x86/xen: split off mmu_pv.c
authorVitaly Kuznetsov <vkuznets@redhat.com>
Tue, 4 Apr 2017 12:48:17 +0000 (14:48 +0200)
committerJuergen Gross <jgross@suse.com>
Tue, 2 May 2017 09:08:51 +0000 (11:08 +0200)
Basically, mmu.c is renamed to mmu_pv.c and some code moved out to common
mmu.c.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
arch/x86/xen/Makefile
arch/x86/xen/mmu.c
arch/x86/xen/mmu_pv.c [new file with mode: 0644]

index 6a95a8b..8da1ca9 100644 (file)
@@ -8,12 +8,12 @@ endif
 # Make sure early boot has no stackprotector
 nostackp := $(call cc-option, -fno-stack-protector)
 CFLAGS_enlighten_pv.o          := $(nostackp)
-CFLAGS_mmu.o                   := $(nostackp)
+CFLAGS_mmu_pv.o                := $(nostackp)
 
 obj-y          := enlighten.o setup.o multicalls.o mmu.o irq.o \
                        time.o xen-asm.o xen-asm_$(BITS).o \
                        grant-table.o suspend.o platform-pci-unplug.o \
-                       p2m.o apic.o pmu.o enlighten_pv.o
+                       p2m.o apic.o pmu.o enlighten_pv.o mmu_pv.o
 
 obj-$(CONFIG_XEN_PVHVM)                += enlighten_hvm.o mmu_hvm.o
 obj-$(CONFIG_XEN_PVH)                  += enlighten_pvh.o
index a376f5e..5e375a5 100644 (file)
-/*
- * Xen mmu operations
- *
- * This file contains the various mmu fetch and update operations.
- * The most important job they must perform is the mapping between the
- * domain's pfn and the overall machine mfns.
- *
- * Xen allows guests to directly update the pagetable, in a controlled
- * fashion.  In other words, the guest modifies the same pagetable
- * that the CPU actually uses, which eliminates the overhead of having
- * a separate shadow pagetable.
- *
- * In order to allow this, it falls on the guest domain to map its
- * notion of a "physical" pfn - which is just a domain-local linear
- * address - into a real "machine address" which the CPU's MMU can
- * use.
- *
- * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
- * inserted directly into the pagetable.  When creating a new
- * pte/pmd/pgd, it converts the passed pfn into an mfn.  Conversely,
- * when reading the content back with __(pgd|pmd|pte)_val, it converts
- * the mfn back into a pfn.
- *
- * The other constraint is that all pages which make up a pagetable
- * must be mapped read-only in the guest.  This prevents uncontrolled
- * guest updates to the pagetable.  Xen strictly enforces this, and
- * will disallow any pagetable update which will end up mapping a
- * pagetable page RW, and will disallow using any writable page as a
- * pagetable.
- *
- * Naively, when loading %cr3 with the base of a new pagetable, Xen
- * would need to validate the whole pagetable before going on.
- * Naturally, this is quite slow.  The solution is to "pin" a
- * pagetable, which enforces all the constraints on the pagetable even
- * when it is not actively in use.  This menas that Xen can be assured
- * that it is still valid when you do load it into %cr3, and doesn't
- * need to revalidate it.
- *
- * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
- */
-#include <linux/sched/mm.h>
-#include <linux/highmem.h>
-#include <linux/debugfs.h>
-#include <linux/bug.h>
-#include <linux/vmalloc.h>
-#include <linux/export.h>
-#include <linux/init.h>
-#include <linux/gfp.h>
-#include <linux/memblock.h>
-#include <linux/seq_file.h>
-#include <linux/crash_dump.h>
-
-#include <trace/events/xen.h>
-
-#include <asm/pgtable.h>
-#include <asm/tlbflush.h>
-#include <asm/fixmap.h>
-#include <asm/mmu_context.h>
-#include <asm/setup.h>
-#include <asm/paravirt.h>
-#include <asm/e820/api.h>
-#include <asm/linkage.h>
-#include <asm/page.h>
-#include <asm/init.h>
-#include <asm/pat.h>
-#include <asm/smp.h>
-
-#include <asm/xen/hypercall.h>
-#include <asm/xen/hypervisor.h>
-
-#include <xen/xen.h>
-#include <xen/page.h>
-#include <xen/interface/xen.h>
-#include <xen/interface/hvm/hvm_op.h>
-#include <xen/interface/version.h>
-#include <xen/interface/memory.h>
-#include <xen/hvc-console.h>
-
-#include "multicalls.h"
-#include "mmu.h"
-#include "debugfs.h"
-
-/*
- * Protects atomic reservation decrease/increase against concurrent increases.
- * Also protects non-atomic updates of current_pages and balloon lists.
- */
-DEFINE_SPINLOCK(xen_reservation_lock);
-
-#ifdef CONFIG_X86_32
-/*
- * Identity map, in addition to plain kernel map.  This needs to be
- * large enough to allocate page table pages to allocate the rest.
- * Each page can map 2MB.
- */
-#define LEVEL1_IDENT_ENTRIES   (PTRS_PER_PTE * 4)
-static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
-#endif
-#ifdef CONFIG_X86_64
-/* l3 pud for userspace vsyscall mapping */
-static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
-#endif /* CONFIG_X86_64 */
-
-/*
- * Note about cr3 (pagetable base) values:
- *
- * xen_cr3 contains the current logical cr3 value; it contains the
- * last set cr3.  This may not be the current effective cr3, because
- * its update may be being lazily deferred.  However, a vcpu looking
- * at its own cr3 can use this value knowing that it everything will
- * be self-consistent.
- *
- * xen_current_cr3 contains the actual vcpu cr3; it is set once the
- * hypercall to set the vcpu cr3 is complete (so it may be a little
- * out of date, but it will never be set early).  If one vcpu is
- * looking at another vcpu's cr3 value, it should use this variable.
- */
-DEFINE_PER_CPU(unsigned long, xen_cr3);         /* cr3 stored as physaddr */
-DEFINE_PER_CPU(unsigned long, xen_current_cr3);         /* actual vcpu cr3 */
-
-static phys_addr_t xen_pt_base, xen_pt_size __initdata;
-
-/*
- * Just beyond the highest usermode address.  STACK_TOP_MAX has a
- * redzone above it, so round it up to a PGD boundary.
- */
-#define USER_LIMIT     ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
-
-unsigned long arbitrary_virt_to_mfn(void *vaddr)
-{
-       xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
-
-       return PFN_DOWN(maddr.maddr);
-}
-
-xmaddr_t arbitrary_virt_to_machine(void *vaddr)
-{
-       unsigned long address = (unsigned long)vaddr;
-       unsigned int level;
-       pte_t *pte;
-       unsigned offset;
-
-       /*
-        * if the PFN is in the linear mapped vaddr range, we can just use
-        * the (quick) virt_to_machine() p2m lookup
-        */
-       if (virt_addr_valid(vaddr))
-               return virt_to_machine(vaddr);
-
-       /* otherwise we have to do a (slower) full page-table walk */
-
-       pte = lookup_address(address, &level);
-       BUG_ON(pte == NULL);
-       offset = address & ~PAGE_MASK;
-       return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
-}
-EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
-
-void make_lowmem_page_readonly(void *vaddr)
-{
-       pte_t *pte, ptev;
-       unsigned long address = (unsigned long)vaddr;
-       unsigned int level;
-
-       pte = lookup_address(address, &level);
-       if (pte == NULL)
-               return;         /* vaddr missing */
-
-       ptev = pte_wrprotect(*pte);
-
-       if (HYPERVISOR_update_va_mapping(address, ptev, 0))
-               BUG();
-}
-
-void make_lowmem_page_readwrite(void *vaddr)
-{
-       pte_t *pte, ptev;
-       unsigned long address = (unsigned long)vaddr;
-       unsigned int level;
-
-       pte = lookup_address(address, &level);
-       if (pte == NULL)
-               return;         /* vaddr missing */
-
-       ptev = pte_mkwrite(*pte);
-
-       if (HYPERVISOR_update_va_mapping(address, ptev, 0))
-               BUG();
-}
-
-
-static bool xen_page_pinned(void *ptr)
-{
-       struct page *page = virt_to_page(ptr);
-
-       return PagePinned(page);
-}
-
-void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
-{
-       struct multicall_space mcs;
-       struct mmu_update *u;
-
-       trace_xen_mmu_set_domain_pte(ptep, pteval, domid);
-
-       mcs = xen_mc_entry(sizeof(*u));
-       u = mcs.args;
-
-       /* ptep might be kmapped when using 32-bit HIGHPTE */
-       u->ptr = virt_to_machine(ptep).maddr;
-       u->val = pte_val_ma(pteval);
-
-       MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-}
-EXPORT_SYMBOL_GPL(xen_set_domain_pte);
-
-static void xen_extend_mmu_update(const struct mmu_update *update)
-{
-       struct multicall_space mcs;
-       struct mmu_update *u;
-
-       mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
-
-       if (mcs.mc != NULL) {
-               mcs.mc->args[1]++;
-       } else {
-               mcs = __xen_mc_entry(sizeof(*u));
-               MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
-       }
-
-       u = mcs.args;
-       *u = *update;
-}
-
-static void xen_extend_mmuext_op(const struct mmuext_op *op)
-{
-       struct multicall_space mcs;
-       struct mmuext_op *u;
-
-       mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
-
-       if (mcs.mc != NULL) {
-               mcs.mc->args[1]++;
-       } else {
-               mcs = __xen_mc_entry(sizeof(*u));
-               MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
-       }
-
-       u = mcs.args;
-       *u = *op;
-}
-
-static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
-{
-       struct mmu_update u;
-
-       preempt_disable();
-
-       xen_mc_batch();
-
-       /* ptr may be ioremapped for 64-bit pagetable setup */
-       u.ptr = arbitrary_virt_to_machine(ptr).maddr;
-       u.val = pmd_val_ma(val);
-       xen_extend_mmu_update(&u);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       preempt_enable();
-}
-
-static void xen_set_pmd(pmd_t *ptr, pmd_t val)
-{
-       trace_xen_mmu_set_pmd(ptr, val);
-
-       /* If page is not pinned, we can just update the entry
-          directly */
-       if (!xen_page_pinned(ptr)) {
-               *ptr = val;
-               return;
-       }
-
-       xen_set_pmd_hyper(ptr, val);
-}
-
-/*
- * Associate a virtual page frame with a given physical page frame
- * and protection flags for that frame.
- */
-void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
-{
-       set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
-}
-
-static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
-{
-       struct mmu_update u;
-
-       if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
-               return false;
-
-       xen_mc_batch();
-
-       u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
-       u.val = pte_val_ma(pteval);
-       xen_extend_mmu_update(&u);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       return true;
-}
-
-static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
-{
-       if (!xen_batched_set_pte(ptep, pteval)) {
-               /*
-                * Could call native_set_pte() here and trap and
-                * emulate the PTE write but with 32-bit guests this
-                * needs two traps (one for each of the two 32-bit
-                * words in the PTE) so do one hypercall directly
-                * instead.
-                */
-               struct mmu_update u;
-
-               u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
-               u.val = pte_val_ma(pteval);
-               HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
-       }
-}
-
-static void xen_set_pte(pte_t *ptep, pte_t pteval)
-{
-       trace_xen_mmu_set_pte(ptep, pteval);
-       __xen_set_pte(ptep, pteval);
-}
-
-static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
-                   pte_t *ptep, pte_t pteval)
-{
-       trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
-       __xen_set_pte(ptep, pteval);
-}
-
-pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
-                                unsigned long addr, pte_t *ptep)
-{
-       /* Just return the pte as-is.  We preserve the bits on commit */
-       trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
-       return *ptep;
-}
-
-void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
-                                pte_t *ptep, pte_t pte)
-{
-       struct mmu_update u;
-
-       trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
-       xen_mc_batch();
-
-       u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
-       u.val = pte_val_ma(pte);
-       xen_extend_mmu_update(&u);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-}
-
-/* Assume pteval_t is equivalent to all the other *val_t types. */
-static pteval_t pte_mfn_to_pfn(pteval_t val)
-{
-       if (val & _PAGE_PRESENT) {
-               unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
-               unsigned long pfn = mfn_to_pfn(mfn);
-
-               pteval_t flags = val & PTE_FLAGS_MASK;
-               if (unlikely(pfn == ~0))
-                       val = flags & ~_PAGE_PRESENT;
-               else
-                       val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
-       }
-
-       return val;
-}
-
-static pteval_t pte_pfn_to_mfn(pteval_t val)
-{
-       if (val & _PAGE_PRESENT) {
-               unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
-               pteval_t flags = val & PTE_FLAGS_MASK;
-               unsigned long mfn;
-
-               if (!xen_feature(XENFEAT_auto_translated_physmap))
-                       mfn = __pfn_to_mfn(pfn);
-               else
-                       mfn = pfn;
-               /*
-                * If there's no mfn for the pfn, then just create an
-                * empty non-present pte.  Unfortunately this loses
-                * information about the original pfn, so
-                * pte_mfn_to_pfn is asymmetric.
-                */
-               if (unlikely(mfn == INVALID_P2M_ENTRY)) {
-                       mfn = 0;
-                       flags = 0;
-               } else
-                       mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
-               val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
-       }
-
-       return val;
-}
-
-__visible pteval_t xen_pte_val(pte_t pte)
-{
-       pteval_t pteval = pte.pte;
-
-       return pte_mfn_to_pfn(pteval);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
-
-__visible pgdval_t xen_pgd_val(pgd_t pgd)
-{
-       return pte_mfn_to_pfn(pgd.pgd);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
-
-__visible pte_t xen_make_pte(pteval_t pte)
-{
-       pte = pte_pfn_to_mfn(pte);
-
-       return native_make_pte(pte);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
-
-__visible pgd_t xen_make_pgd(pgdval_t pgd)
-{
-       pgd = pte_pfn_to_mfn(pgd);
-       return native_make_pgd(pgd);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
-
-__visible pmdval_t xen_pmd_val(pmd_t pmd)
-{
-       return pte_mfn_to_pfn(pmd.pmd);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
-
-static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
-{
-       struct mmu_update u;
-
-       preempt_disable();
-
-       xen_mc_batch();
-
-       /* ptr may be ioremapped for 64-bit pagetable setup */
-       u.ptr = arbitrary_virt_to_machine(ptr).maddr;
-       u.val = pud_val_ma(val);
-       xen_extend_mmu_update(&u);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       preempt_enable();
-}
-
-static void xen_set_pud(pud_t *ptr, pud_t val)
-{
-       trace_xen_mmu_set_pud(ptr, val);
-
-       /* If page is not pinned, we can just update the entry
-          directly */
-       if (!xen_page_pinned(ptr)) {
-               *ptr = val;
-               return;
-       }
-
-       xen_set_pud_hyper(ptr, val);
-}
-
-#ifdef CONFIG_X86_PAE
-static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
-{
-       trace_xen_mmu_set_pte_atomic(ptep, pte);
-       set_64bit((u64 *)ptep, native_pte_val(pte));
-}
-
-static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
-{
-       trace_xen_mmu_pte_clear(mm, addr, ptep);
-       if (!xen_batched_set_pte(ptep, native_make_pte(0)))
-               native_pte_clear(mm, addr, ptep);
-}
-
-static void xen_pmd_clear(pmd_t *pmdp)
-{
-       trace_xen_mmu_pmd_clear(pmdp);
-       set_pmd(pmdp, __pmd(0));
-}
-#endif /* CONFIG_X86_PAE */
-
-__visible pmd_t xen_make_pmd(pmdval_t pmd)
-{
-       pmd = pte_pfn_to_mfn(pmd);
-       return native_make_pmd(pmd);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
-
-#if CONFIG_PGTABLE_LEVELS == 4
-__visible pudval_t xen_pud_val(pud_t pud)
-{
-       return pte_mfn_to_pfn(pud.pud);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
-
-__visible pud_t xen_make_pud(pudval_t pud)
-{
-       pud = pte_pfn_to_mfn(pud);
-
-       return native_make_pud(pud);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
-
-static pgd_t *xen_get_user_pgd(pgd_t *pgd)
-{
-       pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
-       unsigned offset = pgd - pgd_page;
-       pgd_t *user_ptr = NULL;
-
-       if (offset < pgd_index(USER_LIMIT)) {
-               struct page *page = virt_to_page(pgd_page);
-               user_ptr = (pgd_t *)page->private;
-               if (user_ptr)
-                       user_ptr += offset;
-       }
-
-       return user_ptr;
-}
-
-static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
-{
-       struct mmu_update u;
-
-       u.ptr = virt_to_machine(ptr).maddr;
-       u.val = p4d_val_ma(val);
-       xen_extend_mmu_update(&u);
-}
-
-/*
- * Raw hypercall-based set_p4d, intended for in early boot before
- * there's a page structure.  This implies:
- *  1. The only existing pagetable is the kernel's
- *  2. It is always pinned
- *  3. It has no user pagetable attached to it
- */
-static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
-{
-       preempt_disable();
-
-       xen_mc_batch();
-
-       __xen_set_p4d_hyper(ptr, val);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       preempt_enable();
-}
-
-static void xen_set_p4d(p4d_t *ptr, p4d_t val)
-{
-       pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
-       pgd_t pgd_val;
-
-       trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
-
-       /* If page is not pinned, we can just update the entry
-          directly */
-       if (!xen_page_pinned(ptr)) {
-               *ptr = val;
-               if (user_ptr) {
-                       WARN_ON(xen_page_pinned(user_ptr));
-                       pgd_val.pgd = p4d_val_ma(val);
-                       *user_ptr = pgd_val;
-               }
-               return;
-       }
-
-       /* If it's pinned, then we can at least batch the kernel and
-          user updates together. */
-       xen_mc_batch();
-
-       __xen_set_p4d_hyper(ptr, val);
-       if (user_ptr)
-               __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-}
-#endif /* CONFIG_PGTABLE_LEVELS == 4 */
-
-static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
-               int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
-               bool last, unsigned long limit)
-{
-       int i, nr, flush = 0;
-
-       nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
-       for (i = 0; i < nr; i++) {
-               if (!pmd_none(pmd[i]))
-                       flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE);
-       }
-       return flush;
-}
-
-static int xen_pud_walk(struct mm_struct *mm, pud_t *pud,
-               int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
-               bool last, unsigned long limit)
-{
-       int i, nr, flush = 0;
-
-       nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
-       for (i = 0; i < nr; i++) {
-               pmd_t *pmd;
-
-               if (pud_none(pud[i]))
-                       continue;
-
-               pmd = pmd_offset(&pud[i], 0);
-               if (PTRS_PER_PMD > 1)
-                       flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
-               flush |= xen_pmd_walk(mm, pmd, func,
-                               last && i == nr - 1, limit);
-       }
-       return flush;
-}
-
-static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
-               int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
-               bool last, unsigned long limit)
-{
-       int i, nr, flush = 0;
-
-       nr = last ? p4d_index(limit) + 1 : PTRS_PER_P4D;
-       for (i = 0; i < nr; i++) {
-               pud_t *pud;
-
-               if (p4d_none(p4d[i]))
-                       continue;
-
-               pud = pud_offset(&p4d[i], 0);
-               if (PTRS_PER_PUD > 1)
-                       flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
-               flush |= xen_pud_walk(mm, pud, func,
-                               last && i == nr - 1, limit);
-       }
-       return flush;
-}
-
-/*
- * (Yet another) pagetable walker.  This one is intended for pinning a
- * pagetable.  This means that it walks a pagetable and calls the
- * callback function on each page it finds making up the page table,
- * at every level.  It walks the entire pagetable, but it only bothers
- * pinning pte pages which are below limit.  In the normal case this
- * will be STACK_TOP_MAX, but at boot we need to pin up to
- * FIXADDR_TOP.
- *
- * For 32-bit the important bit is that we don't pin beyond there,
- * because then we start getting into Xen's ptes.
- *
- * For 64-bit, we must skip the Xen hole in the middle of the address
- * space, just after the big x86-64 virtual hole.
- */
-static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
-                         int (*func)(struct mm_struct *mm, struct page *,
-                                     enum pt_level),
-                         unsigned long limit)
-{
-       int i, nr, flush = 0;
-       unsigned hole_low, hole_high;
-
-       /* The limit is the last byte to be touched */
-       limit--;
-       BUG_ON(limit >= FIXADDR_TOP);
-
-       if (xen_feature(XENFEAT_auto_translated_physmap))
-               return 0;
-
-       /*
-        * 64-bit has a great big hole in the middle of the address
-        * space, which contains the Xen mappings.  On 32-bit these
-        * will end up making a zero-sized hole and so is a no-op.
-        */
-       hole_low = pgd_index(USER_LIMIT);
-       hole_high = pgd_index(PAGE_OFFSET);
-
-       nr = pgd_index(limit) + 1;
-       for (i = 0; i < nr; i++) {
-               p4d_t *p4d;
-
-               if (i >= hole_low && i < hole_high)
-                       continue;
-
-               if (pgd_none(pgd[i]))
-                       continue;
-
-               p4d = p4d_offset(&pgd[i], 0);
-               if (PTRS_PER_P4D > 1)
-                       flush |= (*func)(mm, virt_to_page(p4d), PT_P4D);
-               flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
-       }
-
-       /* Do the top level last, so that the callbacks can use it as
-          a cue to do final things like tlb flushes. */
-       flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
-
-       return flush;
-}
-
-static int xen_pgd_walk(struct mm_struct *mm,
-                       int (*func)(struct mm_struct *mm, struct page *,
-                                   enum pt_level),
-                       unsigned long limit)
-{
-       return __xen_pgd_walk(mm, mm->pgd, func, limit);
-}
-
-/* If we're using split pte locks, then take the page's lock and
-   return a pointer to it.  Otherwise return NULL. */
-static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
-{
-       spinlock_t *ptl = NULL;
-
-#if USE_SPLIT_PTE_PTLOCKS
-       ptl = ptlock_ptr(page);
-       spin_lock_nest_lock(ptl, &mm->page_table_lock);
-#endif
-
-       return ptl;
-}
-
-static void xen_pte_unlock(void *v)
-{
-       spinlock_t *ptl = v;
-       spin_unlock(ptl);
-}
-
-static void xen_do_pin(unsigned level, unsigned long pfn)
-{
-       struct mmuext_op op;
-
-       op.cmd = level;
-       op.arg1.mfn = pfn_to_mfn(pfn);
-
-       xen_extend_mmuext_op(&op);
-}
-
-static int xen_pin_page(struct mm_struct *mm, struct page *page,
-                       enum pt_level level)
-{
-       unsigned pgfl = TestSetPagePinned(page);
-       int flush;
-
-       if (pgfl)
-               flush = 0;              /* already pinned */
-       else if (PageHighMem(page))
-               /* kmaps need flushing if we found an unpinned
-                  highpage */
-               flush = 1;
-       else {
-               void *pt = lowmem_page_address(page);
-               unsigned long pfn = page_to_pfn(page);
-               struct multicall_space mcs = __xen_mc_entry(0);
-               spinlock_t *ptl;
-
-               flush = 0;
-
-               /*
-                * We need to hold the pagetable lock between the time
-                * we make the pagetable RO and when we actually pin
-                * it.  If we don't, then other users may come in and
-                * attempt to update the pagetable by writing it,
-                * which will fail because the memory is RO but not
-                * pinned, so Xen won't do the trap'n'emulate.
-                *
-                * If we're using split pte locks, we can't hold the
-                * entire pagetable's worth of locks during the
-                * traverse, because we may wrap the preempt count (8
-                * bits).  The solution is to mark RO and pin each PTE
-                * page while holding the lock.  This means the number
-                * of locks we end up holding is never more than a
-                * batch size (~32 entries, at present).
-                *
-                * If we're not using split pte locks, we needn't pin
-                * the PTE pages independently, because we're
-                * protected by the overall pagetable lock.
-                */
-               ptl = NULL;
-               if (level == PT_PTE)
-                       ptl = xen_pte_lock(page, mm);
-
-               MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
-                                       pfn_pte(pfn, PAGE_KERNEL_RO),
-                                       level == PT_PGD ? UVMF_TLB_FLUSH : 0);
-
-               if (ptl) {
-                       xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
-
-                       /* Queue a deferred unlock for when this batch
-                          is completed. */
-                       xen_mc_callback(xen_pte_unlock, ptl);
-               }
-       }
-
-       return flush;
-}
-
-/* This is called just after a mm has been created, but it has not
-   been used yet.  We need to make sure that its pagetable is all
-   read-only, and can be pinned. */
-static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
-{
-       trace_xen_mmu_pgd_pin(mm, pgd);
-
-       xen_mc_batch();
-
-       if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
-               /* re-enable interrupts for flushing */
-               xen_mc_issue(0);
-
-               kmap_flush_unused();
-
-               xen_mc_batch();
-       }
-
-#ifdef CONFIG_X86_64
-       {
-               pgd_t *user_pgd = xen_get_user_pgd(pgd);
-
-               xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
-
-               if (user_pgd) {
-                       xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
-                       xen_do_pin(MMUEXT_PIN_L4_TABLE,
-                                  PFN_DOWN(__pa(user_pgd)));
-               }
-       }
-#else /* CONFIG_X86_32 */
-#ifdef CONFIG_X86_PAE
-       /* Need to make sure unshared kernel PMD is pinnable */
-       xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
-                    PT_PMD);
-#endif
-       xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
-#endif /* CONFIG_X86_64 */
-       xen_mc_issue(0);
-}
-
-static void xen_pgd_pin(struct mm_struct *mm)
-{
-       __xen_pgd_pin(mm, mm->pgd);
-}
-
-/*
- * On save, we need to pin all pagetables to make sure they get their
- * mfns turned into pfns.  Search the list for any unpinned pgds and pin
- * them (unpinned pgds are not currently in use, probably because the
- * process is under construction or destruction).
- *
- * Expected to be called in stop_machine() ("equivalent to taking
- * every spinlock in the system"), so the locking doesn't really
- * matter all that much.
- */
-void xen_mm_pin_all(void)
-{
-       struct page *page;
-
-       spin_lock(&pgd_lock);
-
-       list_for_each_entry(page, &pgd_list, lru) {
-               if (!PagePinned(page)) {
-                       __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
-                       SetPageSavePinned(page);
-               }
-       }
-
-       spin_unlock(&pgd_lock);
-}
-
-/*
- * The init_mm pagetable is really pinned as soon as its created, but
- * that's before we have page structures to store the bits.  So do all
- * the book-keeping now.
- */
-static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
-                                 enum pt_level level)
-{
-       SetPagePinned(page);
-       return 0;
-}
-
-static void __init xen_mark_init_mm_pinned(void)
-{
-       xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
-}
-
-static int xen_unpin_page(struct mm_struct *mm, struct page *page,
-                         enum pt_level level)
-{
-       unsigned pgfl = TestClearPagePinned(page);
-
-       if (pgfl && !PageHighMem(page)) {
-               void *pt = lowmem_page_address(page);
-               unsigned long pfn = page_to_pfn(page);
-               spinlock_t *ptl = NULL;
-               struct multicall_space mcs;
-
-               /*
-                * Do the converse to pin_page.  If we're using split
-                * pte locks, we must be holding the lock for while
-                * the pte page is unpinned but still RO to prevent
-                * concurrent updates from seeing it in this
-                * partially-pinned state.
-                */
-               if (level == PT_PTE) {
-                       ptl = xen_pte_lock(page, mm);
-
-                       if (ptl)
-                               xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
-               }
-
-               mcs = __xen_mc_entry(0);
-
-               MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
-                                       pfn_pte(pfn, PAGE_KERNEL),
-                                       level == PT_PGD ? UVMF_TLB_FLUSH : 0);
-
-               if (ptl) {
-                       /* unlock when batch completed */
-                       xen_mc_callback(xen_pte_unlock, ptl);
-               }
-       }
-
-       return 0;               /* never need to flush on unpin */
-}
-
-/* Release a pagetables pages back as normal RW */
-static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
-{
-       trace_xen_mmu_pgd_unpin(mm, pgd);
-
-       xen_mc_batch();
-
-       xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
-
-#ifdef CONFIG_X86_64
-       {
-               pgd_t *user_pgd = xen_get_user_pgd(pgd);
-
-               if (user_pgd) {
-                       xen_do_pin(MMUEXT_UNPIN_TABLE,
-                                  PFN_DOWN(__pa(user_pgd)));
-                       xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
-               }
-       }
-#endif
-
-#ifdef CONFIG_X86_PAE
-       /* Need to make sure unshared kernel PMD is unpinned */
-       xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
-                      PT_PMD);
-#endif
-
-       __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
-
-       xen_mc_issue(0);
-}
-
-static void xen_pgd_unpin(struct mm_struct *mm)
-{
-       __xen_pgd_unpin(mm, mm->pgd);
-}
-
-/*
- * On resume, undo any pinning done at save, so that the rest of the
- * kernel doesn't see any unexpected pinned pagetables.
- */
-void xen_mm_unpin_all(void)
-{
-       struct page *page;
-
-       spin_lock(&pgd_lock);
-
-       list_for_each_entry(page, &pgd_list, lru) {
-               if (PageSavePinned(page)) {
-                       BUG_ON(!PagePinned(page));
-                       __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
-                       ClearPageSavePinned(page);
-               }
-       }
-
-       spin_unlock(&pgd_lock);
-}
-
-static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
-{
-       spin_lock(&next->page_table_lock);
-       xen_pgd_pin(next);
-       spin_unlock(&next->page_table_lock);
-}
-
-static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
-{
-       spin_lock(&mm->page_table_lock);
-       xen_pgd_pin(mm);
-       spin_unlock(&mm->page_table_lock);
-}
-
-
-#ifdef CONFIG_SMP
-/* Another cpu may still have their %cr3 pointing at the pagetable, so
-   we need to repoint it somewhere else before we can unpin it. */
-static void drop_other_mm_ref(void *info)
-{
-       struct mm_struct *mm = info;
-       struct mm_struct *active_mm;
-
-       active_mm = this_cpu_read(cpu_tlbstate.active_mm);
-
-       if (active_mm == mm && this_cpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
-               leave_mm(smp_processor_id());
-
-       /* If this cpu still has a stale cr3 reference, then make sure
-          it has been flushed. */
-       if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
-               load_cr3(swapper_pg_dir);
-}
-
-static void xen_drop_mm_ref(struct mm_struct *mm)
-{
-       cpumask_var_t mask;
-       unsigned cpu;
-
-       if (current->active_mm == mm) {
-               if (current->mm == mm)
-                       load_cr3(swapper_pg_dir);
-               else
-                       leave_mm(smp_processor_id());
-       }
-
-       /* Get the "official" set of cpus referring to our pagetable. */
-       if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
-               for_each_online_cpu(cpu) {
-                       if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
-                           && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
-                               continue;
-                       smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
-               }
-               return;
-       }
-       cpumask_copy(mask, mm_cpumask(mm));
-
-       /* It's possible that a vcpu may have a stale reference to our
-          cr3, because its in lazy mode, and it hasn't yet flushed
-          its set of pending hypercalls yet.  In this case, we can
-          look at its actual current cr3 value, and force it to flush
-          if needed. */
-       for_each_online_cpu(cpu) {
-               if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
-                       cpumask_set_cpu(cpu, mask);
-       }
-
-       if (!cpumask_empty(mask))
-               smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
-       free_cpumask_var(mask);
-}
-#else
-static void xen_drop_mm_ref(struct mm_struct *mm)
-{
-       if (current->active_mm == mm)
-               load_cr3(swapper_pg_dir);
-}
-#endif
-
-/*
- * While a process runs, Xen pins its pagetables, which means that the
- * hypervisor forces it to be read-only, and it controls all updates
- * to it.  This means that all pagetable updates have to go via the
- * hypervisor, which is moderately expensive.
- *
- * Since we're pulling the pagetable down, we switch to use init_mm,
- * unpin old process pagetable and mark it all read-write, which
- * allows further operations on it to be simple memory accesses.
- *
- * The only subtle point is that another CPU may be still using the
- * pagetable because of lazy tlb flushing.  This means we need need to
- * switch all CPUs off this pagetable before we can unpin it.
- */
-static void xen_exit_mmap(struct mm_struct *mm)
-{
-       get_cpu();              /* make sure we don't move around */
-       xen_drop_mm_ref(mm);
-       put_cpu();
-
-       spin_lock(&mm->page_table_lock);
-
-       /* pgd may not be pinned in the error exit path of execve */
-       if (xen_page_pinned(mm->pgd))
-               xen_pgd_unpin(mm);
-
-       spin_unlock(&mm->page_table_lock);
-}
-
-static void xen_post_allocator_init(void);
-
-static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
-{
-       struct mmuext_op op;
-
-       op.cmd = cmd;
-       op.arg1.mfn = pfn_to_mfn(pfn);
-       if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
-               BUG();
-}
-
-#ifdef CONFIG_X86_64
-static void __init xen_cleanhighmap(unsigned long vaddr,
-                                   unsigned long vaddr_end)
-{
-       unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
-       pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
-
-       /* NOTE: The loop is more greedy than the cleanup_highmap variant.
-        * We include the PMD passed in on _both_ boundaries. */
-       for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
-                       pmd++, vaddr += PMD_SIZE) {
-               if (pmd_none(*pmd))
-                       continue;
-               if (vaddr < (unsigned long) _text || vaddr > kernel_end)
-                       set_pmd(pmd, __pmd(0));
-       }
-       /* In case we did something silly, we should crash in this function
-        * instead of somewhere later and be confusing. */
-       xen_mc_flush();
-}
-
-/*
- * Make a page range writeable and free it.
- */
-static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
-{
-       void *vaddr = __va(paddr);
-       void *vaddr_end = vaddr + size;
-
-       for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
-               make_lowmem_page_readwrite(vaddr);
-
-       memblock_free(paddr, size);
-}
-
-static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
-{
-       unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
-
-       if (unpin)
-               pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
-       ClearPagePinned(virt_to_page(__va(pa)));
-       xen_free_ro_pages(pa, PAGE_SIZE);
-}
-
-static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
-{
-       unsigned long pa;
-       pte_t *pte_tbl;
-       int i;
-
-       if (pmd_large(*pmd)) {
-               pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
-               xen_free_ro_pages(pa, PMD_SIZE);
-               return;
-       }
-
-       pte_tbl = pte_offset_kernel(pmd, 0);
-       for (i = 0; i < PTRS_PER_PTE; i++) {
-               if (pte_none(pte_tbl[i]))
-                       continue;
-               pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
-               xen_free_ro_pages(pa, PAGE_SIZE);
-       }
-       set_pmd(pmd, __pmd(0));
-       xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
-}
-
-static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
-{
-       unsigned long pa;
-       pmd_t *pmd_tbl;
-       int i;
-
-       if (pud_large(*pud)) {
-               pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
-               xen_free_ro_pages(pa, PUD_SIZE);
-               return;
-       }
-
-       pmd_tbl = pmd_offset(pud, 0);
-       for (i = 0; i < PTRS_PER_PMD; i++) {
-               if (pmd_none(pmd_tbl[i]))
-                       continue;
-               xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
-       }
-       set_pud(pud, __pud(0));
-       xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
-}
-
-static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
-{
-       unsigned long pa;
-       pud_t *pud_tbl;
-       int i;
-
-       if (p4d_large(*p4d)) {
-               pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
-               xen_free_ro_pages(pa, P4D_SIZE);
-               return;
-       }
-
-       pud_tbl = pud_offset(p4d, 0);
-       for (i = 0; i < PTRS_PER_PUD; i++) {
-               if (pud_none(pud_tbl[i]))
-                       continue;
-               xen_cleanmfnmap_pud(pud_tbl + i, unpin);
-       }
-       set_p4d(p4d, __p4d(0));
-       xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
-}
-
-/*
- * Since it is well isolated we can (and since it is perhaps large we should)
- * also free the page tables mapping the initial P->M table.
- */
-static void __init xen_cleanmfnmap(unsigned long vaddr)
-{
-       pgd_t *pgd;
-       p4d_t *p4d;
-       unsigned int i;
-       bool unpin;
-
-       unpin = (vaddr == 2 * PGDIR_SIZE);
-       vaddr &= PMD_MASK;
-       pgd = pgd_offset_k(vaddr);
-       p4d = p4d_offset(pgd, 0);
-       for (i = 0; i < PTRS_PER_P4D; i++) {
-               if (p4d_none(p4d[i]))
-                       continue;
-               xen_cleanmfnmap_p4d(p4d + i, unpin);
-       }
-       if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
-               set_pgd(pgd, __pgd(0));
-               xen_cleanmfnmap_free_pgtbl(p4d, unpin);
-       }
-}
-
-static void __init xen_pagetable_p2m_free(void)
-{
-       unsigned long size;
-       unsigned long addr;
-
-       size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
-
-       /* No memory or already called. */
-       if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
-               return;
-
-       /* using __ka address and sticking INVALID_P2M_ENTRY! */
-       memset((void *)xen_start_info->mfn_list, 0xff, size);
-
-       addr = xen_start_info->mfn_list;
-       /*
-        * We could be in __ka space.
-        * We roundup to the PMD, which means that if anybody at this stage is
-        * using the __ka address of xen_start_info or
-        * xen_start_info->shared_info they are in going to crash. Fortunatly
-        * we have already revectored in xen_setup_kernel_pagetable and in
-        * xen_setup_shared_info.
-        */
-       size = roundup(size, PMD_SIZE);
-
-       if (addr >= __START_KERNEL_map) {
-               xen_cleanhighmap(addr, addr + size);
-               size = PAGE_ALIGN(xen_start_info->nr_pages *
-                                 sizeof(unsigned long));
-               memblock_free(__pa(addr), size);
-       } else {
-               xen_cleanmfnmap(addr);
-       }
-}
-
-static void __init xen_pagetable_cleanhighmap(void)
-{
-       unsigned long size;
-       unsigned long addr;
-
-       /* At this stage, cleanup_highmap has already cleaned __ka space
-        * from _brk_limit way up to the max_pfn_mapped (which is the end of
-        * the ramdisk). We continue on, erasing PMD entries that point to page
-        * tables - do note that they are accessible at this stage via __va.
-        * For good measure we also round up to the PMD - which means that if
-        * anybody is using __ka address to the initial boot-stack - and try
-        * to use it - they are going to crash. The xen_start_info has been
-        * taken care of already in xen_setup_kernel_pagetable. */
-       addr = xen_start_info->pt_base;
-       size = roundup(xen_start_info->nr_pt_frames * PAGE_SIZE, PMD_SIZE);
-
-       xen_cleanhighmap(addr, addr + size);
-       xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
-#ifdef DEBUG
-       /* This is superfluous and is not necessary, but you know what
-        * lets do it. The MODULES_VADDR -> MODULES_END should be clear of
-        * anything at this stage. */
-       xen_cleanhighmap(MODULES_VADDR, roundup(MODULES_VADDR, PUD_SIZE) - 1);
-#endif
-}
-#endif
-
-static void __init xen_pagetable_p2m_setup(void)
-{
-       if (xen_feature(XENFEAT_auto_translated_physmap))
-               return;
-
-       xen_vmalloc_p2m_tree();
-
-#ifdef CONFIG_X86_64
-       xen_pagetable_p2m_free();
-
-       xen_pagetable_cleanhighmap();
-#endif
-       /* And revector! Bye bye old array */
-       xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
-}
-
-static void __init xen_pagetable_init(void)
-{
-       paging_init();
-       xen_post_allocator_init();
-
-       xen_pagetable_p2m_setup();
-
-       /* Allocate and initialize top and mid mfn levels for p2m structure */
-       xen_build_mfn_list_list();
-
-       /* Remap memory freed due to conflicts with E820 map */
-       if (!xen_feature(XENFEAT_auto_translated_physmap))
-               xen_remap_memory();
-
-       xen_setup_shared_info();
-}
-static void xen_write_cr2(unsigned long cr2)
-{
-       this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
-}
-
-static unsigned long xen_read_cr2(void)
-{
-       return this_cpu_read(xen_vcpu)->arch.cr2;
-}
-
-unsigned long xen_read_cr2_direct(void)
-{
-       return this_cpu_read(xen_vcpu_info.arch.cr2);
-}
-
-void xen_flush_tlb_all(void)
-{
-       struct mmuext_op *op;
-       struct multicall_space mcs;
-
-       trace_xen_mmu_flush_tlb_all(0);
-
-       preempt_disable();
-
-       mcs = xen_mc_entry(sizeof(*op));
-
-       op = mcs.args;
-       op->cmd = MMUEXT_TLB_FLUSH_ALL;
-       MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       preempt_enable();
-}
-static void xen_flush_tlb(void)
-{
-       struct mmuext_op *op;
-       struct multicall_space mcs;
-
-       trace_xen_mmu_flush_tlb(0);
-
-       preempt_disable();
-
-       mcs = xen_mc_entry(sizeof(*op));
-
-       op = mcs.args;
-       op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
-       MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       preempt_enable();
-}
-
-static void xen_flush_tlb_single(unsigned long addr)
-{
-       struct mmuext_op *op;
-       struct multicall_space mcs;
-
-       trace_xen_mmu_flush_tlb_single(addr);
-
-       preempt_disable();
-
-       mcs = xen_mc_entry(sizeof(*op));
-       op = mcs.args;
-       op->cmd = MMUEXT_INVLPG_LOCAL;
-       op->arg1.linear_addr = addr & PAGE_MASK;
-       MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-
-       preempt_enable();
-}
-
-static void xen_flush_tlb_others(const struct cpumask *cpus,
-                                struct mm_struct *mm, unsigned long start,
-                                unsigned long end)
-{
-       struct {
-               struct mmuext_op op;
-#ifdef CONFIG_SMP
-               DECLARE_BITMAP(mask, num_processors);
-#else
-               DECLARE_BITMAP(mask, NR_CPUS);
-#endif
-       } *args;
-       struct multicall_space mcs;
-
-       trace_xen_mmu_flush_tlb_others(cpus, mm, start, end);
-
-       if (cpumask_empty(cpus))
-               return;         /* nothing to do */
-
-       mcs = xen_mc_entry(sizeof(*args));
-       args = mcs.args;
-       args->op.arg2.vcpumask = to_cpumask(args->mask);
-
-       /* Remove us, and any offline CPUS. */
-       cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
-       cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
-
-       args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
-       if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) {
-               args->op.cmd = MMUEXT_INVLPG_MULTI;
-               args->op.arg1.linear_addr = start;
-       }
-
-       MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
-
-       xen_mc_issue(PARAVIRT_LAZY_MMU);
-}
-
-static unsigned long xen_read_cr3(void)
-{
-       return this_cpu_read(xen_cr3);
-}
-
-static void set_current_cr3(void *v)
-{
-       this_cpu_write(xen_current_cr3, (unsigned long)v);
-}
-
-static void __xen_write_cr3(bool kernel, unsigned long cr3)
-{
-       struct mmuext_op op;
-       unsigned long mfn;
-
-       trace_xen_mmu_write_cr3(kernel, cr3);
-
-       if (cr3)
-               mfn = pfn_to_mfn(PFN_DOWN(cr3));
-       else
-               mfn = 0;
-
-       WARN_ON(mfn == 0 && kernel);
-
-       op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
-       op.arg1.mfn = mfn;
-
-       xen_extend_mmuext_op(&op);
-
-       if (kernel) {
-               this_cpu_write(xen_cr3, cr3);
-
-               /* Update xen_current_cr3 once the batch has actually
-                  been submitted. */
-               xen_mc_callback(set_current_cr3, (void *)cr3);
-       }
-}
-static void xen_write_cr3(unsigned long cr3)
-{
-       BUG_ON(preemptible());
-
-       xen_mc_batch();  /* disables interrupts */
-
-       /* Update while interrupts are disabled, so its atomic with
-          respect to ipis */
-       this_cpu_write(xen_cr3, cr3);
-
-       __xen_write_cr3(true, cr3);
-
-#ifdef CONFIG_X86_64
-       {
-               pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
-               if (user_pgd)
-                       __xen_write_cr3(false, __pa(user_pgd));
-               else
-                       __xen_write_cr3(false, 0);
-       }
-#endif
-
-       xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
-}
-
-#ifdef CONFIG_X86_64
-/*
- * At the start of the day - when Xen launches a guest, it has already
- * built pagetables for the guest. We diligently look over them
- * in xen_setup_kernel_pagetable and graft as appropriate them in the
- * init_level4_pgt and its friends. Then when we are happy we load
- * the new init_level4_pgt - and continue on.
- *
- * The generic code starts (start_kernel) and 'init_mem_mapping' sets
- * up the rest of the pagetables. When it has completed it loads the cr3.
- * N.B. that baremetal would start at 'start_kernel' (and the early
- * #PF handler would create bootstrap pagetables) - so we are running
- * with the same assumptions as what to do when write_cr3 is executed
- * at this point.
- *
- * Since there are no user-page tables at all, we have two variants
- * of xen_write_cr3 - the early bootup (this one), and the late one
- * (xen_write_cr3). The reason we have to do that is that in 64-bit
- * the Linux kernel and user-space are both in ring 3 while the
- * hypervisor is in ring 0.
- */
-static void __init xen_write_cr3_init(unsigned long cr3)
-{
-       BUG_ON(preemptible());
-
-       xen_mc_batch();  /* disables interrupts */
-
-       /* Update while interrupts are disabled, so its atomic with
-          respect to ipis */
-       this_cpu_write(xen_cr3, cr3);
-
-       __xen_write_cr3(true, cr3);
-
-       xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
-}
-#endif
-
-static int xen_pgd_alloc(struct mm_struct *mm)
-{
-       pgd_t *pgd = mm->pgd;
-       int ret = 0;
-
-       BUG_ON(PagePinned(virt_to_page(pgd)));
-
-#ifdef CONFIG_X86_64
-       {
-               struct page *page = virt_to_page(pgd);
-               pgd_t *user_pgd;
-
-               BUG_ON(page->private != 0);
-
-               ret = -ENOMEM;
-
-               user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
-               page->private = (unsigned long)user_pgd;
-
-               if (user_pgd != NULL) {
-#ifdef CONFIG_X86_VSYSCALL_EMULATION
-                       user_pgd[pgd_index(VSYSCALL_ADDR)] =
-                               __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
-#endif
-                       ret = 0;
-               }
-
-               BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
-       }
-#endif
-       return ret;
-}
-
-static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
-{
-#ifdef CONFIG_X86_64
-       pgd_t *user_pgd = xen_get_user_pgd(pgd);
-
-       if (user_pgd)
-               free_page((unsigned long)user_pgd);
-#endif
-}
-
-/*
- * Init-time set_pte while constructing initial pagetables, which
- * doesn't allow RO page table pages to be remapped RW.
- *
- * If there is no MFN for this PFN then this page is initially
- * ballooned out so clear the PTE (as in decrease_reservation() in
- * drivers/xen/balloon.c).
- *
- * Many of these PTE updates are done on unpinned and writable pages
- * and doing a hypercall for these is unnecessary and expensive.  At
- * this point it is not possible to tell if a page is pinned or not,
- * so always write the PTE directly and rely on Xen trapping and
- * emulating any updates as necessary.
- */
-__visible pte_t xen_make_pte_init(pteval_t pte)
-{
-#ifdef CONFIG_X86_64
-       unsigned long pfn;
-
-       /*
-        * Pages belonging to the initial p2m list mapped outside the default
-        * address range must be mapped read-only. This region contains the
-        * page tables for mapping the p2m list, too, and page tables MUST be
-        * mapped read-only.
-        */
-       pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
-       if (xen_start_info->mfn_list < __START_KERNEL_map &&
-           pfn >= xen_start_info->first_p2m_pfn &&
-           pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
-               pte &= ~_PAGE_RW;
-#endif
-       pte = pte_pfn_to_mfn(pte);
-       return native_make_pte(pte);
-}
-PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
-
-static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
-{
-#ifdef CONFIG_X86_32
-       /* If there's an existing pte, then don't allow _PAGE_RW to be set */
-       if (pte_mfn(pte) != INVALID_P2M_ENTRY
-           && pte_val_ma(*ptep) & _PAGE_PRESENT)
-               pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
-                              pte_val_ma(pte));
-#endif
-       native_set_pte(ptep, pte);
-}
-
-/* Early in boot, while setting up the initial pagetable, assume
-   everything is pinned. */
-static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
-{
-#ifdef CONFIG_FLATMEM
-       BUG_ON(mem_map);        /* should only be used early */
-#endif
-       make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
-       pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
-}
-
-/* Used for pmd and pud */
-static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
-{
-#ifdef CONFIG_FLATMEM
-       BUG_ON(mem_map);        /* should only be used early */
-#endif
-       make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
-}
-
-/* Early release_pte assumes that all pts are pinned, since there's
-   only init_mm and anything attached to that is pinned. */
-static void __init xen_release_pte_init(unsigned long pfn)
-{
-       pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
-       make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
-}
-
-static void __init xen_release_pmd_init(unsigned long pfn)
-{
-       make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
-}
-
-static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
-{
-       struct multicall_space mcs;
-       struct mmuext_op *op;
-
-       mcs = __xen_mc_entry(sizeof(*op));
-       op = mcs.args;
-       op->cmd = cmd;
-       op->arg1.mfn = pfn_to_mfn(pfn);
-
-       MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
-}
-
-static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
-{
-       struct multicall_space mcs;
-       unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
-
-       mcs = __xen_mc_entry(0);
-       MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
-                               pfn_pte(pfn, prot), 0);
-}
-
-/* This needs to make sure the new pte page is pinned iff its being
-   attached to a pinned pagetable. */
-static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
-                                   unsigned level)
-{
-       bool pinned = PagePinned(virt_to_page(mm->pgd));
-
-       trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
-
-       if (pinned) {
-               struct page *page = pfn_to_page(pfn);
-
-               SetPagePinned(page);
-
-               if (!PageHighMem(page)) {
-                       xen_mc_batch();
-
-                       __set_pfn_prot(pfn, PAGE_KERNEL_RO);
-
-                       if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
-                               __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
-
-                       xen_mc_issue(PARAVIRT_LAZY_MMU);
-               } else {
-                       /* make sure there are no stray mappings of
-                          this page */
-                       kmap_flush_unused();
-               }
-       }
-}
-
-static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
-{
-       xen_alloc_ptpage(mm, pfn, PT_PTE);
-}
-
-static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
-{
-       xen_alloc_ptpage(mm, pfn, PT_PMD);
-}
-
-/* This should never happen until we're OK to use struct page */
-static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
-{
-       struct page *page = pfn_to_page(pfn);
-       bool pinned = PagePinned(page);
-
-       trace_xen_mmu_release_ptpage(pfn, level, pinned);
-
-       if (pinned) {
-               if (!PageHighMem(page)) {
-                       xen_mc_batch();
-
-                       if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
-                               __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
-
-                       __set_pfn_prot(pfn, PAGE_KERNEL);
-
-                       xen_mc_issue(PARAVIRT_LAZY_MMU);
-               }
-               ClearPagePinned(page);
-       }
-}
-
-static void xen_release_pte(unsigned long pfn)
-{
-       xen_release_ptpage(pfn, PT_PTE);
-}
-
-static void xen_release_pmd(unsigned long pfn)
-{
-       xen_release_ptpage(pfn, PT_PMD);
-}
-
-#if CONFIG_PGTABLE_LEVELS >= 4
-static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
-{
-       xen_alloc_ptpage(mm, pfn, PT_PUD);
-}
-
-static void xen_release_pud(unsigned long pfn)
-{
-       xen_release_ptpage(pfn, PT_PUD);
-}
-#endif
-
-void __init xen_reserve_top(void)
-{
-#ifdef CONFIG_X86_32
-       unsigned long top = HYPERVISOR_VIRT_START;
-       struct xen_platform_parameters pp;
-
-       if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
-               top = pp.virt_start;
-
-       reserve_top_address(-top);
-#endif /* CONFIG_X86_32 */
-}
-
-/*
- * Like __va(), but returns address in the kernel mapping (which is
- * all we have until the physical memory mapping has been set up.
- */
-static void * __init __ka(phys_addr_t paddr)
-{
-#ifdef CONFIG_X86_64
-       return (void *)(paddr + __START_KERNEL_map);
-#else
-       return __va(paddr);
-#endif
-}
-
-/* Convert a machine address to physical address */
-static unsigned long __init m2p(phys_addr_t maddr)
-{
-       phys_addr_t paddr;
-
-       maddr &= PTE_PFN_MASK;
-       paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
-
-       return paddr;
-}
-
-/* Convert a machine address to kernel virtual */
-static void * __init m2v(phys_addr_t maddr)
-{
-       return __ka(m2p(maddr));
-}
-
-/* Set the page permissions on an identity-mapped pages */
-static void __init set_page_prot_flags(void *addr, pgprot_t prot,
-                                      unsigned long flags)
-{
-       unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
-       pte_t pte = pfn_pte(pfn, prot);
-
-       if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
-               BUG();
-}
-static void __init set_page_prot(void *addr, pgprot_t prot)
-{
-       return set_page_prot_flags(addr, prot, UVMF_NONE);
-}
-#ifdef CONFIG_X86_32
-static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
-{
-       unsigned pmdidx, pteidx;
-       unsigned ident_pte;
-       unsigned long pfn;
-
-       level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
-                                     PAGE_SIZE);
-
-       ident_pte = 0;
-       pfn = 0;
-       for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
-               pte_t *pte_page;
-
-               /* Reuse or allocate a page of ptes */
-               if (pmd_present(pmd[pmdidx]))
-                       pte_page = m2v(pmd[pmdidx].pmd);
-               else {
-                       /* Check for free pte pages */
-                       if (ident_pte == LEVEL1_IDENT_ENTRIES)
-                               break;
-
-                       pte_page = &level1_ident_pgt[ident_pte];
-                       ident_pte += PTRS_PER_PTE;
-
-                       pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
-               }
-
-               /* Install mappings */
-               for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
-                       pte_t pte;
-
-                       if (pfn > max_pfn_mapped)
-                               max_pfn_mapped = pfn;
-
-                       if (!pte_none(pte_page[pteidx]))
-                               continue;
-
-                       pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
-                       pte_page[pteidx] = pte;
-               }
-       }
-
-       for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
-               set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
-
-       set_page_prot(pmd, PAGE_KERNEL_RO);
-}
-#endif
-void __init xen_setup_machphys_mapping(void)
-{
-       struct xen_machphys_mapping mapping;
-
-       if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
-               machine_to_phys_mapping = (unsigned long *)mapping.v_start;
-               machine_to_phys_nr = mapping.max_mfn + 1;
-       } else {
-               machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
-       }
-#ifdef CONFIG_X86_32
-       WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
-               < machine_to_phys_mapping);
-#endif
-}
-
-#ifdef CONFIG_X86_64
-static void __init convert_pfn_mfn(void *v)
-{
-       pte_t *pte = v;
-       int i;
-
-       /* All levels are converted the same way, so just treat them
-          as ptes. */
-       for (i = 0; i < PTRS_PER_PTE; i++)
-               pte[i] = xen_make_pte(pte[i].pte);
-}
-static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
-                                unsigned long addr)
-{
-       if (*pt_base == PFN_DOWN(__pa(addr))) {
-               set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
-               clear_page((void *)addr);
-               (*pt_base)++;
-       }
-       if (*pt_end == PFN_DOWN(__pa(addr))) {
-               set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
-               clear_page((void *)addr);
-               (*pt_end)--;
-       }
-}
-/*
- * Set up the initial kernel pagetable.
- *
- * We can construct this by grafting the Xen provided pagetable into
- * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
- * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
- * kernel has a physical mapping to start with - but that's enough to
- * get __va working.  We need to fill in the rest of the physical
- * mapping once some sort of allocator has been set up.
- */
-void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
-{
-       pud_t *l3;
-       pmd_t *l2;
-       unsigned long addr[3];
-       unsigned long pt_base, pt_end;
-       unsigned i;
-
-       /* max_pfn_mapped is the last pfn mapped in the initial memory
-        * mappings. Considering that on Xen after the kernel mappings we
-        * have the mappings of some pages that don't exist in pfn space, we
-        * set max_pfn_mapped to the last real pfn mapped. */
-       if (xen_start_info->mfn_list < __START_KERNEL_map)
-               max_pfn_mapped = xen_start_info->first_p2m_pfn;
-       else
-               max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
-
-       pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
-       pt_end = pt_base + xen_start_info->nr_pt_frames;
-
-       /* Zap identity mapping */
-       init_level4_pgt[0] = __pgd(0);
-
-       if (!xen_feature(XENFEAT_auto_translated_physmap)) {
-               /* Pre-constructed entries are in pfn, so convert to mfn */
-               /* L4[272] -> level3_ident_pgt
-                * L4[511] -> level3_kernel_pgt */
-               convert_pfn_mfn(init_level4_pgt);
-
-               /* L3_i[0] -> level2_ident_pgt */
-               convert_pfn_mfn(level3_ident_pgt);
-               /* L3_k[510] -> level2_kernel_pgt
-                * L3_k[511] -> level2_fixmap_pgt */
-               convert_pfn_mfn(level3_kernel_pgt);
-
-               /* L3_k[511][506] -> level1_fixmap_pgt */
-               convert_pfn_mfn(level2_fixmap_pgt);
-       }
-       /* We get [511][511] and have Xen's version of level2_kernel_pgt */
-       l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
-       l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
-
-       addr[0] = (unsigned long)pgd;
-       addr[1] = (unsigned long)l3;
-       addr[2] = (unsigned long)l2;
-       /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
-        * Both L4[272][0] and L4[511][510] have entries that point to the same
-        * L2 (PMD) tables. Meaning that if you modify it in __va space
-        * it will be also modified in the __ka space! (But if you just
-        * modify the PMD table to point to other PTE's or none, then you
-        * are OK - which is what cleanup_highmap does) */
-       copy_page(level2_ident_pgt, l2);
-       /* Graft it onto L4[511][510] */
-       copy_page(level2_kernel_pgt, l2);
-
-       /* Copy the initial P->M table mappings if necessary. */
-       i = pgd_index(xen_start_info->mfn_list);
-       if (i && i < pgd_index(__START_KERNEL_map))
-               init_level4_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
-
-       if (!xen_feature(XENFEAT_auto_translated_physmap)) {
-               /* Make pagetable pieces RO */
-               set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
-               set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
-               set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
-               set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
-               set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
-               set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
-               set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
-               set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
-
-               /* Pin down new L4 */
-               pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
-                                 PFN_DOWN(__pa_symbol(init_level4_pgt)));
-
-               /* Unpin Xen-provided one */
-               pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
-
-               /*
-                * At this stage there can be no user pgd, and no page
-                * structure to attach it to, so make sure we just set kernel
-                * pgd.
-                */
-               xen_mc_batch();
-               __xen_write_cr3(true, __pa(init_level4_pgt));
-               xen_mc_issue(PARAVIRT_LAZY_CPU);
-       } else
-               native_write_cr3(__pa(init_level4_pgt));
-
-       /* We can't that easily rip out L3 and L2, as the Xen pagetables are
-        * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
-        * the initial domain. For guests using the toolstack, they are in:
-        * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
-        * rip out the [L4] (pgd), but for guests we shave off three pages.
-        */
-       for (i = 0; i < ARRAY_SIZE(addr); i++)
-               check_pt_base(&pt_base, &pt_end, addr[i]);
-
-       /* Our (by three pages) smaller Xen pagetable that we are using */
-       xen_pt_base = PFN_PHYS(pt_base);
-       xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
-       memblock_reserve(xen_pt_base, xen_pt_size);
-
-       /* Revector the xen_start_info */
-       xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
-}
-
-/*
- * Read a value from a physical address.
- */
-static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
-{
-       unsigned long *vaddr;
-       unsigned long val;
+#include <linux/pfn.h>
+#include <asm/xen/page.h>
+#include <asm/xen/hypercall.h>
+#include <xen/interface/memory.h>
 
-       vaddr = early_memremap_ro(addr, sizeof(val));
-       val = *vaddr;
-       early_memunmap(vaddr, sizeof(val));
-       return val;
-}
+#include "multicalls.h"
+#include "mmu.h"
 
 /*
- * Translate a virtual address to a physical one without relying on mapped
- * page tables.
+ * Protects atomic reservation decrease/increase against concurrent increases.
+ * Also protects non-atomic updates of current_pages and balloon lists.
  */
-static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
-{
-       phys_addr_t pa;
-       pgd_t pgd;
-       pud_t pud;
-       pmd_t pmd;
-       pte_t pte;
-
-       pa = read_cr3();
-       pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
-                                                      sizeof(pgd)));
-       if (!pgd_present(pgd))
-               return 0;
-
-       pa = pgd_val(pgd) & PTE_PFN_MASK;
-       pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
-                                                      sizeof(pud)));
-       if (!pud_present(pud))
-               return 0;
-       pa = pud_pfn(pud) << PAGE_SHIFT;
-       if (pud_large(pud))
-               return pa + (vaddr & ~PUD_MASK);
-
-       pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
-                                                      sizeof(pmd)));
-       if (!pmd_present(pmd))
-               return 0;
-       pa = pmd_pfn(pmd) << PAGE_SHIFT;
-       if (pmd_large(pmd))
-               return pa + (vaddr & ~PMD_MASK);
-
-       pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
-                                                      sizeof(pte)));
-       if (!pte_present(pte))
-               return 0;
-       pa = pte_pfn(pte) << PAGE_SHIFT;
-
-       return pa | (vaddr & ~PAGE_MASK);
-}
+DEFINE_SPINLOCK(xen_reservation_lock);
 
-/*
- * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
- * this area.
- */
-void __init xen_relocate_p2m(void)
+unsigned long arbitrary_virt_to_mfn(void *vaddr)
 {
-       phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys, p4d_phys;
-       unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
-       int n_pte, n_pt, n_pmd, n_pud, n_p4d, idx_pte, idx_pt, idx_pmd, idx_pud, idx_p4d;
-       pte_t *pt;
-       pmd_t *pmd;
-       pud_t *pud;
-       p4d_t *p4d = NULL;
-       pgd_t *pgd;
-       unsigned long *new_p2m;
-       int save_pud;
-
-       size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
-       n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
-       n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
-       n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
-       n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
-       if (PTRS_PER_P4D > 1)
-               n_p4d = roundup(size, PGDIR_SIZE) >> PGDIR_SHIFT;
-       else
-               n_p4d = 0;
-       n_frames = n_pte + n_pt + n_pmd + n_pud + n_p4d;
-
-       new_area = xen_find_free_area(PFN_PHYS(n_frames));
-       if (!new_area) {
-               xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
-               BUG();
-       }
-
-       /*
-        * Setup the page tables for addressing the new p2m list.
-        * We have asked the hypervisor to map the p2m list at the user address
-        * PUD_SIZE. It may have done so, or it may have used a kernel space
-        * address depending on the Xen version.
-        * To avoid any possible virtual address collision, just use
-        * 2 * PUD_SIZE for the new area.
-        */
-       p4d_phys = new_area;
-       pud_phys = p4d_phys + PFN_PHYS(n_p4d);
-       pmd_phys = pud_phys + PFN_PHYS(n_pud);
-       pt_phys = pmd_phys + PFN_PHYS(n_pmd);
-       p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
-
-       pgd = __va(read_cr3());
-       new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
-       idx_p4d = 0;
-       save_pud = n_pud;
-       do {
-               if (n_p4d > 0) {
-                       p4d = early_memremap(p4d_phys, PAGE_SIZE);
-                       clear_page(p4d);
-                       n_pud = min(save_pud, PTRS_PER_P4D);
-               }
-               for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
-                       pud = early_memremap(pud_phys, PAGE_SIZE);
-                       clear_page(pud);
-                       for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
-                                idx_pmd++) {
-                               pmd = early_memremap(pmd_phys, PAGE_SIZE);
-                               clear_page(pmd);
-                               for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
-                                        idx_pt++) {
-                                       pt = early_memremap(pt_phys, PAGE_SIZE);
-                                       clear_page(pt);
-                                       for (idx_pte = 0;
-                                                idx_pte < min(n_pte, PTRS_PER_PTE);
-                                                idx_pte++) {
-                                               set_pte(pt + idx_pte,
-                                                               pfn_pte(p2m_pfn, PAGE_KERNEL));
-                                               p2m_pfn++;
-                                       }
-                                       n_pte -= PTRS_PER_PTE;
-                                       early_memunmap(pt, PAGE_SIZE);
-                                       make_lowmem_page_readonly(__va(pt_phys));
-                                       pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
-                                                       PFN_DOWN(pt_phys));
-                                       set_pmd(pmd + idx_pt,
-                                                       __pmd(_PAGE_TABLE | pt_phys));
-                                       pt_phys += PAGE_SIZE;
-                               }
-                               n_pt -= PTRS_PER_PMD;
-                               early_memunmap(pmd, PAGE_SIZE);
-                               make_lowmem_page_readonly(__va(pmd_phys));
-                               pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
-                                               PFN_DOWN(pmd_phys));
-                               set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
-                               pmd_phys += PAGE_SIZE;
-                       }
-                       n_pmd -= PTRS_PER_PUD;
-                       early_memunmap(pud, PAGE_SIZE);
-                       make_lowmem_page_readonly(__va(pud_phys));
-                       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
-                       if (n_p4d > 0)
-                               set_p4d(p4d + idx_pud, __p4d(_PAGE_TABLE | pud_phys));
-                       else
-                               set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
-                       pud_phys += PAGE_SIZE;
-               }
-               if (n_p4d > 0) {
-                       save_pud -= PTRS_PER_P4D;
-                       early_memunmap(p4d, PAGE_SIZE);
-                       make_lowmem_page_readonly(__va(p4d_phys));
-                       pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, PFN_DOWN(p4d_phys));
-                       set_pgd(pgd + 2 + idx_p4d, __pgd(_PAGE_TABLE | p4d_phys));
-                       p4d_phys += PAGE_SIZE;
-               }
-       } while (++idx_p4d < n_p4d);
-
-       /* Now copy the old p2m info to the new area. */
-       memcpy(new_p2m, xen_p2m_addr, size);
-       xen_p2m_addr = new_p2m;
-
-       /* Release the old p2m list and set new list info. */
-       p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
-       BUG_ON(!p2m_pfn);
-       p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
-
-       if (xen_start_info->mfn_list < __START_KERNEL_map) {
-               pfn = xen_start_info->first_p2m_pfn;
-               pfn_end = xen_start_info->first_p2m_pfn +
-                         xen_start_info->nr_p2m_frames;
-               set_pgd(pgd + 1, __pgd(0));
-       } else {
-               pfn = p2m_pfn;
-               pfn_end = p2m_pfn_end;
-       }
-
-       memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
-       while (pfn < pfn_end) {
-               if (pfn == p2m_pfn) {
-                       pfn = p2m_pfn_end;
-                       continue;
-               }
-               make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
-               pfn++;
-       }
+       xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
 
-       xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
-       xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
-       xen_start_info->nr_p2m_frames = n_frames;
+       return PFN_DOWN(maddr.maddr);
 }
 
-#else  /* !CONFIG_X86_64 */
-static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
-static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
-
-static void __init xen_write_cr3_init(unsigned long cr3)
+xmaddr_t arbitrary_virt_to_machine(void *vaddr)
 {
-       unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
-
-       BUG_ON(read_cr3() != __pa(initial_page_table));
-       BUG_ON(cr3 != __pa(swapper_pg_dir));
+       unsigned long address = (unsigned long)vaddr;
+       unsigned int level;
+       pte_t *pte;
+       unsigned offset;
 
        /*
-        * We are switching to swapper_pg_dir for the first time (from
-        * initial_page_table) and therefore need to mark that page
-        * read-only and then pin it.
-        *
-        * Xen disallows sharing of kernel PMDs for PAE
-        * guests. Therefore we must copy the kernel PMD from
-        * initial_page_table into a new kernel PMD to be used in
-        * swapper_pg_dir.
+        * if the PFN is in the linear mapped vaddr range, we can just use
+        * the (quick) virt_to_machine() p2m lookup
         */
-       swapper_kernel_pmd =
-               extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
-       copy_page(swapper_kernel_pmd, initial_kernel_pmd);
-       swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
-               __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
-       set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
-
-       set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
-       xen_write_cr3(cr3);
-       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
-
-       pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
-                         PFN_DOWN(__pa(initial_page_table)));
-       set_page_prot(initial_page_table, PAGE_KERNEL);
-       set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
-
-       pv_mmu_ops.write_cr3 = &xen_write_cr3;
-}
-
-/*
- * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
- * not the first page table in the page table pool.
- * Iterate through the initial page tables to find the real page table base.
- */
-static phys_addr_t xen_find_pt_base(pmd_t *pmd)
-{
-       phys_addr_t pt_base, paddr;
-       unsigned pmdidx;
-
-       pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
-
-       for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
-               if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
-                       paddr = m2p(pmd[pmdidx].pmd);
-                       pt_base = min(pt_base, paddr);
-               }
-
-       return pt_base;
-}
-
-void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
-{
-       pmd_t *kernel_pmd;
-
-       kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
-
-       xen_pt_base = xen_find_pt_base(kernel_pmd);
-       xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
-
-       initial_kernel_pmd =
-               extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
-
-       max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
-
-       copy_page(initial_kernel_pmd, kernel_pmd);
-
-       xen_map_identity_early(initial_kernel_pmd, max_pfn);
-
-       copy_page(initial_page_table, pgd);
-       initial_page_table[KERNEL_PGD_BOUNDARY] =
-               __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
-
-       set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
-       set_page_prot(initial_page_table, PAGE_KERNEL_RO);
-       set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
-
-       pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
-
-       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
-                         PFN_DOWN(__pa(initial_page_table)));
-       xen_write_cr3(__pa(initial_page_table));
-
-       memblock_reserve(xen_pt_base, xen_pt_size);
-}
-#endif /* CONFIG_X86_64 */
-
-void __init xen_reserve_special_pages(void)
-{
-       phys_addr_t paddr;
-
-       memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
-       if (xen_start_info->store_mfn) {
-               paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
-               memblock_reserve(paddr, PAGE_SIZE);
-       }
-       if (!xen_initial_domain()) {
-               paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
-               memblock_reserve(paddr, PAGE_SIZE);
-       }
-}
-
-void __init xen_pt_check_e820(void)
-{
-       if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
-               xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
-               BUG();
-       }
-}
-
-static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
-
-static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
-{
-       pte_t pte;
-
-       phys >>= PAGE_SHIFT;
-
-       switch (idx) {
-       case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
-       case FIX_RO_IDT:
-#ifdef CONFIG_X86_32
-       case FIX_WP_TEST:
-# ifdef CONFIG_HIGHMEM
-       case FIX_KMAP_BEGIN ... FIX_KMAP_END:
-# endif
-#elif defined(CONFIG_X86_VSYSCALL_EMULATION)
-       case VSYSCALL_PAGE:
-#endif
-       case FIX_TEXT_POKE0:
-       case FIX_TEXT_POKE1:
-       case FIX_GDT_REMAP_BEGIN ... FIX_GDT_REMAP_END:
-               /* All local page mappings */
-               pte = pfn_pte(phys, prot);
-               break;
-
-#ifdef CONFIG_X86_LOCAL_APIC
-       case FIX_APIC_BASE:     /* maps dummy local APIC */
-               pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
-               break;
-#endif
-
-#ifdef CONFIG_X86_IO_APIC
-       case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
-               /*
-                * We just don't map the IO APIC - all access is via
-                * hypercalls.  Keep the address in the pte for reference.
-                */
-               pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
-               break;
-#endif
-
-       case FIX_PARAVIRT_BOOTMAP:
-               /* This is an MFN, but it isn't an IO mapping from the
-                  IO domain */
-               pte = mfn_pte(phys, prot);
-               break;
-
-       default:
-               /* By default, set_fixmap is used for hardware mappings */
-               pte = mfn_pte(phys, prot);
-               break;
-       }
-
-       __native_set_fixmap(idx, pte);
-
-#ifdef CONFIG_X86_VSYSCALL_EMULATION
-       /* Replicate changes to map the vsyscall page into the user
-          pagetable vsyscall mapping. */
-       if (idx == VSYSCALL_PAGE) {
-               unsigned long vaddr = __fix_to_virt(idx);
-               set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
-       }
-#endif
-}
-
-static void __init xen_post_allocator_init(void)
-{
-       if (xen_feature(XENFEAT_auto_translated_physmap))
-               return;
-
-       pv_mmu_ops.set_pte = xen_set_pte;
-       pv_mmu_ops.set_pmd = xen_set_pmd;
-       pv_mmu_ops.set_pud = xen_set_pud;
-#if CONFIG_PGTABLE_LEVELS >= 4
-       pv_mmu_ops.set_p4d = xen_set_p4d;
-#endif
-
-       /* This will work as long as patching hasn't happened yet
-          (which it hasn't) */
-       pv_mmu_ops.alloc_pte = xen_alloc_pte;
-       pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
-       pv_mmu_ops.release_pte = xen_release_pte;
-       pv_mmu_ops.release_pmd = xen_release_pmd;
-#if CONFIG_PGTABLE_LEVELS >= 4
-       pv_mmu_ops.alloc_pud = xen_alloc_pud;
-       pv_mmu_ops.release_pud = xen_release_pud;
-#endif
-       pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
-
-#ifdef CONFIG_X86_64
-       pv_mmu_ops.write_cr3 = &xen_write_cr3;
-       SetPagePinned(virt_to_page(level3_user_vsyscall));
-#endif
-       xen_mark_init_mm_pinned();
-}
-
-static void xen_leave_lazy_mmu(void)
-{
-       preempt_disable();
-       xen_mc_flush();
-       paravirt_leave_lazy_mmu();
-       preempt_enable();
-}
-
-static const struct pv_mmu_ops xen_mmu_ops __initconst = {
-       .read_cr2 = xen_read_cr2,
-       .write_cr2 = xen_write_cr2,
-
-       .read_cr3 = xen_read_cr3,
-       .write_cr3 = xen_write_cr3_init,
-
-       .flush_tlb_user = xen_flush_tlb,
-       .flush_tlb_kernel = xen_flush_tlb,
-       .flush_tlb_single = xen_flush_tlb_single,
-       .flush_tlb_others = xen_flush_tlb_others,
-
-       .pte_update = paravirt_nop,
-
-       .pgd_alloc = xen_pgd_alloc,
-       .pgd_free = xen_pgd_free,
-
-       .alloc_pte = xen_alloc_pte_init,
-       .release_pte = xen_release_pte_init,
-       .alloc_pmd = xen_alloc_pmd_init,
-       .release_pmd = xen_release_pmd_init,
-
-       .set_pte = xen_set_pte_init,
-       .set_pte_at = xen_set_pte_at,
-       .set_pmd = xen_set_pmd_hyper,
-
-       .ptep_modify_prot_start = __ptep_modify_prot_start,
-       .ptep_modify_prot_commit = __ptep_modify_prot_commit,
-
-       .pte_val = PV_CALLEE_SAVE(xen_pte_val),
-       .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
-
-       .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
-       .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
-
-#ifdef CONFIG_X86_PAE
-       .set_pte_atomic = xen_set_pte_atomic,
-       .pte_clear = xen_pte_clear,
-       .pmd_clear = xen_pmd_clear,
-#endif /* CONFIG_X86_PAE */
-       .set_pud = xen_set_pud_hyper,
-
-       .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
-       .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
-
-#if CONFIG_PGTABLE_LEVELS >= 4
-       .pud_val = PV_CALLEE_SAVE(xen_pud_val),
-       .make_pud = PV_CALLEE_SAVE(xen_make_pud),
-       .set_p4d = xen_set_p4d_hyper,
-
-       .alloc_pud = xen_alloc_pmd_init,
-       .release_pud = xen_release_pmd_init,
-#endif /* CONFIG_PGTABLE_LEVELS == 4 */
-
-       .activate_mm = xen_activate_mm,
-       .dup_mmap = xen_dup_mmap,
-       .exit_mmap = xen_exit_mmap,
-
-       .lazy_mode = {
-               .enter = paravirt_enter_lazy_mmu,
-               .leave = xen_leave_lazy_mmu,
-               .flush = paravirt_flush_lazy_mmu,
-       },
-
-       .set_fixmap = xen_set_fixmap,
-};
-
-void __init xen_init_mmu_ops(void)
-{
-       x86_init.paging.pagetable_init = xen_pagetable_init;
-
-       if (xen_feature(XENFEAT_auto_translated_physmap))
-               return;
+       if (virt_addr_valid(vaddr))
+               return virt_to_machine(vaddr);
 
-       pv_mmu_ops = xen_mmu_ops;
+       /* otherwise we have to do a (slower) full page-table walk */
 
-       memset(dummy_mapping, 0xff, PAGE_SIZE);
+       pte = lookup_address(address, &level);
+       BUG_ON(pte == NULL);
+       offset = address & ~PAGE_MASK;
+       return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
 }
+EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
 
-/* Protected by xen_reservation_lock. */
-#define MAX_CONTIG_ORDER 9 /* 2MB */
-static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
-
-#define VOID_PTE (mfn_pte(0, __pgprot(0)))
-static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
-                               unsigned long *in_frames,
-                               unsigned long *out_frames)
+void xen_flush_tlb_all(void)
 {
-       int i;
+       struct mmuext_op *op;
        struct multicall_space mcs;
 
-       xen_mc_batch();
-       for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
-               mcs = __xen_mc_entry(0);
-
-               if (in_frames)
-                       in_frames[i] = virt_to_mfn(vaddr);
-
-               MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
-               __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
-
-               if (out_frames)
-                       out_frames[i] = virt_to_pfn(vaddr);
-       }
-       xen_mc_issue(0);
-}
-
-/*
- * Update the pfn-to-mfn mappings for a virtual address range, either to
- * point to an array of mfns, or contiguously from a single starting
- * mfn.
- */
-static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
-                                    unsigned long *mfns,
-                                    unsigned long first_mfn)
-{
-       unsigned i, limit;
-       unsigned long mfn;
-
-       xen_mc_batch();
-
-       limit = 1u << order;
-       for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
-               struct multicall_space mcs;
-               unsigned flags;
-
-               mcs = __xen_mc_entry(0);
-               if (mfns)
-                       mfn = mfns[i];
-               else
-                       mfn = first_mfn + i;
-
-               if (i < (limit - 1))
-                       flags = 0;
-               else {
-                       if (order == 0)
-                               flags = UVMF_INVLPG | UVMF_ALL;
-                       else
-                               flags = UVMF_TLB_FLUSH | UVMF_ALL;
-               }
-
-               MULTI_update_va_mapping(mcs.mc, vaddr,
-                               mfn_pte(mfn, PAGE_KERNEL), flags);
-
-               set_phys_to_machine(virt_to_pfn(vaddr), mfn);
-       }
-
-       xen_mc_issue(0);
-}
-
-/*
- * Perform the hypercall to exchange a region of our pfns to point to
- * memory with the required contiguous alignment.  Takes the pfns as
- * input, and populates mfns as output.
- *
- * Returns a success code indicating whether the hypervisor was able to
- * satisfy the request or not.
- */
-static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
-                              unsigned long *pfns_in,
-                              unsigned long extents_out,
-                              unsigned int order_out,
-                              unsigned long *mfns_out,
-                              unsigned int address_bits)
-{
-       long rc;
-       int success;
-
-       struct xen_memory_exchange exchange = {
-               .in = {
-                       .nr_extents   = extents_in,
-                       .extent_order = order_in,
-                       .extent_start = pfns_in,
-                       .domid        = DOMID_SELF
-               },
-               .out = {
-                       .nr_extents   = extents_out,
-                       .extent_order = order_out,
-                       .extent_start = mfns_out,
-                       .address_bits = address_bits,
-                       .domid        = DOMID_SELF
-               }
-       };
-
-       BUG_ON(extents_in << order_in != extents_out << order_out);
-
-       rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
-       success = (exchange.nr_exchanged == extents_in);
-
-       BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
-       BUG_ON(success && (rc != 0));
-
-       return success;
-}
-
-int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
-                                unsigned int address_bits,
-                                dma_addr_t *dma_handle)
-{
-       unsigned long *in_frames = discontig_frames, out_frame;
-       unsigned long  flags;
-       int            success;
-       unsigned long vstart = (unsigned long)phys_to_virt(pstart);
-
-       /*
-        * Currently an auto-translated guest will not perform I/O, nor will
-        * it require PAE page directories below 4GB. Therefore any calls to
-        * this function are redundant and can be ignored.
-        */
-
-       if (xen_feature(XENFEAT_auto_translated_physmap))
-               return 0;
-
-       if (unlikely(order > MAX_CONTIG_ORDER))
-               return -ENOMEM;
-
-       memset((void *) vstart, 0, PAGE_SIZE << order);
-
-       spin_lock_irqsave(&xen_reservation_lock, flags);
-
-       /* 1. Zap current PTEs, remembering MFNs. */
-       xen_zap_pfn_range(vstart, order, in_frames, NULL);
-
-       /* 2. Get a new contiguous memory extent. */
-       out_frame = virt_to_pfn(vstart);
-       success = xen_exchange_memory(1UL << order, 0, in_frames,
-                                     1, order, &out_frame,
-                                     address_bits);
-
-       /* 3. Map the new extent in place of old pages. */
-       if (success)
-               xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
-       else
-               xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
-
-       spin_unlock_irqrestore(&xen_reservation_lock, flags);
-
-       *dma_handle = virt_to_machine(vstart).maddr;
-       return success ? 0 : -ENOMEM;
-}
-EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
-
-void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
-{
-       unsigned long *out_frames = discontig_frames, in_frame;
-       unsigned long  flags;
-       int success;
-       unsigned long vstart;
-
-       if (xen_feature(XENFEAT_auto_translated_physmap))
-               return;
-
-       if (unlikely(order > MAX_CONTIG_ORDER))
-               return;
-
-       vstart = (unsigned long)phys_to_virt(pstart);
-       memset((void *) vstart, 0, PAGE_SIZE << order);
-
-       spin_lock_irqsave(&xen_reservation_lock, flags);
+       trace_xen_mmu_flush_tlb_all(0);
 
-       /* 1. Find start MFN of contiguous extent. */
-       in_frame = virt_to_mfn(vstart);
+       preempt_disable();
 
-       /* 2. Zap current PTEs. */
-       xen_zap_pfn_range(vstart, order, NULL, out_frames);
+       mcs = xen_mc_entry(sizeof(*op));
 
-       /* 3. Do the exchange for non-contiguous MFNs. */
-       success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
-                                       0, out_frames, 0);
+       op = mcs.args;
+       op->cmd = MMUEXT_TLB_FLUSH_ALL;
+       MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 
-       /* 4. Map new pages in place of old pages. */
-       if (success)
-               xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
-       else
-               xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
 
-       spin_unlock_irqrestore(&xen_reservation_lock, flags);
+       preempt_enable();
 }
-EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
 
 #define REMAP_BATCH_SIZE 16
 
@@ -2900,7 +191,6 @@ int xen_remap_domain_gfn_array(struct vm_area_struct *vma,
 }
 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array);
 
-
 /* Returns: 0 success */
 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma,
                               int numpgs, struct page **pages)
diff --git a/arch/x86/xen/mmu_pv.c b/arch/x86/xen/mmu_pv.c
new file mode 100644 (file)
index 0000000..c98bfb9
--- /dev/null
@@ -0,0 +1,2717 @@
+/*
+ * Xen mmu operations
+ *
+ * This file contains the various mmu fetch and update operations.
+ * The most important job they must perform is the mapping between the
+ * domain's pfn and the overall machine mfns.
+ *
+ * Xen allows guests to directly update the pagetable, in a controlled
+ * fashion.  In other words, the guest modifies the same pagetable
+ * that the CPU actually uses, which eliminates the overhead of having
+ * a separate shadow pagetable.
+ *
+ * In order to allow this, it falls on the guest domain to map its
+ * notion of a "physical" pfn - which is just a domain-local linear
+ * address - into a real "machine address" which the CPU's MMU can
+ * use.
+ *
+ * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
+ * inserted directly into the pagetable.  When creating a new
+ * pte/pmd/pgd, it converts the passed pfn into an mfn.  Conversely,
+ * when reading the content back with __(pgd|pmd|pte)_val, it converts
+ * the mfn back into a pfn.
+ *
+ * The other constraint is that all pages which make up a pagetable
+ * must be mapped read-only in the guest.  This prevents uncontrolled
+ * guest updates to the pagetable.  Xen strictly enforces this, and
+ * will disallow any pagetable update which will end up mapping a
+ * pagetable page RW, and will disallow using any writable page as a
+ * pagetable.
+ *
+ * Naively, when loading %cr3 with the base of a new pagetable, Xen
+ * would need to validate the whole pagetable before going on.
+ * Naturally, this is quite slow.  The solution is to "pin" a
+ * pagetable, which enforces all the constraints on the pagetable even
+ * when it is not actively in use.  This menas that Xen can be assured
+ * that it is still valid when you do load it into %cr3, and doesn't
+ * need to revalidate it.
+ *
+ * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
+ */
+#include <linux/sched/mm.h>
+#include <linux/highmem.h>
+#include <linux/debugfs.h>
+#include <linux/bug.h>
+#include <linux/vmalloc.h>
+#include <linux/export.h>
+#include <linux/init.h>
+#include <linux/gfp.h>
+#include <linux/memblock.h>
+#include <linux/seq_file.h>
+#include <linux/crash_dump.h>
+
+#include <trace/events/xen.h>
+
+#include <asm/pgtable.h>
+#include <asm/tlbflush.h>
+#include <asm/fixmap.h>
+#include <asm/mmu_context.h>
+#include <asm/setup.h>
+#include <asm/paravirt.h>
+#include <asm/e820/api.h>
+#include <asm/linkage.h>
+#include <asm/page.h>
+#include <asm/init.h>
+#include <asm/pat.h>
+#include <asm/smp.h>
+
+#include <asm/xen/hypercall.h>
+#include <asm/xen/hypervisor.h>
+
+#include <xen/xen.h>
+#include <xen/page.h>
+#include <xen/interface/xen.h>
+#include <xen/interface/hvm/hvm_op.h>
+#include <xen/interface/version.h>
+#include <xen/interface/memory.h>
+#include <xen/hvc-console.h>
+
+#include "multicalls.h"
+#include "mmu.h"
+#include "debugfs.h"
+
+#ifdef CONFIG_X86_32
+/*
+ * Identity map, in addition to plain kernel map.  This needs to be
+ * large enough to allocate page table pages to allocate the rest.
+ * Each page can map 2MB.
+ */
+#define LEVEL1_IDENT_ENTRIES   (PTRS_PER_PTE * 4)
+static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
+#endif
+#ifdef CONFIG_X86_64
+/* l3 pud for userspace vsyscall mapping */
+static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
+#endif /* CONFIG_X86_64 */
+
+/*
+ * Note about cr3 (pagetable base) values:
+ *
+ * xen_cr3 contains the current logical cr3 value; it contains the
+ * last set cr3.  This may not be the current effective cr3, because
+ * its update may be being lazily deferred.  However, a vcpu looking
+ * at its own cr3 can use this value knowing that it everything will
+ * be self-consistent.
+ *
+ * xen_current_cr3 contains the actual vcpu cr3; it is set once the
+ * hypercall to set the vcpu cr3 is complete (so it may be a little
+ * out of date, but it will never be set early).  If one vcpu is
+ * looking at another vcpu's cr3 value, it should use this variable.
+ */
+DEFINE_PER_CPU(unsigned long, xen_cr3);         /* cr3 stored as physaddr */
+DEFINE_PER_CPU(unsigned long, xen_current_cr3);         /* actual vcpu cr3 */
+
+static phys_addr_t xen_pt_base, xen_pt_size __initdata;
+
+/*
+ * Just beyond the highest usermode address.  STACK_TOP_MAX has a
+ * redzone above it, so round it up to a PGD boundary.
+ */
+#define USER_LIMIT     ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
+
+void make_lowmem_page_readonly(void *vaddr)
+{
+       pte_t *pte, ptev;
+       unsigned long address = (unsigned long)vaddr;
+       unsigned int level;
+
+       pte = lookup_address(address, &level);
+       if (pte == NULL)
+               return;         /* vaddr missing */
+
+       ptev = pte_wrprotect(*pte);
+
+       if (HYPERVISOR_update_va_mapping(address, ptev, 0))
+               BUG();
+}
+
+void make_lowmem_page_readwrite(void *vaddr)
+{
+       pte_t *pte, ptev;
+       unsigned long address = (unsigned long)vaddr;
+       unsigned int level;
+
+       pte = lookup_address(address, &level);
+       if (pte == NULL)
+               return;         /* vaddr missing */
+
+       ptev = pte_mkwrite(*pte);
+
+       if (HYPERVISOR_update_va_mapping(address, ptev, 0))
+               BUG();
+}
+
+
+static bool xen_page_pinned(void *ptr)
+{
+       struct page *page = virt_to_page(ptr);
+
+       return PagePinned(page);
+}
+
+void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
+{
+       struct multicall_space mcs;
+       struct mmu_update *u;
+
+       trace_xen_mmu_set_domain_pte(ptep, pteval, domid);
+
+       mcs = xen_mc_entry(sizeof(*u));
+       u = mcs.args;
+
+       /* ptep might be kmapped when using 32-bit HIGHPTE */
+       u->ptr = virt_to_machine(ptep).maddr;
+       u->val = pte_val_ma(pteval);
+
+       MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+}
+EXPORT_SYMBOL_GPL(xen_set_domain_pte);
+
+static void xen_extend_mmu_update(const struct mmu_update *update)
+{
+       struct multicall_space mcs;
+       struct mmu_update *u;
+
+       mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
+
+       if (mcs.mc != NULL) {
+               mcs.mc->args[1]++;
+       } else {
+               mcs = __xen_mc_entry(sizeof(*u));
+               MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
+       }
+
+       u = mcs.args;
+       *u = *update;
+}
+
+static void xen_extend_mmuext_op(const struct mmuext_op *op)
+{
+       struct multicall_space mcs;
+       struct mmuext_op *u;
+
+       mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
+
+       if (mcs.mc != NULL) {
+               mcs.mc->args[1]++;
+       } else {
+               mcs = __xen_mc_entry(sizeof(*u));
+               MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
+       }
+
+       u = mcs.args;
+       *u = *op;
+}
+
+static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
+{
+       struct mmu_update u;
+
+       preempt_disable();
+
+       xen_mc_batch();
+
+       /* ptr may be ioremapped for 64-bit pagetable setup */
+       u.ptr = arbitrary_virt_to_machine(ptr).maddr;
+       u.val = pmd_val_ma(val);
+       xen_extend_mmu_update(&u);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
+}
+
+static void xen_set_pmd(pmd_t *ptr, pmd_t val)
+{
+       trace_xen_mmu_set_pmd(ptr, val);
+
+       /* If page is not pinned, we can just update the entry
+          directly */
+       if (!xen_page_pinned(ptr)) {
+               *ptr = val;
+               return;
+       }
+
+       xen_set_pmd_hyper(ptr, val);
+}
+
+/*
+ * Associate a virtual page frame with a given physical page frame
+ * and protection flags for that frame.
+ */
+void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
+{
+       set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
+}
+
+static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
+{
+       struct mmu_update u;
+
+       if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
+               return false;
+
+       xen_mc_batch();
+
+       u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
+       u.val = pte_val_ma(pteval);
+       xen_extend_mmu_update(&u);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       return true;
+}
+
+static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
+{
+       if (!xen_batched_set_pte(ptep, pteval)) {
+               /*
+                * Could call native_set_pte() here and trap and
+                * emulate the PTE write but with 32-bit guests this
+                * needs two traps (one for each of the two 32-bit
+                * words in the PTE) so do one hypercall directly
+                * instead.
+                */
+               struct mmu_update u;
+
+               u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
+               u.val = pte_val_ma(pteval);
+               HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
+       }
+}
+
+static void xen_set_pte(pte_t *ptep, pte_t pteval)
+{
+       trace_xen_mmu_set_pte(ptep, pteval);
+       __xen_set_pte(ptep, pteval);
+}
+
+static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
+                   pte_t *ptep, pte_t pteval)
+{
+       trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
+       __xen_set_pte(ptep, pteval);
+}
+
+pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
+                                unsigned long addr, pte_t *ptep)
+{
+       /* Just return the pte as-is.  We preserve the bits on commit */
+       trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
+       return *ptep;
+}
+
+void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
+                                pte_t *ptep, pte_t pte)
+{
+       struct mmu_update u;
+
+       trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
+       xen_mc_batch();
+
+       u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
+       u.val = pte_val_ma(pte);
+       xen_extend_mmu_update(&u);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+}
+
+/* Assume pteval_t is equivalent to all the other *val_t types. */
+static pteval_t pte_mfn_to_pfn(pteval_t val)
+{
+       if (val & _PAGE_PRESENT) {
+               unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
+               unsigned long pfn = mfn_to_pfn(mfn);
+
+               pteval_t flags = val & PTE_FLAGS_MASK;
+               if (unlikely(pfn == ~0))
+                       val = flags & ~_PAGE_PRESENT;
+               else
+                       val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
+       }
+
+       return val;
+}
+
+static pteval_t pte_pfn_to_mfn(pteval_t val)
+{
+       if (val & _PAGE_PRESENT) {
+               unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
+               pteval_t flags = val & PTE_FLAGS_MASK;
+               unsigned long mfn;
+
+               if (!xen_feature(XENFEAT_auto_translated_physmap))
+                       mfn = __pfn_to_mfn(pfn);
+               else
+                       mfn = pfn;
+               /*
+                * If there's no mfn for the pfn, then just create an
+                * empty non-present pte.  Unfortunately this loses
+                * information about the original pfn, so
+                * pte_mfn_to_pfn is asymmetric.
+                */
+               if (unlikely(mfn == INVALID_P2M_ENTRY)) {
+                       mfn = 0;
+                       flags = 0;
+               } else
+                       mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
+               val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
+       }
+
+       return val;
+}
+
+__visible pteval_t xen_pte_val(pte_t pte)
+{
+       pteval_t pteval = pte.pte;
+
+       return pte_mfn_to_pfn(pteval);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
+
+__visible pgdval_t xen_pgd_val(pgd_t pgd)
+{
+       return pte_mfn_to_pfn(pgd.pgd);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
+
+__visible pte_t xen_make_pte(pteval_t pte)
+{
+       pte = pte_pfn_to_mfn(pte);
+
+       return native_make_pte(pte);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
+
+__visible pgd_t xen_make_pgd(pgdval_t pgd)
+{
+       pgd = pte_pfn_to_mfn(pgd);
+       return native_make_pgd(pgd);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
+
+__visible pmdval_t xen_pmd_val(pmd_t pmd)
+{
+       return pte_mfn_to_pfn(pmd.pmd);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
+
+static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
+{
+       struct mmu_update u;
+
+       preempt_disable();
+
+       xen_mc_batch();
+
+       /* ptr may be ioremapped for 64-bit pagetable setup */
+       u.ptr = arbitrary_virt_to_machine(ptr).maddr;
+       u.val = pud_val_ma(val);
+       xen_extend_mmu_update(&u);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
+}
+
+static void xen_set_pud(pud_t *ptr, pud_t val)
+{
+       trace_xen_mmu_set_pud(ptr, val);
+
+       /* If page is not pinned, we can just update the entry
+          directly */
+       if (!xen_page_pinned(ptr)) {
+               *ptr = val;
+               return;
+       }
+
+       xen_set_pud_hyper(ptr, val);
+}
+
+#ifdef CONFIG_X86_PAE
+static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
+{
+       trace_xen_mmu_set_pte_atomic(ptep, pte);
+       set_64bit((u64 *)ptep, native_pte_val(pte));
+}
+
+static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
+{
+       trace_xen_mmu_pte_clear(mm, addr, ptep);
+       if (!xen_batched_set_pte(ptep, native_make_pte(0)))
+               native_pte_clear(mm, addr, ptep);
+}
+
+static void xen_pmd_clear(pmd_t *pmdp)
+{
+       trace_xen_mmu_pmd_clear(pmdp);
+       set_pmd(pmdp, __pmd(0));
+}
+#endif /* CONFIG_X86_PAE */
+
+__visible pmd_t xen_make_pmd(pmdval_t pmd)
+{
+       pmd = pte_pfn_to_mfn(pmd);
+       return native_make_pmd(pmd);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
+
+#if CONFIG_PGTABLE_LEVELS == 4
+__visible pudval_t xen_pud_val(pud_t pud)
+{
+       return pte_mfn_to_pfn(pud.pud);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
+
+__visible pud_t xen_make_pud(pudval_t pud)
+{
+       pud = pte_pfn_to_mfn(pud);
+
+       return native_make_pud(pud);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
+
+static pgd_t *xen_get_user_pgd(pgd_t *pgd)
+{
+       pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
+       unsigned offset = pgd - pgd_page;
+       pgd_t *user_ptr = NULL;
+
+       if (offset < pgd_index(USER_LIMIT)) {
+               struct page *page = virt_to_page(pgd_page);
+               user_ptr = (pgd_t *)page->private;
+               if (user_ptr)
+                       user_ptr += offset;
+       }
+
+       return user_ptr;
+}
+
+static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
+{
+       struct mmu_update u;
+
+       u.ptr = virt_to_machine(ptr).maddr;
+       u.val = p4d_val_ma(val);
+       xen_extend_mmu_update(&u);
+}
+
+/*
+ * Raw hypercall-based set_p4d, intended for in early boot before
+ * there's a page structure.  This implies:
+ *  1. The only existing pagetable is the kernel's
+ *  2. It is always pinned
+ *  3. It has no user pagetable attached to it
+ */
+static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
+{
+       preempt_disable();
+
+       xen_mc_batch();
+
+       __xen_set_p4d_hyper(ptr, val);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
+}
+
+static void xen_set_p4d(p4d_t *ptr, p4d_t val)
+{
+       pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
+       pgd_t pgd_val;
+
+       trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
+
+       /* If page is not pinned, we can just update the entry
+          directly */
+       if (!xen_page_pinned(ptr)) {
+               *ptr = val;
+               if (user_ptr) {
+                       WARN_ON(xen_page_pinned(user_ptr));
+                       pgd_val.pgd = p4d_val_ma(val);
+                       *user_ptr = pgd_val;
+               }
+               return;
+       }
+
+       /* If it's pinned, then we can at least batch the kernel and
+          user updates together. */
+       xen_mc_batch();
+
+       __xen_set_p4d_hyper(ptr, val);
+       if (user_ptr)
+               __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+}
+#endif /* CONFIG_PGTABLE_LEVELS == 4 */
+
+static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
+               int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
+               bool last, unsigned long limit)
+{
+       int i, nr, flush = 0;
+
+       nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
+       for (i = 0; i < nr; i++) {
+               if (!pmd_none(pmd[i]))
+                       flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE);
+       }
+       return flush;
+}
+
+static int xen_pud_walk(struct mm_struct *mm, pud_t *pud,
+               int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
+               bool last, unsigned long limit)
+{
+       int i, nr, flush = 0;
+
+       nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
+       for (i = 0; i < nr; i++) {
+               pmd_t *pmd;
+
+               if (pud_none(pud[i]))
+                       continue;
+
+               pmd = pmd_offset(&pud[i], 0);
+               if (PTRS_PER_PMD > 1)
+                       flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
+               flush |= xen_pmd_walk(mm, pmd, func,
+                               last && i == nr - 1, limit);
+       }
+       return flush;
+}
+
+static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
+               int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
+               bool last, unsigned long limit)
+{
+       int i, nr, flush = 0;
+
+       nr = last ? p4d_index(limit) + 1 : PTRS_PER_P4D;
+       for (i = 0; i < nr; i++) {
+               pud_t *pud;
+
+               if (p4d_none(p4d[i]))
+                       continue;
+
+               pud = pud_offset(&p4d[i], 0);
+               if (PTRS_PER_PUD > 1)
+                       flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
+               flush |= xen_pud_walk(mm, pud, func,
+                               last && i == nr - 1, limit);
+       }
+       return flush;
+}
+
+/*
+ * (Yet another) pagetable walker.  This one is intended for pinning a
+ * pagetable.  This means that it walks a pagetable and calls the
+ * callback function on each page it finds making up the page table,
+ * at every level.  It walks the entire pagetable, but it only bothers
+ * pinning pte pages which are below limit.  In the normal case this
+ * will be STACK_TOP_MAX, but at boot we need to pin up to
+ * FIXADDR_TOP.
+ *
+ * For 32-bit the important bit is that we don't pin beyond there,
+ * because then we start getting into Xen's ptes.
+ *
+ * For 64-bit, we must skip the Xen hole in the middle of the address
+ * space, just after the big x86-64 virtual hole.
+ */
+static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
+                         int (*func)(struct mm_struct *mm, struct page *,
+                                     enum pt_level),
+                         unsigned long limit)
+{
+       int i, nr, flush = 0;
+       unsigned hole_low, hole_high;
+
+       /* The limit is the last byte to be touched */
+       limit--;
+       BUG_ON(limit >= FIXADDR_TOP);
+
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return 0;
+
+       /*
+        * 64-bit has a great big hole in the middle of the address
+        * space, which contains the Xen mappings.  On 32-bit these
+        * will end up making a zero-sized hole and so is a no-op.
+        */
+       hole_low = pgd_index(USER_LIMIT);
+       hole_high = pgd_index(PAGE_OFFSET);
+
+       nr = pgd_index(limit) + 1;
+       for (i = 0; i < nr; i++) {
+               p4d_t *p4d;
+
+               if (i >= hole_low && i < hole_high)
+                       continue;
+
+               if (pgd_none(pgd[i]))
+                       continue;
+
+               p4d = p4d_offset(&pgd[i], 0);
+               if (PTRS_PER_P4D > 1)
+                       flush |= (*func)(mm, virt_to_page(p4d), PT_P4D);
+               flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
+       }
+
+       /* Do the top level last, so that the callbacks can use it as
+          a cue to do final things like tlb flushes. */
+       flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
+
+       return flush;
+}
+
+static int xen_pgd_walk(struct mm_struct *mm,
+                       int (*func)(struct mm_struct *mm, struct page *,
+                                   enum pt_level),
+                       unsigned long limit)
+{
+       return __xen_pgd_walk(mm, mm->pgd, func, limit);
+}
+
+/* If we're using split pte locks, then take the page's lock and
+   return a pointer to it.  Otherwise return NULL. */
+static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
+{
+       spinlock_t *ptl = NULL;
+
+#if USE_SPLIT_PTE_PTLOCKS
+       ptl = ptlock_ptr(page);
+       spin_lock_nest_lock(ptl, &mm->page_table_lock);
+#endif
+
+       return ptl;
+}
+
+static void xen_pte_unlock(void *v)
+{
+       spinlock_t *ptl = v;
+       spin_unlock(ptl);
+}
+
+static void xen_do_pin(unsigned level, unsigned long pfn)
+{
+       struct mmuext_op op;
+
+       op.cmd = level;
+       op.arg1.mfn = pfn_to_mfn(pfn);
+
+       xen_extend_mmuext_op(&op);
+}
+
+static int xen_pin_page(struct mm_struct *mm, struct page *page,
+                       enum pt_level level)
+{
+       unsigned pgfl = TestSetPagePinned(page);
+       int flush;
+
+       if (pgfl)
+               flush = 0;              /* already pinned */
+       else if (PageHighMem(page))
+               /* kmaps need flushing if we found an unpinned
+                  highpage */
+               flush = 1;
+       else {
+               void *pt = lowmem_page_address(page);
+               unsigned long pfn = page_to_pfn(page);
+               struct multicall_space mcs = __xen_mc_entry(0);
+               spinlock_t *ptl;
+
+               flush = 0;
+
+               /*
+                * We need to hold the pagetable lock between the time
+                * we make the pagetable RO and when we actually pin
+                * it.  If we don't, then other users may come in and
+                * attempt to update the pagetable by writing it,
+                * which will fail because the memory is RO but not
+                * pinned, so Xen won't do the trap'n'emulate.
+                *
+                * If we're using split pte locks, we can't hold the
+                * entire pagetable's worth of locks during the
+                * traverse, because we may wrap the preempt count (8
+                * bits).  The solution is to mark RO and pin each PTE
+                * page while holding the lock.  This means the number
+                * of locks we end up holding is never more than a
+                * batch size (~32 entries, at present).
+                *
+                * If we're not using split pte locks, we needn't pin
+                * the PTE pages independently, because we're
+                * protected by the overall pagetable lock.
+                */
+               ptl = NULL;
+               if (level == PT_PTE)
+                       ptl = xen_pte_lock(page, mm);
+
+               MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
+                                       pfn_pte(pfn, PAGE_KERNEL_RO),
+                                       level == PT_PGD ? UVMF_TLB_FLUSH : 0);
+
+               if (ptl) {
+                       xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
+
+                       /* Queue a deferred unlock for when this batch
+                          is completed. */
+                       xen_mc_callback(xen_pte_unlock, ptl);
+               }
+       }
+
+       return flush;
+}
+
+/* This is called just after a mm has been created, but it has not
+   been used yet.  We need to make sure that its pagetable is all
+   read-only, and can be pinned. */
+static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
+{
+       trace_xen_mmu_pgd_pin(mm, pgd);
+
+       xen_mc_batch();
+
+       if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
+               /* re-enable interrupts for flushing */
+               xen_mc_issue(0);
+
+               kmap_flush_unused();
+
+               xen_mc_batch();
+       }
+
+#ifdef CONFIG_X86_64
+       {
+               pgd_t *user_pgd = xen_get_user_pgd(pgd);
+
+               xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
+
+               if (user_pgd) {
+                       xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
+                       xen_do_pin(MMUEXT_PIN_L4_TABLE,
+                                  PFN_DOWN(__pa(user_pgd)));
+               }
+       }
+#else /* CONFIG_X86_32 */
+#ifdef CONFIG_X86_PAE
+       /* Need to make sure unshared kernel PMD is pinnable */
+       xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
+                    PT_PMD);
+#endif
+       xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
+#endif /* CONFIG_X86_64 */
+       xen_mc_issue(0);
+}
+
+static void xen_pgd_pin(struct mm_struct *mm)
+{
+       __xen_pgd_pin(mm, mm->pgd);
+}
+
+/*
+ * On save, we need to pin all pagetables to make sure they get their
+ * mfns turned into pfns.  Search the list for any unpinned pgds and pin
+ * them (unpinned pgds are not currently in use, probably because the
+ * process is under construction or destruction).
+ *
+ * Expected to be called in stop_machine() ("equivalent to taking
+ * every spinlock in the system"), so the locking doesn't really
+ * matter all that much.
+ */
+void xen_mm_pin_all(void)
+{
+       struct page *page;
+
+       spin_lock(&pgd_lock);
+
+       list_for_each_entry(page, &pgd_list, lru) {
+               if (!PagePinned(page)) {
+                       __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
+                       SetPageSavePinned(page);
+               }
+       }
+
+       spin_unlock(&pgd_lock);
+}
+
+/*
+ * The init_mm pagetable is really pinned as soon as its created, but
+ * that's before we have page structures to store the bits.  So do all
+ * the book-keeping now.
+ */
+static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
+                                 enum pt_level level)
+{
+       SetPagePinned(page);
+       return 0;
+}
+
+static void __init xen_mark_init_mm_pinned(void)
+{
+       xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
+}
+
+static int xen_unpin_page(struct mm_struct *mm, struct page *page,
+                         enum pt_level level)
+{
+       unsigned pgfl = TestClearPagePinned(page);
+
+       if (pgfl && !PageHighMem(page)) {
+               void *pt = lowmem_page_address(page);
+               unsigned long pfn = page_to_pfn(page);
+               spinlock_t *ptl = NULL;
+               struct multicall_space mcs;
+
+               /*
+                * Do the converse to pin_page.  If we're using split
+                * pte locks, we must be holding the lock for while
+                * the pte page is unpinned but still RO to prevent
+                * concurrent updates from seeing it in this
+                * partially-pinned state.
+                */
+               if (level == PT_PTE) {
+                       ptl = xen_pte_lock(page, mm);
+
+                       if (ptl)
+                               xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
+               }
+
+               mcs = __xen_mc_entry(0);
+
+               MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
+                                       pfn_pte(pfn, PAGE_KERNEL),
+                                       level == PT_PGD ? UVMF_TLB_FLUSH : 0);
+
+               if (ptl) {
+                       /* unlock when batch completed */
+                       xen_mc_callback(xen_pte_unlock, ptl);
+               }
+       }
+
+       return 0;               /* never need to flush on unpin */
+}
+
+/* Release a pagetables pages back as normal RW */
+static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
+{
+       trace_xen_mmu_pgd_unpin(mm, pgd);
+
+       xen_mc_batch();
+
+       xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
+
+#ifdef CONFIG_X86_64
+       {
+               pgd_t *user_pgd = xen_get_user_pgd(pgd);
+
+               if (user_pgd) {
+                       xen_do_pin(MMUEXT_UNPIN_TABLE,
+                                  PFN_DOWN(__pa(user_pgd)));
+                       xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
+               }
+       }
+#endif
+
+#ifdef CONFIG_X86_PAE
+       /* Need to make sure unshared kernel PMD is unpinned */
+       xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
+                      PT_PMD);
+#endif
+
+       __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
+
+       xen_mc_issue(0);
+}
+
+static void xen_pgd_unpin(struct mm_struct *mm)
+{
+       __xen_pgd_unpin(mm, mm->pgd);
+}
+
+/*
+ * On resume, undo any pinning done at save, so that the rest of the
+ * kernel doesn't see any unexpected pinned pagetables.
+ */
+void xen_mm_unpin_all(void)
+{
+       struct page *page;
+
+       spin_lock(&pgd_lock);
+
+       list_for_each_entry(page, &pgd_list, lru) {
+               if (PageSavePinned(page)) {
+                       BUG_ON(!PagePinned(page));
+                       __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
+                       ClearPageSavePinned(page);
+               }
+       }
+
+       spin_unlock(&pgd_lock);
+}
+
+static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
+{
+       spin_lock(&next->page_table_lock);
+       xen_pgd_pin(next);
+       spin_unlock(&next->page_table_lock);
+}
+
+static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
+{
+       spin_lock(&mm->page_table_lock);
+       xen_pgd_pin(mm);
+       spin_unlock(&mm->page_table_lock);
+}
+
+
+#ifdef CONFIG_SMP
+/* Another cpu may still have their %cr3 pointing at the pagetable, so
+   we need to repoint it somewhere else before we can unpin it. */
+static void drop_other_mm_ref(void *info)
+{
+       struct mm_struct *mm = info;
+       struct mm_struct *active_mm;
+
+       active_mm = this_cpu_read(cpu_tlbstate.active_mm);
+
+       if (active_mm == mm && this_cpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
+               leave_mm(smp_processor_id());
+
+       /* If this cpu still has a stale cr3 reference, then make sure
+          it has been flushed. */
+       if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
+               load_cr3(swapper_pg_dir);
+}
+
+static void xen_drop_mm_ref(struct mm_struct *mm)
+{
+       cpumask_var_t mask;
+       unsigned cpu;
+
+       if (current->active_mm == mm) {
+               if (current->mm == mm)
+                       load_cr3(swapper_pg_dir);
+               else
+                       leave_mm(smp_processor_id());
+       }
+
+       /* Get the "official" set of cpus referring to our pagetable. */
+       if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
+               for_each_online_cpu(cpu) {
+                       if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
+                           && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
+                               continue;
+                       smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
+               }
+               return;
+       }
+       cpumask_copy(mask, mm_cpumask(mm));
+
+       /* It's possible that a vcpu may have a stale reference to our
+          cr3, because its in lazy mode, and it hasn't yet flushed
+          its set of pending hypercalls yet.  In this case, we can
+          look at its actual current cr3 value, and force it to flush
+          if needed. */
+       for_each_online_cpu(cpu) {
+               if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
+                       cpumask_set_cpu(cpu, mask);
+       }
+
+       if (!cpumask_empty(mask))
+               smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
+       free_cpumask_var(mask);
+}
+#else
+static void xen_drop_mm_ref(struct mm_struct *mm)
+{
+       if (current->active_mm == mm)
+               load_cr3(swapper_pg_dir);
+}
+#endif
+
+/*
+ * While a process runs, Xen pins its pagetables, which means that the
+ * hypervisor forces it to be read-only, and it controls all updates
+ * to it.  This means that all pagetable updates have to go via the
+ * hypervisor, which is moderately expensive.
+ *
+ * Since we're pulling the pagetable down, we switch to use init_mm,
+ * unpin old process pagetable and mark it all read-write, which
+ * allows further operations on it to be simple memory accesses.
+ *
+ * The only subtle point is that another CPU may be still using the
+ * pagetable because of lazy tlb flushing.  This means we need need to
+ * switch all CPUs off this pagetable before we can unpin it.
+ */
+static void xen_exit_mmap(struct mm_struct *mm)
+{
+       get_cpu();              /* make sure we don't move around */
+       xen_drop_mm_ref(mm);
+       put_cpu();
+
+       spin_lock(&mm->page_table_lock);
+
+       /* pgd may not be pinned in the error exit path of execve */
+       if (xen_page_pinned(mm->pgd))
+               xen_pgd_unpin(mm);
+
+       spin_unlock(&mm->page_table_lock);
+}
+
+static void xen_post_allocator_init(void);
+
+static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
+{
+       struct mmuext_op op;
+
+       op.cmd = cmd;
+       op.arg1.mfn = pfn_to_mfn(pfn);
+       if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
+               BUG();
+}
+
+#ifdef CONFIG_X86_64
+static void __init xen_cleanhighmap(unsigned long vaddr,
+                                   unsigned long vaddr_end)
+{
+       unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
+       pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
+
+       /* NOTE: The loop is more greedy than the cleanup_highmap variant.
+        * We include the PMD passed in on _both_ boundaries. */
+       for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
+                       pmd++, vaddr += PMD_SIZE) {
+               if (pmd_none(*pmd))
+                       continue;
+               if (vaddr < (unsigned long) _text || vaddr > kernel_end)
+                       set_pmd(pmd, __pmd(0));
+       }
+       /* In case we did something silly, we should crash in this function
+        * instead of somewhere later and be confusing. */
+       xen_mc_flush();
+}
+
+/*
+ * Make a page range writeable and free it.
+ */
+static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
+{
+       void *vaddr = __va(paddr);
+       void *vaddr_end = vaddr + size;
+
+       for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
+               make_lowmem_page_readwrite(vaddr);
+
+       memblock_free(paddr, size);
+}
+
+static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
+{
+       unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
+
+       if (unpin)
+               pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
+       ClearPagePinned(virt_to_page(__va(pa)));
+       xen_free_ro_pages(pa, PAGE_SIZE);
+}
+
+static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
+{
+       unsigned long pa;
+       pte_t *pte_tbl;
+       int i;
+
+       if (pmd_large(*pmd)) {
+               pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
+               xen_free_ro_pages(pa, PMD_SIZE);
+               return;
+       }
+
+       pte_tbl = pte_offset_kernel(pmd, 0);
+       for (i = 0; i < PTRS_PER_PTE; i++) {
+               if (pte_none(pte_tbl[i]))
+                       continue;
+               pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
+               xen_free_ro_pages(pa, PAGE_SIZE);
+       }
+       set_pmd(pmd, __pmd(0));
+       xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
+}
+
+static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
+{
+       unsigned long pa;
+       pmd_t *pmd_tbl;
+       int i;
+
+       if (pud_large(*pud)) {
+               pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
+               xen_free_ro_pages(pa, PUD_SIZE);
+               return;
+       }
+
+       pmd_tbl = pmd_offset(pud, 0);
+       for (i = 0; i < PTRS_PER_PMD; i++) {
+               if (pmd_none(pmd_tbl[i]))
+                       continue;
+               xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
+       }
+       set_pud(pud, __pud(0));
+       xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
+}
+
+static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
+{
+       unsigned long pa;
+       pud_t *pud_tbl;
+       int i;
+
+       if (p4d_large(*p4d)) {
+               pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
+               xen_free_ro_pages(pa, P4D_SIZE);
+               return;
+       }
+
+       pud_tbl = pud_offset(p4d, 0);
+       for (i = 0; i < PTRS_PER_PUD; i++) {
+               if (pud_none(pud_tbl[i]))
+                       continue;
+               xen_cleanmfnmap_pud(pud_tbl + i, unpin);
+       }
+       set_p4d(p4d, __p4d(0));
+       xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
+}
+
+/*
+ * Since it is well isolated we can (and since it is perhaps large we should)
+ * also free the page tables mapping the initial P->M table.
+ */
+static void __init xen_cleanmfnmap(unsigned long vaddr)
+{
+       pgd_t *pgd;
+       p4d_t *p4d;
+       unsigned int i;
+       bool unpin;
+
+       unpin = (vaddr == 2 * PGDIR_SIZE);
+       vaddr &= PMD_MASK;
+       pgd = pgd_offset_k(vaddr);
+       p4d = p4d_offset(pgd, 0);
+       for (i = 0; i < PTRS_PER_P4D; i++) {
+               if (p4d_none(p4d[i]))
+                       continue;
+               xen_cleanmfnmap_p4d(p4d + i, unpin);
+       }
+       if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
+               set_pgd(pgd, __pgd(0));
+               xen_cleanmfnmap_free_pgtbl(p4d, unpin);
+       }
+}
+
+static void __init xen_pagetable_p2m_free(void)
+{
+       unsigned long size;
+       unsigned long addr;
+
+       size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
+
+       /* No memory or already called. */
+       if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
+               return;
+
+       /* using __ka address and sticking INVALID_P2M_ENTRY! */
+       memset((void *)xen_start_info->mfn_list, 0xff, size);
+
+       addr = xen_start_info->mfn_list;
+       /*
+        * We could be in __ka space.
+        * We roundup to the PMD, which means that if anybody at this stage is
+        * using the __ka address of xen_start_info or
+        * xen_start_info->shared_info they are in going to crash. Fortunatly
+        * we have already revectored in xen_setup_kernel_pagetable and in
+        * xen_setup_shared_info.
+        */
+       size = roundup(size, PMD_SIZE);
+
+       if (addr >= __START_KERNEL_map) {
+               xen_cleanhighmap(addr, addr + size);
+               size = PAGE_ALIGN(xen_start_info->nr_pages *
+                                 sizeof(unsigned long));
+               memblock_free(__pa(addr), size);
+       } else {
+               xen_cleanmfnmap(addr);
+       }
+}
+
+static void __init xen_pagetable_cleanhighmap(void)
+{
+       unsigned long size;
+       unsigned long addr;
+
+       /* At this stage, cleanup_highmap has already cleaned __ka space
+        * from _brk_limit way up to the max_pfn_mapped (which is the end of
+        * the ramdisk). We continue on, erasing PMD entries that point to page
+        * tables - do note that they are accessible at this stage via __va.
+        * For good measure we also round up to the PMD - which means that if
+        * anybody is using __ka address to the initial boot-stack - and try
+        * to use it - they are going to crash. The xen_start_info has been
+        * taken care of already in xen_setup_kernel_pagetable. */
+       addr = xen_start_info->pt_base;
+       size = roundup(xen_start_info->nr_pt_frames * PAGE_SIZE, PMD_SIZE);
+
+       xen_cleanhighmap(addr, addr + size);
+       xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
+#ifdef DEBUG
+       /* This is superfluous and is not necessary, but you know what
+        * lets do it. The MODULES_VADDR -> MODULES_END should be clear of
+        * anything at this stage. */
+       xen_cleanhighmap(MODULES_VADDR, roundup(MODULES_VADDR, PUD_SIZE) - 1);
+#endif
+}
+#endif
+
+static void __init xen_pagetable_p2m_setup(void)
+{
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return;
+
+       xen_vmalloc_p2m_tree();
+
+#ifdef CONFIG_X86_64
+       xen_pagetable_p2m_free();
+
+       xen_pagetable_cleanhighmap();
+#endif
+       /* And revector! Bye bye old array */
+       xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
+}
+
+static void __init xen_pagetable_init(void)
+{
+       paging_init();
+       xen_post_allocator_init();
+
+       xen_pagetable_p2m_setup();
+
+       /* Allocate and initialize top and mid mfn levels for p2m structure */
+       xen_build_mfn_list_list();
+
+       /* Remap memory freed due to conflicts with E820 map */
+       if (!xen_feature(XENFEAT_auto_translated_physmap))
+               xen_remap_memory();
+
+       xen_setup_shared_info();
+}
+static void xen_write_cr2(unsigned long cr2)
+{
+       this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
+}
+
+static unsigned long xen_read_cr2(void)
+{
+       return this_cpu_read(xen_vcpu)->arch.cr2;
+}
+
+unsigned long xen_read_cr2_direct(void)
+{
+       return this_cpu_read(xen_vcpu_info.arch.cr2);
+}
+
+static void xen_flush_tlb(void)
+{
+       struct mmuext_op *op;
+       struct multicall_space mcs;
+
+       trace_xen_mmu_flush_tlb(0);
+
+       preempt_disable();
+
+       mcs = xen_mc_entry(sizeof(*op));
+
+       op = mcs.args;
+       op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
+       MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
+}
+
+static void xen_flush_tlb_single(unsigned long addr)
+{
+       struct mmuext_op *op;
+       struct multicall_space mcs;
+
+       trace_xen_mmu_flush_tlb_single(addr);
+
+       preempt_disable();
+
+       mcs = xen_mc_entry(sizeof(*op));
+       op = mcs.args;
+       op->cmd = MMUEXT_INVLPG_LOCAL;
+       op->arg1.linear_addr = addr & PAGE_MASK;
+       MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+
+       preempt_enable();
+}
+
+static void xen_flush_tlb_others(const struct cpumask *cpus,
+                                struct mm_struct *mm, unsigned long start,
+                                unsigned long end)
+{
+       struct {
+               struct mmuext_op op;
+#ifdef CONFIG_SMP
+               DECLARE_BITMAP(mask, num_processors);
+#else
+               DECLARE_BITMAP(mask, NR_CPUS);
+#endif
+       } *args;
+       struct multicall_space mcs;
+
+       trace_xen_mmu_flush_tlb_others(cpus, mm, start, end);
+
+       if (cpumask_empty(cpus))
+               return;         /* nothing to do */
+
+       mcs = xen_mc_entry(sizeof(*args));
+       args = mcs.args;
+       args->op.arg2.vcpumask = to_cpumask(args->mask);
+
+       /* Remove us, and any offline CPUS. */
+       cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
+       cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
+
+       args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
+       if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) {
+               args->op.cmd = MMUEXT_INVLPG_MULTI;
+               args->op.arg1.linear_addr = start;
+       }
+
+       MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
+
+       xen_mc_issue(PARAVIRT_LAZY_MMU);
+}
+
+static unsigned long xen_read_cr3(void)
+{
+       return this_cpu_read(xen_cr3);
+}
+
+static void set_current_cr3(void *v)
+{
+       this_cpu_write(xen_current_cr3, (unsigned long)v);
+}
+
+static void __xen_write_cr3(bool kernel, unsigned long cr3)
+{
+       struct mmuext_op op;
+       unsigned long mfn;
+
+       trace_xen_mmu_write_cr3(kernel, cr3);
+
+       if (cr3)
+               mfn = pfn_to_mfn(PFN_DOWN(cr3));
+       else
+               mfn = 0;
+
+       WARN_ON(mfn == 0 && kernel);
+
+       op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
+       op.arg1.mfn = mfn;
+
+       xen_extend_mmuext_op(&op);
+
+       if (kernel) {
+               this_cpu_write(xen_cr3, cr3);
+
+               /* Update xen_current_cr3 once the batch has actually
+                  been submitted. */
+               xen_mc_callback(set_current_cr3, (void *)cr3);
+       }
+}
+static void xen_write_cr3(unsigned long cr3)
+{
+       BUG_ON(preemptible());
+
+       xen_mc_batch();  /* disables interrupts */
+
+       /* Update while interrupts are disabled, so its atomic with
+          respect to ipis */
+       this_cpu_write(xen_cr3, cr3);
+
+       __xen_write_cr3(true, cr3);
+
+#ifdef CONFIG_X86_64
+       {
+               pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
+               if (user_pgd)
+                       __xen_write_cr3(false, __pa(user_pgd));
+               else
+                       __xen_write_cr3(false, 0);
+       }
+#endif
+
+       xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
+}
+
+#ifdef CONFIG_X86_64
+/*
+ * At the start of the day - when Xen launches a guest, it has already
+ * built pagetables for the guest. We diligently look over them
+ * in xen_setup_kernel_pagetable and graft as appropriate them in the
+ * init_level4_pgt and its friends. Then when we are happy we load
+ * the new init_level4_pgt - and continue on.
+ *
+ * The generic code starts (start_kernel) and 'init_mem_mapping' sets
+ * up the rest of the pagetables. When it has completed it loads the cr3.
+ * N.B. that baremetal would start at 'start_kernel' (and the early
+ * #PF handler would create bootstrap pagetables) - so we are running
+ * with the same assumptions as what to do when write_cr3 is executed
+ * at this point.
+ *
+ * Since there are no user-page tables at all, we have two variants
+ * of xen_write_cr3 - the early bootup (this one), and the late one
+ * (xen_write_cr3). The reason we have to do that is that in 64-bit
+ * the Linux kernel and user-space are both in ring 3 while the
+ * hypervisor is in ring 0.
+ */
+static void __init xen_write_cr3_init(unsigned long cr3)
+{
+       BUG_ON(preemptible());
+
+       xen_mc_batch();  /* disables interrupts */
+
+       /* Update while interrupts are disabled, so its atomic with
+          respect to ipis */
+       this_cpu_write(xen_cr3, cr3);
+
+       __xen_write_cr3(true, cr3);
+
+       xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
+}
+#endif
+
+static int xen_pgd_alloc(struct mm_struct *mm)
+{
+       pgd_t *pgd = mm->pgd;
+       int ret = 0;
+
+       BUG_ON(PagePinned(virt_to_page(pgd)));
+
+#ifdef CONFIG_X86_64
+       {
+               struct page *page = virt_to_page(pgd);
+               pgd_t *user_pgd;
+
+               BUG_ON(page->private != 0);
+
+               ret = -ENOMEM;
+
+               user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
+               page->private = (unsigned long)user_pgd;
+
+               if (user_pgd != NULL) {
+#ifdef CONFIG_X86_VSYSCALL_EMULATION
+                       user_pgd[pgd_index(VSYSCALL_ADDR)] =
+                               __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
+#endif
+                       ret = 0;
+               }
+
+               BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
+       }
+#endif
+       return ret;
+}
+
+static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
+{
+#ifdef CONFIG_X86_64
+       pgd_t *user_pgd = xen_get_user_pgd(pgd);
+
+       if (user_pgd)
+               free_page((unsigned long)user_pgd);
+#endif
+}
+
+/*
+ * Init-time set_pte while constructing initial pagetables, which
+ * doesn't allow RO page table pages to be remapped RW.
+ *
+ * If there is no MFN for this PFN then this page is initially
+ * ballooned out so clear the PTE (as in decrease_reservation() in
+ * drivers/xen/balloon.c).
+ *
+ * Many of these PTE updates are done on unpinned and writable pages
+ * and doing a hypercall for these is unnecessary and expensive.  At
+ * this point it is not possible to tell if a page is pinned or not,
+ * so always write the PTE directly and rely on Xen trapping and
+ * emulating any updates as necessary.
+ */
+__visible pte_t xen_make_pte_init(pteval_t pte)
+{
+#ifdef CONFIG_X86_64
+       unsigned long pfn;
+
+       /*
+        * Pages belonging to the initial p2m list mapped outside the default
+        * address range must be mapped read-only. This region contains the
+        * page tables for mapping the p2m list, too, and page tables MUST be
+        * mapped read-only.
+        */
+       pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
+       if (xen_start_info->mfn_list < __START_KERNEL_map &&
+           pfn >= xen_start_info->first_p2m_pfn &&
+           pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
+               pte &= ~_PAGE_RW;
+#endif
+       pte = pte_pfn_to_mfn(pte);
+       return native_make_pte(pte);
+}
+PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
+
+static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
+{
+#ifdef CONFIG_X86_32
+       /* If there's an existing pte, then don't allow _PAGE_RW to be set */
+       if (pte_mfn(pte) != INVALID_P2M_ENTRY
+           && pte_val_ma(*ptep) & _PAGE_PRESENT)
+               pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
+                              pte_val_ma(pte));
+#endif
+       native_set_pte(ptep, pte);
+}
+
+/* Early in boot, while setting up the initial pagetable, assume
+   everything is pinned. */
+static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
+{
+#ifdef CONFIG_FLATMEM
+       BUG_ON(mem_map);        /* should only be used early */
+#endif
+       make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
+       pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
+}
+
+/* Used for pmd and pud */
+static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
+{
+#ifdef CONFIG_FLATMEM
+       BUG_ON(mem_map);        /* should only be used early */
+#endif
+       make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
+}
+
+/* Early release_pte assumes that all pts are pinned, since there's
+   only init_mm and anything attached to that is pinned. */
+static void __init xen_release_pte_init(unsigned long pfn)
+{
+       pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
+       make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
+}
+
+static void __init xen_release_pmd_init(unsigned long pfn)
+{
+       make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
+}
+
+static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
+{
+       struct multicall_space mcs;
+       struct mmuext_op *op;
+
+       mcs = __xen_mc_entry(sizeof(*op));
+       op = mcs.args;
+       op->cmd = cmd;
+       op->arg1.mfn = pfn_to_mfn(pfn);
+
+       MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
+}
+
+static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
+{
+       struct multicall_space mcs;
+       unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
+
+       mcs = __xen_mc_entry(0);
+       MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
+                               pfn_pte(pfn, prot), 0);
+}
+
+/* This needs to make sure the new pte page is pinned iff its being
+   attached to a pinned pagetable. */
+static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
+                                   unsigned level)
+{
+       bool pinned = PagePinned(virt_to_page(mm->pgd));
+
+       trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
+
+       if (pinned) {
+               struct page *page = pfn_to_page(pfn);
+
+               SetPagePinned(page);
+
+               if (!PageHighMem(page)) {
+                       xen_mc_batch();
+
+                       __set_pfn_prot(pfn, PAGE_KERNEL_RO);
+
+                       if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
+                               __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
+
+                       xen_mc_issue(PARAVIRT_LAZY_MMU);
+               } else {
+                       /* make sure there are no stray mappings of
+                          this page */
+                       kmap_flush_unused();
+               }
+       }
+}
+
+static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
+{
+       xen_alloc_ptpage(mm, pfn, PT_PTE);
+}
+
+static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
+{
+       xen_alloc_ptpage(mm, pfn, PT_PMD);
+}
+
+/* This should never happen until we're OK to use struct page */
+static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
+{
+       struct page *page = pfn_to_page(pfn);
+       bool pinned = PagePinned(page);
+
+       trace_xen_mmu_release_ptpage(pfn, level, pinned);
+
+       if (pinned) {
+               if (!PageHighMem(page)) {
+                       xen_mc_batch();
+
+                       if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
+                               __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
+
+                       __set_pfn_prot(pfn, PAGE_KERNEL);
+
+                       xen_mc_issue(PARAVIRT_LAZY_MMU);
+               }
+               ClearPagePinned(page);
+       }
+}
+
+static void xen_release_pte(unsigned long pfn)
+{
+       xen_release_ptpage(pfn, PT_PTE);
+}
+
+static void xen_release_pmd(unsigned long pfn)
+{
+       xen_release_ptpage(pfn, PT_PMD);
+}
+
+#if CONFIG_PGTABLE_LEVELS >= 4
+static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
+{
+       xen_alloc_ptpage(mm, pfn, PT_PUD);
+}
+
+static void xen_release_pud(unsigned long pfn)
+{
+       xen_release_ptpage(pfn, PT_PUD);
+}
+#endif
+
+void __init xen_reserve_top(void)
+{
+#ifdef CONFIG_X86_32
+       unsigned long top = HYPERVISOR_VIRT_START;
+       struct xen_platform_parameters pp;
+
+       if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
+               top = pp.virt_start;
+
+       reserve_top_address(-top);
+#endif /* CONFIG_X86_32 */
+}
+
+/*
+ * Like __va(), but returns address in the kernel mapping (which is
+ * all we have until the physical memory mapping has been set up.
+ */
+static void * __init __ka(phys_addr_t paddr)
+{
+#ifdef CONFIG_X86_64
+       return (void *)(paddr + __START_KERNEL_map);
+#else
+       return __va(paddr);
+#endif
+}
+
+/* Convert a machine address to physical address */
+static unsigned long __init m2p(phys_addr_t maddr)
+{
+       phys_addr_t paddr;
+
+       maddr &= PTE_PFN_MASK;
+       paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
+
+       return paddr;
+}
+
+/* Convert a machine address to kernel virtual */
+static void * __init m2v(phys_addr_t maddr)
+{
+       return __ka(m2p(maddr));
+}
+
+/* Set the page permissions on an identity-mapped pages */
+static void __init set_page_prot_flags(void *addr, pgprot_t prot,
+                                      unsigned long flags)
+{
+       unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
+       pte_t pte = pfn_pte(pfn, prot);
+
+       if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
+               BUG();
+}
+static void __init set_page_prot(void *addr, pgprot_t prot)
+{
+       return set_page_prot_flags(addr, prot, UVMF_NONE);
+}
+#ifdef CONFIG_X86_32
+static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
+{
+       unsigned pmdidx, pteidx;
+       unsigned ident_pte;
+       unsigned long pfn;
+
+       level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
+                                     PAGE_SIZE);
+
+       ident_pte = 0;
+       pfn = 0;
+       for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
+               pte_t *pte_page;
+
+               /* Reuse or allocate a page of ptes */
+               if (pmd_present(pmd[pmdidx]))
+                       pte_page = m2v(pmd[pmdidx].pmd);
+               else {
+                       /* Check for free pte pages */
+                       if (ident_pte == LEVEL1_IDENT_ENTRIES)
+                               break;
+
+                       pte_page = &level1_ident_pgt[ident_pte];
+                       ident_pte += PTRS_PER_PTE;
+
+                       pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
+               }
+
+               /* Install mappings */
+               for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
+                       pte_t pte;
+
+                       if (pfn > max_pfn_mapped)
+                               max_pfn_mapped = pfn;
+
+                       if (!pte_none(pte_page[pteidx]))
+                               continue;
+
+                       pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
+                       pte_page[pteidx] = pte;
+               }
+       }
+
+       for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
+               set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
+
+       set_page_prot(pmd, PAGE_KERNEL_RO);
+}
+#endif
+void __init xen_setup_machphys_mapping(void)
+{
+       struct xen_machphys_mapping mapping;
+
+       if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
+               machine_to_phys_mapping = (unsigned long *)mapping.v_start;
+               machine_to_phys_nr = mapping.max_mfn + 1;
+       } else {
+               machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
+       }
+#ifdef CONFIG_X86_32
+       WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
+               < machine_to_phys_mapping);
+#endif
+}
+
+#ifdef CONFIG_X86_64
+static void __init convert_pfn_mfn(void *v)
+{
+       pte_t *pte = v;
+       int i;
+
+       /* All levels are converted the same way, so just treat them
+          as ptes. */
+       for (i = 0; i < PTRS_PER_PTE; i++)
+               pte[i] = xen_make_pte(pte[i].pte);
+}
+static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
+                                unsigned long addr)
+{
+       if (*pt_base == PFN_DOWN(__pa(addr))) {
+               set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
+               clear_page((void *)addr);
+               (*pt_base)++;
+       }
+       if (*pt_end == PFN_DOWN(__pa(addr))) {
+               set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
+               clear_page((void *)addr);
+               (*pt_end)--;
+       }
+}
+/*
+ * Set up the initial kernel pagetable.
+ *
+ * We can construct this by grafting the Xen provided pagetable into
+ * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
+ * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
+ * kernel has a physical mapping to start with - but that's enough to
+ * get __va working.  We need to fill in the rest of the physical
+ * mapping once some sort of allocator has been set up.
+ */
+void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
+{
+       pud_t *l3;
+       pmd_t *l2;
+       unsigned long addr[3];
+       unsigned long pt_base, pt_end;
+       unsigned i;
+
+       /* max_pfn_mapped is the last pfn mapped in the initial memory
+        * mappings. Considering that on Xen after the kernel mappings we
+        * have the mappings of some pages that don't exist in pfn space, we
+        * set max_pfn_mapped to the last real pfn mapped. */
+       if (xen_start_info->mfn_list < __START_KERNEL_map)
+               max_pfn_mapped = xen_start_info->first_p2m_pfn;
+       else
+               max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
+
+       pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
+       pt_end = pt_base + xen_start_info->nr_pt_frames;
+
+       /* Zap identity mapping */
+       init_level4_pgt[0] = __pgd(0);
+
+       if (!xen_feature(XENFEAT_auto_translated_physmap)) {
+               /* Pre-constructed entries are in pfn, so convert to mfn */
+               /* L4[272] -> level3_ident_pgt
+                * L4[511] -> level3_kernel_pgt */
+               convert_pfn_mfn(init_level4_pgt);
+
+               /* L3_i[0] -> level2_ident_pgt */
+               convert_pfn_mfn(level3_ident_pgt);
+               /* L3_k[510] -> level2_kernel_pgt
+                * L3_k[511] -> level2_fixmap_pgt */
+               convert_pfn_mfn(level3_kernel_pgt);
+
+               /* L3_k[511][506] -> level1_fixmap_pgt */
+               convert_pfn_mfn(level2_fixmap_pgt);
+       }
+       /* We get [511][511] and have Xen's version of level2_kernel_pgt */
+       l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
+       l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
+
+       addr[0] = (unsigned long)pgd;
+       addr[1] = (unsigned long)l3;
+       addr[2] = (unsigned long)l2;
+       /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
+        * Both L4[272][0] and L4[511][510] have entries that point to the same
+        * L2 (PMD) tables. Meaning that if you modify it in __va space
+        * it will be also modified in the __ka space! (But if you just
+        * modify the PMD table to point to other PTE's or none, then you
+        * are OK - which is what cleanup_highmap does) */
+       copy_page(level2_ident_pgt, l2);
+       /* Graft it onto L4[511][510] */
+       copy_page(level2_kernel_pgt, l2);
+
+       /* Copy the initial P->M table mappings if necessary. */
+       i = pgd_index(xen_start_info->mfn_list);
+       if (i && i < pgd_index(__START_KERNEL_map))
+               init_level4_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
+
+       if (!xen_feature(XENFEAT_auto_translated_physmap)) {
+               /* Make pagetable pieces RO */
+               set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
+               set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
+               set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
+               set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
+               set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
+               set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
+               set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
+               set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
+
+               /* Pin down new L4 */
+               pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
+                                 PFN_DOWN(__pa_symbol(init_level4_pgt)));
+
+               /* Unpin Xen-provided one */
+               pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
+
+               /*
+                * At this stage there can be no user pgd, and no page
+                * structure to attach it to, so make sure we just set kernel
+                * pgd.
+                */
+               xen_mc_batch();
+               __xen_write_cr3(true, __pa(init_level4_pgt));
+               xen_mc_issue(PARAVIRT_LAZY_CPU);
+       } else
+               native_write_cr3(__pa(init_level4_pgt));
+
+       /* We can't that easily rip out L3 and L2, as the Xen pagetables are
+        * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
+        * the initial domain. For guests using the toolstack, they are in:
+        * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
+        * rip out the [L4] (pgd), but for guests we shave off three pages.
+        */
+       for (i = 0; i < ARRAY_SIZE(addr); i++)
+               check_pt_base(&pt_base, &pt_end, addr[i]);
+
+       /* Our (by three pages) smaller Xen pagetable that we are using */
+       xen_pt_base = PFN_PHYS(pt_base);
+       xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
+       memblock_reserve(xen_pt_base, xen_pt_size);
+
+       /* Revector the xen_start_info */
+       xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
+}
+
+/*
+ * Read a value from a physical address.
+ */
+static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
+{
+       unsigned long *vaddr;
+       unsigned long val;
+
+       vaddr = early_memremap_ro(addr, sizeof(val));
+       val = *vaddr;
+       early_memunmap(vaddr, sizeof(val));
+       return val;
+}
+
+/*
+ * Translate a virtual address to a physical one without relying on mapped
+ * page tables.
+ */
+static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
+{
+       phys_addr_t pa;
+       pgd_t pgd;
+       pud_t pud;
+       pmd_t pmd;
+       pte_t pte;
+
+       pa = read_cr3();
+       pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
+                                                      sizeof(pgd)));
+       if (!pgd_present(pgd))
+               return 0;
+
+       pa = pgd_val(pgd) & PTE_PFN_MASK;
+       pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
+                                                      sizeof(pud)));
+       if (!pud_present(pud))
+               return 0;
+       pa = pud_pfn(pud) << PAGE_SHIFT;
+       if (pud_large(pud))
+               return pa + (vaddr & ~PUD_MASK);
+
+       pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
+                                                      sizeof(pmd)));
+       if (!pmd_present(pmd))
+               return 0;
+       pa = pmd_pfn(pmd) << PAGE_SHIFT;
+       if (pmd_large(pmd))
+               return pa + (vaddr & ~PMD_MASK);
+
+       pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
+                                                      sizeof(pte)));
+       if (!pte_present(pte))
+               return 0;
+       pa = pte_pfn(pte) << PAGE_SHIFT;
+
+       return pa | (vaddr & ~PAGE_MASK);
+}
+
+/*
+ * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
+ * this area.
+ */
+void __init xen_relocate_p2m(void)
+{
+       phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys, p4d_phys;
+       unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
+       int n_pte, n_pt, n_pmd, n_pud, n_p4d, idx_pte, idx_pt, idx_pmd, idx_pud, idx_p4d;
+       pte_t *pt;
+       pmd_t *pmd;
+       pud_t *pud;
+       p4d_t *p4d = NULL;
+       pgd_t *pgd;
+       unsigned long *new_p2m;
+       int save_pud;
+
+       size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
+       n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
+       n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
+       n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
+       n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
+       if (PTRS_PER_P4D > 1)
+               n_p4d = roundup(size, PGDIR_SIZE) >> PGDIR_SHIFT;
+       else
+               n_p4d = 0;
+       n_frames = n_pte + n_pt + n_pmd + n_pud + n_p4d;
+
+       new_area = xen_find_free_area(PFN_PHYS(n_frames));
+       if (!new_area) {
+               xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
+               BUG();
+       }
+
+       /*
+        * Setup the page tables for addressing the new p2m list.
+        * We have asked the hypervisor to map the p2m list at the user address
+        * PUD_SIZE. It may have done so, or it may have used a kernel space
+        * address depending on the Xen version.
+        * To avoid any possible virtual address collision, just use
+        * 2 * PUD_SIZE for the new area.
+        */
+       p4d_phys = new_area;
+       pud_phys = p4d_phys + PFN_PHYS(n_p4d);
+       pmd_phys = pud_phys + PFN_PHYS(n_pud);
+       pt_phys = pmd_phys + PFN_PHYS(n_pmd);
+       p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
+
+       pgd = __va(read_cr3());
+       new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
+       idx_p4d = 0;
+       save_pud = n_pud;
+       do {
+               if (n_p4d > 0) {
+                       p4d = early_memremap(p4d_phys, PAGE_SIZE);
+                       clear_page(p4d);
+                       n_pud = min(save_pud, PTRS_PER_P4D);
+               }
+               for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
+                       pud = early_memremap(pud_phys, PAGE_SIZE);
+                       clear_page(pud);
+                       for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
+                                idx_pmd++) {
+                               pmd = early_memremap(pmd_phys, PAGE_SIZE);
+                               clear_page(pmd);
+                               for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
+                                        idx_pt++) {
+                                       pt = early_memremap(pt_phys, PAGE_SIZE);
+                                       clear_page(pt);
+                                       for (idx_pte = 0;
+                                                idx_pte < min(n_pte, PTRS_PER_PTE);
+                                                idx_pte++) {
+                                               set_pte(pt + idx_pte,
+                                                               pfn_pte(p2m_pfn, PAGE_KERNEL));
+                                               p2m_pfn++;
+                                       }
+                                       n_pte -= PTRS_PER_PTE;
+                                       early_memunmap(pt, PAGE_SIZE);
+                                       make_lowmem_page_readonly(__va(pt_phys));
+                                       pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
+                                                       PFN_DOWN(pt_phys));
+                                       set_pmd(pmd + idx_pt,
+                                                       __pmd(_PAGE_TABLE | pt_phys));
+                                       pt_phys += PAGE_SIZE;
+                               }
+                               n_pt -= PTRS_PER_PMD;
+                               early_memunmap(pmd, PAGE_SIZE);
+                               make_lowmem_page_readonly(__va(pmd_phys));
+                               pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
+                                               PFN_DOWN(pmd_phys));
+                               set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
+                               pmd_phys += PAGE_SIZE;
+                       }
+                       n_pmd -= PTRS_PER_PUD;
+                       early_memunmap(pud, PAGE_SIZE);
+                       make_lowmem_page_readonly(__va(pud_phys));
+                       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
+                       if (n_p4d > 0)
+                               set_p4d(p4d + idx_pud, __p4d(_PAGE_TABLE | pud_phys));
+                       else
+                               set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
+                       pud_phys += PAGE_SIZE;
+               }
+               if (n_p4d > 0) {
+                       save_pud -= PTRS_PER_P4D;
+                       early_memunmap(p4d, PAGE_SIZE);
+                       make_lowmem_page_readonly(__va(p4d_phys));
+                       pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, PFN_DOWN(p4d_phys));
+                       set_pgd(pgd + 2 + idx_p4d, __pgd(_PAGE_TABLE | p4d_phys));
+                       p4d_phys += PAGE_SIZE;
+               }
+       } while (++idx_p4d < n_p4d);
+
+       /* Now copy the old p2m info to the new area. */
+       memcpy(new_p2m, xen_p2m_addr, size);
+       xen_p2m_addr = new_p2m;
+
+       /* Release the old p2m list and set new list info. */
+       p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
+       BUG_ON(!p2m_pfn);
+       p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
+
+       if (xen_start_info->mfn_list < __START_KERNEL_map) {
+               pfn = xen_start_info->first_p2m_pfn;
+               pfn_end = xen_start_info->first_p2m_pfn +
+                         xen_start_info->nr_p2m_frames;
+               set_pgd(pgd + 1, __pgd(0));
+       } else {
+               pfn = p2m_pfn;
+               pfn_end = p2m_pfn_end;
+       }
+
+       memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
+       while (pfn < pfn_end) {
+               if (pfn == p2m_pfn) {
+                       pfn = p2m_pfn_end;
+                       continue;
+               }
+               make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
+               pfn++;
+       }
+
+       xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
+       xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
+       xen_start_info->nr_p2m_frames = n_frames;
+}
+
+#else  /* !CONFIG_X86_64 */
+static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
+static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
+
+static void __init xen_write_cr3_init(unsigned long cr3)
+{
+       unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
+
+       BUG_ON(read_cr3() != __pa(initial_page_table));
+       BUG_ON(cr3 != __pa(swapper_pg_dir));
+
+       /*
+        * We are switching to swapper_pg_dir for the first time (from
+        * initial_page_table) and therefore need to mark that page
+        * read-only and then pin it.
+        *
+        * Xen disallows sharing of kernel PMDs for PAE
+        * guests. Therefore we must copy the kernel PMD from
+        * initial_page_table into a new kernel PMD to be used in
+        * swapper_pg_dir.
+        */
+       swapper_kernel_pmd =
+               extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
+       copy_page(swapper_kernel_pmd, initial_kernel_pmd);
+       swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
+               __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
+       set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
+
+       set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
+       xen_write_cr3(cr3);
+       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
+
+       pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
+                         PFN_DOWN(__pa(initial_page_table)));
+       set_page_prot(initial_page_table, PAGE_KERNEL);
+       set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
+
+       pv_mmu_ops.write_cr3 = &xen_write_cr3;
+}
+
+/*
+ * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
+ * not the first page table in the page table pool.
+ * Iterate through the initial page tables to find the real page table base.
+ */
+static phys_addr_t xen_find_pt_base(pmd_t *pmd)
+{
+       phys_addr_t pt_base, paddr;
+       unsigned pmdidx;
+
+       pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
+
+       for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
+               if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
+                       paddr = m2p(pmd[pmdidx].pmd);
+                       pt_base = min(pt_base, paddr);
+               }
+
+       return pt_base;
+}
+
+void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
+{
+       pmd_t *kernel_pmd;
+
+       kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
+
+       xen_pt_base = xen_find_pt_base(kernel_pmd);
+       xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
+
+       initial_kernel_pmd =
+               extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
+
+       max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
+
+       copy_page(initial_kernel_pmd, kernel_pmd);
+
+       xen_map_identity_early(initial_kernel_pmd, max_pfn);
+
+       copy_page(initial_page_table, pgd);
+       initial_page_table[KERNEL_PGD_BOUNDARY] =
+               __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
+
+       set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
+       set_page_prot(initial_page_table, PAGE_KERNEL_RO);
+       set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
+
+       pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
+
+       pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
+                         PFN_DOWN(__pa(initial_page_table)));
+       xen_write_cr3(__pa(initial_page_table));
+
+       memblock_reserve(xen_pt_base, xen_pt_size);
+}
+#endif /* CONFIG_X86_64 */
+
+void __init xen_reserve_special_pages(void)
+{
+       phys_addr_t paddr;
+
+       memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
+       if (xen_start_info->store_mfn) {
+               paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
+               memblock_reserve(paddr, PAGE_SIZE);
+       }
+       if (!xen_initial_domain()) {
+               paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
+               memblock_reserve(paddr, PAGE_SIZE);
+       }
+}
+
+void __init xen_pt_check_e820(void)
+{
+       if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
+               xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
+               BUG();
+       }
+}
+
+static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
+
+static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
+{
+       pte_t pte;
+
+       phys >>= PAGE_SHIFT;
+
+       switch (idx) {
+       case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
+       case FIX_RO_IDT:
+#ifdef CONFIG_X86_32
+       case FIX_WP_TEST:
+# ifdef CONFIG_HIGHMEM
+       case FIX_KMAP_BEGIN ... FIX_KMAP_END:
+# endif
+#elif defined(CONFIG_X86_VSYSCALL_EMULATION)
+       case VSYSCALL_PAGE:
+#endif
+       case FIX_TEXT_POKE0:
+       case FIX_TEXT_POKE1:
+       case FIX_GDT_REMAP_BEGIN ... FIX_GDT_REMAP_END:
+               /* All local page mappings */
+               pte = pfn_pte(phys, prot);
+               break;
+
+#ifdef CONFIG_X86_LOCAL_APIC
+       case FIX_APIC_BASE:     /* maps dummy local APIC */
+               pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
+               break;
+#endif
+
+#ifdef CONFIG_X86_IO_APIC
+       case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
+               /*
+                * We just don't map the IO APIC - all access is via
+                * hypercalls.  Keep the address in the pte for reference.
+                */
+               pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
+               break;
+#endif
+
+       case FIX_PARAVIRT_BOOTMAP:
+               /* This is an MFN, but it isn't an IO mapping from the
+                  IO domain */
+               pte = mfn_pte(phys, prot);
+               break;
+
+       default:
+               /* By default, set_fixmap is used for hardware mappings */
+               pte = mfn_pte(phys, prot);
+               break;
+       }
+
+       __native_set_fixmap(idx, pte);
+
+#ifdef CONFIG_X86_VSYSCALL_EMULATION
+       /* Replicate changes to map the vsyscall page into the user
+          pagetable vsyscall mapping. */
+       if (idx == VSYSCALL_PAGE) {
+               unsigned long vaddr = __fix_to_virt(idx);
+               set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
+       }
+#endif
+}
+
+static void __init xen_post_allocator_init(void)
+{
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return;
+
+       pv_mmu_ops.set_pte = xen_set_pte;
+       pv_mmu_ops.set_pmd = xen_set_pmd;
+       pv_mmu_ops.set_pud = xen_set_pud;
+#if CONFIG_PGTABLE_LEVELS >= 4
+       pv_mmu_ops.set_p4d = xen_set_p4d;
+#endif
+
+       /* This will work as long as patching hasn't happened yet
+          (which it hasn't) */
+       pv_mmu_ops.alloc_pte = xen_alloc_pte;
+       pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
+       pv_mmu_ops.release_pte = xen_release_pte;
+       pv_mmu_ops.release_pmd = xen_release_pmd;
+#if CONFIG_PGTABLE_LEVELS >= 4
+       pv_mmu_ops.alloc_pud = xen_alloc_pud;
+       pv_mmu_ops.release_pud = xen_release_pud;
+#endif
+       pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
+
+#ifdef CONFIG_X86_64
+       pv_mmu_ops.write_cr3 = &xen_write_cr3;
+       SetPagePinned(virt_to_page(level3_user_vsyscall));
+#endif
+       xen_mark_init_mm_pinned();
+}
+
+static void xen_leave_lazy_mmu(void)
+{
+       preempt_disable();
+       xen_mc_flush();
+       paravirt_leave_lazy_mmu();
+       preempt_enable();
+}
+
+static const struct pv_mmu_ops xen_mmu_ops __initconst = {
+       .read_cr2 = xen_read_cr2,
+       .write_cr2 = xen_write_cr2,
+
+       .read_cr3 = xen_read_cr3,
+       .write_cr3 = xen_write_cr3_init,
+
+       .flush_tlb_user = xen_flush_tlb,
+       .flush_tlb_kernel = xen_flush_tlb,
+       .flush_tlb_single = xen_flush_tlb_single,
+       .flush_tlb_others = xen_flush_tlb_others,
+
+       .pte_update = paravirt_nop,
+
+       .pgd_alloc = xen_pgd_alloc,
+       .pgd_free = xen_pgd_free,
+
+       .alloc_pte = xen_alloc_pte_init,
+       .release_pte = xen_release_pte_init,
+       .alloc_pmd = xen_alloc_pmd_init,
+       .release_pmd = xen_release_pmd_init,
+
+       .set_pte = xen_set_pte_init,
+       .set_pte_at = xen_set_pte_at,
+       .set_pmd = xen_set_pmd_hyper,
+
+       .ptep_modify_prot_start = __ptep_modify_prot_start,
+       .ptep_modify_prot_commit = __ptep_modify_prot_commit,
+
+       .pte_val = PV_CALLEE_SAVE(xen_pte_val),
+       .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
+
+       .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
+       .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
+
+#ifdef CONFIG_X86_PAE
+       .set_pte_atomic = xen_set_pte_atomic,
+       .pte_clear = xen_pte_clear,
+       .pmd_clear = xen_pmd_clear,
+#endif /* CONFIG_X86_PAE */
+       .set_pud = xen_set_pud_hyper,
+
+       .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
+       .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
+
+#if CONFIG_PGTABLE_LEVELS >= 4
+       .pud_val = PV_CALLEE_SAVE(xen_pud_val),
+       .make_pud = PV_CALLEE_SAVE(xen_make_pud),
+       .set_p4d = xen_set_p4d_hyper,
+
+       .alloc_pud = xen_alloc_pmd_init,
+       .release_pud = xen_release_pmd_init,
+#endif /* CONFIG_PGTABLE_LEVELS == 4 */
+
+       .activate_mm = xen_activate_mm,
+       .dup_mmap = xen_dup_mmap,
+       .exit_mmap = xen_exit_mmap,
+
+       .lazy_mode = {
+               .enter = paravirt_enter_lazy_mmu,
+               .leave = xen_leave_lazy_mmu,
+               .flush = paravirt_flush_lazy_mmu,
+       },
+
+       .set_fixmap = xen_set_fixmap,
+};
+
+void __init xen_init_mmu_ops(void)
+{
+       x86_init.paging.pagetable_init = xen_pagetable_init;
+
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return;
+
+       pv_mmu_ops = xen_mmu_ops;
+
+       memset(dummy_mapping, 0xff, PAGE_SIZE);
+}
+
+/* Protected by xen_reservation_lock. */
+#define MAX_CONTIG_ORDER 9 /* 2MB */
+static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
+
+#define VOID_PTE (mfn_pte(0, __pgprot(0)))
+static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
+                               unsigned long *in_frames,
+                               unsigned long *out_frames)
+{
+       int i;
+       struct multicall_space mcs;
+
+       xen_mc_batch();
+       for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
+               mcs = __xen_mc_entry(0);
+
+               if (in_frames)
+                       in_frames[i] = virt_to_mfn(vaddr);
+
+               MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
+               __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
+
+               if (out_frames)
+                       out_frames[i] = virt_to_pfn(vaddr);
+       }
+       xen_mc_issue(0);
+}
+
+/*
+ * Update the pfn-to-mfn mappings for a virtual address range, either to
+ * point to an array of mfns, or contiguously from a single starting
+ * mfn.
+ */
+static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
+                                    unsigned long *mfns,
+                                    unsigned long first_mfn)
+{
+       unsigned i, limit;
+       unsigned long mfn;
+
+       xen_mc_batch();
+
+       limit = 1u << order;
+       for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
+               struct multicall_space mcs;
+               unsigned flags;
+
+               mcs = __xen_mc_entry(0);
+               if (mfns)
+                       mfn = mfns[i];
+               else
+                       mfn = first_mfn + i;
+
+               if (i < (limit - 1))
+                       flags = 0;
+               else {
+                       if (order == 0)
+                               flags = UVMF_INVLPG | UVMF_ALL;
+                       else
+                               flags = UVMF_TLB_FLUSH | UVMF_ALL;
+               }
+
+               MULTI_update_va_mapping(mcs.mc, vaddr,
+                               mfn_pte(mfn, PAGE_KERNEL), flags);
+
+               set_phys_to_machine(virt_to_pfn(vaddr), mfn);
+       }
+
+       xen_mc_issue(0);
+}
+
+/*
+ * Perform the hypercall to exchange a region of our pfns to point to
+ * memory with the required contiguous alignment.  Takes the pfns as
+ * input, and populates mfns as output.
+ *
+ * Returns a success code indicating whether the hypervisor was able to
+ * satisfy the request or not.
+ */
+static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
+                              unsigned long *pfns_in,
+                              unsigned long extents_out,
+                              unsigned int order_out,
+                              unsigned long *mfns_out,
+                              unsigned int address_bits)
+{
+       long rc;
+       int success;
+
+       struct xen_memory_exchange exchange = {
+               .in = {
+                       .nr_extents   = extents_in,
+                       .extent_order = order_in,
+                       .extent_start = pfns_in,
+                       .domid        = DOMID_SELF
+               },
+               .out = {
+                       .nr_extents   = extents_out,
+                       .extent_order = order_out,
+                       .extent_start = mfns_out,
+                       .address_bits = address_bits,
+                       .domid        = DOMID_SELF
+               }
+       };
+
+       BUG_ON(extents_in << order_in != extents_out << order_out);
+
+       rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
+       success = (exchange.nr_exchanged == extents_in);
+
+       BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
+       BUG_ON(success && (rc != 0));
+
+       return success;
+}
+
+int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
+                                unsigned int address_bits,
+                                dma_addr_t *dma_handle)
+{
+       unsigned long *in_frames = discontig_frames, out_frame;
+       unsigned long  flags;
+       int            success;
+       unsigned long vstart = (unsigned long)phys_to_virt(pstart);
+
+       /*
+        * Currently an auto-translated guest will not perform I/O, nor will
+        * it require PAE page directories below 4GB. Therefore any calls to
+        * this function are redundant and can be ignored.
+        */
+
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return 0;
+
+       if (unlikely(order > MAX_CONTIG_ORDER))
+               return -ENOMEM;
+
+       memset((void *) vstart, 0, PAGE_SIZE << order);
+
+       spin_lock_irqsave(&xen_reservation_lock, flags);
+
+       /* 1. Zap current PTEs, remembering MFNs. */
+       xen_zap_pfn_range(vstart, order, in_frames, NULL);
+
+       /* 2. Get a new contiguous memory extent. */
+       out_frame = virt_to_pfn(vstart);
+       success = xen_exchange_memory(1UL << order, 0, in_frames,
+                                     1, order, &out_frame,
+                                     address_bits);
+
+       /* 3. Map the new extent in place of old pages. */
+       if (success)
+               xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
+       else
+               xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
+
+       spin_unlock_irqrestore(&xen_reservation_lock, flags);
+
+       *dma_handle = virt_to_machine(vstart).maddr;
+       return success ? 0 : -ENOMEM;
+}
+EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
+
+void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
+{
+       unsigned long *out_frames = discontig_frames, in_frame;
+       unsigned long  flags;
+       int success;
+       unsigned long vstart;
+
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return;
+
+       if (unlikely(order > MAX_CONTIG_ORDER))
+               return;
+
+       vstart = (unsigned long)phys_to_virt(pstart);
+       memset((void *) vstart, 0, PAGE_SIZE << order);
+
+       spin_lock_irqsave(&xen_reservation_lock, flags);
+
+       /* 1. Find start MFN of contiguous extent. */
+       in_frame = virt_to_mfn(vstart);
+
+       /* 2. Zap current PTEs. */
+       xen_zap_pfn_range(vstart, order, NULL, out_frames);
+
+       /* 3. Do the exchange for non-contiguous MFNs. */
+       success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
+                                       0, out_frames, 0);
+
+       /* 4. Map new pages in place of old pages. */
+       if (success)
+               xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
+       else
+               xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
+
+       spin_unlock_irqrestore(&xen_reservation_lock, flags);
+}
+EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);