mmu_pv.c

   1// SPDX-License-Identifier: GPL-2.0
   2
   3/*
   4 * Xen mmu operations
   5 *
   6 * This file contains the various mmu fetch and update operations.
   7 * The most important job they must perform is the mapping between the
   8 * domain's pfn and the overall machine mfns.
   9 *
  10 * Xen allows guests to directly update the pagetable, in a controlled
  11 * fashion.  In other words, the guest modifies the same pagetable
  12 * that the CPU actually uses, which eliminates the overhead of having
  13 * a separate shadow pagetable.
  14 *
  15 * In order to allow this, it falls on the guest domain to map its
  16 * notion of a "physical" pfn - which is just a domain-local linear
  17 * address - into a real "machine address" which the CPU's MMU can
  18 * use.
  19 *
  20 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
  21 * inserted directly into the pagetable.  When creating a new
  22 * pte/pmd/pgd, it converts the passed pfn into an mfn.  Conversely,
  23 * when reading the content back with __(pgd|pmd|pte)_val, it converts
  24 * the mfn back into a pfn.
  25 *
  26 * The other constraint is that all pages which make up a pagetable
  27 * must be mapped read-only in the guest.  This prevents uncontrolled
  28 * guest updates to the pagetable.  Xen strictly enforces this, and
  29 * will disallow any pagetable update which will end up mapping a
  30 * pagetable page RW, and will disallow using any writable page as a
  31 * pagetable.
  32 *
  33 * Naively, when loading %cr3 with the base of a new pagetable, Xen
  34 * would need to validate the whole pagetable before going on.
  35 * Naturally, this is quite slow.  The solution is to "pin" a
  36 * pagetable, which enforces all the constraints on the pagetable even
  37 * when it is not actively in use.  This means that Xen can be assured
  38 * that it is still valid when you do load it into %cr3, and doesn't
  39 * need to revalidate it.
  40 *
  41 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  42 */
  43#include <linux/sched/mm.h>
  44#include <linux/debugfs.h>
  45#include <linux/bug.h>
  46#include <linux/vmalloc.h>
  47#include <linux/export.h>
  48#include <linux/init.h>
  49#include <linux/gfp.h>
  50#include <linux/memblock.h>
  51#include <linux/seq_file.h>
  52#include <linux/crash_dump.h>
  53#include <linux/pgtable.h>
  54#ifdef CONFIG_KEXEC_CORE
  55#include <linux/kexec.h>
  56#endif
  57
  58#include <trace/events/xen.h>
  59
  60#include <asm/tlbflush.h>
  61#include <asm/fixmap.h>
  62#include <asm/mmu_context.h>
  63#include <asm/setup.h>
  64#include <asm/paravirt.h>
  65#include <asm/e820/api.h>
  66#include <asm/linkage.h>
  67#include <asm/page.h>
  68#include <asm/init.h>
  69#include <asm/memtype.h>
  70#include <asm/smp.h>
  71#include <asm/tlb.h>
  72
  73#include <asm/xen/hypercall.h>
  74#include <asm/xen/hypervisor.h>
  75
  76#include <xen/xen.h>
  77#include <xen/page.h>
  78#include <xen/interface/xen.h>
  79#include <xen/interface/hvm/hvm_op.h>
  80#include <xen/interface/version.h>
  81#include <xen/interface/memory.h>
  82#include <xen/hvc-console.h>
  83#include <xen/swiotlb-xen.h>
  84
  85#include "xen-ops.h"
  86
  87/*
  88 * Prototypes for functions called via PV_CALLEE_SAVE_REGS_THUNK() in order
  89 * to avoid warnings with "-Wmissing-prototypes".
  90 */
  91pteval_t xen_pte_val(pte_t pte);
  92pgdval_t xen_pgd_val(pgd_t pgd);
  93pmdval_t xen_pmd_val(pmd_t pmd);
  94pudval_t xen_pud_val(pud_t pud);
  95p4dval_t xen_p4d_val(p4d_t p4d);
  96pte_t xen_make_pte(pteval_t pte);
  97pgd_t xen_make_pgd(pgdval_t pgd);
  98pmd_t xen_make_pmd(pmdval_t pmd);
  99pud_t xen_make_pud(pudval_t pud);
 100p4d_t xen_make_p4d(p4dval_t p4d);
 101pte_t xen_make_pte_init(pteval_t pte);
 102
 103#ifdef CONFIG_X86_VSYSCALL_EMULATION
 104/* l3 pud for userspace vsyscall mapping */
 105static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
 106#endif
 107
 108/*
 109 * Protects atomic reservation decrease/increase against concurrent increases.
 110 * Also protects non-atomic updates of current_pages and balloon lists.
 111 */
 112static DEFINE_SPINLOCK(xen_reservation_lock);
 113
 114/* Protected by xen_reservation_lock. */
 115#define MIN_CONTIG_ORDER 9 /* 2MB */
 116static unsigned int discontig_frames_order = MIN_CONTIG_ORDER;
 117static unsigned long discontig_frames_early[1UL << MIN_CONTIG_ORDER] __initdata;
 118static unsigned long *discontig_frames __refdata = discontig_frames_early;
 119static bool discontig_frames_dyn;
 120
 121static int alloc_discontig_frames(unsigned int order)
 122{
 123	unsigned long *new_array, *old_array;
 124	unsigned int old_order;
 125	unsigned long flags;
 126
 127	BUG_ON(order < MIN_CONTIG_ORDER);
 128	BUILD_BUG_ON(sizeof(discontig_frames_early) != PAGE_SIZE);
 129
 130	new_array = (unsigned long *)__get_free_pages(GFP_KERNEL,
 131						      order - MIN_CONTIG_ORDER);
 132	if (!new_array)
 133		return -ENOMEM;
 134
 135	spin_lock_irqsave(&xen_reservation_lock, flags);
 136
 137	old_order = discontig_frames_order;
 138
 139	if (order > discontig_frames_order || !discontig_frames_dyn) {
 140		if (!discontig_frames_dyn)
 141			old_array = NULL;
 142		else
 143			old_array = discontig_frames;
 144
 145		discontig_frames = new_array;
 146		discontig_frames_order = order;
 147		discontig_frames_dyn = true;
 148	} else {
 149		old_array = new_array;
 150	}
 151
 152	spin_unlock_irqrestore(&xen_reservation_lock, flags);
 153
 154	free_pages((unsigned long)old_array, old_order - MIN_CONTIG_ORDER);
 155
 156	return 0;
 157}
 158
 159/*
 160 * Note about cr3 (pagetable base) values:
 161 *
 162 * xen_cr3 contains the current logical cr3 value; it contains the
 163 * last set cr3.  This may not be the current effective cr3, because
 164 * its update may be being lazily deferred.  However, a vcpu looking
 165 * at its own cr3 can use this value knowing that it everything will
 166 * be self-consistent.
 167 *
 168 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
 169 * hypercall to set the vcpu cr3 is complete (so it may be a little
 170 * out of date, but it will never be set early).  If one vcpu is
 171 * looking at another vcpu's cr3 value, it should use this variable.
 172 */
 173DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
 174static DEFINE_PER_CPU(unsigned long, xen_current_cr3);	/* actual vcpu cr3 */
 175
 176static phys_addr_t xen_pt_base, xen_pt_size __initdata;
 177
 178static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
 179
 180/*
 181 * Just beyond the highest usermode address.  STACK_TOP_MAX has a
 182 * redzone above it, so round it up to a PGD boundary.
 183 */
 184#define USER_LIMIT	((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
 185
 186void make_lowmem_page_readonly(void *vaddr)
 187{
 188	pte_t *pte, ptev;
 189	unsigned long address = (unsigned long)vaddr;
 190	unsigned int level;
 191
 192	pte = lookup_address(address, &level);
 193	if (pte == NULL)
 194		return;		/* vaddr missing */
 195
 196	ptev = pte_wrprotect(*pte);
 197
 198	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
 199		BUG();
 200}
 201
 202void make_lowmem_page_readwrite(void *vaddr)
 203{
 204	pte_t *pte, ptev;
 205	unsigned long address = (unsigned long)vaddr;
 206	unsigned int level;
 207
 208	pte = lookup_address(address, &level);
 209	if (pte == NULL)
 210		return;		/* vaddr missing */
 211
 212	ptev = pte_mkwrite_novma(*pte);
 213
 214	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
 215		BUG();
 216}
 217
 218
 219/*
 220 * During early boot all page table pages are pinned, but we do not have struct
 221 * pages, so return true until struct pages are ready.
 222 */
 223static bool xen_page_pinned(void *ptr)
 224{
 225	if (static_branch_likely(&xen_struct_pages_ready)) {
 226		struct page *page = virt_to_page(ptr);
 227
 228		return PagePinned(page);
 229	}
 230	return true;
 231}
 232
 233static void xen_extend_mmu_update(const struct mmu_update *update)
 234{
 235	struct multicall_space mcs;
 236	struct mmu_update *u;
 237
 238	mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
 239
 240	if (mcs.mc != NULL) {
 241		mcs.mc->args[1]++;
 242	} else {
 243		mcs = __xen_mc_entry(sizeof(*u));
 244		MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
 245	}
 246
 247	u = mcs.args;
 248	*u = *update;
 249}
 250
 251static void xen_extend_mmuext_op(const struct mmuext_op *op)
 252{
 253	struct multicall_space mcs;
 254	struct mmuext_op *u;
 255
 256	mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
 257
 258	if (mcs.mc != NULL) {
 259		mcs.mc->args[1]++;
 260	} else {
 261		mcs = __xen_mc_entry(sizeof(*u));
 262		MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
 263	}
 264
 265	u = mcs.args;
 266	*u = *op;
 267}
 268
 269static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
 270{
 271	struct mmu_update u;
 272
 273	preempt_disable();
 274
 275	xen_mc_batch();
 276
 277	/* ptr may be ioremapped for 64-bit pagetable setup */
 278	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
 279	u.val = pmd_val_ma(val);
 280	xen_extend_mmu_update(&u);
 281
 282	xen_mc_issue(XEN_LAZY_MMU);
 283
 284	preempt_enable();
 285}
 286
 287static void xen_set_pmd(pmd_t *ptr, pmd_t val)
 288{
 289	trace_xen_mmu_set_pmd(ptr, val);
 290
 291	/* If page is not pinned, we can just update the entry
 292	   directly */
 293	if (!xen_page_pinned(ptr)) {
 294		*ptr = val;
 295		return;
 296	}
 297
 298	xen_set_pmd_hyper(ptr, val);
 299}
 300
 301/*
 302 * Associate a virtual page frame with a given physical page frame
 303 * and protection flags for that frame.
 304 */
 305void __init set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
 306{
 307	if (HYPERVISOR_update_va_mapping(vaddr, mfn_pte(mfn, flags),
 308					 UVMF_INVLPG))
 309		BUG();
 310}
 311
 312static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
 313{
 314	struct mmu_update u;
 315
 316	if (xen_get_lazy_mode() != XEN_LAZY_MMU)
 317		return false;
 318
 319	xen_mc_batch();
 320
 321	u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
 322	u.val = pte_val_ma(pteval);
 323	xen_extend_mmu_update(&u);
 324
 325	xen_mc_issue(XEN_LAZY_MMU);
 326
 327	return true;
 328}
 329
 330static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
 331{
 332	if (!xen_batched_set_pte(ptep, pteval)) {
 333		/*
 334		 * Could call native_set_pte() here and trap and
 335		 * emulate the PTE write, but a hypercall is much cheaper.
 336		 */
 337		struct mmu_update u;
 338
 339		u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
 340		u.val = pte_val_ma(pteval);
 341		HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
 342	}
 343}
 344
 345static void xen_set_pte(pte_t *ptep, pte_t pteval)
 346{
 347	trace_xen_mmu_set_pte(ptep, pteval);
 348	__xen_set_pte(ptep, pteval);
 349}
 350
 351static pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
 352					unsigned long addr, pte_t *ptep)
 353{
 354	/* Just return the pte as-is.  We preserve the bits on commit */
 355	trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
 356	return *ptep;
 357}
 358
 359static void xen_ptep_modify_prot_commit(struct vm_area_struct *vma,
 360					unsigned long addr,
 361					pte_t *ptep, pte_t pte)
 362{
 363	struct mmu_update u;
 364
 365	trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
 366	xen_mc_batch();
 367
 368	u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
 369	u.val = pte_val_ma(pte);
 370	xen_extend_mmu_update(&u);
 371
 372	xen_mc_issue(XEN_LAZY_MMU);
 373}
 374
 375/* Assume pteval_t is equivalent to all the other *val_t types. */
 376static pteval_t pte_mfn_to_pfn(pteval_t val)
 377{
 378	if (val & _PAGE_PRESENT) {
 379		unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
 380		unsigned long pfn = mfn_to_pfn(mfn);
 381
 382		pteval_t flags = val & PTE_FLAGS_MASK;
 383		if (unlikely(pfn == ~0))
 384			val = flags & ~_PAGE_PRESENT;
 385		else
 386			val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
 387	}
 388
 389	return val;
 390}
 391
 392static pteval_t pte_pfn_to_mfn(pteval_t val)
 393{
 394	if (val & _PAGE_PRESENT) {
 395		unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
 396		pteval_t flags = val & PTE_FLAGS_MASK;
 397		unsigned long mfn;
 398
 399		mfn = __pfn_to_mfn(pfn);
 400
 401		/*
 402		 * If there's no mfn for the pfn, then just create an
 403		 * empty non-present pte.  Unfortunately this loses
 404		 * information about the original pfn, so
 405		 * pte_mfn_to_pfn is asymmetric.
 406		 */
 407		if (unlikely(mfn == INVALID_P2M_ENTRY)) {
 408			mfn = 0;
 409			flags = 0;
 410		} else
 411			mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
 412		val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
 413	}
 414
 415	return val;
 416}
 417
 418__visible pteval_t xen_pte_val(pte_t pte)
 419{
 420	pteval_t pteval = pte.pte;
 421
 422	return pte_mfn_to_pfn(pteval);
 423}
 424PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
 425
 426__visible pgdval_t xen_pgd_val(pgd_t pgd)
 427{
 428	return pte_mfn_to_pfn(pgd.pgd);
 429}
 430PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
 431
 432__visible pte_t xen_make_pte(pteval_t pte)
 433{
 434	pte = pte_pfn_to_mfn(pte);
 435
 436	return native_make_pte(pte);
 437}
 438PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
 439
 440__visible pgd_t xen_make_pgd(pgdval_t pgd)
 441{
 442	pgd = pte_pfn_to_mfn(pgd);
 443	return native_make_pgd(pgd);
 444}
 445PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
 446
 447__visible pmdval_t xen_pmd_val(pmd_t pmd)
 448{
 449	return pte_mfn_to_pfn(pmd.pmd);
 450}
 451PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
 452
 453static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
 454{
 455	struct mmu_update u;
 456
 457	preempt_disable();
 458
 459	xen_mc_batch();
 460
 461	/* ptr may be ioremapped for 64-bit pagetable setup */
 462	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
 463	u.val = pud_val_ma(val);
 464	xen_extend_mmu_update(&u);
 465
 466	xen_mc_issue(XEN_LAZY_MMU);
 467
 468	preempt_enable();
 469}
 470
 471static void xen_set_pud(pud_t *ptr, pud_t val)
 472{
 473	trace_xen_mmu_set_pud(ptr, val);
 474
 475	/* If page is not pinned, we can just update the entry
 476	   directly */
 477	if (!xen_page_pinned(ptr)) {
 478		*ptr = val;
 479		return;
 480	}
 481
 482	xen_set_pud_hyper(ptr, val);
 483}
 484
 485__visible pmd_t xen_make_pmd(pmdval_t pmd)
 486{
 487	pmd = pte_pfn_to_mfn(pmd);
 488	return native_make_pmd(pmd);
 489}
 490PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
 491
 492__visible pudval_t xen_pud_val(pud_t pud)
 493{
 494	return pte_mfn_to_pfn(pud.pud);
 495}
 496PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
 497
 498__visible pud_t xen_make_pud(pudval_t pud)
 499{
 500	pud = pte_pfn_to_mfn(pud);
 501
 502	return native_make_pud(pud);
 503}
 504PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
 505
 506static pgd_t *xen_get_user_pgd(pgd_t *pgd)
 507{
 508	pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
 509	unsigned offset = pgd - pgd_page;
 510	pgd_t *user_ptr = NULL;
 511
 512	if (offset < pgd_index(USER_LIMIT)) {
 513		struct page *page = virt_to_page(pgd_page);
 514		user_ptr = (pgd_t *)page->private;
 515		if (user_ptr)
 516			user_ptr += offset;
 517	}
 518
 519	return user_ptr;
 520}
 521
 522static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
 523{
 524	struct mmu_update u;
 525
 526	u.ptr = virt_to_machine(ptr).maddr;
 527	u.val = p4d_val_ma(val);
 528	xen_extend_mmu_update(&u);
 529}
 530
 531/*
 532 * Raw hypercall-based set_p4d, intended for in early boot before
 533 * there's a page structure.  This implies:
 534 *  1. The only existing pagetable is the kernel's
 535 *  2. It is always pinned
 536 *  3. It has no user pagetable attached to it
 537 */
 538static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
 539{
 540	preempt_disable();
 541
 542	xen_mc_batch();
 543
 544	__xen_set_p4d_hyper(ptr, val);
 545
 546	xen_mc_issue(XEN_LAZY_MMU);
 547
 548	preempt_enable();
 549}
 550
 551static void xen_set_p4d(p4d_t *ptr, p4d_t val)
 552{
 553	pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
 554	pgd_t pgd_val;
 555
 556	trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
 557
 558	/* If page is not pinned, we can just update the entry
 559	   directly */
 560	if (!xen_page_pinned(ptr)) {
 561		*ptr = val;
 562		if (user_ptr) {
 563			WARN_ON(xen_page_pinned(user_ptr));
 564			pgd_val.pgd = p4d_val_ma(val);
 565			*user_ptr = pgd_val;
 566		}
 567		return;
 568	}
 569
 570	/* If it's pinned, then we can at least batch the kernel and
 571	   user updates together. */
 572	xen_mc_batch();
 573
 574	__xen_set_p4d_hyper(ptr, val);
 575	if (user_ptr)
 576		__xen_set_p4d_hyper((p4d_t *)user_ptr, val);
 577
 578	xen_mc_issue(XEN_LAZY_MMU);
 579}
 580
 581__visible p4dval_t xen_p4d_val(p4d_t p4d)
 582{
 583	return pte_mfn_to_pfn(p4d.p4d);
 584}
 585PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
 586
 587__visible p4d_t xen_make_p4d(p4dval_t p4d)
 588{
 589	p4d = pte_pfn_to_mfn(p4d);
 590
 591	return native_make_p4d(p4d);
 592}
 593PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
 594
 595static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
 596			 void (*func)(struct mm_struct *mm, struct page *,
 597				      enum pt_level),
 598			 bool last, unsigned long limit)
 599{
 600	int i, nr;
 601
 602	nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
 603	for (i = 0; i < nr; i++) {
 604		if (!pmd_none(pmd[i]))
 605			(*func)(mm, pmd_page(pmd[i]), PT_PTE);
 606	}
 607}
 608
 609static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
 610			 void (*func)(struct mm_struct *mm, struct page *,
 611				      enum pt_level),
 612			 bool last, unsigned long limit)
 613{
 614	int i, nr;
 615
 616	nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
 617	for (i = 0; i < nr; i++) {
 618		pmd_t *pmd;
 619
 620		if (pud_none(pud[i]))
 621			continue;
 622
 623		pmd = pmd_offset(&pud[i], 0);
 624		if (PTRS_PER_PMD > 1)
 625			(*func)(mm, virt_to_page(pmd), PT_PMD);
 626		xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
 627	}
 628}
 629
 630static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
 631			 void (*func)(struct mm_struct *mm, struct page *,
 632				      enum pt_level),
 633			 bool last, unsigned long limit)
 634{
 635	pud_t *pud;
 636
 637
 638	if (p4d_none(*p4d))
 639		return;
 640
 641	pud = pud_offset(p4d, 0);
 642	if (PTRS_PER_PUD > 1)
 643		(*func)(mm, virt_to_page(pud), PT_PUD);
 644	xen_pud_walk(mm, pud, func, last, limit);
 645}
 646
 647/*
 648 * (Yet another) pagetable walker.  This one is intended for pinning a
 649 * pagetable.  This means that it walks a pagetable and calls the
 650 * callback function on each page it finds making up the page table,
 651 * at every level.  It walks the entire pagetable, but it only bothers
 652 * pinning pte pages which are below limit.  In the normal case this
 653 * will be STACK_TOP_MAX, but at boot we need to pin up to
 654 * FIXADDR_TOP.
 655 *
 656 * We must skip the Xen hole in the middle of the address space, just after
 657 * the big x86-64 virtual hole.
 658 */
 659static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
 660			   void (*func)(struct mm_struct *mm, struct page *,
 661					enum pt_level),
 662			   unsigned long limit)
 663{
 664	int i, nr;
 665	unsigned hole_low = 0, hole_high = 0;
 666
 667	/* The limit is the last byte to be touched */
 668	limit--;
 669	BUG_ON(limit >= FIXADDR_TOP);
 670
 671	/*
 672	 * 64-bit has a great big hole in the middle of the address
 673	 * space, which contains the Xen mappings.
 674	 */
 675	hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
 676	hole_high = pgd_index(GUARD_HOLE_END_ADDR);
 677
 678	nr = pgd_index(limit) + 1;
 679	for (i = 0; i < nr; i++) {
 680		p4d_t *p4d;
 681
 682		if (i >= hole_low && i < hole_high)
 683			continue;
 684
 685		if (pgd_none(pgd[i]))
 686			continue;
 687
 688		p4d = p4d_offset(&pgd[i], 0);
 689		xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
 690	}
 691
 692	/* Do the top level last, so that the callbacks can use it as
 693	   a cue to do final things like tlb flushes. */
 694	(*func)(mm, virt_to_page(pgd), PT_PGD);
 695}
 696
 697static void xen_pgd_walk(struct mm_struct *mm,
 698			 void (*func)(struct mm_struct *mm, struct page *,
 699				      enum pt_level),
 700			 unsigned long limit)
 701{
 702	__xen_pgd_walk(mm, mm->pgd, func, limit);
 703}
 704
 705/* If we're using split pte locks, then take the page's lock and
 706   return a pointer to it.  Otherwise return NULL. */
 707static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
 708{
 709	spinlock_t *ptl = NULL;
 710
 711#if defined(CONFIG_SPLIT_PTE_PTLOCKS)
 712	ptl = ptlock_ptr(page_ptdesc(page));
 713	spin_lock_nest_lock(ptl, &mm->page_table_lock);
 714#endif
 715
 716	return ptl;
 717}
 718
 719static void xen_pte_unlock(void *v)
 720{
 721	spinlock_t *ptl = v;
 722	spin_unlock(ptl);
 723}
 724
 725static void xen_do_pin(unsigned level, unsigned long pfn)
 726{
 727	struct mmuext_op op;
 728
 729	op.cmd = level;
 730	op.arg1.mfn = pfn_to_mfn(pfn);
 731
 732	xen_extend_mmuext_op(&op);
 733}
 734
 735static void xen_pin_page(struct mm_struct *mm, struct page *page,
 736			 enum pt_level level)
 737{
 738	unsigned pgfl = TestSetPagePinned(page);
 739
 740	if (!pgfl) {
 741		void *pt = lowmem_page_address(page);
 742		unsigned long pfn = page_to_pfn(page);
 743		struct multicall_space mcs = __xen_mc_entry(0);
 744		spinlock_t *ptl;
 745
 746		/*
 747		 * We need to hold the pagetable lock between the time
 748		 * we make the pagetable RO and when we actually pin
 749		 * it.  If we don't, then other users may come in and
 750		 * attempt to update the pagetable by writing it,
 751		 * which will fail because the memory is RO but not
 752		 * pinned, so Xen won't do the trap'n'emulate.
 753		 *
 754		 * If we're using split pte locks, we can't hold the
 755		 * entire pagetable's worth of locks during the
 756		 * traverse, because we may wrap the preempt count (8
 757		 * bits).  The solution is to mark RO and pin each PTE
 758		 * page while holding the lock.  This means the number
 759		 * of locks we end up holding is never more than a
 760		 * batch size (~32 entries, at present).
 761		 *
 762		 * If we're not using split pte locks, we needn't pin
 763		 * the PTE pages independently, because we're
 764		 * protected by the overall pagetable lock.
 765		 */
 766		ptl = NULL;
 767		if (level == PT_PTE)
 768			ptl = xen_pte_lock(page, mm);
 769
 770		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
 771					pfn_pte(pfn, PAGE_KERNEL_RO),
 772					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
 773
 774		if (ptl) {
 775			xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
 776
 777			/* Queue a deferred unlock for when this batch
 778			   is completed. */
 779			xen_mc_callback(xen_pte_unlock, ptl);
 780		}
 781	}
 782}
 783
 784/* This is called just after a mm has been created, but it has not
 785   been used yet.  We need to make sure that its pagetable is all
 786   read-only, and can be pinned. */
 787static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
 788{
 789	pgd_t *user_pgd = xen_get_user_pgd(pgd);
 790
 791	trace_xen_mmu_pgd_pin(mm, pgd);
 792
 793	xen_mc_batch();
 794
 795	__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
 796
 797	xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
 798
 799	if (user_pgd) {
 800		xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
 801		xen_do_pin(MMUEXT_PIN_L4_TABLE,
 802			   PFN_DOWN(__pa(user_pgd)));
 803	}
 804
 805	xen_mc_issue(0);
 806}
 807
 808static void xen_pgd_pin(struct mm_struct *mm)
 809{
 810	__xen_pgd_pin(mm, mm->pgd);
 811}
 812
 813/*
 814 * On save, we need to pin all pagetables to make sure they get their
 815 * mfns turned into pfns.  Search the list for any unpinned pgds and pin
 816 * them (unpinned pgds are not currently in use, probably because the
 817 * process is under construction or destruction).
 818 *
 819 * Expected to be called in stop_machine() ("equivalent to taking
 820 * every spinlock in the system"), so the locking doesn't really
 821 * matter all that much.
 822 */
 823void xen_mm_pin_all(void)
 824{
 825	struct page *page;
 826
 827	spin_lock(&init_mm.page_table_lock);
 828	spin_lock(&pgd_lock);
 829
 830	list_for_each_entry(page, &pgd_list, lru) {
 831		if (!PagePinned(page)) {
 832			__xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
 833			SetPageSavePinned(page);
 834		}
 835	}
 836
 837	spin_unlock(&pgd_lock);
 838	spin_unlock(&init_mm.page_table_lock);
 839}
 840
 841static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
 842				   enum pt_level level)
 843{
 844	SetPagePinned(page);
 845}
 846
 847/*
 848 * The init_mm pagetable is really pinned as soon as its created, but
 849 * that's before we have page structures to store the bits.  So do all
 850 * the book-keeping now once struct pages for allocated pages are
 851 * initialized. This happens only after memblock_free_all() is called.
 852 */
 853static void __init xen_after_bootmem(void)
 854{
 855	static_branch_enable(&xen_struct_pages_ready);
 856#ifdef CONFIG_X86_VSYSCALL_EMULATION
 857	SetPagePinned(virt_to_page(level3_user_vsyscall));
 858#endif
 859	xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
 860
 861	if (alloc_discontig_frames(MIN_CONTIG_ORDER))
 862		BUG();
 863}
 864
 865static void xen_unpin_page(struct mm_struct *mm, struct page *page,
 866			   enum pt_level level)
 867{
 868	unsigned pgfl = TestClearPagePinned(page);
 869
 870	if (pgfl) {
 871		void *pt = lowmem_page_address(page);
 872		unsigned long pfn = page_to_pfn(page);
 873		spinlock_t *ptl = NULL;
 874		struct multicall_space mcs;
 875
 876		/*
 877		 * Do the converse to pin_page.  If we're using split
 878		 * pte locks, we must be holding the lock for while
 879		 * the pte page is unpinned but still RO to prevent
 880		 * concurrent updates from seeing it in this
 881		 * partially-pinned state.
 882		 */
 883		if (level == PT_PTE) {
 884			ptl = xen_pte_lock(page, mm);
 885
 886			if (ptl)
 887				xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
 888		}
 889
 890		mcs = __xen_mc_entry(0);
 891
 892		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
 893					pfn_pte(pfn, PAGE_KERNEL),
 894					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
 895
 896		if (ptl) {
 897			/* unlock when batch completed */
 898			xen_mc_callback(xen_pte_unlock, ptl);
 899		}
 900	}
 901}
 902
 903/* Release a pagetables pages back as normal RW */
 904static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
 905{
 906	pgd_t *user_pgd = xen_get_user_pgd(pgd);
 907
 908	trace_xen_mmu_pgd_unpin(mm, pgd);
 909
 910	xen_mc_batch();
 911
 912	xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
 913
 914	if (user_pgd) {
 915		xen_do_pin(MMUEXT_UNPIN_TABLE,
 916			   PFN_DOWN(__pa(user_pgd)));
 917		xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
 918	}
 919
 920	__xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
 921
 922	xen_mc_issue(0);
 923}
 924
 925static void xen_pgd_unpin(struct mm_struct *mm)
 926{
 927	__xen_pgd_unpin(mm, mm->pgd);
 928}
 929
 930/*
 931 * On resume, undo any pinning done at save, so that the rest of the
 932 * kernel doesn't see any unexpected pinned pagetables.
 933 */
 934void xen_mm_unpin_all(void)
 935{
 936	struct page *page;
 937
 938	spin_lock(&init_mm.page_table_lock);
 939	spin_lock(&pgd_lock);
 940
 941	list_for_each_entry(page, &pgd_list, lru) {
 942		if (PageSavePinned(page)) {
 943			BUG_ON(!PagePinned(page));
 944			__xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
 945			ClearPageSavePinned(page);
 946		}
 947	}
 948
 949	spin_unlock(&pgd_lock);
 950	spin_unlock(&init_mm.page_table_lock);
 951}
 952
 953static void xen_enter_mmap(struct mm_struct *mm)
 954{
 955	spin_lock(&mm->page_table_lock);
 956	xen_pgd_pin(mm);
 957	spin_unlock(&mm->page_table_lock);
 958}
 959
 960static void drop_mm_ref_this_cpu(void *info)
 961{
 962	struct mm_struct *mm = info;
 963
 964	if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
 965		leave_mm();
 966
 967	/*
 968	 * If this cpu still has a stale cr3 reference, then make sure
 969	 * it has been flushed.
 970	 */
 971	if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
 972		xen_mc_flush();
 973}
 974
 975#ifdef CONFIG_SMP
 976/*
 977 * Another cpu may still have their %cr3 pointing at the pagetable, so
 978 * we need to repoint it somewhere else before we can unpin it.
 979 */
 980static void xen_drop_mm_ref(struct mm_struct *mm)
 981{
 982	cpumask_var_t mask;
 983	unsigned cpu;
 984
 985	drop_mm_ref_this_cpu(mm);
 986
 987	/* Get the "official" set of cpus referring to our pagetable. */
 988	if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
 989		for_each_online_cpu(cpu) {
 990			if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
 991				continue;
 992			smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
 993		}
 994		return;
 995	}
 996
 997	/*
 998	 * It's possible that a vcpu may have a stale reference to our
 999	 * cr3, because its in lazy mode, and it hasn't yet flushed
1000	 * its set of pending hypercalls yet.  In this case, we can
1001	 * look at its actual current cr3 value, and force it to flush
1002	 * if needed.
1003	 */
1004	cpumask_clear(mask);
1005	for_each_online_cpu(cpu) {
1006		if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1007			cpumask_set_cpu(cpu, mask);
1008	}
1009
1010	smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
1011	free_cpumask_var(mask);
1012}
1013#else
1014static void xen_drop_mm_ref(struct mm_struct *mm)
1015{
1016	drop_mm_ref_this_cpu(mm);
1017}
1018#endif
1019
1020/*
1021 * While a process runs, Xen pins its pagetables, which means that the
1022 * hypervisor forces it to be read-only, and it controls all updates
1023 * to it.  This means that all pagetable updates have to go via the
1024 * hypervisor, which is moderately expensive.
1025 *
1026 * Since we're pulling the pagetable down, we switch to use init_mm,
1027 * unpin old process pagetable and mark it all read-write, which
1028 * allows further operations on it to be simple memory accesses.
1029 *
1030 * The only subtle point is that another CPU may be still using the
1031 * pagetable because of lazy tlb flushing.  This means we need need to
1032 * switch all CPUs off this pagetable before we can unpin it.
1033 */
1034static void xen_exit_mmap(struct mm_struct *mm)
1035{
1036	get_cpu();		/* make sure we don't move around */
1037	xen_drop_mm_ref(mm);
1038	put_cpu();
1039
1040	spin_lock(&mm->page_table_lock);
1041
1042	/* pgd may not be pinned in the error exit path of execve */
1043	if (xen_page_pinned(mm->pgd))
1044		xen_pgd_unpin(mm);
1045
1046	spin_unlock(&mm->page_table_lock);
1047}
1048
1049static void xen_post_allocator_init(void);
1050
1051static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1052{
1053	struct mmuext_op op;
1054
1055	op.cmd = cmd;
1056	op.arg1.mfn = pfn_to_mfn(pfn);
1057	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1058		BUG();
1059}
1060
1061static void __init xen_cleanhighmap(unsigned long vaddr,
1062				    unsigned long vaddr_end)
1063{
1064	unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1065	pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1066
1067	/* NOTE: The loop is more greedy than the cleanup_highmap variant.
1068	 * We include the PMD passed in on _both_ boundaries. */
1069	for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1070			pmd++, vaddr += PMD_SIZE) {
1071		if (pmd_none(*pmd))
1072			continue;
1073		if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1074			set_pmd(pmd, __pmd(0));
1075	}
1076	/* In case we did something silly, we should crash in this function
1077	 * instead of somewhere later and be confusing. */
1078	xen_mc_flush();
1079}
1080
1081/*
1082 * Make a page range writeable and free it.
1083 */
1084static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1085{
1086	void *vaddr = __va(paddr);
1087	void *vaddr_end = vaddr + size;
1088
1089	for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1090		make_lowmem_page_readwrite(vaddr);
1091
1092	memblock_phys_free(paddr, size);
1093}
1094
1095static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1096{
1097	unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1098
1099	if (unpin)
1100		pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1101	ClearPagePinned(virt_to_page(__va(pa)));
1102	xen_free_ro_pages(pa, PAGE_SIZE);
1103}
1104
1105static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1106{
1107	unsigned long pa;
1108	pte_t *pte_tbl;
1109	int i;
1110
1111	if (pmd_leaf(*pmd)) {
1112		pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1113		xen_free_ro_pages(pa, PMD_SIZE);
1114		return;
1115	}
1116
1117	pte_tbl = pte_offset_kernel(pmd, 0);
1118	for (i = 0; i < PTRS_PER_PTE; i++) {
1119		if (pte_none(pte_tbl[i]))
1120			continue;
1121		pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1122		xen_free_ro_pages(pa, PAGE_SIZE);
1123	}
1124	set_pmd(pmd, __pmd(0));
1125	xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1126}
1127
1128static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1129{
1130	unsigned long pa;
1131	pmd_t *pmd_tbl;
1132	int i;
1133
1134	if (pud_leaf(*pud)) {
1135		pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1136		xen_free_ro_pages(pa, PUD_SIZE);
1137		return;
1138	}
1139
1140	pmd_tbl = pmd_offset(pud, 0);
1141	for (i = 0; i < PTRS_PER_PMD; i++) {
1142		if (pmd_none(pmd_tbl[i]))
1143			continue;
1144		xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1145	}
1146	set_pud(pud, __pud(0));
1147	xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1148}
1149
1150static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1151{
1152	unsigned long pa;
1153	pud_t *pud_tbl;
1154	int i;
1155
1156	if (p4d_leaf(*p4d)) {
1157		pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1158		xen_free_ro_pages(pa, P4D_SIZE);
1159		return;
1160	}
1161
1162	pud_tbl = pud_offset(p4d, 0);
1163	for (i = 0; i < PTRS_PER_PUD; i++) {
1164		if (pud_none(pud_tbl[i]))
1165			continue;
1166		xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1167	}
1168	set_p4d(p4d, __p4d(0));
1169	xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1170}
1171
1172/*
1173 * Since it is well isolated we can (and since it is perhaps large we should)
1174 * also free the page tables mapping the initial P->M table.
1175 */
1176static void __init xen_cleanmfnmap(unsigned long vaddr)
1177{
1178	pgd_t *pgd;
1179	p4d_t *p4d;
1180	bool unpin;
1181
1182	unpin = (vaddr == 2 * PGDIR_SIZE);
1183	vaddr &= PMD_MASK;
1184	pgd = pgd_offset_k(vaddr);
1185	p4d = p4d_offset(pgd, 0);
1186	if (!p4d_none(*p4d))
1187		xen_cleanmfnmap_p4d(p4d, unpin);
1188}
1189
1190static void __init xen_pagetable_p2m_free(void)
1191{
1192	unsigned long size;
1193	unsigned long addr;
1194
1195	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1196
1197	/* No memory or already called. */
1198	if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1199		return;
1200
1201	/* using __ka address and sticking INVALID_P2M_ENTRY! */
1202	memset((void *)xen_start_info->mfn_list, 0xff, size);
1203
1204	addr = xen_start_info->mfn_list;
1205	/*
1206	 * We could be in __ka space.
1207	 * We roundup to the PMD, which means that if anybody at this stage is
1208	 * using the __ka address of xen_start_info or
1209	 * xen_start_info->shared_info they are in going to crash. Fortunately
1210	 * we have already revectored in xen_setup_kernel_pagetable.
1211	 */
1212	size = roundup(size, PMD_SIZE);
1213
1214	if (addr >= __START_KERNEL_map) {
1215		xen_cleanhighmap(addr, addr + size);
1216		size = PAGE_ALIGN(xen_start_info->nr_pages *
1217				  sizeof(unsigned long));
1218		memblock_free((void *)addr, size);
1219	} else {
1220		xen_cleanmfnmap(addr);
1221	}
1222}
1223
1224static void __init xen_pagetable_cleanhighmap(void)
1225{
1226	unsigned long size;
1227	unsigned long addr;
1228
1229	/* At this stage, cleanup_highmap has already cleaned __ka space
1230	 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1231	 * the ramdisk). We continue on, erasing PMD entries that point to page
1232	 * tables - do note that they are accessible at this stage via __va.
1233	 * As Xen is aligning the memory end to a 4MB boundary, for good
1234	 * measure we also round up to PMD_SIZE * 2 - which means that if
1235	 * anybody is using __ka address to the initial boot-stack - and try
1236	 * to use it - they are going to crash. The xen_start_info has been
1237	 * taken care of already in xen_setup_kernel_pagetable. */
1238	addr = xen_start_info->pt_base;
1239	size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1240
1241	xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1242	xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1243}
1244
1245static void __init xen_pagetable_p2m_setup(void)
1246{
1247	xen_vmalloc_p2m_tree();
1248
1249	xen_pagetable_p2m_free();
1250
1251	xen_pagetable_cleanhighmap();
1252
1253	/* And revector! Bye bye old array */
1254	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1255}
1256
1257static void __init xen_pagetable_init(void)
1258{
1259	/*
1260	 * The majority of further PTE writes is to pagetables already
1261	 * announced as such to Xen. Hence it is more efficient to use
1262	 * hypercalls for these updates.
1263	 */
1264	pv_ops.mmu.set_pte = __xen_set_pte;
1265
1266	paging_init();
1267	xen_post_allocator_init();
1268
1269	xen_pagetable_p2m_setup();
1270
1271	/* Allocate and initialize top and mid mfn levels for p2m structure */
1272	xen_build_mfn_list_list();
1273
1274	/* Remap memory freed due to conflicts with E820 map */
1275	xen_remap_memory();
1276	xen_setup_mfn_list_list();
1277}
1278
1279static noinstr void xen_write_cr2(unsigned long cr2)
1280{
1281	this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1282}
1283
1284static noinline void xen_flush_tlb(void)
1285{
1286	struct mmuext_op *op;
1287	struct multicall_space mcs;
1288
1289	preempt_disable();
1290
1291	mcs = xen_mc_entry(sizeof(*op));
1292
1293	op = mcs.args;
1294	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1295	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1296
1297	xen_mc_issue(XEN_LAZY_MMU);
1298
1299	preempt_enable();
1300}
1301
1302static void xen_flush_tlb_one_user(unsigned long addr)
1303{
1304	struct mmuext_op *op;
1305	struct multicall_space mcs;
1306
1307	trace_xen_mmu_flush_tlb_one_user(addr);
1308
1309	preempt_disable();
1310
1311	mcs = xen_mc_entry(sizeof(*op));
1312	op = mcs.args;
1313	op->cmd = MMUEXT_INVLPG_LOCAL;
1314	op->arg1.linear_addr = addr & PAGE_MASK;
1315	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1316
1317	xen_mc_issue(XEN_LAZY_MMU);
1318
1319	preempt_enable();
1320}
1321
1322static void xen_flush_tlb_multi(const struct cpumask *cpus,
1323				const struct flush_tlb_info *info)
1324{
1325	struct {
1326		struct mmuext_op op;
1327		DECLARE_BITMAP(mask, NR_CPUS);
1328	} *args;
1329	struct multicall_space mcs;
1330	const size_t mc_entry_size = sizeof(args->op) +
1331		sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1332
1333	trace_xen_mmu_flush_tlb_multi(cpus, info->mm, info->start, info->end);
1334
1335	if (cpumask_empty(cpus))
1336		return;		/* nothing to do */
1337
1338	mcs = xen_mc_entry(mc_entry_size);
1339	args = mcs.args;
1340	args->op.arg2.vcpumask = to_cpumask(args->mask);
1341
1342	/* Remove any offline CPUs */
1343	cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1344
1345	args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1346	if (info->end != TLB_FLUSH_ALL &&
1347	    (info->end - info->start) <= PAGE_SIZE) {
1348		args->op.cmd = MMUEXT_INVLPG_MULTI;
1349		args->op.arg1.linear_addr = info->start;
1350	}
1351
1352	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1353
1354	xen_mc_issue(XEN_LAZY_MMU);
1355}
1356
1357static unsigned long xen_read_cr3(void)
1358{
1359	return this_cpu_read(xen_cr3);
1360}
1361
1362static void set_current_cr3(void *v)
1363{
1364	this_cpu_write(xen_current_cr3, (unsigned long)v);
1365}
1366
1367static void __xen_write_cr3(bool kernel, unsigned long cr3)
1368{
1369	struct mmuext_op op;
1370	unsigned long mfn;
1371
1372	trace_xen_mmu_write_cr3(kernel, cr3);
1373
1374	if (cr3)
1375		mfn = pfn_to_mfn(PFN_DOWN(cr3));
1376	else
1377		mfn = 0;
1378
1379	WARN_ON(mfn == 0 && kernel);
1380
1381	op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1382	op.arg1.mfn = mfn;
1383
1384	xen_extend_mmuext_op(&op);
1385
1386	if (kernel) {
1387		this_cpu_write(xen_cr3, cr3);
1388
1389		/* Update xen_current_cr3 once the batch has actually
1390		   been submitted. */
1391		xen_mc_callback(set_current_cr3, (void *)cr3);
1392	}
1393}
1394static void xen_write_cr3(unsigned long cr3)
1395{
1396	pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1397
1398	BUG_ON(preemptible());
1399
1400	xen_mc_batch();  /* disables interrupts */
1401
1402	/* Update while interrupts are disabled, so its atomic with
1403	   respect to ipis */
1404	this_cpu_write(xen_cr3, cr3);
1405
1406	__xen_write_cr3(true, cr3);
1407
1408	if (user_pgd)
1409		__xen_write_cr3(false, __pa(user_pgd));
1410	else
1411		__xen_write_cr3(false, 0);
1412
1413	xen_mc_issue(XEN_LAZY_CPU);  /* interrupts restored */
1414}
1415
1416/*
1417 * At the start of the day - when Xen launches a guest, it has already
1418 * built pagetables for the guest. We diligently look over them
1419 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1420 * init_top_pgt and its friends. Then when we are happy we load
1421 * the new init_top_pgt - and continue on.
1422 *
1423 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1424 * up the rest of the pagetables. When it has completed it loads the cr3.
1425 * N.B. that baremetal would start at 'start_kernel' (and the early
1426 * #PF handler would create bootstrap pagetables) - so we are running
1427 * with the same assumptions as what to do when write_cr3 is executed
1428 * at this point.
1429 *
1430 * Since there are no user-page tables at all, we have two variants
1431 * of xen_write_cr3 - the early bootup (this one), and the late one
1432 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1433 * the Linux kernel and user-space are both in ring 3 while the
1434 * hypervisor is in ring 0.
1435 */
1436static void __init xen_write_cr3_init(unsigned long cr3)
1437{
1438	BUG_ON(preemptible());
1439
1440	xen_mc_batch();  /* disables interrupts */
1441
1442	/* Update while interrupts are disabled, so its atomic with
1443	   respect to ipis */
1444	this_cpu_write(xen_cr3, cr3);
1445
1446	__xen_write_cr3(true, cr3);
1447
1448	xen_mc_issue(XEN_LAZY_CPU);  /* interrupts restored */
1449}
1450
1451static int xen_pgd_alloc(struct mm_struct *mm)
1452{
1453	pgd_t *pgd = mm->pgd;
1454	struct page *page = virt_to_page(pgd);
1455	pgd_t *user_pgd;
1456	int ret = -ENOMEM;
1457
1458	BUG_ON(PagePinned(virt_to_page(pgd)));
1459	BUG_ON(page->private != 0);
1460
1461	user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1462	page->private = (unsigned long)user_pgd;
1463
1464	if (user_pgd != NULL) {
1465#ifdef CONFIG_X86_VSYSCALL_EMULATION
1466		user_pgd[pgd_index(VSYSCALL_ADDR)] =
1467			__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1468#endif
1469		ret = 0;
1470	}
1471
1472	BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1473
1474	return ret;
1475}
1476
1477static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1478{
1479	pgd_t *user_pgd = xen_get_user_pgd(pgd);
1480
1481	if (user_pgd)
1482		free_page((unsigned long)user_pgd);
1483}
1484
1485/*
1486 * Init-time set_pte while constructing initial pagetables, which
1487 * doesn't allow RO page table pages to be remapped RW.
1488 *
1489 * If there is no MFN for this PFN then this page is initially
1490 * ballooned out so clear the PTE (as in decrease_reservation() in
1491 * drivers/xen/balloon.c).
1492 *
1493 * Many of these PTE updates are done on unpinned and writable pages
1494 * and doing a hypercall for these is unnecessary and expensive.  At
1495 * this point it is rarely possible to tell if a page is pinned, so
1496 * mostly write the PTE directly and rely on Xen trapping and
1497 * emulating any updates as necessary.
1498 */
1499static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1500{
1501	if (unlikely(is_early_ioremap_ptep(ptep)))
1502		__xen_set_pte(ptep, pte);
1503	else
1504		native_set_pte(ptep, pte);
1505}
1506
1507__visible pte_t xen_make_pte_init(pteval_t pte)
1508{
1509	unsigned long pfn;
1510
1511	/*
1512	 * Pages belonging to the initial p2m list mapped outside the default
1513	 * address range must be mapped read-only. This region contains the
1514	 * page tables for mapping the p2m list, too, and page tables MUST be
1515	 * mapped read-only.
1516	 */
1517	pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1518	if (xen_start_info->mfn_list < __START_KERNEL_map &&
1519	    pfn >= xen_start_info->first_p2m_pfn &&
1520	    pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1521		pte &= ~_PAGE_RW;
1522
1523	pte = pte_pfn_to_mfn(pte);
1524	return native_make_pte(pte);
1525}
1526PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1527
1528/* Early in boot, while setting up the initial pagetable, assume
1529   everything is pinned. */
1530static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1531{
1532#ifdef CONFIG_FLATMEM
1533	BUG_ON(mem_map);	/* should only be used early */
1534#endif
1535	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1536	pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1537}
1538
1539/* Used for pmd and pud */
1540static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1541{
1542#ifdef CONFIG_FLATMEM
1543	BUG_ON(mem_map);	/* should only be used early */
1544#endif
1545	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1546}
1547
1548/* Early release_pte assumes that all pts are pinned, since there's
1549   only init_mm and anything attached to that is pinned. */
1550static void __init xen_release_pte_init(unsigned long pfn)
1551{
1552	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1553	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1554}
1555
1556static void __init xen_release_pmd_init(unsigned long pfn)
1557{
1558	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1559}
1560
1561static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1562{
1563	struct multicall_space mcs;
1564	struct mmuext_op *op;
1565
1566	mcs = __xen_mc_entry(sizeof(*op));
1567	op = mcs.args;
1568	op->cmd = cmd;
1569	op->arg1.mfn = pfn_to_mfn(pfn);
1570
1571	MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1572}
1573
1574static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1575{
1576	struct multicall_space mcs;
1577	unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1578
1579	mcs = __xen_mc_entry(0);
1580	MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1581				pfn_pte(pfn, prot), 0);
1582}
1583
1584/* This needs to make sure the new pte page is pinned iff its being
1585   attached to a pinned pagetable. */
1586static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1587				    unsigned level)
1588{
1589	bool pinned = xen_page_pinned(mm->pgd);
1590
1591	trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1592
1593	if (pinned) {
1594		struct page *page = pfn_to_page(pfn);
1595
1596		pinned = false;
1597		if (static_branch_likely(&xen_struct_pages_ready)) {
1598			pinned = PagePinned(page);
1599			SetPagePinned(page);
1600		}
1601
1602		xen_mc_batch();
1603
1604		__set_pfn_prot(pfn, PAGE_KERNEL_RO);
1605
1606		if (level == PT_PTE && IS_ENABLED(CONFIG_SPLIT_PTE_PTLOCKS) &&
1607		    !pinned)
1608			__pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1609
1610		xen_mc_issue(XEN_LAZY_MMU);
1611	}
1612}
1613
1614static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1615{
1616	xen_alloc_ptpage(mm, pfn, PT_PTE);
1617}
1618
1619static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1620{
1621	xen_alloc_ptpage(mm, pfn, PT_PMD);
1622}
1623
1624/* This should never happen until we're OK to use struct page */
1625static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1626{
1627	struct page *page = pfn_to_page(pfn);
1628	bool pinned = PagePinned(page);
1629
1630	trace_xen_mmu_release_ptpage(pfn, level, pinned);
1631
1632	if (pinned) {
1633		xen_mc_batch();
1634
1635		if (level == PT_PTE && IS_ENABLED(CONFIG_SPLIT_PTE_PTLOCKS))
1636			__pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1637
1638		__set_pfn_prot(pfn, PAGE_KERNEL);
1639
1640		xen_mc_issue(XEN_LAZY_MMU);
1641
1642		ClearPagePinned(page);
1643	}
1644}
1645
1646static void xen_release_pte(unsigned long pfn)
1647{
1648	xen_release_ptpage(pfn, PT_PTE);
1649}
1650
1651static void xen_release_pmd(unsigned long pfn)
1652{
1653	xen_release_ptpage(pfn, PT_PMD);
1654}
1655
1656static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1657{
1658	xen_alloc_ptpage(mm, pfn, PT_PUD);
1659}
1660
1661static void xen_release_pud(unsigned long pfn)
1662{
1663	xen_release_ptpage(pfn, PT_PUD);
1664}
1665
1666/*
1667 * Like __va(), but returns address in the kernel mapping (which is
1668 * all we have until the physical memory mapping has been set up.
1669 */
1670static void * __init __ka(phys_addr_t paddr)
1671{
1672	return (void *)(paddr + __START_KERNEL_map);
1673}
1674
1675/* Convert a machine address to physical address */
1676static unsigned long __init m2p(phys_addr_t maddr)
1677{
1678	phys_addr_t paddr;
1679
1680	maddr &= XEN_PTE_MFN_MASK;
1681	paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1682
1683	return paddr;
1684}
1685
1686/* Convert a machine address to kernel virtual */
1687static void * __init m2v(phys_addr_t maddr)
1688{
1689	return __ka(m2p(maddr));
1690}
1691
1692/* Set the page permissions on an identity-mapped pages */
1693static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1694				       unsigned long flags)
1695{
1696	unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1697	pte_t pte = pfn_pte(pfn, prot);
1698
1699	if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1700		BUG();
1701}
1702static void __init set_page_prot(void *addr, pgprot_t prot)
1703{
1704	return set_page_prot_flags(addr, prot, UVMF_NONE);
1705}
1706
1707void __init xen_setup_machphys_mapping(void)
1708{
1709	struct xen_machphys_mapping mapping;
1710
1711	if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1712		machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1713		machine_to_phys_nr = mapping.max_mfn + 1;
1714	} else {
1715		machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1716	}
1717}
1718
1719static void __init convert_pfn_mfn(void *v)
1720{
1721	pte_t *pte = v;
1722	int i;
1723
1724	/* All levels are converted the same way, so just treat them
1725	   as ptes. */
1726	for (i = 0; i < PTRS_PER_PTE; i++)
1727		pte[i] = xen_make_pte(pte[i].pte);
1728}
1729static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1730				 unsigned long addr)
1731{
1732	if (*pt_base == PFN_DOWN(__pa(addr))) {
1733		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1734		clear_page((void *)addr);
1735		(*pt_base)++;
1736	}
1737	if (*pt_end == PFN_DOWN(__pa(addr))) {
1738		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1739		clear_page((void *)addr);
1740		(*pt_end)--;
1741	}
1742}
1743/*
1744 * Set up the initial kernel pagetable.
1745 *
1746 * We can construct this by grafting the Xen provided pagetable into
1747 * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1748 * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1749 * kernel has a physical mapping to start with - but that's enough to
1750 * get __va working.  We need to fill in the rest of the physical
1751 * mapping once some sort of allocator has been set up.
1752 */
1753void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1754{
1755	pud_t *l3;
1756	pmd_t *l2;
1757	unsigned long addr[3];
1758	unsigned long pt_base, pt_end;
1759	unsigned i;
1760
1761	/* max_pfn_mapped is the last pfn mapped in the initial memory
1762	 * mappings. Considering that on Xen after the kernel mappings we
1763	 * have the mappings of some pages that don't exist in pfn space, we
1764	 * set max_pfn_mapped to the last real pfn mapped. */
1765	if (xen_start_info->mfn_list < __START_KERNEL_map)
1766		max_pfn_mapped = xen_start_info->first_p2m_pfn;
1767	else
1768		max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1769
1770	pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1771	pt_end = pt_base + xen_start_info->nr_pt_frames;
1772
1773	/* Zap identity mapping */
1774	init_top_pgt[0] = __pgd(0);
1775
1776	/* Pre-constructed entries are in pfn, so convert to mfn */
1777	/* L4[273] -> level3_ident_pgt  */
1778	/* L4[511] -> level3_kernel_pgt */
1779	convert_pfn_mfn(init_top_pgt);
1780
1781	/* L3_i[0] -> level2_ident_pgt */
1782	convert_pfn_mfn(level3_ident_pgt);
1783	/* L3_k[510] -> level2_kernel_pgt */
1784	/* L3_k[511] -> level2_fixmap_pgt */
1785	convert_pfn_mfn(level3_kernel_pgt);
1786
1787	/* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1788	convert_pfn_mfn(level2_fixmap_pgt);
1789
1790	/* We get [511][511] and have Xen's version of level2_kernel_pgt */
1791	l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1792	l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1793
1794	addr[0] = (unsigned long)pgd;
1795	addr[1] = (unsigned long)l3;
1796	addr[2] = (unsigned long)l2;
1797	/* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1798	 * Both L4[273][0] and L4[511][510] have entries that point to the same
1799	 * L2 (PMD) tables. Meaning that if you modify it in __va space
1800	 * it will be also modified in the __ka space! (But if you just
1801	 * modify the PMD table to point to other PTE's or none, then you
1802	 * are OK - which is what cleanup_highmap does) */
1803	copy_page(level2_ident_pgt, l2);
1804	/* Graft it onto L4[511][510] */
1805	copy_page(level2_kernel_pgt, l2);
1806
1807	/*
1808	 * Zap execute permission from the ident map. Due to the sharing of
1809	 * L1 entries we need to do this in the L2.
1810	 */
1811	if (__supported_pte_mask & _PAGE_NX) {
1812		for (i = 0; i < PTRS_PER_PMD; ++i) {
1813			if (pmd_none(level2_ident_pgt[i]))
1814				continue;
1815			level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1816		}
1817	}
1818
1819	/* Copy the initial P->M table mappings if necessary. */
1820	i = pgd_index(xen_start_info->mfn_list);
1821	if (i && i < pgd_index(__START_KERNEL_map))
1822		init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1823
1824	/* Make pagetable pieces RO */
1825	set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1826	set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1827	set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1828	set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1829	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1830	set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1831
1832	for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1833		set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1834			      PAGE_KERNEL_RO);
1835	}
1836
1837	/* Pin down new L4 */
1838	pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1839			  PFN_DOWN(__pa_symbol(init_top_pgt)));
1840
1841	/* Unpin Xen-provided one */
1842	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1843
1844#ifdef CONFIG_X86_VSYSCALL_EMULATION
1845	/* Pin user vsyscall L3 */
1846	set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1847	pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1848			  PFN_DOWN(__pa_symbol(level3_user_vsyscall)));
1849#endif
1850
1851	/*
1852	 * At this stage there can be no user pgd, and no page structure to
1853	 * attach it to, so make sure we just set kernel pgd.
1854	 */
1855	xen_mc_batch();
1856	__xen_write_cr3(true, __pa(init_top_pgt));
1857	xen_mc_issue(XEN_LAZY_CPU);
1858
1859	/* We can't that easily rip out L3 and L2, as the Xen pagetables are
1860	 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1861	 * the initial domain. For guests using the toolstack, they are in:
1862	 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1863	 * rip out the [L4] (pgd), but for guests we shave off three pages.
1864	 */
1865	for (i = 0; i < ARRAY_SIZE(addr); i++)
1866		check_pt_base(&pt_base, &pt_end, addr[i]);
1867
1868	/* Our (by three pages) smaller Xen pagetable that we are using */
1869	xen_pt_base = PFN_PHYS(pt_base);
1870	xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1871	memblock_reserve(xen_pt_base, xen_pt_size);
1872
1873	/* Revector the xen_start_info */
1874	xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1875}
1876
1877/*
1878 * Read a value from a physical address.
1879 */
1880static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1881{
1882	unsigned long *vaddr;
1883	unsigned long val;
1884
1885	vaddr = early_memremap_ro(addr, sizeof(val));
1886	val = *vaddr;
1887	early_memunmap(vaddr, sizeof(val));
1888	return val;
1889}
1890
1891/*
1892 * Translate a virtual address to a physical one without relying on mapped
1893 * page tables. Don't rely on big pages being aligned in (guest) physical
1894 * space!
1895 */
1896static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1897{
1898	phys_addr_t pa;
1899	pgd_t pgd;
1900	pud_t pud;
1901	pmd_t pmd;
1902	pte_t pte;
1903
1904	pa = read_cr3_pa();
1905	pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1906						       sizeof(pgd)));
1907	if (!pgd_present(pgd))
1908		return 0;
1909
1910	pa = pgd_val(pgd) & PTE_PFN_MASK;
1911	pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1912						       sizeof(pud)));
1913	if (!pud_present(pud))
1914		return 0;
1915	pa = pud_val(pud) & PTE_PFN_MASK;
1916	if (pud_leaf(pud))
1917		return pa + (vaddr & ~PUD_MASK);
1918
1919	pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1920						       sizeof(pmd)));
1921	if (!pmd_present(pmd))
1922		return 0;
1923	pa = pmd_val(pmd) & PTE_PFN_MASK;
1924	if (pmd_leaf(pmd))
1925		return pa + (vaddr & ~PMD_MASK);
1926
1927	pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1928						       sizeof(pte)));
1929	if (!pte_present(pte))
1930		return 0;
1931	pa = pte_pfn(pte) << PAGE_SHIFT;
1932
1933	return pa | (vaddr & ~PAGE_MASK);
1934}
1935
1936/*
1937 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1938 * this area.
1939 */
1940void __init xen_relocate_p2m(void)
1941{
1942	phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1943	unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1944	int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1945	pte_t *pt;
1946	pmd_t *pmd;
1947	pud_t *pud;
1948	pgd_t *pgd;
1949	unsigned long *new_p2m;
1950
1951	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1952	n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1953	n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1954	n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1955	n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1956	n_frames = n_pte + n_pt + n_pmd + n_pud;
1957
1958	new_area = xen_find_free_area(PFN_PHYS(n_frames));
1959	if (!new_area) {
1960		xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1961		BUG();
1962	}
1963
1964	/*
1965	 * Setup the page tables for addressing the new p2m list.
1966	 * We have asked the hypervisor to map the p2m list at the user address
1967	 * PUD_SIZE. It may have done so, or it may have used a kernel space
1968	 * address depending on the Xen version.
1969	 * To avoid any possible virtual address collision, just use
1970	 * 2 * PUD_SIZE for the new area.
1971	 */
1972	pud_phys = new_area;
1973	pmd_phys = pud_phys + PFN_PHYS(n_pud);
1974	pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1975	p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1976
1977	pgd = __va(read_cr3_pa());
1978	new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1979	for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1980		pud = early_memremap(pud_phys, PAGE_SIZE);
1981		clear_page(pud);
1982		for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1983				idx_pmd++) {
1984			pmd = early_memremap(pmd_phys, PAGE_SIZE);
1985			clear_page(pmd);
1986			for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1987					idx_pt++) {
1988				pt = early_memremap(pt_phys, PAGE_SIZE);
1989				clear_page(pt);
1990				for (idx_pte = 0;
1991				     idx_pte < min(n_pte, PTRS_PER_PTE);
1992				     idx_pte++) {
1993					pt[idx_pte] = pfn_pte(p2m_pfn,
1994							      PAGE_KERNEL);
1995					p2m_pfn++;
1996				}
1997				n_pte -= PTRS_PER_PTE;
1998				early_memunmap(pt, PAGE_SIZE);
1999				make_lowmem_page_readonly(__va(pt_phys));
2000				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2001						PFN_DOWN(pt_phys));
2002				pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
2003				pt_phys += PAGE_SIZE;
2004			}
2005			n_pt -= PTRS_PER_PMD;
2006			early_memunmap(pmd, PAGE_SIZE);
2007			make_lowmem_page_readonly(__va(pmd_phys));
2008			pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2009					PFN_DOWN(pmd_phys));
2010			pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
2011			pmd_phys += PAGE_SIZE;
2012		}
2013		n_pmd -= PTRS_PER_PUD;
2014		early_memunmap(pud, PAGE_SIZE);
2015		make_lowmem_page_readonly(__va(pud_phys));
2016		pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2017		set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2018		pud_phys += PAGE_SIZE;
2019	}
2020
2021	/* Now copy the old p2m info to the new area. */
2022	memcpy(new_p2m, xen_p2m_addr, size);
2023	xen_p2m_addr = new_p2m;
2024
2025	/* Release the old p2m list and set new list info. */
2026	p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2027	BUG_ON(!p2m_pfn);
2028	p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2029
2030	if (xen_start_info->mfn_list < __START_KERNEL_map) {
2031		pfn = xen_start_info->first_p2m_pfn;
2032		pfn_end = xen_start_info->first_p2m_pfn +
2033			  xen_start_info->nr_p2m_frames;
2034		set_pgd(pgd + 1, __pgd(0));
2035	} else {
2036		pfn = p2m_pfn;
2037		pfn_end = p2m_pfn_end;
2038	}
2039
2040	memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2041	while (pfn < pfn_end) {
2042		if (pfn == p2m_pfn) {
2043			pfn = p2m_pfn_end;
2044			continue;
2045		}
2046		make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2047		pfn++;
2048	}
2049
2050	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2051	xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
2052	xen_start_info->nr_p2m_frames = n_frames;
2053}
2054
2055void __init xen_reserve_special_pages(void)
2056{
2057	phys_addr_t paddr;
2058
2059	memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2060	if (xen_start_info->store_mfn) {
2061		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2062		memblock_reserve(paddr, PAGE_SIZE);
2063	}
2064	if (!xen_initial_domain()) {
2065		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2066		memblock_reserve(paddr, PAGE_SIZE);
2067	}
2068}
2069
2070void __init xen_pt_check_e820(void)
2071{
2072	xen_chk_is_e820_usable(xen_pt_base, xen_pt_size, "page table");
2073}
2074
2075static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2076
2077static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2078{
2079	pte_t pte;
2080	unsigned long vaddr;
2081
2082	phys >>= PAGE_SHIFT;
2083
2084	switch (idx) {
2085	case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2086#ifdef CONFIG_X86_VSYSCALL_EMULATION
2087	case VSYSCALL_PAGE:
2088#endif
2089		/* All local page mappings */
2090		pte = pfn_pte(phys, prot);
2091		break;
2092
2093#ifdef CONFIG_X86_LOCAL_APIC
2094	case FIX_APIC_BASE:	/* maps dummy local APIC */
2095		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2096		break;
2097#endif
2098
2099#ifdef CONFIG_X86_IO_APIC
2100	case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2101		/*
2102		 * We just don't map the IO APIC - all access is via
2103		 * hypercalls.  Keep the address in the pte for reference.
2104		 */
2105		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2106		break;
2107#endif
2108
2109	case FIX_PARAVIRT_BOOTMAP:
2110		/* This is an MFN, but it isn't an IO mapping from the
2111		   IO domain */
2112		pte = mfn_pte(phys, prot);
2113		break;
2114
2115	default:
2116		/* By default, set_fixmap is used for hardware mappings */
2117		pte = mfn_pte(phys, prot);
2118		break;
2119	}
2120
2121	vaddr = __fix_to_virt(idx);
2122	if (HYPERVISOR_update_va_mapping(vaddr, pte, UVMF_INVLPG))
2123		BUG();
2124
2125#ifdef CONFIG_X86_VSYSCALL_EMULATION
2126	/* Replicate changes to map the vsyscall page into the user
2127	   pagetable vsyscall mapping. */
2128	if (idx == VSYSCALL_PAGE)
2129		set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2130#endif
2131}
2132
2133static void xen_enter_lazy_mmu(void)
2134{
2135	enter_lazy(XEN_LAZY_MMU);
2136}
2137
2138static void xen_flush_lazy_mmu(void)
2139{
2140	preempt_disable();
2141
2142	if (xen_get_lazy_mode() == XEN_LAZY_MMU) {
2143		arch_leave_lazy_mmu_mode();
2144		arch_enter_lazy_mmu_mode();
2145	}
2146
2147	preempt_enable();
2148}
2149
2150static void __init xen_post_allocator_init(void)
2151{
2152	pv_ops.mmu.set_pte = xen_set_pte;
2153	pv_ops.mmu.set_pmd = xen_set_pmd;
2154	pv_ops.mmu.set_pud = xen_set_pud;
2155	pv_ops.mmu.set_p4d = xen_set_p4d;
2156
2157	/* This will work as long as patching hasn't happened yet
2158	   (which it hasn't) */
2159	pv_ops.mmu.alloc_pte = xen_alloc_pte;
2160	pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2161	pv_ops.mmu.release_pte = xen_release_pte;
2162	pv_ops.mmu.release_pmd = xen_release_pmd;
2163	pv_ops.mmu.alloc_pud = xen_alloc_pud;
2164	pv_ops.mmu.release_pud = xen_release_pud;
2165	pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2166
2167	pv_ops.mmu.write_cr3 = &xen_write_cr3;
2168}
2169
2170static void xen_leave_lazy_mmu(void)
2171{
2172	preempt_disable();
2173	xen_mc_flush();
2174	leave_lazy(XEN_LAZY_MMU);
2175	preempt_enable();
2176}
2177
2178static const typeof(pv_ops) xen_mmu_ops __initconst = {
2179	.mmu = {
2180		.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2181		.write_cr2 = xen_write_cr2,
2182
2183		.read_cr3 = xen_read_cr3,
2184		.write_cr3 = xen_write_cr3_init,
2185
2186		.flush_tlb_user = xen_flush_tlb,
2187		.flush_tlb_kernel = xen_flush_tlb,
2188		.flush_tlb_one_user = xen_flush_tlb_one_user,
2189		.flush_tlb_multi = xen_flush_tlb_multi,
2190
2191		.pgd_alloc = xen_pgd_alloc,
2192		.pgd_free = xen_pgd_free,
2193
2194		.alloc_pte = xen_alloc_pte_init,
2195		.release_pte = xen_release_pte_init,
2196		.alloc_pmd = xen_alloc_pmd_init,
2197		.release_pmd = xen_release_pmd_init,
2198
2199		.set_pte = xen_set_pte_init,
2200		.set_pmd = xen_set_pmd_hyper,
2201
2202		.ptep_modify_prot_start = xen_ptep_modify_prot_start,
2203		.ptep_modify_prot_commit = xen_ptep_modify_prot_commit,
2204
2205		.pte_val = PV_CALLEE_SAVE(xen_pte_val),
2206		.pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2207
2208		.make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2209		.make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2210
2211		.set_pud = xen_set_pud_hyper,
2212
2213		.make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2214		.pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2215
2216		.pud_val = PV_CALLEE_SAVE(xen_pud_val),
2217		.make_pud = PV_CALLEE_SAVE(xen_make_pud),
2218		.set_p4d = xen_set_p4d_hyper,
2219
2220		.alloc_pud = xen_alloc_pmd_init,
2221		.release_pud = xen_release_pmd_init,
2222
2223		.p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2224		.make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2225
2226		.enter_mmap = xen_enter_mmap,
2227		.exit_mmap = xen_exit_mmap,
2228
2229		.lazy_mode = {
2230			.enter = xen_enter_lazy_mmu,
2231			.leave = xen_leave_lazy_mmu,
2232			.flush = xen_flush_lazy_mmu,
2233		},
2234
2235		.set_fixmap = xen_set_fixmap,
2236	},
2237};
2238
2239void __init xen_init_mmu_ops(void)
2240{
2241	x86_init.paging.pagetable_init = xen_pagetable_init;
2242	x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2243
2244	pv_ops.mmu = xen_mmu_ops.mmu;
2245
2246	memset(dummy_mapping, 0xff, PAGE_SIZE);
2247}
2248
2249#define VOID_PTE (mfn_pte(0, __pgprot(0)))
2250static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2251				unsigned long *in_frames,
2252				unsigned long *out_frames)
2253{
2254	int i;
2255	struct multicall_space mcs;
2256
2257	xen_mc_batch();
2258	for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2259		mcs = __xen_mc_entry(0);
2260
2261		if (in_frames)
2262			in_frames[i] = virt_to_mfn((void *)vaddr);
2263
2264		MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2265		__set_phys_to_machine(virt_to_pfn((void *)vaddr), INVALID_P2M_ENTRY);
2266
2267		if (out_frames)
2268			out_frames[i] = virt_to_pfn((void *)vaddr);
2269	}
2270	xen_mc_issue(0);
2271}
2272
2273/*
2274 * Update the pfn-to-mfn mappings for a virtual address range, either to
2275 * point to an array of mfns, or contiguously from a single starting
2276 * mfn.
2277 */
2278static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2279				     unsigned long *mfns,
2280				     unsigned long first_mfn)
2281{
2282	unsigned i, limit;
2283	unsigned long mfn;
2284
2285	xen_mc_batch();
2286
2287	limit = 1u << order;
2288	for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2289		struct multicall_space mcs;
2290		unsigned flags;
2291
2292		mcs = __xen_mc_entry(0);
2293		if (mfns)
2294			mfn = mfns[i];
2295		else
2296			mfn = first_mfn + i;
2297
2298		if (i < (limit - 1))
2299			flags = 0;
2300		else {
2301			if (order == 0)
2302				flags = UVMF_INVLPG | UVMF_ALL;
2303			else
2304				flags = UVMF_TLB_FLUSH | UVMF_ALL;
2305		}
2306
2307		MULTI_update_va_mapping(mcs.mc, vaddr,
2308				mfn_pte(mfn, PAGE_KERNEL), flags);
2309
2310		set_phys_to_machine(virt_to_pfn((void *)vaddr), mfn);
2311	}
2312
2313	xen_mc_issue(0);
2314}
2315
2316/*
2317 * Perform the hypercall to exchange a region of our pfns to point to
2318 * memory with the required contiguous alignment.  Takes the pfns as
2319 * input, and populates mfns as output.
2320 *
2321 * Returns a success code indicating whether the hypervisor was able to
2322 * satisfy the request or not.
2323 */
2324static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2325			       unsigned long *pfns_in,
2326			       unsigned long extents_out,
2327			       unsigned int order_out,
2328			       unsigned long *mfns_out,
2329			       unsigned int address_bits)
2330{
2331	long rc;
2332	int success;
2333
2334	struct xen_memory_exchange exchange = {
2335		.in = {
2336			.nr_extents   = extents_in,
2337			.extent_order = order_in,
2338			.extent_start = pfns_in,
2339			.domid        = DOMID_SELF
2340		},
2341		.out = {
2342			.nr_extents   = extents_out,
2343			.extent_order = order_out,
2344			.extent_start = mfns_out,
2345			.address_bits = address_bits,
2346			.domid        = DOMID_SELF
2347		}
2348	};
2349
2350	BUG_ON(extents_in << order_in != extents_out << order_out);
2351
2352	rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2353	success = (exchange.nr_exchanged == extents_in);
2354
2355	BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2356	BUG_ON(success && (rc != 0));
2357
2358	return success;
2359}
2360
2361int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2362				 unsigned int address_bits,
2363				 dma_addr_t *dma_handle)
2364{
2365	unsigned long *in_frames, out_frame;
2366	unsigned long  flags;
2367	int            success;
2368	unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2369
2370	if (unlikely(order > discontig_frames_order)) {
2371		if (!discontig_frames_dyn)
2372			return -ENOMEM;
2373
2374		if (alloc_discontig_frames(order))
2375			return -ENOMEM;
2376	}
2377
2378	memset((void *) vstart, 0, PAGE_SIZE << order);
2379
2380	spin_lock_irqsave(&xen_reservation_lock, flags);
2381
2382	in_frames = discontig_frames;
2383
2384	/* 1. Zap current PTEs, remembering MFNs. */
2385	xen_zap_pfn_range(vstart, order, in_frames, NULL);
2386
2387	/* 2. Get a new contiguous memory extent. */
2388	out_frame = virt_to_pfn((void *)vstart);
2389	success = xen_exchange_memory(1UL << order, 0, in_frames,
2390				      1, order, &out_frame,
2391				      address_bits);
2392
2393	/* 3. Map the new extent in place of old pages. */
2394	if (success)
2395		xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2396	else
2397		xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2398
2399	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2400
2401	*dma_handle = virt_to_machine(vstart).maddr;
2402	return success ? 0 : -ENOMEM;
2403}
2404
2405void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2406{
2407	unsigned long *out_frames, in_frame;
2408	unsigned long  flags;
2409	int success;
2410	unsigned long vstart;
2411
2412	if (unlikely(order > discontig_frames_order))
2413		return;
2414
2415	vstart = (unsigned long)phys_to_virt(pstart);
2416	memset((void *) vstart, 0, PAGE_SIZE << order);
2417
2418	spin_lock_irqsave(&xen_reservation_lock, flags);
2419
2420	out_frames = discontig_frames;
2421
2422	/* 1. Find start MFN of contiguous extent. */
2423	in_frame = virt_to_mfn((void *)vstart);
2424
2425	/* 2. Zap current PTEs. */
2426	xen_zap_pfn_range(vstart, order, NULL, out_frames);
2427
2428	/* 3. Do the exchange for non-contiguous MFNs. */
2429	success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2430					0, out_frames, 0);
2431
2432	/* 4. Map new pages in place of old pages. */
2433	if (success)
2434		xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2435	else
2436		xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2437
2438	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2439}
2440
2441static noinline void xen_flush_tlb_all(void)
2442{
2443	struct mmuext_op *op;
2444	struct multicall_space mcs;
2445
2446	preempt_disable();
2447
2448	mcs = xen_mc_entry(sizeof(*op));
2449
2450	op = mcs.args;
2451	op->cmd = MMUEXT_TLB_FLUSH_ALL;
2452	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2453
2454	xen_mc_issue(XEN_LAZY_MMU);
2455
2456	preempt_enable();
2457}
2458
2459#define REMAP_BATCH_SIZE 16
2460
2461struct remap_data {
2462	xen_pfn_t *pfn;
2463	bool contiguous;
2464	bool no_translate;
2465	pgprot_t prot;
2466	struct mmu_update *mmu_update;
2467};
2468
2469static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2470{
2471	struct remap_data *rmd = data;
2472	pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2473
2474	/*
2475	 * If we have a contiguous range, just update the pfn itself,
2476	 * else update pointer to be "next pfn".
2477	 */
2478	if (rmd->contiguous)
2479		(*rmd->pfn)++;
2480	else
2481		rmd->pfn++;
2482
2483	rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2484	rmd->mmu_update->ptr |= rmd->no_translate ?
2485		MMU_PT_UPDATE_NO_TRANSLATE :
2486		MMU_NORMAL_PT_UPDATE;
2487	rmd->mmu_update->val = pte_val_ma(pte);
2488	rmd->mmu_update++;
2489
2490	return 0;
2491}
2492
2493int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2494		  xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2495		  unsigned int domid, bool no_translate)
2496{
2497	int err = 0;
2498	struct remap_data rmd;
2499	struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2500	unsigned long range;
2501	int mapped = 0;
2502
2503	BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2504
2505	rmd.pfn = pfn;
2506	rmd.prot = prot;
2507	/*
2508	 * We use the err_ptr to indicate if there we are doing a contiguous
2509	 * mapping or a discontiguous mapping.
2510	 */
2511	rmd.contiguous = !err_ptr;
2512	rmd.no_translate = no_translate;
2513
2514	while (nr) {
2515		int index = 0;
2516		int done = 0;
2517		int batch = min(REMAP_BATCH_SIZE, nr);
2518		int batch_left = batch;
2519
2520		range = (unsigned long)batch << PAGE_SHIFT;
2521
2522		rmd.mmu_update = mmu_update;
2523		err = apply_to_page_range(vma->vm_mm, addr, range,
2524					  remap_area_pfn_pte_fn, &rmd);
2525		if (err)
2526			goto out;
2527
2528		/*
2529		 * We record the error for each page that gives an error, but
2530		 * continue mapping until the whole set is done
2531		 */
2532		do {
2533			int i;
2534
2535			err = HYPERVISOR_mmu_update(&mmu_update[index],
2536						    batch_left, &done, domid);
2537
2538			/*
2539			 * @err_ptr may be the same buffer as @gfn, so
2540			 * only clear it after each chunk of @gfn is
2541			 * used.
2542			 */
2543			if (err_ptr) {
2544				for (i = index; i < index + done; i++)
2545					err_ptr[i] = 0;
2546			}
2547			if (err < 0) {
2548				if (!err_ptr)
2549					goto out;
2550				err_ptr[i] = err;
2551				done++; /* Skip failed frame. */
2552			} else
2553				mapped += done;
2554			batch_left -= done;
2555			index += done;
2556		} while (batch_left);
2557
2558		nr -= batch;
2559		addr += range;
2560		if (err_ptr)
2561			err_ptr += batch;
2562		cond_resched();
2563	}
2564out:
2565
2566	xen_flush_tlb_all();
2567
2568	return err < 0 ? err : mapped;
2569}
2570EXPORT_SYMBOL_GPL(xen_remap_pfn);
2571
2572#ifdef CONFIG_VMCORE_INFO
2573phys_addr_t paddr_vmcoreinfo_note(void)
2574{
2575	if (xen_pv_domain())
2576		return virt_to_machine(vmcoreinfo_note).maddr;
2577	else
2578		return __pa(vmcoreinfo_note);
2579}
2580#endif /* CONFIG_KEXEC_CORE */