Linux Audio

Check our new training course

Embedded Linux training

Mar 10-20, 2025, special US time zones
Register
Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  KVM guest address space mapping code
   4 *
   5 *    Copyright IBM Corp. 2007, 2020
   6 *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
   7 *		 David Hildenbrand <david@redhat.com>
   8 *		 Janosch Frank <frankja@linux.vnet.ibm.com>
   9 */
  10
  11#include <linux/kernel.h>
  12#include <linux/pagewalk.h>
  13#include <linux/swap.h>
  14#include <linux/smp.h>
  15#include <linux/spinlock.h>
  16#include <linux/slab.h>
  17#include <linux/swapops.h>
  18#include <linux/ksm.h>
  19#include <linux/mman.h>
  20#include <linux/pgtable.h>
  21
 
  22#include <asm/pgalloc.h>
  23#include <asm/gmap.h>
  24#include <asm/tlb.h>
  25
  26#define GMAP_SHADOW_FAKE_TABLE 1ULL
  27
  28/**
  29 * gmap_alloc - allocate and initialize a guest address space
 
  30 * @limit: maximum address of the gmap address space
  31 *
  32 * Returns a guest address space structure.
  33 */
  34static struct gmap *gmap_alloc(unsigned long limit)
  35{
  36	struct gmap *gmap;
  37	struct page *page;
  38	unsigned long *table;
  39	unsigned long etype, atype;
  40
  41	if (limit < _REGION3_SIZE) {
  42		limit = _REGION3_SIZE - 1;
  43		atype = _ASCE_TYPE_SEGMENT;
  44		etype = _SEGMENT_ENTRY_EMPTY;
  45	} else if (limit < _REGION2_SIZE) {
  46		limit = _REGION2_SIZE - 1;
  47		atype = _ASCE_TYPE_REGION3;
  48		etype = _REGION3_ENTRY_EMPTY;
  49	} else if (limit < _REGION1_SIZE) {
  50		limit = _REGION1_SIZE - 1;
  51		atype = _ASCE_TYPE_REGION2;
  52		etype = _REGION2_ENTRY_EMPTY;
  53	} else {
  54		limit = -1UL;
  55		atype = _ASCE_TYPE_REGION1;
  56		etype = _REGION1_ENTRY_EMPTY;
  57	}
  58	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL_ACCOUNT);
  59	if (!gmap)
  60		goto out;
  61	INIT_LIST_HEAD(&gmap->crst_list);
  62	INIT_LIST_HEAD(&gmap->children);
  63	INIT_LIST_HEAD(&gmap->pt_list);
  64	INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL_ACCOUNT);
  65	INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC | __GFP_ACCOUNT);
  66	INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC | __GFP_ACCOUNT);
  67	spin_lock_init(&gmap->guest_table_lock);
  68	spin_lock_init(&gmap->shadow_lock);
  69	refcount_set(&gmap->ref_count, 1);
  70	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
  71	if (!page)
  72		goto out_free;
  73	page->index = 0;
  74	list_add(&page->lru, &gmap->crst_list);
  75	table = page_to_virt(page);
  76	crst_table_init(table, etype);
  77	gmap->table = table;
  78	gmap->asce = atype | _ASCE_TABLE_LENGTH |
  79		_ASCE_USER_BITS | __pa(table);
  80	gmap->asce_end = limit;
  81	return gmap;
  82
  83out_free:
  84	kfree(gmap);
  85out:
  86	return NULL;
  87}
  88
  89/**
  90 * gmap_create - create a guest address space
  91 * @mm: pointer to the parent mm_struct
  92 * @limit: maximum size of the gmap address space
  93 *
  94 * Returns a guest address space structure.
  95 */
  96struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
  97{
  98	struct gmap *gmap;
  99	unsigned long gmap_asce;
 100
 101	gmap = gmap_alloc(limit);
 102	if (!gmap)
 103		return NULL;
 104	gmap->mm = mm;
 105	spin_lock(&mm->context.lock);
 106	list_add_rcu(&gmap->list, &mm->context.gmap_list);
 107	if (list_is_singular(&mm->context.gmap_list))
 108		gmap_asce = gmap->asce;
 109	else
 110		gmap_asce = -1UL;
 111	WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
 112	spin_unlock(&mm->context.lock);
 113	return gmap;
 114}
 115EXPORT_SYMBOL_GPL(gmap_create);
 116
 117static void gmap_flush_tlb(struct gmap *gmap)
 118{
 119	if (MACHINE_HAS_IDTE)
 120		__tlb_flush_idte(gmap->asce);
 121	else
 122		__tlb_flush_global();
 123}
 124
 125static void gmap_radix_tree_free(struct radix_tree_root *root)
 126{
 127	struct radix_tree_iter iter;
 128	unsigned long indices[16];
 129	unsigned long index;
 130	void __rcu **slot;
 131	int i, nr;
 132
 133	/* A radix tree is freed by deleting all of its entries */
 134	index = 0;
 135	do {
 136		nr = 0;
 137		radix_tree_for_each_slot(slot, root, &iter, index) {
 138			indices[nr] = iter.index;
 139			if (++nr == 16)
 140				break;
 141		}
 142		for (i = 0; i < nr; i++) {
 143			index = indices[i];
 144			radix_tree_delete(root, index);
 145		}
 146	} while (nr > 0);
 147}
 148
 149static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
 150{
 151	struct gmap_rmap *rmap, *rnext, *head;
 152	struct radix_tree_iter iter;
 153	unsigned long indices[16];
 154	unsigned long index;
 155	void __rcu **slot;
 156	int i, nr;
 157
 158	/* A radix tree is freed by deleting all of its entries */
 159	index = 0;
 160	do {
 161		nr = 0;
 162		radix_tree_for_each_slot(slot, root, &iter, index) {
 163			indices[nr] = iter.index;
 164			if (++nr == 16)
 165				break;
 166		}
 167		for (i = 0; i < nr; i++) {
 168			index = indices[i];
 169			head = radix_tree_delete(root, index);
 170			gmap_for_each_rmap_safe(rmap, rnext, head)
 171				kfree(rmap);
 172		}
 173	} while (nr > 0);
 174}
 175
 176/**
 177 * gmap_free - free a guest address space
 178 * @gmap: pointer to the guest address space structure
 179 *
 180 * No locks required. There are no references to this gmap anymore.
 181 */
 182static void gmap_free(struct gmap *gmap)
 183{
 184	struct page *page, *next;
 185
 186	/* Flush tlb of all gmaps (if not already done for shadows) */
 187	if (!(gmap_is_shadow(gmap) && gmap->removed))
 188		gmap_flush_tlb(gmap);
 189	/* Free all segment & region tables. */
 190	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
 191		__free_pages(page, CRST_ALLOC_ORDER);
 192	gmap_radix_tree_free(&gmap->guest_to_host);
 193	gmap_radix_tree_free(&gmap->host_to_guest);
 194
 195	/* Free additional data for a shadow gmap */
 196	if (gmap_is_shadow(gmap)) {
 197		/* Free all page tables. */
 198		list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
 199			page_table_free_pgste(page);
 200		gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
 201		/* Release reference to the parent */
 202		gmap_put(gmap->parent);
 203	}
 204
 205	kfree(gmap);
 206}
 207
 208/**
 209 * gmap_get - increase reference counter for guest address space
 210 * @gmap: pointer to the guest address space structure
 211 *
 212 * Returns the gmap pointer
 213 */
 214struct gmap *gmap_get(struct gmap *gmap)
 215{
 216	refcount_inc(&gmap->ref_count);
 217	return gmap;
 218}
 219EXPORT_SYMBOL_GPL(gmap_get);
 220
 221/**
 222 * gmap_put - decrease reference counter for guest address space
 223 * @gmap: pointer to the guest address space structure
 224 *
 225 * If the reference counter reaches zero the guest address space is freed.
 226 */
 227void gmap_put(struct gmap *gmap)
 228{
 229	if (refcount_dec_and_test(&gmap->ref_count))
 230		gmap_free(gmap);
 231}
 232EXPORT_SYMBOL_GPL(gmap_put);
 233
 234/**
 235 * gmap_remove - remove a guest address space but do not free it yet
 236 * @gmap: pointer to the guest address space structure
 237 */
 238void gmap_remove(struct gmap *gmap)
 239{
 240	struct gmap *sg, *next;
 241	unsigned long gmap_asce;
 242
 243	/* Remove all shadow gmaps linked to this gmap */
 244	if (!list_empty(&gmap->children)) {
 245		spin_lock(&gmap->shadow_lock);
 246		list_for_each_entry_safe(sg, next, &gmap->children, list) {
 247			list_del(&sg->list);
 248			gmap_put(sg);
 249		}
 250		spin_unlock(&gmap->shadow_lock);
 251	}
 252	/* Remove gmap from the pre-mm list */
 253	spin_lock(&gmap->mm->context.lock);
 254	list_del_rcu(&gmap->list);
 255	if (list_empty(&gmap->mm->context.gmap_list))
 256		gmap_asce = 0;
 257	else if (list_is_singular(&gmap->mm->context.gmap_list))
 258		gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
 259					     struct gmap, list)->asce;
 260	else
 261		gmap_asce = -1UL;
 262	WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
 263	spin_unlock(&gmap->mm->context.lock);
 264	synchronize_rcu();
 265	/* Put reference */
 266	gmap_put(gmap);
 267}
 268EXPORT_SYMBOL_GPL(gmap_remove);
 269
 270/**
 271 * gmap_enable - switch primary space to the guest address space
 272 * @gmap: pointer to the guest address space structure
 273 */
 274void gmap_enable(struct gmap *gmap)
 275{
 276	S390_lowcore.gmap = (unsigned long) gmap;
 277}
 278EXPORT_SYMBOL_GPL(gmap_enable);
 279
 280/**
 281 * gmap_disable - switch back to the standard primary address space
 282 * @gmap: pointer to the guest address space structure
 283 */
 284void gmap_disable(struct gmap *gmap)
 285{
 286	S390_lowcore.gmap = 0UL;
 287}
 288EXPORT_SYMBOL_GPL(gmap_disable);
 289
 290/**
 291 * gmap_get_enabled - get a pointer to the currently enabled gmap
 292 *
 293 * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
 294 */
 295struct gmap *gmap_get_enabled(void)
 296{
 297	return (struct gmap *) S390_lowcore.gmap;
 298}
 299EXPORT_SYMBOL_GPL(gmap_get_enabled);
 300
 301/*
 302 * gmap_alloc_table is assumed to be called with mmap_lock held
 303 */
 304static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
 305			    unsigned long init, unsigned long gaddr)
 306{
 307	struct page *page;
 308	unsigned long *new;
 309
 310	/* since we dont free the gmap table until gmap_free we can unlock */
 311	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
 312	if (!page)
 313		return -ENOMEM;
 314	new = page_to_virt(page);
 315	crst_table_init(new, init);
 316	spin_lock(&gmap->guest_table_lock);
 317	if (*table & _REGION_ENTRY_INVALID) {
 318		list_add(&page->lru, &gmap->crst_list);
 319		*table = __pa(new) | _REGION_ENTRY_LENGTH |
 320			(*table & _REGION_ENTRY_TYPE_MASK);
 321		page->index = gaddr;
 322		page = NULL;
 323	}
 324	spin_unlock(&gmap->guest_table_lock);
 325	if (page)
 326		__free_pages(page, CRST_ALLOC_ORDER);
 327	return 0;
 328}
 329
 330/**
 331 * __gmap_segment_gaddr - find virtual address from segment pointer
 332 * @entry: pointer to a segment table entry in the guest address space
 333 *
 334 * Returns the virtual address in the guest address space for the segment
 335 */
 336static unsigned long __gmap_segment_gaddr(unsigned long *entry)
 337{
 338	struct page *page;
 339	unsigned long offset;
 340
 341	offset = (unsigned long) entry / sizeof(unsigned long);
 342	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
 343	page = pmd_pgtable_page((pmd_t *) entry);
 
 344	return page->index + offset;
 345}
 346
 347/**
 348 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
 349 * @gmap: pointer to the guest address space structure
 350 * @vmaddr: address in the host process address space
 351 *
 352 * Returns 1 if a TLB flush is required
 353 */
 354static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
 355{
 356	unsigned long *entry;
 357	int flush = 0;
 358
 359	BUG_ON(gmap_is_shadow(gmap));
 360	spin_lock(&gmap->guest_table_lock);
 361	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
 362	if (entry) {
 363		flush = (*entry != _SEGMENT_ENTRY_EMPTY);
 364		*entry = _SEGMENT_ENTRY_EMPTY;
 365	}
 366	spin_unlock(&gmap->guest_table_lock);
 367	return flush;
 368}
 369
 370/**
 371 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
 372 * @gmap: pointer to the guest address space structure
 373 * @gaddr: address in the guest address space
 374 *
 375 * Returns 1 if a TLB flush is required
 376 */
 377static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
 378{
 379	unsigned long vmaddr;
 380
 381	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
 382						   gaddr >> PMD_SHIFT);
 383	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
 384}
 385
 386/**
 387 * gmap_unmap_segment - unmap segment from the guest address space
 388 * @gmap: pointer to the guest address space structure
 389 * @to: address in the guest address space
 390 * @len: length of the memory area to unmap
 391 *
 392 * Returns 0 if the unmap succeeded, -EINVAL if not.
 393 */
 394int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
 395{
 396	unsigned long off;
 397	int flush;
 398
 399	BUG_ON(gmap_is_shadow(gmap));
 400	if ((to | len) & (PMD_SIZE - 1))
 401		return -EINVAL;
 402	if (len == 0 || to + len < to)
 403		return -EINVAL;
 404
 405	flush = 0;
 406	mmap_write_lock(gmap->mm);
 407	for (off = 0; off < len; off += PMD_SIZE)
 408		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
 409	mmap_write_unlock(gmap->mm);
 410	if (flush)
 411		gmap_flush_tlb(gmap);
 412	return 0;
 413}
 414EXPORT_SYMBOL_GPL(gmap_unmap_segment);
 415
 416/**
 417 * gmap_map_segment - map a segment to the guest address space
 418 * @gmap: pointer to the guest address space structure
 419 * @from: source address in the parent address space
 420 * @to: target address in the guest address space
 421 * @len: length of the memory area to map
 422 *
 423 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
 424 */
 425int gmap_map_segment(struct gmap *gmap, unsigned long from,
 426		     unsigned long to, unsigned long len)
 427{
 428	unsigned long off;
 429	int flush;
 430
 431	BUG_ON(gmap_is_shadow(gmap));
 432	if ((from | to | len) & (PMD_SIZE - 1))
 433		return -EINVAL;
 434	if (len == 0 || from + len < from || to + len < to ||
 435	    from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
 436		return -EINVAL;
 437
 438	flush = 0;
 439	mmap_write_lock(gmap->mm);
 440	for (off = 0; off < len; off += PMD_SIZE) {
 441		/* Remove old translation */
 442		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
 443		/* Store new translation */
 444		if (radix_tree_insert(&gmap->guest_to_host,
 445				      (to + off) >> PMD_SHIFT,
 446				      (void *) from + off))
 447			break;
 448	}
 449	mmap_write_unlock(gmap->mm);
 450	if (flush)
 451		gmap_flush_tlb(gmap);
 452	if (off >= len)
 453		return 0;
 454	gmap_unmap_segment(gmap, to, len);
 455	return -ENOMEM;
 456}
 457EXPORT_SYMBOL_GPL(gmap_map_segment);
 458
 459/**
 460 * __gmap_translate - translate a guest address to a user space address
 461 * @gmap: pointer to guest mapping meta data structure
 462 * @gaddr: guest address
 463 *
 464 * Returns user space address which corresponds to the guest address or
 465 * -EFAULT if no such mapping exists.
 466 * This function does not establish potentially missing page table entries.
 467 * The mmap_lock of the mm that belongs to the address space must be held
 468 * when this function gets called.
 469 *
 470 * Note: Can also be called for shadow gmaps.
 471 */
 472unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
 473{
 474	unsigned long vmaddr;
 475
 476	vmaddr = (unsigned long)
 477		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
 478	/* Note: guest_to_host is empty for a shadow gmap */
 479	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
 480}
 481EXPORT_SYMBOL_GPL(__gmap_translate);
 482
 483/**
 484 * gmap_translate - translate a guest address to a user space address
 485 * @gmap: pointer to guest mapping meta data structure
 486 * @gaddr: guest address
 487 *
 488 * Returns user space address which corresponds to the guest address or
 489 * -EFAULT if no such mapping exists.
 490 * This function does not establish potentially missing page table entries.
 491 */
 492unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
 493{
 494	unsigned long rc;
 495
 496	mmap_read_lock(gmap->mm);
 497	rc = __gmap_translate(gmap, gaddr);
 498	mmap_read_unlock(gmap->mm);
 499	return rc;
 500}
 501EXPORT_SYMBOL_GPL(gmap_translate);
 502
 503/**
 504 * gmap_unlink - disconnect a page table from the gmap shadow tables
 505 * @mm: pointer to the parent mm_struct
 506 * @table: pointer to the host page table
 507 * @vmaddr: vm address associated with the host page table
 508 */
 509void gmap_unlink(struct mm_struct *mm, unsigned long *table,
 510		 unsigned long vmaddr)
 511{
 512	struct gmap *gmap;
 513	int flush;
 514
 515	rcu_read_lock();
 516	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
 517		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
 518		if (flush)
 519			gmap_flush_tlb(gmap);
 520	}
 521	rcu_read_unlock();
 522}
 523
 524static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
 525			   unsigned long gaddr);
 526
 527/**
 528 * __gmap_link - set up shadow page tables to connect a host to a guest address
 529 * @gmap: pointer to guest mapping meta data structure
 530 * @gaddr: guest address
 531 * @vmaddr: vm address
 532 *
 533 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
 534 * if the vm address is already mapped to a different guest segment.
 535 * The mmap_lock of the mm that belongs to the address space must be held
 536 * when this function gets called.
 537 */
 538int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
 539{
 540	struct mm_struct *mm;
 541	unsigned long *table;
 542	spinlock_t *ptl;
 543	pgd_t *pgd;
 544	p4d_t *p4d;
 545	pud_t *pud;
 546	pmd_t *pmd;
 547	u64 unprot;
 548	int rc;
 549
 550	BUG_ON(gmap_is_shadow(gmap));
 551	/* Create higher level tables in the gmap page table */
 552	table = gmap->table;
 553	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
 554		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
 555		if ((*table & _REGION_ENTRY_INVALID) &&
 556		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
 557				     gaddr & _REGION1_MASK))
 558			return -ENOMEM;
 559		table = __va(*table & _REGION_ENTRY_ORIGIN);
 560	}
 561	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
 562		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
 563		if ((*table & _REGION_ENTRY_INVALID) &&
 564		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
 565				     gaddr & _REGION2_MASK))
 566			return -ENOMEM;
 567		table = __va(*table & _REGION_ENTRY_ORIGIN);
 568	}
 569	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
 570		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
 571		if ((*table & _REGION_ENTRY_INVALID) &&
 572		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
 573				     gaddr & _REGION3_MASK))
 574			return -ENOMEM;
 575		table = __va(*table & _REGION_ENTRY_ORIGIN);
 576	}
 577	table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 578	/* Walk the parent mm page table */
 579	mm = gmap->mm;
 580	pgd = pgd_offset(mm, vmaddr);
 581	VM_BUG_ON(pgd_none(*pgd));
 582	p4d = p4d_offset(pgd, vmaddr);
 583	VM_BUG_ON(p4d_none(*p4d));
 584	pud = pud_offset(p4d, vmaddr);
 585	VM_BUG_ON(pud_none(*pud));
 586	/* large puds cannot yet be handled */
 587	if (pud_large(*pud))
 588		return -EFAULT;
 589	pmd = pmd_offset(pud, vmaddr);
 590	VM_BUG_ON(pmd_none(*pmd));
 591	/* Are we allowed to use huge pages? */
 592	if (pmd_large(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
 593		return -EFAULT;
 594	/* Link gmap segment table entry location to page table. */
 595	rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
 596	if (rc)
 597		return rc;
 598	ptl = pmd_lock(mm, pmd);
 599	spin_lock(&gmap->guest_table_lock);
 600	if (*table == _SEGMENT_ENTRY_EMPTY) {
 601		rc = radix_tree_insert(&gmap->host_to_guest,
 602				       vmaddr >> PMD_SHIFT, table);
 603		if (!rc) {
 604			if (pmd_large(*pmd)) {
 605				*table = (pmd_val(*pmd) &
 606					  _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
 607					| _SEGMENT_ENTRY_GMAP_UC;
 608			} else
 609				*table = pmd_val(*pmd) &
 610					_SEGMENT_ENTRY_HARDWARE_BITS;
 611		}
 612	} else if (*table & _SEGMENT_ENTRY_PROTECT &&
 613		   !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
 614		unprot = (u64)*table;
 615		unprot &= ~_SEGMENT_ENTRY_PROTECT;
 616		unprot |= _SEGMENT_ENTRY_GMAP_UC;
 617		gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
 618	}
 619	spin_unlock(&gmap->guest_table_lock);
 620	spin_unlock(ptl);
 621	radix_tree_preload_end();
 622	return rc;
 623}
 624
 625/**
 626 * gmap_fault - resolve a fault on a guest address
 627 * @gmap: pointer to guest mapping meta data structure
 628 * @gaddr: guest address
 629 * @fault_flags: flags to pass down to handle_mm_fault()
 630 *
 631 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
 632 * if the vm address is already mapped to a different guest segment.
 633 */
 634int gmap_fault(struct gmap *gmap, unsigned long gaddr,
 635	       unsigned int fault_flags)
 636{
 637	unsigned long vmaddr;
 638	int rc;
 639	bool unlocked;
 640
 641	mmap_read_lock(gmap->mm);
 642
 643retry:
 644	unlocked = false;
 645	vmaddr = __gmap_translate(gmap, gaddr);
 646	if (IS_ERR_VALUE(vmaddr)) {
 647		rc = vmaddr;
 648		goto out_up;
 649	}
 650	if (fixup_user_fault(gmap->mm, vmaddr, fault_flags,
 651			     &unlocked)) {
 652		rc = -EFAULT;
 653		goto out_up;
 654	}
 655	/*
 656	 * In the case that fixup_user_fault unlocked the mmap_lock during
 657	 * faultin redo __gmap_translate to not race with a map/unmap_segment.
 658	 */
 659	if (unlocked)
 660		goto retry;
 661
 662	rc = __gmap_link(gmap, gaddr, vmaddr);
 663out_up:
 664	mmap_read_unlock(gmap->mm);
 665	return rc;
 666}
 667EXPORT_SYMBOL_GPL(gmap_fault);
 668
 669/*
 670 * this function is assumed to be called with mmap_lock held
 671 */
 672void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
 673{
 674	struct vm_area_struct *vma;
 675	unsigned long vmaddr;
 676	spinlock_t *ptl;
 677	pte_t *ptep;
 678
 679	/* Find the vm address for the guest address */
 680	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
 681						   gaddr >> PMD_SHIFT);
 682	if (vmaddr) {
 683		vmaddr |= gaddr & ~PMD_MASK;
 684
 685		vma = vma_lookup(gmap->mm, vmaddr);
 686		if (!vma || is_vm_hugetlb_page(vma))
 687			return;
 688
 689		/* Get pointer to the page table entry */
 690		ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
 691		if (likely(ptep)) {
 692			ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
 693			pte_unmap_unlock(ptep, ptl);
 694		}
 695	}
 696}
 697EXPORT_SYMBOL_GPL(__gmap_zap);
 698
 699void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
 700{
 701	unsigned long gaddr, vmaddr, size;
 702	struct vm_area_struct *vma;
 703
 704	mmap_read_lock(gmap->mm);
 705	for (gaddr = from; gaddr < to;
 706	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
 707		/* Find the vm address for the guest address */
 708		vmaddr = (unsigned long)
 709			radix_tree_lookup(&gmap->guest_to_host,
 710					  gaddr >> PMD_SHIFT);
 711		if (!vmaddr)
 712			continue;
 713		vmaddr |= gaddr & ~PMD_MASK;
 714		/* Find vma in the parent mm */
 715		vma = find_vma(gmap->mm, vmaddr);
 716		if (!vma)
 717			continue;
 718		/*
 719		 * We do not discard pages that are backed by
 720		 * hugetlbfs, so we don't have to refault them.
 721		 */
 722		if (is_vm_hugetlb_page(vma))
 723			continue;
 724		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
 725		zap_page_range(vma, vmaddr, size);
 726	}
 727	mmap_read_unlock(gmap->mm);
 728}
 729EXPORT_SYMBOL_GPL(gmap_discard);
 730
 731static LIST_HEAD(gmap_notifier_list);
 732static DEFINE_SPINLOCK(gmap_notifier_lock);
 733
 734/**
 735 * gmap_register_pte_notifier - register a pte invalidation callback
 736 * @nb: pointer to the gmap notifier block
 737 */
 738void gmap_register_pte_notifier(struct gmap_notifier *nb)
 739{
 740	spin_lock(&gmap_notifier_lock);
 741	list_add_rcu(&nb->list, &gmap_notifier_list);
 742	spin_unlock(&gmap_notifier_lock);
 743}
 744EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
 745
 746/**
 747 * gmap_unregister_pte_notifier - remove a pte invalidation callback
 748 * @nb: pointer to the gmap notifier block
 749 */
 750void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
 751{
 752	spin_lock(&gmap_notifier_lock);
 753	list_del_rcu(&nb->list);
 754	spin_unlock(&gmap_notifier_lock);
 755	synchronize_rcu();
 756}
 757EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
 758
 759/**
 760 * gmap_call_notifier - call all registered invalidation callbacks
 761 * @gmap: pointer to guest mapping meta data structure
 762 * @start: start virtual address in the guest address space
 763 * @end: end virtual address in the guest address space
 764 */
 765static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
 766			       unsigned long end)
 767{
 768	struct gmap_notifier *nb;
 769
 770	list_for_each_entry(nb, &gmap_notifier_list, list)
 771		nb->notifier_call(gmap, start, end);
 772}
 773
 774/**
 775 * gmap_table_walk - walk the gmap page tables
 776 * @gmap: pointer to guest mapping meta data structure
 777 * @gaddr: virtual address in the guest address space
 778 * @level: page table level to stop at
 779 *
 780 * Returns a table entry pointer for the given guest address and @level
 781 * @level=0 : returns a pointer to a page table table entry (or NULL)
 782 * @level=1 : returns a pointer to a segment table entry (or NULL)
 783 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
 784 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
 785 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
 786 *
 787 * Returns NULL if the gmap page tables could not be walked to the
 788 * requested level.
 789 *
 790 * Note: Can also be called for shadow gmaps.
 791 */
 792static inline unsigned long *gmap_table_walk(struct gmap *gmap,
 793					     unsigned long gaddr, int level)
 794{
 795	const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
 796	unsigned long *table = gmap->table;
 797
 798	if (gmap_is_shadow(gmap) && gmap->removed)
 799		return NULL;
 800
 801	if (WARN_ON_ONCE(level > (asce_type >> 2) + 1))
 802		return NULL;
 803
 804	if (asce_type != _ASCE_TYPE_REGION1 &&
 805	    gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
 806		return NULL;
 807
 808	switch (asce_type) {
 809	case _ASCE_TYPE_REGION1:
 810		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
 811		if (level == 4)
 812			break;
 813		if (*table & _REGION_ENTRY_INVALID)
 814			return NULL;
 815		table = __va(*table & _REGION_ENTRY_ORIGIN);
 816		fallthrough;
 817	case _ASCE_TYPE_REGION2:
 818		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
 819		if (level == 3)
 820			break;
 821		if (*table & _REGION_ENTRY_INVALID)
 822			return NULL;
 823		table = __va(*table & _REGION_ENTRY_ORIGIN);
 824		fallthrough;
 825	case _ASCE_TYPE_REGION3:
 826		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
 827		if (level == 2)
 828			break;
 829		if (*table & _REGION_ENTRY_INVALID)
 830			return NULL;
 831		table = __va(*table & _REGION_ENTRY_ORIGIN);
 832		fallthrough;
 833	case _ASCE_TYPE_SEGMENT:
 834		table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 835		if (level == 1)
 836			break;
 837		if (*table & _REGION_ENTRY_INVALID)
 838			return NULL;
 839		table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
 840		table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
 841	}
 842	return table;
 843}
 844
 845/**
 846 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
 847 *		      and return the pte pointer
 848 * @gmap: pointer to guest mapping meta data structure
 849 * @gaddr: virtual address in the guest address space
 850 * @ptl: pointer to the spinlock pointer
 851 *
 852 * Returns a pointer to the locked pte for a guest address, or NULL
 853 */
 854static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
 855			       spinlock_t **ptl)
 856{
 857	unsigned long *table;
 858
 859	BUG_ON(gmap_is_shadow(gmap));
 860	/* Walk the gmap page table, lock and get pte pointer */
 861	table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
 862	if (!table || *table & _SEGMENT_ENTRY_INVALID)
 863		return NULL;
 864	return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
 865}
 866
 867/**
 868 * gmap_pte_op_fixup - force a page in and connect the gmap page table
 869 * @gmap: pointer to guest mapping meta data structure
 870 * @gaddr: virtual address in the guest address space
 871 * @vmaddr: address in the host process address space
 872 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
 873 *
 874 * Returns 0 if the caller can retry __gmap_translate (might fail again),
 875 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
 876 * up or connecting the gmap page table.
 877 */
 878static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
 879			     unsigned long vmaddr, int prot)
 880{
 881	struct mm_struct *mm = gmap->mm;
 882	unsigned int fault_flags;
 883	bool unlocked = false;
 884
 885	BUG_ON(gmap_is_shadow(gmap));
 886	fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
 887	if (fixup_user_fault(mm, vmaddr, fault_flags, &unlocked))
 888		return -EFAULT;
 889	if (unlocked)
 890		/* lost mmap_lock, caller has to retry __gmap_translate */
 891		return 0;
 892	/* Connect the page tables */
 893	return __gmap_link(gmap, gaddr, vmaddr);
 894}
 895
 896/**
 897 * gmap_pte_op_end - release the page table lock
 898 * @ptl: pointer to the spinlock pointer
 899 */
 900static void gmap_pte_op_end(spinlock_t *ptl)
 901{
 902	if (ptl)
 903		spin_unlock(ptl);
 904}
 905
 906/**
 907 * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
 908 *		      and return the pmd pointer
 909 * @gmap: pointer to guest mapping meta data structure
 910 * @gaddr: virtual address in the guest address space
 911 *
 912 * Returns a pointer to the pmd for a guest address, or NULL
 913 */
 914static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
 915{
 916	pmd_t *pmdp;
 917
 918	BUG_ON(gmap_is_shadow(gmap));
 919	pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
 920	if (!pmdp)
 921		return NULL;
 922
 923	/* without huge pages, there is no need to take the table lock */
 924	if (!gmap->mm->context.allow_gmap_hpage_1m)
 925		return pmd_none(*pmdp) ? NULL : pmdp;
 926
 927	spin_lock(&gmap->guest_table_lock);
 928	if (pmd_none(*pmdp)) {
 929		spin_unlock(&gmap->guest_table_lock);
 930		return NULL;
 931	}
 932
 933	/* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
 934	if (!pmd_large(*pmdp))
 935		spin_unlock(&gmap->guest_table_lock);
 936	return pmdp;
 937}
 938
 939/**
 940 * gmap_pmd_op_end - release the guest_table_lock if needed
 941 * @gmap: pointer to the guest mapping meta data structure
 942 * @pmdp: pointer to the pmd
 943 */
 944static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
 945{
 946	if (pmd_large(*pmdp))
 947		spin_unlock(&gmap->guest_table_lock);
 948}
 949
 950/*
 951 * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
 952 * @pmdp: pointer to the pmd to be protected
 953 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
 954 * @bits: notification bits to set
 955 *
 956 * Returns:
 957 * 0 if successfully protected
 958 * -EAGAIN if a fixup is needed
 959 * -EINVAL if unsupported notifier bits have been specified
 960 *
 961 * Expected to be called with sg->mm->mmap_lock in read and
 962 * guest_table_lock held.
 963 */
 964static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
 965			    pmd_t *pmdp, int prot, unsigned long bits)
 966{
 967	int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
 968	int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
 969	pmd_t new = *pmdp;
 970
 971	/* Fixup needed */
 972	if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
 973		return -EAGAIN;
 974
 975	if (prot == PROT_NONE && !pmd_i) {
 976		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
 977		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
 978	}
 979
 980	if (prot == PROT_READ && !pmd_p) {
 981		new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
 982		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_PROTECT));
 983		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
 984	}
 985
 986	if (bits & GMAP_NOTIFY_MPROT)
 987		set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
 988
 989	/* Shadow GMAP protection needs split PMDs */
 990	if (bits & GMAP_NOTIFY_SHADOW)
 991		return -EINVAL;
 992
 993	return 0;
 994}
 995
 996/*
 997 * gmap_protect_pte - remove access rights to memory and set pgste bits
 998 * @gmap: pointer to guest mapping meta data structure
 999 * @gaddr: virtual address in the guest address space
1000 * @pmdp: pointer to the pmd associated with the pte
1001 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1002 * @bits: notification bits to set
1003 *
1004 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1005 * -EAGAIN if a fixup is needed.
1006 *
1007 * Expected to be called with sg->mm->mmap_lock in read
1008 */
1009static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
1010			    pmd_t *pmdp, int prot, unsigned long bits)
1011{
1012	int rc;
1013	pte_t *ptep;
1014	spinlock_t *ptl = NULL;
1015	unsigned long pbits = 0;
1016
1017	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1018		return -EAGAIN;
1019
1020	ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1021	if (!ptep)
1022		return -ENOMEM;
1023
1024	pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1025	pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1026	/* Protect and unlock. */
1027	rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1028	gmap_pte_op_end(ptl);
1029	return rc;
1030}
1031
1032/*
1033 * gmap_protect_range - remove access rights to memory and set pgste bits
1034 * @gmap: pointer to guest mapping meta data structure
1035 * @gaddr: virtual address in the guest address space
1036 * @len: size of area
1037 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1038 * @bits: pgste notification bits to set
1039 *
1040 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1041 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1042 *
1043 * Called with sg->mm->mmap_lock in read.
1044 */
1045static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1046			      unsigned long len, int prot, unsigned long bits)
1047{
1048	unsigned long vmaddr, dist;
1049	pmd_t *pmdp;
1050	int rc;
1051
1052	BUG_ON(gmap_is_shadow(gmap));
1053	while (len) {
1054		rc = -EAGAIN;
1055		pmdp = gmap_pmd_op_walk(gmap, gaddr);
1056		if (pmdp) {
1057			if (!pmd_large(*pmdp)) {
1058				rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1059						      bits);
1060				if (!rc) {
1061					len -= PAGE_SIZE;
1062					gaddr += PAGE_SIZE;
1063				}
1064			} else {
1065				rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1066						      bits);
1067				if (!rc) {
1068					dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1069					len = len < dist ? 0 : len - dist;
1070					gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1071				}
1072			}
1073			gmap_pmd_op_end(gmap, pmdp);
1074		}
1075		if (rc) {
1076			if (rc == -EINVAL)
1077				return rc;
1078
1079			/* -EAGAIN, fixup of userspace mm and gmap */
1080			vmaddr = __gmap_translate(gmap, gaddr);
1081			if (IS_ERR_VALUE(vmaddr))
1082				return vmaddr;
1083			rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1084			if (rc)
1085				return rc;
1086		}
1087	}
1088	return 0;
1089}
1090
1091/**
1092 * gmap_mprotect_notify - change access rights for a range of ptes and
1093 *                        call the notifier if any pte changes again
1094 * @gmap: pointer to guest mapping meta data structure
1095 * @gaddr: virtual address in the guest address space
1096 * @len: size of area
1097 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1098 *
1099 * Returns 0 if for each page in the given range a gmap mapping exists,
1100 * the new access rights could be set and the notifier could be armed.
1101 * If the gmap mapping is missing for one or more pages -EFAULT is
1102 * returned. If no memory could be allocated -ENOMEM is returned.
1103 * This function establishes missing page table entries.
1104 */
1105int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1106			 unsigned long len, int prot)
1107{
1108	int rc;
1109
1110	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1111		return -EINVAL;
1112	if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1113		return -EINVAL;
1114	mmap_read_lock(gmap->mm);
1115	rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1116	mmap_read_unlock(gmap->mm);
1117	return rc;
1118}
1119EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1120
1121/**
1122 * gmap_read_table - get an unsigned long value from a guest page table using
1123 *                   absolute addressing, without marking the page referenced.
1124 * @gmap: pointer to guest mapping meta data structure
1125 * @gaddr: virtual address in the guest address space
1126 * @val: pointer to the unsigned long value to return
1127 *
1128 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1129 * if reading using the virtual address failed. -EINVAL if called on a gmap
1130 * shadow.
1131 *
1132 * Called with gmap->mm->mmap_lock in read.
1133 */
1134int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1135{
1136	unsigned long address, vmaddr;
1137	spinlock_t *ptl;
1138	pte_t *ptep, pte;
1139	int rc;
1140
1141	if (gmap_is_shadow(gmap))
1142		return -EINVAL;
1143
1144	while (1) {
1145		rc = -EAGAIN;
1146		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1147		if (ptep) {
1148			pte = *ptep;
1149			if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1150				address = pte_val(pte) & PAGE_MASK;
1151				address += gaddr & ~PAGE_MASK;
1152				*val = *(unsigned long *)__va(address);
1153				set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_YOUNG)));
1154				/* Do *NOT* clear the _PAGE_INVALID bit! */
1155				rc = 0;
1156			}
1157			gmap_pte_op_end(ptl);
1158		}
1159		if (!rc)
1160			break;
1161		vmaddr = __gmap_translate(gmap, gaddr);
1162		if (IS_ERR_VALUE(vmaddr)) {
1163			rc = vmaddr;
1164			break;
1165		}
1166		rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1167		if (rc)
1168			break;
1169	}
1170	return rc;
1171}
1172EXPORT_SYMBOL_GPL(gmap_read_table);
1173
1174/**
1175 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1176 * @sg: pointer to the shadow guest address space structure
1177 * @vmaddr: vm address associated with the rmap
1178 * @rmap: pointer to the rmap structure
1179 *
1180 * Called with the sg->guest_table_lock
1181 */
1182static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1183				    struct gmap_rmap *rmap)
1184{
1185	struct gmap_rmap *temp;
1186	void __rcu **slot;
1187
1188	BUG_ON(!gmap_is_shadow(sg));
1189	slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1190	if (slot) {
1191		rmap->next = radix_tree_deref_slot_protected(slot,
1192							&sg->guest_table_lock);
1193		for (temp = rmap->next; temp; temp = temp->next) {
1194			if (temp->raddr == rmap->raddr) {
1195				kfree(rmap);
1196				return;
1197			}
1198		}
1199		radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1200	} else {
1201		rmap->next = NULL;
1202		radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1203				  rmap);
1204	}
1205}
1206
1207/**
1208 * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1209 * @sg: pointer to the shadow guest address space structure
1210 * @raddr: rmap address in the shadow gmap
1211 * @paddr: address in the parent guest address space
1212 * @len: length of the memory area to protect
1213 *
1214 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1215 * if out of memory and -EFAULT if paddr is invalid.
1216 */
1217static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1218			     unsigned long paddr, unsigned long len)
1219{
1220	struct gmap *parent;
1221	struct gmap_rmap *rmap;
1222	unsigned long vmaddr;
1223	spinlock_t *ptl;
1224	pte_t *ptep;
1225	int rc;
1226
1227	BUG_ON(!gmap_is_shadow(sg));
1228	parent = sg->parent;
1229	while (len) {
1230		vmaddr = __gmap_translate(parent, paddr);
1231		if (IS_ERR_VALUE(vmaddr))
1232			return vmaddr;
1233		rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
1234		if (!rmap)
1235			return -ENOMEM;
1236		rmap->raddr = raddr;
1237		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
1238		if (rc) {
1239			kfree(rmap);
1240			return rc;
1241		}
1242		rc = -EAGAIN;
1243		ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1244		if (ptep) {
1245			spin_lock(&sg->guest_table_lock);
1246			rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1247					     PGSTE_VSIE_BIT);
1248			if (!rc)
1249				gmap_insert_rmap(sg, vmaddr, rmap);
1250			spin_unlock(&sg->guest_table_lock);
1251			gmap_pte_op_end(ptl);
1252		}
1253		radix_tree_preload_end();
1254		if (rc) {
1255			kfree(rmap);
1256			rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1257			if (rc)
1258				return rc;
1259			continue;
1260		}
1261		paddr += PAGE_SIZE;
1262		len -= PAGE_SIZE;
1263	}
1264	return 0;
1265}
1266
1267#define _SHADOW_RMAP_MASK	0x7
1268#define _SHADOW_RMAP_REGION1	0x5
1269#define _SHADOW_RMAP_REGION2	0x4
1270#define _SHADOW_RMAP_REGION3	0x3
1271#define _SHADOW_RMAP_SEGMENT	0x2
1272#define _SHADOW_RMAP_PGTABLE	0x1
1273
1274/**
1275 * gmap_idte_one - invalidate a single region or segment table entry
1276 * @asce: region or segment table *origin* + table-type bits
1277 * @vaddr: virtual address to identify the table entry to flush
1278 *
1279 * The invalid bit of a single region or segment table entry is set
1280 * and the associated TLB entries depending on the entry are flushed.
1281 * The table-type of the @asce identifies the portion of the @vaddr
1282 * that is used as the invalidation index.
1283 */
1284static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1285{
1286	asm volatile(
1287		"	idte	%0,0,%1"
1288		: : "a" (asce), "a" (vaddr) : "cc", "memory");
1289}
1290
1291/**
1292 * gmap_unshadow_page - remove a page from a shadow page table
1293 * @sg: pointer to the shadow guest address space structure
1294 * @raddr: rmap address in the shadow guest address space
1295 *
1296 * Called with the sg->guest_table_lock
1297 */
1298static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1299{
1300	unsigned long *table;
1301
1302	BUG_ON(!gmap_is_shadow(sg));
1303	table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1304	if (!table || *table & _PAGE_INVALID)
1305		return;
1306	gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1307	ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1308}
1309
1310/**
1311 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1312 * @sg: pointer to the shadow guest address space structure
1313 * @raddr: rmap address in the shadow guest address space
1314 * @pgt: pointer to the start of a shadow page table
1315 *
1316 * Called with the sg->guest_table_lock
1317 */
1318static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1319				unsigned long *pgt)
1320{
1321	int i;
1322
1323	BUG_ON(!gmap_is_shadow(sg));
1324	for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1325		pgt[i] = _PAGE_INVALID;
1326}
1327
1328/**
1329 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1330 * @sg: pointer to the shadow guest address space structure
1331 * @raddr: address in the shadow guest address space
1332 *
1333 * Called with the sg->guest_table_lock
1334 */
1335static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1336{
1337	unsigned long *ste;
1338	phys_addr_t sto, pgt;
1339	struct page *page;
1340
1341	BUG_ON(!gmap_is_shadow(sg));
1342	ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1343	if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1344		return;
1345	gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1346	sto = __pa(ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1347	gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1348	pgt = *ste & _SEGMENT_ENTRY_ORIGIN;
1349	*ste = _SEGMENT_ENTRY_EMPTY;
1350	__gmap_unshadow_pgt(sg, raddr, __va(pgt));
1351	/* Free page table */
1352	page = phys_to_page(pgt);
1353	list_del(&page->lru);
1354	page_table_free_pgste(page);
1355}
1356
1357/**
1358 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1359 * @sg: pointer to the shadow guest address space structure
1360 * @raddr: rmap address in the shadow guest address space
1361 * @sgt: pointer to the start of a shadow segment table
1362 *
1363 * Called with the sg->guest_table_lock
1364 */
1365static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1366				unsigned long *sgt)
1367{
 
1368	struct page *page;
1369	phys_addr_t pgt;
1370	int i;
1371
1372	BUG_ON(!gmap_is_shadow(sg));
1373	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1374		if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1375			continue;
1376		pgt = sgt[i] & _REGION_ENTRY_ORIGIN;
1377		sgt[i] = _SEGMENT_ENTRY_EMPTY;
1378		__gmap_unshadow_pgt(sg, raddr, __va(pgt));
1379		/* Free page table */
1380		page = phys_to_page(pgt);
1381		list_del(&page->lru);
1382		page_table_free_pgste(page);
1383	}
1384}
1385
1386/**
1387 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1388 * @sg: pointer to the shadow guest address space structure
1389 * @raddr: rmap address in the shadow guest address space
1390 *
1391 * Called with the shadow->guest_table_lock
1392 */
1393static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1394{
1395	unsigned long r3o, *r3e;
1396	phys_addr_t sgt;
1397	struct page *page;
1398
1399	BUG_ON(!gmap_is_shadow(sg));
1400	r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1401	if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1402		return;
1403	gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1404	r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1405	gmap_idte_one(__pa(r3o) | _ASCE_TYPE_REGION3, raddr);
1406	sgt = *r3e & _REGION_ENTRY_ORIGIN;
1407	*r3e = _REGION3_ENTRY_EMPTY;
1408	__gmap_unshadow_sgt(sg, raddr, __va(sgt));
1409	/* Free segment table */
1410	page = phys_to_page(sgt);
1411	list_del(&page->lru);
1412	__free_pages(page, CRST_ALLOC_ORDER);
1413}
1414
1415/**
1416 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1417 * @sg: pointer to the shadow guest address space structure
1418 * @raddr: address in the shadow guest address space
1419 * @r3t: pointer to the start of a shadow region-3 table
1420 *
1421 * Called with the sg->guest_table_lock
1422 */
1423static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1424				unsigned long *r3t)
1425{
 
1426	struct page *page;
1427	phys_addr_t sgt;
1428	int i;
1429
1430	BUG_ON(!gmap_is_shadow(sg));
1431	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1432		if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1433			continue;
1434		sgt = r3t[i] & _REGION_ENTRY_ORIGIN;
1435		r3t[i] = _REGION3_ENTRY_EMPTY;
1436		__gmap_unshadow_sgt(sg, raddr, __va(sgt));
1437		/* Free segment table */
1438		page = phys_to_page(sgt);
1439		list_del(&page->lru);
1440		__free_pages(page, CRST_ALLOC_ORDER);
1441	}
1442}
1443
1444/**
1445 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1446 * @sg: pointer to the shadow guest address space structure
1447 * @raddr: rmap address in the shadow guest address space
1448 *
1449 * Called with the sg->guest_table_lock
1450 */
1451static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1452{
1453	unsigned long r2o, *r2e;
1454	phys_addr_t r3t;
1455	struct page *page;
1456
1457	BUG_ON(!gmap_is_shadow(sg));
1458	r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1459	if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1460		return;
1461	gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1462	r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1463	gmap_idte_one(__pa(r2o) | _ASCE_TYPE_REGION2, raddr);
1464	r3t = *r2e & _REGION_ENTRY_ORIGIN;
1465	*r2e = _REGION2_ENTRY_EMPTY;
1466	__gmap_unshadow_r3t(sg, raddr, __va(r3t));
1467	/* Free region 3 table */
1468	page = phys_to_page(r3t);
1469	list_del(&page->lru);
1470	__free_pages(page, CRST_ALLOC_ORDER);
1471}
1472
1473/**
1474 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1475 * @sg: pointer to the shadow guest address space structure
1476 * @raddr: rmap address in the shadow guest address space
1477 * @r2t: pointer to the start of a shadow region-2 table
1478 *
1479 * Called with the sg->guest_table_lock
1480 */
1481static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1482				unsigned long *r2t)
1483{
1484	phys_addr_t r3t;
1485	struct page *page;
1486	int i;
1487
1488	BUG_ON(!gmap_is_shadow(sg));
1489	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1490		if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1491			continue;
1492		r3t = r2t[i] & _REGION_ENTRY_ORIGIN;
1493		r2t[i] = _REGION2_ENTRY_EMPTY;
1494		__gmap_unshadow_r3t(sg, raddr, __va(r3t));
1495		/* Free region 3 table */
1496		page = phys_to_page(r3t);
1497		list_del(&page->lru);
1498		__free_pages(page, CRST_ALLOC_ORDER);
1499	}
1500}
1501
1502/**
1503 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1504 * @sg: pointer to the shadow guest address space structure
1505 * @raddr: rmap address in the shadow guest address space
1506 *
1507 * Called with the sg->guest_table_lock
1508 */
1509static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1510{
1511	unsigned long r1o, *r1e;
1512	struct page *page;
1513	phys_addr_t r2t;
1514
1515	BUG_ON(!gmap_is_shadow(sg));
1516	r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1517	if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1518		return;
1519	gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1520	r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1521	gmap_idte_one(__pa(r1o) | _ASCE_TYPE_REGION1, raddr);
1522	r2t = *r1e & _REGION_ENTRY_ORIGIN;
1523	*r1e = _REGION1_ENTRY_EMPTY;
1524	__gmap_unshadow_r2t(sg, raddr, __va(r2t));
1525	/* Free region 2 table */
1526	page = phys_to_page(r2t);
1527	list_del(&page->lru);
1528	__free_pages(page, CRST_ALLOC_ORDER);
1529}
1530
1531/**
1532 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1533 * @sg: pointer to the shadow guest address space structure
1534 * @raddr: rmap address in the shadow guest address space
1535 * @r1t: pointer to the start of a shadow region-1 table
1536 *
1537 * Called with the shadow->guest_table_lock
1538 */
1539static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1540				unsigned long *r1t)
1541{
1542	unsigned long asce;
1543	struct page *page;
1544	phys_addr_t r2t;
1545	int i;
1546
1547	BUG_ON(!gmap_is_shadow(sg));
1548	asce = __pa(r1t) | _ASCE_TYPE_REGION1;
1549	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1550		if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1551			continue;
1552		r2t = r1t[i] & _REGION_ENTRY_ORIGIN;
1553		__gmap_unshadow_r2t(sg, raddr, __va(r2t));
1554		/* Clear entry and flush translation r1t -> r2t */
1555		gmap_idte_one(asce, raddr);
1556		r1t[i] = _REGION1_ENTRY_EMPTY;
1557		/* Free region 2 table */
1558		page = phys_to_page(r2t);
1559		list_del(&page->lru);
1560		__free_pages(page, CRST_ALLOC_ORDER);
1561	}
1562}
1563
1564/**
1565 * gmap_unshadow - remove a shadow page table completely
1566 * @sg: pointer to the shadow guest address space structure
1567 *
1568 * Called with sg->guest_table_lock
1569 */
1570static void gmap_unshadow(struct gmap *sg)
1571{
1572	unsigned long *table;
1573
1574	BUG_ON(!gmap_is_shadow(sg));
1575	if (sg->removed)
1576		return;
1577	sg->removed = 1;
1578	gmap_call_notifier(sg, 0, -1UL);
1579	gmap_flush_tlb(sg);
1580	table = __va(sg->asce & _ASCE_ORIGIN);
1581	switch (sg->asce & _ASCE_TYPE_MASK) {
1582	case _ASCE_TYPE_REGION1:
1583		__gmap_unshadow_r1t(sg, 0, table);
1584		break;
1585	case _ASCE_TYPE_REGION2:
1586		__gmap_unshadow_r2t(sg, 0, table);
1587		break;
1588	case _ASCE_TYPE_REGION3:
1589		__gmap_unshadow_r3t(sg, 0, table);
1590		break;
1591	case _ASCE_TYPE_SEGMENT:
1592		__gmap_unshadow_sgt(sg, 0, table);
1593		break;
1594	}
1595}
1596
1597/**
1598 * gmap_find_shadow - find a specific asce in the list of shadow tables
1599 * @parent: pointer to the parent gmap
1600 * @asce: ASCE for which the shadow table is created
1601 * @edat_level: edat level to be used for the shadow translation
1602 *
1603 * Returns the pointer to a gmap if a shadow table with the given asce is
1604 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1605 * otherwise NULL
1606 */
1607static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1608				     int edat_level)
1609{
1610	struct gmap *sg;
1611
1612	list_for_each_entry(sg, &parent->children, list) {
1613		if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1614		    sg->removed)
1615			continue;
1616		if (!sg->initialized)
1617			return ERR_PTR(-EAGAIN);
1618		refcount_inc(&sg->ref_count);
1619		return sg;
1620	}
1621	return NULL;
1622}
1623
1624/**
1625 * gmap_shadow_valid - check if a shadow guest address space matches the
1626 *                     given properties and is still valid
1627 * @sg: pointer to the shadow guest address space structure
1628 * @asce: ASCE for which the shadow table is requested
1629 * @edat_level: edat level to be used for the shadow translation
1630 *
1631 * Returns 1 if the gmap shadow is still valid and matches the given
1632 * properties, the caller can continue using it. Returns 0 otherwise, the
1633 * caller has to request a new shadow gmap in this case.
1634 *
1635 */
1636int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1637{
1638	if (sg->removed)
1639		return 0;
1640	return sg->orig_asce == asce && sg->edat_level == edat_level;
1641}
1642EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1643
1644/**
1645 * gmap_shadow - create/find a shadow guest address space
1646 * @parent: pointer to the parent gmap
1647 * @asce: ASCE for which the shadow table is created
1648 * @edat_level: edat level to be used for the shadow translation
1649 *
1650 * The pages of the top level page table referred by the asce parameter
1651 * will be set to read-only and marked in the PGSTEs of the kvm process.
1652 * The shadow table will be removed automatically on any change to the
1653 * PTE mapping for the source table.
1654 *
1655 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1656 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1657 * parent gmap table could not be protected.
1658 */
1659struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1660			 int edat_level)
1661{
1662	struct gmap *sg, *new;
1663	unsigned long limit;
1664	int rc;
1665
1666	BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1667	BUG_ON(gmap_is_shadow(parent));
1668	spin_lock(&parent->shadow_lock);
1669	sg = gmap_find_shadow(parent, asce, edat_level);
1670	spin_unlock(&parent->shadow_lock);
1671	if (sg)
1672		return sg;
1673	/* Create a new shadow gmap */
1674	limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1675	if (asce & _ASCE_REAL_SPACE)
1676		limit = -1UL;
1677	new = gmap_alloc(limit);
1678	if (!new)
1679		return ERR_PTR(-ENOMEM);
1680	new->mm = parent->mm;
1681	new->parent = gmap_get(parent);
1682	new->orig_asce = asce;
1683	new->edat_level = edat_level;
1684	new->initialized = false;
1685	spin_lock(&parent->shadow_lock);
1686	/* Recheck if another CPU created the same shadow */
1687	sg = gmap_find_shadow(parent, asce, edat_level);
1688	if (sg) {
1689		spin_unlock(&parent->shadow_lock);
1690		gmap_free(new);
1691		return sg;
1692	}
1693	if (asce & _ASCE_REAL_SPACE) {
1694		/* only allow one real-space gmap shadow */
1695		list_for_each_entry(sg, &parent->children, list) {
1696			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1697				spin_lock(&sg->guest_table_lock);
1698				gmap_unshadow(sg);
1699				spin_unlock(&sg->guest_table_lock);
1700				list_del(&sg->list);
1701				gmap_put(sg);
1702				break;
1703			}
1704		}
1705	}
1706	refcount_set(&new->ref_count, 2);
1707	list_add(&new->list, &parent->children);
1708	if (asce & _ASCE_REAL_SPACE) {
1709		/* nothing to protect, return right away */
1710		new->initialized = true;
1711		spin_unlock(&parent->shadow_lock);
1712		return new;
1713	}
1714	spin_unlock(&parent->shadow_lock);
1715	/* protect after insertion, so it will get properly invalidated */
1716	mmap_read_lock(parent->mm);
1717	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1718				((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1719				PROT_READ, GMAP_NOTIFY_SHADOW);
1720	mmap_read_unlock(parent->mm);
1721	spin_lock(&parent->shadow_lock);
1722	new->initialized = true;
1723	if (rc) {
1724		list_del(&new->list);
1725		gmap_free(new);
1726		new = ERR_PTR(rc);
1727	}
1728	spin_unlock(&parent->shadow_lock);
1729	return new;
1730}
1731EXPORT_SYMBOL_GPL(gmap_shadow);
1732
1733/**
1734 * gmap_shadow_r2t - create an empty shadow region 2 table
1735 * @sg: pointer to the shadow guest address space structure
1736 * @saddr: faulting address in the shadow gmap
1737 * @r2t: parent gmap address of the region 2 table to get shadowed
1738 * @fake: r2t references contiguous guest memory block, not a r2t
1739 *
1740 * The r2t parameter specifies the address of the source table. The
1741 * four pages of the source table are made read-only in the parent gmap
1742 * address space. A write to the source table area @r2t will automatically
1743 * remove the shadow r2 table and all of its decendents.
1744 *
1745 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1746 * shadow table structure is incomplete, -ENOMEM if out of memory and
1747 * -EFAULT if an address in the parent gmap could not be resolved.
1748 *
1749 * Called with sg->mm->mmap_lock in read.
1750 */
1751int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1752		    int fake)
1753{
1754	unsigned long raddr, origin, offset, len;
1755	unsigned long *table;
1756	phys_addr_t s_r2t;
1757	struct page *page;
1758	int rc;
1759
1760	BUG_ON(!gmap_is_shadow(sg));
1761	/* Allocate a shadow region second table */
1762	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
1763	if (!page)
1764		return -ENOMEM;
1765	page->index = r2t & _REGION_ENTRY_ORIGIN;
1766	if (fake)
1767		page->index |= GMAP_SHADOW_FAKE_TABLE;
1768	s_r2t = page_to_phys(page);
1769	/* Install shadow region second table */
1770	spin_lock(&sg->guest_table_lock);
1771	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1772	if (!table) {
1773		rc = -EAGAIN;		/* Race with unshadow */
1774		goto out_free;
1775	}
1776	if (!(*table & _REGION_ENTRY_INVALID)) {
1777		rc = 0;			/* Already established */
1778		goto out_free;
1779	} else if (*table & _REGION_ENTRY_ORIGIN) {
1780		rc = -EAGAIN;		/* Race with shadow */
1781		goto out_free;
1782	}
1783	crst_table_init(__va(s_r2t), _REGION2_ENTRY_EMPTY);
1784	/* mark as invalid as long as the parent table is not protected */
1785	*table = s_r2t | _REGION_ENTRY_LENGTH |
1786		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1787	if (sg->edat_level >= 1)
1788		*table |= (r2t & _REGION_ENTRY_PROTECT);
1789	list_add(&page->lru, &sg->crst_list);
1790	if (fake) {
1791		/* nothing to protect for fake tables */
1792		*table &= ~_REGION_ENTRY_INVALID;
1793		spin_unlock(&sg->guest_table_lock);
1794		return 0;
1795	}
1796	spin_unlock(&sg->guest_table_lock);
1797	/* Make r2t read-only in parent gmap page table */
1798	raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1799	origin = r2t & _REGION_ENTRY_ORIGIN;
1800	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1801	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1802	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1803	spin_lock(&sg->guest_table_lock);
1804	if (!rc) {
1805		table = gmap_table_walk(sg, saddr, 4);
1806		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r2t)
 
1807			rc = -EAGAIN;		/* Race with unshadow */
1808		else
1809			*table &= ~_REGION_ENTRY_INVALID;
1810	} else {
1811		gmap_unshadow_r2t(sg, raddr);
1812	}
1813	spin_unlock(&sg->guest_table_lock);
1814	return rc;
1815out_free:
1816	spin_unlock(&sg->guest_table_lock);
1817	__free_pages(page, CRST_ALLOC_ORDER);
1818	return rc;
1819}
1820EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1821
1822/**
1823 * gmap_shadow_r3t - create a shadow region 3 table
1824 * @sg: pointer to the shadow guest address space structure
1825 * @saddr: faulting address in the shadow gmap
1826 * @r3t: parent gmap address of the region 3 table to get shadowed
1827 * @fake: r3t references contiguous guest memory block, not a r3t
1828 *
1829 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1830 * shadow table structure is incomplete, -ENOMEM if out of memory and
1831 * -EFAULT if an address in the parent gmap could not be resolved.
1832 *
1833 * Called with sg->mm->mmap_lock in read.
1834 */
1835int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1836		    int fake)
1837{
1838	unsigned long raddr, origin, offset, len;
1839	unsigned long *table;
1840	phys_addr_t s_r3t;
1841	struct page *page;
1842	int rc;
1843
1844	BUG_ON(!gmap_is_shadow(sg));
1845	/* Allocate a shadow region second table */
1846	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
1847	if (!page)
1848		return -ENOMEM;
1849	page->index = r3t & _REGION_ENTRY_ORIGIN;
1850	if (fake)
1851		page->index |= GMAP_SHADOW_FAKE_TABLE;
1852	s_r3t = page_to_phys(page);
1853	/* Install shadow region second table */
1854	spin_lock(&sg->guest_table_lock);
1855	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1856	if (!table) {
1857		rc = -EAGAIN;		/* Race with unshadow */
1858		goto out_free;
1859	}
1860	if (!(*table & _REGION_ENTRY_INVALID)) {
1861		rc = 0;			/* Already established */
1862		goto out_free;
1863	} else if (*table & _REGION_ENTRY_ORIGIN) {
1864		rc = -EAGAIN;		/* Race with shadow */
1865		goto out_free;
1866	}
1867	crst_table_init(__va(s_r3t), _REGION3_ENTRY_EMPTY);
1868	/* mark as invalid as long as the parent table is not protected */
1869	*table = s_r3t | _REGION_ENTRY_LENGTH |
1870		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1871	if (sg->edat_level >= 1)
1872		*table |= (r3t & _REGION_ENTRY_PROTECT);
1873	list_add(&page->lru, &sg->crst_list);
1874	if (fake) {
1875		/* nothing to protect for fake tables */
1876		*table &= ~_REGION_ENTRY_INVALID;
1877		spin_unlock(&sg->guest_table_lock);
1878		return 0;
1879	}
1880	spin_unlock(&sg->guest_table_lock);
1881	/* Make r3t read-only in parent gmap page table */
1882	raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1883	origin = r3t & _REGION_ENTRY_ORIGIN;
1884	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1885	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1886	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1887	spin_lock(&sg->guest_table_lock);
1888	if (!rc) {
1889		table = gmap_table_walk(sg, saddr, 3);
1890		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r3t)
 
1891			rc = -EAGAIN;		/* Race with unshadow */
1892		else
1893			*table &= ~_REGION_ENTRY_INVALID;
1894	} else {
1895		gmap_unshadow_r3t(sg, raddr);
1896	}
1897	spin_unlock(&sg->guest_table_lock);
1898	return rc;
1899out_free:
1900	spin_unlock(&sg->guest_table_lock);
1901	__free_pages(page, CRST_ALLOC_ORDER);
1902	return rc;
1903}
1904EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1905
1906/**
1907 * gmap_shadow_sgt - create a shadow segment table
1908 * @sg: pointer to the shadow guest address space structure
1909 * @saddr: faulting address in the shadow gmap
1910 * @sgt: parent gmap address of the segment table to get shadowed
1911 * @fake: sgt references contiguous guest memory block, not a sgt
1912 *
1913 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1914 * shadow table structure is incomplete, -ENOMEM if out of memory and
1915 * -EFAULT if an address in the parent gmap could not be resolved.
1916 *
1917 * Called with sg->mm->mmap_lock in read.
1918 */
1919int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1920		    int fake)
1921{
1922	unsigned long raddr, origin, offset, len;
1923	unsigned long *table;
1924	phys_addr_t s_sgt;
1925	struct page *page;
1926	int rc;
1927
1928	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1929	/* Allocate a shadow segment table */
1930	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
1931	if (!page)
1932		return -ENOMEM;
1933	page->index = sgt & _REGION_ENTRY_ORIGIN;
1934	if (fake)
1935		page->index |= GMAP_SHADOW_FAKE_TABLE;
1936	s_sgt = page_to_phys(page);
1937	/* Install shadow region second table */
1938	spin_lock(&sg->guest_table_lock);
1939	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1940	if (!table) {
1941		rc = -EAGAIN;		/* Race with unshadow */
1942		goto out_free;
1943	}
1944	if (!(*table & _REGION_ENTRY_INVALID)) {
1945		rc = 0;			/* Already established */
1946		goto out_free;
1947	} else if (*table & _REGION_ENTRY_ORIGIN) {
1948		rc = -EAGAIN;		/* Race with shadow */
1949		goto out_free;
1950	}
1951	crst_table_init(__va(s_sgt), _SEGMENT_ENTRY_EMPTY);
1952	/* mark as invalid as long as the parent table is not protected */
1953	*table = s_sgt | _REGION_ENTRY_LENGTH |
1954		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1955	if (sg->edat_level >= 1)
1956		*table |= sgt & _REGION_ENTRY_PROTECT;
1957	list_add(&page->lru, &sg->crst_list);
1958	if (fake) {
1959		/* nothing to protect for fake tables */
1960		*table &= ~_REGION_ENTRY_INVALID;
1961		spin_unlock(&sg->guest_table_lock);
1962		return 0;
1963	}
1964	spin_unlock(&sg->guest_table_lock);
1965	/* Make sgt read-only in parent gmap page table */
1966	raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1967	origin = sgt & _REGION_ENTRY_ORIGIN;
1968	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1969	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1970	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1971	spin_lock(&sg->guest_table_lock);
1972	if (!rc) {
1973		table = gmap_table_walk(sg, saddr, 2);
1974		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_sgt)
 
1975			rc = -EAGAIN;		/* Race with unshadow */
1976		else
1977			*table &= ~_REGION_ENTRY_INVALID;
1978	} else {
1979		gmap_unshadow_sgt(sg, raddr);
1980	}
1981	spin_unlock(&sg->guest_table_lock);
1982	return rc;
1983out_free:
1984	spin_unlock(&sg->guest_table_lock);
1985	__free_pages(page, CRST_ALLOC_ORDER);
1986	return rc;
1987}
1988EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1989
1990/**
1991 * gmap_shadow_pgt_lookup - find a shadow page table
1992 * @sg: pointer to the shadow guest address space structure
1993 * @saddr: the address in the shadow aguest address space
1994 * @pgt: parent gmap address of the page table to get shadowed
1995 * @dat_protection: if the pgtable is marked as protected by dat
1996 * @fake: pgt references contiguous guest memory block, not a pgtable
1997 *
1998 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1999 * table was not found.
2000 *
2001 * Called with sg->mm->mmap_lock in read.
2002 */
2003int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
2004			   unsigned long *pgt, int *dat_protection,
2005			   int *fake)
2006{
2007	unsigned long *table;
2008	struct page *page;
2009	int rc;
2010
2011	BUG_ON(!gmap_is_shadow(sg));
2012	spin_lock(&sg->guest_table_lock);
2013	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2014	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
2015		/* Shadow page tables are full pages (pte+pgste) */
2016		page = pfn_to_page(*table >> PAGE_SHIFT);
2017		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
2018		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
2019		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2020		rc = 0;
2021	} else  {
2022		rc = -EAGAIN;
2023	}
2024	spin_unlock(&sg->guest_table_lock);
2025	return rc;
2026
2027}
2028EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2029
2030/**
2031 * gmap_shadow_pgt - instantiate a shadow page table
2032 * @sg: pointer to the shadow guest address space structure
2033 * @saddr: faulting address in the shadow gmap
2034 * @pgt: parent gmap address of the page table to get shadowed
2035 * @fake: pgt references contiguous guest memory block, not a pgtable
2036 *
2037 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2038 * shadow table structure is incomplete, -ENOMEM if out of memory,
2039 * -EFAULT if an address in the parent gmap could not be resolved and
2040 *
2041 * Called with gmap->mm->mmap_lock in read
2042 */
2043int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2044		    int fake)
2045{
2046	unsigned long raddr, origin;
2047	unsigned long *table;
2048	struct page *page;
2049	phys_addr_t s_pgt;
2050	int rc;
2051
2052	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2053	/* Allocate a shadow page table */
2054	page = page_table_alloc_pgste(sg->mm);
2055	if (!page)
2056		return -ENOMEM;
2057	page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
2058	if (fake)
2059		page->index |= GMAP_SHADOW_FAKE_TABLE;
2060	s_pgt = page_to_phys(page);
2061	/* Install shadow page table */
2062	spin_lock(&sg->guest_table_lock);
2063	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2064	if (!table) {
2065		rc = -EAGAIN;		/* Race with unshadow */
2066		goto out_free;
2067	}
2068	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2069		rc = 0;			/* Already established */
2070		goto out_free;
2071	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2072		rc = -EAGAIN;		/* Race with shadow */
2073		goto out_free;
2074	}
2075	/* mark as invalid as long as the parent table is not protected */
2076	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2077		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2078	list_add(&page->lru, &sg->pt_list);
2079	if (fake) {
2080		/* nothing to protect for fake tables */
2081		*table &= ~_SEGMENT_ENTRY_INVALID;
2082		spin_unlock(&sg->guest_table_lock);
2083		return 0;
2084	}
2085	spin_unlock(&sg->guest_table_lock);
2086	/* Make pgt read-only in parent gmap page table (not the pgste) */
2087	raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2088	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2089	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2090	spin_lock(&sg->guest_table_lock);
2091	if (!rc) {
2092		table = gmap_table_walk(sg, saddr, 1);
2093		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) != s_pgt)
 
2094			rc = -EAGAIN;		/* Race with unshadow */
2095		else
2096			*table &= ~_SEGMENT_ENTRY_INVALID;
2097	} else {
2098		gmap_unshadow_pgt(sg, raddr);
2099	}
2100	spin_unlock(&sg->guest_table_lock);
2101	return rc;
2102out_free:
2103	spin_unlock(&sg->guest_table_lock);
2104	page_table_free_pgste(page);
2105	return rc;
2106
2107}
2108EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2109
2110/**
2111 * gmap_shadow_page - create a shadow page mapping
2112 * @sg: pointer to the shadow guest address space structure
2113 * @saddr: faulting address in the shadow gmap
2114 * @pte: pte in parent gmap address space to get shadowed
2115 *
2116 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2117 * shadow table structure is incomplete, -ENOMEM if out of memory and
2118 * -EFAULT if an address in the parent gmap could not be resolved.
2119 *
2120 * Called with sg->mm->mmap_lock in read.
2121 */
2122int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2123{
2124	struct gmap *parent;
2125	struct gmap_rmap *rmap;
2126	unsigned long vmaddr, paddr;
2127	spinlock_t *ptl;
2128	pte_t *sptep, *tptep;
2129	int prot;
2130	int rc;
2131
2132	BUG_ON(!gmap_is_shadow(sg));
2133	parent = sg->parent;
2134	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2135
2136	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
2137	if (!rmap)
2138		return -ENOMEM;
2139	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2140
2141	while (1) {
2142		paddr = pte_val(pte) & PAGE_MASK;
2143		vmaddr = __gmap_translate(parent, paddr);
2144		if (IS_ERR_VALUE(vmaddr)) {
2145			rc = vmaddr;
2146			break;
2147		}
2148		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
2149		if (rc)
2150			break;
2151		rc = -EAGAIN;
2152		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2153		if (sptep) {
2154			spin_lock(&sg->guest_table_lock);
2155			/* Get page table pointer */
2156			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2157			if (!tptep) {
2158				spin_unlock(&sg->guest_table_lock);
2159				gmap_pte_op_end(ptl);
2160				radix_tree_preload_end();
2161				break;
2162			}
2163			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2164			if (rc > 0) {
2165				/* Success and a new mapping */
2166				gmap_insert_rmap(sg, vmaddr, rmap);
2167				rmap = NULL;
2168				rc = 0;
2169			}
2170			gmap_pte_op_end(ptl);
2171			spin_unlock(&sg->guest_table_lock);
2172		}
2173		radix_tree_preload_end();
2174		if (!rc)
2175			break;
2176		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2177		if (rc)
2178			break;
2179	}
2180	kfree(rmap);
2181	return rc;
2182}
2183EXPORT_SYMBOL_GPL(gmap_shadow_page);
2184
2185/*
2186 * gmap_shadow_notify - handle notifications for shadow gmap
2187 *
2188 * Called with sg->parent->shadow_lock.
2189 */
2190static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2191			       unsigned long gaddr)
2192{
2193	struct gmap_rmap *rmap, *rnext, *head;
2194	unsigned long start, end, bits, raddr;
2195
2196	BUG_ON(!gmap_is_shadow(sg));
2197
2198	spin_lock(&sg->guest_table_lock);
2199	if (sg->removed) {
2200		spin_unlock(&sg->guest_table_lock);
2201		return;
2202	}
2203	/* Check for top level table */
2204	start = sg->orig_asce & _ASCE_ORIGIN;
2205	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2206	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2207	    gaddr < end) {
2208		/* The complete shadow table has to go */
2209		gmap_unshadow(sg);
2210		spin_unlock(&sg->guest_table_lock);
2211		list_del(&sg->list);
2212		gmap_put(sg);
2213		return;
2214	}
2215	/* Remove the page table tree from on specific entry */
2216	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2217	gmap_for_each_rmap_safe(rmap, rnext, head) {
2218		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2219		raddr = rmap->raddr ^ bits;
2220		switch (bits) {
2221		case _SHADOW_RMAP_REGION1:
2222			gmap_unshadow_r2t(sg, raddr);
2223			break;
2224		case _SHADOW_RMAP_REGION2:
2225			gmap_unshadow_r3t(sg, raddr);
2226			break;
2227		case _SHADOW_RMAP_REGION3:
2228			gmap_unshadow_sgt(sg, raddr);
2229			break;
2230		case _SHADOW_RMAP_SEGMENT:
2231			gmap_unshadow_pgt(sg, raddr);
2232			break;
2233		case _SHADOW_RMAP_PGTABLE:
2234			gmap_unshadow_page(sg, raddr);
2235			break;
2236		}
2237		kfree(rmap);
2238	}
2239	spin_unlock(&sg->guest_table_lock);
2240}
2241
2242/**
2243 * ptep_notify - call all invalidation callbacks for a specific pte.
2244 * @mm: pointer to the process mm_struct
2245 * @vmaddr: virtual address in the process address space
2246 * @pte: pointer to the page table entry
2247 * @bits: bits from the pgste that caused the notify call
2248 *
2249 * This function is assumed to be called with the page table lock held
2250 * for the pte to notify.
2251 */
2252void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2253		 pte_t *pte, unsigned long bits)
2254{
2255	unsigned long offset, gaddr = 0;
2256	unsigned long *table;
2257	struct gmap *gmap, *sg, *next;
2258
2259	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2260	offset = offset * (PAGE_SIZE / sizeof(pte_t));
2261	rcu_read_lock();
2262	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2263		spin_lock(&gmap->guest_table_lock);
2264		table = radix_tree_lookup(&gmap->host_to_guest,
2265					  vmaddr >> PMD_SHIFT);
2266		if (table)
2267			gaddr = __gmap_segment_gaddr(table) + offset;
2268		spin_unlock(&gmap->guest_table_lock);
2269		if (!table)
2270			continue;
2271
2272		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2273			spin_lock(&gmap->shadow_lock);
2274			list_for_each_entry_safe(sg, next,
2275						 &gmap->children, list)
2276				gmap_shadow_notify(sg, vmaddr, gaddr);
2277			spin_unlock(&gmap->shadow_lock);
2278		}
2279		if (bits & PGSTE_IN_BIT)
2280			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2281	}
2282	rcu_read_unlock();
2283}
2284EXPORT_SYMBOL_GPL(ptep_notify);
2285
2286static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2287			     unsigned long gaddr)
2288{
2289	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
2290	gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2291}
2292
2293/**
2294 * gmap_pmdp_xchg - exchange a gmap pmd with another
2295 * @gmap: pointer to the guest address space structure
2296 * @pmdp: pointer to the pmd entry
2297 * @new: replacement entry
2298 * @gaddr: the affected guest address
2299 *
2300 * This function is assumed to be called with the guest_table_lock
2301 * held.
2302 */
2303static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2304			   unsigned long gaddr)
2305{
2306	gaddr &= HPAGE_MASK;
2307	pmdp_notify_gmap(gmap, pmdp, gaddr);
2308	new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_GMAP_IN));
2309	if (MACHINE_HAS_TLB_GUEST)
2310		__pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2311			    IDTE_GLOBAL);
2312	else if (MACHINE_HAS_IDTE)
2313		__pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2314	else
2315		__pmdp_csp(pmdp);
2316	set_pmd(pmdp, new);
2317}
2318
2319static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2320			    int purge)
2321{
2322	pmd_t *pmdp;
2323	struct gmap *gmap;
2324	unsigned long gaddr;
2325
2326	rcu_read_lock();
2327	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2328		spin_lock(&gmap->guest_table_lock);
2329		pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2330						  vmaddr >> PMD_SHIFT);
2331		if (pmdp) {
2332			gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2333			pmdp_notify_gmap(gmap, pmdp, gaddr);
2334			WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2335						   _SEGMENT_ENTRY_GMAP_UC));
2336			if (purge)
2337				__pmdp_csp(pmdp);
2338			set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
2339		}
2340		spin_unlock(&gmap->guest_table_lock);
2341	}
2342	rcu_read_unlock();
2343}
2344
2345/**
2346 * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2347 *                        flushing
2348 * @mm: pointer to the process mm_struct
2349 * @vmaddr: virtual address in the process address space
2350 */
2351void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2352{
2353	gmap_pmdp_clear(mm, vmaddr, 0);
2354}
2355EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2356
2357/**
2358 * gmap_pmdp_csp - csp all affected guest pmd entries
2359 * @mm: pointer to the process mm_struct
2360 * @vmaddr: virtual address in the process address space
2361 */
2362void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2363{
2364	gmap_pmdp_clear(mm, vmaddr, 1);
2365}
2366EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2367
2368/**
2369 * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2370 * @mm: pointer to the process mm_struct
2371 * @vmaddr: virtual address in the process address space
2372 */
2373void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2374{
2375	unsigned long *entry, gaddr;
2376	struct gmap *gmap;
2377	pmd_t *pmdp;
2378
2379	rcu_read_lock();
2380	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2381		spin_lock(&gmap->guest_table_lock);
2382		entry = radix_tree_delete(&gmap->host_to_guest,
2383					  vmaddr >> PMD_SHIFT);
2384		if (entry) {
2385			pmdp = (pmd_t *)entry;
2386			gaddr = __gmap_segment_gaddr(entry);
2387			pmdp_notify_gmap(gmap, pmdp, gaddr);
2388			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2389					   _SEGMENT_ENTRY_GMAP_UC));
2390			if (MACHINE_HAS_TLB_GUEST)
2391				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2392					    gmap->asce, IDTE_LOCAL);
2393			else if (MACHINE_HAS_IDTE)
2394				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2395			*entry = _SEGMENT_ENTRY_EMPTY;
2396		}
2397		spin_unlock(&gmap->guest_table_lock);
2398	}
2399	rcu_read_unlock();
2400}
2401EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2402
2403/**
2404 * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2405 * @mm: pointer to the process mm_struct
2406 * @vmaddr: virtual address in the process address space
2407 */
2408void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2409{
2410	unsigned long *entry, gaddr;
2411	struct gmap *gmap;
2412	pmd_t *pmdp;
2413
2414	rcu_read_lock();
2415	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2416		spin_lock(&gmap->guest_table_lock);
2417		entry = radix_tree_delete(&gmap->host_to_guest,
2418					  vmaddr >> PMD_SHIFT);
2419		if (entry) {
2420			pmdp = (pmd_t *)entry;
2421			gaddr = __gmap_segment_gaddr(entry);
2422			pmdp_notify_gmap(gmap, pmdp, gaddr);
2423			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2424					   _SEGMENT_ENTRY_GMAP_UC));
2425			if (MACHINE_HAS_TLB_GUEST)
2426				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2427					    gmap->asce, IDTE_GLOBAL);
2428			else if (MACHINE_HAS_IDTE)
2429				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2430			else
2431				__pmdp_csp(pmdp);
2432			*entry = _SEGMENT_ENTRY_EMPTY;
2433		}
2434		spin_unlock(&gmap->guest_table_lock);
2435	}
2436	rcu_read_unlock();
2437}
2438EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2439
2440/**
2441 * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2442 * @gmap: pointer to guest address space
2443 * @pmdp: pointer to the pmd to be tested
2444 * @gaddr: virtual address in the guest address space
2445 *
2446 * This function is assumed to be called with the guest_table_lock
2447 * held.
2448 */
2449static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2450					  unsigned long gaddr)
2451{
2452	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2453		return false;
2454
2455	/* Already protected memory, which did not change is clean */
2456	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2457	    !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2458		return false;
2459
2460	/* Clear UC indication and reset protection */
2461	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_UC)));
2462	gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2463	return true;
2464}
2465
2466/**
2467 * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2468 * @gmap: pointer to guest address space
2469 * @bitmap: dirty bitmap for this pmd
2470 * @gaddr: virtual address in the guest address space
2471 * @vmaddr: virtual address in the host address space
2472 *
2473 * This function is assumed to be called with the guest_table_lock
2474 * held.
2475 */
2476void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2477			     unsigned long gaddr, unsigned long vmaddr)
2478{
2479	int i;
2480	pmd_t *pmdp;
2481	pte_t *ptep;
2482	spinlock_t *ptl;
2483
2484	pmdp = gmap_pmd_op_walk(gmap, gaddr);
2485	if (!pmdp)
2486		return;
2487
2488	if (pmd_large(*pmdp)) {
2489		if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2490			bitmap_fill(bitmap, _PAGE_ENTRIES);
2491	} else {
2492		for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2493			ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2494			if (!ptep)
2495				continue;
2496			if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2497				set_bit(i, bitmap);
2498			spin_unlock(ptl);
2499		}
2500	}
2501	gmap_pmd_op_end(gmap, pmdp);
2502}
2503EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2504
2505#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2506static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
2507				    unsigned long end, struct mm_walk *walk)
2508{
2509	struct vm_area_struct *vma = walk->vma;
2510
2511	split_huge_pmd(vma, pmd, addr);
2512	return 0;
2513}
2514
2515static const struct mm_walk_ops thp_split_walk_ops = {
2516	.pmd_entry	= thp_split_walk_pmd_entry,
2517};
2518
2519static inline void thp_split_mm(struct mm_struct *mm)
2520{
 
2521	struct vm_area_struct *vma;
2522	VMA_ITERATOR(vmi, mm, 0);
2523
2524	for_each_vma(vmi, vma) {
 
 
 
 
2525		vma->vm_flags &= ~VM_HUGEPAGE;
2526		vma->vm_flags |= VM_NOHUGEPAGE;
2527		walk_page_vma(vma, &thp_split_walk_ops, NULL);
2528	}
2529	mm->def_flags |= VM_NOHUGEPAGE;
 
2530}
2531#else
2532static inline void thp_split_mm(struct mm_struct *mm)
2533{
2534}
2535#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2536
2537/*
2538 * Remove all empty zero pages from the mapping for lazy refaulting
2539 * - This must be called after mm->context.has_pgste is set, to avoid
2540 *   future creation of zero pages
2541 * - This must be called after THP was enabled
2542 */
2543static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2544			   unsigned long end, struct mm_walk *walk)
2545{
2546	unsigned long addr;
2547
2548	for (addr = start; addr != end; addr += PAGE_SIZE) {
2549		pte_t *ptep;
2550		spinlock_t *ptl;
2551
2552		ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2553		if (is_zero_pfn(pte_pfn(*ptep)))
2554			ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2555		pte_unmap_unlock(ptep, ptl);
2556	}
2557	return 0;
2558}
2559
2560static const struct mm_walk_ops zap_zero_walk_ops = {
2561	.pmd_entry	= __zap_zero_pages,
2562};
2563
2564/*
2565 * switch on pgstes for its userspace process (for kvm)
2566 */
2567int s390_enable_sie(void)
2568{
2569	struct mm_struct *mm = current->mm;
2570
2571	/* Do we have pgstes? if yes, we are done */
2572	if (mm_has_pgste(mm))
2573		return 0;
2574	/* Fail if the page tables are 2K */
2575	if (!mm_alloc_pgste(mm))
2576		return -EINVAL;
2577	mmap_write_lock(mm);
2578	mm->context.has_pgste = 1;
2579	/* split thp mappings and disable thp for future mappings */
2580	thp_split_mm(mm);
2581	walk_page_range(mm, 0, TASK_SIZE, &zap_zero_walk_ops, NULL);
2582	mmap_write_unlock(mm);
2583	return 0;
2584}
2585EXPORT_SYMBOL_GPL(s390_enable_sie);
2586
2587int gmap_mark_unmergeable(void)
2588{
2589	struct mm_struct *mm = current->mm;
2590	struct vm_area_struct *vma;
2591	int ret;
2592	VMA_ITERATOR(vmi, mm, 0);
2593
2594	for_each_vma(vmi, vma) {
2595		ret = ksm_madvise(vma, vma->vm_start, vma->vm_end,
2596				  MADV_UNMERGEABLE, &vma->vm_flags);
2597		if (ret)
2598			return ret;
2599	}
2600	mm->def_flags &= ~VM_MERGEABLE;
2601	return 0;
2602}
2603EXPORT_SYMBOL_GPL(gmap_mark_unmergeable);
2604
2605/*
2606 * Enable storage key handling from now on and initialize the storage
2607 * keys with the default key.
2608 */
2609static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2610				  unsigned long next, struct mm_walk *walk)
2611{
2612	/* Clear storage key */
2613	ptep_zap_key(walk->mm, addr, pte);
2614	return 0;
2615}
2616
2617/*
2618 * Give a chance to schedule after setting a key to 256 pages.
2619 * We only hold the mm lock, which is a rwsem and the kvm srcu.
2620 * Both can sleep.
2621 */
2622static int __s390_enable_skey_pmd(pmd_t *pmd, unsigned long addr,
2623				  unsigned long next, struct mm_walk *walk)
2624{
2625	cond_resched();
2626	return 0;
2627}
2628
2629static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2630				      unsigned long hmask, unsigned long next,
2631				      struct mm_walk *walk)
2632{
2633	pmd_t *pmd = (pmd_t *)pte;
2634	unsigned long start, end;
2635	struct page *page = pmd_page(*pmd);
2636
2637	/*
2638	 * The write check makes sure we do not set a key on shared
2639	 * memory. This is needed as the walker does not differentiate
2640	 * between actual guest memory and the process executable or
2641	 * shared libraries.
2642	 */
2643	if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2644	    !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2645		return 0;
2646
2647	start = pmd_val(*pmd) & HPAGE_MASK;
2648	end = start + HPAGE_SIZE - 1;
2649	__storage_key_init_range(start, end);
2650	set_bit(PG_arch_1, &page->flags);
2651	cond_resched();
2652	return 0;
2653}
2654
2655static const struct mm_walk_ops enable_skey_walk_ops = {
2656	.hugetlb_entry		= __s390_enable_skey_hugetlb,
2657	.pte_entry		= __s390_enable_skey_pte,
2658	.pmd_entry		= __s390_enable_skey_pmd,
2659};
2660
2661int s390_enable_skey(void)
2662{
2663	struct mm_struct *mm = current->mm;
 
2664	int rc = 0;
2665
2666	mmap_write_lock(mm);
2667	if (mm_uses_skeys(mm))
2668		goto out_up;
2669
2670	mm->context.uses_skeys = 1;
2671	rc = gmap_mark_unmergeable();
2672	if (rc) {
2673		mm->context.uses_skeys = 0;
2674		goto out_up;
 
 
 
2675	}
 
 
2676	walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2677
2678out_up:
2679	mmap_write_unlock(mm);
2680	return rc;
2681}
2682EXPORT_SYMBOL_GPL(s390_enable_skey);
2683
2684/*
2685 * Reset CMMA state, make all pages stable again.
2686 */
2687static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2688			     unsigned long next, struct mm_walk *walk)
2689{
2690	ptep_zap_unused(walk->mm, addr, pte, 1);
2691	return 0;
2692}
2693
2694static const struct mm_walk_ops reset_cmma_walk_ops = {
2695	.pte_entry		= __s390_reset_cmma,
2696};
2697
2698void s390_reset_cmma(struct mm_struct *mm)
2699{
2700	mmap_write_lock(mm);
2701	walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2702	mmap_write_unlock(mm);
2703}
2704EXPORT_SYMBOL_GPL(s390_reset_cmma);
2705
2706#define GATHER_GET_PAGES 32
2707
2708struct reset_walk_state {
2709	unsigned long next;
2710	unsigned long count;
2711	unsigned long pfns[GATHER_GET_PAGES];
2712};
2713
2714static int s390_gather_pages(pte_t *ptep, unsigned long addr,
2715			     unsigned long next, struct mm_walk *walk)
2716{
2717	struct reset_walk_state *p = walk->private;
2718	pte_t pte = READ_ONCE(*ptep);
2719
2720	if (pte_present(pte)) {
2721		/* we have a reference from the mapping, take an extra one */
2722		get_page(phys_to_page(pte_val(pte)));
2723		p->pfns[p->count] = phys_to_pfn(pte_val(pte));
2724		p->next = next;
2725		p->count++;
2726	}
2727	return p->count >= GATHER_GET_PAGES;
2728}
2729
2730static const struct mm_walk_ops gather_pages_ops = {
2731	.pte_entry = s390_gather_pages,
2732};
2733
2734/*
2735 * Call the Destroy secure page UVC on each page in the given array of PFNs.
2736 * Each page needs to have an extra reference, which will be released here.
2737 */
2738void s390_uv_destroy_pfns(unsigned long count, unsigned long *pfns)
2739{
2740	unsigned long i;
2741
2742	for (i = 0; i < count; i++) {
2743		/* we always have an extra reference */
2744		uv_destroy_owned_page(pfn_to_phys(pfns[i]));
2745		/* get rid of the extra reference */
2746		put_page(pfn_to_page(pfns[i]));
2747		cond_resched();
2748	}
2749}
2750EXPORT_SYMBOL_GPL(s390_uv_destroy_pfns);
2751
2752/**
2753 * __s390_uv_destroy_range - Call the destroy secure page UVC on each page
2754 * in the given range of the given address space.
2755 * @mm: the mm to operate on
2756 * @start: the start of the range
2757 * @end: the end of the range
2758 * @interruptible: if not 0, stop when a fatal signal is received
2759 *
2760 * Walk the given range of the given address space and call the destroy
2761 * secure page UVC on each page. Optionally exit early if a fatal signal is
2762 * pending.
2763 *
2764 * Return: 0 on success, -EINTR if the function stopped before completing
2765 */
2766int __s390_uv_destroy_range(struct mm_struct *mm, unsigned long start,
2767			    unsigned long end, bool interruptible)
2768{
2769	struct reset_walk_state state = { .next = start };
2770	int r = 1;
2771
2772	while (r > 0) {
2773		state.count = 0;
2774		mmap_read_lock(mm);
2775		r = walk_page_range(mm, state.next, end, &gather_pages_ops, &state);
2776		mmap_read_unlock(mm);
2777		cond_resched();
2778		s390_uv_destroy_pfns(state.count, state.pfns);
2779		if (interruptible && fatal_signal_pending(current))
2780			return -EINTR;
2781	}
2782	return 0;
2783}
2784EXPORT_SYMBOL_GPL(__s390_uv_destroy_range);
2785
2786/**
2787 * s390_unlist_old_asce - Remove the topmost level of page tables from the
2788 * list of page tables of the gmap.
2789 * @gmap: the gmap whose table is to be removed
2790 *
2791 * On s390x, KVM keeps a list of all pages containing the page tables of the
2792 * gmap (the CRST list). This list is used at tear down time to free all
2793 * pages that are now not needed anymore.
2794 *
2795 * This function removes the topmost page of the tree (the one pointed to by
2796 * the ASCE) from the CRST list.
2797 *
2798 * This means that it will not be freed when the VM is torn down, and needs
2799 * to be handled separately by the caller, unless a leak is actually
2800 * intended. Notice that this function will only remove the page from the
2801 * list, the page will still be used as a top level page table (and ASCE).
2802 */
2803void s390_unlist_old_asce(struct gmap *gmap)
2804{
2805	struct page *old;
2806
2807	old = virt_to_page(gmap->table);
2808	spin_lock(&gmap->guest_table_lock);
2809	list_del(&old->lru);
2810	/*
2811	 * Sometimes the topmost page might need to be "removed" multiple
2812	 * times, for example if the VM is rebooted into secure mode several
2813	 * times concurrently, or if s390_replace_asce fails after calling
2814	 * s390_remove_old_asce and is attempted again later. In that case
2815	 * the old asce has been removed from the list, and therefore it
2816	 * will not be freed when the VM terminates, but the ASCE is still
2817	 * in use and still pointed to.
2818	 * A subsequent call to replace_asce will follow the pointer and try
2819	 * to remove the same page from the list again.
2820	 * Therefore it's necessary that the page of the ASCE has valid
2821	 * pointers, so list_del can work (and do nothing) without
2822	 * dereferencing stale or invalid pointers.
2823	 */
2824	INIT_LIST_HEAD(&old->lru);
2825	spin_unlock(&gmap->guest_table_lock);
2826}
2827EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
2828
2829/**
2830 * s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
2831 * @gmap: the gmap whose ASCE needs to be replaced
2832 *
2833 * If the allocation of the new top level page table fails, the ASCE is not
2834 * replaced.
2835 * In any case, the old ASCE is always removed from the gmap CRST list.
2836 * Therefore the caller has to make sure to save a pointer to it
2837 * beforehand, unless a leak is actually intended.
2838 */
2839int s390_replace_asce(struct gmap *gmap)
2840{
2841	unsigned long asce;
2842	struct page *page;
2843	void *table;
2844
2845	s390_unlist_old_asce(gmap);
2846
2847	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
2848	if (!page)
2849		return -ENOMEM;
2850	table = page_to_virt(page);
2851	memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
2852
2853	/*
2854	 * The caller has to deal with the old ASCE, but here we make sure
2855	 * the new one is properly added to the CRST list, so that
2856	 * it will be freed when the VM is torn down.
2857	 */
2858	spin_lock(&gmap->guest_table_lock);
2859	list_add(&page->lru, &gmap->crst_list);
2860	spin_unlock(&gmap->guest_table_lock);
2861
2862	/* Set new table origin while preserving existing ASCE control bits */
2863	asce = (gmap->asce & ~_ASCE_ORIGIN) | __pa(table);
2864	WRITE_ONCE(gmap->asce, asce);
2865	WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
2866	WRITE_ONCE(gmap->table, table);
2867
2868	return 0;
2869}
2870EXPORT_SYMBOL_GPL(s390_replace_asce);
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  KVM guest address space mapping code
   4 *
   5 *    Copyright IBM Corp. 2007, 2016, 2018
   6 *    Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
   7 *		 David Hildenbrand <david@redhat.com>
   8 *		 Janosch Frank <frankja@linux.vnet.ibm.com>
   9 */
  10
  11#include <linux/kernel.h>
  12#include <linux/pagewalk.h>
  13#include <linux/swap.h>
  14#include <linux/smp.h>
  15#include <linux/spinlock.h>
  16#include <linux/slab.h>
  17#include <linux/swapops.h>
  18#include <linux/ksm.h>
  19#include <linux/mman.h>
 
  20
  21#include <asm/pgtable.h>
  22#include <asm/pgalloc.h>
  23#include <asm/gmap.h>
  24#include <asm/tlb.h>
  25
  26#define GMAP_SHADOW_FAKE_TABLE 1ULL
  27
  28/**
  29 * gmap_alloc - allocate and initialize a guest address space
  30 * @mm: pointer to the parent mm_struct
  31 * @limit: maximum address of the gmap address space
  32 *
  33 * Returns a guest address space structure.
  34 */
  35static struct gmap *gmap_alloc(unsigned long limit)
  36{
  37	struct gmap *gmap;
  38	struct page *page;
  39	unsigned long *table;
  40	unsigned long etype, atype;
  41
  42	if (limit < _REGION3_SIZE) {
  43		limit = _REGION3_SIZE - 1;
  44		atype = _ASCE_TYPE_SEGMENT;
  45		etype = _SEGMENT_ENTRY_EMPTY;
  46	} else if (limit < _REGION2_SIZE) {
  47		limit = _REGION2_SIZE - 1;
  48		atype = _ASCE_TYPE_REGION3;
  49		etype = _REGION3_ENTRY_EMPTY;
  50	} else if (limit < _REGION1_SIZE) {
  51		limit = _REGION1_SIZE - 1;
  52		atype = _ASCE_TYPE_REGION2;
  53		etype = _REGION2_ENTRY_EMPTY;
  54	} else {
  55		limit = -1UL;
  56		atype = _ASCE_TYPE_REGION1;
  57		etype = _REGION1_ENTRY_EMPTY;
  58	}
  59	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
  60	if (!gmap)
  61		goto out;
  62	INIT_LIST_HEAD(&gmap->crst_list);
  63	INIT_LIST_HEAD(&gmap->children);
  64	INIT_LIST_HEAD(&gmap->pt_list);
  65	INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL);
  66	INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC);
  67	INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC);
  68	spin_lock_init(&gmap->guest_table_lock);
  69	spin_lock_init(&gmap->shadow_lock);
  70	refcount_set(&gmap->ref_count, 1);
  71	page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
  72	if (!page)
  73		goto out_free;
  74	page->index = 0;
  75	list_add(&page->lru, &gmap->crst_list);
  76	table = (unsigned long *) page_to_phys(page);
  77	crst_table_init(table, etype);
  78	gmap->table = table;
  79	gmap->asce = atype | _ASCE_TABLE_LENGTH |
  80		_ASCE_USER_BITS | __pa(table);
  81	gmap->asce_end = limit;
  82	return gmap;
  83
  84out_free:
  85	kfree(gmap);
  86out:
  87	return NULL;
  88}
  89
  90/**
  91 * gmap_create - create a guest address space
  92 * @mm: pointer to the parent mm_struct
  93 * @limit: maximum size of the gmap address space
  94 *
  95 * Returns a guest address space structure.
  96 */
  97struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
  98{
  99	struct gmap *gmap;
 100	unsigned long gmap_asce;
 101
 102	gmap = gmap_alloc(limit);
 103	if (!gmap)
 104		return NULL;
 105	gmap->mm = mm;
 106	spin_lock(&mm->context.lock);
 107	list_add_rcu(&gmap->list, &mm->context.gmap_list);
 108	if (list_is_singular(&mm->context.gmap_list))
 109		gmap_asce = gmap->asce;
 110	else
 111		gmap_asce = -1UL;
 112	WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
 113	spin_unlock(&mm->context.lock);
 114	return gmap;
 115}
 116EXPORT_SYMBOL_GPL(gmap_create);
 117
 118static void gmap_flush_tlb(struct gmap *gmap)
 119{
 120	if (MACHINE_HAS_IDTE)
 121		__tlb_flush_idte(gmap->asce);
 122	else
 123		__tlb_flush_global();
 124}
 125
 126static void gmap_radix_tree_free(struct radix_tree_root *root)
 127{
 128	struct radix_tree_iter iter;
 129	unsigned long indices[16];
 130	unsigned long index;
 131	void __rcu **slot;
 132	int i, nr;
 133
 134	/* A radix tree is freed by deleting all of its entries */
 135	index = 0;
 136	do {
 137		nr = 0;
 138		radix_tree_for_each_slot(slot, root, &iter, index) {
 139			indices[nr] = iter.index;
 140			if (++nr == 16)
 141				break;
 142		}
 143		for (i = 0; i < nr; i++) {
 144			index = indices[i];
 145			radix_tree_delete(root, index);
 146		}
 147	} while (nr > 0);
 148}
 149
 150static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
 151{
 152	struct gmap_rmap *rmap, *rnext, *head;
 153	struct radix_tree_iter iter;
 154	unsigned long indices[16];
 155	unsigned long index;
 156	void __rcu **slot;
 157	int i, nr;
 158
 159	/* A radix tree is freed by deleting all of its entries */
 160	index = 0;
 161	do {
 162		nr = 0;
 163		radix_tree_for_each_slot(slot, root, &iter, index) {
 164			indices[nr] = iter.index;
 165			if (++nr == 16)
 166				break;
 167		}
 168		for (i = 0; i < nr; i++) {
 169			index = indices[i];
 170			head = radix_tree_delete(root, index);
 171			gmap_for_each_rmap_safe(rmap, rnext, head)
 172				kfree(rmap);
 173		}
 174	} while (nr > 0);
 175}
 176
 177/**
 178 * gmap_free - free a guest address space
 179 * @gmap: pointer to the guest address space structure
 180 *
 181 * No locks required. There are no references to this gmap anymore.
 182 */
 183static void gmap_free(struct gmap *gmap)
 184{
 185	struct page *page, *next;
 186
 187	/* Flush tlb of all gmaps (if not already done for shadows) */
 188	if (!(gmap_is_shadow(gmap) && gmap->removed))
 189		gmap_flush_tlb(gmap);
 190	/* Free all segment & region tables. */
 191	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
 192		__free_pages(page, CRST_ALLOC_ORDER);
 193	gmap_radix_tree_free(&gmap->guest_to_host);
 194	gmap_radix_tree_free(&gmap->host_to_guest);
 195
 196	/* Free additional data for a shadow gmap */
 197	if (gmap_is_shadow(gmap)) {
 198		/* Free all page tables. */
 199		list_for_each_entry_safe(page, next, &gmap->pt_list, lru)
 200			page_table_free_pgste(page);
 201		gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
 202		/* Release reference to the parent */
 203		gmap_put(gmap->parent);
 204	}
 205
 206	kfree(gmap);
 207}
 208
 209/**
 210 * gmap_get - increase reference counter for guest address space
 211 * @gmap: pointer to the guest address space structure
 212 *
 213 * Returns the gmap pointer
 214 */
 215struct gmap *gmap_get(struct gmap *gmap)
 216{
 217	refcount_inc(&gmap->ref_count);
 218	return gmap;
 219}
 220EXPORT_SYMBOL_GPL(gmap_get);
 221
 222/**
 223 * gmap_put - decrease reference counter for guest address space
 224 * @gmap: pointer to the guest address space structure
 225 *
 226 * If the reference counter reaches zero the guest address space is freed.
 227 */
 228void gmap_put(struct gmap *gmap)
 229{
 230	if (refcount_dec_and_test(&gmap->ref_count))
 231		gmap_free(gmap);
 232}
 233EXPORT_SYMBOL_GPL(gmap_put);
 234
 235/**
 236 * gmap_remove - remove a guest address space but do not free it yet
 237 * @gmap: pointer to the guest address space structure
 238 */
 239void gmap_remove(struct gmap *gmap)
 240{
 241	struct gmap *sg, *next;
 242	unsigned long gmap_asce;
 243
 244	/* Remove all shadow gmaps linked to this gmap */
 245	if (!list_empty(&gmap->children)) {
 246		spin_lock(&gmap->shadow_lock);
 247		list_for_each_entry_safe(sg, next, &gmap->children, list) {
 248			list_del(&sg->list);
 249			gmap_put(sg);
 250		}
 251		spin_unlock(&gmap->shadow_lock);
 252	}
 253	/* Remove gmap from the pre-mm list */
 254	spin_lock(&gmap->mm->context.lock);
 255	list_del_rcu(&gmap->list);
 256	if (list_empty(&gmap->mm->context.gmap_list))
 257		gmap_asce = 0;
 258	else if (list_is_singular(&gmap->mm->context.gmap_list))
 259		gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
 260					     struct gmap, list)->asce;
 261	else
 262		gmap_asce = -1UL;
 263	WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
 264	spin_unlock(&gmap->mm->context.lock);
 265	synchronize_rcu();
 266	/* Put reference */
 267	gmap_put(gmap);
 268}
 269EXPORT_SYMBOL_GPL(gmap_remove);
 270
 271/**
 272 * gmap_enable - switch primary space to the guest address space
 273 * @gmap: pointer to the guest address space structure
 274 */
 275void gmap_enable(struct gmap *gmap)
 276{
 277	S390_lowcore.gmap = (unsigned long) gmap;
 278}
 279EXPORT_SYMBOL_GPL(gmap_enable);
 280
 281/**
 282 * gmap_disable - switch back to the standard primary address space
 283 * @gmap: pointer to the guest address space structure
 284 */
 285void gmap_disable(struct gmap *gmap)
 286{
 287	S390_lowcore.gmap = 0UL;
 288}
 289EXPORT_SYMBOL_GPL(gmap_disable);
 290
 291/**
 292 * gmap_get_enabled - get a pointer to the currently enabled gmap
 293 *
 294 * Returns a pointer to the currently enabled gmap. 0 if none is enabled.
 295 */
 296struct gmap *gmap_get_enabled(void)
 297{
 298	return (struct gmap *) S390_lowcore.gmap;
 299}
 300EXPORT_SYMBOL_GPL(gmap_get_enabled);
 301
 302/*
 303 * gmap_alloc_table is assumed to be called with mmap_sem held
 304 */
 305static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
 306			    unsigned long init, unsigned long gaddr)
 307{
 308	struct page *page;
 309	unsigned long *new;
 310
 311	/* since we dont free the gmap table until gmap_free we can unlock */
 312	page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
 313	if (!page)
 314		return -ENOMEM;
 315	new = (unsigned long *) page_to_phys(page);
 316	crst_table_init(new, init);
 317	spin_lock(&gmap->guest_table_lock);
 318	if (*table & _REGION_ENTRY_INVALID) {
 319		list_add(&page->lru, &gmap->crst_list);
 320		*table = (unsigned long) new | _REGION_ENTRY_LENGTH |
 321			(*table & _REGION_ENTRY_TYPE_MASK);
 322		page->index = gaddr;
 323		page = NULL;
 324	}
 325	spin_unlock(&gmap->guest_table_lock);
 326	if (page)
 327		__free_pages(page, CRST_ALLOC_ORDER);
 328	return 0;
 329}
 330
 331/**
 332 * __gmap_segment_gaddr - find virtual address from segment pointer
 333 * @entry: pointer to a segment table entry in the guest address space
 334 *
 335 * Returns the virtual address in the guest address space for the segment
 336 */
 337static unsigned long __gmap_segment_gaddr(unsigned long *entry)
 338{
 339	struct page *page;
 340	unsigned long offset, mask;
 341
 342	offset = (unsigned long) entry / sizeof(unsigned long);
 343	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
 344	mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
 345	page = virt_to_page((void *)((unsigned long) entry & mask));
 346	return page->index + offset;
 347}
 348
 349/**
 350 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
 351 * @gmap: pointer to the guest address space structure
 352 * @vmaddr: address in the host process address space
 353 *
 354 * Returns 1 if a TLB flush is required
 355 */
 356static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
 357{
 358	unsigned long *entry;
 359	int flush = 0;
 360
 361	BUG_ON(gmap_is_shadow(gmap));
 362	spin_lock(&gmap->guest_table_lock);
 363	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
 364	if (entry) {
 365		flush = (*entry != _SEGMENT_ENTRY_EMPTY);
 366		*entry = _SEGMENT_ENTRY_EMPTY;
 367	}
 368	spin_unlock(&gmap->guest_table_lock);
 369	return flush;
 370}
 371
 372/**
 373 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
 374 * @gmap: pointer to the guest address space structure
 375 * @gaddr: address in the guest address space
 376 *
 377 * Returns 1 if a TLB flush is required
 378 */
 379static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
 380{
 381	unsigned long vmaddr;
 382
 383	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
 384						   gaddr >> PMD_SHIFT);
 385	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
 386}
 387
 388/**
 389 * gmap_unmap_segment - unmap segment from the guest address space
 390 * @gmap: pointer to the guest address space structure
 391 * @to: address in the guest address space
 392 * @len: length of the memory area to unmap
 393 *
 394 * Returns 0 if the unmap succeeded, -EINVAL if not.
 395 */
 396int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
 397{
 398	unsigned long off;
 399	int flush;
 400
 401	BUG_ON(gmap_is_shadow(gmap));
 402	if ((to | len) & (PMD_SIZE - 1))
 403		return -EINVAL;
 404	if (len == 0 || to + len < to)
 405		return -EINVAL;
 406
 407	flush = 0;
 408	down_write(&gmap->mm->mmap_sem);
 409	for (off = 0; off < len; off += PMD_SIZE)
 410		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
 411	up_write(&gmap->mm->mmap_sem);
 412	if (flush)
 413		gmap_flush_tlb(gmap);
 414	return 0;
 415}
 416EXPORT_SYMBOL_GPL(gmap_unmap_segment);
 417
 418/**
 419 * gmap_map_segment - map a segment to the guest address space
 420 * @gmap: pointer to the guest address space structure
 421 * @from: source address in the parent address space
 422 * @to: target address in the guest address space
 423 * @len: length of the memory area to map
 424 *
 425 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
 426 */
 427int gmap_map_segment(struct gmap *gmap, unsigned long from,
 428		     unsigned long to, unsigned long len)
 429{
 430	unsigned long off;
 431	int flush;
 432
 433	BUG_ON(gmap_is_shadow(gmap));
 434	if ((from | to | len) & (PMD_SIZE - 1))
 435		return -EINVAL;
 436	if (len == 0 || from + len < from || to + len < to ||
 437	    from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
 438		return -EINVAL;
 439
 440	flush = 0;
 441	down_write(&gmap->mm->mmap_sem);
 442	for (off = 0; off < len; off += PMD_SIZE) {
 443		/* Remove old translation */
 444		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
 445		/* Store new translation */
 446		if (radix_tree_insert(&gmap->guest_to_host,
 447				      (to + off) >> PMD_SHIFT,
 448				      (void *) from + off))
 449			break;
 450	}
 451	up_write(&gmap->mm->mmap_sem);
 452	if (flush)
 453		gmap_flush_tlb(gmap);
 454	if (off >= len)
 455		return 0;
 456	gmap_unmap_segment(gmap, to, len);
 457	return -ENOMEM;
 458}
 459EXPORT_SYMBOL_GPL(gmap_map_segment);
 460
 461/**
 462 * __gmap_translate - translate a guest address to a user space address
 463 * @gmap: pointer to guest mapping meta data structure
 464 * @gaddr: guest address
 465 *
 466 * Returns user space address which corresponds to the guest address or
 467 * -EFAULT if no such mapping exists.
 468 * This function does not establish potentially missing page table entries.
 469 * The mmap_sem of the mm that belongs to the address space must be held
 470 * when this function gets called.
 471 *
 472 * Note: Can also be called for shadow gmaps.
 473 */
 474unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
 475{
 476	unsigned long vmaddr;
 477
 478	vmaddr = (unsigned long)
 479		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
 480	/* Note: guest_to_host is empty for a shadow gmap */
 481	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
 482}
 483EXPORT_SYMBOL_GPL(__gmap_translate);
 484
 485/**
 486 * gmap_translate - translate a guest address to a user space address
 487 * @gmap: pointer to guest mapping meta data structure
 488 * @gaddr: guest address
 489 *
 490 * Returns user space address which corresponds to the guest address or
 491 * -EFAULT if no such mapping exists.
 492 * This function does not establish potentially missing page table entries.
 493 */
 494unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
 495{
 496	unsigned long rc;
 497
 498	down_read(&gmap->mm->mmap_sem);
 499	rc = __gmap_translate(gmap, gaddr);
 500	up_read(&gmap->mm->mmap_sem);
 501	return rc;
 502}
 503EXPORT_SYMBOL_GPL(gmap_translate);
 504
 505/**
 506 * gmap_unlink - disconnect a page table from the gmap shadow tables
 507 * @gmap: pointer to guest mapping meta data structure
 508 * @table: pointer to the host page table
 509 * @vmaddr: vm address associated with the host page table
 510 */
 511void gmap_unlink(struct mm_struct *mm, unsigned long *table,
 512		 unsigned long vmaddr)
 513{
 514	struct gmap *gmap;
 515	int flush;
 516
 517	rcu_read_lock();
 518	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
 519		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
 520		if (flush)
 521			gmap_flush_tlb(gmap);
 522	}
 523	rcu_read_unlock();
 524}
 525
 526static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
 527			   unsigned long gaddr);
 528
 529/**
 530 * gmap_link - set up shadow page tables to connect a host to a guest address
 531 * @gmap: pointer to guest mapping meta data structure
 532 * @gaddr: guest address
 533 * @vmaddr: vm address
 534 *
 535 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
 536 * if the vm address is already mapped to a different guest segment.
 537 * The mmap_sem of the mm that belongs to the address space must be held
 538 * when this function gets called.
 539 */
 540int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
 541{
 542	struct mm_struct *mm;
 543	unsigned long *table;
 544	spinlock_t *ptl;
 545	pgd_t *pgd;
 546	p4d_t *p4d;
 547	pud_t *pud;
 548	pmd_t *pmd;
 549	u64 unprot;
 550	int rc;
 551
 552	BUG_ON(gmap_is_shadow(gmap));
 553	/* Create higher level tables in the gmap page table */
 554	table = gmap->table;
 555	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
 556		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
 557		if ((*table & _REGION_ENTRY_INVALID) &&
 558		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
 559				     gaddr & _REGION1_MASK))
 560			return -ENOMEM;
 561		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 562	}
 563	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
 564		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
 565		if ((*table & _REGION_ENTRY_INVALID) &&
 566		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
 567				     gaddr & _REGION2_MASK))
 568			return -ENOMEM;
 569		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 570	}
 571	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
 572		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
 573		if ((*table & _REGION_ENTRY_INVALID) &&
 574		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
 575				     gaddr & _REGION3_MASK))
 576			return -ENOMEM;
 577		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 578	}
 579	table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 580	/* Walk the parent mm page table */
 581	mm = gmap->mm;
 582	pgd = pgd_offset(mm, vmaddr);
 583	VM_BUG_ON(pgd_none(*pgd));
 584	p4d = p4d_offset(pgd, vmaddr);
 585	VM_BUG_ON(p4d_none(*p4d));
 586	pud = pud_offset(p4d, vmaddr);
 587	VM_BUG_ON(pud_none(*pud));
 588	/* large puds cannot yet be handled */
 589	if (pud_large(*pud))
 590		return -EFAULT;
 591	pmd = pmd_offset(pud, vmaddr);
 592	VM_BUG_ON(pmd_none(*pmd));
 593	/* Are we allowed to use huge pages? */
 594	if (pmd_large(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
 595		return -EFAULT;
 596	/* Link gmap segment table entry location to page table. */
 597	rc = radix_tree_preload(GFP_KERNEL);
 598	if (rc)
 599		return rc;
 600	ptl = pmd_lock(mm, pmd);
 601	spin_lock(&gmap->guest_table_lock);
 602	if (*table == _SEGMENT_ENTRY_EMPTY) {
 603		rc = radix_tree_insert(&gmap->host_to_guest,
 604				       vmaddr >> PMD_SHIFT, table);
 605		if (!rc) {
 606			if (pmd_large(*pmd)) {
 607				*table = (pmd_val(*pmd) &
 608					  _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
 609					| _SEGMENT_ENTRY_GMAP_UC;
 610			} else
 611				*table = pmd_val(*pmd) &
 612					_SEGMENT_ENTRY_HARDWARE_BITS;
 613		}
 614	} else if (*table & _SEGMENT_ENTRY_PROTECT &&
 615		   !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
 616		unprot = (u64)*table;
 617		unprot &= ~_SEGMENT_ENTRY_PROTECT;
 618		unprot |= _SEGMENT_ENTRY_GMAP_UC;
 619		gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
 620	}
 621	spin_unlock(&gmap->guest_table_lock);
 622	spin_unlock(ptl);
 623	radix_tree_preload_end();
 624	return rc;
 625}
 626
 627/**
 628 * gmap_fault - resolve a fault on a guest address
 629 * @gmap: pointer to guest mapping meta data structure
 630 * @gaddr: guest address
 631 * @fault_flags: flags to pass down to handle_mm_fault()
 632 *
 633 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
 634 * if the vm address is already mapped to a different guest segment.
 635 */
 636int gmap_fault(struct gmap *gmap, unsigned long gaddr,
 637	       unsigned int fault_flags)
 638{
 639	unsigned long vmaddr;
 640	int rc;
 641	bool unlocked;
 642
 643	down_read(&gmap->mm->mmap_sem);
 644
 645retry:
 646	unlocked = false;
 647	vmaddr = __gmap_translate(gmap, gaddr);
 648	if (IS_ERR_VALUE(vmaddr)) {
 649		rc = vmaddr;
 650		goto out_up;
 651	}
 652	if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags,
 653			     &unlocked)) {
 654		rc = -EFAULT;
 655		goto out_up;
 656	}
 657	/*
 658	 * In the case that fixup_user_fault unlocked the mmap_sem during
 659	 * faultin redo __gmap_translate to not race with a map/unmap_segment.
 660	 */
 661	if (unlocked)
 662		goto retry;
 663
 664	rc = __gmap_link(gmap, gaddr, vmaddr);
 665out_up:
 666	up_read(&gmap->mm->mmap_sem);
 667	return rc;
 668}
 669EXPORT_SYMBOL_GPL(gmap_fault);
 670
 671/*
 672 * this function is assumed to be called with mmap_sem held
 673 */
 674void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
 675{
 
 676	unsigned long vmaddr;
 677	spinlock_t *ptl;
 678	pte_t *ptep;
 679
 680	/* Find the vm address for the guest address */
 681	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
 682						   gaddr >> PMD_SHIFT);
 683	if (vmaddr) {
 684		vmaddr |= gaddr & ~PMD_MASK;
 
 
 
 
 
 685		/* Get pointer to the page table entry */
 686		ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
 687		if (likely(ptep))
 688			ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
 689		pte_unmap_unlock(ptep, ptl);
 
 690	}
 691}
 692EXPORT_SYMBOL_GPL(__gmap_zap);
 693
 694void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
 695{
 696	unsigned long gaddr, vmaddr, size;
 697	struct vm_area_struct *vma;
 698
 699	down_read(&gmap->mm->mmap_sem);
 700	for (gaddr = from; gaddr < to;
 701	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
 702		/* Find the vm address for the guest address */
 703		vmaddr = (unsigned long)
 704			radix_tree_lookup(&gmap->guest_to_host,
 705					  gaddr >> PMD_SHIFT);
 706		if (!vmaddr)
 707			continue;
 708		vmaddr |= gaddr & ~PMD_MASK;
 709		/* Find vma in the parent mm */
 710		vma = find_vma(gmap->mm, vmaddr);
 711		if (!vma)
 712			continue;
 713		/*
 714		 * We do not discard pages that are backed by
 715		 * hugetlbfs, so we don't have to refault them.
 716		 */
 717		if (is_vm_hugetlb_page(vma))
 718			continue;
 719		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
 720		zap_page_range(vma, vmaddr, size);
 721	}
 722	up_read(&gmap->mm->mmap_sem);
 723}
 724EXPORT_SYMBOL_GPL(gmap_discard);
 725
 726static LIST_HEAD(gmap_notifier_list);
 727static DEFINE_SPINLOCK(gmap_notifier_lock);
 728
 729/**
 730 * gmap_register_pte_notifier - register a pte invalidation callback
 731 * @nb: pointer to the gmap notifier block
 732 */
 733void gmap_register_pte_notifier(struct gmap_notifier *nb)
 734{
 735	spin_lock(&gmap_notifier_lock);
 736	list_add_rcu(&nb->list, &gmap_notifier_list);
 737	spin_unlock(&gmap_notifier_lock);
 738}
 739EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
 740
 741/**
 742 * gmap_unregister_pte_notifier - remove a pte invalidation callback
 743 * @nb: pointer to the gmap notifier block
 744 */
 745void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
 746{
 747	spin_lock(&gmap_notifier_lock);
 748	list_del_rcu(&nb->list);
 749	spin_unlock(&gmap_notifier_lock);
 750	synchronize_rcu();
 751}
 752EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
 753
 754/**
 755 * gmap_call_notifier - call all registered invalidation callbacks
 756 * @gmap: pointer to guest mapping meta data structure
 757 * @start: start virtual address in the guest address space
 758 * @end: end virtual address in the guest address space
 759 */
 760static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
 761			       unsigned long end)
 762{
 763	struct gmap_notifier *nb;
 764
 765	list_for_each_entry(nb, &gmap_notifier_list, list)
 766		nb->notifier_call(gmap, start, end);
 767}
 768
 769/**
 770 * gmap_table_walk - walk the gmap page tables
 771 * @gmap: pointer to guest mapping meta data structure
 772 * @gaddr: virtual address in the guest address space
 773 * @level: page table level to stop at
 774 *
 775 * Returns a table entry pointer for the given guest address and @level
 776 * @level=0 : returns a pointer to a page table table entry (or NULL)
 777 * @level=1 : returns a pointer to a segment table entry (or NULL)
 778 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
 779 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
 780 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
 781 *
 782 * Returns NULL if the gmap page tables could not be walked to the
 783 * requested level.
 784 *
 785 * Note: Can also be called for shadow gmaps.
 786 */
 787static inline unsigned long *gmap_table_walk(struct gmap *gmap,
 788					     unsigned long gaddr, int level)
 789{
 790	unsigned long *table;
 
 791
 792	if ((gmap->asce & _ASCE_TYPE_MASK) + 4 < (level * 4))
 793		return NULL;
 794	if (gmap_is_shadow(gmap) && gmap->removed)
 
 795		return NULL;
 796	if (gaddr & (-1UL << (31 + ((gmap->asce & _ASCE_TYPE_MASK) >> 2)*11)))
 
 
 797		return NULL;
 798	table = gmap->table;
 799	switch (gmap->asce & _ASCE_TYPE_MASK) {
 800	case _ASCE_TYPE_REGION1:
 801		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
 802		if (level == 4)
 803			break;
 804		if (*table & _REGION_ENTRY_INVALID)
 805			return NULL;
 806		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 807		/* Fallthrough */
 808	case _ASCE_TYPE_REGION2:
 809		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
 810		if (level == 3)
 811			break;
 812		if (*table & _REGION_ENTRY_INVALID)
 813			return NULL;
 814		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 815		/* Fallthrough */
 816	case _ASCE_TYPE_REGION3:
 817		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
 818		if (level == 2)
 819			break;
 820		if (*table & _REGION_ENTRY_INVALID)
 821			return NULL;
 822		table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
 823		/* Fallthrough */
 824	case _ASCE_TYPE_SEGMENT:
 825		table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 826		if (level == 1)
 827			break;
 828		if (*table & _REGION_ENTRY_INVALID)
 829			return NULL;
 830		table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN);
 831		table += (gaddr & _PAGE_INDEX) >> _PAGE_SHIFT;
 832	}
 833	return table;
 834}
 835
 836/**
 837 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
 838 *		      and return the pte pointer
 839 * @gmap: pointer to guest mapping meta data structure
 840 * @gaddr: virtual address in the guest address space
 841 * @ptl: pointer to the spinlock pointer
 842 *
 843 * Returns a pointer to the locked pte for a guest address, or NULL
 844 */
 845static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
 846			       spinlock_t **ptl)
 847{
 848	unsigned long *table;
 849
 850	BUG_ON(gmap_is_shadow(gmap));
 851	/* Walk the gmap page table, lock and get pte pointer */
 852	table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
 853	if (!table || *table & _SEGMENT_ENTRY_INVALID)
 854		return NULL;
 855	return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
 856}
 857
 858/**
 859 * gmap_pte_op_fixup - force a page in and connect the gmap page table
 860 * @gmap: pointer to guest mapping meta data structure
 861 * @gaddr: virtual address in the guest address space
 862 * @vmaddr: address in the host process address space
 863 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
 864 *
 865 * Returns 0 if the caller can retry __gmap_translate (might fail again),
 866 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
 867 * up or connecting the gmap page table.
 868 */
 869static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
 870			     unsigned long vmaddr, int prot)
 871{
 872	struct mm_struct *mm = gmap->mm;
 873	unsigned int fault_flags;
 874	bool unlocked = false;
 875
 876	BUG_ON(gmap_is_shadow(gmap));
 877	fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
 878	if (fixup_user_fault(current, mm, vmaddr, fault_flags, &unlocked))
 879		return -EFAULT;
 880	if (unlocked)
 881		/* lost mmap_sem, caller has to retry __gmap_translate */
 882		return 0;
 883	/* Connect the page tables */
 884	return __gmap_link(gmap, gaddr, vmaddr);
 885}
 886
 887/**
 888 * gmap_pte_op_end - release the page table lock
 889 * @ptl: pointer to the spinlock pointer
 890 */
 891static void gmap_pte_op_end(spinlock_t *ptl)
 892{
 893	if (ptl)
 894		spin_unlock(ptl);
 895}
 896
 897/**
 898 * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
 899 *		      and return the pmd pointer
 900 * @gmap: pointer to guest mapping meta data structure
 901 * @gaddr: virtual address in the guest address space
 902 *
 903 * Returns a pointer to the pmd for a guest address, or NULL
 904 */
 905static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
 906{
 907	pmd_t *pmdp;
 908
 909	BUG_ON(gmap_is_shadow(gmap));
 910	pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
 911	if (!pmdp)
 912		return NULL;
 913
 914	/* without huge pages, there is no need to take the table lock */
 915	if (!gmap->mm->context.allow_gmap_hpage_1m)
 916		return pmd_none(*pmdp) ? NULL : pmdp;
 917
 918	spin_lock(&gmap->guest_table_lock);
 919	if (pmd_none(*pmdp)) {
 920		spin_unlock(&gmap->guest_table_lock);
 921		return NULL;
 922	}
 923
 924	/* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
 925	if (!pmd_large(*pmdp))
 926		spin_unlock(&gmap->guest_table_lock);
 927	return pmdp;
 928}
 929
 930/**
 931 * gmap_pmd_op_end - release the guest_table_lock if needed
 932 * @gmap: pointer to the guest mapping meta data structure
 933 * @pmdp: pointer to the pmd
 934 */
 935static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
 936{
 937	if (pmd_large(*pmdp))
 938		spin_unlock(&gmap->guest_table_lock);
 939}
 940
 941/*
 942 * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
 943 * @pmdp: pointer to the pmd to be protected
 944 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
 945 * @bits: notification bits to set
 946 *
 947 * Returns:
 948 * 0 if successfully protected
 949 * -EAGAIN if a fixup is needed
 950 * -EINVAL if unsupported notifier bits have been specified
 951 *
 952 * Expected to be called with sg->mm->mmap_sem in read and
 953 * guest_table_lock held.
 954 */
 955static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
 956			    pmd_t *pmdp, int prot, unsigned long bits)
 957{
 958	int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
 959	int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
 960	pmd_t new = *pmdp;
 961
 962	/* Fixup needed */
 963	if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
 964		return -EAGAIN;
 965
 966	if (prot == PROT_NONE && !pmd_i) {
 967		pmd_val(new) |= _SEGMENT_ENTRY_INVALID;
 968		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
 969	}
 970
 971	if (prot == PROT_READ && !pmd_p) {
 972		pmd_val(new) &= ~_SEGMENT_ENTRY_INVALID;
 973		pmd_val(new) |= _SEGMENT_ENTRY_PROTECT;
 974		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
 975	}
 976
 977	if (bits & GMAP_NOTIFY_MPROT)
 978		pmd_val(*pmdp) |= _SEGMENT_ENTRY_GMAP_IN;
 979
 980	/* Shadow GMAP protection needs split PMDs */
 981	if (bits & GMAP_NOTIFY_SHADOW)
 982		return -EINVAL;
 983
 984	return 0;
 985}
 986
 987/*
 988 * gmap_protect_pte - remove access rights to memory and set pgste bits
 989 * @gmap: pointer to guest mapping meta data structure
 990 * @gaddr: virtual address in the guest address space
 991 * @pmdp: pointer to the pmd associated with the pte
 992 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
 993 * @bits: notification bits to set
 994 *
 995 * Returns 0 if successfully protected, -ENOMEM if out of memory and
 996 * -EAGAIN if a fixup is needed.
 997 *
 998 * Expected to be called with sg->mm->mmap_sem in read
 999 */
1000static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
1001			    pmd_t *pmdp, int prot, unsigned long bits)
1002{
1003	int rc;
1004	pte_t *ptep;
1005	spinlock_t *ptl = NULL;
1006	unsigned long pbits = 0;
1007
1008	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1009		return -EAGAIN;
1010
1011	ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1012	if (!ptep)
1013		return -ENOMEM;
1014
1015	pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1016	pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1017	/* Protect and unlock. */
1018	rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1019	gmap_pte_op_end(ptl);
1020	return rc;
1021}
1022
1023/*
1024 * gmap_protect_range - remove access rights to memory and set pgste bits
1025 * @gmap: pointer to guest mapping meta data structure
1026 * @gaddr: virtual address in the guest address space
1027 * @len: size of area
1028 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1029 * @bits: pgste notification bits to set
1030 *
1031 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1032 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1033 *
1034 * Called with sg->mm->mmap_sem in read.
1035 */
1036static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1037			      unsigned long len, int prot, unsigned long bits)
1038{
1039	unsigned long vmaddr, dist;
1040	pmd_t *pmdp;
1041	int rc;
1042
1043	BUG_ON(gmap_is_shadow(gmap));
1044	while (len) {
1045		rc = -EAGAIN;
1046		pmdp = gmap_pmd_op_walk(gmap, gaddr);
1047		if (pmdp) {
1048			if (!pmd_large(*pmdp)) {
1049				rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1050						      bits);
1051				if (!rc) {
1052					len -= PAGE_SIZE;
1053					gaddr += PAGE_SIZE;
1054				}
1055			} else {
1056				rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1057						      bits);
1058				if (!rc) {
1059					dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1060					len = len < dist ? 0 : len - dist;
1061					gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1062				}
1063			}
1064			gmap_pmd_op_end(gmap, pmdp);
1065		}
1066		if (rc) {
1067			if (rc == -EINVAL)
1068				return rc;
1069
1070			/* -EAGAIN, fixup of userspace mm and gmap */
1071			vmaddr = __gmap_translate(gmap, gaddr);
1072			if (IS_ERR_VALUE(vmaddr))
1073				return vmaddr;
1074			rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1075			if (rc)
1076				return rc;
1077		}
1078	}
1079	return 0;
1080}
1081
1082/**
1083 * gmap_mprotect_notify - change access rights for a range of ptes and
1084 *                        call the notifier if any pte changes again
1085 * @gmap: pointer to guest mapping meta data structure
1086 * @gaddr: virtual address in the guest address space
1087 * @len: size of area
1088 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1089 *
1090 * Returns 0 if for each page in the given range a gmap mapping exists,
1091 * the new access rights could be set and the notifier could be armed.
1092 * If the gmap mapping is missing for one or more pages -EFAULT is
1093 * returned. If no memory could be allocated -ENOMEM is returned.
1094 * This function establishes missing page table entries.
1095 */
1096int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1097			 unsigned long len, int prot)
1098{
1099	int rc;
1100
1101	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1102		return -EINVAL;
1103	if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1104		return -EINVAL;
1105	down_read(&gmap->mm->mmap_sem);
1106	rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1107	up_read(&gmap->mm->mmap_sem);
1108	return rc;
1109}
1110EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1111
1112/**
1113 * gmap_read_table - get an unsigned long value from a guest page table using
1114 *                   absolute addressing, without marking the page referenced.
1115 * @gmap: pointer to guest mapping meta data structure
1116 * @gaddr: virtual address in the guest address space
1117 * @val: pointer to the unsigned long value to return
1118 *
1119 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1120 * if reading using the virtual address failed. -EINVAL if called on a gmap
1121 * shadow.
1122 *
1123 * Called with gmap->mm->mmap_sem in read.
1124 */
1125int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1126{
1127	unsigned long address, vmaddr;
1128	spinlock_t *ptl;
1129	pte_t *ptep, pte;
1130	int rc;
1131
1132	if (gmap_is_shadow(gmap))
1133		return -EINVAL;
1134
1135	while (1) {
1136		rc = -EAGAIN;
1137		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1138		if (ptep) {
1139			pte = *ptep;
1140			if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1141				address = pte_val(pte) & PAGE_MASK;
1142				address += gaddr & ~PAGE_MASK;
1143				*val = *(unsigned long *) address;
1144				pte_val(*ptep) |= _PAGE_YOUNG;
1145				/* Do *NOT* clear the _PAGE_INVALID bit! */
1146				rc = 0;
1147			}
1148			gmap_pte_op_end(ptl);
1149		}
1150		if (!rc)
1151			break;
1152		vmaddr = __gmap_translate(gmap, gaddr);
1153		if (IS_ERR_VALUE(vmaddr)) {
1154			rc = vmaddr;
1155			break;
1156		}
1157		rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1158		if (rc)
1159			break;
1160	}
1161	return rc;
1162}
1163EXPORT_SYMBOL_GPL(gmap_read_table);
1164
1165/**
1166 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1167 * @sg: pointer to the shadow guest address space structure
1168 * @vmaddr: vm address associated with the rmap
1169 * @rmap: pointer to the rmap structure
1170 *
1171 * Called with the sg->guest_table_lock
1172 */
1173static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1174				    struct gmap_rmap *rmap)
1175{
 
1176	void __rcu **slot;
1177
1178	BUG_ON(!gmap_is_shadow(sg));
1179	slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1180	if (slot) {
1181		rmap->next = radix_tree_deref_slot_protected(slot,
1182							&sg->guest_table_lock);
 
 
 
 
 
 
1183		radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1184	} else {
1185		rmap->next = NULL;
1186		radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1187				  rmap);
1188	}
1189}
1190
1191/**
1192 * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1193 * @sg: pointer to the shadow guest address space structure
1194 * @raddr: rmap address in the shadow gmap
1195 * @paddr: address in the parent guest address space
1196 * @len: length of the memory area to protect
1197 *
1198 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1199 * if out of memory and -EFAULT if paddr is invalid.
1200 */
1201static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1202			     unsigned long paddr, unsigned long len)
1203{
1204	struct gmap *parent;
1205	struct gmap_rmap *rmap;
1206	unsigned long vmaddr;
1207	spinlock_t *ptl;
1208	pte_t *ptep;
1209	int rc;
1210
1211	BUG_ON(!gmap_is_shadow(sg));
1212	parent = sg->parent;
1213	while (len) {
1214		vmaddr = __gmap_translate(parent, paddr);
1215		if (IS_ERR_VALUE(vmaddr))
1216			return vmaddr;
1217		rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
1218		if (!rmap)
1219			return -ENOMEM;
1220		rmap->raddr = raddr;
1221		rc = radix_tree_preload(GFP_KERNEL);
1222		if (rc) {
1223			kfree(rmap);
1224			return rc;
1225		}
1226		rc = -EAGAIN;
1227		ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1228		if (ptep) {
1229			spin_lock(&sg->guest_table_lock);
1230			rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1231					     PGSTE_VSIE_BIT);
1232			if (!rc)
1233				gmap_insert_rmap(sg, vmaddr, rmap);
1234			spin_unlock(&sg->guest_table_lock);
1235			gmap_pte_op_end(ptl);
1236		}
1237		radix_tree_preload_end();
1238		if (rc) {
1239			kfree(rmap);
1240			rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1241			if (rc)
1242				return rc;
1243			continue;
1244		}
1245		paddr += PAGE_SIZE;
1246		len -= PAGE_SIZE;
1247	}
1248	return 0;
1249}
1250
1251#define _SHADOW_RMAP_MASK	0x7
1252#define _SHADOW_RMAP_REGION1	0x5
1253#define _SHADOW_RMAP_REGION2	0x4
1254#define _SHADOW_RMAP_REGION3	0x3
1255#define _SHADOW_RMAP_SEGMENT	0x2
1256#define _SHADOW_RMAP_PGTABLE	0x1
1257
1258/**
1259 * gmap_idte_one - invalidate a single region or segment table entry
1260 * @asce: region or segment table *origin* + table-type bits
1261 * @vaddr: virtual address to identify the table entry to flush
1262 *
1263 * The invalid bit of a single region or segment table entry is set
1264 * and the associated TLB entries depending on the entry are flushed.
1265 * The table-type of the @asce identifies the portion of the @vaddr
1266 * that is used as the invalidation index.
1267 */
1268static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1269{
1270	asm volatile(
1271		"	.insn	rrf,0xb98e0000,%0,%1,0,0"
1272		: : "a" (asce), "a" (vaddr) : "cc", "memory");
1273}
1274
1275/**
1276 * gmap_unshadow_page - remove a page from a shadow page table
1277 * @sg: pointer to the shadow guest address space structure
1278 * @raddr: rmap address in the shadow guest address space
1279 *
1280 * Called with the sg->guest_table_lock
1281 */
1282static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1283{
1284	unsigned long *table;
1285
1286	BUG_ON(!gmap_is_shadow(sg));
1287	table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1288	if (!table || *table & _PAGE_INVALID)
1289		return;
1290	gmap_call_notifier(sg, raddr, raddr + _PAGE_SIZE - 1);
1291	ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1292}
1293
1294/**
1295 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1296 * @sg: pointer to the shadow guest address space structure
1297 * @raddr: rmap address in the shadow guest address space
1298 * @pgt: pointer to the start of a shadow page table
1299 *
1300 * Called with the sg->guest_table_lock
1301 */
1302static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1303				unsigned long *pgt)
1304{
1305	int i;
1306
1307	BUG_ON(!gmap_is_shadow(sg));
1308	for (i = 0; i < _PAGE_ENTRIES; i++, raddr += _PAGE_SIZE)
1309		pgt[i] = _PAGE_INVALID;
1310}
1311
1312/**
1313 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1314 * @sg: pointer to the shadow guest address space structure
1315 * @raddr: address in the shadow guest address space
1316 *
1317 * Called with the sg->guest_table_lock
1318 */
1319static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1320{
1321	unsigned long sto, *ste, *pgt;
 
1322	struct page *page;
1323
1324	BUG_ON(!gmap_is_shadow(sg));
1325	ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1326	if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1327		return;
1328	gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1329	sto = (unsigned long) (ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1330	gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1331	pgt = (unsigned long *)(*ste & _SEGMENT_ENTRY_ORIGIN);
1332	*ste = _SEGMENT_ENTRY_EMPTY;
1333	__gmap_unshadow_pgt(sg, raddr, pgt);
1334	/* Free page table */
1335	page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1336	list_del(&page->lru);
1337	page_table_free_pgste(page);
1338}
1339
1340/**
1341 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1342 * @sg: pointer to the shadow guest address space structure
1343 * @raddr: rmap address in the shadow guest address space
1344 * @sgt: pointer to the start of a shadow segment table
1345 *
1346 * Called with the sg->guest_table_lock
1347 */
1348static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1349				unsigned long *sgt)
1350{
1351	unsigned long *pgt;
1352	struct page *page;
 
1353	int i;
1354
1355	BUG_ON(!gmap_is_shadow(sg));
1356	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1357		if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1358			continue;
1359		pgt = (unsigned long *)(sgt[i] & _REGION_ENTRY_ORIGIN);
1360		sgt[i] = _SEGMENT_ENTRY_EMPTY;
1361		__gmap_unshadow_pgt(sg, raddr, pgt);
1362		/* Free page table */
1363		page = pfn_to_page(__pa(pgt) >> PAGE_SHIFT);
1364		list_del(&page->lru);
1365		page_table_free_pgste(page);
1366	}
1367}
1368
1369/**
1370 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1371 * @sg: pointer to the shadow guest address space structure
1372 * @raddr: rmap address in the shadow guest address space
1373 *
1374 * Called with the shadow->guest_table_lock
1375 */
1376static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1377{
1378	unsigned long r3o, *r3e, *sgt;
 
1379	struct page *page;
1380
1381	BUG_ON(!gmap_is_shadow(sg));
1382	r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1383	if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1384		return;
1385	gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1386	r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1387	gmap_idte_one(r3o | _ASCE_TYPE_REGION3, raddr);
1388	sgt = (unsigned long *)(*r3e & _REGION_ENTRY_ORIGIN);
1389	*r3e = _REGION3_ENTRY_EMPTY;
1390	__gmap_unshadow_sgt(sg, raddr, sgt);
1391	/* Free segment table */
1392	page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1393	list_del(&page->lru);
1394	__free_pages(page, CRST_ALLOC_ORDER);
1395}
1396
1397/**
1398 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1399 * @sg: pointer to the shadow guest address space structure
1400 * @raddr: address in the shadow guest address space
1401 * @r3t: pointer to the start of a shadow region-3 table
1402 *
1403 * Called with the sg->guest_table_lock
1404 */
1405static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1406				unsigned long *r3t)
1407{
1408	unsigned long *sgt;
1409	struct page *page;
 
1410	int i;
1411
1412	BUG_ON(!gmap_is_shadow(sg));
1413	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1414		if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1415			continue;
1416		sgt = (unsigned long *)(r3t[i] & _REGION_ENTRY_ORIGIN);
1417		r3t[i] = _REGION3_ENTRY_EMPTY;
1418		__gmap_unshadow_sgt(sg, raddr, sgt);
1419		/* Free segment table */
1420		page = pfn_to_page(__pa(sgt) >> PAGE_SHIFT);
1421		list_del(&page->lru);
1422		__free_pages(page, CRST_ALLOC_ORDER);
1423	}
1424}
1425
1426/**
1427 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1428 * @sg: pointer to the shadow guest address space structure
1429 * @raddr: rmap address in the shadow guest address space
1430 *
1431 * Called with the sg->guest_table_lock
1432 */
1433static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1434{
1435	unsigned long r2o, *r2e, *r3t;
 
1436	struct page *page;
1437
1438	BUG_ON(!gmap_is_shadow(sg));
1439	r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1440	if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1441		return;
1442	gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1443	r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1444	gmap_idte_one(r2o | _ASCE_TYPE_REGION2, raddr);
1445	r3t = (unsigned long *)(*r2e & _REGION_ENTRY_ORIGIN);
1446	*r2e = _REGION2_ENTRY_EMPTY;
1447	__gmap_unshadow_r3t(sg, raddr, r3t);
1448	/* Free region 3 table */
1449	page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1450	list_del(&page->lru);
1451	__free_pages(page, CRST_ALLOC_ORDER);
1452}
1453
1454/**
1455 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1456 * @sg: pointer to the shadow guest address space structure
1457 * @raddr: rmap address in the shadow guest address space
1458 * @r2t: pointer to the start of a shadow region-2 table
1459 *
1460 * Called with the sg->guest_table_lock
1461 */
1462static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1463				unsigned long *r2t)
1464{
1465	unsigned long *r3t;
1466	struct page *page;
1467	int i;
1468
1469	BUG_ON(!gmap_is_shadow(sg));
1470	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1471		if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1472			continue;
1473		r3t = (unsigned long *)(r2t[i] & _REGION_ENTRY_ORIGIN);
1474		r2t[i] = _REGION2_ENTRY_EMPTY;
1475		__gmap_unshadow_r3t(sg, raddr, r3t);
1476		/* Free region 3 table */
1477		page = pfn_to_page(__pa(r3t) >> PAGE_SHIFT);
1478		list_del(&page->lru);
1479		__free_pages(page, CRST_ALLOC_ORDER);
1480	}
1481}
1482
1483/**
1484 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1485 * @sg: pointer to the shadow guest address space structure
1486 * @raddr: rmap address in the shadow guest address space
1487 *
1488 * Called with the sg->guest_table_lock
1489 */
1490static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1491{
1492	unsigned long r1o, *r1e, *r2t;
1493	struct page *page;
 
1494
1495	BUG_ON(!gmap_is_shadow(sg));
1496	r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1497	if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1498		return;
1499	gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1500	r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1501	gmap_idte_one(r1o | _ASCE_TYPE_REGION1, raddr);
1502	r2t = (unsigned long *)(*r1e & _REGION_ENTRY_ORIGIN);
1503	*r1e = _REGION1_ENTRY_EMPTY;
1504	__gmap_unshadow_r2t(sg, raddr, r2t);
1505	/* Free region 2 table */
1506	page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1507	list_del(&page->lru);
1508	__free_pages(page, CRST_ALLOC_ORDER);
1509}
1510
1511/**
1512 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1513 * @sg: pointer to the shadow guest address space structure
1514 * @raddr: rmap address in the shadow guest address space
1515 * @r1t: pointer to the start of a shadow region-1 table
1516 *
1517 * Called with the shadow->guest_table_lock
1518 */
1519static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1520				unsigned long *r1t)
1521{
1522	unsigned long asce, *r2t;
1523	struct page *page;
 
1524	int i;
1525
1526	BUG_ON(!gmap_is_shadow(sg));
1527	asce = (unsigned long) r1t | _ASCE_TYPE_REGION1;
1528	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1529		if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1530			continue;
1531		r2t = (unsigned long *)(r1t[i] & _REGION_ENTRY_ORIGIN);
1532		__gmap_unshadow_r2t(sg, raddr, r2t);
1533		/* Clear entry and flush translation r1t -> r2t */
1534		gmap_idte_one(asce, raddr);
1535		r1t[i] = _REGION1_ENTRY_EMPTY;
1536		/* Free region 2 table */
1537		page = pfn_to_page(__pa(r2t) >> PAGE_SHIFT);
1538		list_del(&page->lru);
1539		__free_pages(page, CRST_ALLOC_ORDER);
1540	}
1541}
1542
1543/**
1544 * gmap_unshadow - remove a shadow page table completely
1545 * @sg: pointer to the shadow guest address space structure
1546 *
1547 * Called with sg->guest_table_lock
1548 */
1549static void gmap_unshadow(struct gmap *sg)
1550{
1551	unsigned long *table;
1552
1553	BUG_ON(!gmap_is_shadow(sg));
1554	if (sg->removed)
1555		return;
1556	sg->removed = 1;
1557	gmap_call_notifier(sg, 0, -1UL);
1558	gmap_flush_tlb(sg);
1559	table = (unsigned long *)(sg->asce & _ASCE_ORIGIN);
1560	switch (sg->asce & _ASCE_TYPE_MASK) {
1561	case _ASCE_TYPE_REGION1:
1562		__gmap_unshadow_r1t(sg, 0, table);
1563		break;
1564	case _ASCE_TYPE_REGION2:
1565		__gmap_unshadow_r2t(sg, 0, table);
1566		break;
1567	case _ASCE_TYPE_REGION3:
1568		__gmap_unshadow_r3t(sg, 0, table);
1569		break;
1570	case _ASCE_TYPE_SEGMENT:
1571		__gmap_unshadow_sgt(sg, 0, table);
1572		break;
1573	}
1574}
1575
1576/**
1577 * gmap_find_shadow - find a specific asce in the list of shadow tables
1578 * @parent: pointer to the parent gmap
1579 * @asce: ASCE for which the shadow table is created
1580 * @edat_level: edat level to be used for the shadow translation
1581 *
1582 * Returns the pointer to a gmap if a shadow table with the given asce is
1583 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1584 * otherwise NULL
1585 */
1586static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1587				     int edat_level)
1588{
1589	struct gmap *sg;
1590
1591	list_for_each_entry(sg, &parent->children, list) {
1592		if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1593		    sg->removed)
1594			continue;
1595		if (!sg->initialized)
1596			return ERR_PTR(-EAGAIN);
1597		refcount_inc(&sg->ref_count);
1598		return sg;
1599	}
1600	return NULL;
1601}
1602
1603/**
1604 * gmap_shadow_valid - check if a shadow guest address space matches the
1605 *                     given properties and is still valid
1606 * @sg: pointer to the shadow guest address space structure
1607 * @asce: ASCE for which the shadow table is requested
1608 * @edat_level: edat level to be used for the shadow translation
1609 *
1610 * Returns 1 if the gmap shadow is still valid and matches the given
1611 * properties, the caller can continue using it. Returns 0 otherwise, the
1612 * caller has to request a new shadow gmap in this case.
1613 *
1614 */
1615int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1616{
1617	if (sg->removed)
1618		return 0;
1619	return sg->orig_asce == asce && sg->edat_level == edat_level;
1620}
1621EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1622
1623/**
1624 * gmap_shadow - create/find a shadow guest address space
1625 * @parent: pointer to the parent gmap
1626 * @asce: ASCE for which the shadow table is created
1627 * @edat_level: edat level to be used for the shadow translation
1628 *
1629 * The pages of the top level page table referred by the asce parameter
1630 * will be set to read-only and marked in the PGSTEs of the kvm process.
1631 * The shadow table will be removed automatically on any change to the
1632 * PTE mapping for the source table.
1633 *
1634 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1635 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1636 * parent gmap table could not be protected.
1637 */
1638struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1639			 int edat_level)
1640{
1641	struct gmap *sg, *new;
1642	unsigned long limit;
1643	int rc;
1644
1645	BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1646	BUG_ON(gmap_is_shadow(parent));
1647	spin_lock(&parent->shadow_lock);
1648	sg = gmap_find_shadow(parent, asce, edat_level);
1649	spin_unlock(&parent->shadow_lock);
1650	if (sg)
1651		return sg;
1652	/* Create a new shadow gmap */
1653	limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1654	if (asce & _ASCE_REAL_SPACE)
1655		limit = -1UL;
1656	new = gmap_alloc(limit);
1657	if (!new)
1658		return ERR_PTR(-ENOMEM);
1659	new->mm = parent->mm;
1660	new->parent = gmap_get(parent);
1661	new->orig_asce = asce;
1662	new->edat_level = edat_level;
1663	new->initialized = false;
1664	spin_lock(&parent->shadow_lock);
1665	/* Recheck if another CPU created the same shadow */
1666	sg = gmap_find_shadow(parent, asce, edat_level);
1667	if (sg) {
1668		spin_unlock(&parent->shadow_lock);
1669		gmap_free(new);
1670		return sg;
1671	}
1672	if (asce & _ASCE_REAL_SPACE) {
1673		/* only allow one real-space gmap shadow */
1674		list_for_each_entry(sg, &parent->children, list) {
1675			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1676				spin_lock(&sg->guest_table_lock);
1677				gmap_unshadow(sg);
1678				spin_unlock(&sg->guest_table_lock);
1679				list_del(&sg->list);
1680				gmap_put(sg);
1681				break;
1682			}
1683		}
1684	}
1685	refcount_set(&new->ref_count, 2);
1686	list_add(&new->list, &parent->children);
1687	if (asce & _ASCE_REAL_SPACE) {
1688		/* nothing to protect, return right away */
1689		new->initialized = true;
1690		spin_unlock(&parent->shadow_lock);
1691		return new;
1692	}
1693	spin_unlock(&parent->shadow_lock);
1694	/* protect after insertion, so it will get properly invalidated */
1695	down_read(&parent->mm->mmap_sem);
1696	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1697				((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1698				PROT_READ, GMAP_NOTIFY_SHADOW);
1699	up_read(&parent->mm->mmap_sem);
1700	spin_lock(&parent->shadow_lock);
1701	new->initialized = true;
1702	if (rc) {
1703		list_del(&new->list);
1704		gmap_free(new);
1705		new = ERR_PTR(rc);
1706	}
1707	spin_unlock(&parent->shadow_lock);
1708	return new;
1709}
1710EXPORT_SYMBOL_GPL(gmap_shadow);
1711
1712/**
1713 * gmap_shadow_r2t - create an empty shadow region 2 table
1714 * @sg: pointer to the shadow guest address space structure
1715 * @saddr: faulting address in the shadow gmap
1716 * @r2t: parent gmap address of the region 2 table to get shadowed
1717 * @fake: r2t references contiguous guest memory block, not a r2t
1718 *
1719 * The r2t parameter specifies the address of the source table. The
1720 * four pages of the source table are made read-only in the parent gmap
1721 * address space. A write to the source table area @r2t will automatically
1722 * remove the shadow r2 table and all of its decendents.
1723 *
1724 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1725 * shadow table structure is incomplete, -ENOMEM if out of memory and
1726 * -EFAULT if an address in the parent gmap could not be resolved.
1727 *
1728 * Called with sg->mm->mmap_sem in read.
1729 */
1730int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1731		    int fake)
1732{
1733	unsigned long raddr, origin, offset, len;
1734	unsigned long *s_r2t, *table;
 
1735	struct page *page;
1736	int rc;
1737
1738	BUG_ON(!gmap_is_shadow(sg));
1739	/* Allocate a shadow region second table */
1740	page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1741	if (!page)
1742		return -ENOMEM;
1743	page->index = r2t & _REGION_ENTRY_ORIGIN;
1744	if (fake)
1745		page->index |= GMAP_SHADOW_FAKE_TABLE;
1746	s_r2t = (unsigned long *) page_to_phys(page);
1747	/* Install shadow region second table */
1748	spin_lock(&sg->guest_table_lock);
1749	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1750	if (!table) {
1751		rc = -EAGAIN;		/* Race with unshadow */
1752		goto out_free;
1753	}
1754	if (!(*table & _REGION_ENTRY_INVALID)) {
1755		rc = 0;			/* Already established */
1756		goto out_free;
1757	} else if (*table & _REGION_ENTRY_ORIGIN) {
1758		rc = -EAGAIN;		/* Race with shadow */
1759		goto out_free;
1760	}
1761	crst_table_init(s_r2t, _REGION2_ENTRY_EMPTY);
1762	/* mark as invalid as long as the parent table is not protected */
1763	*table = (unsigned long) s_r2t | _REGION_ENTRY_LENGTH |
1764		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1765	if (sg->edat_level >= 1)
1766		*table |= (r2t & _REGION_ENTRY_PROTECT);
1767	list_add(&page->lru, &sg->crst_list);
1768	if (fake) {
1769		/* nothing to protect for fake tables */
1770		*table &= ~_REGION_ENTRY_INVALID;
1771		spin_unlock(&sg->guest_table_lock);
1772		return 0;
1773	}
1774	spin_unlock(&sg->guest_table_lock);
1775	/* Make r2t read-only in parent gmap page table */
1776	raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1777	origin = r2t & _REGION_ENTRY_ORIGIN;
1778	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1779	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1780	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1781	spin_lock(&sg->guest_table_lock);
1782	if (!rc) {
1783		table = gmap_table_walk(sg, saddr, 4);
1784		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1785			      (unsigned long) s_r2t)
1786			rc = -EAGAIN;		/* Race with unshadow */
1787		else
1788			*table &= ~_REGION_ENTRY_INVALID;
1789	} else {
1790		gmap_unshadow_r2t(sg, raddr);
1791	}
1792	spin_unlock(&sg->guest_table_lock);
1793	return rc;
1794out_free:
1795	spin_unlock(&sg->guest_table_lock);
1796	__free_pages(page, CRST_ALLOC_ORDER);
1797	return rc;
1798}
1799EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1800
1801/**
1802 * gmap_shadow_r3t - create a shadow region 3 table
1803 * @sg: pointer to the shadow guest address space structure
1804 * @saddr: faulting address in the shadow gmap
1805 * @r3t: parent gmap address of the region 3 table to get shadowed
1806 * @fake: r3t references contiguous guest memory block, not a r3t
1807 *
1808 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1809 * shadow table structure is incomplete, -ENOMEM if out of memory and
1810 * -EFAULT if an address in the parent gmap could not be resolved.
1811 *
1812 * Called with sg->mm->mmap_sem in read.
1813 */
1814int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1815		    int fake)
1816{
1817	unsigned long raddr, origin, offset, len;
1818	unsigned long *s_r3t, *table;
 
1819	struct page *page;
1820	int rc;
1821
1822	BUG_ON(!gmap_is_shadow(sg));
1823	/* Allocate a shadow region second table */
1824	page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1825	if (!page)
1826		return -ENOMEM;
1827	page->index = r3t & _REGION_ENTRY_ORIGIN;
1828	if (fake)
1829		page->index |= GMAP_SHADOW_FAKE_TABLE;
1830	s_r3t = (unsigned long *) page_to_phys(page);
1831	/* Install shadow region second table */
1832	spin_lock(&sg->guest_table_lock);
1833	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1834	if (!table) {
1835		rc = -EAGAIN;		/* Race with unshadow */
1836		goto out_free;
1837	}
1838	if (!(*table & _REGION_ENTRY_INVALID)) {
1839		rc = 0;			/* Already established */
1840		goto out_free;
1841	} else if (*table & _REGION_ENTRY_ORIGIN) {
1842		rc = -EAGAIN;		/* Race with shadow */
 
1843	}
1844	crst_table_init(s_r3t, _REGION3_ENTRY_EMPTY);
1845	/* mark as invalid as long as the parent table is not protected */
1846	*table = (unsigned long) s_r3t | _REGION_ENTRY_LENGTH |
1847		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1848	if (sg->edat_level >= 1)
1849		*table |= (r3t & _REGION_ENTRY_PROTECT);
1850	list_add(&page->lru, &sg->crst_list);
1851	if (fake) {
1852		/* nothing to protect for fake tables */
1853		*table &= ~_REGION_ENTRY_INVALID;
1854		spin_unlock(&sg->guest_table_lock);
1855		return 0;
1856	}
1857	spin_unlock(&sg->guest_table_lock);
1858	/* Make r3t read-only in parent gmap page table */
1859	raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1860	origin = r3t & _REGION_ENTRY_ORIGIN;
1861	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1862	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1863	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1864	spin_lock(&sg->guest_table_lock);
1865	if (!rc) {
1866		table = gmap_table_walk(sg, saddr, 3);
1867		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1868			      (unsigned long) s_r3t)
1869			rc = -EAGAIN;		/* Race with unshadow */
1870		else
1871			*table &= ~_REGION_ENTRY_INVALID;
1872	} else {
1873		gmap_unshadow_r3t(sg, raddr);
1874	}
1875	spin_unlock(&sg->guest_table_lock);
1876	return rc;
1877out_free:
1878	spin_unlock(&sg->guest_table_lock);
1879	__free_pages(page, CRST_ALLOC_ORDER);
1880	return rc;
1881}
1882EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1883
1884/**
1885 * gmap_shadow_sgt - create a shadow segment table
1886 * @sg: pointer to the shadow guest address space structure
1887 * @saddr: faulting address in the shadow gmap
1888 * @sgt: parent gmap address of the segment table to get shadowed
1889 * @fake: sgt references contiguous guest memory block, not a sgt
1890 *
1891 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1892 * shadow table structure is incomplete, -ENOMEM if out of memory and
1893 * -EFAULT if an address in the parent gmap could not be resolved.
1894 *
1895 * Called with sg->mm->mmap_sem in read.
1896 */
1897int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1898		    int fake)
1899{
1900	unsigned long raddr, origin, offset, len;
1901	unsigned long *s_sgt, *table;
 
1902	struct page *page;
1903	int rc;
1904
1905	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1906	/* Allocate a shadow segment table */
1907	page = alloc_pages(GFP_KERNEL, CRST_ALLOC_ORDER);
1908	if (!page)
1909		return -ENOMEM;
1910	page->index = sgt & _REGION_ENTRY_ORIGIN;
1911	if (fake)
1912		page->index |= GMAP_SHADOW_FAKE_TABLE;
1913	s_sgt = (unsigned long *) page_to_phys(page);
1914	/* Install shadow region second table */
1915	spin_lock(&sg->guest_table_lock);
1916	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
1917	if (!table) {
1918		rc = -EAGAIN;		/* Race with unshadow */
1919		goto out_free;
1920	}
1921	if (!(*table & _REGION_ENTRY_INVALID)) {
1922		rc = 0;			/* Already established */
1923		goto out_free;
1924	} else if (*table & _REGION_ENTRY_ORIGIN) {
1925		rc = -EAGAIN;		/* Race with shadow */
1926		goto out_free;
1927	}
1928	crst_table_init(s_sgt, _SEGMENT_ENTRY_EMPTY);
1929	/* mark as invalid as long as the parent table is not protected */
1930	*table = (unsigned long) s_sgt | _REGION_ENTRY_LENGTH |
1931		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
1932	if (sg->edat_level >= 1)
1933		*table |= sgt & _REGION_ENTRY_PROTECT;
1934	list_add(&page->lru, &sg->crst_list);
1935	if (fake) {
1936		/* nothing to protect for fake tables */
1937		*table &= ~_REGION_ENTRY_INVALID;
1938		spin_unlock(&sg->guest_table_lock);
1939		return 0;
1940	}
1941	spin_unlock(&sg->guest_table_lock);
1942	/* Make sgt read-only in parent gmap page table */
1943	raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
1944	origin = sgt & _REGION_ENTRY_ORIGIN;
1945	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1946	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1947	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1948	spin_lock(&sg->guest_table_lock);
1949	if (!rc) {
1950		table = gmap_table_walk(sg, saddr, 2);
1951		if (!table || (*table & _REGION_ENTRY_ORIGIN) !=
1952			      (unsigned long) s_sgt)
1953			rc = -EAGAIN;		/* Race with unshadow */
1954		else
1955			*table &= ~_REGION_ENTRY_INVALID;
1956	} else {
1957		gmap_unshadow_sgt(sg, raddr);
1958	}
1959	spin_unlock(&sg->guest_table_lock);
1960	return rc;
1961out_free:
1962	spin_unlock(&sg->guest_table_lock);
1963	__free_pages(page, CRST_ALLOC_ORDER);
1964	return rc;
1965}
1966EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
1967
1968/**
1969 * gmap_shadow_lookup_pgtable - find a shadow page table
1970 * @sg: pointer to the shadow guest address space structure
1971 * @saddr: the address in the shadow aguest address space
1972 * @pgt: parent gmap address of the page table to get shadowed
1973 * @dat_protection: if the pgtable is marked as protected by dat
1974 * @fake: pgt references contiguous guest memory block, not a pgtable
1975 *
1976 * Returns 0 if the shadow page table was found and -EAGAIN if the page
1977 * table was not found.
1978 *
1979 * Called with sg->mm->mmap_sem in read.
1980 */
1981int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
1982			   unsigned long *pgt, int *dat_protection,
1983			   int *fake)
1984{
1985	unsigned long *table;
1986	struct page *page;
1987	int rc;
1988
1989	BUG_ON(!gmap_is_shadow(sg));
1990	spin_lock(&sg->guest_table_lock);
1991	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
1992	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
1993		/* Shadow page tables are full pages (pte+pgste) */
1994		page = pfn_to_page(*table >> PAGE_SHIFT);
1995		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
1996		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
1997		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
1998		rc = 0;
1999	} else  {
2000		rc = -EAGAIN;
2001	}
2002	spin_unlock(&sg->guest_table_lock);
2003	return rc;
2004
2005}
2006EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2007
2008/**
2009 * gmap_shadow_pgt - instantiate a shadow page table
2010 * @sg: pointer to the shadow guest address space structure
2011 * @saddr: faulting address in the shadow gmap
2012 * @pgt: parent gmap address of the page table to get shadowed
2013 * @fake: pgt references contiguous guest memory block, not a pgtable
2014 *
2015 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2016 * shadow table structure is incomplete, -ENOMEM if out of memory,
2017 * -EFAULT if an address in the parent gmap could not be resolved and
2018 *
2019 * Called with gmap->mm->mmap_sem in read
2020 */
2021int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2022		    int fake)
2023{
2024	unsigned long raddr, origin;
2025	unsigned long *s_pgt, *table;
2026	struct page *page;
 
2027	int rc;
2028
2029	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2030	/* Allocate a shadow page table */
2031	page = page_table_alloc_pgste(sg->mm);
2032	if (!page)
2033		return -ENOMEM;
2034	page->index = pgt & _SEGMENT_ENTRY_ORIGIN;
2035	if (fake)
2036		page->index |= GMAP_SHADOW_FAKE_TABLE;
2037	s_pgt = (unsigned long *) page_to_phys(page);
2038	/* Install shadow page table */
2039	spin_lock(&sg->guest_table_lock);
2040	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2041	if (!table) {
2042		rc = -EAGAIN;		/* Race with unshadow */
2043		goto out_free;
2044	}
2045	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2046		rc = 0;			/* Already established */
2047		goto out_free;
2048	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2049		rc = -EAGAIN;		/* Race with shadow */
2050		goto out_free;
2051	}
2052	/* mark as invalid as long as the parent table is not protected */
2053	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2054		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2055	list_add(&page->lru, &sg->pt_list);
2056	if (fake) {
2057		/* nothing to protect for fake tables */
2058		*table &= ~_SEGMENT_ENTRY_INVALID;
2059		spin_unlock(&sg->guest_table_lock);
2060		return 0;
2061	}
2062	spin_unlock(&sg->guest_table_lock);
2063	/* Make pgt read-only in parent gmap page table (not the pgste) */
2064	raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2065	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2066	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2067	spin_lock(&sg->guest_table_lock);
2068	if (!rc) {
2069		table = gmap_table_walk(sg, saddr, 1);
2070		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) !=
2071			      (unsigned long) s_pgt)
2072			rc = -EAGAIN;		/* Race with unshadow */
2073		else
2074			*table &= ~_SEGMENT_ENTRY_INVALID;
2075	} else {
2076		gmap_unshadow_pgt(sg, raddr);
2077	}
2078	spin_unlock(&sg->guest_table_lock);
2079	return rc;
2080out_free:
2081	spin_unlock(&sg->guest_table_lock);
2082	page_table_free_pgste(page);
2083	return rc;
2084
2085}
2086EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2087
2088/**
2089 * gmap_shadow_page - create a shadow page mapping
2090 * @sg: pointer to the shadow guest address space structure
2091 * @saddr: faulting address in the shadow gmap
2092 * @pte: pte in parent gmap address space to get shadowed
2093 *
2094 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2095 * shadow table structure is incomplete, -ENOMEM if out of memory and
2096 * -EFAULT if an address in the parent gmap could not be resolved.
2097 *
2098 * Called with sg->mm->mmap_sem in read.
2099 */
2100int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2101{
2102	struct gmap *parent;
2103	struct gmap_rmap *rmap;
2104	unsigned long vmaddr, paddr;
2105	spinlock_t *ptl;
2106	pte_t *sptep, *tptep;
2107	int prot;
2108	int rc;
2109
2110	BUG_ON(!gmap_is_shadow(sg));
2111	parent = sg->parent;
2112	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2113
2114	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL);
2115	if (!rmap)
2116		return -ENOMEM;
2117	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2118
2119	while (1) {
2120		paddr = pte_val(pte) & PAGE_MASK;
2121		vmaddr = __gmap_translate(parent, paddr);
2122		if (IS_ERR_VALUE(vmaddr)) {
2123			rc = vmaddr;
2124			break;
2125		}
2126		rc = radix_tree_preload(GFP_KERNEL);
2127		if (rc)
2128			break;
2129		rc = -EAGAIN;
2130		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2131		if (sptep) {
2132			spin_lock(&sg->guest_table_lock);
2133			/* Get page table pointer */
2134			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2135			if (!tptep) {
2136				spin_unlock(&sg->guest_table_lock);
2137				gmap_pte_op_end(ptl);
2138				radix_tree_preload_end();
2139				break;
2140			}
2141			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2142			if (rc > 0) {
2143				/* Success and a new mapping */
2144				gmap_insert_rmap(sg, vmaddr, rmap);
2145				rmap = NULL;
2146				rc = 0;
2147			}
2148			gmap_pte_op_end(ptl);
2149			spin_unlock(&sg->guest_table_lock);
2150		}
2151		radix_tree_preload_end();
2152		if (!rc)
2153			break;
2154		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2155		if (rc)
2156			break;
2157	}
2158	kfree(rmap);
2159	return rc;
2160}
2161EXPORT_SYMBOL_GPL(gmap_shadow_page);
2162
2163/**
2164 * gmap_shadow_notify - handle notifications for shadow gmap
2165 *
2166 * Called with sg->parent->shadow_lock.
2167 */
2168static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2169			       unsigned long gaddr)
2170{
2171	struct gmap_rmap *rmap, *rnext, *head;
2172	unsigned long start, end, bits, raddr;
2173
2174	BUG_ON(!gmap_is_shadow(sg));
2175
2176	spin_lock(&sg->guest_table_lock);
2177	if (sg->removed) {
2178		spin_unlock(&sg->guest_table_lock);
2179		return;
2180	}
2181	/* Check for top level table */
2182	start = sg->orig_asce & _ASCE_ORIGIN;
2183	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2184	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2185	    gaddr < end) {
2186		/* The complete shadow table has to go */
2187		gmap_unshadow(sg);
2188		spin_unlock(&sg->guest_table_lock);
2189		list_del(&sg->list);
2190		gmap_put(sg);
2191		return;
2192	}
2193	/* Remove the page table tree from on specific entry */
2194	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2195	gmap_for_each_rmap_safe(rmap, rnext, head) {
2196		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2197		raddr = rmap->raddr ^ bits;
2198		switch (bits) {
2199		case _SHADOW_RMAP_REGION1:
2200			gmap_unshadow_r2t(sg, raddr);
2201			break;
2202		case _SHADOW_RMAP_REGION2:
2203			gmap_unshadow_r3t(sg, raddr);
2204			break;
2205		case _SHADOW_RMAP_REGION3:
2206			gmap_unshadow_sgt(sg, raddr);
2207			break;
2208		case _SHADOW_RMAP_SEGMENT:
2209			gmap_unshadow_pgt(sg, raddr);
2210			break;
2211		case _SHADOW_RMAP_PGTABLE:
2212			gmap_unshadow_page(sg, raddr);
2213			break;
2214		}
2215		kfree(rmap);
2216	}
2217	spin_unlock(&sg->guest_table_lock);
2218}
2219
2220/**
2221 * ptep_notify - call all invalidation callbacks for a specific pte.
2222 * @mm: pointer to the process mm_struct
2223 * @addr: virtual address in the process address space
2224 * @pte: pointer to the page table entry
2225 * @bits: bits from the pgste that caused the notify call
2226 *
2227 * This function is assumed to be called with the page table lock held
2228 * for the pte to notify.
2229 */
2230void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2231		 pte_t *pte, unsigned long bits)
2232{
2233	unsigned long offset, gaddr = 0;
2234	unsigned long *table;
2235	struct gmap *gmap, *sg, *next;
2236
2237	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2238	offset = offset * (PAGE_SIZE / sizeof(pte_t));
2239	rcu_read_lock();
2240	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2241		spin_lock(&gmap->guest_table_lock);
2242		table = radix_tree_lookup(&gmap->host_to_guest,
2243					  vmaddr >> PMD_SHIFT);
2244		if (table)
2245			gaddr = __gmap_segment_gaddr(table) + offset;
2246		spin_unlock(&gmap->guest_table_lock);
2247		if (!table)
2248			continue;
2249
2250		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2251			spin_lock(&gmap->shadow_lock);
2252			list_for_each_entry_safe(sg, next,
2253						 &gmap->children, list)
2254				gmap_shadow_notify(sg, vmaddr, gaddr);
2255			spin_unlock(&gmap->shadow_lock);
2256		}
2257		if (bits & PGSTE_IN_BIT)
2258			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2259	}
2260	rcu_read_unlock();
2261}
2262EXPORT_SYMBOL_GPL(ptep_notify);
2263
2264static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2265			     unsigned long gaddr)
2266{
2267	pmd_val(*pmdp) &= ~_SEGMENT_ENTRY_GMAP_IN;
2268	gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2269}
2270
2271/**
2272 * gmap_pmdp_xchg - exchange a gmap pmd with another
2273 * @gmap: pointer to the guest address space structure
2274 * @pmdp: pointer to the pmd entry
2275 * @new: replacement entry
2276 * @gaddr: the affected guest address
2277 *
2278 * This function is assumed to be called with the guest_table_lock
2279 * held.
2280 */
2281static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2282			   unsigned long gaddr)
2283{
2284	gaddr &= HPAGE_MASK;
2285	pmdp_notify_gmap(gmap, pmdp, gaddr);
2286	pmd_val(new) &= ~_SEGMENT_ENTRY_GMAP_IN;
2287	if (MACHINE_HAS_TLB_GUEST)
2288		__pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2289			    IDTE_GLOBAL);
2290	else if (MACHINE_HAS_IDTE)
2291		__pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2292	else
2293		__pmdp_csp(pmdp);
2294	*pmdp = new;
2295}
2296
2297static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2298			    int purge)
2299{
2300	pmd_t *pmdp;
2301	struct gmap *gmap;
2302	unsigned long gaddr;
2303
2304	rcu_read_lock();
2305	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2306		spin_lock(&gmap->guest_table_lock);
2307		pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2308						  vmaddr >> PMD_SHIFT);
2309		if (pmdp) {
2310			gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2311			pmdp_notify_gmap(gmap, pmdp, gaddr);
2312			WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2313						   _SEGMENT_ENTRY_GMAP_UC));
2314			if (purge)
2315				__pmdp_csp(pmdp);
2316			pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
2317		}
2318		spin_unlock(&gmap->guest_table_lock);
2319	}
2320	rcu_read_unlock();
2321}
2322
2323/**
2324 * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2325 *                        flushing
2326 * @mm: pointer to the process mm_struct
2327 * @vmaddr: virtual address in the process address space
2328 */
2329void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2330{
2331	gmap_pmdp_clear(mm, vmaddr, 0);
2332}
2333EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2334
2335/**
2336 * gmap_pmdp_csp - csp all affected guest pmd entries
2337 * @mm: pointer to the process mm_struct
2338 * @vmaddr: virtual address in the process address space
2339 */
2340void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2341{
2342	gmap_pmdp_clear(mm, vmaddr, 1);
2343}
2344EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2345
2346/**
2347 * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2348 * @mm: pointer to the process mm_struct
2349 * @vmaddr: virtual address in the process address space
2350 */
2351void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2352{
2353	unsigned long *entry, gaddr;
2354	struct gmap *gmap;
2355	pmd_t *pmdp;
2356
2357	rcu_read_lock();
2358	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2359		spin_lock(&gmap->guest_table_lock);
2360		entry = radix_tree_delete(&gmap->host_to_guest,
2361					  vmaddr >> PMD_SHIFT);
2362		if (entry) {
2363			pmdp = (pmd_t *)entry;
2364			gaddr = __gmap_segment_gaddr(entry);
2365			pmdp_notify_gmap(gmap, pmdp, gaddr);
2366			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2367					   _SEGMENT_ENTRY_GMAP_UC));
2368			if (MACHINE_HAS_TLB_GUEST)
2369				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2370					    gmap->asce, IDTE_LOCAL);
2371			else if (MACHINE_HAS_IDTE)
2372				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2373			*entry = _SEGMENT_ENTRY_EMPTY;
2374		}
2375		spin_unlock(&gmap->guest_table_lock);
2376	}
2377	rcu_read_unlock();
2378}
2379EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2380
2381/**
2382 * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2383 * @mm: pointer to the process mm_struct
2384 * @vmaddr: virtual address in the process address space
2385 */
2386void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2387{
2388	unsigned long *entry, gaddr;
2389	struct gmap *gmap;
2390	pmd_t *pmdp;
2391
2392	rcu_read_lock();
2393	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2394		spin_lock(&gmap->guest_table_lock);
2395		entry = radix_tree_delete(&gmap->host_to_guest,
2396					  vmaddr >> PMD_SHIFT);
2397		if (entry) {
2398			pmdp = (pmd_t *)entry;
2399			gaddr = __gmap_segment_gaddr(entry);
2400			pmdp_notify_gmap(gmap, pmdp, gaddr);
2401			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2402					   _SEGMENT_ENTRY_GMAP_UC));
2403			if (MACHINE_HAS_TLB_GUEST)
2404				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2405					    gmap->asce, IDTE_GLOBAL);
2406			else if (MACHINE_HAS_IDTE)
2407				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2408			else
2409				__pmdp_csp(pmdp);
2410			*entry = _SEGMENT_ENTRY_EMPTY;
2411		}
2412		spin_unlock(&gmap->guest_table_lock);
2413	}
2414	rcu_read_unlock();
2415}
2416EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2417
2418/**
2419 * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2420 * @gmap: pointer to guest address space
2421 * @pmdp: pointer to the pmd to be tested
2422 * @gaddr: virtual address in the guest address space
2423 *
2424 * This function is assumed to be called with the guest_table_lock
2425 * held.
2426 */
2427static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2428					  unsigned long gaddr)
2429{
2430	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2431		return false;
2432
2433	/* Already protected memory, which did not change is clean */
2434	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2435	    !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2436		return false;
2437
2438	/* Clear UC indication and reset protection */
2439	pmd_val(*pmdp) &= ~_SEGMENT_ENTRY_GMAP_UC;
2440	gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2441	return true;
2442}
2443
2444/**
2445 * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2446 * @gmap: pointer to guest address space
2447 * @bitmap: dirty bitmap for this pmd
2448 * @gaddr: virtual address in the guest address space
2449 * @vmaddr: virtual address in the host address space
2450 *
2451 * This function is assumed to be called with the guest_table_lock
2452 * held.
2453 */
2454void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2455			     unsigned long gaddr, unsigned long vmaddr)
2456{
2457	int i;
2458	pmd_t *pmdp;
2459	pte_t *ptep;
2460	spinlock_t *ptl;
2461
2462	pmdp = gmap_pmd_op_walk(gmap, gaddr);
2463	if (!pmdp)
2464		return;
2465
2466	if (pmd_large(*pmdp)) {
2467		if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2468			bitmap_fill(bitmap, _PAGE_ENTRIES);
2469	} else {
2470		for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2471			ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2472			if (!ptep)
2473				continue;
2474			if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2475				set_bit(i, bitmap);
2476			spin_unlock(ptl);
2477		}
2478	}
2479	gmap_pmd_op_end(gmap, pmdp);
2480}
2481EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2482
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2483static inline void thp_split_mm(struct mm_struct *mm)
2484{
2485#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2486	struct vm_area_struct *vma;
2487	unsigned long addr;
2488
2489	for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
2490		for (addr = vma->vm_start;
2491		     addr < vma->vm_end;
2492		     addr += PAGE_SIZE)
2493			follow_page(vma, addr, FOLL_SPLIT);
2494		vma->vm_flags &= ~VM_HUGEPAGE;
2495		vma->vm_flags |= VM_NOHUGEPAGE;
 
2496	}
2497	mm->def_flags |= VM_NOHUGEPAGE;
2498#endif
2499}
 
 
 
 
 
2500
2501/*
2502 * Remove all empty zero pages from the mapping for lazy refaulting
2503 * - This must be called after mm->context.has_pgste is set, to avoid
2504 *   future creation of zero pages
2505 * - This must be called after THP was enabled
2506 */
2507static int __zap_zero_pages(pmd_t *pmd, unsigned long start,
2508			   unsigned long end, struct mm_walk *walk)
2509{
2510	unsigned long addr;
2511
2512	for (addr = start; addr != end; addr += PAGE_SIZE) {
2513		pte_t *ptep;
2514		spinlock_t *ptl;
2515
2516		ptep = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
2517		if (is_zero_pfn(pte_pfn(*ptep)))
2518			ptep_xchg_direct(walk->mm, addr, ptep, __pte(_PAGE_INVALID));
2519		pte_unmap_unlock(ptep, ptl);
2520	}
2521	return 0;
2522}
2523
2524static const struct mm_walk_ops zap_zero_walk_ops = {
2525	.pmd_entry	= __zap_zero_pages,
2526};
2527
2528/*
2529 * switch on pgstes for its userspace process (for kvm)
2530 */
2531int s390_enable_sie(void)
2532{
2533	struct mm_struct *mm = current->mm;
2534
2535	/* Do we have pgstes? if yes, we are done */
2536	if (mm_has_pgste(mm))
2537		return 0;
2538	/* Fail if the page tables are 2K */
2539	if (!mm_alloc_pgste(mm))
2540		return -EINVAL;
2541	down_write(&mm->mmap_sem);
2542	mm->context.has_pgste = 1;
2543	/* split thp mappings and disable thp for future mappings */
2544	thp_split_mm(mm);
2545	walk_page_range(mm, 0, TASK_SIZE, &zap_zero_walk_ops, NULL);
2546	up_write(&mm->mmap_sem);
2547	return 0;
2548}
2549EXPORT_SYMBOL_GPL(s390_enable_sie);
2550
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2551/*
2552 * Enable storage key handling from now on and initialize the storage
2553 * keys with the default key.
2554 */
2555static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2556				  unsigned long next, struct mm_walk *walk)
2557{
2558	/* Clear storage key */
2559	ptep_zap_key(walk->mm, addr, pte);
2560	return 0;
2561}
2562
 
 
 
 
 
 
 
 
 
 
 
 
2563static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2564				      unsigned long hmask, unsigned long next,
2565				      struct mm_walk *walk)
2566{
2567	pmd_t *pmd = (pmd_t *)pte;
2568	unsigned long start, end;
2569	struct page *page = pmd_page(*pmd);
2570
2571	/*
2572	 * The write check makes sure we do not set a key on shared
2573	 * memory. This is needed as the walker does not differentiate
2574	 * between actual guest memory and the process executable or
2575	 * shared libraries.
2576	 */
2577	if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2578	    !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2579		return 0;
2580
2581	start = pmd_val(*pmd) & HPAGE_MASK;
2582	end = start + HPAGE_SIZE - 1;
2583	__storage_key_init_range(start, end);
2584	set_bit(PG_arch_1, &page->flags);
 
2585	return 0;
2586}
2587
2588static const struct mm_walk_ops enable_skey_walk_ops = {
2589	.hugetlb_entry		= __s390_enable_skey_hugetlb,
2590	.pte_entry		= __s390_enable_skey_pte,
 
2591};
2592
2593int s390_enable_skey(void)
2594{
2595	struct mm_struct *mm = current->mm;
2596	struct vm_area_struct *vma;
2597	int rc = 0;
2598
2599	down_write(&mm->mmap_sem);
2600	if (mm_uses_skeys(mm))
2601		goto out_up;
2602
2603	mm->context.uses_skeys = 1;
2604	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2605		if (ksm_madvise(vma, vma->vm_start, vma->vm_end,
2606				MADV_UNMERGEABLE, &vma->vm_flags)) {
2607			mm->context.uses_skeys = 0;
2608			rc = -ENOMEM;
2609			goto out_up;
2610		}
2611	}
2612	mm->def_flags &= ~VM_MERGEABLE;
2613
2614	walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2615
2616out_up:
2617	up_write(&mm->mmap_sem);
2618	return rc;
2619}
2620EXPORT_SYMBOL_GPL(s390_enable_skey);
2621
2622/*
2623 * Reset CMMA state, make all pages stable again.
2624 */
2625static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2626			     unsigned long next, struct mm_walk *walk)
2627{
2628	ptep_zap_unused(walk->mm, addr, pte, 1);
2629	return 0;
2630}
2631
2632static const struct mm_walk_ops reset_cmma_walk_ops = {
2633	.pte_entry		= __s390_reset_cmma,
2634};
2635
2636void s390_reset_cmma(struct mm_struct *mm)
2637{
2638	down_write(&mm->mmap_sem);
2639	walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2640	up_write(&mm->mmap_sem);
2641}
2642EXPORT_SYMBOL_GPL(s390_reset_cmma);