WebSVN – HelenOS – Blame – /branches/tracing/kernel/generic/src/mm/as.c

Rev	Author	Line No.	Line
703	jermar	1	/*
2071	jermar	2	* Copyright (c) 2001-2006 Jakub Jermar
703	jermar	3	* All rights reserved.
		4	*
		5	* Redistribution and use in source and binary forms, with or without
		6	* modification, are permitted provided that the following conditions
		7	* are met:
		8	*
		9	* - Redistributions of source code must retain the above copyright
		10	* notice, this list of conditions and the following disclaimer.
		11	* - Redistributions in binary form must reproduce the above copyright
		12	* notice, this list of conditions and the following disclaimer in the
		13	* documentation and/or other materials provided with the distribution.
		14	* - The name of the author may not be used to endorse or promote products
		15	* derived from this software without specific prior written permission.
		16	*
		17	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
		18	* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
		19	* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
		20	* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
		21	* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
		22	* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
		23	* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
		24	* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
		25	* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
		26	* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
		27	*/
		28
1757	jermar	29	/** @addtogroup genericmm
1702	cejka	30	* @{
		31	*/
		32
1248	jermar	33	/**
1702	cejka	34	* @file
1248	jermar	35	* @brief Address space related functions.
		36	*
703	jermar	37	* This file contains address space manipulation functions.
		38	* Roughly speaking, this is a higher-level client of
		39	* Virtual Address Translation (VAT) subsystem.
1248	jermar	40	*
		41	* Functionality provided by this file allows one to
1757	jermar	42	* create address spaces and create, resize and share
1248	jermar	43	* address space areas.
		44	*
		45	* @see page.c
		46	*
703	jermar	47	*/
		48
		49	#include <mm/as.h>
756	jermar	50	#include <arch/mm/as.h>
703	jermar	51	#include <mm/page.h>
		52	#include <mm/frame.h>
814	palkovsky	53	#include <mm/slab.h>
703	jermar	54	#include <mm/tlb.h>
		55	#include <arch/mm/page.h>
		56	#include <genarch/mm/page_pt.h>
1108	jermar	57	#include <genarch/mm/page_ht.h>
727	jermar	58	#include <mm/asid.h>
703	jermar	59	#include <arch/mm/asid.h>
2183	jermar	60	#include <preemption.h>
703	jermar	61	#include <synch/spinlock.h>
1380	jermar	62	#include <synch/mutex.h>
788	jermar	63	#include <adt/list.h>
1147	jermar	64	#include <adt/btree.h>
1235	jermar	65	#include <proc/task.h>
1288	jermar	66	#include <proc/thread.h>
1235	jermar	67	#include <arch/asm.h>
703	jermar	68	#include <panic.h>
		69	#include <debug.h>
1235	jermar	70	#include <print.h>
703	jermar	71	#include <memstr.h>
1070	jermar	72	#include <macros.h>
703	jermar	73	#include <arch.h>
1235	jermar	74	#include <errno.h>
		75	#include <config.h>
1387	jermar	76	#include <align.h>
1235	jermar	77	#include <arch/types.h>
1288	jermar	78	#include <syscall/copy.h>
		79	#include <arch/interrupt.h>
703	jermar	80
2009	jermar	81	#ifdef CONFIG_VIRT_IDX_DCACHE
		82	#include <arch/mm/cache.h>
		83	#endif /* CONFIG_VIRT_IDX_DCACHE */
		84
1757	jermar	85	/**
		86	* Each architecture decides what functions will be used to carry out
		87	* address space operations such as creating or locking page tables.
		88	*/
756	jermar	89	as_operations_t *as_operations = NULL;
703	jermar	90
1890	jermar	91	/**
		92	* Slab for as_t objects.
		93	*/
		94	static slab_cache_t *as_slab;
		95
2087	jermar	96	/**
2170	jermar	97	* This lock serializes access to the ASID subsystem.
		98	* It protects:
		99	* - inactive_as_with_asid_head list
		100	* - as->asid for each as of the as_t type
		101	* - asids_allocated counter
2087	jermar	102	*/
2170	jermar	103	SPINLOCK_INITIALIZE(asidlock);
823	jermar	104
		105	/**
		106	* This list contains address spaces that are not active on any
		107	* processor and that have valid ASID.
		108	*/
		109	LIST_INITIALIZE(inactive_as_with_asid_head);
		110
757	jermar	111	/** Kernel address space. */
		112	as_t *AS_KERNEL = NULL;
		113
3425	svoboda	114	static int area_flags_to_page_flags(int);
		115	static as_area_t find_area_and_lock(as_t , uintptr_t);
		116	static bool check_area_conflicts(as_t , uintptr_t, size_t, as_area_t );
		117	static void sh_info_remove_reference(share_info_t *);
703	jermar	118
1891	jermar	119	static int as_constructor(void *obj, int flags)
		120	{
		121	as_t as = (as_t ) obj;
		122	int rc;
		123
		124	link_initialize(&as->inactive_as_with_asid_link);
3425	svoboda	125	mutex_initialize(&as->lock, MUTEX_PASSIVE);
1891	jermar	126
		127	rc = as_constructor_arch(as, flags);
		128
		129	return rc;
		130	}
		131
		132	static int as_destructor(void *obj)
		133	{
		134	as_t as = (as_t ) obj;
		135
		136	return as_destructor_arch(as);
		137	}
		138
756	jermar	139	/** Initialize address space subsystem. */
		140	void as_init(void)
		141	{
		142	as_arch_init();
2126	decky	143
1891	jermar	144	as_slab = slab_cache_create("as_slab", sizeof(as_t), 0,
2087	jermar	145	as_constructor, as_destructor, SLAB_CACHE_MAGDEFERRED);
1890	jermar	146
789	palkovsky	147	AS_KERNEL = as_create(FLAG_AS_KERNEL);
1383	decky	148	if (!AS_KERNEL)
		149	panic("can't create kernel address space\n");
		150
756	jermar	151	}
		152
757	jermar	153	/** Create address space.
		154	*
3425	svoboda	155	* @param flags Flags that influence the way in wich the address space
		156	* is created.
757	jermar	157	*/
756	jermar	158	as_t *as_create(int flags)
703	jermar	159	{
		160	as_t *as;
		161
1890	jermar	162	as = (as_t *) slab_alloc(as_slab, 0);
1891	jermar	163	(void) as_create_arch(as, 0);
		164
1147	jermar	165	btree_create(&as->as_area_btree);
822	palkovsky	166
		167	if (flags & FLAG_AS_KERNEL)
		168	as->asid = ASID_KERNEL;
		169	else
		170	as->asid = ASID_INVALID;
		171
2183	jermar	172	atomic_set(&as->refcount, 0);
1415	jermar	173	as->cpu_refcount = 0;
2089	decky	174	#ifdef AS_PAGE_TABLE
2106	jermar	175	as->genarch.page_table = page_table_create(flags);
2089	decky	176	#else
		177	page_table_create(flags);
		178	#endif
703	jermar	179
		180	return as;
		181	}
		182
1468	jermar	183	/** Destroy adress space.
		184	*
2087	jermar	185	* When there are no tasks referencing this address space (i.e. its refcount is
		186	* zero), the address space can be destroyed.
2183	jermar	187	*
		188	* We know that we don't hold any spinlock.
3425	svoboda	189	*
		190	* @param as Address space to be destroyed.
1468	jermar	191	*/
		192	void as_destroy(as_t *as)
973	palkovsky	193	{
1468	jermar	194	ipl_t ipl;
1594	jermar	195	bool cond;
2183	jermar	196	DEADLOCK_PROBE_INIT(p_asidlock);
973	palkovsky	197
2183	jermar	198	ASSERT(atomic_get(&as->refcount) == 0);
1468	jermar	199
		200	/*
		201	* Since there is no reference to this area,
		202	* it is safe not to lock its mutex.
		203	*/
2170	jermar	204
2183	jermar	205	/*
		206	* We need to avoid deadlock between TLB shootdown and asidlock.
		207	* We therefore try to take asid conditionally and if we don't succeed,
		208	* we enable interrupts and try again. This is done while preemption is
		209	* disabled to prevent nested context switches. We also depend on the
		210	* fact that so far no spinlocks are held.
		211	*/
		212	preemption_disable();
		213	ipl = interrupts_read();
		214	retry:
		215	interrupts_disable();
		216	if (!spinlock_trylock(&asidlock)) {
		217	interrupts_enable();
		218	DEADLOCK_PROBE(p_asidlock, DEADLOCK_THRESHOLD);
		219	goto retry;
		220	}
		221	preemption_enable(); /* Interrupts disabled, enable preemption */
1587	jermar	222	if (as->asid != ASID_INVALID && as != AS_KERNEL) {
1594	jermar	223	if (as != AS && as->cpu_refcount == 0)
1587	jermar	224	list_remove(&as->inactive_as_with_asid_link);
1468	jermar	225	asid_put(as->asid);
		226	}
2170	jermar	227	spinlock_unlock(&asidlock);
1468	jermar	228
		229	/*
		230	* Destroy address space areas of the address space.
1954	jermar	231	* The B+tree must be walked carefully because it is
1594	jermar	232	* also being destroyed.
1468	jermar	233	*/
1594	jermar	234	for (cond = true; cond; ) {
1468	jermar	235	btree_node_t *node;
1594	jermar	236
		237	ASSERT(!list_empty(&as->as_area_btree.leaf_head));
2087	jermar	238	node = list_get_instance(as->as_area_btree.leaf_head.next,
		239	btree_node_t, leaf_link);
1594	jermar	240
		241	if ((cond = node->keys)) {
		242	as_area_destroy(as, node->key[0]);
		243	}
1468	jermar	244	}
1495	jermar	245
1483	jermar	246	btree_destroy(&as->as_area_btree);
2089	decky	247	#ifdef AS_PAGE_TABLE
2106	jermar	248	page_table_destroy(as->genarch.page_table);
2089	decky	249	#else
		250	page_table_destroy(NULL);
		251	#endif
1468	jermar	252
		253	interrupts_restore(ipl);
2126	decky	254
1890	jermar	255	slab_free(as_slab, as);
973	palkovsky	256	}
		257
703	jermar	258	/** Create address space area of common attributes.
		259	*
		260	* The created address space area is added to the target address space.
		261	*
3425	svoboda	262	* @param as Target address space.
		263	* @param flags Flags of the area memory.
		264	* @param size Size of area.
		265	* @param base Base address of area.
		266	* @param attrs Attributes of the area.
		267	* @param backend Address space area backend. NULL if no backend is used.
		268	* @param backend_data NULL or a pointer to an array holding two void *.
703	jermar	269	*
3425	svoboda	270	* @return Address space area on success or NULL on failure.
703	jermar	271	*/
2069	jermar	272	as_area_t *
		273	as_area_create(as_t *as, int flags, size_t size, uintptr_t base, int attrs,
3425	svoboda	274	mem_backend_t backend, mem_backend_data_t backend_data)
703	jermar	275	{
		276	ipl_t ipl;
		277	as_area_t *a;
		278
		279	if (base % PAGE_SIZE)
1048	jermar	280	return NULL;
		281
1233	jermar	282	if (!size)
		283	return NULL;
		284
1048	jermar	285	/* Writeable executable areas are not supported. */
		286	if ((flags & AS_AREA_EXEC) && (flags & AS_AREA_WRITE))
		287	return NULL;
703	jermar	288
		289	ipl = interrupts_disable();
1380	jermar	290	mutex_lock(&as->lock);
703	jermar	291
1048	jermar	292	if (!check_area_conflicts(as, base, size, NULL)) {
1380	jermar	293	mutex_unlock(&as->lock);
1048	jermar	294	interrupts_restore(ipl);
		295	return NULL;
		296	}
703	jermar	297
822	palkovsky	298	a = (as_area_t *) malloc(sizeof(as_area_t), 0);
703	jermar	299
3425	svoboda	300	mutex_initialize(&a->lock, MUTEX_PASSIVE);
822	palkovsky	301
1424	jermar	302	a->as = as;
1026	jermar	303	a->flags = flags;
1239	jermar	304	a->attributes = attrs;
1048	jermar	305	a->pages = SIZE2FRAMES(size);
822	palkovsky	306	a->base = base;
1409	jermar	307	a->sh_info = NULL;
		308	a->backend = backend;
1424	jermar	309	if (backend_data)
		310	a->backend_data = *backend_data;
		311	else
3424	svoboda	312	memsetb(&a->backend_data, sizeof(a->backend_data), 0);
1424	jermar	313
1387	jermar	314	btree_create(&a->used_space);
822	palkovsky	315
1147	jermar	316	btree_insert(&as->as_area_btree, base, (void *) a, NULL);
822	palkovsky	317
1380	jermar	318	mutex_unlock(&as->lock);
703	jermar	319	interrupts_restore(ipl);
704	jermar	320
703	jermar	321	return a;
		322	}
		323
1235	jermar	324	/** Find address space area and change it.
		325	*
3425	svoboda	326	* @param as Address space.
		327	* @param address Virtual address belonging to the area to be changed.
		328	* Must be page-aligned.
		329	* @param size New size of the virtual memory block starting at
		330	* address.
		331	* @param flags Flags influencing the remap operation. Currently unused.
1235	jermar	332	*
3425	svoboda	333	* @return Zero on success or a value from @ref errno.h otherwise.
1235	jermar	334	*/
1780	jermar	335	int as_area_resize(as_t *as, uintptr_t address, size_t size, int flags)
1235	jermar	336	{
1306	jermar	337	as_area_t *area;
1235	jermar	338	ipl_t ipl;
		339	size_t pages;
		340
		341	ipl = interrupts_disable();
1380	jermar	342	mutex_lock(&as->lock);
1235	jermar	343
		344	/*
		345	* Locate the area.
		346	*/
		347	area = find_area_and_lock(as, address);
		348	if (!area) {
1380	jermar	349	mutex_unlock(&as->lock);
1235	jermar	350	interrupts_restore(ipl);
1306	jermar	351	return ENOENT;
1235	jermar	352	}
		353
1424	jermar	354	if (area->backend == &phys_backend) {
1235	jermar	355	/*
		356	* Remapping of address space areas associated
		357	* with memory mapped devices is not supported.
		358	*/
1380	jermar	359	mutex_unlock(&area->lock);
		360	mutex_unlock(&as->lock);
1235	jermar	361	interrupts_restore(ipl);
1306	jermar	362	return ENOTSUP;
1235	jermar	363	}
1409	jermar	364	if (area->sh_info) {
		365	/*
		366	* Remapping of shared address space areas
		367	* is not supported.
		368	*/
		369	mutex_unlock(&area->lock);
		370	mutex_unlock(&as->lock);
		371	interrupts_restore(ipl);
		372	return ENOTSUP;
		373	}
1235	jermar	374
		375	pages = SIZE2FRAMES((address - area->base) + size);
		376	if (!pages) {
		377	/*
		378	* Zero size address space areas are not allowed.
		379	*/
1380	jermar	380	mutex_unlock(&area->lock);
		381	mutex_unlock(&as->lock);
1235	jermar	382	interrupts_restore(ipl);
1306	jermar	383	return EPERM;
1235	jermar	384	}
		385
		386	if (pages < area->pages) {
1403	jermar	387	bool cond;
3425	svoboda	388	uintptr_t start_free = area->base + pages * PAGE_SIZE;
1235	jermar	389
		390	/*
		391	* Shrinking the area.
		392	* No need to check for overlaps.
		393	*/
1403	jermar	394
		395	/*
1436	jermar	396	* Start TLB shootdown sequence.
		397	*/
3425	svoboda	398	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base +
2087	jermar	399	pages * PAGE_SIZE, area->pages - pages);
1436	jermar	400
		401	/*
1403	jermar	402	* Remove frames belonging to used space starting from
		403	* the highest addresses downwards until an overlap with
		404	* the resized address space area is found. Note that this
		405	* is also the right way to remove part of the used_space
		406	* B+tree leaf list.
		407	*/
		408	for (cond = true; cond;) {
		409	btree_node_t *node;
		410
		411	ASSERT(!list_empty(&area->used_space.leaf_head));
2087	jermar	412	node =
		413	list_get_instance(area->used_space.leaf_head.prev,
		414	btree_node_t, leaf_link);
1403	jermar	415	if ((cond = (bool) node->keys)) {
1780	jermar	416	uintptr_t b = node->key[node->keys - 1];
2087	jermar	417	count_t c =
		418	(count_t) node->value[node->keys - 1];
2745	decky	419	unsigned int i = 0;
1235	jermar	420
2087	jermar	421	if (overlaps(b, c * PAGE_SIZE, area->base,
2133	jermar	422	pages * PAGE_SIZE)) {
1403	jermar	423
2087	jermar	424	if (b + c * PAGE_SIZE <= start_free) {
1403	jermar	425	/*
2087	jermar	426	* The whole interval fits
		427	* completely in the resized
		428	* address space area.
1403	jermar	429	*/
		430	break;
		431	}
		432
		433	/*
2087	jermar	434	* Part of the interval corresponding
		435	* to b and c overlaps with the resized
		436	* address space area.
1403	jermar	437	*/
		438
		439	cond = false; /* we are almost done */
		440	i = (start_free - b) >> PAGE_WIDTH;
3425	svoboda	441	if (!used_space_remove(area, start_free,
		442	c - i))
		443	panic("Could not remove used "
		444	"space.\n");
1403	jermar	445	} else {
		446	/*
2087	jermar	447	* The interval of used space can be
		448	* completely removed.
1403	jermar	449	*/
		450	if (!used_space_remove(area, b, c))
3425	svoboda	451	panic("Could not remove used "
		452	"space.\n");
1403	jermar	453	}
		454
		455	for (; i < c; i++) {
		456	pte_t *pte;
		457
		458	page_table_lock(as, false);
2087	jermar	459	pte = page_mapping_find(as, b +
		460	i * PAGE_SIZE);
		461	ASSERT(pte && PTE_VALID(pte) &&
		462	PTE_PRESENT(pte));
		463	if (area->backend &&
		464	area->backend->frame_free) {
1424	jermar	465	area->backend->frame_free(area,
2087	jermar	466	b + i * PAGE_SIZE,
		467	PTE_GET_FRAME(pte));
1409	jermar	468	}
2087	jermar	469	page_mapping_remove(as, b +
		470	i * PAGE_SIZE);
1403	jermar	471	page_table_unlock(as, false);
		472	}
1235	jermar	473	}
		474	}
1436	jermar	475
1235	jermar	476	/*
1436	jermar	477	* Finish TLB shootdown sequence.
1235	jermar	478	*/
2183	jermar	479
2087	jermar	480	tlb_invalidate_pages(as->asid, area->base + pages * PAGE_SIZE,
		481	area->pages - pages);
1889	jermar	482	/*
		483	* Invalidate software translation caches (e.g. TSB on sparc64).
		484	*/
2087	jermar	485	as_invalidate_translation_cache(as, area->base +
		486	pages * PAGE_SIZE, area->pages - pages);
2183	jermar	487	tlb_shootdown_finalize();
		488
1235	jermar	489	} else {
		490	/*
		491	* Growing the area.
		492	* Check for overlaps with other address space areas.
		493	*/
2087	jermar	494	if (!check_area_conflicts(as, address, pages * PAGE_SIZE,
		495	area)) {
1380	jermar	496	mutex_unlock(&area->lock);
		497	mutex_unlock(&as->lock);
1235	jermar	498	interrupts_restore(ipl);
1306	jermar	499	return EADDRNOTAVAIL;
1235	jermar	500	}
		501	}
		502
		503	area->pages = pages;
		504
1380	jermar	505	mutex_unlock(&area->lock);
		506	mutex_unlock(&as->lock);
1235	jermar	507	interrupts_restore(ipl);
		508
1306	jermar	509	return 0;
1235	jermar	510	}
		511
1306	jermar	512	/** Destroy address space area.
		513	*
3425	svoboda	514	* @param as Address space.
		515	* @param address Address within the area to be deleted.
1306	jermar	516	*
3425	svoboda	517	* @return Zero on success or a value from @ref errno.h on failure.
1306	jermar	518	*/
1780	jermar	519	int as_area_destroy(as_t *as, uintptr_t address)
1306	jermar	520	{
		521	as_area_t *area;
1780	jermar	522	uintptr_t base;
1495	jermar	523	link_t *cur;
1306	jermar	524	ipl_t ipl;
		525
		526	ipl = interrupts_disable();
1380	jermar	527	mutex_lock(&as->lock);
1306	jermar	528
		529	area = find_area_and_lock(as, address);
		530	if (!area) {
1380	jermar	531	mutex_unlock(&as->lock);
1306	jermar	532	interrupts_restore(ipl);
		533	return ENOENT;
		534	}
		535
1403	jermar	536	base = area->base;
		537
1411	jermar	538	/*
1436	jermar	539	* Start TLB shootdown sequence.
		540	*/
1889	jermar	541	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
1436	jermar	542
		543	/*
1411	jermar	544	* Visit only the pages mapped by used_space B+tree.
		545	*/
2087	jermar	546	for (cur = area->used_space.leaf_head.next;
		547	cur != &area->used_space.leaf_head; cur = cur->next) {
1411	jermar	548	btree_node_t *node;
2745	decky	549	unsigned int i;
1403	jermar	550
1495	jermar	551	node = list_get_instance(cur, btree_node_t, leaf_link);
		552	for (i = 0; i < node->keys; i++) {
1780	jermar	553	uintptr_t b = node->key[i];
1495	jermar	554	count_t j;
1411	jermar	555	pte_t *pte;
1403	jermar	556
1495	jermar	557	for (j = 0; j < (count_t) node->value[i]; j++) {
1411	jermar	558	page_table_lock(as, false);
2087	jermar	559	pte = page_mapping_find(as, b + j * PAGE_SIZE);
		560	ASSERT(pte && PTE_VALID(pte) &&
		561	PTE_PRESENT(pte));
		562	if (area->backend &&
		563	area->backend->frame_free) {
		564	area->backend->frame_free(area, b +
2133	jermar	565	j * PAGE_SIZE, PTE_GET_FRAME(pte));
1403	jermar	566	}
2087	jermar	567	page_mapping_remove(as, b + j * PAGE_SIZE);
1411	jermar	568	page_table_unlock(as, false);
1306	jermar	569	}
		570	}
		571	}
1403	jermar	572
1306	jermar	573	/*
1436	jermar	574	* Finish TLB shootdown sequence.
1306	jermar	575	*/
2183	jermar	576
1889	jermar	577	tlb_invalidate_pages(as->asid, area->base, area->pages);
		578	/*
2087	jermar	579	* Invalidate potential software translation caches (e.g. TSB on
		580	* sparc64).
1889	jermar	581	*/
		582	as_invalidate_translation_cache(as, area->base, area->pages);
2183	jermar	583	tlb_shootdown_finalize();
1889	jermar	584
1436	jermar	585	btree_destroy(&area->used_space);
1306	jermar	586
1309	jermar	587	area->attributes \|= AS_AREA_ATTR_PARTIAL;
1409	jermar	588
		589	if (area->sh_info)
		590	sh_info_remove_reference(area->sh_info);
		591
1380	jermar	592	mutex_unlock(&area->lock);
1306	jermar	593
		594	/*
		595	* Remove the empty area from address space.
		596	*/
1889	jermar	597	btree_remove(&as->as_area_btree, base, NULL);
1306	jermar	598
1309	jermar	599	free(area);
		600
1889	jermar	601	mutex_unlock(&as->lock);
1306	jermar	602	interrupts_restore(ipl);
		603	return 0;
		604	}
		605
1413	jermar	606	/** Share address space area with another or the same address space.
1235	jermar	607	*
1424	jermar	608	* Address space area mapping is shared with a new address space area.
		609	* If the source address space area has not been shared so far,
		610	* a new sh_info is created. The new address space area simply gets the
		611	* sh_info of the source area. The process of duplicating the
		612	* mapping is done through the backend share function.
1413	jermar	613	*
3425	svoboda	614	* @param src_as Pointer to source address space.
		615	* @param src_base Base address of the source address space area.
		616	* @param acc_size Expected size of the source area.
		617	* @param dst_as Pointer to destination address space.
		618	* @param dst_base Target base address.
1417	jermar	619	* @param dst_flags_mask Destination address space area flags mask.
1235	jermar	620	*
3425	svoboda	621	* @return Zero on success or ENOENT if there is no such task or if
		622	* there is no such address space area, EPERM if there was
		623	* a problem in accepting the area or ENOMEM if there was a
		624	* problem in allocating destination address space area.
		625	* ENOTSUP is returned if the address space area backend
		626	* does not support sharing.
1235	jermar	627	*/
1780	jermar	628	int as_area_share(as_t *src_as, uintptr_t src_base, size_t acc_size,
2647	jermar	629	as_t *dst_as, uintptr_t dst_base, int dst_flags_mask)
1235	jermar	630	{
		631	ipl_t ipl;
1239	jermar	632	int src_flags;
		633	size_t src_size;
		634	as_area_t src_area, dst_area;
1413	jermar	635	share_info_t *sh_info;
1424	jermar	636	mem_backend_t *src_backend;
		637	mem_backend_data_t src_backend_data;
1434	palkovsky	638
1235	jermar	639	ipl = interrupts_disable();
1380	jermar	640	mutex_lock(&src_as->lock);
1329	palkovsky	641	src_area = find_area_and_lock(src_as, src_base);
1239	jermar	642	if (!src_area) {
1238	jermar	643	/*
		644	* Could not find the source address space area.
		645	*/
1380	jermar	646	mutex_unlock(&src_as->lock);
1238	jermar	647	interrupts_restore(ipl);
		648	return ENOENT;
		649	}
2007	jermar	650
1424	jermar	651	if (!src_area->backend \|\| !src_area->backend->share) {
1413	jermar	652	/*
1851	jermar	653	* There is no backend or the backend does not
1424	jermar	654	* know how to share the area.
1413	jermar	655	*/
		656	mutex_unlock(&src_area->lock);
		657	mutex_unlock(&src_as->lock);
		658	interrupts_restore(ipl);
		659	return ENOTSUP;
		660	}
		661
1239	jermar	662	src_size = src_area->pages * PAGE_SIZE;
		663	src_flags = src_area->flags;
1424	jermar	664	src_backend = src_area->backend;
		665	src_backend_data = src_area->backend_data;
1544	palkovsky	666
		667	/* Share the cacheable flag from the original mapping */
		668	if (src_flags & AS_AREA_CACHEABLE)
		669	dst_flags_mask \|= AS_AREA_CACHEABLE;
		670
2087	jermar	671	if (src_size != acc_size \|\|
		672	(src_flags & dst_flags_mask) != dst_flags_mask) {
1413	jermar	673	mutex_unlock(&src_area->lock);
		674	mutex_unlock(&src_as->lock);
1235	jermar	675	interrupts_restore(ipl);
		676	return EPERM;
		677	}
1413	jermar	678
1235	jermar	679	/*
1413	jermar	680	* Now we are committed to sharing the area.
1954	jermar	681	* First, prepare the area for sharing.
1413	jermar	682	* Then it will be safe to unlock it.
		683	*/
		684	sh_info = src_area->sh_info;
		685	if (!sh_info) {
		686	sh_info = (share_info_t *) malloc(sizeof(share_info_t), 0);
3425	svoboda	687	mutex_initialize(&sh_info->lock, MUTEX_PASSIVE);
1413	jermar	688	sh_info->refcount = 2;
		689	btree_create(&sh_info->pagemap);
		690	src_area->sh_info = sh_info;
2647	jermar	691	/*
		692	* Call the backend to setup sharing.
		693	*/
		694	src_area->backend->share(src_area);
1413	jermar	695	} else {
		696	mutex_lock(&sh_info->lock);
		697	sh_info->refcount++;
		698	mutex_unlock(&sh_info->lock);
		699	}
		700
		701	mutex_unlock(&src_area->lock);
		702	mutex_unlock(&src_as->lock);
		703
		704	/*
1239	jermar	705	* Create copy of the source address space area.
		706	* The destination area is created with AS_AREA_ATTR_PARTIAL
		707	* attribute set which prevents race condition with
		708	* preliminary as_page_fault() calls.
1417	jermar	709	* The flags of the source area are masked against dst_flags_mask
		710	* to support sharing in less privileged mode.
1235	jermar	711	*/
1461	palkovsky	712	dst_area = as_area_create(dst_as, dst_flags_mask, src_size, dst_base,
2087	jermar	713	AS_AREA_ATTR_PARTIAL, src_backend, &src_backend_data);
1239	jermar	714	if (!dst_area) {
1235	jermar	715	/*
		716	* Destination address space area could not be created.
		717	*/
1413	jermar	718	sh_info_remove_reference(sh_info);
		719
1235	jermar	720	interrupts_restore(ipl);
		721	return ENOMEM;
		722	}
2009	jermar	723
1235	jermar	724	/*
1239	jermar	725	* Now the destination address space area has been
		726	* fully initialized. Clear the AS_AREA_ATTR_PARTIAL
1413	jermar	727	* attribute and set the sh_info.
1239	jermar	728	*/
2009	jermar	729	mutex_lock(&dst_as->lock);
1380	jermar	730	mutex_lock(&dst_area->lock);
1239	jermar	731	dst_area->attributes &= ~AS_AREA_ATTR_PARTIAL;
1413	jermar	732	dst_area->sh_info = sh_info;
1380	jermar	733	mutex_unlock(&dst_area->lock);
2009	jermar	734	mutex_unlock(&dst_as->lock);
		735
1235	jermar	736	interrupts_restore(ipl);
		737
		738	return 0;
		739	}
		740
1423	jermar	741	/** Check access mode for address space area.
		742	*
		743	* The address space area must be locked prior to this call.
		744	*
3425	svoboda	745	* @param area Address space area.
		746	* @param access Access mode.
1423	jermar	747	*
3425	svoboda	748	* @return False if access violates area's permissions, true
		749	* otherwise.
1423	jermar	750	*/
		751	bool as_area_check_access(as_area_t *area, pf_access_t access)
		752	{
		753	int flagmap[] = {
		754	[PF_ACCESS_READ] = AS_AREA_READ,
		755	[PF_ACCESS_WRITE] = AS_AREA_WRITE,
		756	[PF_ACCESS_EXEC] = AS_AREA_EXEC
		757	};
		758
		759	if (!(area->flags & flagmap[access]))
		760	return false;
		761
		762	return true;
		763	}
		764
3425	svoboda	765	/** Change adress space area flags.
		766	*
		767	* The idea is to have the same data, but with a different access mode.
		768	* This is needed e.g. for writing code into memory and then executing it.
		769	* In order for this to work properly, this may copy the data
		770	* into private anonymous memory (unless it's already there).
		771	*
		772	* @param as Address space.
		773	* @param flags Flags of the area memory.
		774	* @param address Address withing the area to be changed.
		775	*
		776	* @return Zero on success or a value from @ref errno.h on failure.
		777	*/
		778	int as_area_change_flags(as_t *as, int flags, uintptr_t address)
		779	{
		780	as_area_t *area;
		781	uintptr_t base;
		782	link_t *cur;
		783	ipl_t ipl;
		784	int page_flags;
		785	uintptr_t *old_frame;
		786	index_t frame_idx;
		787	count_t used_pages;
		788
		789	/* Flags for the new memory mapping */
		790	page_flags = area_flags_to_page_flags(flags);
		791
		792	ipl = interrupts_disable();
		793	mutex_lock(&as->lock);
		794
		795	area = find_area_and_lock(as, address);
		796	if (!area) {
		797	mutex_unlock(&as->lock);
		798	interrupts_restore(ipl);
		799	return ENOENT;
		800	}
		801
		802	if (area->sh_info \|\| area->backend != &anon_backend) {
		803	/* Copying shared areas not supported yet */
		804	/* Copying non-anonymous memory not supported yet */
		805	mutex_unlock(&area->lock);
		806	mutex_unlock(&as->lock);
		807	interrupts_restore(ipl);
		808	return ENOTSUP;
		809	}
		810
		811	base = area->base;
		812
		813	/*
		814	* Compute total number of used pages in the used_space B+tree
		815	*/
		816	used_pages = 0;
		817
		818	for (cur = area->used_space.leaf_head.next;
		819	cur != &area->used_space.leaf_head; cur = cur->next) {
		820	btree_node_t *node;
		821	unsigned int i;
		822
		823	node = list_get_instance(cur, btree_node_t, leaf_link);
		824	for (i = 0; i < node->keys; i++) {
		825	used_pages += (count_t) node->value[i];
		826	}
		827	}
		828
		829	/* An array for storing frame numbers */
		830	old_frame = malloc(used_pages * sizeof(uintptr_t), 0);
		831
		832	/*
		833	* Start TLB shootdown sequence.
		834	*/
		835	tlb_shootdown_start(TLB_INVL_PAGES, as->asid, area->base, area->pages);
		836
		837	/*
		838	* Remove used pages from page tables and remember their frame
		839	* numbers.
		840	*/
		841	frame_idx = 0;
		842
		843	for (cur = area->used_space.leaf_head.next;
		844	cur != &area->used_space.leaf_head; cur = cur->next) {
		845	btree_node_t *node;
		846	unsigned int i;
		847
		848	node = list_get_instance(cur, btree_node_t, leaf_link);
		849	for (i = 0; i < node->keys; i++) {
		850	uintptr_t b = node->key[i];
		851	count_t j;
		852	pte_t *pte;
		853
		854	for (j = 0; j < (count_t) node->value[i]; j++) {
		855	page_table_lock(as, false);
		856	pte = page_mapping_find(as, b + j * PAGE_SIZE);
		857	ASSERT(pte && PTE_VALID(pte) &&
		858	PTE_PRESENT(pte));
		859	old_frame[frame_idx++] = PTE_GET_FRAME(pte);
		860
		861	/* Remove old mapping */
		862	page_mapping_remove(as, b + j * PAGE_SIZE);
		863	page_table_unlock(as, false);
		864	}
		865	}
		866	}
		867
		868	/*
		869	* Finish TLB shootdown sequence.
		870	*/
		871
		872	tlb_invalidate_pages(as->asid, area->base, area->pages);
		873	/*
		874	* Invalidate potential software translation caches (e.g. TSB on
		875	* sparc64).
		876	*/
		877	as_invalidate_translation_cache(as, area->base, area->pages);
		878	tlb_shootdown_finalize();
		879
		880	/*
		881	* Set the new flags.
		882	*/
		883	area->flags = flags;
		884
		885	/*
		886	* Map pages back in with new flags. This step is kept separate
		887	* so that the memory area could not be accesed with both the old and
		888	* the new flags at once.
		889	*/
		890	frame_idx = 0;
		891
		892	for (cur = area->used_space.leaf_head.next;
		893	cur != &area->used_space.leaf_head; cur = cur->next) {
		894	btree_node_t *node;
		895	unsigned int i;
		896
		897	node = list_get_instance(cur, btree_node_t, leaf_link);
		898	for (i = 0; i < node->keys; i++) {
		899	uintptr_t b = node->key[i];
		900	count_t j;
		901
		902	for (j = 0; j < (count_t) node->value[i]; j++) {
		903	page_table_lock(as, false);
		904
		905	/* Insert the new mapping */
		906	page_mapping_insert(as, b + j * PAGE_SIZE,
		907	old_frame[frame_idx++], page_flags);
		908
		909	page_table_unlock(as, false);
		910	}
		911	}
		912	}
		913
		914	free(old_frame);
		915
		916	mutex_unlock(&area->lock);
		917	mutex_unlock(&as->lock);
		918	interrupts_restore(ipl);
		919
		920	return 0;
		921	}
		922
		923
703	jermar	924	/** Handle page fault within the current address space.
		925	*
3425	svoboda	926	* This is the high-level page fault handler. It decides whether the page fault
		927	* can be resolved by any backend and if so, it invokes the backend to resolve
		928	* the page fault.
1409	jermar	929	*
703	jermar	930	* Interrupts are assumed disabled.
		931	*
3425	svoboda	932	* @param page Faulting page.
		933	* @param access Access mode that caused the page fault (i.e.
		934	* read/write/exec).
		935	* @param istate Pointer to the interrupted state.
703	jermar	936	*
3425	svoboda	937	* @return AS_PF_FAULT on page fault, AS_PF_OK on success or
		938	* AS_PF_DEFER if the fault was caused by copy_to_uspace()
		939	* or copy_from_uspace().
703	jermar	940	*/
1780	jermar	941	int as_page_fault(uintptr_t page, pf_access_t access, istate_t *istate)
703	jermar	942	{
1044	jermar	943	pte_t *pte;
977	jermar	944	as_area_t *area;
703	jermar	945
1380	jermar	946	if (!THREAD)
1409	jermar	947	return AS_PF_FAULT;
1380	jermar	948
703	jermar	949	ASSERT(AS);
1044	jermar	950
1380	jermar	951	mutex_lock(&AS->lock);
977	jermar	952	area = find_area_and_lock(AS, page);
703	jermar	953	if (!area) {
		954	/*
		955	* No area contained mapping for 'page'.
		956	* Signal page fault to low-level handler.
		957	*/
1380	jermar	958	mutex_unlock(&AS->lock);
1288	jermar	959	goto page_fault;
703	jermar	960	}
		961
1239	jermar	962	if (area->attributes & AS_AREA_ATTR_PARTIAL) {
		963	/*
		964	* The address space area is not fully initialized.
		965	* Avoid possible race by returning error.
		966	*/
1380	jermar	967	mutex_unlock(&area->lock);
		968	mutex_unlock(&AS->lock);
1288	jermar	969	goto page_fault;
1239	jermar	970	}
		971
1424	jermar	972	if (!area->backend \|\| !area->backend->page_fault) {
1409	jermar	973	/*
		974	* The address space area is not backed by any backend
		975	* or the backend cannot handle page faults.
		976	*/
		977	mutex_unlock(&area->lock);
		978	mutex_unlock(&AS->lock);
		979	goto page_fault;
		980	}
1179	jermar	981
1044	jermar	982	page_table_lock(AS, false);
		983
703	jermar	984	/*
3425	svoboda	985	* To avoid race condition between two page faults on the same address,
		986	* we need to make sure the mapping has not been already inserted.
1044	jermar	987	*/
		988	if ((pte = page_mapping_find(AS, page))) {
		989	if (PTE_PRESENT(pte)) {
1423	jermar	990	if (((access == PF_ACCESS_READ) && PTE_READABLE(pte)) \|\|
2087	jermar	991	(access == PF_ACCESS_WRITE && PTE_WRITABLE(pte)) \|\|
		992	(access == PF_ACCESS_EXEC && PTE_EXECUTABLE(pte))) {
1423	jermar	993	page_table_unlock(AS, false);
		994	mutex_unlock(&area->lock);
		995	mutex_unlock(&AS->lock);
		996	return AS_PF_OK;
		997	}
1044	jermar	998	}
		999	}
1409	jermar	1000
1044	jermar	1001	/*
1409	jermar	1002	* Resort to the backend page fault handler.
703	jermar	1003	*/
1424	jermar	1004	if (area->backend->page_fault(area, page, access) != AS_PF_OK) {
1409	jermar	1005	page_table_unlock(AS, false);
		1006	mutex_unlock(&area->lock);
		1007	mutex_unlock(&AS->lock);
		1008	goto page_fault;
		1009	}
703	jermar	1010
1044	jermar	1011	page_table_unlock(AS, false);
1380	jermar	1012	mutex_unlock(&area->lock);
		1013	mutex_unlock(&AS->lock);
1288	jermar	1014	return AS_PF_OK;
		1015
		1016	page_fault:
		1017	if (THREAD->in_copy_from_uspace) {
		1018	THREAD->in_copy_from_uspace = false;
2087	jermar	1019	istate_set_retaddr(istate,
		1020	(uintptr_t) &memcpy_from_uspace_failover_address);
1288	jermar	1021	} else if (THREAD->in_copy_to_uspace) {
		1022	THREAD->in_copy_to_uspace = false;
2087	jermar	1023	istate_set_retaddr(istate,
		1024	(uintptr_t) &memcpy_to_uspace_failover_address);
1288	jermar	1025	} else {
		1026	return AS_PF_FAULT;
		1027	}
		1028
		1029	return AS_PF_DEFER;
703	jermar	1030	}
		1031
823	jermar	1032	/** Switch address spaces.
703	jermar	1033	*
1380	jermar	1034	* Note that this function cannot sleep as it is essentially a part of
2170	jermar	1035	* scheduling. Sleeping here would lead to deadlock on wakeup. Another
		1036	* thing which is forbidden in this context is locking the address space.
1380	jermar	1037	*
2183	jermar	1038	* When this function is enetered, no spinlocks may be held.
		1039	*
3425	svoboda	1040	* @param old Old address space or NULL.
		1041	* @param new New address space.
703	jermar	1042	*/
2106	jermar	1043	void as_switch(as_t old_as, as_t new_as)
703	jermar	1044	{
2183	jermar	1045	DEADLOCK_PROBE_INIT(p_asidlock);
		1046	preemption_disable();
		1047	retry:
		1048	(void) interrupts_disable();
		1049	if (!spinlock_trylock(&asidlock)) {
		1050	/*
		1051	* Avoid deadlock with TLB shootdown.
		1052	* We can enable interrupts here because
		1053	* preemption is disabled. We should not be
		1054	* holding any other lock.
		1055	*/
		1056	(void) interrupts_enable();
		1057	DEADLOCK_PROBE(p_asidlock, DEADLOCK_THRESHOLD);
		1058	goto retry;
		1059	}
		1060	preemption_enable();
703	jermar	1061
		1062	/*
823	jermar	1063	* First, take care of the old address space.
		1064	*/
2106	jermar	1065	if (old_as) {
		1066	ASSERT(old_as->cpu_refcount);
		1067	if((--old_as->cpu_refcount == 0) && (old_as != AS_KERNEL)) {
823	jermar	1068	/*
		1069	* The old address space is no longer active on
		1070	* any processor. It can be appended to the
		1071	* list of inactive address spaces with assigned
		1072	* ASID.
		1073	*/
2141	jermar	1074	ASSERT(old_as->asid != ASID_INVALID);
		1075	list_append(&old_as->inactive_as_with_asid_link,
		1076	&inactive_as_with_asid_head);
823	jermar	1077	}
1890	jermar	1078
		1079	/*
		1080	* Perform architecture-specific tasks when the address space
		1081	* is being removed from the CPU.
		1082	*/
2106	jermar	1083	as_deinstall_arch(old_as);
823	jermar	1084	}
		1085
		1086	/*
		1087	* Second, prepare the new address space.
		1088	*/
2106	jermar	1089	if ((new_as->cpu_refcount++ == 0) && (new_as != AS_KERNEL)) {
2170	jermar	1090	if (new_as->asid != ASID_INVALID)
2106	jermar	1091	list_remove(&new_as->inactive_as_with_asid_link);
2170	jermar	1092	else
		1093	new_as->asid = asid_get();
823	jermar	1094	}
2106	jermar	1095	#ifdef AS_PAGE_TABLE
		1096	SET_PTL0_ADDRESS(new_as->genarch.page_table);
		1097	#endif
823	jermar	1098
		1099	/*
703	jermar	1100	* Perform architecture-specific steps.
727	jermar	1101	* (e.g. write ASID to hardware register etc.)
703	jermar	1102	*/
2106	jermar	1103	as_install_arch(new_as);
2170	jermar	1104
		1105	spinlock_unlock(&asidlock);
703	jermar	1106
2106	jermar	1107	AS = new_as;
703	jermar	1108	}
754	jermar	1109
3431	svoboda	1110	#ifdef CONFIG_UDEBUG
		1111
3013	svoboda	1112	/** Write directly into a page, bypassing area flags.
		1113	*
		1114	* This allows a debugger to write into a page that is mapped read-only
3596	svoboda	1115	* (such as the text segment). Naturally, this can only be done if the
		1116	* correspoinding area is private (not shared) and anonymous.
3013	svoboda	1117	*
3596	svoboda	1118	* If this is not the case, this function calls as_area_make_writeable()
		1119	* first.
3013	svoboda	1120	*/
		1121	static int debug_write_inside_page(uintptr_t va, void *data, size_t n)
		1122	{
		1123	uintptr_t page;
		1124	pte_t *pte;
		1125	as_area_t *area;
		1126	uintptr_t frame;
		1127	ipl_t ipl;
		1128	int rc;
		1129
		1130	page = ALIGN_DOWN(va, PAGE_SIZE);
		1131	ASSERT(ALIGN_DOWN(va + n - 1, PAGE_SIZE) == page);
		1132
		1133	restart:
		1134	mutex_lock(&AS->lock);
		1135	ipl = interrupts_disable();
		1136	area = find_area_and_lock(AS, page);
3036	svoboda	1137	if (area->backend != &anon_backend \|\| area->sh_info != NULL) {
		1138	mutex_unlock(&area->lock);
		1139	mutex_unlock(&AS->lock);
		1140	interrupts_restore(ipl);
3013	svoboda	1141
3036	svoboda	1142	rc = as_area_make_writeable(area->base);
		1143	if (rc != 0) return rc;
		1144
		1145	goto restart;
		1146	}
		1147
3013	svoboda	1148	pte = page_mapping_find(AS, page);
		1149	if (! (pte && PTE_VALID(pte) && PTE_PRESENT(pte)) ) {
		1150	mutex_unlock(&area->lock);
		1151	mutex_unlock(&AS->lock);
		1152	interrupts_restore(ipl);
		1153
		1154	rc = as_page_fault(page, PF_ACCESS_WRITE, NULL);
		1155	if (rc == AS_PF_FAULT) return EINVAL;
		1156
		1157	goto restart;
		1158	}
		1159
		1160	frame = PTE_GET_FRAME(pte);
		1161	memcpy((void *)(PA2KA(frame) + (va - page)), data, n);
		1162
		1163	mutex_unlock(&area->lock);
		1164	mutex_unlock(&AS->lock);
		1165	interrupts_restore(ipl);
		1166
		1167	return EOK;
		1168	}
		1169
		1170	/** Write data bypassing area flags.
		1171	*
		1172	* See debug_write_inside_page().
		1173	*/
		1174	int as_debug_write(uintptr_t va, void *data, size_t n)
		1175	{
		1176	size_t now;
		1177	int rc;
		1178
		1179	while (n > 0) {
3127	svoboda	1180	/* Number of bytes until the end of page */
		1181	now = ALIGN_DOWN(va, PAGE_SIZE) + PAGE_SIZE - va;
3013	svoboda	1182	if (now > n) now = n;
		1183
		1184	rc = debug_write_inside_page(va, data, now);
		1185	if (rc != EOK) return rc;
		1186
		1187	va += now;
		1188	data += now;
		1189	n -= now;
		1190	}
		1191
		1192	return EOK;
		1193	}
		1194
3036	svoboda	1195	/** Make sure area is private and anonymous.
		1196	*
		1197	* Not atomic atm.
		1198	* @param address Virtual address in AS.
		1199	*/
		1200	int as_area_make_writeable(uintptr_t address)
		1201	{
		1202	ipl_t ipl;
		1203	as_area_t *area;
		1204	uintptr_t base, page;
		1205	uintptr_t old_frame, frame;
		1206	size_t size;
		1207	int flags;
		1208	int page_flags;
		1209	pte_t *pte;
		1210	int rc;
		1211	uintptr_t *pagemap;
		1212
		1213	ipl = interrupts_disable();
		1214	mutex_lock(&AS->lock);
		1215	area = find_area_and_lock(AS, address);
		1216	if (!area) {
		1217	/*
		1218	* Could not find the address space area.
		1219	*/
		1220	mutex_unlock(&AS->lock);
		1221	interrupts_restore(ipl);
		1222	return ENOENT;
		1223	}
		1224
		1225	if (area->backend == &anon_backend && !area->sh_info) {
		1226	/* Nothing to do */
		1227	mutex_unlock(&area->lock);
		1228	mutex_unlock(&AS->lock);
		1229	interrupts_restore(ipl);
		1230	return EOK;
		1231	}
		1232
		1233	base = area->base;
		1234	size = area->pages * PAGE_SIZE;
		1235	flags = area->flags;
		1236	page_flags = as_area_get_flags(area);
		1237
		1238	pagemap = malloc(area->pages * sizeof(uintptr_t), 0);
		1239	page_table_lock(AS, false);
		1240
		1241	for (page = base; page < base + size; page += PAGE_SIZE) {
		1242	pte = page_mapping_find(AS, page);
		1243	if (!pte \|\| !PTE_PRESENT(pte) \|\| !PTE_READABLE(pte)) {
		1244	/* Fetch the missing page */
		1245	if (!area->backend \|\| !area->backend->page_fault) {
		1246	page_table_unlock(AS, false);
		1247	mutex_unlock(&area->lock);
		1248	mutex_unlock(&AS->lock);
		1249	interrupts_restore(ipl);
		1250	return EINVAL;
		1251	}
		1252	if (area->backend->page_fault(area, page, PF_ACCESS_READ) != AS_PF_OK) {
		1253	page_table_unlock(AS, false);
		1254	mutex_unlock(&area->lock);
		1255	mutex_unlock(&AS->lock);
		1256	interrupts_restore(ipl);
		1257	return EINVAL;
		1258	}
		1259	}
		1260	ASSERT(PTE_VALID(pte));
		1261
		1262	old_frame = PTE_GET_FRAME(pte);
		1263
		1264	frame = (uintptr_t)frame_alloc(ONE_FRAME, 0);
		1265	memcpy((void ) PA2KA(frame), (void )PA2KA(old_frame),
		1266	FRAME_SIZE);
		1267
		1268	pagemap[(page - base) / PAGE_SIZE] = frame;
		1269	}
		1270
		1271	page_table_unlock(AS, false);
		1272	mutex_unlock(&area->lock);
		1273	mutex_unlock(&AS->lock);
		1274	interrupts_restore(ipl);
		1275
		1276	rc = as_area_destroy(AS, address);
		1277	if (rc < 0) {
		1278	free(pagemap);
		1279	return rc;
		1280	}
		1281
		1282	area = as_area_create(AS, flags, size, base, AS_AREA_ATTR_PARTIAL,
		1283	&anon_backend, NULL);
		1284	if (area == NULL) {
		1285	free(pagemap);
		1286	return rc;
		1287	}
		1288
		1289	mutex_lock(&AS->lock);
		1290	mutex_lock(&area->lock);
		1291	page_table_lock(AS, false);
		1292	for (page = base; page < base + size; page += PAGE_SIZE) {
		1293	frame = pagemap[(page - base) / PAGE_SIZE];
		1294
		1295	page_mapping_insert(AS, page, frame, page_flags);
		1296	if (!used_space_insert(area, page, 1))
		1297	panic("Could not insert used space.\n");
		1298	}
		1299
		1300	page_table_unlock(AS, false);
		1301
		1302	area->attributes &= ~AS_AREA_ATTR_PARTIAL;
		1303
		1304	mutex_unlock(&area->lock);
		1305	mutex_unlock(&AS->lock);
		1306
		1307	free(pagemap);
		1308
		1309	return EOK;
		1310	}
		1311
3431	svoboda	1312	#endif /* defined(CONFIG_UDEBUG) */
		1313
1235	jermar	1314	/** Convert address space area flags to page flags.
754	jermar	1315	*
3425	svoboda	1316	* @param aflags Flags of some address space area.
754	jermar	1317	*
3425	svoboda	1318	* @return Flags to be passed to page_mapping_insert().
754	jermar	1319	*/
1235	jermar	1320	int area_flags_to_page_flags(int aflags)
754	jermar	1321	{
		1322	int flags;
		1323
1178	jermar	1324	flags = PAGE_USER \| PAGE_PRESENT;
754	jermar	1325
1235	jermar	1326	if (aflags & AS_AREA_READ)
1026	jermar	1327	flags \|= PAGE_READ;
		1328
1235	jermar	1329	if (aflags & AS_AREA_WRITE)
1026	jermar	1330	flags \|= PAGE_WRITE;
		1331
1235	jermar	1332	if (aflags & AS_AREA_EXEC)
1026	jermar	1333	flags \|= PAGE_EXEC;
		1334
1424	jermar	1335	if (aflags & AS_AREA_CACHEABLE)
1178	jermar	1336	flags \|= PAGE_CACHEABLE;
		1337
754	jermar	1338	return flags;
		1339	}
756	jermar	1340
1235	jermar	1341	/** Compute flags for virtual address translation subsytem.
		1342	*
		1343	* The address space area must be locked.
		1344	* Interrupts must be disabled.
		1345	*
3425	svoboda	1346	* @param a Address space area.
1235	jermar	1347	*
3425	svoboda	1348	* @return Flags to be used in page_mapping_insert().
1235	jermar	1349	*/
1409	jermar	1350	int as_area_get_flags(as_area_t *a)
1235	jermar	1351	{
		1352	return area_flags_to_page_flags(a->flags);
		1353	}
		1354
756	jermar	1355	/** Create page table.
		1356	*
3425	svoboda	1357	* Depending on architecture, create either address space private or global page
		1358	* table.
756	jermar	1359	*
3425	svoboda	1360	* @param flags Flags saying whether the page table is for the kernel
		1361	* address space.
756	jermar	1362	*
3425	svoboda	1363	* @return First entry of the page table.
756	jermar	1364	*/
		1365	pte_t *page_table_create(int flags)
		1366	{
2125	decky	1367	ASSERT(as_operations);
		1368	ASSERT(as_operations->page_table_create);
		1369
		1370	return as_operations->page_table_create(flags);
756	jermar	1371	}
977	jermar	1372
1468	jermar	1373	/** Destroy page table.
		1374	*
		1375	* Destroy page table in architecture specific way.
		1376	*
3425	svoboda	1377	* @param page_table Physical address of PTL0.
1468	jermar	1378	*/
		1379	void page_table_destroy(pte_t *page_table)
		1380	{
2125	decky	1381	ASSERT(as_operations);
		1382	ASSERT(as_operations->page_table_destroy);
		1383
		1384	as_operations->page_table_destroy(page_table);
1468	jermar	1385	}
		1386
1044	jermar	1387	/** Lock page table.
		1388	*
		1389	* This function should be called before any page_mapping_insert(),
		1390	* page_mapping_remove() and page_mapping_find().
		1391	*
		1392	* Locking order is such that address space areas must be locked
		1393	* prior to this call. Address space can be locked prior to this
		1394	* call in which case the lock argument is false.
		1395	*
3425	svoboda	1396	* @param as Address space.
		1397	* @param lock If false, do not attempt to lock as->lock.
1044	jermar	1398	*/
		1399	void page_table_lock(as_t *as, bool lock)
		1400	{
		1401	ASSERT(as_operations);
		1402	ASSERT(as_operations->page_table_lock);
2125	decky	1403
1044	jermar	1404	as_operations->page_table_lock(as, lock);
		1405	}
		1406
		1407	/** Unlock page table.
		1408	*
3425	svoboda	1409	* @param as Address space.
		1410	* @param unlock If false, do not attempt to unlock as->lock.
1044	jermar	1411	*/
		1412	void page_table_unlock(as_t *as, bool unlock)
		1413	{
		1414	ASSERT(as_operations);
		1415	ASSERT(as_operations->page_table_unlock);
2125	decky	1416
1044	jermar	1417	as_operations->page_table_unlock(as, unlock);
		1418	}
		1419
977	jermar	1420
		1421	/** Find address space area and lock it.
		1422	*
		1423	* The address space must be locked and interrupts must be disabled.
		1424	*
3425	svoboda	1425	* @param as Address space.
		1426	* @param va Virtual address.
977	jermar	1427	*
3425	svoboda	1428	* @return Locked address space area containing va on success or
		1429	* NULL on failure.
977	jermar	1430	*/
1780	jermar	1431	as_area_t find_area_and_lock(as_t as, uintptr_t va)
977	jermar	1432	{
		1433	as_area_t *a;
1147	jermar	1434	btree_node_t leaf, lnode;
2745	decky	1435	unsigned int i;
977	jermar	1436
1147	jermar	1437	a = (as_area_t *) btree_search(&as->as_area_btree, va, &leaf);
		1438	if (a) {
		1439	/* va is the base address of an address space area */
1380	jermar	1440	mutex_lock(&a->lock);
1147	jermar	1441	return a;
		1442	}
		1443
		1444	/*
1150	jermar	1445	* Search the leaf node and the righmost record of its left neighbour
1147	jermar	1446	* to find out whether this is a miss or va belongs to an address
		1447	* space area found there.
		1448	*/
		1449
		1450	/* First, search the leaf node itself. */
		1451	for (i = 0; i < leaf->keys; i++) {
		1452	a = (as_area_t *) leaf->value[i];
1380	jermar	1453	mutex_lock(&a->lock);
1147	jermar	1454	if ((a->base <= va) && (va < a->base + a->pages * PAGE_SIZE)) {
		1455	return a;
		1456	}
1380	jermar	1457	mutex_unlock(&a->lock);
1147	jermar	1458	}
977	jermar	1459
1147	jermar	1460	/*
1150	jermar	1461	* Second, locate the left neighbour and test its last record.
1148	jermar	1462	* Because of its position in the B+tree, it must have base < va.
1147	jermar	1463	*/
2087	jermar	1464	lnode = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf);
		1465	if (lnode) {
1147	jermar	1466	a = (as_area_t *) lnode->value[lnode->keys - 1];
1380	jermar	1467	mutex_lock(&a->lock);
1147	jermar	1468	if (va < a->base + a->pages * PAGE_SIZE) {
1048	jermar	1469	return a;
1147	jermar	1470	}
1380	jermar	1471	mutex_unlock(&a->lock);
977	jermar	1472	}
		1473
		1474	return NULL;
		1475	}
1048	jermar	1476
		1477	/** Check area conflicts with other areas.
		1478	*
		1479	* The address space must be locked and interrupts must be disabled.
		1480	*
3425	svoboda	1481	* @param as Address space.
		1482	* @param va Starting virtual address of the area being tested.
		1483	* @param size Size of the area being tested.
		1484	* @param avoid_area Do not touch this area.
1048	jermar	1485	*
3425	svoboda	1486	* @return True if there is no conflict, false otherwise.
1048	jermar	1487	*/
3425	svoboda	1488	bool
		1489	check_area_conflicts(as_t as, uintptr_t va, size_t size, as_area_t avoid_area)
1048	jermar	1490	{
		1491	as_area_t *a;
1147	jermar	1492	btree_node_t leaf, node;
2745	decky	1493	unsigned int i;
1048	jermar	1494
1070	jermar	1495	/*
		1496	* We don't want any area to have conflicts with NULL page.
		1497	*/
		1498	if (overlaps(va, size, NULL, PAGE_SIZE))
		1499	return false;
		1500
1147	jermar	1501	/*
		1502	* The leaf node is found in O(log n), where n is proportional to
		1503	* the number of address space areas belonging to as.
		1504	* The check for conflicts is then attempted on the rightmost
1150	jermar	1505	* record in the left neighbour, the leftmost record in the right
		1506	* neighbour and all records in the leaf node itself.
1147	jermar	1507	*/
1048	jermar	1508
1147	jermar	1509	if ((a = (as_area_t *) btree_search(&as->as_area_btree, va, &leaf))) {
		1510	if (a != avoid_area)
		1511	return false;
		1512	}
		1513
		1514	/* First, check the two border cases. */
1150	jermar	1515	if ((node = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf))) {
1147	jermar	1516	a = (as_area_t *) node->value[node->keys - 1];
1380	jermar	1517	mutex_lock(&a->lock);
1147	jermar	1518	if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
1380	jermar	1519	mutex_unlock(&a->lock);
1147	jermar	1520	return false;
		1521	}
1380	jermar	1522	mutex_unlock(&a->lock);
1147	jermar	1523	}
2087	jermar	1524	node = btree_leaf_node_right_neighbour(&as->as_area_btree, leaf);
		1525	if (node) {
1147	jermar	1526	a = (as_area_t *) node->value[0];
1380	jermar	1527	mutex_lock(&a->lock);
1147	jermar	1528	if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
1380	jermar	1529	mutex_unlock(&a->lock);
1147	jermar	1530	return false;
		1531	}
1380	jermar	1532	mutex_unlock(&a->lock);
1147	jermar	1533	}
		1534
		1535	/* Second, check the leaf node. */
		1536	for (i = 0; i < leaf->keys; i++) {
		1537	a = (as_area_t *) leaf->value[i];
		1538
1048	jermar	1539	if (a == avoid_area)
		1540	continue;
1147	jermar	1541
1380	jermar	1542	mutex_lock(&a->lock);
1147	jermar	1543	if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
1380	jermar	1544	mutex_unlock(&a->lock);
1147	jermar	1545	return false;
		1546	}
1380	jermar	1547	mutex_unlock(&a->lock);
1048	jermar	1548	}
		1549
1070	jermar	1550	/*
		1551	* So far, the area does not conflict with other areas.
		1552	* Check if it doesn't conflict with kernel address space.
		1553	*/
		1554	if (!KERNEL_ADDRESS_SPACE_SHADOWED) {
		1555	return !overlaps(va, size,
2087	jermar	1556	KERNEL_ADDRESS_SPACE_START,
		1557	KERNEL_ADDRESS_SPACE_END - KERNEL_ADDRESS_SPACE_START);
1070	jermar	1558	}
		1559
1048	jermar	1560	return true;
		1561	}
1235	jermar	1562
2556	jermar	1563	/** Return size of the address space area with given base.
		1564	*
		1565	* @param base Arbitrary address insede the address space area.
		1566	*
		1567	* @return Size of the address space area in bytes or zero if it
		1568	* does not exist.
		1569	*/
		1570	size_t as_area_get_size(uintptr_t base)
1329	palkovsky	1571	{
		1572	ipl_t ipl;
		1573	as_area_t *src_area;
		1574	size_t size;
		1575
		1576	ipl = interrupts_disable();
		1577	src_area = find_area_and_lock(AS, base);
3425	svoboda	1578	if (src_area) {
1329	palkovsky	1579	size = src_area->pages * PAGE_SIZE;
1380	jermar	1580	mutex_unlock(&src_area->lock);
1329	palkovsky	1581	} else {
		1582	size = 0;
		1583	}
		1584	interrupts_restore(ipl);
		1585	return size;
		1586	}
		1587
1387	jermar	1588	/** Mark portion of address space area as used.
		1589	*
		1590	* The address space area must be already locked.
		1591	*
3425	svoboda	1592	* @param a Address space area.
		1593	* @param page First page to be marked.
		1594	* @param count Number of page to be marked.
1387	jermar	1595	*
3425	svoboda	1596	* @return Zero on failure and non-zero on success.
1387	jermar	1597	*/
1780	jermar	1598	int used_space_insert(as_area_t *a, uintptr_t page, count_t count)
1387	jermar	1599	{
		1600	btree_node_t leaf, node;
		1601	count_t pages;
2745	decky	1602	unsigned int i;
1387	jermar	1603
		1604	ASSERT(page == ALIGN_DOWN(page, PAGE_SIZE));
		1605	ASSERT(count);
		1606
		1607	pages = (count_t) btree_search(&a->used_space, page, &leaf);
		1608	if (pages) {
		1609	/*
		1610	* We hit the beginning of some used space.
		1611	*/
		1612	return 0;
		1613	}
		1614
1437	jermar	1615	if (!leaf->keys) {
		1616	btree_insert(&a->used_space, page, (void *) count, leaf);
		1617	return 1;
		1618	}
		1619
1387	jermar	1620	node = btree_leaf_node_left_neighbour(&a->used_space, leaf);
		1621	if (node) {
2087	jermar	1622	uintptr_t left_pg = node->key[node->keys - 1];
		1623	uintptr_t right_pg = leaf->key[0];
		1624	count_t left_cnt = (count_t) node->value[node->keys - 1];
		1625	count_t right_cnt = (count_t) leaf->value[0];
1387	jermar	1626
		1627	/*
		1628	* Examine the possibility that the interval fits
		1629	* somewhere between the rightmost interval of
		1630	* the left neigbour and the first interval of the leaf.
		1631	*/
		1632
		1633	if (page >= right_pg) {
		1634	/* Do nothing. */
2087	jermar	1635	} else if (overlaps(page, count * PAGE_SIZE, left_pg,
		1636	left_cnt * PAGE_SIZE)) {
1387	jermar	1637	/* The interval intersects with the left interval. */
		1638	return 0;
2087	jermar	1639	} else if (overlaps(page, count * PAGE_SIZE, right_pg,
		1640	right_cnt * PAGE_SIZE)) {
1387	jermar	1641	/* The interval intersects with the right interval. */
		1642	return 0;
2087	jermar	1643	} else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
		1644	(page + count * PAGE_SIZE == right_pg)) {
		1645	/*
		1646	* The interval can be added by merging the two already
		1647	* present intervals.
		1648	*/
1403	jermar	1649	node->value[node->keys - 1] += count + right_cnt;
1387	jermar	1650	btree_remove(&a->used_space, right_pg, leaf);
		1651	return 1;
2087	jermar	1652	} else if (page == left_pg + left_cnt * PAGE_SIZE) {
		1653	/*
		1654	* The interval can be added by simply growing the left
		1655	* interval.
		1656	*/
1403	jermar	1657	node->value[node->keys - 1] += count;
1387	jermar	1658	return 1;
2087	jermar	1659	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1660	/*
2087	jermar	1661	* The interval can be addded by simply moving base of
		1662	* the right interval down and increasing its size
		1663	* accordingly.
1387	jermar	1664	*/
1403	jermar	1665	leaf->value[0] += count;
1387	jermar	1666	leaf->key[0] = page;
		1667	return 1;
		1668	} else {
		1669	/*
		1670	* The interval is between both neigbouring intervals,
		1671	* but cannot be merged with any of them.
		1672	*/
2087	jermar	1673	btree_insert(&a->used_space, page, (void *) count,
		1674	leaf);
1387	jermar	1675	return 1;
		1676	}
		1677	} else if (page < leaf->key[0]) {
1780	jermar	1678	uintptr_t right_pg = leaf->key[0];
1387	jermar	1679	count_t right_cnt = (count_t) leaf->value[0];
		1680
		1681	/*
2087	jermar	1682	* Investigate the border case in which the left neighbour does
		1683	* not exist but the interval fits from the left.
1387	jermar	1684	*/
		1685
2087	jermar	1686	if (overlaps(page, count * PAGE_SIZE, right_pg,
		1687	right_cnt * PAGE_SIZE)) {
1387	jermar	1688	/* The interval intersects with the right interval. */
		1689	return 0;
2087	jermar	1690	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1691	/*
2087	jermar	1692	* The interval can be added by moving the base of the
		1693	* right interval down and increasing its size
		1694	* accordingly.
1387	jermar	1695	*/
		1696	leaf->key[0] = page;
1403	jermar	1697	leaf->value[0] += count;
1387	jermar	1698	return 1;
		1699	} else {
		1700	/*
		1701	* The interval doesn't adjoin with the right interval.
		1702	* It must be added individually.
		1703	*/
2087	jermar	1704	btree_insert(&a->used_space, page, (void *) count,
		1705	leaf);
1387	jermar	1706	return 1;
		1707	}
		1708	}
		1709
		1710	node = btree_leaf_node_right_neighbour(&a->used_space, leaf);
		1711	if (node) {
2087	jermar	1712	uintptr_t left_pg = leaf->key[leaf->keys - 1];
		1713	uintptr_t right_pg = node->key[0];
		1714	count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
		1715	count_t right_cnt = (count_t) node->value[0];
1387	jermar	1716
		1717	/*
		1718	* Examine the possibility that the interval fits
		1719	* somewhere between the leftmost interval of
		1720	* the right neigbour and the last interval of the leaf.
		1721	*/
		1722
		1723	if (page < left_pg) {
		1724	/* Do nothing. */
2087	jermar	1725	} else if (overlaps(page, count * PAGE_SIZE, left_pg,
		1726	left_cnt * PAGE_SIZE)) {
1387	jermar	1727	/* The interval intersects with the left interval. */
		1728	return 0;
2087	jermar	1729	} else if (overlaps(page, count * PAGE_SIZE, right_pg,
		1730	right_cnt * PAGE_SIZE)) {
1387	jermar	1731	/* The interval intersects with the right interval. */
		1732	return 0;
2087	jermar	1733	} else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
		1734	(page + count * PAGE_SIZE == right_pg)) {
		1735	/*
		1736	* The interval can be added by merging the two already
		1737	* present intervals.
		1738	* */
1403	jermar	1739	leaf->value[leaf->keys - 1] += count + right_cnt;
1387	jermar	1740	btree_remove(&a->used_space, right_pg, node);
		1741	return 1;
2087	jermar	1742	} else if (page == left_pg + left_cnt * PAGE_SIZE) {
		1743	/*
		1744	* The interval can be added by simply growing the left
		1745	* interval.
		1746	* */
1403	jermar	1747	leaf->value[leaf->keys - 1] += count;
1387	jermar	1748	return 1;
2087	jermar	1749	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1750	/*
2087	jermar	1751	* The interval can be addded by simply moving base of
		1752	* the right interval down and increasing its size
		1753	* accordingly.
1387	jermar	1754	*/
1403	jermar	1755	node->value[0] += count;
1387	jermar	1756	node->key[0] = page;
		1757	return 1;
		1758	} else {
		1759	/*
		1760	* The interval is between both neigbouring intervals,
		1761	* but cannot be merged with any of them.
		1762	*/
2087	jermar	1763	btree_insert(&a->used_space, page, (void *) count,
		1764	leaf);
1387	jermar	1765	return 1;
		1766	}
		1767	} else if (page >= leaf->key[leaf->keys - 1]) {
1780	jermar	1768	uintptr_t left_pg = leaf->key[leaf->keys - 1];
1387	jermar	1769	count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
		1770
		1771	/*
2087	jermar	1772	* Investigate the border case in which the right neighbour
		1773	* does not exist but the interval fits from the right.
1387	jermar	1774	*/
		1775
2087	jermar	1776	if (overlaps(page, count * PAGE_SIZE, left_pg,
		1777	left_cnt * PAGE_SIZE)) {
1403	jermar	1778	/* The interval intersects with the left interval. */
1387	jermar	1779	return 0;
2087	jermar	1780	} else if (left_pg + left_cnt * PAGE_SIZE == page) {
		1781	/*
		1782	* The interval can be added by growing the left
		1783	* interval.
		1784	*/
1403	jermar	1785	leaf->value[leaf->keys - 1] += count;
1387	jermar	1786	return 1;
		1787	} else {
		1788	/*
		1789	* The interval doesn't adjoin with the left interval.
		1790	* It must be added individually.
		1791	*/
2087	jermar	1792	btree_insert(&a->used_space, page, (void *) count,
		1793	leaf);
1387	jermar	1794	return 1;
		1795	}
		1796	}
		1797
		1798	/*
2087	jermar	1799	* Note that if the algorithm made it thus far, the interval can fit
		1800	* only between two other intervals of the leaf. The two border cases
		1801	* were already resolved.
1387	jermar	1802	*/
		1803	for (i = 1; i < leaf->keys; i++) {
		1804	if (page < leaf->key[i]) {
2087	jermar	1805	uintptr_t left_pg = leaf->key[i - 1];
		1806	uintptr_t right_pg = leaf->key[i];
		1807	count_t left_cnt = (count_t) leaf->value[i - 1];
		1808	count_t right_cnt = (count_t) leaf->value[i];
1387	jermar	1809
		1810	/*
		1811	* The interval fits between left_pg and right_pg.
		1812	*/
		1813
2087	jermar	1814	if (overlaps(page, count * PAGE_SIZE, left_pg,
		1815	left_cnt * PAGE_SIZE)) {
		1816	/*
		1817	* The interval intersects with the left
		1818	* interval.
		1819	*/
1387	jermar	1820	return 0;
2087	jermar	1821	} else if (overlaps(page, count * PAGE_SIZE, right_pg,
		1822	right_cnt * PAGE_SIZE)) {
		1823	/*
		1824	* The interval intersects with the right
		1825	* interval.
		1826	*/
1387	jermar	1827	return 0;
2087	jermar	1828	} else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
		1829	(page + count * PAGE_SIZE == right_pg)) {
		1830	/*
		1831	* The interval can be added by merging the two
		1832	* already present intervals.
		1833	*/
1403	jermar	1834	leaf->value[i - 1] += count + right_cnt;
1387	jermar	1835	btree_remove(&a->used_space, right_pg, leaf);
		1836	return 1;
2087	jermar	1837	} else if (page == left_pg + left_cnt * PAGE_SIZE) {
		1838	/*
		1839	* The interval can be added by simply growing
		1840	* the left interval.
		1841	*/
1403	jermar	1842	leaf->value[i - 1] += count;
1387	jermar	1843	return 1;
2087	jermar	1844	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1845	/*
2087	jermar	1846	* The interval can be addded by simply moving
		1847	* base of the right interval down and
		1848	* increasing its size accordingly.
1387	jermar	1849	*/
1403	jermar	1850	leaf->value[i] += count;
1387	jermar	1851	leaf->key[i] = page;
		1852	return 1;
		1853	} else {
		1854	/*
2087	jermar	1855	* The interval is between both neigbouring
		1856	* intervals, but cannot be merged with any of
		1857	* them.
1387	jermar	1858	*/
2087	jermar	1859	btree_insert(&a->used_space, page,
		1860	(void *) count, leaf);
1387	jermar	1861	return 1;
		1862	}
		1863	}
		1864	}
		1865
3425	svoboda	1866	panic("Inconsistency detected while adding %" PRIc " pages of used "
		1867	"space at %p.\n", count, page);
1387	jermar	1868	}
		1869
		1870	/** Mark portion of address space area as unused.
		1871	*
		1872	* The address space area must be already locked.
		1873	*
3425	svoboda	1874	* @param a Address space area.
		1875	* @param page First page to be marked.
		1876	* @param count Number of page to be marked.
1387	jermar	1877	*
3425	svoboda	1878	* @return Zero on failure and non-zero on success.
1387	jermar	1879	*/
1780	jermar	1880	int used_space_remove(as_area_t *a, uintptr_t page, count_t count)
1387	jermar	1881	{
		1882	btree_node_t leaf, node;
		1883	count_t pages;
2745	decky	1884	unsigned int i;
1387	jermar	1885
		1886	ASSERT(page == ALIGN_DOWN(page, PAGE_SIZE));
		1887	ASSERT(count);
		1888
		1889	pages = (count_t) btree_search(&a->used_space, page, &leaf);
		1890	if (pages) {
		1891	/*
		1892	* We are lucky, page is the beginning of some interval.
		1893	*/
		1894	if (count > pages) {
		1895	return 0;
		1896	} else if (count == pages) {
		1897	btree_remove(&a->used_space, page, leaf);
1403	jermar	1898	return 1;
1387	jermar	1899	} else {
		1900	/*
		1901	* Find the respective interval.
		1902	* Decrease its size and relocate its start address.
		1903	*/
		1904	for (i = 0; i < leaf->keys; i++) {
		1905	if (leaf->key[i] == page) {
2087	jermar	1906	leaf->key[i] += count * PAGE_SIZE;
1403	jermar	1907	leaf->value[i] -= count;
1387	jermar	1908	return 1;
		1909	}
		1910	}
		1911	goto error;
		1912	}
		1913	}
		1914
		1915	node = btree_leaf_node_left_neighbour(&a->used_space, leaf);
		1916	if (node && page < leaf->key[0]) {
1780	jermar	1917	uintptr_t left_pg = node->key[node->keys - 1];
1387	jermar	1918	count_t left_cnt = (count_t) node->value[node->keys - 1];
		1919
2087	jermar	1920	if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
		1921	count * PAGE_SIZE)) {
		1922	if (page + count * PAGE_SIZE ==
		1923	left_pg + left_cnt * PAGE_SIZE) {
1387	jermar	1924	/*
2087	jermar	1925	* The interval is contained in the rightmost
		1926	* interval of the left neighbour and can be
		1927	* removed by updating the size of the bigger
		1928	* interval.
1387	jermar	1929	*/
1403	jermar	1930	node->value[node->keys - 1] -= count;
1387	jermar	1931	return 1;
2087	jermar	1932	} else if (page + count * PAGE_SIZE <
		1933	left_pg + left_cnt*PAGE_SIZE) {
1403	jermar	1934	count_t new_cnt;
1387	jermar	1935
		1936	/*
2087	jermar	1937	* The interval is contained in the rightmost
		1938	* interval of the left neighbour but its
		1939	* removal requires both updating the size of
		1940	* the original interval and also inserting a
		1941	* new interval.
1387	jermar	1942	*/
2087	jermar	1943	new_cnt = ((left_pg + left_cnt * PAGE_SIZE) -
		1944	(page + count*PAGE_SIZE)) >> PAGE_WIDTH;
1403	jermar	1945	node->value[node->keys - 1] -= count + new_cnt;
2087	jermar	1946	btree_insert(&a->used_space, page +
		1947	count * PAGE_SIZE, (void *) new_cnt, leaf);
1387	jermar	1948	return 1;
		1949	}
		1950	}
		1951	return 0;
		1952	} else if (page < leaf->key[0]) {
		1953	return 0;
		1954	}
		1955
		1956	if (page > leaf->key[leaf->keys - 1]) {
1780	jermar	1957	uintptr_t left_pg = leaf->key[leaf->keys - 1];
1387	jermar	1958	count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
		1959
2087	jermar	1960	if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
		1961	count * PAGE_SIZE)) {
		1962	if (page + count * PAGE_SIZE ==
		1963	left_pg + left_cnt * PAGE_SIZE) {
1387	jermar	1964	/*
2087	jermar	1965	* The interval is contained in the rightmost
		1966	* interval of the leaf and can be removed by
		1967	* updating the size of the bigger interval.
1387	jermar	1968	*/
1403	jermar	1969	leaf->value[leaf->keys - 1] -= count;
1387	jermar	1970	return 1;
2087	jermar	1971	} else if (page + count * PAGE_SIZE < left_pg +
		1972	left_cnt * PAGE_SIZE) {
1403	jermar	1973	count_t new_cnt;
1387	jermar	1974
		1975	/*
2087	jermar	1976	* The interval is contained in the rightmost
		1977	* interval of the leaf but its removal
		1978	* requires both updating the size of the
		1979	* original interval and also inserting a new
		1980	* interval.
1387	jermar	1981	*/
2087	jermar	1982	new_cnt = ((left_pg + left_cnt * PAGE_SIZE) -
		1983	(page + count * PAGE_SIZE)) >> PAGE_WIDTH;
1403	jermar	1984	leaf->value[leaf->keys - 1] -= count + new_cnt;
2087	jermar	1985	btree_insert(&a->used_space, page +
		1986	count * PAGE_SIZE, (void *) new_cnt, leaf);
1387	jermar	1987	return 1;
		1988	}
		1989	}
		1990	return 0;
		1991	}
		1992
		1993	/*
		1994	* The border cases have been already resolved.
		1995	* Now the interval can be only between intervals of the leaf.
		1996	*/
		1997	for (i = 1; i < leaf->keys - 1; i++) {
		1998	if (page < leaf->key[i]) {
1780	jermar	1999	uintptr_t left_pg = leaf->key[i - 1];
1387	jermar	2000	count_t left_cnt = (count_t) leaf->value[i - 1];
		2001
		2002	/*
2087	jermar	2003	* Now the interval is between intervals corresponding
		2004	* to (i - 1) and i.
1387	jermar	2005	*/
2087	jermar	2006	if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
		2007	count * PAGE_SIZE)) {
		2008	if (page + count * PAGE_SIZE ==
		2009	left_pg + left_cnt*PAGE_SIZE) {
1387	jermar	2010	/*
2087	jermar	2011	* The interval is contained in the
		2012	* interval (i - 1) of the leaf and can
		2013	* be removed by updating the size of
		2014	* the bigger interval.
1387	jermar	2015	*/
1403	jermar	2016	leaf->value[i - 1] -= count;
1387	jermar	2017	return 1;
2087	jermar	2018	} else if (page + count * PAGE_SIZE <
		2019	left_pg + left_cnt * PAGE_SIZE) {
1403	jermar	2020	count_t new_cnt;
1387	jermar	2021
		2022	/*
2087	jermar	2023	* The interval is contained in the
		2024	* interval (i - 1) of the leaf but its
		2025	* removal requires both updating the
		2026	* size of the original interval and
1387	jermar	2027	* also inserting a new interval.
		2028	*/
2087	jermar	2029	new_cnt = ((left_pg +
		2030	left_cnt * PAGE_SIZE) -
		2031	(page + count * PAGE_SIZE)) >>
		2032	PAGE_WIDTH;
1403	jermar	2033	leaf->value[i - 1] -= count + new_cnt;
2087	jermar	2034	btree_insert(&a->used_space, page +
		2035	count * PAGE_SIZE, (void *) new_cnt,
		2036	leaf);
1387	jermar	2037	return 1;
		2038	}
		2039	}
		2040	return 0;
		2041	}
		2042	}
		2043
		2044	error:
3425	svoboda	2045	panic("Inconsistency detected while removing %" PRIc " pages of used "
		2046	"space from %p.\n", count, page);
1387	jermar	2047	}
		2048
1409	jermar	2049	/** Remove reference to address space area share info.
		2050	*
		2051	* If the reference count drops to 0, the sh_info is deallocated.
		2052	*
3425	svoboda	2053	* @param sh_info Pointer to address space area share info.
1409	jermar	2054	*/
		2055	void sh_info_remove_reference(share_info_t *sh_info)
		2056	{
		2057	bool dealloc = false;
		2058
		2059	mutex_lock(&sh_info->lock);
		2060	ASSERT(sh_info->refcount);
		2061	if (--sh_info->refcount == 0) {
		2062	dealloc = true;
1495	jermar	2063	link_t *cur;
1409	jermar	2064
		2065	/*
		2066	* Now walk carefully the pagemap B+tree and free/remove
		2067	* reference from all frames found there.
		2068	*/
2087	jermar	2069	for (cur = sh_info->pagemap.leaf_head.next;
		2070	cur != &sh_info->pagemap.leaf_head; cur = cur->next) {
1409	jermar	2071	btree_node_t *node;
2745	decky	2072	unsigned int i;
1409	jermar	2073
1495	jermar	2074	node = list_get_instance(cur, btree_node_t, leaf_link);
		2075	for (i = 0; i < node->keys; i++)
1780	jermar	2076	frame_free((uintptr_t) node->value[i]);
1409	jermar	2077	}
		2078
		2079	}
		2080	mutex_unlock(&sh_info->lock);
		2081
		2082	if (dealloc) {
		2083	btree_destroy(&sh_info->pagemap);
		2084	free(sh_info);
		2085	}
		2086	}
		2087
1235	jermar	2088	/*
		2089	* Address space related syscalls.
		2090	*/
		2091
		2092	/** Wrapper for as_area_create(). */
1780	jermar	2093	unative_t sys_as_area_create(uintptr_t address, size_t size, int flags)
1235	jermar	2094	{
2087	jermar	2095	if (as_area_create(AS, flags \| AS_AREA_CACHEABLE, size, address,
		2096	AS_AREA_ATTR_NONE, &anon_backend, NULL))
1780	jermar	2097	return (unative_t) address;
1235	jermar	2098	else
1780	jermar	2099	return (unative_t) -1;
1235	jermar	2100	}
		2101
1793	jermar	2102	/** Wrapper for as_area_resize(). */
1780	jermar	2103	unative_t sys_as_area_resize(uintptr_t address, size_t size, int flags)
1235	jermar	2104	{
1780	jermar	2105	return (unative_t) as_area_resize(AS, address, size, 0);
1235	jermar	2106	}
		2107
3425	svoboda	2108	/** Wrapper for as_area_change_flags(). */
		2109	unative_t sys_as_area_change_flags(uintptr_t address, int flags)
		2110	{
		2111	return (unative_t) as_area_change_flags(AS, flags, address);
		2112	}
		2113
1793	jermar	2114	/** Wrapper for as_area_destroy(). */
1780	jermar	2115	unative_t sys_as_area_destroy(uintptr_t address)
1306	jermar	2116	{
1780	jermar	2117	return (unative_t) as_area_destroy(AS, address);
1306	jermar	2118	}
1702	cejka	2119
1914	jermar	2120	/** Print out information about address space.
		2121	*
3425	svoboda	2122	* @param as Address space.
1914	jermar	2123	*/
		2124	void as_print(as_t *as)
		2125	{
		2126	ipl_t ipl;
		2127
		2128	ipl = interrupts_disable();
		2129	mutex_lock(&as->lock);
		2130
		2131	/* print out info about address space areas */
		2132	link_t *cur;
2087	jermar	2133	for (cur = as->as_area_btree.leaf_head.next;
		2134	cur != &as->as_area_btree.leaf_head; cur = cur->next) {
		2135	btree_node_t *node;
1914	jermar	2136
2087	jermar	2137	node = list_get_instance(cur, btree_node_t, leaf_link);
		2138
2745	decky	2139	unsigned int i;
1914	jermar	2140	for (i = 0; i < node->keys; i++) {
1915	jermar	2141	as_area_t *area = node->value[i];
1914	jermar	2142
		2143	mutex_lock(&area->lock);
3425	svoboda	2144	printf("as_area: %p, base=%p, pages=%" PRIc
		2145	" (%p - %p)\n", area, area->base, area->pages,
		2146	area->base, area->base + FRAMES2SIZE(area->pages));
1914	jermar	2147	mutex_unlock(&area->lock);
		2148	}
		2149	}
		2150
		2151	mutex_unlock(&as->lock);
		2152	interrupts_restore(ipl);
		2153	}
		2154
1757	jermar	2155	/** @}
1702	cejka	2156	*/

Subversion Repositories HelenOS

(root)/branches/tracing/kernel/generic/src/mm/as.c – Rev 3596