4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/module.h>
47 #include <linux/gfp.h>
49 #include <asm/pgtable.h>
50 #include <asm/tlbflush.h>
51 #include <asm/fixmap.h>
52 #include <asm/mmu_context.h>
53 #include <asm/setup.h>
54 #include <asm/paravirt.h>
56 #include <asm/linkage.h>
59 #include <asm/xen/hypercall.h>
60 #include <asm/xen/hypervisor.h>
64 #include <xen/interface/xen.h>
65 #include <xen/interface/hvm/hvm_op.h>
66 #include <xen/interface/version.h>
67 #include <xen/interface/memory.h>
68 #include <xen/hvc-console.h>
70 #include "multicalls.h"
74 #define MMU_UPDATE_HISTO 30
77 * Protects atomic reservation decrease/increase against concurrent increases.
78 * Also protects non-atomic updates of current_pages and driver_pages, and
81 DEFINE_SPINLOCK(xen_reservation_lock);
83 #ifdef CONFIG_XEN_DEBUG_FS
87 u32 pgd_update_pinned;
88 u32 pgd_update_batched;
91 u32 pud_update_pinned;
92 u32 pud_update_batched;
95 u32 pmd_update_pinned;
96 u32 pmd_update_batched;
99 u32 pte_update_pinned;
100 u32 pte_update_batched;
103 u32 mmu_update_extended;
104 u32 mmu_update_histo[MMU_UPDATE_HISTO];
107 u32 prot_commit_batched;
110 u32 set_pte_at_batched;
111 u32 set_pte_at_pinned;
112 u32 set_pte_at_current;
113 u32 set_pte_at_kernel;
116 static u8 zero_stats;
118 static inline void check_zero(void)
120 if (unlikely(zero_stats)) {
121 memset(&mmu_stats, 0, sizeof(mmu_stats));
126 #define ADD_STATS(elem, val) \
127 do { check_zero(); mmu_stats.elem += (val); } while(0)
129 #else /* !CONFIG_XEN_DEBUG_FS */
131 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
133 #endif /* CONFIG_XEN_DEBUG_FS */
137 * Identity map, in addition to plain kernel map. This needs to be
138 * large enough to allocate page table pages to allocate the rest.
139 * Each page can map 2MB.
141 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
142 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
145 /* l3 pud for userspace vsyscall mapping */
146 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
147 #endif /* CONFIG_X86_64 */
150 * Note about cr3 (pagetable base) values:
152 * xen_cr3 contains the current logical cr3 value; it contains the
153 * last set cr3. This may not be the current effective cr3, because
154 * its update may be being lazily deferred. However, a vcpu looking
155 * at its own cr3 can use this value knowing that it everything will
156 * be self-consistent.
158 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
159 * hypercall to set the vcpu cr3 is complete (so it may be a little
160 * out of date, but it will never be set early). If one vcpu is
161 * looking at another vcpu's cr3 value, it should use this variable.
163 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
164 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
168 * Just beyond the highest usermode address. STACK_TOP_MAX has a
169 * redzone above it, so round it up to a PGD boundary.
171 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
173 static unsigned long max_p2m_pfn __read_mostly = MAX_DOMAIN_PAGES;
175 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
176 #define TOP_ENTRIES(pages) ((pages) / P2M_ENTRIES_PER_PAGE)
177 #define MAX_TOP_ENTRIES TOP_ENTRIES(MAX_DOMAIN_PAGES)
179 /* Placeholder for holes in the address space */
180 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_ENTRIES_PER_PAGE);
182 /* Array of pointers to pages containing p2m entries */
183 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top, MAX_TOP_ENTRIES);
185 /* Arrays of p2m arrays expressed in mfns used for save/restore */
186 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, MAX_TOP_ENTRIES);
188 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn_list,
189 (MAX_TOP_ENTRIES / P2M_ENTRIES_PER_PAGE));
191 static inline unsigned p2m_top_index(unsigned long pfn)
193 BUG_ON(pfn >= max_p2m_pfn);
194 return pfn / P2M_ENTRIES_PER_PAGE;
197 static inline unsigned p2m_index(unsigned long pfn)
199 return pfn % P2M_ENTRIES_PER_PAGE;
202 /* Build the parallel p2m_top_mfn structures */
203 void xen_build_mfn_list_list(void)
207 for (pfn = 0; pfn < max_p2m_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
208 unsigned topidx = p2m_top_index(pfn);
210 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
214 idx < TOP_ENTRIES(max_p2m_pfn)/P2M_ENTRIES_PER_PAGE;
216 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
217 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
221 void xen_setup_mfn_list_list(void)
223 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
225 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
226 virt_to_mfn(p2m_top_mfn_list);
227 HYPERVISOR_shared_info->arch.max_pfn = max_p2m_mfn;
230 /* Set up p2m_top to point to the domain-builder provided p2m pages */
231 void __init xen_build_dynamic_phys_to_machine(void)
233 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
234 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
238 max_p2m_pfn = max_pfn;
240 p2m_missing = extend_brk(sizeof(*p2m_missing) * P2M_ENTRIES_PER_PAGE,
242 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
243 p2m_missing[i] = ~0UL;
245 p2m_top = extend_brk(sizeof(*p2m_top) * TOP_ENTRIES(max_pfn),
247 for (i = 0; i < TOP_ENTRIES(max_pfn); i++)
248 p2m_top[i] = p2m_missing;
250 p2m_top_mfn = extend_brk(sizeof(*p2m_top_mfn) * TOP_ENTRIES(max_pfn),
252 p2m_top_mfn_list = extend_brk(sizeof(*p2m_top_mfn_list) *
253 (TOP_ENTRIES(max_pfn) / P2M_ENTRIES_PER_PAGE),
256 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
257 unsigned topidx = p2m_top_index(pfn);
259 p2m_top[topidx] = &mfn_list[pfn];
262 xen_build_mfn_list_list();
265 unsigned long get_phys_to_machine(unsigned long pfn)
267 unsigned topidx, idx;
269 if (unlikely(pfn >= max_p2m_pfn))
270 return INVALID_P2M_ENTRY;
272 topidx = p2m_top_index(pfn);
273 idx = p2m_index(pfn);
274 return p2m_top[topidx][idx];
276 EXPORT_SYMBOL_GPL(get_phys_to_machine);
278 /* install a new p2m_top page */
279 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
281 unsigned topidx = p2m_top_index(pfn);
282 unsigned long **pfnp, *mfnp;
285 pfnp = &p2m_top[topidx];
286 mfnp = &p2m_top_mfn[topidx];
288 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
289 p[i] = INVALID_P2M_ENTRY;
291 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
292 *mfnp = virt_to_mfn(p);
299 static void alloc_p2m(unsigned long pfn)
303 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
306 if (!install_p2mtop_page(pfn, p))
307 free_page((unsigned long)p);
310 /* Try to install p2m mapping; fail if intermediate bits missing */
311 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
313 unsigned topidx, idx;
315 if (unlikely(pfn >= max_p2m_pfn)) {
316 BUG_ON(mfn != INVALID_P2M_ENTRY);
320 topidx = p2m_top_index(pfn);
321 if (p2m_top[topidx] == p2m_missing) {
322 if (mfn == INVALID_P2M_ENTRY)
327 idx = p2m_index(pfn);
328 p2m_top[topidx][idx] = mfn;
333 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
335 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
336 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
340 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
343 if (!__set_phys_to_machine(pfn, mfn))
348 unsigned long arbitrary_virt_to_mfn(void *vaddr)
350 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
352 return PFN_DOWN(maddr.maddr);
355 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
357 unsigned long address = (unsigned long)vaddr;
363 * if the PFN is in the linear mapped vaddr range, we can just use
364 * the (quick) virt_to_machine() p2m lookup
366 if (virt_addr_valid(vaddr))
367 return virt_to_machine(vaddr);
369 /* otherwise we have to do a (slower) full page-table walk */
371 pte = lookup_address(address, &level);
373 offset = address & ~PAGE_MASK;
374 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
377 void make_lowmem_page_readonly(void *vaddr)
380 unsigned long address = (unsigned long)vaddr;
383 pte = lookup_address(address, &level);
386 ptev = pte_wrprotect(*pte);
388 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
392 void make_lowmem_page_readwrite(void *vaddr)
395 unsigned long address = (unsigned long)vaddr;
398 pte = lookup_address(address, &level);
401 ptev = pte_mkwrite(*pte);
403 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
408 static bool xen_page_pinned(void *ptr)
410 struct page *page = virt_to_page(ptr);
412 return PagePinned(page);
415 static bool xen_iomap_pte(pte_t pte)
417 return pte_flags(pte) & _PAGE_IOMAP;
420 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
422 struct multicall_space mcs;
423 struct mmu_update *u;
425 mcs = xen_mc_entry(sizeof(*u));
428 /* ptep might be kmapped when using 32-bit HIGHPTE */
429 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
430 u->val = pte_val_ma(pteval);
432 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_IO);
434 xen_mc_issue(PARAVIRT_LAZY_MMU);
437 static void xen_extend_mmu_update(const struct mmu_update *update)
439 struct multicall_space mcs;
440 struct mmu_update *u;
442 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
444 if (mcs.mc != NULL) {
445 ADD_STATS(mmu_update_extended, 1);
446 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
450 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
451 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
453 ADD_STATS(mmu_update_histo[0], 1);
455 ADD_STATS(mmu_update, 1);
456 mcs = __xen_mc_entry(sizeof(*u));
457 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
458 ADD_STATS(mmu_update_histo[1], 1);
465 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
473 /* ptr may be ioremapped for 64-bit pagetable setup */
474 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
475 u.val = pmd_val_ma(val);
476 xen_extend_mmu_update(&u);
478 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
480 xen_mc_issue(PARAVIRT_LAZY_MMU);
485 void xen_set_pmd(pmd_t *ptr, pmd_t val)
487 ADD_STATS(pmd_update, 1);
489 /* If page is not pinned, we can just update the entry
491 if (!xen_page_pinned(ptr)) {
496 ADD_STATS(pmd_update_pinned, 1);
498 xen_set_pmd_hyper(ptr, val);
502 * Associate a virtual page frame with a given physical page frame
503 * and protection flags for that frame.
505 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
507 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
510 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
511 pte_t *ptep, pte_t pteval)
513 if (xen_iomap_pte(pteval)) {
514 xen_set_iomap_pte(ptep, pteval);
518 ADD_STATS(set_pte_at, 1);
519 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
520 ADD_STATS(set_pte_at_current, mm == current->mm);
521 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
523 if (mm == current->mm || mm == &init_mm) {
524 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
525 struct multicall_space mcs;
526 mcs = xen_mc_entry(0);
528 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
529 ADD_STATS(set_pte_at_batched, 1);
530 xen_mc_issue(PARAVIRT_LAZY_MMU);
533 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
536 xen_set_pte(ptep, pteval);
541 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
542 unsigned long addr, pte_t *ptep)
544 /* Just return the pte as-is. We preserve the bits on commit */
548 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
549 pte_t *ptep, pte_t pte)
555 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
556 u.val = pte_val_ma(pte);
557 xen_extend_mmu_update(&u);
559 ADD_STATS(prot_commit, 1);
560 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
562 xen_mc_issue(PARAVIRT_LAZY_MMU);
565 /* Assume pteval_t is equivalent to all the other *val_t types. */
566 static pteval_t pte_mfn_to_pfn(pteval_t val)
568 if (val & _PAGE_PRESENT) {
569 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
570 pteval_t flags = val & PTE_FLAGS_MASK;
571 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
577 static pteval_t pte_pfn_to_mfn(pteval_t val)
579 if (val & _PAGE_PRESENT) {
580 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
581 pteval_t flags = val & PTE_FLAGS_MASK;
582 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
588 static pteval_t iomap_pte(pteval_t val)
590 if (val & _PAGE_PRESENT) {
591 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
592 pteval_t flags = val & PTE_FLAGS_MASK;
594 /* We assume the pte frame number is a MFN, so
595 just use it as-is. */
596 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
602 pteval_t xen_pte_val(pte_t pte)
604 if (xen_initial_domain() && (pte.pte & _PAGE_IOMAP))
607 return pte_mfn_to_pfn(pte.pte);
609 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
611 pgdval_t xen_pgd_val(pgd_t pgd)
613 return pte_mfn_to_pfn(pgd.pgd);
615 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
617 pte_t xen_make_pte(pteval_t pte)
619 phys_addr_t addr = (pte & PTE_PFN_MASK);
622 * Unprivileged domains are allowed to do IOMAPpings for
623 * PCI passthrough, but not map ISA space. The ISA
624 * mappings are just dummy local mappings to keep other
625 * parts of the kernel happy.
627 if (unlikely(pte & _PAGE_IOMAP) &&
628 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
629 pte = iomap_pte(pte);
632 pte = pte_pfn_to_mfn(pte);
635 return native_make_pte(pte);
637 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
639 pgd_t xen_make_pgd(pgdval_t pgd)
641 pgd = pte_pfn_to_mfn(pgd);
642 return native_make_pgd(pgd);
644 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
646 pmdval_t xen_pmd_val(pmd_t pmd)
648 return pte_mfn_to_pfn(pmd.pmd);
650 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
652 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
660 /* ptr may be ioremapped for 64-bit pagetable setup */
661 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
662 u.val = pud_val_ma(val);
663 xen_extend_mmu_update(&u);
665 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
667 xen_mc_issue(PARAVIRT_LAZY_MMU);
672 void xen_set_pud(pud_t *ptr, pud_t val)
674 ADD_STATS(pud_update, 1);
676 /* If page is not pinned, we can just update the entry
678 if (!xen_page_pinned(ptr)) {
683 ADD_STATS(pud_update_pinned, 1);
685 xen_set_pud_hyper(ptr, val);
688 void xen_set_pte(pte_t *ptep, pte_t pte)
690 if (xen_iomap_pte(pte)) {
691 xen_set_iomap_pte(ptep, pte);
695 ADD_STATS(pte_update, 1);
696 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
697 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
699 #ifdef CONFIG_X86_PAE
700 ptep->pte_high = pte.pte_high;
702 ptep->pte_low = pte.pte_low;
708 #ifdef CONFIG_X86_PAE
709 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
711 if (xen_iomap_pte(pte)) {
712 xen_set_iomap_pte(ptep, pte);
716 set_64bit((u64 *)ptep, native_pte_val(pte));
719 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
722 smp_wmb(); /* make sure low gets written first */
726 void xen_pmd_clear(pmd_t *pmdp)
728 set_pmd(pmdp, __pmd(0));
730 #endif /* CONFIG_X86_PAE */
732 pmd_t xen_make_pmd(pmdval_t pmd)
734 pmd = pte_pfn_to_mfn(pmd);
735 return native_make_pmd(pmd);
737 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
739 #if PAGETABLE_LEVELS == 4
740 pudval_t xen_pud_val(pud_t pud)
742 return pte_mfn_to_pfn(pud.pud);
744 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
746 pud_t xen_make_pud(pudval_t pud)
748 pud = pte_pfn_to_mfn(pud);
750 return native_make_pud(pud);
752 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
754 pgd_t *xen_get_user_pgd(pgd_t *pgd)
756 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
757 unsigned offset = pgd - pgd_page;
758 pgd_t *user_ptr = NULL;
760 if (offset < pgd_index(USER_LIMIT)) {
761 struct page *page = virt_to_page(pgd_page);
762 user_ptr = (pgd_t *)page->private;
770 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
774 u.ptr = virt_to_machine(ptr).maddr;
775 u.val = pgd_val_ma(val);
776 xen_extend_mmu_update(&u);
780 * Raw hypercall-based set_pgd, intended for in early boot before
781 * there's a page structure. This implies:
782 * 1. The only existing pagetable is the kernel's
783 * 2. It is always pinned
784 * 3. It has no user pagetable attached to it
786 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
792 __xen_set_pgd_hyper(ptr, val);
794 xen_mc_issue(PARAVIRT_LAZY_MMU);
799 void xen_set_pgd(pgd_t *ptr, pgd_t val)
801 pgd_t *user_ptr = xen_get_user_pgd(ptr);
803 ADD_STATS(pgd_update, 1);
805 /* If page is not pinned, we can just update the entry
807 if (!xen_page_pinned(ptr)) {
810 WARN_ON(xen_page_pinned(user_ptr));
816 ADD_STATS(pgd_update_pinned, 1);
817 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
819 /* If it's pinned, then we can at least batch the kernel and
820 user updates together. */
823 __xen_set_pgd_hyper(ptr, val);
825 __xen_set_pgd_hyper(user_ptr, val);
827 xen_mc_issue(PARAVIRT_LAZY_MMU);
829 #endif /* PAGETABLE_LEVELS == 4 */
832 * (Yet another) pagetable walker. This one is intended for pinning a
833 * pagetable. This means that it walks a pagetable and calls the
834 * callback function on each page it finds making up the page table,
835 * at every level. It walks the entire pagetable, but it only bothers
836 * pinning pte pages which are below limit. In the normal case this
837 * will be STACK_TOP_MAX, but at boot we need to pin up to
840 * For 32-bit the important bit is that we don't pin beyond there,
841 * because then we start getting into Xen's ptes.
843 * For 64-bit, we must skip the Xen hole in the middle of the address
844 * space, just after the big x86-64 virtual hole.
846 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
847 int (*func)(struct mm_struct *mm, struct page *,
852 unsigned hole_low, hole_high;
853 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
854 unsigned pgdidx, pudidx, pmdidx;
856 /* The limit is the last byte to be touched */
858 BUG_ON(limit >= FIXADDR_TOP);
860 if (xen_feature(XENFEAT_auto_translated_physmap))
864 * 64-bit has a great big hole in the middle of the address
865 * space, which contains the Xen mappings. On 32-bit these
866 * will end up making a zero-sized hole and so is a no-op.
868 hole_low = pgd_index(USER_LIMIT);
869 hole_high = pgd_index(PAGE_OFFSET);
871 pgdidx_limit = pgd_index(limit);
873 pudidx_limit = pud_index(limit);
878 pmdidx_limit = pmd_index(limit);
883 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
886 if (pgdidx >= hole_low && pgdidx < hole_high)
889 if (!pgd_val(pgd[pgdidx]))
892 pud = pud_offset(&pgd[pgdidx], 0);
894 if (PTRS_PER_PUD > 1) /* not folded */
895 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
897 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
900 if (pgdidx == pgdidx_limit &&
901 pudidx > pudidx_limit)
904 if (pud_none(pud[pudidx]))
907 pmd = pmd_offset(&pud[pudidx], 0);
909 if (PTRS_PER_PMD > 1) /* not folded */
910 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
912 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
915 if (pgdidx == pgdidx_limit &&
916 pudidx == pudidx_limit &&
917 pmdidx > pmdidx_limit)
920 if (pmd_none(pmd[pmdidx]))
923 pte = pmd_page(pmd[pmdidx]);
924 flush |= (*func)(mm, pte, PT_PTE);
930 /* Do the top level last, so that the callbacks can use it as
931 a cue to do final things like tlb flushes. */
932 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
937 static int xen_pgd_walk(struct mm_struct *mm,
938 int (*func)(struct mm_struct *mm, struct page *,
942 return __xen_pgd_walk(mm, mm->pgd, func, limit);
945 /* If we're using split pte locks, then take the page's lock and
946 return a pointer to it. Otherwise return NULL. */
947 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
949 spinlock_t *ptl = NULL;
951 #if USE_SPLIT_PTLOCKS
952 ptl = __pte_lockptr(page);
953 spin_lock_nest_lock(ptl, &mm->page_table_lock);
959 static void xen_pte_unlock(void *v)
965 static void xen_do_pin(unsigned level, unsigned long pfn)
967 struct mmuext_op *op;
968 struct multicall_space mcs;
970 mcs = __xen_mc_entry(sizeof(*op));
973 op->arg1.mfn = pfn_to_mfn(pfn);
974 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
977 static int xen_pin_page(struct mm_struct *mm, struct page *page,
980 unsigned pgfl = TestSetPagePinned(page);
984 flush = 0; /* already pinned */
985 else if (PageHighMem(page))
986 /* kmaps need flushing if we found an unpinned
990 void *pt = lowmem_page_address(page);
991 unsigned long pfn = page_to_pfn(page);
992 struct multicall_space mcs = __xen_mc_entry(0);
998 * We need to hold the pagetable lock between the time
999 * we make the pagetable RO and when we actually pin
1000 * it. If we don't, then other users may come in and
1001 * attempt to update the pagetable by writing it,
1002 * which will fail because the memory is RO but not
1003 * pinned, so Xen won't do the trap'n'emulate.
1005 * If we're using split pte locks, we can't hold the
1006 * entire pagetable's worth of locks during the
1007 * traverse, because we may wrap the preempt count (8
1008 * bits). The solution is to mark RO and pin each PTE
1009 * page while holding the lock. This means the number
1010 * of locks we end up holding is never more than a
1011 * batch size (~32 entries, at present).
1013 * If we're not using split pte locks, we needn't pin
1014 * the PTE pages independently, because we're
1015 * protected by the overall pagetable lock.
1018 if (level == PT_PTE)
1019 ptl = xen_pte_lock(page, mm);
1021 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1022 pfn_pte(pfn, PAGE_KERNEL_RO),
1023 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1026 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
1028 /* Queue a deferred unlock for when this batch
1030 xen_mc_callback(xen_pte_unlock, ptl);
1037 /* This is called just after a mm has been created, but it has not
1038 been used yet. We need to make sure that its pagetable is all
1039 read-only, and can be pinned. */
1040 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
1044 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
1045 /* re-enable interrupts for flushing */
1048 kmap_flush_unused();
1053 #ifdef CONFIG_X86_64
1055 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1057 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
1060 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
1061 xen_do_pin(MMUEXT_PIN_L4_TABLE,
1062 PFN_DOWN(__pa(user_pgd)));
1065 #else /* CONFIG_X86_32 */
1066 #ifdef CONFIG_X86_PAE
1067 /* Need to make sure unshared kernel PMD is pinnable */
1068 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1071 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
1072 #endif /* CONFIG_X86_64 */
1076 static void xen_pgd_pin(struct mm_struct *mm)
1078 __xen_pgd_pin(mm, mm->pgd);
1082 * On save, we need to pin all pagetables to make sure they get their
1083 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1084 * them (unpinned pgds are not currently in use, probably because the
1085 * process is under construction or destruction).
1087 * Expected to be called in stop_machine() ("equivalent to taking
1088 * every spinlock in the system"), so the locking doesn't really
1089 * matter all that much.
1091 void xen_mm_pin_all(void)
1093 unsigned long flags;
1096 spin_lock_irqsave(&pgd_lock, flags);
1098 list_for_each_entry(page, &pgd_list, lru) {
1099 if (!PagePinned(page)) {
1100 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1101 SetPageSavePinned(page);
1105 spin_unlock_irqrestore(&pgd_lock, flags);
1109 * The init_mm pagetable is really pinned as soon as its created, but
1110 * that's before we have page structures to store the bits. So do all
1111 * the book-keeping now.
1113 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1114 enum pt_level level)
1116 SetPagePinned(page);
1120 static void __init xen_mark_init_mm_pinned(void)
1122 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1125 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1126 enum pt_level level)
1128 unsigned pgfl = TestClearPagePinned(page);
1130 if (pgfl && !PageHighMem(page)) {
1131 void *pt = lowmem_page_address(page);
1132 unsigned long pfn = page_to_pfn(page);
1133 spinlock_t *ptl = NULL;
1134 struct multicall_space mcs;
1137 * Do the converse to pin_page. If we're using split
1138 * pte locks, we must be holding the lock for while
1139 * the pte page is unpinned but still RO to prevent
1140 * concurrent updates from seeing it in this
1141 * partially-pinned state.
1143 if (level == PT_PTE) {
1144 ptl = xen_pte_lock(page, mm);
1147 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1150 mcs = __xen_mc_entry(0);
1152 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1153 pfn_pte(pfn, PAGE_KERNEL),
1154 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1157 /* unlock when batch completed */
1158 xen_mc_callback(xen_pte_unlock, ptl);
1162 return 0; /* never need to flush on unpin */
1165 /* Release a pagetables pages back as normal RW */
1166 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1170 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1172 #ifdef CONFIG_X86_64
1174 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1177 xen_do_pin(MMUEXT_UNPIN_TABLE,
1178 PFN_DOWN(__pa(user_pgd)));
1179 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1184 #ifdef CONFIG_X86_PAE
1185 /* Need to make sure unshared kernel PMD is unpinned */
1186 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1190 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1195 static void xen_pgd_unpin(struct mm_struct *mm)
1197 __xen_pgd_unpin(mm, mm->pgd);
1201 * On resume, undo any pinning done at save, so that the rest of the
1202 * kernel doesn't see any unexpected pinned pagetables.
1204 void xen_mm_unpin_all(void)
1206 unsigned long flags;
1209 spin_lock_irqsave(&pgd_lock, flags);
1211 list_for_each_entry(page, &pgd_list, lru) {
1212 if (PageSavePinned(page)) {
1213 BUG_ON(!PagePinned(page));
1214 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1215 ClearPageSavePinned(page);
1219 spin_unlock_irqrestore(&pgd_lock, flags);
1222 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1224 spin_lock(&next->page_table_lock);
1226 spin_unlock(&next->page_table_lock);
1229 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1231 spin_lock(&mm->page_table_lock);
1233 spin_unlock(&mm->page_table_lock);
1238 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1239 we need to repoint it somewhere else before we can unpin it. */
1240 static void drop_other_mm_ref(void *info)
1242 struct mm_struct *mm = info;
1243 struct mm_struct *active_mm;
1245 active_mm = percpu_read(cpu_tlbstate.active_mm);
1247 if (active_mm == mm)
1248 leave_mm(smp_processor_id());
1250 /* If this cpu still has a stale cr3 reference, then make sure
1251 it has been flushed. */
1252 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1253 load_cr3(swapper_pg_dir);
1256 static void xen_drop_mm_ref(struct mm_struct *mm)
1261 if (current->active_mm == mm) {
1262 if (current->mm == mm)
1263 load_cr3(swapper_pg_dir);
1265 leave_mm(smp_processor_id());
1268 /* Get the "official" set of cpus referring to our pagetable. */
1269 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1270 for_each_online_cpu(cpu) {
1271 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1272 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1274 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1278 cpumask_copy(mask, mm_cpumask(mm));
1280 /* It's possible that a vcpu may have a stale reference to our
1281 cr3, because its in lazy mode, and it hasn't yet flushed
1282 its set of pending hypercalls yet. In this case, we can
1283 look at its actual current cr3 value, and force it to flush
1285 for_each_online_cpu(cpu) {
1286 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1287 cpumask_set_cpu(cpu, mask);
1290 if (!cpumask_empty(mask))
1291 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1292 free_cpumask_var(mask);
1295 static void xen_drop_mm_ref(struct mm_struct *mm)
1297 if (current->active_mm == mm)
1298 load_cr3(swapper_pg_dir);
1303 * While a process runs, Xen pins its pagetables, which means that the
1304 * hypervisor forces it to be read-only, and it controls all updates
1305 * to it. This means that all pagetable updates have to go via the
1306 * hypervisor, which is moderately expensive.
1308 * Since we're pulling the pagetable down, we switch to use init_mm,
1309 * unpin old process pagetable and mark it all read-write, which
1310 * allows further operations on it to be simple memory accesses.
1312 * The only subtle point is that another CPU may be still using the
1313 * pagetable because of lazy tlb flushing. This means we need need to
1314 * switch all CPUs off this pagetable before we can unpin it.
1316 void xen_exit_mmap(struct mm_struct *mm)
1318 get_cpu(); /* make sure we don't move around */
1319 xen_drop_mm_ref(mm);
1322 spin_lock(&mm->page_table_lock);
1324 /* pgd may not be pinned in the error exit path of execve */
1325 if (xen_page_pinned(mm->pgd))
1328 spin_unlock(&mm->page_table_lock);
1331 static __init void xen_pagetable_setup_start(pgd_t *base)
1335 static void xen_post_allocator_init(void);
1337 static __init void xen_pagetable_setup_done(pgd_t *base)
1339 xen_setup_shared_info();
1340 xen_post_allocator_init();
1343 static void xen_write_cr2(unsigned long cr2)
1345 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1348 static unsigned long xen_read_cr2(void)
1350 return percpu_read(xen_vcpu)->arch.cr2;
1353 unsigned long xen_read_cr2_direct(void)
1355 return percpu_read(xen_vcpu_info.arch.cr2);
1358 static void xen_flush_tlb(void)
1360 struct mmuext_op *op;
1361 struct multicall_space mcs;
1365 mcs = xen_mc_entry(sizeof(*op));
1368 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1369 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1371 xen_mc_issue(PARAVIRT_LAZY_MMU);
1376 static void xen_flush_tlb_single(unsigned long addr)
1378 struct mmuext_op *op;
1379 struct multicall_space mcs;
1383 mcs = xen_mc_entry(sizeof(*op));
1385 op->cmd = MMUEXT_INVLPG_LOCAL;
1386 op->arg1.linear_addr = addr & PAGE_MASK;
1387 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1389 xen_mc_issue(PARAVIRT_LAZY_MMU);
1394 static void xen_flush_tlb_others(const struct cpumask *cpus,
1395 struct mm_struct *mm, unsigned long va)
1398 struct mmuext_op op;
1399 DECLARE_BITMAP(mask, NR_CPUS);
1401 struct multicall_space mcs;
1403 if (cpumask_empty(cpus))
1404 return; /* nothing to do */
1406 mcs = xen_mc_entry(sizeof(*args));
1408 args->op.arg2.vcpumask = to_cpumask(args->mask);
1410 /* Remove us, and any offline CPUS. */
1411 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1412 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1414 if (va == TLB_FLUSH_ALL) {
1415 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1417 args->op.cmd = MMUEXT_INVLPG_MULTI;
1418 args->op.arg1.linear_addr = va;
1421 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1423 xen_mc_issue(PARAVIRT_LAZY_MMU);
1426 static unsigned long xen_read_cr3(void)
1428 return percpu_read(xen_cr3);
1431 static void set_current_cr3(void *v)
1433 percpu_write(xen_current_cr3, (unsigned long)v);
1436 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1438 struct mmuext_op *op;
1439 struct multicall_space mcs;
1443 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1447 WARN_ON(mfn == 0 && kernel);
1449 mcs = __xen_mc_entry(sizeof(*op));
1452 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1455 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1458 percpu_write(xen_cr3, cr3);
1460 /* Update xen_current_cr3 once the batch has actually
1462 xen_mc_callback(set_current_cr3, (void *)cr3);
1466 static void xen_write_cr3(unsigned long cr3)
1468 BUG_ON(preemptible());
1470 xen_mc_batch(); /* disables interrupts */
1472 /* Update while interrupts are disabled, so its atomic with
1474 percpu_write(xen_cr3, cr3);
1476 __xen_write_cr3(true, cr3);
1478 #ifdef CONFIG_X86_64
1480 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1482 __xen_write_cr3(false, __pa(user_pgd));
1484 __xen_write_cr3(false, 0);
1488 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1491 static int xen_pgd_alloc(struct mm_struct *mm)
1493 pgd_t *pgd = mm->pgd;
1496 BUG_ON(PagePinned(virt_to_page(pgd)));
1498 #ifdef CONFIG_X86_64
1500 struct page *page = virt_to_page(pgd);
1503 BUG_ON(page->private != 0);
1507 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1508 page->private = (unsigned long)user_pgd;
1510 if (user_pgd != NULL) {
1511 user_pgd[pgd_index(VSYSCALL_START)] =
1512 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1516 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1523 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1525 #ifdef CONFIG_X86_64
1526 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1529 free_page((unsigned long)user_pgd);
1533 #ifdef CONFIG_X86_32
1534 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1536 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1537 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1538 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1544 /* Init-time set_pte while constructing initial pagetables, which
1545 doesn't allow RO pagetable pages to be remapped RW */
1546 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1548 pte = mask_rw_pte(ptep, pte);
1550 xen_set_pte(ptep, pte);
1554 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1556 struct mmuext_op op;
1558 op.arg1.mfn = pfn_to_mfn(pfn);
1559 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1563 /* Early in boot, while setting up the initial pagetable, assume
1564 everything is pinned. */
1565 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1567 #ifdef CONFIG_FLATMEM
1568 BUG_ON(mem_map); /* should only be used early */
1570 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1571 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1574 /* Used for pmd and pud */
1575 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1577 #ifdef CONFIG_FLATMEM
1578 BUG_ON(mem_map); /* should only be used early */
1580 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1583 /* Early release_pte assumes that all pts are pinned, since there's
1584 only init_mm and anything attached to that is pinned. */
1585 static __init void xen_release_pte_init(unsigned long pfn)
1587 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1588 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1591 static __init void xen_release_pmd_init(unsigned long pfn)
1593 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1596 /* This needs to make sure the new pte page is pinned iff its being
1597 attached to a pinned pagetable. */
1598 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1600 struct page *page = pfn_to_page(pfn);
1602 if (PagePinned(virt_to_page(mm->pgd))) {
1603 SetPagePinned(page);
1605 if (!PageHighMem(page)) {
1606 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1607 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1608 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1610 /* make sure there are no stray mappings of
1612 kmap_flush_unused();
1617 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1619 xen_alloc_ptpage(mm, pfn, PT_PTE);
1622 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1624 xen_alloc_ptpage(mm, pfn, PT_PMD);
1627 /* This should never happen until we're OK to use struct page */
1628 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1630 struct page *page = pfn_to_page(pfn);
1632 if (PagePinned(page)) {
1633 if (!PageHighMem(page)) {
1634 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1635 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1636 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1638 ClearPagePinned(page);
1642 static void xen_release_pte(unsigned long pfn)
1644 xen_release_ptpage(pfn, PT_PTE);
1647 static void xen_release_pmd(unsigned long pfn)
1649 xen_release_ptpage(pfn, PT_PMD);
1652 #if PAGETABLE_LEVELS == 4
1653 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1655 xen_alloc_ptpage(mm, pfn, PT_PUD);
1658 static void xen_release_pud(unsigned long pfn)
1660 xen_release_ptpage(pfn, PT_PUD);
1664 void __init xen_reserve_top(void)
1666 #ifdef CONFIG_X86_32
1667 unsigned long top = HYPERVISOR_VIRT_START;
1668 struct xen_platform_parameters pp;
1670 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1671 top = pp.virt_start;
1673 reserve_top_address(-top);
1674 #endif /* CONFIG_X86_32 */
1678 * Like __va(), but returns address in the kernel mapping (which is
1679 * all we have until the physical memory mapping has been set up.
1681 static void *__ka(phys_addr_t paddr)
1683 #ifdef CONFIG_X86_64
1684 return (void *)(paddr + __START_KERNEL_map);
1690 /* Convert a machine address to physical address */
1691 static unsigned long m2p(phys_addr_t maddr)
1695 maddr &= PTE_PFN_MASK;
1696 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1701 /* Convert a machine address to kernel virtual */
1702 static void *m2v(phys_addr_t maddr)
1704 return __ka(m2p(maddr));
1707 static void set_page_prot(void *addr, pgprot_t prot)
1709 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1710 pte_t pte = pfn_pte(pfn, prot);
1712 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1716 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1718 unsigned pmdidx, pteidx;
1722 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1727 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1730 /* Reuse or allocate a page of ptes */
1731 if (pmd_present(pmd[pmdidx]))
1732 pte_page = m2v(pmd[pmdidx].pmd);
1734 /* Check for free pte pages */
1735 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1738 pte_page = &level1_ident_pgt[ident_pte];
1739 ident_pte += PTRS_PER_PTE;
1741 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1744 /* Install mappings */
1745 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1748 if (pfn > max_pfn_mapped)
1749 max_pfn_mapped = pfn;
1751 if (!pte_none(pte_page[pteidx]))
1754 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1755 pte_page[pteidx] = pte;
1759 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1760 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1762 set_page_prot(pmd, PAGE_KERNEL_RO);
1765 #ifdef CONFIG_X86_64
1766 static void convert_pfn_mfn(void *v)
1771 /* All levels are converted the same way, so just treat them
1773 for (i = 0; i < PTRS_PER_PTE; i++)
1774 pte[i] = xen_make_pte(pte[i].pte);
1778 * Set up the inital kernel pagetable.
1780 * We can construct this by grafting the Xen provided pagetable into
1781 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1782 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1783 * means that only the kernel has a physical mapping to start with -
1784 * but that's enough to get __va working. We need to fill in the rest
1785 * of the physical mapping once some sort of allocator has been set
1788 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1789 unsigned long max_pfn)
1794 /* Zap identity mapping */
1795 init_level4_pgt[0] = __pgd(0);
1797 /* Pre-constructed entries are in pfn, so convert to mfn */
1798 convert_pfn_mfn(init_level4_pgt);
1799 convert_pfn_mfn(level3_ident_pgt);
1800 convert_pfn_mfn(level3_kernel_pgt);
1802 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1803 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1805 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1806 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1808 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1809 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1810 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1812 /* Set up identity map */
1813 xen_map_identity_early(level2_ident_pgt, max_pfn);
1815 /* Make pagetable pieces RO */
1816 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1817 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1818 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1819 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1820 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1821 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1823 /* Pin down new L4 */
1824 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1825 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1827 /* Unpin Xen-provided one */
1828 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1831 pgd = init_level4_pgt;
1834 * At this stage there can be no user pgd, and no page
1835 * structure to attach it to, so make sure we just set kernel
1839 __xen_write_cr3(true, __pa(pgd));
1840 xen_mc_issue(PARAVIRT_LAZY_CPU);
1842 reserve_early(__pa(xen_start_info->pt_base),
1843 __pa(xen_start_info->pt_base +
1844 xen_start_info->nr_pt_frames * PAGE_SIZE),
1849 #else /* !CONFIG_X86_64 */
1850 static RESERVE_BRK_ARRAY(pmd_t, level2_kernel_pgt, PTRS_PER_PMD);
1852 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1853 unsigned long max_pfn)
1857 level2_kernel_pgt = extend_brk(sizeof(pmd_t *) * PTRS_PER_PMD, PAGE_SIZE);
1859 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1860 xen_start_info->nr_pt_frames * PAGE_SIZE +
1863 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1864 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1866 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1868 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1869 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1870 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1872 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1873 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1874 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1876 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1878 xen_write_cr3(__pa(swapper_pg_dir));
1880 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1882 reserve_early(__pa(xen_start_info->pt_base),
1883 __pa(xen_start_info->pt_base +
1884 xen_start_info->nr_pt_frames * PAGE_SIZE),
1887 return swapper_pg_dir;
1889 #endif /* CONFIG_X86_64 */
1891 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1895 phys >>= PAGE_SHIFT;
1898 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1899 #ifdef CONFIG_X86_F00F_BUG
1902 #ifdef CONFIG_X86_32
1905 # ifdef CONFIG_HIGHMEM
1906 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1909 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1911 #ifdef CONFIG_X86_LOCAL_APIC
1912 case FIX_APIC_BASE: /* maps dummy local APIC */
1914 case FIX_TEXT_POKE0:
1915 case FIX_TEXT_POKE1:
1916 /* All local page mappings */
1917 pte = pfn_pte(phys, prot);
1920 case FIX_PARAVIRT_BOOTMAP:
1921 /* This is an MFN, but it isn't an IO mapping from the
1923 pte = mfn_pte(phys, prot);
1927 /* By default, set_fixmap is used for hardware mappings */
1928 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1932 __native_set_fixmap(idx, pte);
1934 #ifdef CONFIG_X86_64
1935 /* Replicate changes to map the vsyscall page into the user
1936 pagetable vsyscall mapping. */
1937 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1938 unsigned long vaddr = __fix_to_virt(idx);
1939 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1944 static __init void xen_post_allocator_init(void)
1946 pv_mmu_ops.set_pte = xen_set_pte;
1947 pv_mmu_ops.set_pmd = xen_set_pmd;
1948 pv_mmu_ops.set_pud = xen_set_pud;
1949 #if PAGETABLE_LEVELS == 4
1950 pv_mmu_ops.set_pgd = xen_set_pgd;
1953 /* This will work as long as patching hasn't happened yet
1954 (which it hasn't) */
1955 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1956 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1957 pv_mmu_ops.release_pte = xen_release_pte;
1958 pv_mmu_ops.release_pmd = xen_release_pmd;
1959 #if PAGETABLE_LEVELS == 4
1960 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1961 pv_mmu_ops.release_pud = xen_release_pud;
1964 #ifdef CONFIG_X86_64
1965 SetPagePinned(virt_to_page(level3_user_vsyscall));
1967 xen_mark_init_mm_pinned();
1970 static void xen_leave_lazy_mmu(void)
1974 paravirt_leave_lazy_mmu();
1978 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1979 .read_cr2 = xen_read_cr2,
1980 .write_cr2 = xen_write_cr2,
1982 .read_cr3 = xen_read_cr3,
1983 .write_cr3 = xen_write_cr3,
1985 .flush_tlb_user = xen_flush_tlb,
1986 .flush_tlb_kernel = xen_flush_tlb,
1987 .flush_tlb_single = xen_flush_tlb_single,
1988 .flush_tlb_others = xen_flush_tlb_others,
1990 .pte_update = paravirt_nop,
1991 .pte_update_defer = paravirt_nop,
1993 .pgd_alloc = xen_pgd_alloc,
1994 .pgd_free = xen_pgd_free,
1996 .alloc_pte = xen_alloc_pte_init,
1997 .release_pte = xen_release_pte_init,
1998 .alloc_pmd = xen_alloc_pmd_init,
1999 .alloc_pmd_clone = paravirt_nop,
2000 .release_pmd = xen_release_pmd_init,
2002 #ifdef CONFIG_X86_64
2003 .set_pte = xen_set_pte,
2005 .set_pte = xen_set_pte_init,
2007 .set_pte_at = xen_set_pte_at,
2008 .set_pmd = xen_set_pmd_hyper,
2010 .ptep_modify_prot_start = __ptep_modify_prot_start,
2011 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2013 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2014 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2016 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2017 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2019 #ifdef CONFIG_X86_PAE
2020 .set_pte_atomic = xen_set_pte_atomic,
2021 .pte_clear = xen_pte_clear,
2022 .pmd_clear = xen_pmd_clear,
2023 #endif /* CONFIG_X86_PAE */
2024 .set_pud = xen_set_pud_hyper,
2026 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2027 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2029 #if PAGETABLE_LEVELS == 4
2030 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2031 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2032 .set_pgd = xen_set_pgd_hyper,
2034 .alloc_pud = xen_alloc_pmd_init,
2035 .release_pud = xen_release_pmd_init,
2036 #endif /* PAGETABLE_LEVELS == 4 */
2038 .activate_mm = xen_activate_mm,
2039 .dup_mmap = xen_dup_mmap,
2040 .exit_mmap = xen_exit_mmap,
2043 .enter = paravirt_enter_lazy_mmu,
2044 .leave = xen_leave_lazy_mmu,
2047 .set_fixmap = xen_set_fixmap,
2050 void __init xen_init_mmu_ops(void)
2052 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2053 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2054 pv_mmu_ops = xen_mmu_ops;
2056 vmap_lazy_unmap = false;
2059 /* Protected by xen_reservation_lock. */
2060 #define MAX_CONTIG_ORDER 9 /* 2MB */
2061 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2063 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2064 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2065 unsigned long *in_frames,
2066 unsigned long *out_frames)
2069 struct multicall_space mcs;
2072 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2073 mcs = __xen_mc_entry(0);
2076 in_frames[i] = virt_to_mfn(vaddr);
2078 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2079 set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2082 out_frames[i] = virt_to_pfn(vaddr);
2088 * Update the pfn-to-mfn mappings for a virtual address range, either to
2089 * point to an array of mfns, or contiguously from a single starting
2092 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2093 unsigned long *mfns,
2094 unsigned long first_mfn)
2101 limit = 1u << order;
2102 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2103 struct multicall_space mcs;
2106 mcs = __xen_mc_entry(0);
2110 mfn = first_mfn + i;
2112 if (i < (limit - 1))
2116 flags = UVMF_INVLPG | UVMF_ALL;
2118 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2121 MULTI_update_va_mapping(mcs.mc, vaddr,
2122 mfn_pte(mfn, PAGE_KERNEL), flags);
2124 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2131 * Perform the hypercall to exchange a region of our pfns to point to
2132 * memory with the required contiguous alignment. Takes the pfns as
2133 * input, and populates mfns as output.
2135 * Returns a success code indicating whether the hypervisor was able to
2136 * satisfy the request or not.
2138 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2139 unsigned long *pfns_in,
2140 unsigned long extents_out,
2141 unsigned int order_out,
2142 unsigned long *mfns_out,
2143 unsigned int address_bits)
2148 struct xen_memory_exchange exchange = {
2150 .nr_extents = extents_in,
2151 .extent_order = order_in,
2152 .extent_start = pfns_in,
2156 .nr_extents = extents_out,
2157 .extent_order = order_out,
2158 .extent_start = mfns_out,
2159 .address_bits = address_bits,
2164 BUG_ON(extents_in << order_in != extents_out << order_out);
2166 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2167 success = (exchange.nr_exchanged == extents_in);
2169 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2170 BUG_ON(success && (rc != 0));
2175 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2176 unsigned int address_bits)
2178 unsigned long *in_frames = discontig_frames, out_frame;
2179 unsigned long flags;
2183 * Currently an auto-translated guest will not perform I/O, nor will
2184 * it require PAE page directories below 4GB. Therefore any calls to
2185 * this function are redundant and can be ignored.
2188 if (xen_feature(XENFEAT_auto_translated_physmap))
2191 if (unlikely(order > MAX_CONTIG_ORDER))
2194 memset((void *) vstart, 0, PAGE_SIZE << order);
2196 spin_lock_irqsave(&xen_reservation_lock, flags);
2198 /* 1. Zap current PTEs, remembering MFNs. */
2199 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2201 /* 2. Get a new contiguous memory extent. */
2202 out_frame = virt_to_pfn(vstart);
2203 success = xen_exchange_memory(1UL << order, 0, in_frames,
2204 1, order, &out_frame,
2207 /* 3. Map the new extent in place of old pages. */
2209 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2211 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2213 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2215 return success ? 0 : -ENOMEM;
2217 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2219 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2221 unsigned long *out_frames = discontig_frames, in_frame;
2222 unsigned long flags;
2225 if (xen_feature(XENFEAT_auto_translated_physmap))
2228 if (unlikely(order > MAX_CONTIG_ORDER))
2231 memset((void *) vstart, 0, PAGE_SIZE << order);
2233 spin_lock_irqsave(&xen_reservation_lock, flags);
2235 /* 1. Find start MFN of contiguous extent. */
2236 in_frame = virt_to_mfn(vstart);
2238 /* 2. Zap current PTEs. */
2239 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2241 /* 3. Do the exchange for non-contiguous MFNs. */
2242 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2245 /* 4. Map new pages in place of old pages. */
2247 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2249 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2251 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2253 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2255 #ifdef CONFIG_XEN_PVHVM
2256 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2258 struct xen_hvm_pagetable_dying a;
2261 a.domid = DOMID_SELF;
2262 a.gpa = __pa(mm->pgd);
2263 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2264 WARN_ON_ONCE(rc < 0);
2267 static int is_pagetable_dying_supported(void)
2269 struct xen_hvm_pagetable_dying a;
2272 a.domid = DOMID_SELF;
2274 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2276 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2282 void __init xen_hvm_init_mmu_ops(void)
2284 if (is_pagetable_dying_supported())
2285 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2289 #ifdef CONFIG_XEN_DEBUG_FS
2291 static struct dentry *d_mmu_debug;
2293 static int __init xen_mmu_debugfs(void)
2295 struct dentry *d_xen = xen_init_debugfs();
2300 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2302 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2304 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2305 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2306 &mmu_stats.pgd_update_pinned);
2307 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2308 &mmu_stats.pgd_update_pinned);
2310 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2311 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2312 &mmu_stats.pud_update_pinned);
2313 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2314 &mmu_stats.pud_update_pinned);
2316 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2317 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2318 &mmu_stats.pmd_update_pinned);
2319 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2320 &mmu_stats.pmd_update_pinned);
2322 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2323 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2324 // &mmu_stats.pte_update_pinned);
2325 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2326 &mmu_stats.pte_update_pinned);
2328 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2329 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2330 &mmu_stats.mmu_update_extended);
2331 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2332 mmu_stats.mmu_update_histo, 20);
2334 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2335 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2336 &mmu_stats.set_pte_at_batched);
2337 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2338 &mmu_stats.set_pte_at_current);
2339 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2340 &mmu_stats.set_pte_at_kernel);
2342 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2343 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2344 &mmu_stats.prot_commit_batched);
2348 fs_initcall(xen_mmu_debugfs);
2350 #endif /* CONFIG_XEN_DEBUG_FS */