Linux Audio

Check our new training course

Loading...
v6.13.7
   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#include <asm/page-states.h>
  22#include <asm/pgalloc.h>
  23#include <asm/gmap.h>
  24#include <asm/page.h>
  25#include <asm/tlb.h>
  26
  27#define GMAP_SHADOW_FAKE_TABLE 1ULL
  28
  29static struct page *gmap_alloc_crst(void)
  30{
  31	struct page *page;
  32
  33	page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
  34	if (!page)
  35		return NULL;
  36	__arch_set_page_dat(page_to_virt(page), 1UL << CRST_ALLOC_ORDER);
  37	return page;
  38}
  39
  40/**
  41 * gmap_alloc - allocate and initialize a guest address space
  42 * @limit: maximum address of the gmap address space
  43 *
  44 * Returns a guest address space structure.
  45 */
  46static struct gmap *gmap_alloc(unsigned long limit)
  47{
  48	struct gmap *gmap;
  49	struct page *page;
  50	unsigned long *table;
  51	unsigned long etype, atype;
  52
  53	if (limit < _REGION3_SIZE) {
  54		limit = _REGION3_SIZE - 1;
  55		atype = _ASCE_TYPE_SEGMENT;
  56		etype = _SEGMENT_ENTRY_EMPTY;
  57	} else if (limit < _REGION2_SIZE) {
  58		limit = _REGION2_SIZE - 1;
  59		atype = _ASCE_TYPE_REGION3;
  60		etype = _REGION3_ENTRY_EMPTY;
  61	} else if (limit < _REGION1_SIZE) {
  62		limit = _REGION1_SIZE - 1;
  63		atype = _ASCE_TYPE_REGION2;
  64		etype = _REGION2_ENTRY_EMPTY;
  65	} else {
  66		limit = -1UL;
  67		atype = _ASCE_TYPE_REGION1;
  68		etype = _REGION1_ENTRY_EMPTY;
  69	}
  70	gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL_ACCOUNT);
  71	if (!gmap)
  72		goto out;
  73	INIT_LIST_HEAD(&gmap->crst_list);
  74	INIT_LIST_HEAD(&gmap->children);
  75	INIT_LIST_HEAD(&gmap->pt_list);
  76	INIT_RADIX_TREE(&gmap->guest_to_host, GFP_KERNEL_ACCOUNT);
  77	INIT_RADIX_TREE(&gmap->host_to_guest, GFP_ATOMIC | __GFP_ACCOUNT);
  78	INIT_RADIX_TREE(&gmap->host_to_rmap, GFP_ATOMIC | __GFP_ACCOUNT);
  79	spin_lock_init(&gmap->guest_table_lock);
  80	spin_lock_init(&gmap->shadow_lock);
  81	refcount_set(&gmap->ref_count, 1);
  82	page = gmap_alloc_crst();
  83	if (!page)
  84		goto out_free;
  85	page->index = 0;
  86	list_add(&page->lru, &gmap->crst_list);
  87	table = page_to_virt(page);
  88	crst_table_init(table, etype);
  89	gmap->table = table;
  90	gmap->asce = atype | _ASCE_TABLE_LENGTH |
  91		_ASCE_USER_BITS | __pa(table);
  92	gmap->asce_end = limit;
  93	return gmap;
  94
  95out_free:
  96	kfree(gmap);
  97out:
  98	return NULL;
  99}
 100
 101/**
 102 * gmap_create - create a guest address space
 103 * @mm: pointer to the parent mm_struct
 104 * @limit: maximum size of the gmap address space
 105 *
 106 * Returns a guest address space structure.
 107 */
 108struct gmap *gmap_create(struct mm_struct *mm, unsigned long limit)
 109{
 110	struct gmap *gmap;
 111	unsigned long gmap_asce;
 112
 113	gmap = gmap_alloc(limit);
 114	if (!gmap)
 115		return NULL;
 116	gmap->mm = mm;
 117	spin_lock(&mm->context.lock);
 118	list_add_rcu(&gmap->list, &mm->context.gmap_list);
 119	if (list_is_singular(&mm->context.gmap_list))
 120		gmap_asce = gmap->asce;
 121	else
 122		gmap_asce = -1UL;
 123	WRITE_ONCE(mm->context.gmap_asce, gmap_asce);
 124	spin_unlock(&mm->context.lock);
 125	return gmap;
 126}
 127EXPORT_SYMBOL_GPL(gmap_create);
 128
 129static void gmap_flush_tlb(struct gmap *gmap)
 130{
 131	if (MACHINE_HAS_IDTE)
 132		__tlb_flush_idte(gmap->asce);
 133	else
 134		__tlb_flush_global();
 135}
 136
 137static void gmap_radix_tree_free(struct radix_tree_root *root)
 138{
 139	struct radix_tree_iter iter;
 140	unsigned long indices[16];
 141	unsigned long index;
 142	void __rcu **slot;
 143	int i, nr;
 144
 145	/* A radix tree is freed by deleting all of its entries */
 146	index = 0;
 147	do {
 148		nr = 0;
 149		radix_tree_for_each_slot(slot, root, &iter, index) {
 150			indices[nr] = iter.index;
 151			if (++nr == 16)
 152				break;
 153		}
 154		for (i = 0; i < nr; i++) {
 155			index = indices[i];
 156			radix_tree_delete(root, index);
 157		}
 158	} while (nr > 0);
 159}
 160
 161static void gmap_rmap_radix_tree_free(struct radix_tree_root *root)
 162{
 163	struct gmap_rmap *rmap, *rnext, *head;
 164	struct radix_tree_iter iter;
 165	unsigned long indices[16];
 166	unsigned long index;
 167	void __rcu **slot;
 168	int i, nr;
 169
 170	/* A radix tree is freed by deleting all of its entries */
 171	index = 0;
 172	do {
 173		nr = 0;
 174		radix_tree_for_each_slot(slot, root, &iter, index) {
 175			indices[nr] = iter.index;
 176			if (++nr == 16)
 177				break;
 178		}
 179		for (i = 0; i < nr; i++) {
 180			index = indices[i];
 181			head = radix_tree_delete(root, index);
 182			gmap_for_each_rmap_safe(rmap, rnext, head)
 183				kfree(rmap);
 184		}
 185	} while (nr > 0);
 186}
 187
 188/**
 189 * gmap_free - free a guest address space
 190 * @gmap: pointer to the guest address space structure
 191 *
 192 * No locks required. There are no references to this gmap anymore.
 193 */
 194static void gmap_free(struct gmap *gmap)
 195{
 196	struct page *page, *next;
 197
 198	/* Flush tlb of all gmaps (if not already done for shadows) */
 199	if (!(gmap_is_shadow(gmap) && gmap->removed))
 200		gmap_flush_tlb(gmap);
 201	/* Free all segment & region tables. */
 202	list_for_each_entry_safe(page, next, &gmap->crst_list, lru)
 203		__free_pages(page, CRST_ALLOC_ORDER);
 204	gmap_radix_tree_free(&gmap->guest_to_host);
 205	gmap_radix_tree_free(&gmap->host_to_guest);
 206
 207	/* Free additional data for a shadow gmap */
 208	if (gmap_is_shadow(gmap)) {
 209		struct ptdesc *ptdesc, *n;
 210
 211		/* Free all page tables. */
 212		list_for_each_entry_safe(ptdesc, n, &gmap->pt_list, pt_list)
 213			page_table_free_pgste(ptdesc);
 214		gmap_rmap_radix_tree_free(&gmap->host_to_rmap);
 215		/* Release reference to the parent */
 216		gmap_put(gmap->parent);
 217	}
 218
 219	kfree(gmap);
 220}
 221
 222/**
 223 * gmap_get - increase reference counter for guest address space
 224 * @gmap: pointer to the guest address space structure
 225 *
 226 * Returns the gmap pointer
 227 */
 228struct gmap *gmap_get(struct gmap *gmap)
 229{
 230	refcount_inc(&gmap->ref_count);
 231	return gmap;
 232}
 233EXPORT_SYMBOL_GPL(gmap_get);
 234
 235/**
 236 * gmap_put - decrease reference counter for guest address space
 237 * @gmap: pointer to the guest address space structure
 238 *
 239 * If the reference counter reaches zero the guest address space is freed.
 240 */
 241void gmap_put(struct gmap *gmap)
 242{
 243	if (refcount_dec_and_test(&gmap->ref_count))
 244		gmap_free(gmap);
 245}
 246EXPORT_SYMBOL_GPL(gmap_put);
 247
 248/**
 249 * gmap_remove - remove a guest address space but do not free it yet
 250 * @gmap: pointer to the guest address space structure
 251 */
 252void gmap_remove(struct gmap *gmap)
 253{
 254	struct gmap *sg, *next;
 255	unsigned long gmap_asce;
 256
 257	/* Remove all shadow gmaps linked to this gmap */
 258	if (!list_empty(&gmap->children)) {
 259		spin_lock(&gmap->shadow_lock);
 260		list_for_each_entry_safe(sg, next, &gmap->children, list) {
 261			list_del(&sg->list);
 262			gmap_put(sg);
 263		}
 264		spin_unlock(&gmap->shadow_lock);
 265	}
 266	/* Remove gmap from the pre-mm list */
 267	spin_lock(&gmap->mm->context.lock);
 268	list_del_rcu(&gmap->list);
 269	if (list_empty(&gmap->mm->context.gmap_list))
 270		gmap_asce = 0;
 271	else if (list_is_singular(&gmap->mm->context.gmap_list))
 272		gmap_asce = list_first_entry(&gmap->mm->context.gmap_list,
 273					     struct gmap, list)->asce;
 274	else
 275		gmap_asce = -1UL;
 276	WRITE_ONCE(gmap->mm->context.gmap_asce, gmap_asce);
 277	spin_unlock(&gmap->mm->context.lock);
 278	synchronize_rcu();
 279	/* Put reference */
 280	gmap_put(gmap);
 281}
 282EXPORT_SYMBOL_GPL(gmap_remove);
 283
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 284/*
 285 * gmap_alloc_table is assumed to be called with mmap_lock held
 286 */
 287static int gmap_alloc_table(struct gmap *gmap, unsigned long *table,
 288			    unsigned long init, unsigned long gaddr)
 289{
 290	struct page *page;
 291	unsigned long *new;
 292
 293	/* since we dont free the gmap table until gmap_free we can unlock */
 294	page = gmap_alloc_crst();
 295	if (!page)
 296		return -ENOMEM;
 297	new = page_to_virt(page);
 298	crst_table_init(new, init);
 299	spin_lock(&gmap->guest_table_lock);
 300	if (*table & _REGION_ENTRY_INVALID) {
 301		list_add(&page->lru, &gmap->crst_list);
 302		*table = __pa(new) | _REGION_ENTRY_LENGTH |
 303			(*table & _REGION_ENTRY_TYPE_MASK);
 304		page->index = gaddr;
 305		page = NULL;
 306	}
 307	spin_unlock(&gmap->guest_table_lock);
 308	if (page)
 309		__free_pages(page, CRST_ALLOC_ORDER);
 310	return 0;
 311}
 312
 313/**
 314 * __gmap_segment_gaddr - find virtual address from segment pointer
 315 * @entry: pointer to a segment table entry in the guest address space
 316 *
 317 * Returns the virtual address in the guest address space for the segment
 318 */
 319static unsigned long __gmap_segment_gaddr(unsigned long *entry)
 320{
 321	struct page *page;
 322	unsigned long offset;
 323
 324	offset = (unsigned long) entry / sizeof(unsigned long);
 325	offset = (offset & (PTRS_PER_PMD - 1)) * PMD_SIZE;
 326	page = pmd_pgtable_page((pmd_t *) entry);
 327	return page->index + offset;
 328}
 329
 330/**
 331 * __gmap_unlink_by_vmaddr - unlink a single segment via a host address
 332 * @gmap: pointer to the guest address space structure
 333 * @vmaddr: address in the host process address space
 334 *
 335 * Returns 1 if a TLB flush is required
 336 */
 337static int __gmap_unlink_by_vmaddr(struct gmap *gmap, unsigned long vmaddr)
 338{
 339	unsigned long *entry;
 340	int flush = 0;
 341
 342	BUG_ON(gmap_is_shadow(gmap));
 343	spin_lock(&gmap->guest_table_lock);
 344	entry = radix_tree_delete(&gmap->host_to_guest, vmaddr >> PMD_SHIFT);
 345	if (entry) {
 346		flush = (*entry != _SEGMENT_ENTRY_EMPTY);
 347		*entry = _SEGMENT_ENTRY_EMPTY;
 348	}
 349	spin_unlock(&gmap->guest_table_lock);
 350	return flush;
 351}
 352
 353/**
 354 * __gmap_unmap_by_gaddr - unmap a single segment via a guest address
 355 * @gmap: pointer to the guest address space structure
 356 * @gaddr: address in the guest address space
 357 *
 358 * Returns 1 if a TLB flush is required
 359 */
 360static int __gmap_unmap_by_gaddr(struct gmap *gmap, unsigned long gaddr)
 361{
 362	unsigned long vmaddr;
 363
 364	vmaddr = (unsigned long) radix_tree_delete(&gmap->guest_to_host,
 365						   gaddr >> PMD_SHIFT);
 366	return vmaddr ? __gmap_unlink_by_vmaddr(gmap, vmaddr) : 0;
 367}
 368
 369/**
 370 * gmap_unmap_segment - unmap segment from the guest address space
 371 * @gmap: pointer to the guest address space structure
 372 * @to: address in the guest address space
 373 * @len: length of the memory area to unmap
 374 *
 375 * Returns 0 if the unmap succeeded, -EINVAL if not.
 376 */
 377int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
 378{
 379	unsigned long off;
 380	int flush;
 381
 382	BUG_ON(gmap_is_shadow(gmap));
 383	if ((to | len) & (PMD_SIZE - 1))
 384		return -EINVAL;
 385	if (len == 0 || to + len < to)
 386		return -EINVAL;
 387
 388	flush = 0;
 389	mmap_write_lock(gmap->mm);
 390	for (off = 0; off < len; off += PMD_SIZE)
 391		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
 392	mmap_write_unlock(gmap->mm);
 393	if (flush)
 394		gmap_flush_tlb(gmap);
 395	return 0;
 396}
 397EXPORT_SYMBOL_GPL(gmap_unmap_segment);
 398
 399/**
 400 * gmap_map_segment - map a segment to the guest address space
 401 * @gmap: pointer to the guest address space structure
 402 * @from: source address in the parent address space
 403 * @to: target address in the guest address space
 404 * @len: length of the memory area to map
 405 *
 406 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not.
 407 */
 408int gmap_map_segment(struct gmap *gmap, unsigned long from,
 409		     unsigned long to, unsigned long len)
 410{
 411	unsigned long off;
 412	int flush;
 413
 414	BUG_ON(gmap_is_shadow(gmap));
 415	if ((from | to | len) & (PMD_SIZE - 1))
 416		return -EINVAL;
 417	if (len == 0 || from + len < from || to + len < to ||
 418	    from + len - 1 > TASK_SIZE_MAX || to + len - 1 > gmap->asce_end)
 419		return -EINVAL;
 420
 421	flush = 0;
 422	mmap_write_lock(gmap->mm);
 423	for (off = 0; off < len; off += PMD_SIZE) {
 424		/* Remove old translation */
 425		flush |= __gmap_unmap_by_gaddr(gmap, to + off);
 426		/* Store new translation */
 427		if (radix_tree_insert(&gmap->guest_to_host,
 428				      (to + off) >> PMD_SHIFT,
 429				      (void *) from + off))
 430			break;
 431	}
 432	mmap_write_unlock(gmap->mm);
 433	if (flush)
 434		gmap_flush_tlb(gmap);
 435	if (off >= len)
 436		return 0;
 437	gmap_unmap_segment(gmap, to, len);
 438	return -ENOMEM;
 439}
 440EXPORT_SYMBOL_GPL(gmap_map_segment);
 441
 442/**
 443 * __gmap_translate - translate a guest address to a user space address
 444 * @gmap: pointer to guest mapping meta data structure
 445 * @gaddr: guest address
 446 *
 447 * Returns user space address which corresponds to the guest address or
 448 * -EFAULT if no such mapping exists.
 449 * This function does not establish potentially missing page table entries.
 450 * The mmap_lock of the mm that belongs to the address space must be held
 451 * when this function gets called.
 452 *
 453 * Note: Can also be called for shadow gmaps.
 454 */
 455unsigned long __gmap_translate(struct gmap *gmap, unsigned long gaddr)
 456{
 457	unsigned long vmaddr;
 458
 459	vmaddr = (unsigned long)
 460		radix_tree_lookup(&gmap->guest_to_host, gaddr >> PMD_SHIFT);
 461	/* Note: guest_to_host is empty for a shadow gmap */
 462	return vmaddr ? (vmaddr | (gaddr & ~PMD_MASK)) : -EFAULT;
 463}
 464EXPORT_SYMBOL_GPL(__gmap_translate);
 465
 466/**
 467 * gmap_translate - translate a guest address to a user space address
 468 * @gmap: pointer to guest mapping meta data structure
 469 * @gaddr: guest address
 470 *
 471 * Returns user space address which corresponds to the guest address or
 472 * -EFAULT if no such mapping exists.
 473 * This function does not establish potentially missing page table entries.
 474 */
 475unsigned long gmap_translate(struct gmap *gmap, unsigned long gaddr)
 476{
 477	unsigned long rc;
 478
 479	mmap_read_lock(gmap->mm);
 480	rc = __gmap_translate(gmap, gaddr);
 481	mmap_read_unlock(gmap->mm);
 482	return rc;
 483}
 484EXPORT_SYMBOL_GPL(gmap_translate);
 485
 486/**
 487 * gmap_unlink - disconnect a page table from the gmap shadow tables
 488 * @mm: pointer to the parent mm_struct
 489 * @table: pointer to the host page table
 490 * @vmaddr: vm address associated with the host page table
 491 */
 492void gmap_unlink(struct mm_struct *mm, unsigned long *table,
 493		 unsigned long vmaddr)
 494{
 495	struct gmap *gmap;
 496	int flush;
 497
 498	rcu_read_lock();
 499	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
 500		flush = __gmap_unlink_by_vmaddr(gmap, vmaddr);
 501		if (flush)
 502			gmap_flush_tlb(gmap);
 503	}
 504	rcu_read_unlock();
 505}
 506
 507static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *old, pmd_t new,
 508			   unsigned long gaddr);
 509
 510/**
 511 * __gmap_link - set up shadow page tables to connect a host to a guest address
 512 * @gmap: pointer to guest mapping meta data structure
 513 * @gaddr: guest address
 514 * @vmaddr: vm address
 515 *
 516 * Returns 0 on success, -ENOMEM for out of memory conditions, and -EFAULT
 517 * if the vm address is already mapped to a different guest segment.
 518 * The mmap_lock of the mm that belongs to the address space must be held
 519 * when this function gets called.
 520 */
 521int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
 522{
 523	struct mm_struct *mm;
 524	unsigned long *table;
 525	spinlock_t *ptl;
 526	pgd_t *pgd;
 527	p4d_t *p4d;
 528	pud_t *pud;
 529	pmd_t *pmd;
 530	u64 unprot;
 531	int rc;
 532
 533	BUG_ON(gmap_is_shadow(gmap));
 534	/* Create higher level tables in the gmap page table */
 535	table = gmap->table;
 536	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION1) {
 537		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
 538		if ((*table & _REGION_ENTRY_INVALID) &&
 539		    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY,
 540				     gaddr & _REGION1_MASK))
 541			return -ENOMEM;
 542		table = __va(*table & _REGION_ENTRY_ORIGIN);
 543	}
 544	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION2) {
 545		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
 546		if ((*table & _REGION_ENTRY_INVALID) &&
 547		    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY,
 548				     gaddr & _REGION2_MASK))
 549			return -ENOMEM;
 550		table = __va(*table & _REGION_ENTRY_ORIGIN);
 551	}
 552	if ((gmap->asce & _ASCE_TYPE_MASK) >= _ASCE_TYPE_REGION3) {
 553		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
 554		if ((*table & _REGION_ENTRY_INVALID) &&
 555		    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY,
 556				     gaddr & _REGION3_MASK))
 557			return -ENOMEM;
 558		table = __va(*table & _REGION_ENTRY_ORIGIN);
 559	}
 560	table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 561	/* Walk the parent mm page table */
 562	mm = gmap->mm;
 563	pgd = pgd_offset(mm, vmaddr);
 564	VM_BUG_ON(pgd_none(*pgd));
 565	p4d = p4d_offset(pgd, vmaddr);
 566	VM_BUG_ON(p4d_none(*p4d));
 567	pud = pud_offset(p4d, vmaddr);
 568	VM_BUG_ON(pud_none(*pud));
 569	/* large puds cannot yet be handled */
 570	if (pud_leaf(*pud))
 571		return -EFAULT;
 572	pmd = pmd_offset(pud, vmaddr);
 573	VM_BUG_ON(pmd_none(*pmd));
 574	/* Are we allowed to use huge pages? */
 575	if (pmd_leaf(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
 576		return -EFAULT;
 577	/* Link gmap segment table entry location to page table. */
 578	rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
 579	if (rc)
 580		return rc;
 581	ptl = pmd_lock(mm, pmd);
 582	spin_lock(&gmap->guest_table_lock);
 583	if (*table == _SEGMENT_ENTRY_EMPTY) {
 584		rc = radix_tree_insert(&gmap->host_to_guest,
 585				       vmaddr >> PMD_SHIFT, table);
 586		if (!rc) {
 587			if (pmd_leaf(*pmd)) {
 588				*table = (pmd_val(*pmd) &
 589					  _SEGMENT_ENTRY_HARDWARE_BITS_LARGE)
 590					| _SEGMENT_ENTRY_GMAP_UC
 591					| _SEGMENT_ENTRY;
 592			} else
 593				*table = pmd_val(*pmd) &
 594					_SEGMENT_ENTRY_HARDWARE_BITS;
 595		}
 596	} else if (*table & _SEGMENT_ENTRY_PROTECT &&
 597		   !(pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT)) {
 598		unprot = (u64)*table;
 599		unprot &= ~_SEGMENT_ENTRY_PROTECT;
 600		unprot |= _SEGMENT_ENTRY_GMAP_UC;
 601		gmap_pmdp_xchg(gmap, (pmd_t *)table, __pmd(unprot), gaddr);
 602	}
 603	spin_unlock(&gmap->guest_table_lock);
 604	spin_unlock(ptl);
 605	radix_tree_preload_end();
 606	return rc;
 607}
 608
 609/**
 610 * fixup_user_fault_nowait - manually resolve a user page fault without waiting
 611 * @mm:		mm_struct of target mm
 612 * @address:	user address
 613 * @fault_flags:flags to pass down to handle_mm_fault()
 614 * @unlocked:	did we unlock the mmap_lock while retrying
 615 *
 616 * This function behaves similarly to fixup_user_fault(), but it guarantees
 617 * that the fault will be resolved without waiting. The function might drop
 618 * and re-acquire the mm lock, in which case @unlocked will be set to true.
 619 *
 620 * The guarantee is that the fault is handled without waiting, but the
 621 * function itself might sleep, due to the lock.
 622 *
 623 * Context: Needs to be called with mm->mmap_lock held in read mode, and will
 624 * return with the lock held in read mode; @unlocked will indicate whether
 625 * the lock has been dropped and re-acquired. This is the same behaviour as
 626 * fixup_user_fault().
 627 *
 628 * Return: 0 on success, -EAGAIN if the fault cannot be resolved without
 629 * waiting, -EFAULT if the fault cannot be resolved, -ENOMEM if out of
 630 * memory.
 631 */
 632static int fixup_user_fault_nowait(struct mm_struct *mm, unsigned long address,
 633				   unsigned int fault_flags, bool *unlocked)
 634{
 635	struct vm_area_struct *vma;
 636	unsigned int test_flags;
 637	vm_fault_t fault;
 638	int rc;
 639
 640	fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
 641	test_flags = fault_flags & FAULT_FLAG_WRITE ? VM_WRITE : VM_READ;
 642
 643	vma = find_vma(mm, address);
 644	if (unlikely(!vma || address < vma->vm_start))
 645		return -EFAULT;
 646	if (unlikely(!(vma->vm_flags & test_flags)))
 647		return -EFAULT;
 648
 649	fault = handle_mm_fault(vma, address, fault_flags, NULL);
 650	/* the mm lock has been dropped, take it again */
 651	if (fault & VM_FAULT_COMPLETED) {
 652		*unlocked = true;
 653		mmap_read_lock(mm);
 654		return 0;
 655	}
 656	/* the mm lock has not been dropped */
 657	if (fault & VM_FAULT_ERROR) {
 658		rc = vm_fault_to_errno(fault, 0);
 659		BUG_ON(!rc);
 660		return rc;
 661	}
 662	/* the mm lock has not been dropped because of FAULT_FLAG_RETRY_NOWAIT */
 663	if (fault & VM_FAULT_RETRY)
 664		return -EAGAIN;
 665	/* nothing needed to be done and the mm lock has not been dropped */
 666	return 0;
 667}
 668
 669/**
 670 * __gmap_fault - resolve a fault on a guest address
 671 * @gmap: pointer to guest mapping meta data structure
 672 * @gaddr: guest address
 673 * @fault_flags: flags to pass down to handle_mm_fault()
 674 *
 675 * Context: Needs to be called with mm->mmap_lock held in read mode. Might
 676 * drop and re-acquire the lock. Will always return with the lock held.
 677 */
 678static int __gmap_fault(struct gmap *gmap, unsigned long gaddr, unsigned int fault_flags)
 
 679{
 680	unsigned long vmaddr;
 
 681	bool unlocked;
 682	int rc = 0;
 
 683
 684retry:
 685	unlocked = false;
 686
 687	vmaddr = __gmap_translate(gmap, gaddr);
 688	if (IS_ERR_VALUE(vmaddr))
 689		return vmaddr;
 690
 691	if (fault_flags & FAULT_FLAG_RETRY_NOWAIT)
 692		rc = fixup_user_fault_nowait(gmap->mm, vmaddr, fault_flags, &unlocked);
 693	else
 694		rc = fixup_user_fault(gmap->mm, vmaddr, fault_flags, &unlocked);
 695	if (rc)
 696		return rc;
 697	/*
 698	 * In the case that fixup_user_fault unlocked the mmap_lock during
 699	 * fault-in, redo __gmap_translate() to avoid racing with a
 700	 * map/unmap_segment.
 701	 * In particular, __gmap_translate(), fixup_user_fault{,_nowait}(),
 702	 * and __gmap_link() must all be called atomically in one go; if the
 703	 * lock had been dropped in between, a retry is needed.
 704	 */
 705	if (unlocked)
 706		goto retry;
 707
 708	return __gmap_link(gmap, gaddr, vmaddr);
 709}
 710
 711/**
 712 * gmap_fault - resolve a fault on a guest address
 713 * @gmap: pointer to guest mapping meta data structure
 714 * @gaddr: guest address
 715 * @fault_flags: flags to pass down to handle_mm_fault()
 716 *
 717 * Returns 0 on success, -ENOMEM for out of memory conditions, -EFAULT if the
 718 * vm address is already mapped to a different guest segment, and -EAGAIN if
 719 * FAULT_FLAG_RETRY_NOWAIT was specified and the fault could not be processed
 720 * immediately.
 721 */
 722int gmap_fault(struct gmap *gmap, unsigned long gaddr, unsigned int fault_flags)
 723{
 724	int rc;
 725
 726	mmap_read_lock(gmap->mm);
 727	rc = __gmap_fault(gmap, gaddr, fault_flags);
 728	mmap_read_unlock(gmap->mm);
 729	return rc;
 730}
 731EXPORT_SYMBOL_GPL(gmap_fault);
 732
 733/*
 734 * this function is assumed to be called with mmap_lock held
 735 */
 736void __gmap_zap(struct gmap *gmap, unsigned long gaddr)
 737{
 738	struct vm_area_struct *vma;
 739	unsigned long vmaddr;
 740	spinlock_t *ptl;
 741	pte_t *ptep;
 742
 743	/* Find the vm address for the guest address */
 744	vmaddr = (unsigned long) radix_tree_lookup(&gmap->guest_to_host,
 745						   gaddr >> PMD_SHIFT);
 746	if (vmaddr) {
 747		vmaddr |= gaddr & ~PMD_MASK;
 748
 749		vma = vma_lookup(gmap->mm, vmaddr);
 750		if (!vma || is_vm_hugetlb_page(vma))
 751			return;
 752
 753		/* Get pointer to the page table entry */
 754		ptep = get_locked_pte(gmap->mm, vmaddr, &ptl);
 755		if (likely(ptep)) {
 756			ptep_zap_unused(gmap->mm, vmaddr, ptep, 0);
 757			pte_unmap_unlock(ptep, ptl);
 758		}
 759	}
 760}
 761EXPORT_SYMBOL_GPL(__gmap_zap);
 762
 763void gmap_discard(struct gmap *gmap, unsigned long from, unsigned long to)
 764{
 765	unsigned long gaddr, vmaddr, size;
 766	struct vm_area_struct *vma;
 767
 768	mmap_read_lock(gmap->mm);
 769	for (gaddr = from; gaddr < to;
 770	     gaddr = (gaddr + PMD_SIZE) & PMD_MASK) {
 771		/* Find the vm address for the guest address */
 772		vmaddr = (unsigned long)
 773			radix_tree_lookup(&gmap->guest_to_host,
 774					  gaddr >> PMD_SHIFT);
 775		if (!vmaddr)
 776			continue;
 777		vmaddr |= gaddr & ~PMD_MASK;
 778		/* Find vma in the parent mm */
 779		vma = find_vma(gmap->mm, vmaddr);
 780		if (!vma)
 781			continue;
 782		/*
 783		 * We do not discard pages that are backed by
 784		 * hugetlbfs, so we don't have to refault them.
 785		 */
 786		if (is_vm_hugetlb_page(vma))
 787			continue;
 788		size = min(to - gaddr, PMD_SIZE - (gaddr & ~PMD_MASK));
 789		zap_page_range_single(vma, vmaddr, size, NULL);
 790	}
 791	mmap_read_unlock(gmap->mm);
 792}
 793EXPORT_SYMBOL_GPL(gmap_discard);
 794
 795static LIST_HEAD(gmap_notifier_list);
 796static DEFINE_SPINLOCK(gmap_notifier_lock);
 797
 798/**
 799 * gmap_register_pte_notifier - register a pte invalidation callback
 800 * @nb: pointer to the gmap notifier block
 801 */
 802void gmap_register_pte_notifier(struct gmap_notifier *nb)
 803{
 804	spin_lock(&gmap_notifier_lock);
 805	list_add_rcu(&nb->list, &gmap_notifier_list);
 806	spin_unlock(&gmap_notifier_lock);
 807}
 808EXPORT_SYMBOL_GPL(gmap_register_pte_notifier);
 809
 810/**
 811 * gmap_unregister_pte_notifier - remove a pte invalidation callback
 812 * @nb: pointer to the gmap notifier block
 813 */
 814void gmap_unregister_pte_notifier(struct gmap_notifier *nb)
 815{
 816	spin_lock(&gmap_notifier_lock);
 817	list_del_rcu(&nb->list);
 818	spin_unlock(&gmap_notifier_lock);
 819	synchronize_rcu();
 820}
 821EXPORT_SYMBOL_GPL(gmap_unregister_pte_notifier);
 822
 823/**
 824 * gmap_call_notifier - call all registered invalidation callbacks
 825 * @gmap: pointer to guest mapping meta data structure
 826 * @start: start virtual address in the guest address space
 827 * @end: end virtual address in the guest address space
 828 */
 829static void gmap_call_notifier(struct gmap *gmap, unsigned long start,
 830			       unsigned long end)
 831{
 832	struct gmap_notifier *nb;
 833
 834	list_for_each_entry(nb, &gmap_notifier_list, list)
 835		nb->notifier_call(gmap, start, end);
 836}
 837
 838/**
 839 * gmap_table_walk - walk the gmap page tables
 840 * @gmap: pointer to guest mapping meta data structure
 841 * @gaddr: virtual address in the guest address space
 842 * @level: page table level to stop at
 843 *
 844 * Returns a table entry pointer for the given guest address and @level
 845 * @level=0 : returns a pointer to a page table table entry (or NULL)
 846 * @level=1 : returns a pointer to a segment table entry (or NULL)
 847 * @level=2 : returns a pointer to a region-3 table entry (or NULL)
 848 * @level=3 : returns a pointer to a region-2 table entry (or NULL)
 849 * @level=4 : returns a pointer to a region-1 table entry (or NULL)
 850 *
 851 * Returns NULL if the gmap page tables could not be walked to the
 852 * requested level.
 853 *
 854 * Note: Can also be called for shadow gmaps.
 855 */
 856static inline unsigned long *gmap_table_walk(struct gmap *gmap,
 857					     unsigned long gaddr, int level)
 858{
 859	const int asce_type = gmap->asce & _ASCE_TYPE_MASK;
 860	unsigned long *table = gmap->table;
 861
 862	if (gmap_is_shadow(gmap) && gmap->removed)
 863		return NULL;
 864
 865	if (WARN_ON_ONCE(level > (asce_type >> 2) + 1))
 866		return NULL;
 867
 868	if (asce_type != _ASCE_TYPE_REGION1 &&
 869	    gaddr & (-1UL << (31 + (asce_type >> 2) * 11)))
 870		return NULL;
 871
 872	switch (asce_type) {
 873	case _ASCE_TYPE_REGION1:
 874		table += (gaddr & _REGION1_INDEX) >> _REGION1_SHIFT;
 875		if (level == 4)
 876			break;
 877		if (*table & _REGION_ENTRY_INVALID)
 878			return NULL;
 879		table = __va(*table & _REGION_ENTRY_ORIGIN);
 880		fallthrough;
 881	case _ASCE_TYPE_REGION2:
 882		table += (gaddr & _REGION2_INDEX) >> _REGION2_SHIFT;
 883		if (level == 3)
 884			break;
 885		if (*table & _REGION_ENTRY_INVALID)
 886			return NULL;
 887		table = __va(*table & _REGION_ENTRY_ORIGIN);
 888		fallthrough;
 889	case _ASCE_TYPE_REGION3:
 890		table += (gaddr & _REGION3_INDEX) >> _REGION3_SHIFT;
 891		if (level == 2)
 892			break;
 893		if (*table & _REGION_ENTRY_INVALID)
 894			return NULL;
 895		table = __va(*table & _REGION_ENTRY_ORIGIN);
 896		fallthrough;
 897	case _ASCE_TYPE_SEGMENT:
 898		table += (gaddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT;
 899		if (level == 1)
 900			break;
 901		if (*table & _REGION_ENTRY_INVALID)
 902			return NULL;
 903		table = __va(*table & _SEGMENT_ENTRY_ORIGIN);
 904		table += (gaddr & _PAGE_INDEX) >> PAGE_SHIFT;
 905	}
 906	return table;
 907}
 908
 909/**
 910 * gmap_pte_op_walk - walk the gmap page table, get the page table lock
 911 *		      and return the pte pointer
 912 * @gmap: pointer to guest mapping meta data structure
 913 * @gaddr: virtual address in the guest address space
 914 * @ptl: pointer to the spinlock pointer
 915 *
 916 * Returns a pointer to the locked pte for a guest address, or NULL
 917 */
 918static pte_t *gmap_pte_op_walk(struct gmap *gmap, unsigned long gaddr,
 919			       spinlock_t **ptl)
 920{
 921	unsigned long *table;
 922
 923	BUG_ON(gmap_is_shadow(gmap));
 924	/* Walk the gmap page table, lock and get pte pointer */
 925	table = gmap_table_walk(gmap, gaddr, 1); /* get segment pointer */
 926	if (!table || *table & _SEGMENT_ENTRY_INVALID)
 927		return NULL;
 928	return pte_alloc_map_lock(gmap->mm, (pmd_t *) table, gaddr, ptl);
 929}
 930
 931/**
 932 * gmap_pte_op_fixup - force a page in and connect the gmap page table
 933 * @gmap: pointer to guest mapping meta data structure
 934 * @gaddr: virtual address in the guest address space
 935 * @vmaddr: address in the host process address space
 936 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
 937 *
 938 * Returns 0 if the caller can retry __gmap_translate (might fail again),
 939 * -ENOMEM if out of memory and -EFAULT if anything goes wrong while fixing
 940 * up or connecting the gmap page table.
 941 */
 942static int gmap_pte_op_fixup(struct gmap *gmap, unsigned long gaddr,
 943			     unsigned long vmaddr, int prot)
 944{
 945	struct mm_struct *mm = gmap->mm;
 946	unsigned int fault_flags;
 947	bool unlocked = false;
 948
 949	BUG_ON(gmap_is_shadow(gmap));
 950	fault_flags = (prot == PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
 951	if (fixup_user_fault(mm, vmaddr, fault_flags, &unlocked))
 952		return -EFAULT;
 953	if (unlocked)
 954		/* lost mmap_lock, caller has to retry __gmap_translate */
 955		return 0;
 956	/* Connect the page tables */
 957	return __gmap_link(gmap, gaddr, vmaddr);
 958}
 959
 960/**
 961 * gmap_pte_op_end - release the page table lock
 962 * @ptep: pointer to the locked pte
 963 * @ptl: pointer to the page table spinlock
 964 */
 965static void gmap_pte_op_end(pte_t *ptep, spinlock_t *ptl)
 966{
 967	pte_unmap_unlock(ptep, ptl);
 
 968}
 969
 970/**
 971 * gmap_pmd_op_walk - walk the gmap tables, get the guest table lock
 972 *		      and return the pmd pointer
 973 * @gmap: pointer to guest mapping meta data structure
 974 * @gaddr: virtual address in the guest address space
 975 *
 976 * Returns a pointer to the pmd for a guest address, or NULL
 977 */
 978static inline pmd_t *gmap_pmd_op_walk(struct gmap *gmap, unsigned long gaddr)
 979{
 980	pmd_t *pmdp;
 981
 982	BUG_ON(gmap_is_shadow(gmap));
 983	pmdp = (pmd_t *) gmap_table_walk(gmap, gaddr, 1);
 984	if (!pmdp)
 985		return NULL;
 986
 987	/* without huge pages, there is no need to take the table lock */
 988	if (!gmap->mm->context.allow_gmap_hpage_1m)
 989		return pmd_none(*pmdp) ? NULL : pmdp;
 990
 991	spin_lock(&gmap->guest_table_lock);
 992	if (pmd_none(*pmdp)) {
 993		spin_unlock(&gmap->guest_table_lock);
 994		return NULL;
 995	}
 996
 997	/* 4k page table entries are locked via the pte (pte_alloc_map_lock). */
 998	if (!pmd_leaf(*pmdp))
 999		spin_unlock(&gmap->guest_table_lock);
1000	return pmdp;
1001}
1002
1003/**
1004 * gmap_pmd_op_end - release the guest_table_lock if needed
1005 * @gmap: pointer to the guest mapping meta data structure
1006 * @pmdp: pointer to the pmd
1007 */
1008static inline void gmap_pmd_op_end(struct gmap *gmap, pmd_t *pmdp)
1009{
1010	if (pmd_leaf(*pmdp))
1011		spin_unlock(&gmap->guest_table_lock);
1012}
1013
1014/*
1015 * gmap_protect_pmd - remove access rights to memory and set pmd notification bits
1016 * @pmdp: pointer to the pmd to be protected
1017 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1018 * @bits: notification bits to set
1019 *
1020 * Returns:
1021 * 0 if successfully protected
1022 * -EAGAIN if a fixup is needed
1023 * -EINVAL if unsupported notifier bits have been specified
1024 *
1025 * Expected to be called with sg->mm->mmap_lock in read and
1026 * guest_table_lock held.
1027 */
1028static int gmap_protect_pmd(struct gmap *gmap, unsigned long gaddr,
1029			    pmd_t *pmdp, int prot, unsigned long bits)
1030{
1031	int pmd_i = pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID;
1032	int pmd_p = pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT;
1033	pmd_t new = *pmdp;
1034
1035	/* Fixup needed */
1036	if ((pmd_i && (prot != PROT_NONE)) || (pmd_p && (prot == PROT_WRITE)))
1037		return -EAGAIN;
1038
1039	if (prot == PROT_NONE && !pmd_i) {
1040		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
1041		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
1042	}
1043
1044	if (prot == PROT_READ && !pmd_p) {
1045		new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_INVALID));
1046		new = set_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_PROTECT));
1047		gmap_pmdp_xchg(gmap, pmdp, new, gaddr);
1048	}
1049
1050	if (bits & GMAP_NOTIFY_MPROT)
1051		set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
1052
1053	/* Shadow GMAP protection needs split PMDs */
1054	if (bits & GMAP_NOTIFY_SHADOW)
1055		return -EINVAL;
1056
1057	return 0;
1058}
1059
1060/*
1061 * gmap_protect_pte - remove access rights to memory and set pgste bits
1062 * @gmap: pointer to guest mapping meta data structure
1063 * @gaddr: virtual address in the guest address space
1064 * @pmdp: pointer to the pmd associated with the pte
1065 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1066 * @bits: notification bits to set
1067 *
1068 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1069 * -EAGAIN if a fixup is needed.
1070 *
1071 * Expected to be called with sg->mm->mmap_lock in read
1072 */
1073static int gmap_protect_pte(struct gmap *gmap, unsigned long gaddr,
1074			    pmd_t *pmdp, int prot, unsigned long bits)
1075{
1076	int rc;
1077	pte_t *ptep;
1078	spinlock_t *ptl;
1079	unsigned long pbits = 0;
1080
1081	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1082		return -EAGAIN;
1083
1084	ptep = pte_alloc_map_lock(gmap->mm, pmdp, gaddr, &ptl);
1085	if (!ptep)
1086		return -ENOMEM;
1087
1088	pbits |= (bits & GMAP_NOTIFY_MPROT) ? PGSTE_IN_BIT : 0;
1089	pbits |= (bits & GMAP_NOTIFY_SHADOW) ? PGSTE_VSIE_BIT : 0;
1090	/* Protect and unlock. */
1091	rc = ptep_force_prot(gmap->mm, gaddr, ptep, prot, pbits);
1092	gmap_pte_op_end(ptep, ptl);
1093	return rc;
1094}
1095
1096/*
1097 * gmap_protect_range - remove access rights to memory and set pgste bits
1098 * @gmap: pointer to guest mapping meta data structure
1099 * @gaddr: virtual address in the guest address space
1100 * @len: size of area
1101 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1102 * @bits: pgste notification bits to set
1103 *
1104 * Returns 0 if successfully protected, -ENOMEM if out of memory and
1105 * -EFAULT if gaddr is invalid (or mapping for shadows is missing).
1106 *
1107 * Called with sg->mm->mmap_lock in read.
1108 */
1109static int gmap_protect_range(struct gmap *gmap, unsigned long gaddr,
1110			      unsigned long len, int prot, unsigned long bits)
1111{
1112	unsigned long vmaddr, dist;
1113	pmd_t *pmdp;
1114	int rc;
1115
1116	BUG_ON(gmap_is_shadow(gmap));
1117	while (len) {
1118		rc = -EAGAIN;
1119		pmdp = gmap_pmd_op_walk(gmap, gaddr);
1120		if (pmdp) {
1121			if (!pmd_leaf(*pmdp)) {
1122				rc = gmap_protect_pte(gmap, gaddr, pmdp, prot,
1123						      bits);
1124				if (!rc) {
1125					len -= PAGE_SIZE;
1126					gaddr += PAGE_SIZE;
1127				}
1128			} else {
1129				rc = gmap_protect_pmd(gmap, gaddr, pmdp, prot,
1130						      bits);
1131				if (!rc) {
1132					dist = HPAGE_SIZE - (gaddr & ~HPAGE_MASK);
1133					len = len < dist ? 0 : len - dist;
1134					gaddr = (gaddr & HPAGE_MASK) + HPAGE_SIZE;
1135				}
1136			}
1137			gmap_pmd_op_end(gmap, pmdp);
1138		}
1139		if (rc) {
1140			if (rc == -EINVAL)
1141				return rc;
1142
1143			/* -EAGAIN, fixup of userspace mm and gmap */
1144			vmaddr = __gmap_translate(gmap, gaddr);
1145			if (IS_ERR_VALUE(vmaddr))
1146				return vmaddr;
1147			rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, prot);
1148			if (rc)
1149				return rc;
1150		}
1151	}
1152	return 0;
1153}
1154
1155/**
1156 * gmap_mprotect_notify - change access rights for a range of ptes and
1157 *                        call the notifier if any pte changes again
1158 * @gmap: pointer to guest mapping meta data structure
1159 * @gaddr: virtual address in the guest address space
1160 * @len: size of area
1161 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
1162 *
1163 * Returns 0 if for each page in the given range a gmap mapping exists,
1164 * the new access rights could be set and the notifier could be armed.
1165 * If the gmap mapping is missing for one or more pages -EFAULT is
1166 * returned. If no memory could be allocated -ENOMEM is returned.
1167 * This function establishes missing page table entries.
1168 */
1169int gmap_mprotect_notify(struct gmap *gmap, unsigned long gaddr,
1170			 unsigned long len, int prot)
1171{
1172	int rc;
1173
1174	if ((gaddr & ~PAGE_MASK) || (len & ~PAGE_MASK) || gmap_is_shadow(gmap))
1175		return -EINVAL;
1176	if (!MACHINE_HAS_ESOP && prot == PROT_READ)
1177		return -EINVAL;
1178	mmap_read_lock(gmap->mm);
1179	rc = gmap_protect_range(gmap, gaddr, len, prot, GMAP_NOTIFY_MPROT);
1180	mmap_read_unlock(gmap->mm);
1181	return rc;
1182}
1183EXPORT_SYMBOL_GPL(gmap_mprotect_notify);
1184
1185/**
1186 * gmap_read_table - get an unsigned long value from a guest page table using
1187 *                   absolute addressing, without marking the page referenced.
1188 * @gmap: pointer to guest mapping meta data structure
1189 * @gaddr: virtual address in the guest address space
1190 * @val: pointer to the unsigned long value to return
1191 *
1192 * Returns 0 if the value was read, -ENOMEM if out of memory and -EFAULT
1193 * if reading using the virtual address failed. -EINVAL if called on a gmap
1194 * shadow.
1195 *
1196 * Called with gmap->mm->mmap_lock in read.
1197 */
1198int gmap_read_table(struct gmap *gmap, unsigned long gaddr, unsigned long *val)
1199{
1200	unsigned long address, vmaddr;
1201	spinlock_t *ptl;
1202	pte_t *ptep, pte;
1203	int rc;
1204
1205	if (gmap_is_shadow(gmap))
1206		return -EINVAL;
1207
1208	while (1) {
1209		rc = -EAGAIN;
1210		ptep = gmap_pte_op_walk(gmap, gaddr, &ptl);
1211		if (ptep) {
1212			pte = *ptep;
1213			if (pte_present(pte) && (pte_val(pte) & _PAGE_READ)) {
1214				address = pte_val(pte) & PAGE_MASK;
1215				address += gaddr & ~PAGE_MASK;
1216				*val = *(unsigned long *)__va(address);
1217				set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_YOUNG)));
1218				/* Do *NOT* clear the _PAGE_INVALID bit! */
1219				rc = 0;
1220			}
1221			gmap_pte_op_end(ptep, ptl);
1222		}
1223		if (!rc)
1224			break;
1225		vmaddr = __gmap_translate(gmap, gaddr);
1226		if (IS_ERR_VALUE(vmaddr)) {
1227			rc = vmaddr;
1228			break;
1229		}
1230		rc = gmap_pte_op_fixup(gmap, gaddr, vmaddr, PROT_READ);
1231		if (rc)
1232			break;
1233	}
1234	return rc;
1235}
1236EXPORT_SYMBOL_GPL(gmap_read_table);
1237
1238/**
1239 * gmap_insert_rmap - add a rmap to the host_to_rmap radix tree
1240 * @sg: pointer to the shadow guest address space structure
1241 * @vmaddr: vm address associated with the rmap
1242 * @rmap: pointer to the rmap structure
1243 *
1244 * Called with the sg->guest_table_lock
1245 */
1246static inline void gmap_insert_rmap(struct gmap *sg, unsigned long vmaddr,
1247				    struct gmap_rmap *rmap)
1248{
1249	struct gmap_rmap *temp;
1250	void __rcu **slot;
1251
1252	BUG_ON(!gmap_is_shadow(sg));
1253	slot = radix_tree_lookup_slot(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
1254	if (slot) {
1255		rmap->next = radix_tree_deref_slot_protected(slot,
1256							&sg->guest_table_lock);
1257		for (temp = rmap->next; temp; temp = temp->next) {
1258			if (temp->raddr == rmap->raddr) {
1259				kfree(rmap);
1260				return;
1261			}
1262		}
1263		radix_tree_replace_slot(&sg->host_to_rmap, slot, rmap);
1264	} else {
1265		rmap->next = NULL;
1266		radix_tree_insert(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT,
1267				  rmap);
1268	}
1269}
1270
1271/**
1272 * gmap_protect_rmap - restrict access rights to memory (RO) and create an rmap
1273 * @sg: pointer to the shadow guest address space structure
1274 * @raddr: rmap address in the shadow gmap
1275 * @paddr: address in the parent guest address space
1276 * @len: length of the memory area to protect
1277 *
1278 * Returns 0 if successfully protected and the rmap was created, -ENOMEM
1279 * if out of memory and -EFAULT if paddr is invalid.
1280 */
1281static int gmap_protect_rmap(struct gmap *sg, unsigned long raddr,
1282			     unsigned long paddr, unsigned long len)
1283{
1284	struct gmap *parent;
1285	struct gmap_rmap *rmap;
1286	unsigned long vmaddr;
1287	spinlock_t *ptl;
1288	pte_t *ptep;
1289	int rc;
1290
1291	BUG_ON(!gmap_is_shadow(sg));
1292	parent = sg->parent;
1293	while (len) {
1294		vmaddr = __gmap_translate(parent, paddr);
1295		if (IS_ERR_VALUE(vmaddr))
1296			return vmaddr;
1297		rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
1298		if (!rmap)
1299			return -ENOMEM;
1300		rmap->raddr = raddr;
1301		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
1302		if (rc) {
1303			kfree(rmap);
1304			return rc;
1305		}
1306		rc = -EAGAIN;
1307		ptep = gmap_pte_op_walk(parent, paddr, &ptl);
1308		if (ptep) {
1309			spin_lock(&sg->guest_table_lock);
1310			rc = ptep_force_prot(parent->mm, paddr, ptep, PROT_READ,
1311					     PGSTE_VSIE_BIT);
1312			if (!rc)
1313				gmap_insert_rmap(sg, vmaddr, rmap);
1314			spin_unlock(&sg->guest_table_lock);
1315			gmap_pte_op_end(ptep, ptl);
1316		}
1317		radix_tree_preload_end();
1318		if (rc) {
1319			kfree(rmap);
1320			rc = gmap_pte_op_fixup(parent, paddr, vmaddr, PROT_READ);
1321			if (rc)
1322				return rc;
1323			continue;
1324		}
1325		paddr += PAGE_SIZE;
1326		len -= PAGE_SIZE;
1327	}
1328	return 0;
1329}
1330
1331#define _SHADOW_RMAP_MASK	0x7
1332#define _SHADOW_RMAP_REGION1	0x5
1333#define _SHADOW_RMAP_REGION2	0x4
1334#define _SHADOW_RMAP_REGION3	0x3
1335#define _SHADOW_RMAP_SEGMENT	0x2
1336#define _SHADOW_RMAP_PGTABLE	0x1
1337
1338/**
1339 * gmap_idte_one - invalidate a single region or segment table entry
1340 * @asce: region or segment table *origin* + table-type bits
1341 * @vaddr: virtual address to identify the table entry to flush
1342 *
1343 * The invalid bit of a single region or segment table entry is set
1344 * and the associated TLB entries depending on the entry are flushed.
1345 * The table-type of the @asce identifies the portion of the @vaddr
1346 * that is used as the invalidation index.
1347 */
1348static inline void gmap_idte_one(unsigned long asce, unsigned long vaddr)
1349{
1350	asm volatile(
1351		"	idte	%0,0,%1"
1352		: : "a" (asce), "a" (vaddr) : "cc", "memory");
1353}
1354
1355/**
1356 * gmap_unshadow_page - remove a page from a shadow page table
1357 * @sg: pointer to the shadow guest address space structure
1358 * @raddr: rmap address in the shadow guest address space
1359 *
1360 * Called with the sg->guest_table_lock
1361 */
1362static void gmap_unshadow_page(struct gmap *sg, unsigned long raddr)
1363{
1364	unsigned long *table;
1365
1366	BUG_ON(!gmap_is_shadow(sg));
1367	table = gmap_table_walk(sg, raddr, 0); /* get page table pointer */
1368	if (!table || *table & _PAGE_INVALID)
1369		return;
1370	gmap_call_notifier(sg, raddr, raddr + PAGE_SIZE - 1);
1371	ptep_unshadow_pte(sg->mm, raddr, (pte_t *) table);
1372}
1373
1374/**
1375 * __gmap_unshadow_pgt - remove all entries from a shadow page table
1376 * @sg: pointer to the shadow guest address space structure
1377 * @raddr: rmap address in the shadow guest address space
1378 * @pgt: pointer to the start of a shadow page table
1379 *
1380 * Called with the sg->guest_table_lock
1381 */
1382static void __gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr,
1383				unsigned long *pgt)
1384{
1385	int i;
1386
1387	BUG_ON(!gmap_is_shadow(sg));
1388	for (i = 0; i < _PAGE_ENTRIES; i++, raddr += PAGE_SIZE)
1389		pgt[i] = _PAGE_INVALID;
1390}
1391
1392/**
1393 * gmap_unshadow_pgt - remove a shadow page table from a segment entry
1394 * @sg: pointer to the shadow guest address space structure
1395 * @raddr: address in the shadow guest address space
1396 *
1397 * Called with the sg->guest_table_lock
1398 */
1399static void gmap_unshadow_pgt(struct gmap *sg, unsigned long raddr)
1400{
1401	unsigned long *ste;
1402	phys_addr_t sto, pgt;
1403	struct ptdesc *ptdesc;
1404
1405	BUG_ON(!gmap_is_shadow(sg));
1406	ste = gmap_table_walk(sg, raddr, 1); /* get segment pointer */
1407	if (!ste || !(*ste & _SEGMENT_ENTRY_ORIGIN))
1408		return;
1409	gmap_call_notifier(sg, raddr, raddr + _SEGMENT_SIZE - 1);
1410	sto = __pa(ste - ((raddr & _SEGMENT_INDEX) >> _SEGMENT_SHIFT));
1411	gmap_idte_one(sto | _ASCE_TYPE_SEGMENT, raddr);
1412	pgt = *ste & _SEGMENT_ENTRY_ORIGIN;
1413	*ste = _SEGMENT_ENTRY_EMPTY;
1414	__gmap_unshadow_pgt(sg, raddr, __va(pgt));
1415	/* Free page table */
1416	ptdesc = page_ptdesc(phys_to_page(pgt));
1417	list_del(&ptdesc->pt_list);
1418	page_table_free_pgste(ptdesc);
1419}
1420
1421/**
1422 * __gmap_unshadow_sgt - remove all entries from a shadow segment table
1423 * @sg: pointer to the shadow guest address space structure
1424 * @raddr: rmap address in the shadow guest address space
1425 * @sgt: pointer to the start of a shadow segment table
1426 *
1427 * Called with the sg->guest_table_lock
1428 */
1429static void __gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr,
1430				unsigned long *sgt)
1431{
1432	struct ptdesc *ptdesc;
1433	phys_addr_t pgt;
1434	int i;
1435
1436	BUG_ON(!gmap_is_shadow(sg));
1437	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _SEGMENT_SIZE) {
1438		if (!(sgt[i] & _SEGMENT_ENTRY_ORIGIN))
1439			continue;
1440		pgt = sgt[i] & _REGION_ENTRY_ORIGIN;
1441		sgt[i] = _SEGMENT_ENTRY_EMPTY;
1442		__gmap_unshadow_pgt(sg, raddr, __va(pgt));
1443		/* Free page table */
1444		ptdesc = page_ptdesc(phys_to_page(pgt));
1445		list_del(&ptdesc->pt_list);
1446		page_table_free_pgste(ptdesc);
1447	}
1448}
1449
1450/**
1451 * gmap_unshadow_sgt - remove a shadow segment table from a region-3 entry
1452 * @sg: pointer to the shadow guest address space structure
1453 * @raddr: rmap address in the shadow guest address space
1454 *
1455 * Called with the shadow->guest_table_lock
1456 */
1457static void gmap_unshadow_sgt(struct gmap *sg, unsigned long raddr)
1458{
1459	unsigned long r3o, *r3e;
1460	phys_addr_t sgt;
1461	struct page *page;
1462
1463	BUG_ON(!gmap_is_shadow(sg));
1464	r3e = gmap_table_walk(sg, raddr, 2); /* get region-3 pointer */
1465	if (!r3e || !(*r3e & _REGION_ENTRY_ORIGIN))
1466		return;
1467	gmap_call_notifier(sg, raddr, raddr + _REGION3_SIZE - 1);
1468	r3o = (unsigned long) (r3e - ((raddr & _REGION3_INDEX) >> _REGION3_SHIFT));
1469	gmap_idte_one(__pa(r3o) | _ASCE_TYPE_REGION3, raddr);
1470	sgt = *r3e & _REGION_ENTRY_ORIGIN;
1471	*r3e = _REGION3_ENTRY_EMPTY;
1472	__gmap_unshadow_sgt(sg, raddr, __va(sgt));
1473	/* Free segment table */
1474	page = phys_to_page(sgt);
1475	list_del(&page->lru);
1476	__free_pages(page, CRST_ALLOC_ORDER);
1477}
1478
1479/**
1480 * __gmap_unshadow_r3t - remove all entries from a shadow region-3 table
1481 * @sg: pointer to the shadow guest address space structure
1482 * @raddr: address in the shadow guest address space
1483 * @r3t: pointer to the start of a shadow region-3 table
1484 *
1485 * Called with the sg->guest_table_lock
1486 */
1487static void __gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr,
1488				unsigned long *r3t)
1489{
1490	struct page *page;
1491	phys_addr_t sgt;
1492	int i;
1493
1494	BUG_ON(!gmap_is_shadow(sg));
1495	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION3_SIZE) {
1496		if (!(r3t[i] & _REGION_ENTRY_ORIGIN))
1497			continue;
1498		sgt = r3t[i] & _REGION_ENTRY_ORIGIN;
1499		r3t[i] = _REGION3_ENTRY_EMPTY;
1500		__gmap_unshadow_sgt(sg, raddr, __va(sgt));
1501		/* Free segment table */
1502		page = phys_to_page(sgt);
1503		list_del(&page->lru);
1504		__free_pages(page, CRST_ALLOC_ORDER);
1505	}
1506}
1507
1508/**
1509 * gmap_unshadow_r3t - remove a shadow region-3 table from a region-2 entry
1510 * @sg: pointer to the shadow guest address space structure
1511 * @raddr: rmap address in the shadow guest address space
1512 *
1513 * Called with the sg->guest_table_lock
1514 */
1515static void gmap_unshadow_r3t(struct gmap *sg, unsigned long raddr)
1516{
1517	unsigned long r2o, *r2e;
1518	phys_addr_t r3t;
1519	struct page *page;
1520
1521	BUG_ON(!gmap_is_shadow(sg));
1522	r2e = gmap_table_walk(sg, raddr, 3); /* get region-2 pointer */
1523	if (!r2e || !(*r2e & _REGION_ENTRY_ORIGIN))
1524		return;
1525	gmap_call_notifier(sg, raddr, raddr + _REGION2_SIZE - 1);
1526	r2o = (unsigned long) (r2e - ((raddr & _REGION2_INDEX) >> _REGION2_SHIFT));
1527	gmap_idte_one(__pa(r2o) | _ASCE_TYPE_REGION2, raddr);
1528	r3t = *r2e & _REGION_ENTRY_ORIGIN;
1529	*r2e = _REGION2_ENTRY_EMPTY;
1530	__gmap_unshadow_r3t(sg, raddr, __va(r3t));
1531	/* Free region 3 table */
1532	page = phys_to_page(r3t);
1533	list_del(&page->lru);
1534	__free_pages(page, CRST_ALLOC_ORDER);
1535}
1536
1537/**
1538 * __gmap_unshadow_r2t - remove all entries from a shadow region-2 table
1539 * @sg: pointer to the shadow guest address space structure
1540 * @raddr: rmap address in the shadow guest address space
1541 * @r2t: pointer to the start of a shadow region-2 table
1542 *
1543 * Called with the sg->guest_table_lock
1544 */
1545static void __gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr,
1546				unsigned long *r2t)
1547{
1548	phys_addr_t r3t;
1549	struct page *page;
1550	int i;
1551
1552	BUG_ON(!gmap_is_shadow(sg));
1553	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION2_SIZE) {
1554		if (!(r2t[i] & _REGION_ENTRY_ORIGIN))
1555			continue;
1556		r3t = r2t[i] & _REGION_ENTRY_ORIGIN;
1557		r2t[i] = _REGION2_ENTRY_EMPTY;
1558		__gmap_unshadow_r3t(sg, raddr, __va(r3t));
1559		/* Free region 3 table */
1560		page = phys_to_page(r3t);
1561		list_del(&page->lru);
1562		__free_pages(page, CRST_ALLOC_ORDER);
1563	}
1564}
1565
1566/**
1567 * gmap_unshadow_r2t - remove a shadow region-2 table from a region-1 entry
1568 * @sg: pointer to the shadow guest address space structure
1569 * @raddr: rmap address in the shadow guest address space
1570 *
1571 * Called with the sg->guest_table_lock
1572 */
1573static void gmap_unshadow_r2t(struct gmap *sg, unsigned long raddr)
1574{
1575	unsigned long r1o, *r1e;
1576	struct page *page;
1577	phys_addr_t r2t;
1578
1579	BUG_ON(!gmap_is_shadow(sg));
1580	r1e = gmap_table_walk(sg, raddr, 4); /* get region-1 pointer */
1581	if (!r1e || !(*r1e & _REGION_ENTRY_ORIGIN))
1582		return;
1583	gmap_call_notifier(sg, raddr, raddr + _REGION1_SIZE - 1);
1584	r1o = (unsigned long) (r1e - ((raddr & _REGION1_INDEX) >> _REGION1_SHIFT));
1585	gmap_idte_one(__pa(r1o) | _ASCE_TYPE_REGION1, raddr);
1586	r2t = *r1e & _REGION_ENTRY_ORIGIN;
1587	*r1e = _REGION1_ENTRY_EMPTY;
1588	__gmap_unshadow_r2t(sg, raddr, __va(r2t));
1589	/* Free region 2 table */
1590	page = phys_to_page(r2t);
1591	list_del(&page->lru);
1592	__free_pages(page, CRST_ALLOC_ORDER);
1593}
1594
1595/**
1596 * __gmap_unshadow_r1t - remove all entries from a shadow region-1 table
1597 * @sg: pointer to the shadow guest address space structure
1598 * @raddr: rmap address in the shadow guest address space
1599 * @r1t: pointer to the start of a shadow region-1 table
1600 *
1601 * Called with the shadow->guest_table_lock
1602 */
1603static void __gmap_unshadow_r1t(struct gmap *sg, unsigned long raddr,
1604				unsigned long *r1t)
1605{
1606	unsigned long asce;
1607	struct page *page;
1608	phys_addr_t r2t;
1609	int i;
1610
1611	BUG_ON(!gmap_is_shadow(sg));
1612	asce = __pa(r1t) | _ASCE_TYPE_REGION1;
1613	for (i = 0; i < _CRST_ENTRIES; i++, raddr += _REGION1_SIZE) {
1614		if (!(r1t[i] & _REGION_ENTRY_ORIGIN))
1615			continue;
1616		r2t = r1t[i] & _REGION_ENTRY_ORIGIN;
1617		__gmap_unshadow_r2t(sg, raddr, __va(r2t));
1618		/* Clear entry and flush translation r1t -> r2t */
1619		gmap_idte_one(asce, raddr);
1620		r1t[i] = _REGION1_ENTRY_EMPTY;
1621		/* Free region 2 table */
1622		page = phys_to_page(r2t);
1623		list_del(&page->lru);
1624		__free_pages(page, CRST_ALLOC_ORDER);
1625	}
1626}
1627
1628/**
1629 * gmap_unshadow - remove a shadow page table completely
1630 * @sg: pointer to the shadow guest address space structure
1631 *
1632 * Called with sg->guest_table_lock
1633 */
1634static void gmap_unshadow(struct gmap *sg)
1635{
1636	unsigned long *table;
1637
1638	BUG_ON(!gmap_is_shadow(sg));
1639	if (sg->removed)
1640		return;
1641	sg->removed = 1;
1642	gmap_call_notifier(sg, 0, -1UL);
1643	gmap_flush_tlb(sg);
1644	table = __va(sg->asce & _ASCE_ORIGIN);
1645	switch (sg->asce & _ASCE_TYPE_MASK) {
1646	case _ASCE_TYPE_REGION1:
1647		__gmap_unshadow_r1t(sg, 0, table);
1648		break;
1649	case _ASCE_TYPE_REGION2:
1650		__gmap_unshadow_r2t(sg, 0, table);
1651		break;
1652	case _ASCE_TYPE_REGION3:
1653		__gmap_unshadow_r3t(sg, 0, table);
1654		break;
1655	case _ASCE_TYPE_SEGMENT:
1656		__gmap_unshadow_sgt(sg, 0, table);
1657		break;
1658	}
1659}
1660
1661/**
1662 * gmap_find_shadow - find a specific asce in the list of shadow tables
1663 * @parent: pointer to the parent gmap
1664 * @asce: ASCE for which the shadow table is created
1665 * @edat_level: edat level to be used for the shadow translation
1666 *
1667 * Returns the pointer to a gmap if a shadow table with the given asce is
1668 * already available, ERR_PTR(-EAGAIN) if another one is just being created,
1669 * otherwise NULL
1670 */
1671static struct gmap *gmap_find_shadow(struct gmap *parent, unsigned long asce,
1672				     int edat_level)
1673{
1674	struct gmap *sg;
1675
1676	list_for_each_entry(sg, &parent->children, list) {
1677		if (sg->orig_asce != asce || sg->edat_level != edat_level ||
1678		    sg->removed)
1679			continue;
1680		if (!sg->initialized)
1681			return ERR_PTR(-EAGAIN);
1682		refcount_inc(&sg->ref_count);
1683		return sg;
1684	}
1685	return NULL;
1686}
1687
1688/**
1689 * gmap_shadow_valid - check if a shadow guest address space matches the
1690 *                     given properties and is still valid
1691 * @sg: pointer to the shadow guest address space structure
1692 * @asce: ASCE for which the shadow table is requested
1693 * @edat_level: edat level to be used for the shadow translation
1694 *
1695 * Returns 1 if the gmap shadow is still valid and matches the given
1696 * properties, the caller can continue using it. Returns 0 otherwise, the
1697 * caller has to request a new shadow gmap in this case.
1698 *
1699 */
1700int gmap_shadow_valid(struct gmap *sg, unsigned long asce, int edat_level)
1701{
1702	if (sg->removed)
1703		return 0;
1704	return sg->orig_asce == asce && sg->edat_level == edat_level;
1705}
1706EXPORT_SYMBOL_GPL(gmap_shadow_valid);
1707
1708/**
1709 * gmap_shadow - create/find a shadow guest address space
1710 * @parent: pointer to the parent gmap
1711 * @asce: ASCE for which the shadow table is created
1712 * @edat_level: edat level to be used for the shadow translation
1713 *
1714 * The pages of the top level page table referred by the asce parameter
1715 * will be set to read-only and marked in the PGSTEs of the kvm process.
1716 * The shadow table will be removed automatically on any change to the
1717 * PTE mapping for the source table.
1718 *
1719 * Returns a guest address space structure, ERR_PTR(-ENOMEM) if out of memory,
1720 * ERR_PTR(-EAGAIN) if the caller has to retry and ERR_PTR(-EFAULT) if the
1721 * parent gmap table could not be protected.
1722 */
1723struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
1724			 int edat_level)
1725{
1726	struct gmap *sg, *new;
1727	unsigned long limit;
1728	int rc;
1729
1730	BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
1731	BUG_ON(gmap_is_shadow(parent));
1732	spin_lock(&parent->shadow_lock);
1733	sg = gmap_find_shadow(parent, asce, edat_level);
1734	spin_unlock(&parent->shadow_lock);
1735	if (sg)
1736		return sg;
1737	/* Create a new shadow gmap */
1738	limit = -1UL >> (33 - (((asce & _ASCE_TYPE_MASK) >> 2) * 11));
1739	if (asce & _ASCE_REAL_SPACE)
1740		limit = -1UL;
1741	new = gmap_alloc(limit);
1742	if (!new)
1743		return ERR_PTR(-ENOMEM);
1744	new->mm = parent->mm;
1745	new->parent = gmap_get(parent);
1746	new->private = parent->private;
1747	new->orig_asce = asce;
1748	new->edat_level = edat_level;
1749	new->initialized = false;
1750	spin_lock(&parent->shadow_lock);
1751	/* Recheck if another CPU created the same shadow */
1752	sg = gmap_find_shadow(parent, asce, edat_level);
1753	if (sg) {
1754		spin_unlock(&parent->shadow_lock);
1755		gmap_free(new);
1756		return sg;
1757	}
1758	if (asce & _ASCE_REAL_SPACE) {
1759		/* only allow one real-space gmap shadow */
1760		list_for_each_entry(sg, &parent->children, list) {
1761			if (sg->orig_asce & _ASCE_REAL_SPACE) {
1762				spin_lock(&sg->guest_table_lock);
1763				gmap_unshadow(sg);
1764				spin_unlock(&sg->guest_table_lock);
1765				list_del(&sg->list);
1766				gmap_put(sg);
1767				break;
1768			}
1769		}
1770	}
1771	refcount_set(&new->ref_count, 2);
1772	list_add(&new->list, &parent->children);
1773	if (asce & _ASCE_REAL_SPACE) {
1774		/* nothing to protect, return right away */
1775		new->initialized = true;
1776		spin_unlock(&parent->shadow_lock);
1777		return new;
1778	}
1779	spin_unlock(&parent->shadow_lock);
1780	/* protect after insertion, so it will get properly invalidated */
1781	mmap_read_lock(parent->mm);
1782	rc = gmap_protect_range(parent, asce & _ASCE_ORIGIN,
1783				((asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE,
1784				PROT_READ, GMAP_NOTIFY_SHADOW);
1785	mmap_read_unlock(parent->mm);
1786	spin_lock(&parent->shadow_lock);
1787	new->initialized = true;
1788	if (rc) {
1789		list_del(&new->list);
1790		gmap_free(new);
1791		new = ERR_PTR(rc);
1792	}
1793	spin_unlock(&parent->shadow_lock);
1794	return new;
1795}
1796EXPORT_SYMBOL_GPL(gmap_shadow);
1797
1798/**
1799 * gmap_shadow_r2t - create an empty shadow region 2 table
1800 * @sg: pointer to the shadow guest address space structure
1801 * @saddr: faulting address in the shadow gmap
1802 * @r2t: parent gmap address of the region 2 table to get shadowed
1803 * @fake: r2t references contiguous guest memory block, not a r2t
1804 *
1805 * The r2t parameter specifies the address of the source table. The
1806 * four pages of the source table are made read-only in the parent gmap
1807 * address space. A write to the source table area @r2t will automatically
1808 * remove the shadow r2 table and all of its descendants.
1809 *
1810 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1811 * shadow table structure is incomplete, -ENOMEM if out of memory and
1812 * -EFAULT if an address in the parent gmap could not be resolved.
1813 *
1814 * Called with sg->mm->mmap_lock in read.
1815 */
1816int gmap_shadow_r2t(struct gmap *sg, unsigned long saddr, unsigned long r2t,
1817		    int fake)
1818{
1819	unsigned long raddr, origin, offset, len;
1820	unsigned long *table;
1821	phys_addr_t s_r2t;
1822	struct page *page;
1823	int rc;
1824
1825	BUG_ON(!gmap_is_shadow(sg));
1826	/* Allocate a shadow region second table */
1827	page = gmap_alloc_crst();
1828	if (!page)
1829		return -ENOMEM;
1830	page->index = r2t & _REGION_ENTRY_ORIGIN;
1831	if (fake)
1832		page->index |= GMAP_SHADOW_FAKE_TABLE;
1833	s_r2t = page_to_phys(page);
1834	/* Install shadow region second table */
1835	spin_lock(&sg->guest_table_lock);
1836	table = gmap_table_walk(sg, saddr, 4); /* get region-1 pointer */
1837	if (!table) {
1838		rc = -EAGAIN;		/* Race with unshadow */
1839		goto out_free;
1840	}
1841	if (!(*table & _REGION_ENTRY_INVALID)) {
1842		rc = 0;			/* Already established */
1843		goto out_free;
1844	} else if (*table & _REGION_ENTRY_ORIGIN) {
1845		rc = -EAGAIN;		/* Race with shadow */
1846		goto out_free;
1847	}
1848	crst_table_init(__va(s_r2t), _REGION2_ENTRY_EMPTY);
1849	/* mark as invalid as long as the parent table is not protected */
1850	*table = s_r2t | _REGION_ENTRY_LENGTH |
1851		 _REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID;
1852	if (sg->edat_level >= 1)
1853		*table |= (r2t & _REGION_ENTRY_PROTECT);
1854	list_add(&page->lru, &sg->crst_list);
1855	if (fake) {
1856		/* nothing to protect for fake tables */
1857		*table &= ~_REGION_ENTRY_INVALID;
1858		spin_unlock(&sg->guest_table_lock);
1859		return 0;
1860	}
1861	spin_unlock(&sg->guest_table_lock);
1862	/* Make r2t read-only in parent gmap page table */
1863	raddr = (saddr & _REGION1_MASK) | _SHADOW_RMAP_REGION1;
1864	origin = r2t & _REGION_ENTRY_ORIGIN;
1865	offset = ((r2t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1866	len = ((r2t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1867	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1868	spin_lock(&sg->guest_table_lock);
1869	if (!rc) {
1870		table = gmap_table_walk(sg, saddr, 4);
1871		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r2t)
1872			rc = -EAGAIN;		/* Race with unshadow */
1873		else
1874			*table &= ~_REGION_ENTRY_INVALID;
1875	} else {
1876		gmap_unshadow_r2t(sg, raddr);
1877	}
1878	spin_unlock(&sg->guest_table_lock);
1879	return rc;
1880out_free:
1881	spin_unlock(&sg->guest_table_lock);
1882	__free_pages(page, CRST_ALLOC_ORDER);
1883	return rc;
1884}
1885EXPORT_SYMBOL_GPL(gmap_shadow_r2t);
1886
1887/**
1888 * gmap_shadow_r3t - create a shadow region 3 table
1889 * @sg: pointer to the shadow guest address space structure
1890 * @saddr: faulting address in the shadow gmap
1891 * @r3t: parent gmap address of the region 3 table to get shadowed
1892 * @fake: r3t references contiguous guest memory block, not a r3t
1893 *
1894 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
1895 * shadow table structure is incomplete, -ENOMEM if out of memory and
1896 * -EFAULT if an address in the parent gmap could not be resolved.
1897 *
1898 * Called with sg->mm->mmap_lock in read.
1899 */
1900int gmap_shadow_r3t(struct gmap *sg, unsigned long saddr, unsigned long r3t,
1901		    int fake)
1902{
1903	unsigned long raddr, origin, offset, len;
1904	unsigned long *table;
1905	phys_addr_t s_r3t;
1906	struct page *page;
1907	int rc;
1908
1909	BUG_ON(!gmap_is_shadow(sg));
1910	/* Allocate a shadow region second table */
1911	page = gmap_alloc_crst();
1912	if (!page)
1913		return -ENOMEM;
1914	page->index = r3t & _REGION_ENTRY_ORIGIN;
1915	if (fake)
1916		page->index |= GMAP_SHADOW_FAKE_TABLE;
1917	s_r3t = page_to_phys(page);
1918	/* Install shadow region second table */
1919	spin_lock(&sg->guest_table_lock);
1920	table = gmap_table_walk(sg, saddr, 3); /* get region-2 pointer */
1921	if (!table) {
1922		rc = -EAGAIN;		/* Race with unshadow */
1923		goto out_free;
1924	}
1925	if (!(*table & _REGION_ENTRY_INVALID)) {
1926		rc = 0;			/* Already established */
1927		goto out_free;
1928	} else if (*table & _REGION_ENTRY_ORIGIN) {
1929		rc = -EAGAIN;		/* Race with shadow */
1930		goto out_free;
1931	}
1932	crst_table_init(__va(s_r3t), _REGION3_ENTRY_EMPTY);
1933	/* mark as invalid as long as the parent table is not protected */
1934	*table = s_r3t | _REGION_ENTRY_LENGTH |
1935		 _REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID;
1936	if (sg->edat_level >= 1)
1937		*table |= (r3t & _REGION_ENTRY_PROTECT);
1938	list_add(&page->lru, &sg->crst_list);
1939	if (fake) {
1940		/* nothing to protect for fake tables */
1941		*table &= ~_REGION_ENTRY_INVALID;
1942		spin_unlock(&sg->guest_table_lock);
1943		return 0;
1944	}
1945	spin_unlock(&sg->guest_table_lock);
1946	/* Make r3t read-only in parent gmap page table */
1947	raddr = (saddr & _REGION2_MASK) | _SHADOW_RMAP_REGION2;
1948	origin = r3t & _REGION_ENTRY_ORIGIN;
1949	offset = ((r3t & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
1950	len = ((r3t & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
1951	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
1952	spin_lock(&sg->guest_table_lock);
1953	if (!rc) {
1954		table = gmap_table_walk(sg, saddr, 3);
1955		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_r3t)
1956			rc = -EAGAIN;		/* Race with unshadow */
1957		else
1958			*table &= ~_REGION_ENTRY_INVALID;
1959	} else {
1960		gmap_unshadow_r3t(sg, raddr);
1961	}
1962	spin_unlock(&sg->guest_table_lock);
1963	return rc;
1964out_free:
1965	spin_unlock(&sg->guest_table_lock);
1966	__free_pages(page, CRST_ALLOC_ORDER);
1967	return rc;
1968}
1969EXPORT_SYMBOL_GPL(gmap_shadow_r3t);
1970
1971/**
1972 * gmap_shadow_sgt - create a shadow segment table
1973 * @sg: pointer to the shadow guest address space structure
1974 * @saddr: faulting address in the shadow gmap
1975 * @sgt: parent gmap address of the segment table to get shadowed
1976 * @fake: sgt references contiguous guest memory block, not a sgt
1977 *
1978 * Returns: 0 if successfully shadowed or already shadowed, -EAGAIN if the
1979 * shadow table structure is incomplete, -ENOMEM if out of memory and
1980 * -EFAULT if an address in the parent gmap could not be resolved.
1981 *
1982 * Called with sg->mm->mmap_lock in read.
1983 */
1984int gmap_shadow_sgt(struct gmap *sg, unsigned long saddr, unsigned long sgt,
1985		    int fake)
1986{
1987	unsigned long raddr, origin, offset, len;
1988	unsigned long *table;
1989	phys_addr_t s_sgt;
1990	struct page *page;
1991	int rc;
1992
1993	BUG_ON(!gmap_is_shadow(sg) || (sgt & _REGION3_ENTRY_LARGE));
1994	/* Allocate a shadow segment table */
1995	page = gmap_alloc_crst();
1996	if (!page)
1997		return -ENOMEM;
1998	page->index = sgt & _REGION_ENTRY_ORIGIN;
1999	if (fake)
2000		page->index |= GMAP_SHADOW_FAKE_TABLE;
2001	s_sgt = page_to_phys(page);
2002	/* Install shadow region second table */
2003	spin_lock(&sg->guest_table_lock);
2004	table = gmap_table_walk(sg, saddr, 2); /* get region-3 pointer */
2005	if (!table) {
2006		rc = -EAGAIN;		/* Race with unshadow */
2007		goto out_free;
2008	}
2009	if (!(*table & _REGION_ENTRY_INVALID)) {
2010		rc = 0;			/* Already established */
2011		goto out_free;
2012	} else if (*table & _REGION_ENTRY_ORIGIN) {
2013		rc = -EAGAIN;		/* Race with shadow */
2014		goto out_free;
2015	}
2016	crst_table_init(__va(s_sgt), _SEGMENT_ENTRY_EMPTY);
2017	/* mark as invalid as long as the parent table is not protected */
2018	*table = s_sgt | _REGION_ENTRY_LENGTH |
2019		 _REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID;
2020	if (sg->edat_level >= 1)
2021		*table |= sgt & _REGION_ENTRY_PROTECT;
2022	list_add(&page->lru, &sg->crst_list);
2023	if (fake) {
2024		/* nothing to protect for fake tables */
2025		*table &= ~_REGION_ENTRY_INVALID;
2026		spin_unlock(&sg->guest_table_lock);
2027		return 0;
2028	}
2029	spin_unlock(&sg->guest_table_lock);
2030	/* Make sgt read-only in parent gmap page table */
2031	raddr = (saddr & _REGION3_MASK) | _SHADOW_RMAP_REGION3;
2032	origin = sgt & _REGION_ENTRY_ORIGIN;
2033	offset = ((sgt & _REGION_ENTRY_OFFSET) >> 6) * PAGE_SIZE;
2034	len = ((sgt & _REGION_ENTRY_LENGTH) + 1) * PAGE_SIZE - offset;
2035	rc = gmap_protect_rmap(sg, raddr, origin + offset, len);
2036	spin_lock(&sg->guest_table_lock);
2037	if (!rc) {
2038		table = gmap_table_walk(sg, saddr, 2);
2039		if (!table || (*table & _REGION_ENTRY_ORIGIN) != s_sgt)
2040			rc = -EAGAIN;		/* Race with unshadow */
2041		else
2042			*table &= ~_REGION_ENTRY_INVALID;
2043	} else {
2044		gmap_unshadow_sgt(sg, raddr);
2045	}
2046	spin_unlock(&sg->guest_table_lock);
2047	return rc;
2048out_free:
2049	spin_unlock(&sg->guest_table_lock);
2050	__free_pages(page, CRST_ALLOC_ORDER);
2051	return rc;
2052}
2053EXPORT_SYMBOL_GPL(gmap_shadow_sgt);
2054
2055/**
2056 * gmap_shadow_pgt_lookup - find a shadow page table
2057 * @sg: pointer to the shadow guest address space structure
2058 * @saddr: the address in the shadow aguest address space
2059 * @pgt: parent gmap address of the page table to get shadowed
2060 * @dat_protection: if the pgtable is marked as protected by dat
2061 * @fake: pgt references contiguous guest memory block, not a pgtable
2062 *
2063 * Returns 0 if the shadow page table was found and -EAGAIN if the page
2064 * table was not found.
2065 *
2066 * Called with sg->mm->mmap_lock in read.
2067 */
2068int gmap_shadow_pgt_lookup(struct gmap *sg, unsigned long saddr,
2069			   unsigned long *pgt, int *dat_protection,
2070			   int *fake)
2071{
2072	unsigned long *table;
2073	struct page *page;
2074	int rc;
2075
2076	BUG_ON(!gmap_is_shadow(sg));
2077	spin_lock(&sg->guest_table_lock);
2078	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2079	if (table && !(*table & _SEGMENT_ENTRY_INVALID)) {
2080		/* Shadow page tables are full pages (pte+pgste) */
2081		page = pfn_to_page(*table >> PAGE_SHIFT);
2082		*pgt = page->index & ~GMAP_SHADOW_FAKE_TABLE;
2083		*dat_protection = !!(*table & _SEGMENT_ENTRY_PROTECT);
2084		*fake = !!(page->index & GMAP_SHADOW_FAKE_TABLE);
2085		rc = 0;
2086	} else  {
2087		rc = -EAGAIN;
2088	}
2089	spin_unlock(&sg->guest_table_lock);
2090	return rc;
2091
2092}
2093EXPORT_SYMBOL_GPL(gmap_shadow_pgt_lookup);
2094
2095/**
2096 * gmap_shadow_pgt - instantiate a shadow page table
2097 * @sg: pointer to the shadow guest address space structure
2098 * @saddr: faulting address in the shadow gmap
2099 * @pgt: parent gmap address of the page table to get shadowed
2100 * @fake: pgt references contiguous guest memory block, not a pgtable
2101 *
2102 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2103 * shadow table structure is incomplete, -ENOMEM if out of memory,
2104 * -EFAULT if an address in the parent gmap could not be resolved and
2105 *
2106 * Called with gmap->mm->mmap_lock in read
2107 */
2108int gmap_shadow_pgt(struct gmap *sg, unsigned long saddr, unsigned long pgt,
2109		    int fake)
2110{
2111	unsigned long raddr, origin;
2112	unsigned long *table;
2113	struct ptdesc *ptdesc;
2114	phys_addr_t s_pgt;
2115	int rc;
2116
2117	BUG_ON(!gmap_is_shadow(sg) || (pgt & _SEGMENT_ENTRY_LARGE));
2118	/* Allocate a shadow page table */
2119	ptdesc = page_table_alloc_pgste(sg->mm);
2120	if (!ptdesc)
2121		return -ENOMEM;
2122	ptdesc->pt_index = pgt & _SEGMENT_ENTRY_ORIGIN;
2123	if (fake)
2124		ptdesc->pt_index |= GMAP_SHADOW_FAKE_TABLE;
2125	s_pgt = page_to_phys(ptdesc_page(ptdesc));
2126	/* Install shadow page table */
2127	spin_lock(&sg->guest_table_lock);
2128	table = gmap_table_walk(sg, saddr, 1); /* get segment pointer */
2129	if (!table) {
2130		rc = -EAGAIN;		/* Race with unshadow */
2131		goto out_free;
2132	}
2133	if (!(*table & _SEGMENT_ENTRY_INVALID)) {
2134		rc = 0;			/* Already established */
2135		goto out_free;
2136	} else if (*table & _SEGMENT_ENTRY_ORIGIN) {
2137		rc = -EAGAIN;		/* Race with shadow */
2138		goto out_free;
2139	}
2140	/* mark as invalid as long as the parent table is not protected */
2141	*table = (unsigned long) s_pgt | _SEGMENT_ENTRY |
2142		 (pgt & _SEGMENT_ENTRY_PROTECT) | _SEGMENT_ENTRY_INVALID;
2143	list_add(&ptdesc->pt_list, &sg->pt_list);
2144	if (fake) {
2145		/* nothing to protect for fake tables */
2146		*table &= ~_SEGMENT_ENTRY_INVALID;
2147		spin_unlock(&sg->guest_table_lock);
2148		return 0;
2149	}
2150	spin_unlock(&sg->guest_table_lock);
2151	/* Make pgt read-only in parent gmap page table (not the pgste) */
2152	raddr = (saddr & _SEGMENT_MASK) | _SHADOW_RMAP_SEGMENT;
2153	origin = pgt & _SEGMENT_ENTRY_ORIGIN & PAGE_MASK;
2154	rc = gmap_protect_rmap(sg, raddr, origin, PAGE_SIZE);
2155	spin_lock(&sg->guest_table_lock);
2156	if (!rc) {
2157		table = gmap_table_walk(sg, saddr, 1);
2158		if (!table || (*table & _SEGMENT_ENTRY_ORIGIN) != s_pgt)
2159			rc = -EAGAIN;		/* Race with unshadow */
2160		else
2161			*table &= ~_SEGMENT_ENTRY_INVALID;
2162	} else {
2163		gmap_unshadow_pgt(sg, raddr);
2164	}
2165	spin_unlock(&sg->guest_table_lock);
2166	return rc;
2167out_free:
2168	spin_unlock(&sg->guest_table_lock);
2169	page_table_free_pgste(ptdesc);
2170	return rc;
2171
2172}
2173EXPORT_SYMBOL_GPL(gmap_shadow_pgt);
2174
2175/**
2176 * gmap_shadow_page - create a shadow page mapping
2177 * @sg: pointer to the shadow guest address space structure
2178 * @saddr: faulting address in the shadow gmap
2179 * @pte: pte in parent gmap address space to get shadowed
2180 *
2181 * Returns 0 if successfully shadowed or already shadowed, -EAGAIN if the
2182 * shadow table structure is incomplete, -ENOMEM if out of memory and
2183 * -EFAULT if an address in the parent gmap could not be resolved.
2184 *
2185 * Called with sg->mm->mmap_lock in read.
2186 */
2187int gmap_shadow_page(struct gmap *sg, unsigned long saddr, pte_t pte)
2188{
2189	struct gmap *parent;
2190	struct gmap_rmap *rmap;
2191	unsigned long vmaddr, paddr;
2192	spinlock_t *ptl;
2193	pte_t *sptep, *tptep;
2194	int prot;
2195	int rc;
2196
2197	BUG_ON(!gmap_is_shadow(sg));
2198	parent = sg->parent;
2199	prot = (pte_val(pte) & _PAGE_PROTECT) ? PROT_READ : PROT_WRITE;
2200
2201	rmap = kzalloc(sizeof(*rmap), GFP_KERNEL_ACCOUNT);
2202	if (!rmap)
2203		return -ENOMEM;
2204	rmap->raddr = (saddr & PAGE_MASK) | _SHADOW_RMAP_PGTABLE;
2205
2206	while (1) {
2207		paddr = pte_val(pte) & PAGE_MASK;
2208		vmaddr = __gmap_translate(parent, paddr);
2209		if (IS_ERR_VALUE(vmaddr)) {
2210			rc = vmaddr;
2211			break;
2212		}
2213		rc = radix_tree_preload(GFP_KERNEL_ACCOUNT);
2214		if (rc)
2215			break;
2216		rc = -EAGAIN;
2217		sptep = gmap_pte_op_walk(parent, paddr, &ptl);
2218		if (sptep) {
2219			spin_lock(&sg->guest_table_lock);
2220			/* Get page table pointer */
2221			tptep = (pte_t *) gmap_table_walk(sg, saddr, 0);
2222			if (!tptep) {
2223				spin_unlock(&sg->guest_table_lock);
2224				gmap_pte_op_end(sptep, ptl);
2225				radix_tree_preload_end();
2226				break;
2227			}
2228			rc = ptep_shadow_pte(sg->mm, saddr, sptep, tptep, pte);
2229			if (rc > 0) {
2230				/* Success and a new mapping */
2231				gmap_insert_rmap(sg, vmaddr, rmap);
2232				rmap = NULL;
2233				rc = 0;
2234			}
2235			gmap_pte_op_end(sptep, ptl);
2236			spin_unlock(&sg->guest_table_lock);
2237		}
2238		radix_tree_preload_end();
2239		if (!rc)
2240			break;
2241		rc = gmap_pte_op_fixup(parent, paddr, vmaddr, prot);
2242		if (rc)
2243			break;
2244	}
2245	kfree(rmap);
2246	return rc;
2247}
2248EXPORT_SYMBOL_GPL(gmap_shadow_page);
2249
2250/*
2251 * gmap_shadow_notify - handle notifications for shadow gmap
2252 *
2253 * Called with sg->parent->shadow_lock.
2254 */
2255static void gmap_shadow_notify(struct gmap *sg, unsigned long vmaddr,
2256			       unsigned long gaddr)
2257{
2258	struct gmap_rmap *rmap, *rnext, *head;
2259	unsigned long start, end, bits, raddr;
2260
2261	BUG_ON(!gmap_is_shadow(sg));
2262
2263	spin_lock(&sg->guest_table_lock);
2264	if (sg->removed) {
2265		spin_unlock(&sg->guest_table_lock);
2266		return;
2267	}
2268	/* Check for top level table */
2269	start = sg->orig_asce & _ASCE_ORIGIN;
2270	end = start + ((sg->orig_asce & _ASCE_TABLE_LENGTH) + 1) * PAGE_SIZE;
2271	if (!(sg->orig_asce & _ASCE_REAL_SPACE) && gaddr >= start &&
2272	    gaddr < end) {
2273		/* The complete shadow table has to go */
2274		gmap_unshadow(sg);
2275		spin_unlock(&sg->guest_table_lock);
2276		list_del(&sg->list);
2277		gmap_put(sg);
2278		return;
2279	}
2280	/* Remove the page table tree from on specific entry */
2281	head = radix_tree_delete(&sg->host_to_rmap, vmaddr >> PAGE_SHIFT);
2282	gmap_for_each_rmap_safe(rmap, rnext, head) {
2283		bits = rmap->raddr & _SHADOW_RMAP_MASK;
2284		raddr = rmap->raddr ^ bits;
2285		switch (bits) {
2286		case _SHADOW_RMAP_REGION1:
2287			gmap_unshadow_r2t(sg, raddr);
2288			break;
2289		case _SHADOW_RMAP_REGION2:
2290			gmap_unshadow_r3t(sg, raddr);
2291			break;
2292		case _SHADOW_RMAP_REGION3:
2293			gmap_unshadow_sgt(sg, raddr);
2294			break;
2295		case _SHADOW_RMAP_SEGMENT:
2296			gmap_unshadow_pgt(sg, raddr);
2297			break;
2298		case _SHADOW_RMAP_PGTABLE:
2299			gmap_unshadow_page(sg, raddr);
2300			break;
2301		}
2302		kfree(rmap);
2303	}
2304	spin_unlock(&sg->guest_table_lock);
2305}
2306
2307/**
2308 * ptep_notify - call all invalidation callbacks for a specific pte.
2309 * @mm: pointer to the process mm_struct
2310 * @vmaddr: virtual address in the process address space
2311 * @pte: pointer to the page table entry
2312 * @bits: bits from the pgste that caused the notify call
2313 *
2314 * This function is assumed to be called with the page table lock held
2315 * for the pte to notify.
2316 */
2317void ptep_notify(struct mm_struct *mm, unsigned long vmaddr,
2318		 pte_t *pte, unsigned long bits)
2319{
2320	unsigned long offset, gaddr = 0;
2321	unsigned long *table;
2322	struct gmap *gmap, *sg, *next;
2323
2324	offset = ((unsigned long) pte) & (255 * sizeof(pte_t));
2325	offset = offset * (PAGE_SIZE / sizeof(pte_t));
2326	rcu_read_lock();
2327	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2328		spin_lock(&gmap->guest_table_lock);
2329		table = radix_tree_lookup(&gmap->host_to_guest,
2330					  vmaddr >> PMD_SHIFT);
2331		if (table)
2332			gaddr = __gmap_segment_gaddr(table) + offset;
2333		spin_unlock(&gmap->guest_table_lock);
2334		if (!table)
2335			continue;
2336
2337		if (!list_empty(&gmap->children) && (bits & PGSTE_VSIE_BIT)) {
2338			spin_lock(&gmap->shadow_lock);
2339			list_for_each_entry_safe(sg, next,
2340						 &gmap->children, list)
2341				gmap_shadow_notify(sg, vmaddr, gaddr);
2342			spin_unlock(&gmap->shadow_lock);
2343		}
2344		if (bits & PGSTE_IN_BIT)
2345			gmap_call_notifier(gmap, gaddr, gaddr + PAGE_SIZE - 1);
2346	}
2347	rcu_read_unlock();
2348}
2349EXPORT_SYMBOL_GPL(ptep_notify);
2350
2351static void pmdp_notify_gmap(struct gmap *gmap, pmd_t *pmdp,
2352			     unsigned long gaddr)
2353{
2354	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_IN)));
2355	gmap_call_notifier(gmap, gaddr, gaddr + HPAGE_SIZE - 1);
2356}
2357
2358/**
2359 * gmap_pmdp_xchg - exchange a gmap pmd with another
2360 * @gmap: pointer to the guest address space structure
2361 * @pmdp: pointer to the pmd entry
2362 * @new: replacement entry
2363 * @gaddr: the affected guest address
2364 *
2365 * This function is assumed to be called with the guest_table_lock
2366 * held.
2367 */
2368static void gmap_pmdp_xchg(struct gmap *gmap, pmd_t *pmdp, pmd_t new,
2369			   unsigned long gaddr)
2370{
2371	gaddr &= HPAGE_MASK;
2372	pmdp_notify_gmap(gmap, pmdp, gaddr);
2373	new = clear_pmd_bit(new, __pgprot(_SEGMENT_ENTRY_GMAP_IN));
2374	if (MACHINE_HAS_TLB_GUEST)
2375		__pmdp_idte(gaddr, (pmd_t *)pmdp, IDTE_GUEST_ASCE, gmap->asce,
2376			    IDTE_GLOBAL);
2377	else if (MACHINE_HAS_IDTE)
2378		__pmdp_idte(gaddr, (pmd_t *)pmdp, 0, 0, IDTE_GLOBAL);
2379	else
2380		__pmdp_csp(pmdp);
2381	set_pmd(pmdp, new);
2382}
2383
2384static void gmap_pmdp_clear(struct mm_struct *mm, unsigned long vmaddr,
2385			    int purge)
2386{
2387	pmd_t *pmdp;
2388	struct gmap *gmap;
2389	unsigned long gaddr;
2390
2391	rcu_read_lock();
2392	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2393		spin_lock(&gmap->guest_table_lock);
2394		pmdp = (pmd_t *)radix_tree_delete(&gmap->host_to_guest,
2395						  vmaddr >> PMD_SHIFT);
2396		if (pmdp) {
2397			gaddr = __gmap_segment_gaddr((unsigned long *)pmdp);
2398			pmdp_notify_gmap(gmap, pmdp, gaddr);
2399			WARN_ON(pmd_val(*pmdp) & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2400						   _SEGMENT_ENTRY_GMAP_UC |
2401						   _SEGMENT_ENTRY));
2402			if (purge)
2403				__pmdp_csp(pmdp);
2404			set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
2405		}
2406		spin_unlock(&gmap->guest_table_lock);
2407	}
2408	rcu_read_unlock();
2409}
2410
2411/**
2412 * gmap_pmdp_invalidate - invalidate all affected guest pmd entries without
2413 *                        flushing
2414 * @mm: pointer to the process mm_struct
2415 * @vmaddr: virtual address in the process address space
2416 */
2417void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr)
2418{
2419	gmap_pmdp_clear(mm, vmaddr, 0);
2420}
2421EXPORT_SYMBOL_GPL(gmap_pmdp_invalidate);
2422
2423/**
2424 * gmap_pmdp_csp - csp all affected guest pmd entries
2425 * @mm: pointer to the process mm_struct
2426 * @vmaddr: virtual address in the process address space
2427 */
2428void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr)
2429{
2430	gmap_pmdp_clear(mm, vmaddr, 1);
2431}
2432EXPORT_SYMBOL_GPL(gmap_pmdp_csp);
2433
2434/**
2435 * gmap_pmdp_idte_local - invalidate and clear a guest pmd entry
2436 * @mm: pointer to the process mm_struct
2437 * @vmaddr: virtual address in the process address space
2438 */
2439void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr)
2440{
2441	unsigned long *entry, gaddr;
2442	struct gmap *gmap;
2443	pmd_t *pmdp;
2444
2445	rcu_read_lock();
2446	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2447		spin_lock(&gmap->guest_table_lock);
2448		entry = radix_tree_delete(&gmap->host_to_guest,
2449					  vmaddr >> PMD_SHIFT);
2450		if (entry) {
2451			pmdp = (pmd_t *)entry;
2452			gaddr = __gmap_segment_gaddr(entry);
2453			pmdp_notify_gmap(gmap, pmdp, gaddr);
2454			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2455					   _SEGMENT_ENTRY_GMAP_UC |
2456					   _SEGMENT_ENTRY));
2457			if (MACHINE_HAS_TLB_GUEST)
2458				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2459					    gmap->asce, IDTE_LOCAL);
2460			else if (MACHINE_HAS_IDTE)
2461				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_LOCAL);
2462			*entry = _SEGMENT_ENTRY_EMPTY;
2463		}
2464		spin_unlock(&gmap->guest_table_lock);
2465	}
2466	rcu_read_unlock();
2467}
2468EXPORT_SYMBOL_GPL(gmap_pmdp_idte_local);
2469
2470/**
2471 * gmap_pmdp_idte_global - invalidate and clear a guest pmd entry
2472 * @mm: pointer to the process mm_struct
2473 * @vmaddr: virtual address in the process address space
2474 */
2475void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr)
2476{
2477	unsigned long *entry, gaddr;
2478	struct gmap *gmap;
2479	pmd_t *pmdp;
2480
2481	rcu_read_lock();
2482	list_for_each_entry_rcu(gmap, &mm->context.gmap_list, list) {
2483		spin_lock(&gmap->guest_table_lock);
2484		entry = radix_tree_delete(&gmap->host_to_guest,
2485					  vmaddr >> PMD_SHIFT);
2486		if (entry) {
2487			pmdp = (pmd_t *)entry;
2488			gaddr = __gmap_segment_gaddr(entry);
2489			pmdp_notify_gmap(gmap, pmdp, gaddr);
2490			WARN_ON(*entry & ~(_SEGMENT_ENTRY_HARDWARE_BITS_LARGE |
2491					   _SEGMENT_ENTRY_GMAP_UC |
2492					   _SEGMENT_ENTRY));
2493			if (MACHINE_HAS_TLB_GUEST)
2494				__pmdp_idte(gaddr, pmdp, IDTE_GUEST_ASCE,
2495					    gmap->asce, IDTE_GLOBAL);
2496			else if (MACHINE_HAS_IDTE)
2497				__pmdp_idte(gaddr, pmdp, 0, 0, IDTE_GLOBAL);
2498			else
2499				__pmdp_csp(pmdp);
2500			*entry = _SEGMENT_ENTRY_EMPTY;
2501		}
2502		spin_unlock(&gmap->guest_table_lock);
2503	}
2504	rcu_read_unlock();
2505}
2506EXPORT_SYMBOL_GPL(gmap_pmdp_idte_global);
2507
2508/**
2509 * gmap_test_and_clear_dirty_pmd - test and reset segment dirty status
2510 * @gmap: pointer to guest address space
2511 * @pmdp: pointer to the pmd to be tested
2512 * @gaddr: virtual address in the guest address space
2513 *
2514 * This function is assumed to be called with the guest_table_lock
2515 * held.
2516 */
2517static bool gmap_test_and_clear_dirty_pmd(struct gmap *gmap, pmd_t *pmdp,
2518					  unsigned long gaddr)
2519{
2520	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
2521		return false;
2522
2523	/* Already protected memory, which did not change is clean */
2524	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_PROTECT &&
2525	    !(pmd_val(*pmdp) & _SEGMENT_ENTRY_GMAP_UC))
2526		return false;
2527
2528	/* Clear UC indication and reset protection */
2529	set_pmd(pmdp, clear_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_GMAP_UC)));
2530	gmap_protect_pmd(gmap, gaddr, pmdp, PROT_READ, 0);
2531	return true;
2532}
2533
2534/**
2535 * gmap_sync_dirty_log_pmd - set bitmap based on dirty status of segment
2536 * @gmap: pointer to guest address space
2537 * @bitmap: dirty bitmap for this pmd
2538 * @gaddr: virtual address in the guest address space
2539 * @vmaddr: virtual address in the host address space
2540 *
2541 * This function is assumed to be called with the guest_table_lock
2542 * held.
2543 */
2544void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
2545			     unsigned long gaddr, unsigned long vmaddr)
2546{
2547	int i;
2548	pmd_t *pmdp;
2549	pte_t *ptep;
2550	spinlock_t *ptl;
2551
2552	pmdp = gmap_pmd_op_walk(gmap, gaddr);
2553	if (!pmdp)
2554		return;
2555
2556	if (pmd_leaf(*pmdp)) {
2557		if (gmap_test_and_clear_dirty_pmd(gmap, pmdp, gaddr))
2558			bitmap_fill(bitmap, _PAGE_ENTRIES);
2559	} else {
2560		for (i = 0; i < _PAGE_ENTRIES; i++, vmaddr += PAGE_SIZE) {
2561			ptep = pte_alloc_map_lock(gmap->mm, pmdp, vmaddr, &ptl);
2562			if (!ptep)
2563				continue;
2564			if (ptep_test_and_clear_uc(gmap->mm, vmaddr, ptep))
2565				set_bit(i, bitmap);
2566			pte_unmap_unlock(ptep, ptl);
2567		}
2568	}
2569	gmap_pmd_op_end(gmap, pmdp);
2570}
2571EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);
2572
2573#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2574static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
2575				    unsigned long end, struct mm_walk *walk)
2576{
2577	struct vm_area_struct *vma = walk->vma;
2578
2579	split_huge_pmd(vma, pmd, addr);
2580	return 0;
2581}
2582
2583static const struct mm_walk_ops thp_split_walk_ops = {
2584	.pmd_entry	= thp_split_walk_pmd_entry,
2585	.walk_lock	= PGWALK_WRLOCK_VERIFY,
2586};
2587
2588static inline void thp_split_mm(struct mm_struct *mm)
2589{
2590	struct vm_area_struct *vma;
2591	VMA_ITERATOR(vmi, mm, 0);
2592
2593	for_each_vma(vmi, vma) {
2594		vm_flags_mod(vma, VM_NOHUGEPAGE, VM_HUGEPAGE);
 
2595		walk_page_vma(vma, &thp_split_walk_ops, NULL);
2596	}
2597	mm->def_flags |= VM_NOHUGEPAGE;
2598}
2599#else
2600static inline void thp_split_mm(struct mm_struct *mm)
2601{
2602}
2603#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2604
2605/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2606 * switch on pgstes for its userspace process (for kvm)
2607 */
2608int s390_enable_sie(void)
2609{
2610	struct mm_struct *mm = current->mm;
2611
2612	/* Do we have pgstes? if yes, we are done */
2613	if (mm_has_pgste(mm))
2614		return 0;
2615	/* Fail if the page tables are 2K */
2616	if (!mm_alloc_pgste(mm))
2617		return -EINVAL;
2618	mmap_write_lock(mm);
2619	mm->context.has_pgste = 1;
2620	/* split thp mappings and disable thp for future mappings */
2621	thp_split_mm(mm);
 
2622	mmap_write_unlock(mm);
2623	return 0;
2624}
2625EXPORT_SYMBOL_GPL(s390_enable_sie);
2626
2627static int find_zeropage_pte_entry(pte_t *pte, unsigned long addr,
2628				   unsigned long end, struct mm_walk *walk)
2629{
2630	unsigned long *found_addr = walk->private;
2631
2632	/* Return 1 of the page is a zeropage. */
2633	if (is_zero_pfn(pte_pfn(*pte))) {
2634		/*
2635		 * Shared zeropage in e.g., a FS DAX mapping? We cannot do the
2636		 * right thing and likely don't care: FAULT_FLAG_UNSHARE
2637		 * currently only works in COW mappings, which is also where
2638		 * mm_forbids_zeropage() is checked.
2639		 */
2640		if (!is_cow_mapping(walk->vma->vm_flags))
2641			return -EFAULT;
2642
2643		*found_addr = addr;
2644		return 1;
2645	}
2646	return 0;
2647}
2648
2649static const struct mm_walk_ops find_zeropage_ops = {
2650	.pte_entry	= find_zeropage_pte_entry,
2651	.walk_lock	= PGWALK_WRLOCK,
2652};
2653
2654/*
2655 * Unshare all shared zeropages, replacing them by anonymous pages. Note that
2656 * we cannot simply zap all shared zeropages, because this could later
2657 * trigger unexpected userfaultfd missing events.
2658 *
2659 * This must be called after mm->context.allow_cow_sharing was
2660 * set to 0, to avoid future mappings of shared zeropages.
2661 *
2662 * mm contracts with s390, that even if mm were to remove a page table,
2663 * and racing with walk_page_range_vma() calling pte_offset_map_lock()
2664 * would fail, it will never insert a page table containing empty zero
2665 * pages once mm_forbids_zeropage(mm) i.e.
2666 * mm->context.allow_cow_sharing is set to 0.
2667 */
2668static int __s390_unshare_zeropages(struct mm_struct *mm)
2669{
 
2670	struct vm_area_struct *vma;
 
2671	VMA_ITERATOR(vmi, mm, 0);
2672	unsigned long addr;
2673	vm_fault_t fault;
2674	int rc;
2675
2676	for_each_vma(vmi, vma) {
2677		/*
2678		 * We could only look at COW mappings, but it's more future
2679		 * proof to catch unexpected zeropages in other mappings and
2680		 * fail.
2681		 */
2682		if ((vma->vm_flags & VM_PFNMAP) || is_vm_hugetlb_page(vma))
2683			continue;
2684		addr = vma->vm_start;
2685
2686retry:
2687		rc = walk_page_range_vma(vma, addr, vma->vm_end,
2688					 &find_zeropage_ops, &addr);
2689		if (rc < 0)
2690			return rc;
2691		else if (!rc)
2692			continue;
2693
2694		/* addr was updated by find_zeropage_pte_entry() */
2695		fault = handle_mm_fault(vma, addr,
2696					FAULT_FLAG_UNSHARE | FAULT_FLAG_REMOTE,
2697					NULL);
2698		if (fault & VM_FAULT_OOM)
2699			return -ENOMEM;
2700		/*
2701		 * See break_ksm(): even after handle_mm_fault() returned 0, we
2702		 * must start the lookup from the current address, because
2703		 * handle_mm_fault() may back out if there's any difficulty.
2704		 *
2705		 * VM_FAULT_SIGBUS and VM_FAULT_SIGSEGV are unexpected but
2706		 * maybe they could trigger in the future on concurrent
2707		 * truncation. In that case, the shared zeropage would be gone
2708		 * and we can simply retry and make progress.
2709		 */
2710		cond_resched();
2711		goto retry;
2712	}
2713
2714	return 0;
2715}
2716
2717static int __s390_disable_cow_sharing(struct mm_struct *mm)
2718{
2719	int rc;
2720
2721	if (!mm->context.allow_cow_sharing)
2722		return 0;
2723
2724	mm->context.allow_cow_sharing = 0;
2725
2726	/* Replace all shared zeropages by anonymous pages. */
2727	rc = __s390_unshare_zeropages(mm);
2728	/*
2729	 * Make sure to disable KSM (if enabled for the whole process or
2730	 * individual VMAs). Note that nothing currently hinders user space
2731	 * from re-enabling it.
2732	 */
2733	if (!rc)
2734		rc = ksm_disable(mm);
2735	if (rc)
2736		mm->context.allow_cow_sharing = 1;
2737	return rc;
2738}
2739
2740/*
2741 * Disable most COW-sharing of memory pages for the whole process:
2742 * (1) Disable KSM and unmerge/unshare any KSM pages.
2743 * (2) Disallow shared zeropages and unshare any zerpages that are mapped.
2744 *
2745 * Not that we currently don't bother with COW-shared pages that are shared
2746 * with parent/child processes due to fork().
2747 */
2748int s390_disable_cow_sharing(void)
2749{
2750	int rc;
2751
2752	mmap_write_lock(current->mm);
2753	rc = __s390_disable_cow_sharing(current->mm);
2754	mmap_write_unlock(current->mm);
2755	return rc;
2756}
2757EXPORT_SYMBOL_GPL(s390_disable_cow_sharing);
2758
2759/*
2760 * Enable storage key handling from now on and initialize the storage
2761 * keys with the default key.
2762 */
2763static int __s390_enable_skey_pte(pte_t *pte, unsigned long addr,
2764				  unsigned long next, struct mm_walk *walk)
2765{
2766	/* Clear storage key */
2767	ptep_zap_key(walk->mm, addr, pte);
2768	return 0;
2769}
2770
2771/*
2772 * Give a chance to schedule after setting a key to 256 pages.
2773 * We only hold the mm lock, which is a rwsem and the kvm srcu.
2774 * Both can sleep.
2775 */
2776static int __s390_enable_skey_pmd(pmd_t *pmd, unsigned long addr,
2777				  unsigned long next, struct mm_walk *walk)
2778{
2779	cond_resched();
2780	return 0;
2781}
2782
2783static int __s390_enable_skey_hugetlb(pte_t *pte, unsigned long addr,
2784				      unsigned long hmask, unsigned long next,
2785				      struct mm_walk *walk)
2786{
2787	pmd_t *pmd = (pmd_t *)pte;
2788	unsigned long start, end;
2789	struct folio *folio = page_folio(pmd_page(*pmd));
2790
2791	/*
2792	 * The write check makes sure we do not set a key on shared
2793	 * memory. This is needed as the walker does not differentiate
2794	 * between actual guest memory and the process executable or
2795	 * shared libraries.
2796	 */
2797	if (pmd_val(*pmd) & _SEGMENT_ENTRY_INVALID ||
2798	    !(pmd_val(*pmd) & _SEGMENT_ENTRY_WRITE))
2799		return 0;
2800
2801	start = pmd_val(*pmd) & HPAGE_MASK;
2802	end = start + HPAGE_SIZE;
2803	__storage_key_init_range(start, end);
2804	set_bit(PG_arch_1, &folio->flags);
2805	cond_resched();
2806	return 0;
2807}
2808
2809static const struct mm_walk_ops enable_skey_walk_ops = {
2810	.hugetlb_entry		= __s390_enable_skey_hugetlb,
2811	.pte_entry		= __s390_enable_skey_pte,
2812	.pmd_entry		= __s390_enable_skey_pmd,
2813	.walk_lock		= PGWALK_WRLOCK,
2814};
2815
2816int s390_enable_skey(void)
2817{
2818	struct mm_struct *mm = current->mm;
2819	int rc = 0;
2820
2821	mmap_write_lock(mm);
2822	if (mm_uses_skeys(mm))
2823		goto out_up;
2824
2825	mm->context.uses_skeys = 1;
2826	rc = __s390_disable_cow_sharing(mm);
2827	if (rc) {
2828		mm->context.uses_skeys = 0;
2829		goto out_up;
2830	}
2831	walk_page_range(mm, 0, TASK_SIZE, &enable_skey_walk_ops, NULL);
2832
2833out_up:
2834	mmap_write_unlock(mm);
2835	return rc;
2836}
2837EXPORT_SYMBOL_GPL(s390_enable_skey);
2838
2839/*
2840 * Reset CMMA state, make all pages stable again.
2841 */
2842static int __s390_reset_cmma(pte_t *pte, unsigned long addr,
2843			     unsigned long next, struct mm_walk *walk)
2844{
2845	ptep_zap_unused(walk->mm, addr, pte, 1);
2846	return 0;
2847}
2848
2849static const struct mm_walk_ops reset_cmma_walk_ops = {
2850	.pte_entry		= __s390_reset_cmma,
2851	.walk_lock		= PGWALK_WRLOCK,
2852};
2853
2854void s390_reset_cmma(struct mm_struct *mm)
2855{
2856	mmap_write_lock(mm);
2857	walk_page_range(mm, 0, TASK_SIZE, &reset_cmma_walk_ops, NULL);
2858	mmap_write_unlock(mm);
2859}
2860EXPORT_SYMBOL_GPL(s390_reset_cmma);
2861
2862#define GATHER_GET_PAGES 32
2863
2864struct reset_walk_state {
2865	unsigned long next;
2866	unsigned long count;
2867	unsigned long pfns[GATHER_GET_PAGES];
2868};
2869
2870static int s390_gather_pages(pte_t *ptep, unsigned long addr,
2871			     unsigned long next, struct mm_walk *walk)
2872{
2873	struct reset_walk_state *p = walk->private;
2874	pte_t pte = READ_ONCE(*ptep);
2875
2876	if (pte_present(pte)) {
2877		/* we have a reference from the mapping, take an extra one */
2878		get_page(phys_to_page(pte_val(pte)));
2879		p->pfns[p->count] = phys_to_pfn(pte_val(pte));
2880		p->next = next;
2881		p->count++;
2882	}
2883	return p->count >= GATHER_GET_PAGES;
2884}
2885
2886static const struct mm_walk_ops gather_pages_ops = {
2887	.pte_entry = s390_gather_pages,
2888	.walk_lock = PGWALK_RDLOCK,
2889};
2890
2891/*
2892 * Call the Destroy secure page UVC on each page in the given array of PFNs.
2893 * Each page needs to have an extra reference, which will be released here.
2894 */
2895void s390_uv_destroy_pfns(unsigned long count, unsigned long *pfns)
2896{
2897	struct folio *folio;
2898	unsigned long i;
2899
2900	for (i = 0; i < count; i++) {
2901		folio = pfn_folio(pfns[i]);
2902		/* we always have an extra reference */
2903		uv_destroy_folio(folio);
2904		/* get rid of the extra reference */
2905		folio_put(folio);
2906		cond_resched();
2907	}
2908}
2909EXPORT_SYMBOL_GPL(s390_uv_destroy_pfns);
2910
2911/**
2912 * __s390_uv_destroy_range - Call the destroy secure page UVC on each page
2913 * in the given range of the given address space.
2914 * @mm: the mm to operate on
2915 * @start: the start of the range
2916 * @end: the end of the range
2917 * @interruptible: if not 0, stop when a fatal signal is received
2918 *
2919 * Walk the given range of the given address space and call the destroy
2920 * secure page UVC on each page. Optionally exit early if a fatal signal is
2921 * pending.
2922 *
2923 * Return: 0 on success, -EINTR if the function stopped before completing
2924 */
2925int __s390_uv_destroy_range(struct mm_struct *mm, unsigned long start,
2926			    unsigned long end, bool interruptible)
2927{
2928	struct reset_walk_state state = { .next = start };
2929	int r = 1;
2930
2931	while (r > 0) {
2932		state.count = 0;
2933		mmap_read_lock(mm);
2934		r = walk_page_range(mm, state.next, end, &gather_pages_ops, &state);
2935		mmap_read_unlock(mm);
2936		cond_resched();
2937		s390_uv_destroy_pfns(state.count, state.pfns);
2938		if (interruptible && fatal_signal_pending(current))
2939			return -EINTR;
2940	}
2941	return 0;
2942}
2943EXPORT_SYMBOL_GPL(__s390_uv_destroy_range);
2944
2945/**
2946 * s390_unlist_old_asce - Remove the topmost level of page tables from the
2947 * list of page tables of the gmap.
2948 * @gmap: the gmap whose table is to be removed
2949 *
2950 * On s390x, KVM keeps a list of all pages containing the page tables of the
2951 * gmap (the CRST list). This list is used at tear down time to free all
2952 * pages that are now not needed anymore.
2953 *
2954 * This function removes the topmost page of the tree (the one pointed to by
2955 * the ASCE) from the CRST list.
2956 *
2957 * This means that it will not be freed when the VM is torn down, and needs
2958 * to be handled separately by the caller, unless a leak is actually
2959 * intended. Notice that this function will only remove the page from the
2960 * list, the page will still be used as a top level page table (and ASCE).
2961 */
2962void s390_unlist_old_asce(struct gmap *gmap)
2963{
2964	struct page *old;
2965
2966	old = virt_to_page(gmap->table);
2967	spin_lock(&gmap->guest_table_lock);
2968	list_del(&old->lru);
2969	/*
2970	 * Sometimes the topmost page might need to be "removed" multiple
2971	 * times, for example if the VM is rebooted into secure mode several
2972	 * times concurrently, or if s390_replace_asce fails after calling
2973	 * s390_remove_old_asce and is attempted again later. In that case
2974	 * the old asce has been removed from the list, and therefore it
2975	 * will not be freed when the VM terminates, but the ASCE is still
2976	 * in use and still pointed to.
2977	 * A subsequent call to replace_asce will follow the pointer and try
2978	 * to remove the same page from the list again.
2979	 * Therefore it's necessary that the page of the ASCE has valid
2980	 * pointers, so list_del can work (and do nothing) without
2981	 * dereferencing stale or invalid pointers.
2982	 */
2983	INIT_LIST_HEAD(&old->lru);
2984	spin_unlock(&gmap->guest_table_lock);
2985}
2986EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
2987
2988/**
2989 * s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
2990 * @gmap: the gmap whose ASCE needs to be replaced
2991 *
2992 * If the ASCE is a SEGMENT type then this function will return -EINVAL,
2993 * otherwise the pointers in the host_to_guest radix tree will keep pointing
2994 * to the wrong pages, causing use-after-free and memory corruption.
2995 * If the allocation of the new top level page table fails, the ASCE is not
2996 * replaced.
2997 * In any case, the old ASCE is always removed from the gmap CRST list.
2998 * Therefore the caller has to make sure to save a pointer to it
2999 * beforehand, unless a leak is actually intended.
3000 */
3001int s390_replace_asce(struct gmap *gmap)
3002{
3003	unsigned long asce;
3004	struct page *page;
3005	void *table;
3006
3007	s390_unlist_old_asce(gmap);
3008
3009	/* Replacing segment type ASCEs would cause serious issues */
3010	if ((gmap->asce & _ASCE_TYPE_MASK) == _ASCE_TYPE_SEGMENT)
3011		return -EINVAL;
3012
3013	page = gmap_alloc_crst();
3014	if (!page)
3015		return -ENOMEM;
3016	page->index = 0;
3017	table = page_to_virt(page);
3018	memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
3019
3020	/*
3021	 * The caller has to deal with the old ASCE, but here we make sure
3022	 * the new one is properly added to the CRST list, so that
3023	 * it will be freed when the VM is torn down.
3024	 */
3025	spin_lock(&gmap->guest_table_lock);
3026	list_add(&page->lru, &gmap->crst_list);
3027	spin_unlock(&gmap->guest_table_lock);
3028
3029	/* Set new table origin while preserving existing ASCE control bits */
3030	asce = (gmap->asce & ~_ASCE_ORIGIN) | __pa(table);
3031	WRITE_ONCE(gmap->asce, asce);
3032	WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
3033	WRITE_ONCE(gmap->table, table);
3034
3035	return 0;
3036}
3037EXPORT_SYMBOL_GPL(s390_replace_asce);
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);