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
3013	svoboda	1110	/** Write directly into a page, bypassing area flags.
		1111	*
		1112	* This allows a debugger to write into a page that is mapped read-only
		1113	* (such as the text segment). Naturally, this is only possible if the
		1114	* correspoinding area is not shared and anonymous.
		1115	*
		1116	* FIXME: doesn't take into account that it isn't a good idea to write
		1117	* into the frame if the area is shared or isn't anonymous
		1118	*/
		1119	static int debug_write_inside_page(uintptr_t va, void *data, size_t n)
		1120	{
		1121	uintptr_t page;
		1122	pte_t *pte;
		1123	as_area_t *area;
		1124	uintptr_t frame;
		1125	ipl_t ipl;
		1126	int rc;
		1127
		1128	page = ALIGN_DOWN(va, PAGE_SIZE);
		1129	ASSERT(ALIGN_DOWN(va + n - 1, PAGE_SIZE) == page);
		1130
		1131	restart:
		1132	mutex_lock(&AS->lock);
		1133	ipl = interrupts_disable();
		1134	area = find_area_and_lock(AS, page);
3036	svoboda	1135	if (area->backend != &anon_backend \|\| area->sh_info != NULL) {
		1136	mutex_unlock(&area->lock);
		1137	mutex_unlock(&AS->lock);
		1138	interrupts_restore(ipl);
3013	svoboda	1139
3036	svoboda	1140	rc = as_area_make_writeable(area->base);
		1141	if (rc != 0) return rc;
		1142
		1143	goto restart;
		1144	}
		1145
3013	svoboda	1146	pte = page_mapping_find(AS, page);
		1147	if (! (pte && PTE_VALID(pte) && PTE_PRESENT(pte)) ) {
		1148	mutex_unlock(&area->lock);
		1149	mutex_unlock(&AS->lock);
		1150	interrupts_restore(ipl);
		1151
		1152	rc = as_page_fault(page, PF_ACCESS_WRITE, NULL);
		1153	if (rc == AS_PF_FAULT) return EINVAL;
		1154
		1155	goto restart;
		1156	}
		1157
		1158	frame = PTE_GET_FRAME(pte);
		1159	memcpy((void *)(PA2KA(frame) + (va - page)), data, n);
		1160
		1161	mutex_unlock(&area->lock);
		1162	mutex_unlock(&AS->lock);
		1163	interrupts_restore(ipl);
		1164
		1165	return EOK;
		1166	}
		1167
		1168	/** Write data bypassing area flags.
		1169	*
		1170	* See debug_write_inside_page().
		1171	*/
		1172	int as_debug_write(uintptr_t va, void *data, size_t n)
		1173	{
		1174	size_t now;
		1175	int rc;
		1176
		1177	while (n > 0) {
3127	svoboda	1178	/* Number of bytes until the end of page */
		1179	now = ALIGN_DOWN(va, PAGE_SIZE) + PAGE_SIZE - va;
3013	svoboda	1180	if (now > n) now = n;
		1181
		1182	rc = debug_write_inside_page(va, data, now);
		1183	if (rc != EOK) return rc;
		1184
		1185	va += now;
		1186	data += now;
		1187	n -= now;
		1188	}
		1189
		1190	return EOK;
		1191	}
		1192
3036	svoboda	1193	/** Make sure area is private and anonymous.
		1194	*
		1195	* Not atomic atm.
		1196	* @param address Virtual address in AS.
		1197	*/
		1198	int as_area_make_writeable(uintptr_t address)
		1199	{
		1200	ipl_t ipl;
		1201	as_area_t *area;
		1202	uintptr_t base, page;
		1203	uintptr_t old_frame, frame;
		1204	size_t size;
		1205	int flags;
		1206	int page_flags;
		1207	pte_t *pte;
		1208	int rc;
		1209	uintptr_t *pagemap;
		1210
		1211	ipl = interrupts_disable();
		1212	mutex_lock(&AS->lock);
		1213	area = find_area_and_lock(AS, address);
		1214	if (!area) {
		1215	/*
		1216	* Could not find the address space area.
		1217	*/
		1218	mutex_unlock(&AS->lock);
		1219	interrupts_restore(ipl);
		1220	return ENOENT;
		1221	}
		1222
		1223	if (area->backend == &anon_backend && !area->sh_info) {
		1224	/* Nothing to do */
		1225	mutex_unlock(&area->lock);
		1226	mutex_unlock(&AS->lock);
		1227	interrupts_restore(ipl);
		1228	return EOK;
		1229	}
		1230
		1231	base = area->base;
		1232	size = area->pages * PAGE_SIZE;
		1233	flags = area->flags;
		1234	page_flags = as_area_get_flags(area);
		1235
		1236	pagemap = malloc(area->pages * sizeof(uintptr_t), 0);
		1237	page_table_lock(AS, false);
		1238
		1239	for (page = base; page < base + size; page += PAGE_SIZE) {
		1240	pte = page_mapping_find(AS, page);
		1241	if (!pte \|\| !PTE_PRESENT(pte) \|\| !PTE_READABLE(pte)) {
		1242	/* Fetch the missing page */
		1243	if (!area->backend \|\| !area->backend->page_fault) {
		1244	page_table_unlock(AS, false);
		1245	mutex_unlock(&area->lock);
		1246	mutex_unlock(&AS->lock);
		1247	interrupts_restore(ipl);
		1248	return EINVAL;
		1249	}
		1250	if (area->backend->page_fault(area, page, PF_ACCESS_READ) != AS_PF_OK) {
		1251	page_table_unlock(AS, false);
		1252	mutex_unlock(&area->lock);
		1253	mutex_unlock(&AS->lock);
		1254	interrupts_restore(ipl);
		1255	return EINVAL;
		1256	}
		1257	}
		1258	ASSERT(PTE_VALID(pte));
		1259
		1260	old_frame = PTE_GET_FRAME(pte);
		1261
		1262	frame = (uintptr_t)frame_alloc(ONE_FRAME, 0);
		1263	memcpy((void ) PA2KA(frame), (void )PA2KA(old_frame),
		1264	FRAME_SIZE);
		1265
		1266	pagemap[(page - base) / PAGE_SIZE] = frame;
		1267	}
		1268
		1269	page_table_unlock(AS, false);
		1270	mutex_unlock(&area->lock);
		1271	mutex_unlock(&AS->lock);
		1272	interrupts_restore(ipl);
		1273
		1274	rc = as_area_destroy(AS, address);
		1275	if (rc < 0) {
		1276	free(pagemap);
		1277	return rc;
		1278	}
		1279
		1280	area = as_area_create(AS, flags, size, base, AS_AREA_ATTR_PARTIAL,
		1281	&anon_backend, NULL);
		1282	if (area == NULL) {
		1283	free(pagemap);
		1284	return rc;
		1285	}
		1286
		1287	mutex_lock(&AS->lock);
		1288	mutex_lock(&area->lock);
		1289	page_table_lock(AS, false);
		1290	for (page = base; page < base + size; page += PAGE_SIZE) {
		1291	frame = pagemap[(page - base) / PAGE_SIZE];
		1292
		1293	page_mapping_insert(AS, page, frame, page_flags);
		1294	if (!used_space_insert(area, page, 1))
		1295	panic("Could not insert used space.\n");
		1296	}
		1297
		1298	page_table_unlock(AS, false);
		1299
		1300	area->attributes &= ~AS_AREA_ATTR_PARTIAL;
		1301
		1302	mutex_unlock(&area->lock);
		1303	mutex_unlock(&AS->lock);
		1304
		1305	free(pagemap);
		1306
		1307	return EOK;
		1308	}
		1309
1235	jermar	1310	/** Convert address space area flags to page flags.
754	jermar	1311	*
3425	svoboda	1312	* @param aflags Flags of some address space area.
754	jermar	1313	*
3425	svoboda	1314	* @return Flags to be passed to page_mapping_insert().
754	jermar	1315	*/
1235	jermar	1316	int area_flags_to_page_flags(int aflags)
754	jermar	1317	{
		1318	int flags;
		1319
1178	jermar	1320	flags = PAGE_USER \| PAGE_PRESENT;
754	jermar	1321
1235	jermar	1322	if (aflags & AS_AREA_READ)
1026	jermar	1323	flags \|= PAGE_READ;
		1324
1235	jermar	1325	if (aflags & AS_AREA_WRITE)
1026	jermar	1326	flags \|= PAGE_WRITE;
		1327
1235	jermar	1328	if (aflags & AS_AREA_EXEC)
1026	jermar	1329	flags \|= PAGE_EXEC;
		1330
1424	jermar	1331	if (aflags & AS_AREA_CACHEABLE)
1178	jermar	1332	flags \|= PAGE_CACHEABLE;
		1333
754	jermar	1334	return flags;
		1335	}
756	jermar	1336
1235	jermar	1337	/** Compute flags for virtual address translation subsytem.
		1338	*
		1339	* The address space area must be locked.
		1340	* Interrupts must be disabled.
		1341	*
3425	svoboda	1342	* @param a Address space area.
1235	jermar	1343	*
3425	svoboda	1344	* @return Flags to be used in page_mapping_insert().
1235	jermar	1345	*/
1409	jermar	1346	int as_area_get_flags(as_area_t *a)
1235	jermar	1347	{
		1348	return area_flags_to_page_flags(a->flags);
		1349	}
		1350
756	jermar	1351	/** Create page table.
		1352	*
3425	svoboda	1353	* Depending on architecture, create either address space private or global page
		1354	* table.
756	jermar	1355	*
3425	svoboda	1356	* @param flags Flags saying whether the page table is for the kernel
		1357	* address space.
756	jermar	1358	*
3425	svoboda	1359	* @return First entry of the page table.
756	jermar	1360	*/
		1361	pte_t *page_table_create(int flags)
		1362	{
2125	decky	1363	ASSERT(as_operations);
		1364	ASSERT(as_operations->page_table_create);
		1365
		1366	return as_operations->page_table_create(flags);
756	jermar	1367	}
977	jermar	1368
1468	jermar	1369	/** Destroy page table.
		1370	*
		1371	* Destroy page table in architecture specific way.
		1372	*
3425	svoboda	1373	* @param page_table Physical address of PTL0.
1468	jermar	1374	*/
		1375	void page_table_destroy(pte_t *page_table)
		1376	{
2125	decky	1377	ASSERT(as_operations);
		1378	ASSERT(as_operations->page_table_destroy);
		1379
		1380	as_operations->page_table_destroy(page_table);
1468	jermar	1381	}
		1382
1044	jermar	1383	/** Lock page table.
		1384	*
		1385	* This function should be called before any page_mapping_insert(),
		1386	* page_mapping_remove() and page_mapping_find().
		1387	*
		1388	* Locking order is such that address space areas must be locked
		1389	* prior to this call. Address space can be locked prior to this
		1390	* call in which case the lock argument is false.
		1391	*
3425	svoboda	1392	* @param as Address space.
		1393	* @param lock If false, do not attempt to lock as->lock.
1044	jermar	1394	*/
		1395	void page_table_lock(as_t *as, bool lock)
		1396	{
		1397	ASSERT(as_operations);
		1398	ASSERT(as_operations->page_table_lock);
2125	decky	1399
1044	jermar	1400	as_operations->page_table_lock(as, lock);
		1401	}
		1402
		1403	/** Unlock page table.
		1404	*
3425	svoboda	1405	* @param as Address space.
		1406	* @param unlock If false, do not attempt to unlock as->lock.
1044	jermar	1407	*/
		1408	void page_table_unlock(as_t *as, bool unlock)
		1409	{
		1410	ASSERT(as_operations);
		1411	ASSERT(as_operations->page_table_unlock);
2125	decky	1412
1044	jermar	1413	as_operations->page_table_unlock(as, unlock);
		1414	}
		1415
977	jermar	1416
		1417	/** Find address space area and lock it.
		1418	*
		1419	* The address space must be locked and interrupts must be disabled.
		1420	*
3425	svoboda	1421	* @param as Address space.
		1422	* @param va Virtual address.
977	jermar	1423	*
3425	svoboda	1424	* @return Locked address space area containing va on success or
		1425	* NULL on failure.
977	jermar	1426	*/
1780	jermar	1427	as_area_t find_area_and_lock(as_t as, uintptr_t va)
977	jermar	1428	{
		1429	as_area_t *a;
1147	jermar	1430	btree_node_t leaf, lnode;
2745	decky	1431	unsigned int i;
977	jermar	1432
1147	jermar	1433	a = (as_area_t *) btree_search(&as->as_area_btree, va, &leaf);
		1434	if (a) {
		1435	/* va is the base address of an address space area */
1380	jermar	1436	mutex_lock(&a->lock);
1147	jermar	1437	return a;
		1438	}
		1439
		1440	/*
1150	jermar	1441	* Search the leaf node and the righmost record of its left neighbour
1147	jermar	1442	* to find out whether this is a miss or va belongs to an address
		1443	* space area found there.
		1444	*/
		1445
		1446	/* First, search the leaf node itself. */
		1447	for (i = 0; i < leaf->keys; i++) {
		1448	a = (as_area_t *) leaf->value[i];
1380	jermar	1449	mutex_lock(&a->lock);
1147	jermar	1450	if ((a->base <= va) && (va < a->base + a->pages * PAGE_SIZE)) {
		1451	return a;
		1452	}
1380	jermar	1453	mutex_unlock(&a->lock);
1147	jermar	1454	}
977	jermar	1455
1147	jermar	1456	/*
1150	jermar	1457	* Second, locate the left neighbour and test its last record.
1148	jermar	1458	* Because of its position in the B+tree, it must have base < va.
1147	jermar	1459	*/
2087	jermar	1460	lnode = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf);
		1461	if (lnode) {
1147	jermar	1462	a = (as_area_t *) lnode->value[lnode->keys - 1];
1380	jermar	1463	mutex_lock(&a->lock);
1147	jermar	1464	if (va < a->base + a->pages * PAGE_SIZE) {
1048	jermar	1465	return a;
1147	jermar	1466	}
1380	jermar	1467	mutex_unlock(&a->lock);
977	jermar	1468	}
		1469
		1470	return NULL;
		1471	}
1048	jermar	1472
		1473	/** Check area conflicts with other areas.
		1474	*
		1475	* The address space must be locked and interrupts must be disabled.
		1476	*
3425	svoboda	1477	* @param as Address space.
		1478	* @param va Starting virtual address of the area being tested.
		1479	* @param size Size of the area being tested.
		1480	* @param avoid_area Do not touch this area.
1048	jermar	1481	*
3425	svoboda	1482	* @return True if there is no conflict, false otherwise.
1048	jermar	1483	*/
3425	svoboda	1484	bool
		1485	check_area_conflicts(as_t as, uintptr_t va, size_t size, as_area_t avoid_area)
1048	jermar	1486	{
		1487	as_area_t *a;
1147	jermar	1488	btree_node_t leaf, node;
2745	decky	1489	unsigned int i;
1048	jermar	1490
1070	jermar	1491	/*
		1492	* We don't want any area to have conflicts with NULL page.
		1493	*/
		1494	if (overlaps(va, size, NULL, PAGE_SIZE))
		1495	return false;
		1496
1147	jermar	1497	/*
		1498	* The leaf node is found in O(log n), where n is proportional to
		1499	* the number of address space areas belonging to as.
		1500	* The check for conflicts is then attempted on the rightmost
1150	jermar	1501	* record in the left neighbour, the leftmost record in the right
		1502	* neighbour and all records in the leaf node itself.
1147	jermar	1503	*/
1048	jermar	1504
1147	jermar	1505	if ((a = (as_area_t *) btree_search(&as->as_area_btree, va, &leaf))) {
		1506	if (a != avoid_area)
		1507	return false;
		1508	}
		1509
		1510	/* First, check the two border cases. */
1150	jermar	1511	if ((node = btree_leaf_node_left_neighbour(&as->as_area_btree, leaf))) {
1147	jermar	1512	a = (as_area_t *) node->value[node->keys - 1];
1380	jermar	1513	mutex_lock(&a->lock);
1147	jermar	1514	if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
1380	jermar	1515	mutex_unlock(&a->lock);
1147	jermar	1516	return false;
		1517	}
1380	jermar	1518	mutex_unlock(&a->lock);
1147	jermar	1519	}
2087	jermar	1520	node = btree_leaf_node_right_neighbour(&as->as_area_btree, leaf);
		1521	if (node) {
1147	jermar	1522	a = (as_area_t *) node->value[0];
1380	jermar	1523	mutex_lock(&a->lock);
1147	jermar	1524	if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
1380	jermar	1525	mutex_unlock(&a->lock);
1147	jermar	1526	return false;
		1527	}
1380	jermar	1528	mutex_unlock(&a->lock);
1147	jermar	1529	}
		1530
		1531	/* Second, check the leaf node. */
		1532	for (i = 0; i < leaf->keys; i++) {
		1533	a = (as_area_t *) leaf->value[i];
		1534
1048	jermar	1535	if (a == avoid_area)
		1536	continue;
1147	jermar	1537
1380	jermar	1538	mutex_lock(&a->lock);
1147	jermar	1539	if (overlaps(va, size, a->base, a->pages * PAGE_SIZE)) {
1380	jermar	1540	mutex_unlock(&a->lock);
1147	jermar	1541	return false;
		1542	}
1380	jermar	1543	mutex_unlock(&a->lock);
1048	jermar	1544	}
		1545
1070	jermar	1546	/*
		1547	* So far, the area does not conflict with other areas.
		1548	* Check if it doesn't conflict with kernel address space.
		1549	*/
		1550	if (!KERNEL_ADDRESS_SPACE_SHADOWED) {
		1551	return !overlaps(va, size,
2087	jermar	1552	KERNEL_ADDRESS_SPACE_START,
		1553	KERNEL_ADDRESS_SPACE_END - KERNEL_ADDRESS_SPACE_START);
1070	jermar	1554	}
		1555
1048	jermar	1556	return true;
		1557	}
1235	jermar	1558
2556	jermar	1559	/** Return size of the address space area with given base.
		1560	*
		1561	* @param base Arbitrary address insede the address space area.
		1562	*
		1563	* @return Size of the address space area in bytes or zero if it
		1564	* does not exist.
		1565	*/
		1566	size_t as_area_get_size(uintptr_t base)
1329	palkovsky	1567	{
		1568	ipl_t ipl;
		1569	as_area_t *src_area;
		1570	size_t size;
		1571
		1572	ipl = interrupts_disable();
		1573	src_area = find_area_and_lock(AS, base);
3425	svoboda	1574	if (src_area) {
1329	palkovsky	1575	size = src_area->pages * PAGE_SIZE;
1380	jermar	1576	mutex_unlock(&src_area->lock);
1329	palkovsky	1577	} else {
		1578	size = 0;
		1579	}
		1580	interrupts_restore(ipl);
		1581	return size;
		1582	}
		1583
1387	jermar	1584	/** Mark portion of address space area as used.
		1585	*
		1586	* The address space area must be already locked.
		1587	*
3425	svoboda	1588	* @param a Address space area.
		1589	* @param page First page to be marked.
		1590	* @param count Number of page to be marked.
1387	jermar	1591	*
3425	svoboda	1592	* @return Zero on failure and non-zero on success.
1387	jermar	1593	*/
1780	jermar	1594	int used_space_insert(as_area_t *a, uintptr_t page, count_t count)
1387	jermar	1595	{
		1596	btree_node_t leaf, node;
		1597	count_t pages;
2745	decky	1598	unsigned int i;
1387	jermar	1599
		1600	ASSERT(page == ALIGN_DOWN(page, PAGE_SIZE));
		1601	ASSERT(count);
		1602
		1603	pages = (count_t) btree_search(&a->used_space, page, &leaf);
		1604	if (pages) {
		1605	/*
		1606	* We hit the beginning of some used space.
		1607	*/
		1608	return 0;
		1609	}
		1610
1437	jermar	1611	if (!leaf->keys) {
		1612	btree_insert(&a->used_space, page, (void *) count, leaf);
		1613	return 1;
		1614	}
		1615
1387	jermar	1616	node = btree_leaf_node_left_neighbour(&a->used_space, leaf);
		1617	if (node) {
2087	jermar	1618	uintptr_t left_pg = node->key[node->keys - 1];
		1619	uintptr_t right_pg = leaf->key[0];
		1620	count_t left_cnt = (count_t) node->value[node->keys - 1];
		1621	count_t right_cnt = (count_t) leaf->value[0];
1387	jermar	1622
		1623	/*
		1624	* Examine the possibility that the interval fits
		1625	* somewhere between the rightmost interval of
		1626	* the left neigbour and the first interval of the leaf.
		1627	*/
		1628
		1629	if (page >= right_pg) {
		1630	/* Do nothing. */
2087	jermar	1631	} else if (overlaps(page, count * PAGE_SIZE, left_pg,
		1632	left_cnt * PAGE_SIZE)) {
1387	jermar	1633	/* The interval intersects with the left interval. */
		1634	return 0;
2087	jermar	1635	} else if (overlaps(page, count * PAGE_SIZE, right_pg,
		1636	right_cnt * PAGE_SIZE)) {
1387	jermar	1637	/* The interval intersects with the right interval. */
		1638	return 0;
2087	jermar	1639	} else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
		1640	(page + count * PAGE_SIZE == right_pg)) {
		1641	/*
		1642	* The interval can be added by merging the two already
		1643	* present intervals.
		1644	*/
1403	jermar	1645	node->value[node->keys - 1] += count + right_cnt;
1387	jermar	1646	btree_remove(&a->used_space, right_pg, leaf);
		1647	return 1;
2087	jermar	1648	} else if (page == left_pg + left_cnt * PAGE_SIZE) {
		1649	/*
		1650	* The interval can be added by simply growing the left
		1651	* interval.
		1652	*/
1403	jermar	1653	node->value[node->keys - 1] += count;
1387	jermar	1654	return 1;
2087	jermar	1655	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1656	/*
2087	jermar	1657	* The interval can be addded by simply moving base of
		1658	* the right interval down and increasing its size
		1659	* accordingly.
1387	jermar	1660	*/
1403	jermar	1661	leaf->value[0] += count;
1387	jermar	1662	leaf->key[0] = page;
		1663	return 1;
		1664	} else {
		1665	/*
		1666	* The interval is between both neigbouring intervals,
		1667	* but cannot be merged with any of them.
		1668	*/
2087	jermar	1669	btree_insert(&a->used_space, page, (void *) count,
		1670	leaf);
1387	jermar	1671	return 1;
		1672	}
		1673	} else if (page < leaf->key[0]) {
1780	jermar	1674	uintptr_t right_pg = leaf->key[0];
1387	jermar	1675	count_t right_cnt = (count_t) leaf->value[0];
		1676
		1677	/*
2087	jermar	1678	* Investigate the border case in which the left neighbour does
		1679	* not exist but the interval fits from the left.
1387	jermar	1680	*/
		1681
2087	jermar	1682	if (overlaps(page, count * PAGE_SIZE, right_pg,
		1683	right_cnt * PAGE_SIZE)) {
1387	jermar	1684	/* The interval intersects with the right interval. */
		1685	return 0;
2087	jermar	1686	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1687	/*
2087	jermar	1688	* The interval can be added by moving the base of the
		1689	* right interval down and increasing its size
		1690	* accordingly.
1387	jermar	1691	*/
		1692	leaf->key[0] = page;
1403	jermar	1693	leaf->value[0] += count;
1387	jermar	1694	return 1;
		1695	} else {
		1696	/*
		1697	* The interval doesn't adjoin with the right interval.
		1698	* It must be added individually.
		1699	*/
2087	jermar	1700	btree_insert(&a->used_space, page, (void *) count,
		1701	leaf);
1387	jermar	1702	return 1;
		1703	}
		1704	}
		1705
		1706	node = btree_leaf_node_right_neighbour(&a->used_space, leaf);
		1707	if (node) {
2087	jermar	1708	uintptr_t left_pg = leaf->key[leaf->keys - 1];
		1709	uintptr_t right_pg = node->key[0];
		1710	count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
		1711	count_t right_cnt = (count_t) node->value[0];
1387	jermar	1712
		1713	/*
		1714	* Examine the possibility that the interval fits
		1715	* somewhere between the leftmost interval of
		1716	* the right neigbour and the last interval of the leaf.
		1717	*/
		1718
		1719	if (page < left_pg) {
		1720	/* Do nothing. */
2087	jermar	1721	} else if (overlaps(page, count * PAGE_SIZE, left_pg,
		1722	left_cnt * PAGE_SIZE)) {
1387	jermar	1723	/* The interval intersects with the left interval. */
		1724	return 0;
2087	jermar	1725	} else if (overlaps(page, count * PAGE_SIZE, right_pg,
		1726	right_cnt * PAGE_SIZE)) {
1387	jermar	1727	/* The interval intersects with the right interval. */
		1728	return 0;
2087	jermar	1729	} else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
		1730	(page + count * PAGE_SIZE == right_pg)) {
		1731	/*
		1732	* The interval can be added by merging the two already
		1733	* present intervals.
		1734	* */
1403	jermar	1735	leaf->value[leaf->keys - 1] += count + right_cnt;
1387	jermar	1736	btree_remove(&a->used_space, right_pg, node);
		1737	return 1;
2087	jermar	1738	} else if (page == left_pg + left_cnt * PAGE_SIZE) {
		1739	/*
		1740	* The interval can be added by simply growing the left
		1741	* interval.
		1742	* */
1403	jermar	1743	leaf->value[leaf->keys - 1] += count;
1387	jermar	1744	return 1;
2087	jermar	1745	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1746	/*
2087	jermar	1747	* The interval can be addded by simply moving base of
		1748	* the right interval down and increasing its size
		1749	* accordingly.
1387	jermar	1750	*/
1403	jermar	1751	node->value[0] += count;
1387	jermar	1752	node->key[0] = page;
		1753	return 1;
		1754	} else {
		1755	/*
		1756	* The interval is between both neigbouring intervals,
		1757	* but cannot be merged with any of them.
		1758	*/
2087	jermar	1759	btree_insert(&a->used_space, page, (void *) count,
		1760	leaf);
1387	jermar	1761	return 1;
		1762	}
		1763	} else if (page >= leaf->key[leaf->keys - 1]) {
1780	jermar	1764	uintptr_t left_pg = leaf->key[leaf->keys - 1];
1387	jermar	1765	count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
		1766
		1767	/*
2087	jermar	1768	* Investigate the border case in which the right neighbour
		1769	* does not exist but the interval fits from the right.
1387	jermar	1770	*/
		1771
2087	jermar	1772	if (overlaps(page, count * PAGE_SIZE, left_pg,
		1773	left_cnt * PAGE_SIZE)) {
1403	jermar	1774	/* The interval intersects with the left interval. */
1387	jermar	1775	return 0;
2087	jermar	1776	} else if (left_pg + left_cnt * PAGE_SIZE == page) {
		1777	/*
		1778	* The interval can be added by growing the left
		1779	* interval.
		1780	*/
1403	jermar	1781	leaf->value[leaf->keys - 1] += count;
1387	jermar	1782	return 1;
		1783	} else {
		1784	/*
		1785	* The interval doesn't adjoin with the left interval.
		1786	* It must be added individually.
		1787	*/
2087	jermar	1788	btree_insert(&a->used_space, page, (void *) count,
		1789	leaf);
1387	jermar	1790	return 1;
		1791	}
		1792	}
		1793
		1794	/*
2087	jermar	1795	* Note that if the algorithm made it thus far, the interval can fit
		1796	* only between two other intervals of the leaf. The two border cases
		1797	* were already resolved.
1387	jermar	1798	*/
		1799	for (i = 1; i < leaf->keys; i++) {
		1800	if (page < leaf->key[i]) {
2087	jermar	1801	uintptr_t left_pg = leaf->key[i - 1];
		1802	uintptr_t right_pg = leaf->key[i];
		1803	count_t left_cnt = (count_t) leaf->value[i - 1];
		1804	count_t right_cnt = (count_t) leaf->value[i];
1387	jermar	1805
		1806	/*
		1807	* The interval fits between left_pg and right_pg.
		1808	*/
		1809
2087	jermar	1810	if (overlaps(page, count * PAGE_SIZE, left_pg,
		1811	left_cnt * PAGE_SIZE)) {
		1812	/*
		1813	* The interval intersects with the left
		1814	* interval.
		1815	*/
1387	jermar	1816	return 0;
2087	jermar	1817	} else if (overlaps(page, count * PAGE_SIZE, right_pg,
		1818	right_cnt * PAGE_SIZE)) {
		1819	/*
		1820	* The interval intersects with the right
		1821	* interval.
		1822	*/
1387	jermar	1823	return 0;
2087	jermar	1824	} else if ((page == left_pg + left_cnt * PAGE_SIZE) &&
		1825	(page + count * PAGE_SIZE == right_pg)) {
		1826	/*
		1827	* The interval can be added by merging the two
		1828	* already present intervals.
		1829	*/
1403	jermar	1830	leaf->value[i - 1] += count + right_cnt;
1387	jermar	1831	btree_remove(&a->used_space, right_pg, leaf);
		1832	return 1;
2087	jermar	1833	} else if (page == left_pg + left_cnt * PAGE_SIZE) {
		1834	/*
		1835	* The interval can be added by simply growing
		1836	* the left interval.
		1837	*/
1403	jermar	1838	leaf->value[i - 1] += count;
1387	jermar	1839	return 1;
2087	jermar	1840	} else if (page + count * PAGE_SIZE == right_pg) {
1387	jermar	1841	/*
2087	jermar	1842	* The interval can be addded by simply moving
		1843	* base of the right interval down and
		1844	* increasing its size accordingly.
1387	jermar	1845	*/
1403	jermar	1846	leaf->value[i] += count;
1387	jermar	1847	leaf->key[i] = page;
		1848	return 1;
		1849	} else {
		1850	/*
2087	jermar	1851	* The interval is between both neigbouring
		1852	* intervals, but cannot be merged with any of
		1853	* them.
1387	jermar	1854	*/
2087	jermar	1855	btree_insert(&a->used_space, page,
		1856	(void *) count, leaf);
1387	jermar	1857	return 1;
		1858	}
		1859	}
		1860	}
		1861
3425	svoboda	1862	panic("Inconsistency detected while adding %" PRIc " pages of used "
		1863	"space at %p.\n", count, page);
1387	jermar	1864	}
		1865
		1866	/** Mark portion of address space area as unused.
		1867	*
		1868	* The address space area must be already locked.
		1869	*
3425	svoboda	1870	* @param a Address space area.
		1871	* @param page First page to be marked.
		1872	* @param count Number of page to be marked.
1387	jermar	1873	*
3425	svoboda	1874	* @return Zero on failure and non-zero on success.
1387	jermar	1875	*/
1780	jermar	1876	int used_space_remove(as_area_t *a, uintptr_t page, count_t count)
1387	jermar	1877	{
		1878	btree_node_t leaf, node;
		1879	count_t pages;
2745	decky	1880	unsigned int i;
1387	jermar	1881
		1882	ASSERT(page == ALIGN_DOWN(page, PAGE_SIZE));
		1883	ASSERT(count);
		1884
		1885	pages = (count_t) btree_search(&a->used_space, page, &leaf);
		1886	if (pages) {
		1887	/*
		1888	* We are lucky, page is the beginning of some interval.
		1889	*/
		1890	if (count > pages) {
		1891	return 0;
		1892	} else if (count == pages) {
		1893	btree_remove(&a->used_space, page, leaf);
1403	jermar	1894	return 1;
1387	jermar	1895	} else {
		1896	/*
		1897	* Find the respective interval.
		1898	* Decrease its size and relocate its start address.
		1899	*/
		1900	for (i = 0; i < leaf->keys; i++) {
		1901	if (leaf->key[i] == page) {
2087	jermar	1902	leaf->key[i] += count * PAGE_SIZE;
1403	jermar	1903	leaf->value[i] -= count;
1387	jermar	1904	return 1;
		1905	}
		1906	}
		1907	goto error;
		1908	}
		1909	}
		1910
		1911	node = btree_leaf_node_left_neighbour(&a->used_space, leaf);
		1912	if (node && page < leaf->key[0]) {
1780	jermar	1913	uintptr_t left_pg = node->key[node->keys - 1];
1387	jermar	1914	count_t left_cnt = (count_t) node->value[node->keys - 1];
		1915
2087	jermar	1916	if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
		1917	count * PAGE_SIZE)) {
		1918	if (page + count * PAGE_SIZE ==
		1919	left_pg + left_cnt * PAGE_SIZE) {
1387	jermar	1920	/*
2087	jermar	1921	* The interval is contained in the rightmost
		1922	* interval of the left neighbour and can be
		1923	* removed by updating the size of the bigger
		1924	* interval.
1387	jermar	1925	*/
1403	jermar	1926	node->value[node->keys - 1] -= count;
1387	jermar	1927	return 1;
2087	jermar	1928	} else if (page + count * PAGE_SIZE <
		1929	left_pg + left_cnt*PAGE_SIZE) {
1403	jermar	1930	count_t new_cnt;
1387	jermar	1931
		1932	/*
2087	jermar	1933	* The interval is contained in the rightmost
		1934	* interval of the left neighbour but its
		1935	* removal requires both updating the size of
		1936	* the original interval and also inserting a
		1937	* new interval.
1387	jermar	1938	*/
2087	jermar	1939	new_cnt = ((left_pg + left_cnt * PAGE_SIZE) -
		1940	(page + count*PAGE_SIZE)) >> PAGE_WIDTH;
1403	jermar	1941	node->value[node->keys - 1] -= count + new_cnt;
2087	jermar	1942	btree_insert(&a->used_space, page +
		1943	count * PAGE_SIZE, (void *) new_cnt, leaf);
1387	jermar	1944	return 1;
		1945	}
		1946	}
		1947	return 0;
		1948	} else if (page < leaf->key[0]) {
		1949	return 0;
		1950	}
		1951
		1952	if (page > leaf->key[leaf->keys - 1]) {
1780	jermar	1953	uintptr_t left_pg = leaf->key[leaf->keys - 1];
1387	jermar	1954	count_t left_cnt = (count_t) leaf->value[leaf->keys - 1];
		1955
2087	jermar	1956	if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
		1957	count * PAGE_SIZE)) {
		1958	if (page + count * PAGE_SIZE ==
		1959	left_pg + left_cnt * PAGE_SIZE) {
1387	jermar	1960	/*
2087	jermar	1961	* The interval is contained in the rightmost
		1962	* interval of the leaf and can be removed by
		1963	* updating the size of the bigger interval.
1387	jermar	1964	*/
1403	jermar	1965	leaf->value[leaf->keys - 1] -= count;
1387	jermar	1966	return 1;
2087	jermar	1967	} else if (page + count * PAGE_SIZE < left_pg +
		1968	left_cnt * PAGE_SIZE) {
1403	jermar	1969	count_t new_cnt;
1387	jermar	1970
		1971	/*
2087	jermar	1972	* The interval is contained in the rightmost
		1973	* interval of the leaf but its removal
		1974	* requires both updating the size of the
		1975	* original interval and also inserting a new
		1976	* interval.
1387	jermar	1977	*/
2087	jermar	1978	new_cnt = ((left_pg + left_cnt * PAGE_SIZE) -
		1979	(page + count * PAGE_SIZE)) >> PAGE_WIDTH;
1403	jermar	1980	leaf->value[leaf->keys - 1] -= count + new_cnt;
2087	jermar	1981	btree_insert(&a->used_space, page +
		1982	count * PAGE_SIZE, (void *) new_cnt, leaf);
1387	jermar	1983	return 1;
		1984	}
		1985	}
		1986	return 0;
		1987	}
		1988
		1989	/*
		1990	* The border cases have been already resolved.
		1991	* Now the interval can be only between intervals of the leaf.
		1992	*/
		1993	for (i = 1; i < leaf->keys - 1; i++) {
		1994	if (page < leaf->key[i]) {
1780	jermar	1995	uintptr_t left_pg = leaf->key[i - 1];
1387	jermar	1996	count_t left_cnt = (count_t) leaf->value[i - 1];
		1997
		1998	/*
2087	jermar	1999	* Now the interval is between intervals corresponding
		2000	* to (i - 1) and i.
1387	jermar	2001	*/
2087	jermar	2002	if (overlaps(left_pg, left_cnt * PAGE_SIZE, page,
		2003	count * PAGE_SIZE)) {
		2004	if (page + count * PAGE_SIZE ==
		2005	left_pg + left_cnt*PAGE_SIZE) {
1387	jermar	2006	/*
2087	jermar	2007	* The interval is contained in the
		2008	* interval (i - 1) of the leaf and can
		2009	* be removed by updating the size of
		2010	* the bigger interval.
1387	jermar	2011	*/
1403	jermar	2012	leaf->value[i - 1] -= count;
1387	jermar	2013	return 1;
2087	jermar	2014	} else if (page + count * PAGE_SIZE <
		2015	left_pg + left_cnt * PAGE_SIZE) {
1403	jermar	2016	count_t new_cnt;
1387	jermar	2017
		2018	/*
2087	jermar	2019	* The interval is contained in the
		2020	* interval (i - 1) of the leaf but its
		2021	* removal requires both updating the
		2022	* size of the original interval and
1387	jermar	2023	* also inserting a new interval.
		2024	*/
2087	jermar	2025	new_cnt = ((left_pg +
		2026	left_cnt * PAGE_SIZE) -
		2027	(page + count * PAGE_SIZE)) >>
		2028	PAGE_WIDTH;
1403	jermar	2029	leaf->value[i - 1] -= count + new_cnt;
2087	jermar	2030	btree_insert(&a->used_space, page +
		2031	count * PAGE_SIZE, (void *) new_cnt,
		2032	leaf);
1387	jermar	2033	return 1;
		2034	}
		2035	}
		2036	return 0;
		2037	}
		2038	}
		2039
		2040	error:
3425	svoboda	2041	panic("Inconsistency detected while removing %" PRIc " pages of used "
		2042	"space from %p.\n", count, page);
1387	jermar	2043	}
		2044
1409	jermar	2045	/** Remove reference to address space area share info.
		2046	*
		2047	* If the reference count drops to 0, the sh_info is deallocated.
		2048	*
3425	svoboda	2049	* @param sh_info Pointer to address space area share info.
1409	jermar	2050	*/
		2051	void sh_info_remove_reference(share_info_t *sh_info)
		2052	{
		2053	bool dealloc = false;
		2054
		2055	mutex_lock(&sh_info->lock);
		2056	ASSERT(sh_info->refcount);
		2057	if (--sh_info->refcount == 0) {
		2058	dealloc = true;
1495	jermar	2059	link_t *cur;
1409	jermar	2060
		2061	/*
		2062	* Now walk carefully the pagemap B+tree and free/remove
		2063	* reference from all frames found there.
		2064	*/
2087	jermar	2065	for (cur = sh_info->pagemap.leaf_head.next;
		2066	cur != &sh_info->pagemap.leaf_head; cur = cur->next) {
1409	jermar	2067	btree_node_t *node;
2745	decky	2068	unsigned int i;
1409	jermar	2069
1495	jermar	2070	node = list_get_instance(cur, btree_node_t, leaf_link);
		2071	for (i = 0; i < node->keys; i++)
1780	jermar	2072	frame_free((uintptr_t) node->value[i]);
1409	jermar	2073	}
		2074
		2075	}
		2076	mutex_unlock(&sh_info->lock);
		2077
		2078	if (dealloc) {
		2079	btree_destroy(&sh_info->pagemap);
		2080	free(sh_info);
		2081	}
		2082	}
		2083
1235	jermar	2084	/*
		2085	* Address space related syscalls.
		2086	*/
		2087
		2088	/** Wrapper for as_area_create(). */
1780	jermar	2089	unative_t sys_as_area_create(uintptr_t address, size_t size, int flags)
1235	jermar	2090	{
2087	jermar	2091	if (as_area_create(AS, flags \| AS_AREA_CACHEABLE, size, address,
		2092	AS_AREA_ATTR_NONE, &anon_backend, NULL))
1780	jermar	2093	return (unative_t) address;
1235	jermar	2094	else
1780	jermar	2095	return (unative_t) -1;
1235	jermar	2096	}
		2097
1793	jermar	2098	/** Wrapper for as_area_resize(). */
1780	jermar	2099	unative_t sys_as_area_resize(uintptr_t address, size_t size, int flags)
1235	jermar	2100	{
1780	jermar	2101	return (unative_t) as_area_resize(AS, address, size, 0);
1235	jermar	2102	}
		2103
3425	svoboda	2104	/** Wrapper for as_area_change_flags(). */
		2105	unative_t sys_as_area_change_flags(uintptr_t address, int flags)
		2106	{
		2107	return (unative_t) as_area_change_flags(AS, flags, address);
		2108	}
		2109
1793	jermar	2110	/** Wrapper for as_area_destroy(). */
1780	jermar	2111	unative_t sys_as_area_destroy(uintptr_t address)
1306	jermar	2112	{
1780	jermar	2113	return (unative_t) as_area_destroy(AS, address);
1306	jermar	2114	}
1702	cejka	2115
1914	jermar	2116	/** Print out information about address space.
		2117	*
3425	svoboda	2118	* @param as Address space.
1914	jermar	2119	*/
		2120	void as_print(as_t *as)
		2121	{
		2122	ipl_t ipl;
		2123
		2124	ipl = interrupts_disable();
		2125	mutex_lock(&as->lock);
		2126
		2127	/* print out info about address space areas */
		2128	link_t *cur;
2087	jermar	2129	for (cur = as->as_area_btree.leaf_head.next;
		2130	cur != &as->as_area_btree.leaf_head; cur = cur->next) {
		2131	btree_node_t *node;
1914	jermar	2132
2087	jermar	2133	node = list_get_instance(cur, btree_node_t, leaf_link);
		2134
2745	decky	2135	unsigned int i;
1914	jermar	2136	for (i = 0; i < node->keys; i++) {
1915	jermar	2137	as_area_t *area = node->value[i];
1914	jermar	2138
		2139	mutex_lock(&area->lock);
3425	svoboda	2140	printf("as_area: %p, base=%p, pages=%" PRIc
		2141	" (%p - %p)\n", area, area->base, area->pages,
		2142	area->base, area->base + FRAMES2SIZE(area->pages));
1914	jermar	2143	mutex_unlock(&area->lock);
		2144	}
		2145	}
		2146
		2147	mutex_unlock(&as->lock);
		2148	interrupts_restore(ipl);
		2149	}
		2150
1757	jermar	2151	/** @}
1702	cejka	2152	*/

Subversion Repositories HelenOS

(root)/branches/tracing/kernel/generic/src/mm/as.c @ 4692 – Rev 3425