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

Subversion Repositories HelenOS

(root)/branches/tracing/kernel/generic/src/mm/as.c @ 3596 – Rev 3036