/*H:360
* (ii) Making sure the Guest stack is mapped.
*
* Remember that direct traps into the Guest need a mapped Guest kernel stack.
* pin_stack_pages() calls us here: we could simply call demand_page(), but as
* we've seen that logic is quite long, and usually the stack pages are already
* mapped, so it's overkill.
*
* This is a quick version which answers the question: is this virtual address
* mapped by the shadow page tables, and is it writable?
*/
static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
{
pgd_t *spgd;
unsigned long flags;
#ifdef CONFIG_X86_PAE
pmd_t *spmd;
#endif
/* Look at the current top level entry: is it present? */
spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr);
if (!(pgd_flags(*spgd) & _PAGE_PRESENT))
return false;
#ifdef CONFIG_X86_PAE
spmd = spmd_addr(cpu, *spgd, vaddr);
if (!(pmd_flags(*spmd) & _PAGE_PRESENT))
return false;
#endif
/*
* Check the flags on the pte entry itself: it must be present and
* writable.
*/
flags = pte_flags(*(spte_addr(cpu, *spgd, vaddr)));
return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW);
}
/*H:340
* Converting a Guest page table entry to a shadow (ie. real) page table
* entry can be a little tricky. The flags are (almost) the same, but the
* Guest PTE contains a virtual page number: the CPU needs the real page
* number.
*/
static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write)
{
unsigned long pfn, base, flags;
/*
* The Guest sets the global flag, because it thinks that it is using
* PGE. We only told it to use PGE so it would tell us whether it was
* flushing a kernel mapping or a userspace mapping. We don't actually
* use the global bit, so throw it away.
*/
flags = (pte_flags(gpte) & ~_PAGE_GLOBAL);
/* The Guest's pages are offset inside the Launcher. */
base = (unsigned long)cpu->lg->mem_base / PAGE_SIZE;
/*
* We need a temporary "unsigned long" variable to hold the answer from
* get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't
* fit in spte.pfn. get_pfn() finds the real physical number of the
* page, given the virtual number.
*/
pfn = get_pfn(base + pte_pfn(gpte), write);
if (pfn == -1UL) {
kill_guest(cpu, "failed to get page %lu", pte_pfn(gpte));
/*
* When we destroy the Guest, we'll go through the shadow page
* tables and release_pte() them. Make sure we don't think
* this one is valid!
*/
flags = 0;
}
/* Now we assemble our shadow PTE from the page number and flags. */
return pfn_pte(pfn, __pgprot(flags));
}
/*H:501
* We do need the Switcher code mapped at all times, so we allocate that
* part of the Guest page table here. We map the Switcher code immediately,
* but defer mapping of the guest register page and IDT/LDT etc page until
* just before we run the guest in map_switcher_in_guest().
*
* We *could* do this setup in map_switcher_in_guest(), but at that point
* we've interrupts disabled, and allocating pages like that is fraught: we
* can't sleep if we need to free up some memory.
*/
static bool allocate_switcher_mapping(struct lg_cpu *cpu)
{
int i;
for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
pte_t *pte = find_spte(cpu, switcher_addr + i * PAGE_SIZE, true,
CHECK_GPGD_MASK, _PAGE_TABLE);
if (!pte)
return false;
/*
* Map the switcher page if not already there. It might
* already be there because we call allocate_switcher_mapping()
* in guest_set_pgd() just in case it did discard our Switcher
* mapping, but it probably didn't.
*/
if (i == 0 && !(pte_flags(*pte) & _PAGE_PRESENT)) {
/* Get a reference to the Switcher page. */
get_page(lg_switcher_pages[0]);
/* Create a read-only, exectuable, kernel-style PTE */
set_pte(pte,
mk_pte(lg_switcher_pages[0], PAGE_KERNEL_RX));
}
}
cpu->lg->pgdirs[cpu->cpu_pgd].switcher_mapped = true;
return true;
}
/* We walk down the guest page tables to get a guest-physical address */
unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr)
{
pgd_t gpgd;
pte_t gpte;
#ifdef CONFIG_X86_PAE
pmd_t gpmd;
#endif
/* Still not set up? Just map 1:1. */
if (unlikely(cpu->linear_pages))
return vaddr;
/* First step: get the top-level Guest page table entry. */
gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t);
/* Toplevel not present? We can't map it in. */
if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) {
kill_guest(cpu, "Bad address %#lx", vaddr);
return -1UL;
}
#ifdef CONFIG_X86_PAE
gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
kill_guest(cpu, "Bad address %#lx", vaddr);
gpte = lgread(cpu, gpte_addr(cpu, gpmd, vaddr), pte_t);
#else
gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
#endif
if (!(pte_flags(gpte) & _PAGE_PRESENT))
kill_guest(cpu, "Bad address %#lx", vaddr);
return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK);
}
/*H:460 And to complete the chain, release_pte() looks like this: */
static void release_pte(pte_t pte)
{
/*
* Remember that get_user_pages_fast() took a reference to the page, in
* get_pfn()? We have to put it back now.
*/
if (pte_flags(pte) & _PAGE_PRESENT)
put_page(pte_page(pte));
}
static bool check_gpte(struct lg_cpu *cpu, pte_t gpte)
{
if ((pte_flags(gpte) & _PAGE_PSE) ||
pte_pfn(gpte) >= cpu->lg->pfn_limit) {
kill_guest(cpu, "bad page table entry");
return false;
}
return true;
}
static bool gpte_in_iomem(struct lg_cpu *cpu, pte_t gpte)
{
/* We don't handle large pages. */
if (pte_flags(gpte) & _PAGE_PSE)
return false;
return (pte_pfn(gpte) >= cpu->lg->pfn_limit
&& pte_pfn(gpte) < cpu->lg->device_limit);
}
/*H:420
* This is the routine which actually sets the page table entry for then
* "idx"'th shadow page table.
*
* Normally, we can just throw out the old entry and replace it with 0: if they
* use it demand_page() will put the new entry in. We need to do this anyway:
* The Guest expects _PAGE_ACCESSED to be set on its PTE the first time a page
* is read from, and _PAGE_DIRTY when it's written to.
*
* But Avi Kivity pointed out that most Operating Systems (Linux included) set
* these bits on PTEs immediately anyway. This is done to save the CPU from
* having to update them, but it helps us the same way: if they set
* _PAGE_ACCESSED then we can put a read-only PTE entry in immediately, and if
* they set _PAGE_DIRTY then we can put a writable PTE entry in immediately.
*/
static void __guest_set_pte(struct lg_cpu *cpu, int idx,
unsigned long vaddr, pte_t gpte)
{
/* Look up the matching shadow page directory entry. */
pgd_t *spgd = spgd_addr(cpu, idx, vaddr);
#ifdef CONFIG_X86_PAE
pmd_t *spmd;
#endif
/* If the top level isn't present, there's no entry to update. */
if (pgd_flags(*spgd) & _PAGE_PRESENT) {
#ifdef CONFIG_X86_PAE
spmd = spmd_addr(cpu, *spgd, vaddr);
if (pmd_flags(*spmd) & _PAGE_PRESENT) {
#endif
/* Otherwise, start by releasing the existing entry. */
pte_t *spte = spte_addr(cpu, *spgd, vaddr);
release_pte(*spte);
/*
* If they're setting this entry as dirty or accessed,
* we might as well put that entry they've given us in
* now. This shaves 10% off a copy-on-write
* micro-benchmark.
*/
if ((pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED))
&& !gpte_in_iomem(cpu, gpte)) {
if (!check_gpte(cpu, gpte))
return;
set_pte(spte,
gpte_to_spte(cpu, gpte,
pte_flags(gpte) & _PAGE_DIRTY));
} else {
/*
* Otherwise kill it and we can demand_page()
* it in later.
*/
set_pte(spte, __pte(0));
}
#ifdef CONFIG_X86_PAE
}
#endif
}
}
static void ept_mmu_notifier_change_pte(struct mmu_notifier *mn,
struct mm_struct *mm,
unsigned long address,
pte_t pte)
{
struct vmx_vcpu *vcpu = mmu_notifier_to_vmx(mn);
pr_debug("ept: change_pte addr %lx flags %lx\n", address, pte_flags(pte));
/*
* NOTE: Recent linux kernels (seen on 3.7 at least) hold a lock
* while calling this notifier, making it impossible to call
* get_user_pages_fast(). As a result, we just invalidate the
* page so that the mapping can be recreated later during a fault.
*/
ept_invalidate_page(vcpu, mm, address);
}
/*H:360
* (ii) Making sure the Guest stack is mapped.
*
* Remember that direct traps into the Guest need a mapped Guest kernel stack.
* pin_stack_pages() calls us here: we could simply call demand_page(), but as
* we've seen that logic is quite long, and usually the stack pages are already
* mapped, so it's overkill.
*
* This is a quick version which answers the question: is this virtual address
* mapped by the shadow page tables, and is it writable?
*/
static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
{
pte_t *spte;
unsigned long flags;
/* You can't put your stack in the Switcher! */
if (vaddr >= switcher_addr)
return false;
/* If there's no shadow PTE, it's not writable. */
spte = find_spte(cpu, vaddr, false, 0, 0);
if (!spte)
return false;
/*
* Check the flags on the pte entry itself: it must be present and
* writable.
*/
flags = pte_flags(*spte);
return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW);
}
Pml4e *
copyuvm(Pml4e *oldmap, usize sz)
{
uintptr a;
Pml4e *newmap;
Pte *pte;
uchar *oldmem, *newmem;
uint flags;
newmap = setupkvm();
if (newmap == nil)
return nil;
for (a = 0; a < sz; a += PGSIZE) {
pte = walkpgmap(oldmap, (void *)a, 0);
if (pte == nil)
panic("copyuvm - nil pte");
if (!*pte & PTE_P)
panic("copyuvm - page not present");
oldmem = p2v(pte_addr(*pte));
flags = pte_flags(*pte);
newmem = kalloc();
if (newmem == nil)
goto bad;
memmove(newmem, oldmem, PGSIZE);
if (mappages(newmap, (void *)a, PGSIZE, v2p(newmem), flags) < 0)
goto bad;
}
return newmap;
bad:
freeuvm(newmap);
return nil;
}
/* We walk down the guest page tables to get a guest-physical address */
bool __guest_pa(struct lg_cpu *cpu, unsigned long vaddr, unsigned long *paddr)
{
pgd_t gpgd;
pte_t gpte;
#ifdef CONFIG_X86_PAE
pmd_t gpmd;
#endif
/* Still not set up? Just map 1:1. */
if (unlikely(cpu->linear_pages)) {
*paddr = vaddr;
return true;
}
/* First step: get the top-level Guest page table entry. */
gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t);
/* Toplevel not present? We can't map it in. */
if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
goto fail;
#ifdef CONFIG_X86_PAE
gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
goto fail;
gpte = lgread(cpu, gpte_addr(cpu, gpmd, vaddr), pte_t);
#else
gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t);
#endif
if (!(pte_flags(gpte) & _PAGE_PRESENT))
goto fail;
*paddr = pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK);
return true;
fail:
*paddr = -1UL;
return false;
}
static int pte_testbit(pte_t pte)
{
return pte_flags(pte) & _PAGE_UNUSED1;
}
/*H:330
* (i) Looking up a page table entry when the Guest faults.
*
* We saw this call in run_guest(): when we see a page fault in the Guest, we
* come here. That's because we only set up the shadow page tables lazily as
* they're needed, so we get page faults all the time and quietly fix them up
* and return to the Guest without it knowing.
*
* If we fixed up the fault (ie. we mapped the address), this routine returns
* true. Otherwise, it was a real fault and we need to tell the Guest.
*
* There's a corner case: they're trying to access memory between
* pfn_limit and device_limit, which is I/O memory. In this case, we
* return false and set @iomem to the physical address, so the the
* Launcher can handle the instruction manually.
*/
bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode,
unsigned long *iomem)
{
unsigned long gpte_ptr;
pte_t gpte;
pte_t *spte;
pmd_t gpmd;
pgd_t gpgd;
*iomem = 0;
/* We never demand page the Switcher, so trying is a mistake. */
if (vaddr >= switcher_addr)
return false;
/* First step: get the top-level Guest page table entry. */
if (unlikely(cpu->linear_pages)) {
/* Faking up a linear mapping. */
gpgd = __pgd(CHECK_GPGD_MASK);
} else {
gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t);
/* Toplevel not present? We can't map it in. */
if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
return false;
/*
* This kills the Guest if it has weird flags or tries to
* refer to a "physical" address outside the bounds.
*/
if (!check_gpgd(cpu, gpgd))
return false;
}
/* This "mid-level" entry is only used for non-linear, PAE mode. */
gpmd = __pmd(_PAGE_TABLE);
#ifdef CONFIG_X86_PAE
if (likely(!cpu->linear_pages)) {
gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
/* Middle level not present? We can't map it in. */
if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
return false;
/*
* This kills the Guest if it has weird flags or tries to
* refer to a "physical" address outside the bounds.
*/
if (!check_gpmd(cpu, gpmd))
return false;
}
/*
* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later.
*/
gpte_ptr = gpte_addr(cpu, gpmd, vaddr);
#else
/*
* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later.
*/
gpte_ptr = gpte_addr(cpu, gpgd, vaddr);
#endif
if (unlikely(cpu->linear_pages)) {
/* Linear? Make up a PTE which points to same page. */
gpte = __pte((vaddr & PAGE_MASK) | _PAGE_RW | _PAGE_PRESENT);
} else {
/* Read the actual PTE value. */
gpte = lgread(cpu, gpte_ptr, pte_t);
}
/* If this page isn't in the Guest page tables, we can't page it in. */
if (!(pte_flags(gpte) & _PAGE_PRESENT))
return false;
/*
* Check they're not trying to write to a page the Guest wants
* read-only (bit 2 of errcode == write).
*/
if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW))
return false;
/* User access to a kernel-only page? (bit 3 == user access) */
if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER))
return false;
/* If they're accessing io memory, we expect a fault. */
if (gpte_in_iomem(cpu, gpte)) {
*iomem = (pte_pfn(gpte) << PAGE_SHIFT) | (vaddr & ~PAGE_MASK);
return false;
}
/*
* Check that the Guest PTE flags are OK, and the page number is below
* the pfn_limit (ie. not mapping the Launcher binary).
*/
if (!check_gpte(cpu, gpte))
return false;
/* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */
gpte = pte_mkyoung(gpte);
if (errcode & 2)
gpte = pte_mkdirty(gpte);
/* Get the pointer to the shadow PTE entry we're going to set. */
spte = find_spte(cpu, vaddr, true, pgd_flags(gpgd), pmd_flags(gpmd));
if (!spte)
return false;
/*
* If there was a valid shadow PTE entry here before, we release it.
* This can happen with a write to a previously read-only entry.
*/
release_pte(*spte);
/*
* If this is a write, we insist that the Guest page is writable (the
* final arg to gpte_to_spte()).
*/
if (pte_dirty(gpte))
*spte = gpte_to_spte(cpu, gpte, 1);
else
/*
* If this is a read, don't set the "writable" bit in the page
* table entry, even if the Guest says it's writable. That way
* we will come back here when a write does actually occur, so
* we can update the Guest's _PAGE_DIRTY flag.
*/
set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0));
/*
* Finally, we write the Guest PTE entry back: we've set the
* _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags.
*/
if (likely(!cpu->linear_pages))
lgwrite(cpu, gpte_ptr, pte_t, gpte);
/*
* The fault is fixed, the page table is populated, the mapping
* manipulated, the result returned and the code complete. A small
* delay and a trace of alliteration are the only indications the Guest
* has that a page fault occurred at all.
*/
return true;
}
/*H:330
* (i) Looking up a page table entry when the Guest faults.
*
* We saw this call in run_guest(): when we see a page fault in the Guest, we
* come here. That's because we only set up the shadow page tables lazily as
* they're needed, so we get page faults all the time and quietly fix them up
* and return to the Guest without it knowing.
*
* If we fixed up the fault (ie. we mapped the address), this routine returns
* true. Otherwise, it was a real fault and we need to tell the Guest.
*/
bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
{
pgd_t gpgd;
pgd_t *spgd;
unsigned long gpte_ptr;
pte_t gpte;
pte_t *spte;
/* Mid level for PAE. */
#ifdef CONFIG_X86_PAE
pmd_t *spmd;
pmd_t gpmd;
#endif
/* First step: get the top-level Guest page table entry. */
if (unlikely(cpu->linear_pages)) {
/* Faking up a linear mapping. */
gpgd = __pgd(CHECK_GPGD_MASK);
} else {
gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t);
/* Toplevel not present? We can't map it in. */
if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
return false;
}
/* Now look at the matching shadow entry. */
spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr);
if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) {
/* No shadow entry: allocate a new shadow PTE page. */
unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
/*
* This is not really the Guest's fault, but killing it is
* simple for this corner case.
*/
if (!ptepage) {
kill_guest(cpu, "out of memory allocating pte page");
return false;
}
/* We check that the Guest pgd is OK. */
check_gpgd(cpu, gpgd);
/*
* And we copy the flags to the shadow PGD entry. The page
* number in the shadow PGD is the page we just allocated.
*/
set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags(gpgd)));
}
#ifdef CONFIG_X86_PAE
if (unlikely(cpu->linear_pages)) {
/* Faking up a linear mapping. */
gpmd = __pmd(_PAGE_TABLE);
} else {
gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
/* Middle level not present? We can't map it in. */
if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
return false;
}
/* Now look at the matching shadow entry. */
spmd = spmd_addr(cpu, *spgd, vaddr);
if (!(pmd_flags(*spmd) & _PAGE_PRESENT)) {
/* No shadow entry: allocate a new shadow PTE page. */
unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
/*
* This is not really the Guest's fault, but killing it is
* simple for this corner case.
*/
if (!ptepage) {
kill_guest(cpu, "out of memory allocating pte page");
return false;
}
/* We check that the Guest pmd is OK. */
check_gpmd(cpu, gpmd);
/*
* And we copy the flags to the shadow PMD entry. The page
* number in the shadow PMD is the page we just allocated.
*/
set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd)));
}
/*
* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later.
*/
gpte_ptr = gpte_addr(cpu, gpmd, vaddr);
#else
/*
* OK, now we look at the lower level in the Guest page table: keep its
* address, because we might update it later.
*/
gpte_ptr = gpte_addr(cpu, gpgd, vaddr);
#endif
if (unlikely(cpu->linear_pages)) {
/* Linear? Make up a PTE which points to same page. */
gpte = __pte((vaddr & PAGE_MASK) | _PAGE_RW | _PAGE_PRESENT);
} else {
/* Read the actual PTE value. */
gpte = lgread(cpu, gpte_ptr, pte_t);
}
/* If this page isn't in the Guest page tables, we can't page it in. */
if (!(pte_flags(gpte) & _PAGE_PRESENT))
return false;
/*
* Check they're not trying to write to a page the Guest wants
* read-only (bit 2 of errcode == write).
*/
if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW))
return false;
/* User access to a kernel-only page? (bit 3 == user access) */
if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER))
return false;
/*
* Check that the Guest PTE flags are OK, and the page number is below
* the pfn_limit (ie. not mapping the Launcher binary).
*/
check_gpte(cpu, gpte);
/* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */
gpte = pte_mkyoung(gpte);
if (errcode & 2)
gpte = pte_mkdirty(gpte);
/* Get the pointer to the shadow PTE entry we're going to set. */
spte = spte_addr(cpu, *spgd, vaddr);
/*
* If there was a valid shadow PTE entry here before, we release it.
* This can happen with a write to a previously read-only entry.
*/
release_pte(*spte);
/*
* If this is a write, we insist that the Guest page is writable (the
* final arg to gpte_to_spte()).
*/
if (pte_dirty(gpte))
*spte = gpte_to_spte(cpu, gpte, 1);
else
/*
* If this is a read, don't set the "writable" bit in the page
* table entry, even if the Guest says it's writable. That way
* we will come back here when a write does actually occur, so
* we can update the Guest's _PAGE_DIRTY flag.
*/
set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0));
/*
* Finally, we write the Guest PTE entry back: we've set the
* _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags.
*/
if (likely(!cpu->linear_pages))
lgwrite(cpu, gpte_ptr, pte_t, gpte);
/*
* The fault is fixed, the page table is populated, the mapping
* manipulated, the result returned and the code complete. A small
* delay and a trace of alliteration are the only indications the Guest
* has that a page fault occurred at all.
*/
return true;
}
static void check_gpte(struct lg_cpu *cpu, pte_t gpte)
{
if ((pte_flags(gpte) & _PAGE_PSE) ||
pte_pfn(gpte) >= cpu->lg->pfn_limit)
kill_guest(cpu, "bad page table entry");
}
void __init adjust_boot_vcpu_info(void)
{
unsigned long lpfn, rpfn, lmfn, rmfn;
pte_t *lpte, *rpte;
unsigned int level;
mmu_update_t mmu[2];
/*
* setup_vcpu_info() cannot be used more than once for a given (v)CPU,
* hence we must swap the underlying MFNs of the two pages holding old
* and new vcpu_info of the boot CPU.
*
* Do *not* use __get_cpu_var() or percpu_{write,...}() here, as the per-
* CPU segment didn't get reloaded yet. Using percpu_read(), as in
* arch_use_lazy_mmu_mode(), though undesirable, is safe except for the
* accesses to variables that were updated in setup_percpu_areas().
*/
lpte = lookup_address((unsigned long)&per_cpu_var(vcpu_info)
+ (__per_cpu_load - __per_cpu_start),
&level);
rpte = lookup_address((unsigned long)&per_cpu(vcpu_info, 0), &level);
BUG_ON(!lpte || !(pte_flags(*lpte) & _PAGE_PRESENT));
BUG_ON(!rpte || !(pte_flags(*rpte) & _PAGE_PRESENT));
lmfn = __pte_mfn(*lpte);
rmfn = __pte_mfn(*rpte);
if (lmfn == rmfn)
return;
lpfn = mfn_to_local_pfn(lmfn);
rpfn = mfn_to_local_pfn(rmfn);
printk(KERN_INFO
"Swapping MFNs for PFN %lx and %lx (MFN %lx and %lx)\n",
lpfn, rpfn, lmfn, rmfn);
xen_l1_entry_update(lpte, pfn_pte_ma(rmfn, pte_pgprot(*lpte)));
xen_l1_entry_update(rpte, pfn_pte_ma(lmfn, pte_pgprot(*rpte)));
#ifdef CONFIG_X86_64
if (HYPERVISOR_update_va_mapping((unsigned long)__va(lpfn<<PAGE_SHIFT),
pfn_pte_ma(rmfn, PAGE_KERNEL_RO), 0))
BUG();
#endif
if (HYPERVISOR_update_va_mapping((unsigned long)__va(rpfn<<PAGE_SHIFT),
pfn_pte_ma(lmfn, PAGE_KERNEL),
UVMF_TLB_FLUSH))
BUG();
set_phys_to_machine(lpfn, rmfn);
set_phys_to_machine(rpfn, lmfn);
mmu[0].ptr = ((uint64_t)lmfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE;
mmu[0].val = rpfn;
mmu[1].ptr = ((uint64_t)rmfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE;
mmu[1].val = lpfn;
if (HYPERVISOR_mmu_update(mmu, 2, NULL, DOMID_SELF))
BUG();
/*
* Copy over all contents of the page just replaced, except for the
* vcpu_info itself, as it may have got updated after having been
* copied from __per_cpu_load[].
*/
memcpy(__va(rpfn << PAGE_SHIFT),
__va(lpfn << PAGE_SHIFT),
(unsigned long)&per_cpu_var(vcpu_info) & (PAGE_SIZE - 1));
level = (unsigned long)(&per_cpu_var(vcpu_info) + 1) & (PAGE_SIZE - 1);
if (level)
memcpy(__va(rpfn << PAGE_SHIFT) + level,
__va(lpfn << PAGE_SHIFT) + level,
PAGE_SIZE - level);
}