Context Navigation

source: trunk/kernel/mm/vmm.c @ 303

Last change on this file since 303 was 286, checked in by max@…, 7 years ago
Fix dangerous typos.
File size: 41.4 KB

Rev	Line
[1]	1	/*
	2	* vmm.c - virtual memory manager related operations interface.
	3	*
	4	* Authors Ghassan Almaless (2008,2009,2010,2011, 2012)
	5	* Mohamed Lamine Karaoui (2015)
	6	* Alain Greiner (2016)
[21]	7	*
[1]	8	* Copyright (c) UPMC Sorbonne Universites
	9	*
	10	* This file is part of ALMOS-MKH.
	11	*
	12	* ALMOS-MKH is free software; you can redistribute it and/or modify it
	13	* under the terms of the GNU General Public License as published by
	14	* the Free Software Foundation; version 2.0 of the License.
	15	*
	16	* ALMOS-MKH is distributed in the hope that it will be useful, but
	17	* WITHOUT ANY WARRANTY; without even the implied warranty of
	18	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	19	* General Public License for more details.
	20	*
	21	* You should have received a copy of the GNU General Public License
	22	* along with ALMOS-MKH; if not, write to the Free Software Foundation,
	23	* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
	24	*/
	25
[14]	26	#include <kernel_config.h>
[1]	27	#include <hal_types.h>
	28	#include <hal_special.h>
	29	#include <hal_gpt.h>
	30	#include <printk.h>
[23]	31	#include <memcpy.h>
[1]	32	#include <rwlock.h>
	33	#include <list.h>
	34	#include <bits.h>
	35	#include <process.h>
	36	#include <thread.h>
	37	#include <vseg.h>
	38	#include <cluster.h>
	39	#include <scheduler.h>
	40	#include <vfs.h>
	41	#include <mapper.h>
	42	#include <page.h>
	43	#include <kmem.h>
	44	#include <vmm.h>
	45
	46	//////////////////////////////////////////////////////////////////////////////////
	47	// Extern global variables
	48	//////////////////////////////////////////////////////////////////////////////////
	49
	50	extern process_t process_zero; // defined in cluster.c file
	51
	52
	53	////////////////////////////////////
	54	void vmm_init( process_t * process )
[21]	55	{
[1]	56	error_t error;
	57	vseg_t * vseg_kentry;
	58	vseg_t * vseg_args;
	59	vseg_t * vseg_envs;
	60	vseg_t * vseg_heap;
	61	intptr_t base;
	62	intptr_t size;
	63
[204]	64	vmm_dmsg("\n[INFO] %s : enter for process %x\n", __FUNCTION__ , process->pid );
	65
[1]	66	// get pointer on VMM
	67	vmm_t * vmm = &process->vmm;
	68
[204]	69	assert( ((CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + CONFIG_VMM_ENVS_SIZE)
	70	<= CONFIG_VMM_ELF_BASE) , __FUNCTION__ , "UTILS zone too small\n" );
[21]	71
[204]	72	assert( (CONFIG_THREAD_MAX_PER_CLUSTER <= 32) , __FUNCTION__ ,
	73	"no more than 32 threads per cluster for a single process\n");
[1]	74
[204]	75	assert( ((CONFIG_VMM_STACK_SIZE * CONFIG_THREAD_MAX_PER_CLUSTER) <=
	76	(CONFIG_VMM_VSPACE_SIZE - CONFIG_VMM_STACK_BASE)) , __FUNCTION__ ,
	77	"STACK zone too small\n");
[1]	78
[21]	79	// initialize the rwlock protecting the vsegs list
[1]	80	rwlock_init( &vmm->vsegs_lock );
	81
	82	// initialize local list of vsegs and radix-tree
[23]	83	vmm->vsegs_nr = 0;
[1]	84	list_root_init( &vmm->vsegs_root );
	85	error = grdxt_init( &vmm->grdxt,
	86	CONFIG_VMM_GRDXT_W1,
	87	CONFIG_VMM_GRDXT_W2,
	88	CONFIG_VMM_GRDXT_W3 );
[204]	89
	90	assert( (error == 0) , __FUNCTION__ , "cannot initialize radix tree\n" );
[1]	91
	92	// register kentry vseg in VMM
	93	base = 1 << CONFIG_PPM_PAGE_SHIFT;
	94	size = CONFIG_VMM_KENTRY_SIZE << CONFIG_PPM_PAGE_SHIFT;
	95	vseg_kentry = vmm_create_vseg( process , base , size , VSEG_TYPE_CODE );
[204]	96
	97	assert( (vseg_kentry != NULL) , __FUNCTION__ , "cannot register kentry vseg\n" );
	98
[1]	99	vmm->kent_vpn_base = 1;
	100
	101	// register the args vseg in VMM
	102	base = (CONFIG_VMM_KENTRY_SIZE + 1 )<<CONFIG_PPM_PAGE_SHIFT;
	103	size = CONFIG_VMM_ARGS_SIZE << CONFIG_PPM_PAGE_SHIFT;
	104	vseg_args = vmm_create_vseg( process , base , size , VSEG_TYPE_DATA );
[204]	105
	106	assert( (vseg_args != NULL) , __FUNCTION__ , "cannot register args vseg\n" );
	107
[1]	108	vmm->args_vpn_base = CONFIG_VMM_KENTRY_SIZE + 1;
	109
	110	// register the envs vseg in VMM
	111	base = (CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + 1 )<<CONFIG_PPM_PAGE_SHIFT;
	112	size = CONFIG_VMM_ENVS_SIZE << CONFIG_PPM_PAGE_SHIFT;
	113	vseg_envs = vmm_create_vseg( process , base , size , VSEG_TYPE_DATA );
[204]	114
	115	assert( (vseg_envs != NULL) , __FUNCTION__ , "cannot register envs vseg\n" );
	116
[1]	117	vmm->envs_vpn_base = CONFIG_VMM_KENTRY_SIZE + CONFIG_VMM_ARGS_SIZE + 1;
	118
	119	// register the heap vseg in VMM
	120	base = CONFIG_VMM_HEAP_BASE << CONFIG_PPM_PAGE_SHIFT;
	121	size = (CONFIG_VMM_MMAP_BASE-CONFIG_VMM_HEAP_BASE) << CONFIG_PPM_PAGE_SHIFT;
	122	vseg_heap = vmm_create_vseg( process , base , size , VSEG_TYPE_HEAP );
[204]	123
	124	assert( (vseg_heap != NULL) , __FUNCTION__ , "cannot register heap vseg\n" );
	125
[1]	126	vmm->heap_vpn_base = CONFIG_VMM_HEAP_BASE;
	127
	128	// initialize generic page table
	129	error = hal_gpt_create( &vmm->gpt );
	130
[204]	131	assert( (error == 0) , __FUNCTION__ , "cannot initialize page table\n");
	132
[1]	133	// initialize STACK allocator
	134	vmm->stack_mgr.bitmap = 0;
	135	vmm->stack_mgr.vpn_base = CONFIG_VMM_STACK_BASE;
	136
	137	// initialize MMAP allocator
	138	vmm->mmap_mgr.vpn_base = CONFIG_VMM_MMAP_BASE;
	139	vmm->mmap_mgr.vpn_size = CONFIG_VMM_STACK_BASE - CONFIG_VMM_MMAP_BASE;
	140	vmm->mmap_mgr.first_free_vpn = CONFIG_VMM_MMAP_BASE;
	141	uint32_t i;
	142	for( i = 0 ; i < 32 ; i++ ) list_root_init( &vmm->mmap_mgr.zombi_list[i] );
	143
[21]	144	// initialize instrumentation counters
[1]	145	vmm->pgfault_nr = 0;
	146	vmm->u_err_nr = 0;
	147	vmm->m_err_nr = 0;
	148
[124]	149	hal_fence();
[1]	150
[204]	151	vmm_dmsg("\n[INFO] %s : exit for process %x\n", __FUNCTION__ , process->pid );
	152
	153	} // end vmm_init()
	154
[23]	155	//////////////////////////////////////////
	156	error_t vmm_copy( process_t * dst_process,
	157	process_t * src_process )
	158	{
	159	error_t error;
	160
	161	vmm_t * src_vmm = &src_process->vmm;
	162	vmm_t * dst_vmm = &dst_process->vmm;
	163
	164	// take the src_vmm vsegs_lock
	165	rwlock_wr_lock( &src_vmm->vsegs_lock );
	166
[178]	167	// initialize dst_vmm vsegs_lock
[23]	168	rwlock_init( &dst_vmm->vsegs_lock );
	169
[178]	170	// initialize the dst_vmm vsegs list and the radix tree
[23]	171	dst_vmm->vsegs_nr = 0;
	172	list_root_init( &dst_vmm->vsegs_root );
	173	error = grdxt_init( &dst_vmm->grdxt,
	174	CONFIG_VMM_GRDXT_W1,
	175	CONFIG_VMM_GRDXT_W2,
	176	CONFIG_VMM_GRDXT_W3 );
	177	if( error )
	178	{
[178]	179	printk("\n[ERROR] in %s : cannot initialize radix tree for process %x\n",
[23]	180	__FUNCTION__ , dst_process->pid );
	181	return ENOMEM;
	182	}
	183
	184	// loop on src_vmm list of vsegs to create
	185	// and register vsegs copies in dst_vmm
	186	list_entry_t * iter;
	187	vseg_t * src_vseg;
	188	vseg_t * dst_vseg;
	189	LIST_FOREACH( &src_vmm->vsegs_root , iter )
	190	{
	191	// get pointer on current src_vseg
	192	src_vseg = LIST_ELEMENT( iter , vseg_t , list );
	193
	194	// allocate memory for a new dst_vseg
	195	dst_vseg = vseg_alloc();
	196
[178]	197	if( dst_vseg == NULL )
[23]	198	{
	199	// release all allocated vsegs
	200	LIST_FOREACH( &dst_vmm->vsegs_root , iter )
	201	{
	202	dst_vseg = LIST_ELEMENT( iter , vseg_t , list );
	203	vseg_free( dst_vseg );
	204	}
	205	return ENOMEM;
	206	}
	207
	208	// copy src_vseg to dst_vseg
	209	vseg_init_from_ref( dst_vseg , XPTR( local_cxy , src_vseg ) );
	210
	211	// register dst_vseg in dst_vmm
	212	vseg_attach( dst_vmm , dst_vseg );
	213	}
	214
	215	// release the src_vmm vsegs_lock
	216	rwlock_wr_unlock( &src_vmm->vsegs_lock );
	217
	218	// initialize generic page table
	219	error = hal_gpt_create( &dst_vmm->gpt );
	220
	221	if( error )
	222	{
	223	printk("\n[ERROR] in %s : cannot initialize page table\n", __FUNCTION__ );
	224	return ENOMEM;
	225	}
	226
	227	// initialize STACK allocator
	228	dst_vmm->stack_mgr.bitmap = 0;
	229	dst_vmm->stack_mgr.vpn_base = CONFIG_VMM_STACK_BASE;
	230
	231	// initialize MMAP allocator
	232	dst_vmm->mmap_mgr.vpn_base = CONFIG_VMM_MMAP_BASE;
	233	dst_vmm->mmap_mgr.vpn_size = CONFIG_VMM_STACK_BASE - CONFIG_VMM_MMAP_BASE;
	234	dst_vmm->mmap_mgr.first_free_vpn = CONFIG_VMM_MMAP_BASE;
	235	uint32_t i;
	236	for( i = 0 ; i < 32 ; i++ ) list_root_init( &dst_vmm->mmap_mgr.zombi_list[i] );
	237
[178]	238	// initialize instrumentation counters
[23]	239	dst_vmm->pgfault_nr = 0;
	240	dst_vmm->u_err_nr = 0;
	241	dst_vmm->m_err_nr = 0;
	242
	243	// copy base addresses
[178]	244	dst_vmm->kent_vpn_base = src_vmm->kent_vpn_base;
	245	dst_vmm->args_vpn_base = src_vmm->args_vpn_base;
	246	dst_vmm->envs_vpn_base = src_vmm->envs_vpn_base;
	247	dst_vmm->heap_vpn_base = src_vmm->heap_vpn_base;
	248	dst_vmm->code_vpn_base = src_vmm->code_vpn_base;
	249	dst_vmm->data_vpn_base = src_vmm->data_vpn_base;
[23]	250
	251	dst_vmm->entry_point = src_vmm->entry_point;
	252
	253	// HEAP TODO : new heap for child ???
	254	dst_vmm->heap_vseg = src_vmm->heap_vseg;
	255
	256	// initialize generic page table
	257	error = hal_gpt_create( &dst_vmm->gpt );
	258
	259	if( error )
	260	{
	261	printk("\n[ERROR] in %s : cannot initialize page table\n", __FUNCTION__ );
	262	return ENOMEM;
	263	}
	264
	265	// copy GPT content from src_vmm to dst_vmm, activating "Copy-On-Write"
	266	// TODO register Copy-On_Write in page descriptors
	267	bool_t cow = true;
	268	hal_gpt_copy( &dst_vmm->gpt , &src_vmm->gpt , cow );
	269
[124]	270	hal_fence();
[23]	271
	272	return 0;
	273
[204]	274	} // vmm_copy()
	275
[1]	276	///////////////////////////////////////
	277	void vmm_destroy( process_t * process )
	278	{
	279	vseg_t * vseg;
	280
	281	// get pointer on VMM
	282	vmm_t * vmm = &process->vmm;
	283
	284	// get lock protecting vseg list
	285	rwlock_wr_lock( &vmm->vsegs_lock );
	286
	287	// remove all vsegs registered in vmm
	288	while( !list_is_empty( &vmm->vsegs_root ) )
	289	{
	290	vseg = LIST_FIRST( &vmm->vsegs_root , vseg_t , list );
	291	vseg_detach( vmm , vseg );
	292	vseg_free( vseg );
	293	}
	294
	295	// delete vsegs radix_tree
	296	grdxt_destroy( &vmm->grdxt );
	297
	298	// release lock
	299	rwlock_wr_unlock(&vmm->vsegs_lock);
	300
	301	// remove all vsegs from zombi_lists in MMAP allocator
	302	uint32_t i;
	303	for( i = 0 ; i<32 ; i++ )
	304	{
	305	while( !list_is_empty( &vmm->mmap_mgr.zombi_list[i] ) )
	306	{
	307	vseg = LIST_FIRST( &vmm->mmap_mgr.zombi_list[i] , vseg_t , list );
	308	vseg_detach( vmm , vseg );
	309	vseg_free( vseg );
	310	}
	311	}
	312
	313	// release memory allocated to the local page table
	314	hal_gpt_destroy( &vmm->gpt );
	315
[204]	316	} // end vmm_destroy()
	317
[1]	318	/////////////////////////////////////////////////
	319	vseg_t * vmm_check_conflict( process_t * process,
[21]	320	vpn_t vpn_base,
[1]	321	vpn_t vpn_size )
	322	{
	323	vmm_t * vmm = &process->vmm;
	324	vseg_t * vseg;
	325	list_entry_t * iter;
	326
	327	// scan the list of registered vsegs
	328	LIST_FOREACH( &vmm->vsegs_root , iter )
	329	{
	330	vseg = LIST_ELEMENT( iter , vseg_t , list );
[204]	331
[21]	332	if( ((vpn_base + vpn_size) > vseg->vpn_base) &&
	333	(vpn_base < (vseg->vpn_base + vseg->vpn_size)) ) return vseg;
[1]	334	}
	335	return NULL;
	336
[204]	337	} // end vmm_check_conflict()
	338
[1]	339	////////////////////////////////////////////////////////////////////////////////////////////
	340	// This static function is called by the vmm_create_vseg() function, and implements
	341	// the VMM stack_vseg specific allocator.
	342	////////////////////////////////////////////////////////////////////////////////////////////
	343	// @ vmm : pointer on VMM.
[21]	344	// @ vpn_base : (return value) first allocated page
[1]	345	// @ vpn_size : (return value) number of allocated pages
	346	////////////////////////////////////////////////////////////////////////////////////////////
	347	static error_t vmm_stack_alloc( vmm_t * vmm,
	348	vpn_t * vpn_base,
	349	vpn_t * vpn_size )
	350	{
	351	// get stack allocator pointer
	352	stack_mgr_t * mgr = &vmm->stack_mgr;
	353
	354	// get lock on stack allocator
	355	spinlock_lock( &mgr->lock );
	356
	357	// get first free slot index in bitmap
	358	int32_t index = bitmap_ffc( &mgr->bitmap , 4 );
[179]	359	if( (index < 0) \|\| (index > 31) )
	360	{
	361	spinlock_unlock( &mgr->lock );
	362	return ENOMEM;
	363	}
[1]	364
	365	// update bitmap
	366	bitmap_set( &mgr->bitmap , index );
[21]	367
[1]	368	// release lock on stack allocator
	369	spinlock_unlock( &mgr->lock );
	370
[21]	371	// returns vpn_base, vpn_size (one page non allocated)
[1]	372	vpn_base = mgr->vpn_base + index CONFIG_VMM_STACK_SIZE + 1;
	373	*vpn_size = CONFIG_VMM_STACK_SIZE - 1;
	374	return 0;
	375
[204]	376	} // end vmm_stack_alloc()
	377
[1]	378	////////////////////////////////////////////////////////////////////////////////////////////
	379	// This static function is called by the vmm_create_vseg() function, and implements
	380	// the VMM MMAP specific allocator.
	381	////////////////////////////////////////////////////////////////////////////////////////////
	382	// @ vmm : [in] pointer on VMM.
	383	// @ npages : [in] requested number of pages.
[21]	384	// @ vpn_base : [out] first allocated page.
[1]	385	// @ vpn_size : [out] actual number of allocated pages.
	386	////////////////////////////////////////////////////////////////////////////////////////////
	387	static error_t vmm_mmap_alloc( vmm_t * vmm,
	388	vpn_t npages,
	389	vpn_t * vpn_base,
	390	vpn_t * vpn_size )
	391	{
	392	uint32_t index;
	393	vseg_t * vseg;
	394	vpn_t base;
	395	vpn_t size;
[21]	396	vpn_t free;
[1]	397
[21]	398	// mmap vseg size must be power of 2
[1]	399	// compute actual size and index in zombi_list array
	400	size = POW2_ROUNDUP( npages );
	401	index = bits_log2( size );
	402
	403	// get mmap allocator pointer
	404	mmap_mgr_t * mgr = &vmm->mmap_mgr;
	405
	406	// get lock on mmap allocator
	407	spinlock_lock( &mgr->lock );
	408
	409	// get vseg from zombi_list or from mmap zone
	410	if( list_is_empty( &mgr->zombi_list[index] ) ) // from mmap zone
	411	{
	412	// check overflow
	413	free = mgr->first_free_vpn;
	414	if( (free + size) > mgr->vpn_size ) return ENOMEM;
	415
	416	// update STACK allocator
	417	mgr->first_free_vpn += size;
	418
	419	// compute base
	420	base = free;
	421	}
	422	else // from zombi_list
	423	{
	424	// get pointer on zombi vseg from zombi_list
	425	vseg = LIST_FIRST( &mgr->zombi_list[index] , vseg_t , list );
	426
	427	// remove vseg from free-list
	428	list_unlink( &vseg->list );
	429
	430	// compute base
	431	base = vseg->vpn_base;
[21]	432	}
	433
[1]	434	// release lock on mmap allocator
	435	spinlock_unlock( &mgr->lock );
	436
	437	// returns vpn_base, vpn_size
	438	*vpn_base = base;
	439	*vpn_size = size;
	440	return 0;
	441
[204]	442	} // end vmm_mmap_alloc()
	443
[1]	444	//////////////////////////////////////////////
	445	vseg_t * vmm_create_vseg( process_t * process,
[21]	446	intptr_t base,
	447	intptr_t size,
[1]	448	uint32_t type )
	449	{
	450	vseg_t * vseg; // created vseg pointer
[204]	451	vpn_t vpn_base; // first page index
[1]	452	vpn_t vpn_size; // number of pages
	453	error_t error;
	454
	455	// get pointer on VMM
	456	vmm_t * vmm = &process->vmm;
[21]	457
[204]	458	vmm_dmsg("\n[INFO] %s : enter for process %x / base = %x / size = %x / type = %s\n",
[101]	459	__FUNCTION__ , process->pid , base , size , vseg_type_str(type) );
[21]	460
[204]	461	// compute base, size, vpn_base, vpn_size, depending on vseg type
[1]	462	// we use the VMM specific allocators for STACK and MMAP vsegs
	463	if( type == VSEG_TYPE_STACK )
	464	{
	465	// get vpn_base and vpn_size from STACK allocator
	466	error = vmm_stack_alloc( vmm , &vpn_base , &vpn_size );
	467	if( error )
	468	{
	469	printk("\n[ERROR] in %s : no vspace for stack vseg / process %x in cluster %x\n",
	470	__FUNCTION__ , process->pid , local_cxy );
	471	return NULL;
	472	}
	473
	474	// compute vseg base and size from vpn_base and vpn_size
	475	base = vpn_base << CONFIG_PPM_PAGE_SHIFT;
	476	size = vpn_size << CONFIG_PPM_PAGE_SHIFT;
	477	}
[21]	478	else if( (type == VSEG_TYPE_ANON) \|\|
	479	(type == VSEG_TYPE_FILE) \|\|
[1]	480	(type == VSEG_TYPE_REMOTE) )
	481	{
	482	// get vpn_base and vpn_size from MMAP allocator
	483	vpn_t npages = size >> CONFIG_PPM_PAGE_SHIFT;
	484	error = vmm_mmap_alloc( vmm , npages , &vpn_base , &vpn_size );
	485	if( error )
	486	{
	487	printk("\n[ERROR] in %s : no vspace for mmap vseg / process %x in cluster %x\n",
	488	__FUNCTION__ , process->pid , local_cxy );
	489	return NULL;
	490	}
	491
	492	// compute vseg base and size from vpn_base and vpn_size
	493	base = vpn_base << CONFIG_PPM_PAGE_SHIFT;
	494	size = vpn_size << CONFIG_PPM_PAGE_SHIFT;
	495	}
	496	else
	497	{
[204]	498	uint32_t vpn_min = base >> CONFIG_PPM_PAGE_SHIFT;
	499	uint32_t vpn_max = (base + size - 1) >> CONFIG_PPM_PAGE_SHIFT;
	500
	501	vpn_base = vpn_min;
	502	vpn_size = vpn_max - vpn_min + 1;
[1]	503	}
	504
	505	// check collisions
	506	vseg = vmm_check_conflict( process , vpn_base , vpn_size );
	507	if( vseg != NULL )
	508	{
[21]	509	printk("\n[ERROR] in %s for process %x : new vseg [vpn_base = %x / vpn_size = %x]\n"
[1]	510	" overlap existing vseg [vpn_base = %x / vpn_size = %x]\n",
[21]	511	__FUNCTION__ , process->pid, vpn_base, vpn_size,
	512	vseg->vpn_base, vseg->vpn_size );
[1]	513	return NULL;
	514	}
	515
	516	// allocate physical memory for vseg descriptor
	517	vseg = vseg_alloc();
	518	if( vseg == NULL )
	519	{
	520	printk("\n[ERROR] in %s for process %x : cannot allocate memory for vseg\n",
	521	__FUNCTION__ , process->pid );
	522	return NULL;
	523	}
	524
	525	// initialize vseg descriptor
	526	vseg_init( vseg , base, size , vpn_base , vpn_size , type , local_cxy , 0 , 0 );
	527
	528	// update "heap_vseg" in VMM
	529	process->vmm.heap_vseg = vseg;
	530
	531	// attach vseg to vmm
	532	rwlock_wr_lock( &vmm->vsegs_lock );
	533	vseg_attach( vmm , vseg );
	534	rwlock_wr_unlock( &vmm->vsegs_lock );
	535
[204]	536	vmm_dmsg("\n[INFO] %s : exit for process %x / vseg [%x, %x] has been mapped\n",
[1]	537	__FUNCTION__ , process->pid , vseg->min , vseg->max );
[21]	538
[1]	539	return vseg;
[178]	540	}
[1]	541
	542	/////////////////////////////////////
	543	void vmm_remove_vseg( vseg_t * vseg )
	544	{
	545	// get pointers on calling process and VMM
	546	thread_t * this = CURRENT_THREAD;
	547	process_t * process = this->process;
	548	vmm_t * vmm = &this->process->vmm;
	549	uint32_t type = vseg->type;
	550
	551	// detach vseg from VMM
	552	rwlock_wr_lock( &vmm->vsegs_lock );
	553	vseg_detach( &process->vmm , vseg );
	554	rwlock_wr_unlock( &vmm->vsegs_lock );
	555
	556	// release the stack slot to VMM stack allocator if STACK type
	557	if( type == VSEG_TYPE_STACK )
	558	{
	559	// get pointer on stack allocator
	560	stack_mgr_t * mgr = &vmm->stack_mgr;
	561
	562	// compute slot index
	563	uint32_t index = ((vseg->vpn_base - mgr->vpn_base - 1) / CONFIG_VMM_STACK_SIZE);
	564
	565	// update stacks_bitmap
	566	spinlock_lock( &mgr->lock );
	567	bitmap_clear( &mgr->bitmap , index );
	568	spinlock_unlock( &mgr->lock );
	569	}
	570
	571	// release the vseg to VMM mmap allocator if MMAP type
	572	if( (type == VSEG_TYPE_ANON) \|\| (type == VSEG_TYPE_FILE) \|\| (type == VSEG_TYPE_REMOTE) )
	573	{
	574	// get pointer on mmap allocator
	575	mmap_mgr_t * mgr = &vmm->mmap_mgr;
	576
	577	// compute zombi_list index
	578	uint32_t index = bits_log2( vseg->vpn_size );
	579
	580	// update zombi_list
	581	spinlock_lock( &mgr->lock );
	582	list_add_first( &mgr->zombi_list[index] , &vseg->list );
	583	spinlock_unlock( &mgr->lock );
	584	}
	585
	586	// release physical memory allocated for vseg descriptor if no MMAP type
	587	if( (type != VSEG_TYPE_ANON) && (type != VSEG_TYPE_FILE) && (type != VSEG_TYPE_REMOTE) )
	588	{
	589	vseg_free( vseg );
	590	}
[178]	591	}
[1]	592
[68]	593	//////////////////////////////////////////////
	594	error_t vmm_map_kernel_vseg( vseg_t * vseg,
	595	uint32_t attr )
[1]	596	{
	597	vpn_t vpn; // VPN of PTE to be set
	598	vpn_t vpn_min; // VPN of first PTE to be set
	599	vpn_t vpn_max; // VPN of last PTE to be set (excluded)
	600	ppn_t ppn; // PPN of allocated physical page
	601	uint32_t order; // ln( number of small pages for one single PTE )
	602	page_t * page;
	603	error_t error;
	604
[68]	605	// check vseg type : must be a kernel vseg
[1]	606	uint32_t type = vseg->type;
[68]	607	assert( ((type==VSEG_TYPE_KCODE) \|\| (type==VSEG_TYPE_KDATA) \|\| (type==VSEG_TYPE_KDEV)),
	608	__FUNCTION__ , "not a kernel vseg\n" );
[1]	609
	610	// get pointer on page table
	611	gpt_t * gpt = &process_zero.vmm.gpt;
	612
[21]	613	// define number of small pages per PTE
[1]	614	if( attr & GPT_SMALL ) order = 0; // 1 small page
	615	else order = 9; // 512 small pages
	616
	617	// loop on pages in vseg
	618	vpn_min = vseg->vpn_base;
	619	vpn_max = vpn_min + vseg->vpn_size;
	620	for( vpn = vpn_min ; vpn < vpn_max ; vpn++ )
	621	{
[68]	622	// allocate a physical page from local PPM
[1]	623	kmem_req_t req;
	624	req.type = KMEM_PAGE;
	625	req.size = order;
	626	req.flags = AF_KERNEL \| AF_ZERO;
	627	page = (page_t *)kmem_alloc( &req );
[21]	628	if( page == NULL )
[1]	629	{
	630	printk("\n[ERROR] in %s : cannot allocate physical memory\n", __FUNCTION__ );
	631	return ENOMEM;
	632	}
	633
	634	// set page table entry
	635	ppn = ppm_page2ppn( page );
	636	error = hal_gpt_set_pte( gpt , vpn , ppn , attr );
[21]	637	if( error )
[1]	638	{
	639	printk("\n[ERROR] in %s : cannot register PPE\n", __FUNCTION__ );
	640	return ENOMEM;
	641	}
	642	}
[21]	643
[1]	644	return 0;
[178]	645	}
[1]	646
	647	/////////////////////////////////////////
	648	void vmm_unmap_vseg( process_t * process,
	649	vseg_t * vseg )
	650	{
[21]	651	vpn_t vpn; // VPN of current PTE
	652	vpn_t vpn_min; // VPN of first PTE
[1]	653	vpn_t vpn_max; // VPN of last PTE (excluded)
	654
	655	// get pointer on process page table
	656	gpt_t * gpt = &process->vmm.gpt;
	657
	658	// loop on pages in vseg
	659	vpn_min = vseg->vpn_base;
	660	vpn_max = vpn_min + vseg->vpn_size;
	661	for( vpn = vpn_min ; vpn < vpn_max ; vpn++ )
	662	{
	663	hal_gpt_reset_pte( gpt , vpn );
	664	}
[178]	665	}
[1]	666
	667	/////////////////////////////////////////////
	668	error_t vmm_resize_vseg( process_t * process,
	669	intptr_t base,
	670	intptr_t size )
	671	{
	672	error_t error;
	673
	674	// get pointer on process VMM
	675	vmm_t * vmm = &process->vmm;
	676
	677	intptr_t addr_min = base;
	678	intptr_t addr_max = base + size;
[21]	679	uint32_t shift = CONFIG_PPM_PAGE_SHIFT;
[1]	680
	681	// get pointer on vseg
	682	vseg_t * vseg = grdxt_lookup( &vmm->grdxt , (uint32_t)(base >> shift) );
	683
	684	if( vseg == NULL) return EINVAL;
[21]	685
[1]	686	// get VMM lock protecting vsegs list
	687	rwlock_wr_lock( &vmm->vsegs_lock );
[21]	688
[1]	689	if( (vseg->min > addr_min) \|\| (vseg->max < addr_max) ) // region not included in vseg
	690	{
	691	error = EINVAL;
	692	}
	693	else if( (vseg->min == addr_min) && (vseg->max == addr_max) ) // vseg must be removed
	694	{
	695	vmm_remove_vseg( vseg );
	696	error = 0;
	697	}
	698	else if( vseg->min == addr_min ) // vseg must be resized
	699	{
	700	printk("\n[PANIC] in %s : resize not implemented yet\n", __FUNCTION__ );
	701	hal_core_sleep();
	702	error = 0;
	703	}
	704	else if( vseg->max == addr_max ) // vseg must be resized
	705	{
	706	printk("\n[PANIC] in %s : resize not implemented yet\n", __FUNCTION__ );
	707	hal_core_sleep();
	708	error = 0;
	709	}
	710	else // vseg cut in three regions => vseg must be resized & new vseg created
	711	{
	712	printk("\n[PANIC] in %s : resize not implemented yet\n", __FUNCTION__ );
	713	hal_core_sleep();
	714	error = 0;
	715	}
	716
	717	// release VMM lock
	718	rwlock_wr_unlock( &vmm->vsegs_lock );
	719
	720	return error;
[178]	721	}
[1]	722
	723	///////////////////////////////////////////
	724	vseg_t * vmm_get_vseg( process_t * process,
	725	intptr_t vaddr )
	726	{
	727
	728	// get pointer on process VMM
	729	vmm_t * vmm = &process->vmm;
	730
	731	// get lock protecting the vseg list
	732	rwlock_rd_lock( &vmm->vsegs_lock );
	733
	734	// get pointer on vseg from radix tree
	735	vseg_t * vseg = grdxt_lookup( &vmm->grdxt, (uint32_t)(vaddr >> CONFIG_PPM_PAGE_SHIFT) );
	736
	737	// release the lock
	738	rwlock_rd_unlock( &vmm->vsegs_lock );
	739
	740	return vseg;
[178]	741	}
[1]	742
	743	/////////////////////////////////////////
	744	error_t vmm_get_pte( process_t * process,
	745	vpn_t vpn,
	746	uint32_t * ret_attr,
	747	ppn_t * ret_ppn )
	748	{
	749	vseg_t * vseg; // pointer on vseg containing VPN
	750	ppn_t ppn; // PPN from GPT entry
	751	uint32_t attr; // attributes from GPT entry
	752	error_t error;
	753
[178]	754	// this function must be called by a thread running in the reference cluster
[68]	755	assert( (GET_CXY( process->ref_xp ) == local_cxy ) , __FUNCTION__ ,
	756	" not called in the reference cluster\n" );
[1]	757
	758	// get VMM pointer
	759	vmm_t * vmm = &process->vmm;
	760
	761	// access GPT to get PTE attributes and PPN
	762	hal_gpt_get_pte( &vmm->gpt , vpn , &attr , &ppn );
	763
	764	// if PTE unmapped => allocate one small physical page to map it
[21]	765	if( (attr & GPT_MAPPED) == 0 )
[1]	766	{
	767	// get vseg pointer
	768	vseg = vmm_get_vseg( process , vpn<<CONFIG_PPM_PAGE_SHIFT );
	769
[286]	770	if( vseg == NULL )
[1]	771	{
	772	printk("\n[ERROR] in %s : out of segment / process = %x / vpn = %x\n",
	773	__FUNCTION__ , process->pid , vpn );
	774	return EINVAL;
	775	}
	776
	777	// select the target cluster for physical mapping
	778	uint32_t target_cxy;
	779	if( vseg->flags & VSEG_DISTRIB ) // depends on VPN LSB
	780	{
	781	uint32_t x_width = LOCAL_CLUSTER->x_width;
	782	uint32_t y_width = LOCAL_CLUSTER->y_width;
	783	target_cxy = vpn & ((1<<(x_width + y_width)) - 1);
	784	}
	785	else // defined in vseg descriptor
	786	{
	787	target_cxy = vseg->cxy;
	788	}
	789
	790	// allocate memory for page fault
	791	kmem_req_t req;
	792	page_t * page;
	793	if( target_cxy == local_cxy ) // target cluster is the local cluster
[21]	794	{
[1]	795	req.type = KMEM_PAGE;
	796	req.size = 0;
	797	req.flags = AF_NONE;
	798	page = (page_t *)kmem_alloc( &req );
	799
	800	error = ( page == NULL ) ? 1 : 0;
	801	ppn = ppm_page2ppn( page );
	802	}
	803	else // target cluster is not the local cluster
	804	{
	805	rpc_pmem_get_pages_client( target_cxy , 0 , &error , &ppn );
	806	}
	807
	808	if( error )
	809	{
	810	printk("\n[ERROR] in %s : cannot allocate memory / process = %x / vpn = %x\n",
	811	__FUNCTION__ , process->pid , vpn );
	812	return ENOMEM;
	813	}
	814
[21]	815	// define GPT attributes from vseg flags
[1]	816	attr = GPT_MAPPED \| GPT_SMALL;
	817	if( vseg->flags & VSEG_USER ) attr \|= GPT_USER;
	818	if( vseg->flags & VSEG_WRITE ) attr \|= GPT_WRITABLE;
	819	if( vseg->flags & VSEG_EXEC ) attr \|= GPT_EXECUTABLE;
	820	if( vseg->flags & VSEG_CACHE ) attr \|= GPT_CACHABLE;
	821
	822	// set the missing PTE in local VMM
	823	error = hal_gpt_set_pte( &vmm->gpt , vpn , ppn , attr );
	824	if( error )
	825	{
	826	printk("\n[ERROR] in %s : cannot register PTE / process = %x / vpn = %x\n",
	827	__FUNCTION__ , process->pid , vpn );
	828	return ENOMEM;
	829	}
	830	}
	831
	832	*ret_ppn = ppn;
[21]	833	*ret_attr = attr;
[1]	834	return 0;
[178]	835	}
[1]	836
	837	///////////////////////////////////////////////////
	838	error_t vmm_handle_page_fault( process_t * process,
	839	vseg_t * vseg,
	840	vpn_t vpn )
	841	{
	842	uint32_t attr; // missing page attributes
[21]	843	ppn_t ppn; // missing page PPN
[1]	844	error_t error; // return value
	845
	846	// get local VMM pointer
	847	vmm_t * vmm = &process->vmm;
	848
	849	// get reference process cluster and local pointer
	850	cxy_t ref_cxy = GET_CXY( process->ref_xp );
	851	process_t * ref_ptr = (process_t *)GET_PTR( process->ref_xp );
	852
	853	// get missing PTE attributes and PPN
	854	if( local_cxy != ref_cxy ) // local cluster is not the reference cluster
	855	{
	856	rpc_vmm_get_pte_client( ref_cxy , ref_ptr , vpn , &attr , &ppn , &error );
	857	}
[21]	858	else // local cluster is the reference cluster
[1]	859	{
	860	error = vmm_get_pte( process , vpn , &attr , &ppn );
	861	}
	862
	863	// check page allocation error
	864	if( error )
	865	{
	866	printk("\n[ERROR] in %s : cannot allocate memory / process = %x / vpn = %x\n",
[21]	867	__FUNCTION__ , process->pid , vpn );
[1]	868	return ENOMEM;
	869	}
	870
	871	// set the missing PTE in local VMM
	872	error = hal_gpt_set_pte( &vmm->gpt , vpn , attr , ppn );
	873	if( error )
	874	{
	875	printk("\n[ERROR] in %s : cannot register PTE / process = %x / vpn = %x\n",
[21]	876	__FUNCTION__ , process->pid , vpn );
[1]	877	return ENOMEM;
	878	}
[21]	879
[1]	880	return 0;
[178]	881	}
[1]	882
	883	///////////////////////////////////////////
	884	error_t vmm_v2p_translate( bool_t ident,
	885	void * ptr,
	886	paddr_t * paddr )
	887	{
[23]	888	process_t * process = CURRENT_THREAD->process;
[1]	889
	890	if( ident ) // identity mapping
	891	{
[23]	892	*paddr = (paddr_t)PADDR( local_cxy , (lpa_t)ptr );
[1]	893	return 0;
	894	}
	895
[21]	896	// access page table
[1]	897	error_t error;
	898	vpn_t vpn;
	899	uint32_t attr;
	900	ppn_t ppn;
	901	uint32_t offset;
	902
[23]	903	vpn = (vpn_t)( (intptr_t)ptr >> CONFIG_PPM_PAGE_SHIFT );
	904	offset = (uint32_t)( ((intptr_t)ptr) & CONFIG_PPM_PAGE_MASK );
[1]	905
[50]	906	if( local_cxy == GET_CXY( process->ref_xp) ) // calling process is reference process
[1]	907	{
	908	error = vmm_get_pte( process, vpn , &attr , &ppn );
	909	}
[50]	910	else // calling process is not reference process
[1]	911	{
	912	cxy_t ref_cxy = GET_CXY( process->ref_xp );
	913	process_t * ref_ptr = (process_t *)GET_PTR( process->ref_xp );
	914	rpc_vmm_get_pte_client( ref_cxy , ref_ptr , vpn , &attr , &ppn , &error );
	915	}
	916
[23]	917	// set paddr
[1]	918	*paddr = (((paddr_t)ppn) << CONFIG_PPM_PAGE_SHIFT) \| offset;
[21]	919
[23]	920	return error;
[178]	921	}
[23]	922
[1]	923	/*
	924
	925	///////////////////////////////////////////////////////////////////
	926	///////////////////////////////////////////////////////////////////
	927	error_t vmm_inval_shared_page( vseg_t *vseg, vma_t vaddr, ppn_t ppn)
	928	{
	929	pmm_page_info_t current;
	930	error_t err;
	931
	932	error= pmm_get_page(&vseg->vmm->pmm, vaddr, &current);
	933
	934	if((err) \|\| (current.ppn != ppn))
	935	goto ended;
	936
	937	current.ppn = 0;
	938	current.attr = 0;
	939	current.cluster = NULL;
	940
	941	error= pmm_set_page(&vseg->vmm->pmm, vaddr, &current);
	942
	943	ended:
	944	return err;
	945	}
	946
	947	error_t vmm_update_shared_page( vseg_t *vseg, vma_t vaddr, ppn_t ppn)
	948	{
	949	pmm_page_info_t current;
	950	error_t err;
	951
	952	error= pmm_get_page(&vseg->vmm->pmm, vaddr, &current);
	953
	954	if((err) \|\| (current.attr != 0))
	955	goto ended;
	956
	957	current.ppn = ppn;
	958	current.attr = vseg->vm_pgprot;
[178]	959	current.cluster = NULL; // this function is called after invalidate one
[1]	960
	961	error= pmm_set_page(&vseg->vmm->pmm, vaddr , &current);
	962
	963	ended:
	964	return err;
	965	}
	966
[178]	967	// Hypothesis: the vseg is shared-anon, mapper list is rdlocked, page is locked
[1]	968	error_t vmm_migrate_shared_page_seq( vseg_t vseg, struct page_s page, struct page_s **new)
	969	{
	970	register vseg_t *reg;
	971	register struct process_s *process;
	972	register struct process_s *this_process;
	973	struct page_s *new_pg;
	974	struct list_entry *iter;
	975	kmem_req_t req;
	976	vma_t vaddr;
	977	ppn_t ppn;
	978	error_t err;
	979
	980	vaddr = (page->index << PMM_PAGE_SHIFT) + vseg->vm_start + vseg->vm_offset;
	981	ppn = ppm_page2ppn(page);
	982	this_process = (new == NULL) ? NULL : current_process;
	983	iter = &vseg->vm_shared_list;
	984	error = ECANCELED;
	985
[178]	986	// Invalidate All
[1]	987	do
	988	{
	989	reg = list_element(iter, vseg_t, vm_shared_list);
	990
	991	process = vmm_get_process(reg->vmm);
	992
	993	if(process != this_process)
	994	{
	995	error= vmm_inval_shared_page(reg, vaddr, ppn);
	996
	997	if(err) goto fail_inval;
	998	}
	999
	1000	assert(vseg->vm_mapper.m_home_cid == current_cid);
	1001	iter = list_next(&vseg->vm_mapper.m_reg_root, iter);
	1002
	1003	}while(iter != NULL);
	1004
	1005	req.type = KMEM_PAGE;
	1006	req.size = 0;
	1007	req.excep_code = AF_USER;
	1008
	1009	new_pg = kmem_alloc(&req);
	1010	*new = new_pg;
	1011
	1012	if(new_pg == NULL)
	1013	{
	1014	error= ENOMEM;
	1015	goto fail_alloc;
	1016	}
	1017
	1018	page_copy(new_pg, page);
	1019
	1020	page_lock(new_pg);
	1021
	1022	new_pg->mapper = page->mapper;
	1023	new_pg->index = page->index;
	1024
[178]	1025	// TODO: do the complet job regading dirty page
[1]	1026	if(PAGE_IS(page, PG_DIRTY))
	1027	PAGE_SET(new_pg, PG_DIRTY);
	1028
	1029	ppn = ppm_page2ppn(new_pg);
	1030	iter = &vseg->vm_shared_list;
	1031
	1032	// Update All
	1033	do
	1034	{
	1035	reg = list_element(iter, vseg_t, vm_shared_list);
	1036
	1037	process = vmm_get_process(reg->vmm);
	1038
	1039	if(process != this_process)
	1040	(void) vmm_update_shared_page(reg, vaddr, ppn);
	1041
	1042	assert(vseg->vm_mapper.m_home_cid == current_cid);
	1043	iter = list_next(&vseg->vm_mapper.m_reg_root, iter);
	1044
	1045
	1046	}while(iter != NULL);
	1047
	1048	page_unlock(new_pg);
	1049
	1050	fail_alloc:
	1051	fail_inval:
	1052	return err;
	1053	}
	1054
	1055	//TODO: revisit all manipulation of the page->refcount
	1056	///////////////////////////////////////////////////////////////
[178]	1057	static inline error_t vmm_do_migrate( vseg_t * vseg,
[1]	1058	pmm_page_info_t * pinfo,
	1059	uint32_t vaddr )
	1060	{
	1061	kmem_req_t req;
	1062	pmm_page_info_t current;
	1063	page_t * newpage;
	1064	cluster_t * cluster;
	1065	thread_t * this;
	1066	error_t err;
	1067	ppn_t ppn;
[178]	1068
[1]	1069	assert( pinfo->ppn != 0 );
	1070
	1071	ppn = pinfo->ppn;
	1072	this = current_thread;
	1073	newpage = NULL;
	1074	cluster = current_cluster;
[178]	1075
[1]	1076	current.attr = 0;
	1077	current.ppn = 0;
	1078
	1079	error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &current);
	1080
[178]	1081	if(error\|\| (current.isAtomic == false) \|\|
[1]	1082	(current.ppn != ppn) \|\| !(current.attr & PMM_MIGRATE))
	1083	{
	1084	#if CONFIG_SHOW_SPURIOUS_PGFAULT
[178]	1085	printk(INFO, "%s: pid %d, tid %d, cpu %d, nothing to do for vaddr %x\n",
	1086	__FUNCTION__,
	1087	this->process->pid,
	1088	this->info.order,
[1]	1089	cpu_get_id(),
	1090	vaddr);
	1091	#endif
	1092	this->info.spurious_pgfault_cntr ++;
	1093	pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
	1094	pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
	1095	return 0;
	1096	}
[178]	1097
[1]	1098	if(!ppn_is_local(ppn))
	1099	{
	1100	req.type = KMEM_PAGE;
	1101	req.size = 0;
	1102	req.excep_code = AF_PGFAULT;
	1103
	1104	newpage = kmem_alloc(&req);
	1105
	1106	if(newpage)
	1107	{
	1108	newpage->mapper = NULL;//?
	1109	ppn_copy(ppm_page2ppn(newpage), ppn);
	1110
	1111	if(current.attr & PMM_COW)
	1112	{
	1113	current.attr \|= PMM_WRITE;
	1114	current.attr &= ~(PMM_COW);
	1115	}
	1116
	1117	current.ppn = ppm_page2ppn(newpage);
	1118	}
	1119	}
	1120
	1121	current.attr \|= PMM_PRESENT;
	1122	current.attr &= ~(PMM_MIGRATE);
	1123	current.attr &= ~(PMM_LOCKED);
	1124	current.cluster = NULL;
	1125
	1126	//also unlock the table entry
	1127	error= pmm_set_page(&vseg->vmm->pmm, vaddr, &current);
	1128
	1129	if(err)
	1130	{
[178]	1131	// TODO: we should differ the kmem_free call
[1]	1132	//page_unlock(page);
	1133	(void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
	1134	req.ptr = newpage;
	1135	kmem_free(&req);
	1136	return err;
[178]	1137	}
[1]	1138
	1139
	1140	if(newpage)
	1141	{
	1142	ppn_refcount_down(ppn);
	1143	current_thread->info.remote_pages_cntr ++;
[178]	1144	#if CONFIG_SHOW_REMOTE_PGALLOC
[1]	1145	printk(INFO, "%s: pid %d, tid %x, cpu %d, cid %d: got new remote page from cluster %d (vaddr %x)\n",
	1146	__FUNCTION__,
	1147	current_process->pid,
	1148	current_thread,
	1149	cpu_get_id(),
	1150	cluster->id,
	1151	newpage->cid,
	1152	vaddr);
	1153	#endif
	1154	}
	1155
	1156	#if CONFIG_SHOW_VMMMGRT_MSG
	1157	printk(INFO, "%s: pid %d, tid %d, cpu %d: Asked to migrate page (vaddr %x) from cluster %d to cluster %d, error%d\n",
	1158	__FUNCTION__,
	1159	current_process->pid,
	1160	current_thread->info.order,
	1161	cpu_get_id(),
	1162	vaddr,
	1163	ppn_ppn2cid(ppn),
	1164	cluster->id,
	1165	err);
	1166	#endif
	1167
	1168	return err;
	1169	}
	1170
	1171	error_t vmm_do_cow( vseg_t vseg, pmm_page_info_t pinfo, uint32_t vaddr)
	1172	{
	1173	register struct page_s *newpage;
	1174	register struct page_s *page;
	1175	register struct thread_s *this;
	1176	register error_t err;
	1177	register uint32_t count;
	1178	register bool_t isCountDown;
	1179	pmm_page_info_t old;
	1180	pmm_page_info_t new;
	1181	kmem_req_t req;
	1182
	1183	this = current_thread;
	1184	old.attr = 0;
	1185	newpage = NULL;
	1186	isCountDown = true;
	1187
	1188	vmm_dmsg(2,"%s: pid %d, tid %d, cpu %d, vaddr %x\n",
	1189	__FUNCTION__,
	1190	this->process->pid,
	1191	this->info.order,
	1192	cpu_get_id(),
	1193	vaddr);
	1194
[178]	1195
[1]	1196	error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &old);
	1197
	1198	//TODO: check this condition
	1199	if(error\|\| (old.isAtomic == false) \|\| !(old.attr & PMM_COW))
	1200	{
	1201	#if CONFIG_SHOW_SPURIOUS_PGFAULT
[178]	1202	printk(INFO, "%s: pid %d, tid %d, cpu %d, nothing to do for vaddr %x\n",
	1203	__FUNCTION__,
	1204	this->process->pid,
	1205	this->info.order,
[1]	1206	cpu_get_id(),
	1207	vaddr);
	1208	#endif
	1209	this->info.spurious_pgfault_cntr ++;
	1210	pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
	1211	pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
	1212	return err;
	1213	//goto VMM_COW_END;
	1214	}
	1215
[178]	1216	//if the ppn is local and the others (processus with wich we share the page)
[1]	1217	//has done cow, then use the old.ppn directly
	1218	if(ppn_is_local(old.ppn))
	1219	{
	1220	page = ppm_ppn2page(&current_cluster->ppm, old.ppn);
	1221
	1222	if(page->mapper == NULL)
	1223	{
	1224	count = page_refcount_get(page);
	1225	if(count == 1)
	1226	{
	1227	newpage = page;//don't copy the page. use it directly.
	1228	isCountDown = false;
[178]	1229	vmm_dmsg(2, "%s: pid %d, tid %d, cpu %d, reuse same page for vaddr %x, pg_addr %x\n",
	1230	__FUNCTION__,
	1231	this->process->pid,
	1232	this->info.order,
	1233	cpu_get_id(),
	1234	vaddr,
[1]	1235	ppm_page2addr(page));
	1236	}
	1237	}
	1238	//else: we need to do the cow even if it's local!
	1239
	1240	}
	1241
	1242	//else: alocate newpage and copy the data from the remote node
	1243	//also defcount down the ppn
	1244	if(newpage == NULL)
[178]	1245	{
[1]	1246	req.type = KMEM_PAGE;
	1247	req.size = 0;
	1248	req.excep_code = AF_PGFAULT;
	1249
	1250	if((newpage = kmem_alloc(&req)) == NULL)
	1251	{
	1252	(void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
	1253	return ENOMEM;
	1254	}
	1255
	1256	newpage->mapper = NULL;
	1257
	1258	ppn_copy(ppm_page2ppn(newpage), old.ppn);
	1259	assert(isCountDown);
	1260
[178]	1261	vmm_dmsg(2,
	1262	"%s: pid %d, tid %d, cpu %d, newpage for vaddr %x, pg_addr %x\n",
	1263	__FUNCTION__,
	1264	this->process->pid,
	1265	this->info.order,
	1266	cpu_get_id(),
	1267	vaddr,
[1]	1268	ppm_page2addr(newpage));
	1269
	1270	if(newpage->cid != current_cid)
	1271	this->info.remote_pages_cntr ++;
	1272	}
	1273
	1274	new.attr = vseg->vm_pgprot \| PMM_WRITE;
	1275	new.attr &= ~(PMM_COW \| PMM_MIGRATE);
	1276	new.ppn = ppm_page2ppn(newpage);
	1277	new.cluster = NULL;
	1278
	1279	//this also unlock the table entry (if no error)
	1280	error= pmm_set_page(&vseg->vmm->pmm, vaddr, &new);
	1281
	1282	if(err)
	1283	{
	1284	(void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
	1285	req.ptr = newpage;
	1286	kmem_free(&req);
	1287	vmm_dmsg(3, "%s: ended [ error%d ]\n", __FUNCTION__, err);
	1288	return err;
	1289	}
	1290
	1291	if(isCountDown) ppn_refcount_down(old.ppn);
	1292
[178]	1293	vmm_dmsg(2, "%s, pid %d, tid %d, cpu %d, COW ended [vaddr %x]\n",
	1294	__FUNCTION__,
[1]	1295	this->process->pid,
	1296	this->info.order,
	1297	cpu_get_id(),
	1298	vaddr);
[178]	1299
[1]	1300	return 0;
	1301	}
	1302
	1303
	1304	//refcount is taken on the file at mmap
	1305	static inline error_t vmm_do_mapped( vseg_t *vseg, uint32_t vaddr, uint32_t excep_code)
	1306	{
	1307	ppn_t ppn;
	1308	error_t err;
	1309	uint32_t index;
	1310	bool_t isDone;
	1311	pmm_page_info_t info;
	1312	pmm_page_info_t current;
	1313	struct thread_s *this;
	1314
	1315	this = current_thread;
	1316
	1317	current.attr = 1;
	1318	current.ppn = 1;
	1319	isDone = false;
	1320
	1321	error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &current);
	1322
	1323	if(err) return err;
	1324
	1325	if((current.isAtomic == false) \|\| (current.attr != 0))
	1326	{
	1327	#if CONFIG_SHOW_SPURIOUS_PGFAULT
[178]	1328	printk(INFO, "%s: pid %d, tid %d, cpu %d, nothing to do for vaddr %x\n",
	1329	__FUNCTION__,
	1330	this->process->pid,
	1331	this->info.order,
[1]	1332	cpu_get_id(),
	1333	vaddr);
	1334	#endif
	1335	this->info.spurious_pgfault_cntr ++;
	1336	pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
	1337	return 0;
	1338	}
	1339
	1340	index = ((vaddr - vseg->vm_start) + vseg->vm_offset) >> PMM_PAGE_SHIFT;
	1341
	1342	//also hold a refcount!
[178]	1343	ppn = mapper_get_ppn(&vseg->vm_mapper,
	1344	index,
[1]	1345	MAPPER_SYNC_OP);
	1346
	1347	if(!ppn)
	1348	{
	1349	error= pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
	1350	assert(!err); //FIXME: liberate the ppn ...
	1351	return (VFS_FILE_IS_NULL(vseg->vm_file)) ? EIO : ENOMEM;
	1352	}
	1353
	1354	info.attr = vseg->vm_pgprot;
	1355	info.ppn = ppn;
	1356	info.cluster = NULL;
	1357
	1358	//also unlock the page
	1359	error= pmm_set_page(&vseg->vmm->pmm, vaddr, &info);
	1360
	1361	assert(!err);//FIXME: liberate the ppn and unlock the table entry ...
	1362	//error= pmm_unlock_page(&vseg->vmm->pmm, vaddr, &current);
	1363
	1364	return err;
	1365	}
	1366
	1367	/////////////////////////////////////////////////////
	1368	static inline error_t vmm_do_aod( vseg_t *vseg, uint32_t vaddr)
	1369	{
	1370	register error_t err;
	1371	register struct page_s *page;
	1372	register struct cluster_s *cluster;
	1373	struct thread_s *this;
	1374	pmm_page_info_t old;
	1375	pmm_page_info_t new;
	1376	kmem_req_t req;
	1377
	1378	page = NULL;
	1379	old.attr = 0;
	1380	this = current_thread;
[178]	1381
[1]	1382	error= pmm_lock_page(&vseg->vmm->pmm, vaddr, &old);
	1383
	1384	if(err) return err;
	1385
	1386	if(old.isAtomic == false)
	1387	{
	1388	this->info.spurious_pgfault_cntr ++;
	1389	pmm_tlb_flush_vaddr(vaddr, PMM_DATA);
	1390	return 0;
	1391	}
	1392
	1393	req.type = KMEM_PAGE;
	1394	req.size = 0;
	1395	req.excep_code = AF_PGFAULT \| AF_ZERO;
	1396
	1397	if((page = kmem_alloc(&req)) == NULL)
	1398	{
	1399	(void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
	1400	return ENOMEM;
	1401	}
	1402
	1403	page->mapper = NULL;
	1404
	1405	new.attr = vseg->vm_pgprot;
	1406	new.ppn = ppm_page2ppn(page);
	1407	new.cluster = NULL;
	1408
	1409	error= pmm_set_page(&vseg->vmm->pmm, vaddr, &new);
	1410
	1411	if(err) goto fail_set_pg;
	1412
	1413	cluster = current_cluster;
	1414
	1415	if(page->cid != cluster->id)
	1416	this->info.remote_pages_cntr ++;
	1417
	1418	return 0;
	1419
	1420	fail_set_pg:
	1421	(void)pmm_unlock_page(&vseg->vmm->pmm, vaddr, &old);
	1422	req.ptr = page;
	1423	kmem_free(&req);
	1424
	1425	vmm_dmsg(3, "%s: ended [ error%d ]\n", __FUNCTION__, err);
	1426	return err;
	1427	}
	1428
	1429	VSEGION_PAGE_FAULT(vmm_default_pagefault)
	1430	{
	1431	register struct thread_s *this;
	1432	register error_t err;
	1433	pmm_page_info_t info;
	1434
	1435	if((error= pmm_get_page(&vseg->vmm->pmm, vaddr, &info)))
	1436	return err;
	1437
	1438	if((info.attr != 0) && (info.ppn != 0))
	1439	{
	1440	if((info.attr & PMM_COW) && pmm_except_isWrite(excep_code))
	1441	{
	1442	error= vmm_do_cow(vseg, &info, vaddr);
	1443	return err;
	1444	}
	1445
	1446	if(info.attr & PMM_MIGRATE)
	1447	return vmm_do_migrate(vseg, &info, vaddr);
	1448
	1449	if(info.attr & PMM_PRESENT)
	1450	{
	1451	this = current_thread;
	1452
	1453	#if CONFIG_SHOW_SPURIOUS_PGFAULT
	1454	printk(WARNING, "WARNING: %s: pid %d, tid %d, cpu %d, excep_code %x but vaddr is valid %x, attr %x, ppn %x\n",
	1455	__FUNCTION__,
	1456	this->process->pid,
	1457	this->info.order,
	1458	cpu_get_id(),
	1459	excep_code,
	1460	vaddr,
	1461	info.attr,
	1462	info.ppn);
	1463	#endif
	1464
	1465	current_thread->info.spurious_pgfault_cntr ++;
	1466	pmm_tlb_flush_vaddr(vaddr, PMM_UNKNOWN);
	1467	return 0;
	1468	}
	1469
	1470	current_thread->info.spurious_pgfault_cntr ++;
	1471	pmm_tlb_flush_vaddr(vaddr, PMM_UNKNOWN);
	1472	return 0;
	1473	#if 0
	1474	#if CONFIG_SHOW_VMM_ERROR_MSG
[178]	1475	printk(ERROR,
[1]	1476	"ERROR: %s: pid %d, cpu %d, Unexpected page attributes configuration for vaddr %x, found: ppn %x, attr %x\n",
	1477	__FUNCTION__,
	1478	current_process->pid,
[178]	1479	cpu_get_id(),
	1480	vaddr,
[1]	1481	info.ppn,
	1482	info.attr);
	1483	#endif
[178]	1484
[1]	1485	return EPERM;
	1486	#endif
	1487	}
	1488
	1489	if(!MAPPER_IS_NULL(vseg->vm_mapper))
	1490	return vmm_do_mapped(vseg, vaddr, excep_code);
	1491
	1492	return vmm_do_aod(vseg, vaddr);
	1493	}
	1494	*/
	1495

Note: See TracBrowser for help on using the repository browser.

Download in other formats: