_level++;
558        }
559    }
560    else
561    {
562        node_xp =  XPTR_NULL;
563    }
564
565#if DEBUG_DQDT_GET_ROOT
566printk("\n[%s] thread[%x,%x] exit / root_xp[%x,%x]\n",
567__FUNCTION__, this->process->pid, this->trdid, GET_CXY( node_xp ), GET_PTR( node_xp ) );
568#endif
569
570    return node_xp;
571   
572}
573
574/////////////////////////////////////////////////////////////////////////////////////
575// This recursive function is called by both the dqdt_get_cluster_for_process()
576// and by the dqdt_get_cluster_for_memory() functions to select the cluster with the
577// smallest number of threads per core, or the smallest number of pages per cluster.
578// It traverses the quad tree from root to clusters.
579/////////////////////////////////////////////////////////////////////////////////////
580static cxy_t dqdt_select_cluster( xptr_t node,
581                                  bool_t for_memory )
582{
583    dqdt_node_t   node_copy;     // local copy of the current DQDT node
584    xptr_t        child_xp;      // extended pointer on a DQDT child node
585    uint32_t      x;             // child node X coordinate
586    uint32_t      y;             // child node Y coordinate
587    uint32_t      select_x;      // selected child X coordinate
588    uint32_t      select_y;      // selected child Y coordinate
589    uint32_t      load;          // load of the child (threads or pages)
590    uint32_t      load_min;      // current value of the minimal load
591
592    // get DQDT node local copy
593    hal_remote_memcpy( XPTR( local_cxy , &node_copy ), node , sizeof(dqdt_node_t) );
594
595    // return cluster identifier for a terminal mode
596    if( node_copy.level == 0 ) return GET_CXY(node);
597
598    // analyse load for all children in non terminal node
599    load_min = 0xFFFFFFFF;
600    select_x = 0;
601    select_y = 0;
602    for( x = 0 ; x < 2 ; x++ )
603    {
604        for( y = 0 ; y < 2 ; y++ )
605        {
606            child_xp = node_copy.children[x][y];
607            if( child_xp != XPTR_NULL )
608            {
609                cxy_t         cxy  = GET_CXY( child_xp );
610                dqdt_node_t * ptr  = GET_PTR( child_xp );
611
612                // compute average load  for each child
613                if( for_memory )
614                {
615                    load = hal_remote_l32( XPTR( cxy , &ptr->pages ) ) /
616                           hal_remote_l32( XPTR( cxy , &ptr->clusters ) );
617                }
618                else
619                {
620                    load = hal_remote_l32( XPTR( cxy , &ptr->threads ) ) /
621                           hal_remote_l32( XPTR( cxy , &ptr->cores ) );
622                }
623
624                // select children with smallest load
625                if( load <= load_min )
626                {
627                    load_min = load;
628                    select_x = x;
629                    select_y = y;
630                }
631            }
632        }
633    }
634
635    // select the child with the lowest load
636    return dqdt_select_cluster( node_copy.children[select_x][select_y], for_memory );
637
638}  // end dqdt_select_cluster()
639
640
641///////////////////////////////////////////////////
642cxy_t dqdt_get_cluster_for_thread( xptr_t root_xp )
643{
644    // call recursive function
645    cxy_t cxy = dqdt_select_cluster( root_xp , false );
646
647#if DEBUG_DQDT_SELECT_FOR_THREAD
648uint32_t cycle = hal_get_cycles();
649if( cycle > DEBUG_DQDT_SELECT_FOR_PROCESS )
650printk("\n[DBG] %s : thread %x in process %x select cluster %x / cycle %d\n",
651__FUNCTION__, CURRENT_THREAD->trdid, CURRENT_THREAD->process->pid, cxy, cycle );
652#endif
653
654    return cxy;
655}
656
657///////////////////////////////////////////////////
658cxy_t dqdt_get_cluster_for_memory( xptr_t root_xp )
659{
660    // call recursive function
661    cxy_t cxy = dqdt_select_cluster( root_xp , true );
662
663#if DEBUG_DQDT_SELECT_FOR_MEMORY
664uint32_t cycle = hal_get_cycles();
665if( cycle > DEBUG_DQDT_SELECT_FOR_MEMORY )
666printk("\n[DBG] %s : thread %x in process %x select cluster %x / cycle %d\n",
667__FUNCTION__, CURRENT_THREAD->trdid, CURRENT_THREAD->process->pid, cxy, cycle );
668#endif
669
670    return cxy;
671}
672
Note: See TracBrowser for help on using the repository browser.