source: trunk/kernel/mm/mapper.c @ 21

Last change on this file since 21 was 18, checked in by max@…, 7 years ago

cosmetic, and a few typos

File size: 18.0 KB
Line 
1/*
2 * mapper.c - Map memory, file or device in process virtual address space.
3 *
4 * Authors   Mohamed Lamine Karaoui (2015)
5 *           Alain Greiner (2016)
6 *
7 * Copyright (c)  UPMC Sorbonne Universites
8 *
9 * This file is part of ALMOS-MKH.
10 *
11 * ALMOS-MKH is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; version 2.0 of the License.
14 *
15 * ALMOS-MKH is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
22 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25#include <kernel_config.h>
26#include <hal_types.h>
27#include <hal_special.h>
28#include <grdxt.h>
29#include <rwlock.h>
30#include <printk.h>
31#include <thread.h>
32#include <core.h>
33#include <process.h>
34#include <kmem.h>
35#include <kcm.h>
36#include <page.h>
37#include <cluster.h>
38#include <vfs.h>
39#include <mapper.h>
40
41//////////////////////////
42mapper_t * mapper_create()
43{
44    mapper_t * mapper;
45    kmem_req_t req;
46    error_t    error;
47
48    // allocate memory for associated mapper
49        req.type  = KMEM_MAPPER;
50        req.size  = sizeof(mapper_t);
51    req.flags = AF_KERNEL | AF_ZERO;
52        mapper    = (mapper_t *)kmem_alloc( &req );
53
54    if( mapper == NULL )
55    {
56        printk("\n[ERROR] in %s : no memory for mapper descriptor\n", __FUNCTION__ );
57        return NULL;
58    }
59
60    // initialize refcount & inode
61        mapper->refcount = 0;
62    mapper->inode    = NULL;
63
64    // initialize radix tree
65        error = grdxt_init( &mapper->radix,
66                        CONFIG_VMM_GRDXT_W1,
67                        CONFIG_VMM_GRDXT_W2,
68                        CONFIG_VMM_GRDXT_W3 );
69
70    if( error )
71    {
72        printk("\n[ERROR] in %s : cannot initialize radix tree\n", __FUNCTION__ );
73            req.type  = KMEM_MAPPER;
74        req.ptr   = mapper;
75        kmem_free( &req );
76        return NULL;
77    }
78
79    // initialize mapper lock
80        rwlock_init(  &mapper->lock );
81
82    // initialize waiting threads xlist (empty)
83        xlist_root_init( XPTR( local_cxy , &mapper->wait_root ) );
84
85    // initialize vsegs xlist (empty)
86        xlist_root_init( XPTR( local_cxy , &mapper->vsegs_root ) );
87
88    return mapper;
89
90} // end mapper_create()
91
92///////////////////////////////////////////
93error_t mapper_destroy( mapper_t * mapper )
94{
95    page_t   * page;
96    uint32_t   found_index = 0;
97    uint32_t   start_index = 0;
98    kmem_req_t req;
99    error_t    error;
100
101    // scan radix three and release all registered pages to PPM
102    do
103    {
104        // get page from radix tree
105        page = (page_t *)grdxt_get_first( &mapper->radix , start_index , &found_index );
106
107        if( page != NULL )
108        {
109            // remove page from mapper and release to PPM
110                error = mapper_release_page( mapper , found_index , page );
111
112            if ( error ) return error;
113
114            // update start_key value for next page
115            start_index = found_index;
116        }
117    }
118    while( page != NULL );
119
120    // release the memory allocated to radix-tree itself
121    grdxt_destroy( &mapper->radix );
122
123    // release memory for mapper descriptor
124    req.type = KMEM_MAPPER;
125    req.ptr  = mapper;
126    kmem_free( &req );
127
128    return 0;
129
130}  // end mapper_destroy()
131
132
133////////////////////////////////////////////
134page_t * mapper_get_page( mapper_t * mapper,
135                          uint32_t   index )
136{
137        kmem_req_t    req;
138        page_t      * page;
139        error_t       error;
140
141    thread_t * this = CURRENT_THREAD;
142
143    // take mapper lock in READ_MODE
144    rwlock_rd_lock( &mapper->lock );
145
146    // search page in radix tree
147    page = (page_t *)grdxt_lookup( &mapper->radix , index );
148
149    // test if page available in mapper
150    if( ( page == NULL) || page_is_flag( page , PG_INLOAD ) )  // page not available            /
151    {
152        // release the lock in READ_MODE and take it in WRITE_MODE
153        rwlock_rd_unlock( &mapper->lock );
154        rwlock_wr_lock( &mapper->lock );
155
156        // second test on missing page because the page status can have been modified
157        // by another thread, when passing from READ_MODE to WRITE_MODE.
158        // from this point there is no concurrent accesses to mapper.
159
160        page = grdxt_lookup( &mapper->radix , index );
161
162        if ( page == NULL )   // missing page => load it from file system
163        {
164            // allocate one page from PPM
165            req.type  = KMEM_PAGE;
166            req.size  = 0;
167            req.flags = AF_NONE;
168            page = kmem_alloc( &req );
169
170            if( page == NULL )
171            {
172                printk("\n[ERROR] in %s : thread %x cannot allocate a page in cluster %x\n",
173                       __FUNCTION__ , this->trdid , local_cxy );
174                rwlock_wr_unlock( &mapper->lock );
175                return NULL;
176            }
177
178            // initialize the page descriptor
179            page_init( page );
180            page_set_flag( page , PG_INIT );
181            page_set_flag( page , PG_INLOAD );
182            page_refcount_up( page );
183            page->mapper = mapper;
184            page->index  = index;
185
186            // insert page in mapper radix tree
187            error = grdxt_insert( &mapper->radix, index , page );
188
189            // release mapper lock from WRITE_MODE
190            rwlock_wr_unlock( &mapper->lock );
191
192            if( error )
193            {
194                printk("\n[ERROR] in %s : thread %x cannot insert page in mapper\n",
195                       __FUNCTION__ , this->trdid );
196                mapper_release_page( mapper , index , page );
197                page_clear_flag( page , PG_ALL );
198                req.ptr  = page;
199                req.type = KMEM_PAGE;
200                kmem_free(&req);
201                return NULL;
202            }
203
204            // launch I/O operation to load page from file system
205            error = mapper_updt_page( mapper , index , page );
206
207            if( error )
208            {
209                printk("\n[ERROR] in %s : thread %x cannot load page from device\n",
210                       __FUNCTION__ , this->trdid );
211                mapper_release_page( mapper , index , page );
212                page_clear_flag( page , PG_ALL );
213                req.ptr  = page;
214                req.type = KMEM_PAGE;
215                kmem_free( &req );
216                return NULL;
217            }
218
219            // update the mapper and index fields in page descriptor
220            page->mapper = mapper;
221            page->index  = index;
222
223            // reset the page INLOAD flag to make the page available to all readers
224            page_clear_flag( page , PG_INLOAD );
225
226        }
227        else if( page_is_flag( page , PG_INLOAD ) )   // page is loaded by another thread
228        {
229            // release mapper lock from WRITE_MODE
230            rwlock_wr_unlock( &mapper->lock );
231
232            // deschedule to wait load completion
233            while( 1 )
234            {
235                // exit waiting loop when loaded
236                if(  page_is_flag( page , PG_INLOAD ) ) break;
237
238                // deschedule
239                sched_yield();
240            }
241
242        }
243
244        return page;
245    }
246    else
247    {
248         // release lock from READ_MODE
249         rwlock_rd_unlock( &mapper->lock );
250
251         return page;
252    }
253}  // end mapper_get_page()
254
255///////////////////////////////////////////////
256error_t mapper_release_page( mapper_t * mapper,
257                             uint32_t   index,
258                             page_t   * page )
259{
260    error_t error;
261
262    // lauch IO operation to update page to file system
263    error = mapper_sync_page( mapper , index , page );
264
265    if( error )
266    {
267        printk("\n[ERROR] in %s : cannot update file system\n", __FUNCTION__ );
268        return EIO;
269    }
270
271    // take mapper lock in WRITE_MODE
272    rwlock_wr_lock( &mapper->lock );
273
274    // remove physical page from radix tree
275        grdxt_remove( &mapper->radix , page->index );
276
277    // release mapper lock from WRITE_MODE
278    rwlock_wr_unlock( &mapper->lock );
279
280    // release page to PPM
281        kmem_req_t   req;
282        req.type  = KMEM_PAGE;
283    req.ptr   = page;
284    kmem_free( &req );
285
286    return 0;
287
288}  // end mapper_release_page()
289
290////////////////////////////////////////////
291error_t mapper_updt_page( mapper_t * mapper,
292                          uint32_t   index,
293                          page_t   * page )
294{
295    uint32_t      type;
296    vfs_inode_t * inode;
297    error_t       error = 0;
298
299    if( page == NULL )
300    {
301        printk("\n[PANIC] in %s : page pointer is NULL\n", __FUNCTION__ );
302        hal_core_sleep();
303    }
304
305    if( mapper == NULL )
306    {
307        printk("\n[PANIC] in %s : no mapper for this page\n", __FUNCTION__ );
308        hal_core_sleep();
309    }
310
311    // get file system type and inode pointer
312    inode = mapper->inode;
313    type  = inode->ctx->type;
314
315    // get page lock
316    page_lock( page );
317
318    // get mapper lock in WRITE_MODE
319    rwlock_wr_lock( &mapper->lock );
320
321    // call proper I/O operation to update file system
322    if     ( type == FS_TYPE_FATFS ) error = fatfs_read_page( page );
323    else if( type == FS_TYPE_RAMFS ) error = ramfs_read_page( page );
324    else
325    {
326        printk("\n[PANIC] in %s : undefined file system type\n", __FUNCTION__ );
327        hal_core_sleep();
328    }
329
330    // release mapper lock from WRITE_MODE
331    rwlock_wr_unlock( &mapper->lock );
332
333    // release page lock
334    page_unlock( page );
335
336    if( error )
337    {
338        printk("\n[PANIC] in %s : cannot access file system\n", __FUNCTION__ );
339        return EIO;
340    }
341
342    return 0;
343}  // end mapper_updt_page
344
345////////////////////////////////////////////
346error_t mapper_sync_page( mapper_t * mapper,
347                          uint32_t   index,
348                          page_t   * page )
349{
350    uint32_t      type;
351    vfs_inode_t * inode;
352    error_t       error = 0;
353
354    if( page == NULL )
355    {
356        printk("\n[PANIC] in %s : page pointer is NULL\n", __FUNCTION__ );
357        hal_core_sleep();
358    }
359
360    if( mapper == NULL )
361    {
362        printk("\n[PANIC] in %s : no mapper for this page\n", __FUNCTION__ );
363        hal_core_sleep();
364    }
365
366        if( page_is_flag( page , PG_DIRTY ) )
367        {
368        // get file system type and inode pointer
369        inode = mapper->inode;
370        type  = inode->ctx->type;
371
372        // get page lock
373        page_lock( page );
374
375        // get mapper lock in READ_MODE
376        rwlock_rd_lock( &mapper->lock );
377
378        // call proper I/O operation to update file system
379        if     ( type == FS_TYPE_FATFS ) error = fatfs_write_page( page );
380        else if( type == FS_TYPE_RAMFS ) error = ramfs_write_page( page );
381        else
382        {
383            printk("\n[PANIC] in %s : undefined file system type\n", __FUNCTION__ );
384            hal_core_sleep();
385        }
386
387        // release mapper lock from READ_MODE
388        rwlock_rd_unlock( &mapper->lock );
389
390        // release page lock
391        page_unlock( page );
392
393        if( error )
394        {
395            printk("\n[PANIC] in %s : cannot update file system\n", __FUNCTION__ );
396            return EIO;
397        }
398
399        // clear dirty bit
400                page_undo_dirty( page );
401     }
402
403    return 0;
404
405}  // end mapper_sync_page()
406
407///////////////////////////////////////////////////////////////////////////////////////
408// This static function is called by the mapper_move fragments() function.
409// It moves one fragment between an user buffer and the kernel mapper.
410// Implementation Note: It can require access to one or two pages in mapper:
411//  [max_page_index == min_page_index]     <=>  fragment fit in one mapper page
412//  [max_page index == min_page_index + 1] <=>  fragment spread on two mapper pages
413///////////////////////////////////////////////////////////////////////////////////////
414static error_t mapper_move_one_fragment( mapper_t   * mapper,
415                                         bool_t       to_buffer,
416                                         fragment_t * fragment )
417{
418    uint32_t   size;                 // number of bytes in fragment
419    cxy_t      buf_cxy;              // cluster identifier for user buffer
420    uint8_t  * buf_ptr;              // local pointer on first byte in user buffer
421
422    xptr_t     xp_buf;               // extended pointer on byte in user buffer
423    xptr_t     xp_map;               // extended pointer on byte in kernel mapper
424
425    uint32_t   min_file_offset;      // offset of first byte in file
426    uint32_t   max_file_offset;      // offset of last byte in file
427
428    uint32_t   first_page_index;     // index of first page in mapper
429    uint32_t   first_page_offset;    // offset of first byte in first page in mapper
430    uint32_t   first_page_size;      // offset of first byte in first page in mapper
431
432    uint32_t   second_page_index;    // index of last page in mapper
433    uint32_t   second_page_offset;   // offset of last byte in last page in mapper
434    uint32_t   second_page_size;     // offset of last byte in last page in mapper
435
436    page_t   * page;                 // pointer on one page descriptor in mapper
437    uint8_t  * map_ptr;              // local pointer on first byte in mapper
438
439    // get fragment attributes in user buffer
440    buf_cxy = fragment->buf_cxy;
441    buf_ptr = fragment->buf_ptr;
442    size    = fragment->size;
443
444    if( size > CONFIG_PPM_PAGE_SIZE )
445    {
446        printk("\n[PANIC] in %s : illegal fragment size = %d\n",
447               __FUNCTION__ , size );
448        return EINVAL;
449    }
450
451    // compute offsets of first and last bytes in file
452    min_file_offset = fragment->file_offset;
453    max_file_offset = min_file_offset + size;
454
455    // compute indexes of pages for first and last byte in mapper
456    first_page_index  = min_file_offset >> CONFIG_PPM_PAGE_SHIFT;
457    second_page_index = max_file_offset >> CONFIG_PPM_PAGE_SHIFT;
458
459    if ( first_page_index == second_page_index )  // only one page in mapper
460    {
461        // compute offset and size for page in mapper
462        first_page_offset = min_file_offset & (1<<CONFIG_PPM_PAGE_SHIFT);
463        first_page_size   = size;
464
465        // get pointer on first page in mapper
466        page = mapper_get_page( mapper , first_page_index );
467
468        if ( page == NULL ) return EINVAL;
469
470        // compute pointer on fragment first byte in mapper
471        map_ptr = (uint8_t *)ppm_page2base( page ) + first_page_offset;
472
473        // compute extended pointers in mapper and in buffer
474        xp_map = XPTR( local_cxy , map_ptr );
475        xp_buf = XPTR( buf_cxy , buf_ptr );
476
477        // move fragment
478        if( to_buffer )
479        {
480            hal_remote_memcpy( xp_buf , xp_map , first_page_size );
481        }
482        else
483        {
484            page_do_dirty( page );
485            hal_remote_memcpy( xp_map , xp_buf , first_page_size );
486        }
487    }
488    else                                        // two pages in mapper
489    {
490        // compute offset and size for first page in mapper
491        first_page_offset = min_file_offset & (1<<CONFIG_PPM_PAGE_SHIFT);
492        first_page_size   = CONFIG_PPM_PAGE_SIZE - first_page_offset;
493
494        // get pointer on first page descriptor in mapper
495        page = mapper_get_page( mapper , first_page_index );
496
497        if ( page == NULL ) return EINVAL;
498
499        // compute local pointer on first byte in first page in mapper
500        map_ptr = (uint8_t *)ppm_page2base(page) + first_page_offset;
501
502        // compute extended pointers
503        xp_map = XPTR( local_cxy , map_ptr );
504        xp_buf = XPTR( buf_cxy , buf_ptr );
505
506        // move fragment to/from first page
507        if( to_buffer )
508        {
509            hal_remote_memcpy( xp_buf , xp_map , first_page_size );
510        }
511        else
512        {
513            page_do_dirty( page );
514            hal_remote_memcpy( xp_map , xp_buf , first_page_size );
515        }
516
517        // compute offset and size for second page in mapper
518        second_page_offset = 0;
519        second_page_size   = size - first_page_size;
520
521        // get pointer on second page in mapper
522        page = mapper_get_page( mapper , second_page_index );
523
524        if ( page == NULL ) return EINVAL;
525
526        // compute local pointer on first byte in second page in mapper
527        map_ptr = (uint8_t *)ppm_page2base( page ) + second_page_offset;
528
529        // compute extended pointers
530        xp_map = XPTR( local_cxy , map_ptr );
531        xp_buf = XPTR( buf_cxy , buf_ptr + first_page_offset );
532
533        // move fragment to/from second page
534        if( to_buffer )
535        {
536            hal_remote_memcpy( xp_buf , xp_map , second_page_size );
537        }
538        else
539        {
540            page_do_dirty( page );
541            hal_remote_memcpy( xp_map , xp_buf , second_page_size );
542        }
543    }
544
545    return 0;
546}  // end mapper_move_one_fragment()
547
548/////////////////////////////////////////////////
549error_t mapper_move_fragments( mapper_t * mapper,
550                               bool_t     read,
551                               uint32_t   nb_frags,
552                               xptr_t     xp_frags )
553{
554        uint32_t     index;
555        error_t      error;
556    fragment_t   local_frags[CONFIG_MAPPER_MAX_FRAGMENTS];   // local copy of fragments array
557    fragment_t * frags_array;                                // pointer on fragments array
558
559    // check nb_frags
560    if( nb_frags > CONFIG_MAPPER_MAX_FRAGMENTS )
561    {
562        printk("\n[PANIC] in %s : number of fragments cannot be larger than %d\n",
563               __FUNCTION__ , CONFIG_MAPPER_MAX_FRAGMENTS );
564        return EINVAL;
565    }
566
567    // get client cluster and local pointer on fragments array
568    cxy_t        client_cxy   = GET_CXY( xp_frags );
569    fragment_t * client_frags = (fragment_t *)GET_PTR( xp_frags );
570
571    if ( local_cxy == client_cxy ) // use the local fragments array if possible
572    {
573        frags_array = client_frags;
574    }
575    else                           // make a local copy of fragments array
576    {
577        hal_remote_memcpy( XPTR( local_cxy , local_frags ) , xp_frags ,
578                           sizeof(fragment_t) * nb_frags );
579        frags_array = local_frags;
580    }
581
582    // loop on fragments
583    for( index = 0 ; index < nb_frags ; index ++ )
584    {
585        error = mapper_move_one_fragment( mapper , read , &frags_array[index] );
586        if ( error ) return error;
587    }
588
589    return 0;
590
591}  // end mapper_move_fragments()
592
593
Note: See TracBrowser for help on using the repository browser.