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 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
144 /* l3 pud for userspace vsyscall mapping */
145 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
146 #endif /* CONFIG_X86_64 */
149 * Note about cr3 (pagetable base) values:
151 * xen_cr3 contains the current logical cr3 value; it contains the
152 * last set cr3. This may not be the current effective cr3, because
153 * its update may be being lazily deferred. However, a vcpu looking
154 * at its own cr3 can use this value knowing that it everything will
155 * be self-consistent.
157 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
158 * hypercall to set the vcpu cr3 is complete (so it may be a little
159 * out of date, but it will never be set early). If one vcpu is
160 * looking at another vcpu's cr3 value, it should use this variable.
162 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
163 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
167 * Just beyond the highest usermode address. STACK_TOP_MAX has a
168 * redzone above it, so round it up to a PGD boundary.
170 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
173 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
174 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
176 /* Placeholder for holes in the address space */
177 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
178 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
180 /* Array of pointers to pages containing p2m entries */
181 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
182 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
184 /* Arrays of p2m arrays expressed in mfns used for save/restore */
185 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
187 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
190 static inline unsigned p2m_top_index(unsigned long pfn)
192 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
193 return pfn / P2M_ENTRIES_PER_PAGE;
196 static inline unsigned p2m_index(unsigned long pfn)
198 return pfn % P2M_ENTRIES_PER_PAGE;
201 /* Build the parallel p2m_top_mfn structures */
202 void xen_build_mfn_list_list(void)
206 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
207 unsigned topidx = p2m_top_index(pfn);
209 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
212 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
213 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
214 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
218 void xen_setup_mfn_list_list(void)
220 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
222 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
223 virt_to_mfn(p2m_top_mfn_list);
224 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
227 /* Set up p2m_top to point to the domain-builder provided p2m pages */
228 void __init xen_build_dynamic_phys_to_machine(void)
230 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
231 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
234 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
235 unsigned topidx = p2m_top_index(pfn);
237 p2m_top[topidx] = &mfn_list[pfn];
240 xen_build_mfn_list_list();
243 unsigned long get_phys_to_machine(unsigned long pfn)
245 unsigned topidx, idx;
247 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
248 return INVALID_P2M_ENTRY;
250 topidx = p2m_top_index(pfn);
251 idx = p2m_index(pfn);
252 return p2m_top[topidx][idx];
254 EXPORT_SYMBOL_GPL(get_phys_to_machine);
256 /* install a new p2m_top page */
257 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
259 unsigned topidx = p2m_top_index(pfn);
260 unsigned long **pfnp, *mfnp;
263 pfnp = &p2m_top[topidx];
264 mfnp = &p2m_top_mfn[topidx];
266 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
267 p[i] = INVALID_P2M_ENTRY;
269 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
270 *mfnp = virt_to_mfn(p);
277 static void alloc_p2m(unsigned long pfn)
281 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
284 if (!install_p2mtop_page(pfn, p))
285 free_page((unsigned long)p);
288 /* Try to install p2m mapping; fail if intermediate bits missing */
289 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
291 unsigned topidx, idx;
293 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
294 BUG_ON(mfn != INVALID_P2M_ENTRY);
298 topidx = p2m_top_index(pfn);
299 if (p2m_top[topidx] == p2m_missing) {
300 if (mfn == INVALID_P2M_ENTRY)
305 idx = p2m_index(pfn);
306 p2m_top[topidx][idx] = mfn;
311 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
313 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
314 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
318 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
321 if (!__set_phys_to_machine(pfn, mfn))
326 unsigned long arbitrary_virt_to_mfn(void *vaddr)
328 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
330 return PFN_DOWN(maddr.maddr);
333 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
335 unsigned long address = (unsigned long)vaddr;
341 * if the PFN is in the linear mapped vaddr range, we can just use
342 * the (quick) virt_to_machine() p2m lookup
344 if (virt_addr_valid(vaddr))
345 return virt_to_machine(vaddr);
347 /* otherwise we have to do a (slower) full page-table walk */
349 pte = lookup_address(address, &level);
351 offset = address & ~PAGE_MASK;
352 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
355 void make_lowmem_page_readonly(void *vaddr)
358 unsigned long address = (unsigned long)vaddr;
361 pte = lookup_address(address, &level);
364 ptev = pte_wrprotect(*pte);
366 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
370 void make_lowmem_page_readwrite(void *vaddr)
373 unsigned long address = (unsigned long)vaddr;
376 pte = lookup_address(address, &level);
379 ptev = pte_mkwrite(*pte);
381 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
386 static bool xen_page_pinned(void *ptr)
388 struct page *page = virt_to_page(ptr);
390 return PagePinned(page);
393 static bool xen_iomap_pte(pte_t pte)
395 return pte_flags(pte) & _PAGE_IOMAP;
398 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
400 struct multicall_space mcs;
401 struct mmu_update *u;
403 mcs = xen_mc_entry(sizeof(*u));
406 /* ptep might be kmapped when using 32-bit HIGHPTE */
407 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
408 u->val = pte_val_ma(pteval);
410 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
412 xen_mc_issue(PARAVIRT_LAZY_MMU);
414 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
416 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
418 xen_set_domain_pte(ptep, pteval, DOMID_IO);
421 static void xen_extend_mmu_update(const struct mmu_update *update)
423 struct multicall_space mcs;
424 struct mmu_update *u;
426 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
428 if (mcs.mc != NULL) {
429 ADD_STATS(mmu_update_extended, 1);
430 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
434 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
435 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
437 ADD_STATS(mmu_update_histo[0], 1);
439 ADD_STATS(mmu_update, 1);
440 mcs = __xen_mc_entry(sizeof(*u));
441 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
442 ADD_STATS(mmu_update_histo[1], 1);
449 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
457 /* ptr may be ioremapped for 64-bit pagetable setup */
458 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
459 u.val = pmd_val_ma(val);
460 xen_extend_mmu_update(&u);
462 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
464 xen_mc_issue(PARAVIRT_LAZY_MMU);
469 void xen_set_pmd(pmd_t *ptr, pmd_t val)
471 ADD_STATS(pmd_update, 1);
473 /* If page is not pinned, we can just update the entry
475 if (!xen_page_pinned(ptr)) {
480 ADD_STATS(pmd_update_pinned, 1);
482 xen_set_pmd_hyper(ptr, val);
486 * Associate a virtual page frame with a given physical page frame
487 * and protection flags for that frame.
489 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
491 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
494 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
495 pte_t *ptep, pte_t pteval)
497 if (xen_iomap_pte(pteval)) {
498 xen_set_iomap_pte(ptep, pteval);
502 ADD_STATS(set_pte_at, 1);
503 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
504 ADD_STATS(set_pte_at_current, mm == current->mm);
505 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
507 if (mm == current->mm || mm == &init_mm) {
508 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
509 struct multicall_space mcs;
510 mcs = xen_mc_entry(0);
512 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
513 ADD_STATS(set_pte_at_batched, 1);
514 xen_mc_issue(PARAVIRT_LAZY_MMU);
517 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
520 xen_set_pte(ptep, pteval);
525 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
526 unsigned long addr, pte_t *ptep)
528 /* Just return the pte as-is. We preserve the bits on commit */
532 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
533 pte_t *ptep, pte_t pte)
539 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
540 u.val = pte_val_ma(pte);
541 xen_extend_mmu_update(&u);
543 ADD_STATS(prot_commit, 1);
544 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
546 xen_mc_issue(PARAVIRT_LAZY_MMU);
549 /* Assume pteval_t is equivalent to all the other *val_t types. */
550 static pteval_t pte_mfn_to_pfn(pteval_t val)
552 if (val & _PAGE_PRESENT) {
553 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
554 pteval_t flags = val & PTE_FLAGS_MASK;
555 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
561 static pteval_t pte_pfn_to_mfn(pteval_t val)
563 if (val & _PAGE_PRESENT) {
564 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
565 pteval_t flags = val & PTE_FLAGS_MASK;
566 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
572 static pteval_t iomap_pte(pteval_t val)
574 if (val & _PAGE_PRESENT) {
575 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
576 pteval_t flags = val & PTE_FLAGS_MASK;
578 /* We assume the pte frame number is a MFN, so
579 just use it as-is. */
580 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
586 pteval_t xen_pte_val(pte_t pte)
588 if (xen_initial_domain() && (pte.pte & _PAGE_IOMAP))
591 return pte_mfn_to_pfn(pte.pte);
593 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
595 pgdval_t xen_pgd_val(pgd_t pgd)
597 return pte_mfn_to_pfn(pgd.pgd);
599 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
601 pte_t xen_make_pte(pteval_t pte)
603 phys_addr_t addr = (pte & PTE_PFN_MASK);
606 * Unprivileged domains are allowed to do IOMAPpings for
607 * PCI passthrough, but not map ISA space. The ISA
608 * mappings are just dummy local mappings to keep other
609 * parts of the kernel happy.
611 if (unlikely(pte & _PAGE_IOMAP) &&
612 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
613 pte = iomap_pte(pte);
616 pte = pte_pfn_to_mfn(pte);
619 return native_make_pte(pte);
621 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
623 pgd_t xen_make_pgd(pgdval_t pgd)
625 pgd = pte_pfn_to_mfn(pgd);
626 return native_make_pgd(pgd);
628 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
630 pmdval_t xen_pmd_val(pmd_t pmd)
632 return pte_mfn_to_pfn(pmd.pmd);
634 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
636 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
644 /* ptr may be ioremapped for 64-bit pagetable setup */
645 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
646 u.val = pud_val_ma(val);
647 xen_extend_mmu_update(&u);
649 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
651 xen_mc_issue(PARAVIRT_LAZY_MMU);
656 void xen_set_pud(pud_t *ptr, pud_t val)
658 ADD_STATS(pud_update, 1);
660 /* If page is not pinned, we can just update the entry
662 if (!xen_page_pinned(ptr)) {
667 ADD_STATS(pud_update_pinned, 1);
669 xen_set_pud_hyper(ptr, val);
672 void xen_set_pte(pte_t *ptep, pte_t pte)
674 if (xen_iomap_pte(pte)) {
675 xen_set_iomap_pte(ptep, pte);
679 ADD_STATS(pte_update, 1);
680 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
681 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
683 #ifdef CONFIG_X86_PAE
684 ptep->pte_high = pte.pte_high;
686 ptep->pte_low = pte.pte_low;
692 #ifdef CONFIG_X86_PAE
693 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
695 if (xen_iomap_pte(pte)) {
696 xen_set_iomap_pte(ptep, pte);
700 set_64bit((u64 *)ptep, native_pte_val(pte));
703 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
706 smp_wmb(); /* make sure low gets written first */
710 void xen_pmd_clear(pmd_t *pmdp)
712 set_pmd(pmdp, __pmd(0));
714 #endif /* CONFIG_X86_PAE */
716 pmd_t xen_make_pmd(pmdval_t pmd)
718 pmd = pte_pfn_to_mfn(pmd);
719 return native_make_pmd(pmd);
721 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
723 #if PAGETABLE_LEVELS == 4
724 pudval_t xen_pud_val(pud_t pud)
726 return pte_mfn_to_pfn(pud.pud);
728 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
730 pud_t xen_make_pud(pudval_t pud)
732 pud = pte_pfn_to_mfn(pud);
734 return native_make_pud(pud);
736 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
738 pgd_t *xen_get_user_pgd(pgd_t *pgd)
740 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
741 unsigned offset = pgd - pgd_page;
742 pgd_t *user_ptr = NULL;
744 if (offset < pgd_index(USER_LIMIT)) {
745 struct page *page = virt_to_page(pgd_page);
746 user_ptr = (pgd_t *)page->private;
754 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
758 u.ptr = virt_to_machine(ptr).maddr;
759 u.val = pgd_val_ma(val);
760 xen_extend_mmu_update(&u);
764 * Raw hypercall-based set_pgd, intended for in early boot before
765 * there's a page structure. This implies:
766 * 1. The only existing pagetable is the kernel's
767 * 2. It is always pinned
768 * 3. It has no user pagetable attached to it
770 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
776 __xen_set_pgd_hyper(ptr, val);
778 xen_mc_issue(PARAVIRT_LAZY_MMU);
783 void xen_set_pgd(pgd_t *ptr, pgd_t val)
785 pgd_t *user_ptr = xen_get_user_pgd(ptr);
787 ADD_STATS(pgd_update, 1);
789 /* If page is not pinned, we can just update the entry
791 if (!xen_page_pinned(ptr)) {
794 WARN_ON(xen_page_pinned(user_ptr));
800 ADD_STATS(pgd_update_pinned, 1);
801 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
803 /* If it's pinned, then we can at least batch the kernel and
804 user updates together. */
807 __xen_set_pgd_hyper(ptr, val);
809 __xen_set_pgd_hyper(user_ptr, val);
811 xen_mc_issue(PARAVIRT_LAZY_MMU);
813 #endif /* PAGETABLE_LEVELS == 4 */
816 * (Yet another) pagetable walker. This one is intended for pinning a
817 * pagetable. This means that it walks a pagetable and calls the
818 * callback function on each page it finds making up the page table,
819 * at every level. It walks the entire pagetable, but it only bothers
820 * pinning pte pages which are below limit. In the normal case this
821 * will be STACK_TOP_MAX, but at boot we need to pin up to
824 * For 32-bit the important bit is that we don't pin beyond there,
825 * because then we start getting into Xen's ptes.
827 * For 64-bit, we must skip the Xen hole in the middle of the address
828 * space, just after the big x86-64 virtual hole.
830 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
831 int (*func)(struct mm_struct *mm, struct page *,
836 unsigned hole_low, hole_high;
837 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
838 unsigned pgdidx, pudidx, pmdidx;
840 /* The limit is the last byte to be touched */
842 BUG_ON(limit >= FIXADDR_TOP);
844 if (xen_feature(XENFEAT_auto_translated_physmap))
848 * 64-bit has a great big hole in the middle of the address
849 * space, which contains the Xen mappings. On 32-bit these
850 * will end up making a zero-sized hole and so is a no-op.
852 hole_low = pgd_index(USER_LIMIT);
853 hole_high = pgd_index(PAGE_OFFSET);
855 pgdidx_limit = pgd_index(limit);
857 pudidx_limit = pud_index(limit);
862 pmdidx_limit = pmd_index(limit);
867 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
870 if (pgdidx >= hole_low && pgdidx < hole_high)
873 if (!pgd_val(pgd[pgdidx]))
876 pud = pud_offset(&pgd[pgdidx], 0);
878 if (PTRS_PER_PUD > 1) /* not folded */
879 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
881 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
884 if (pgdidx == pgdidx_limit &&
885 pudidx > pudidx_limit)
888 if (pud_none(pud[pudidx]))
891 pmd = pmd_offset(&pud[pudidx], 0);
893 if (PTRS_PER_PMD > 1) /* not folded */
894 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
896 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
899 if (pgdidx == pgdidx_limit &&
900 pudidx == pudidx_limit &&
901 pmdidx > pmdidx_limit)
904 if (pmd_none(pmd[pmdidx]))
907 pte = pmd_page(pmd[pmdidx]);
908 flush |= (*func)(mm, pte, PT_PTE);
914 /* Do the top level last, so that the callbacks can use it as
915 a cue to do final things like tlb flushes. */
916 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
921 static int xen_pgd_walk(struct mm_struct *mm,
922 int (*func)(struct mm_struct *mm, struct page *,
926 return __xen_pgd_walk(mm, mm->pgd, func, limit);
929 /* If we're using split pte locks, then take the page's lock and
930 return a pointer to it. Otherwise return NULL. */
931 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
933 spinlock_t *ptl = NULL;
935 #if USE_SPLIT_PTLOCKS
936 ptl = __pte_lockptr(page);
937 spin_lock_nest_lock(ptl, &mm->page_table_lock);
943 static void xen_pte_unlock(void *v)
949 static void xen_do_pin(unsigned level, unsigned long pfn)
951 struct mmuext_op *op;
952 struct multicall_space mcs;
954 mcs = __xen_mc_entry(sizeof(*op));
957 op->arg1.mfn = pfn_to_mfn(pfn);
958 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
961 static int xen_pin_page(struct mm_struct *mm, struct page *page,
964 unsigned pgfl = TestSetPagePinned(page);
968 flush = 0; /* already pinned */
969 else if (PageHighMem(page))
970 /* kmaps need flushing if we found an unpinned
974 void *pt = lowmem_page_address(page);
975 unsigned long pfn = page_to_pfn(page);
976 struct multicall_space mcs = __xen_mc_entry(0);
982 * We need to hold the pagetable lock between the time
983 * we make the pagetable RO and when we actually pin
984 * it. If we don't, then other users may come in and
985 * attempt to update the pagetable by writing it,
986 * which will fail because the memory is RO but not
987 * pinned, so Xen won't do the trap'n'emulate.
989 * If we're using split pte locks, we can't hold the
990 * entire pagetable's worth of locks during the
991 * traverse, because we may wrap the preempt count (8
992 * bits). The solution is to mark RO and pin each PTE
993 * page while holding the lock. This means the number
994 * of locks we end up holding is never more than a
995 * batch size (~32 entries, at present).
997 * If we're not using split pte locks, we needn't pin
998 * the PTE pages independently, because we're
999 * protected by the overall pagetable lock.
1002 if (level == PT_PTE)
1003 ptl = xen_pte_lock(page, mm);
1005 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1006 pfn_pte(pfn, PAGE_KERNEL_RO),
1007 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1010 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
1012 /* Queue a deferred unlock for when this batch
1014 xen_mc_callback(xen_pte_unlock, ptl);
1021 /* This is called just after a mm has been created, but it has not
1022 been used yet. We need to make sure that its pagetable is all
1023 read-only, and can be pinned. */
1024 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
1028 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
1029 /* re-enable interrupts for flushing */
1032 kmap_flush_unused();
1037 #ifdef CONFIG_X86_64
1039 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1041 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
1044 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
1045 xen_do_pin(MMUEXT_PIN_L4_TABLE,
1046 PFN_DOWN(__pa(user_pgd)));
1049 #else /* CONFIG_X86_32 */
1050 #ifdef CONFIG_X86_PAE
1051 /* Need to make sure unshared kernel PMD is pinnable */
1052 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1055 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
1056 #endif /* CONFIG_X86_64 */
1060 static void xen_pgd_pin(struct mm_struct *mm)
1062 __xen_pgd_pin(mm, mm->pgd);
1066 * On save, we need to pin all pagetables to make sure they get their
1067 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1068 * them (unpinned pgds are not currently in use, probably because the
1069 * process is under construction or destruction).
1071 * Expected to be called in stop_machine() ("equivalent to taking
1072 * every spinlock in the system"), so the locking doesn't really
1073 * matter all that much.
1075 void xen_mm_pin_all(void)
1077 unsigned long flags;
1080 spin_lock_irqsave(&pgd_lock, flags);
1082 list_for_each_entry(page, &pgd_list, lru) {
1083 if (!PagePinned(page)) {
1084 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1085 SetPageSavePinned(page);
1089 spin_unlock_irqrestore(&pgd_lock, flags);
1093 * The init_mm pagetable is really pinned as soon as its created, but
1094 * that's before we have page structures to store the bits. So do all
1095 * the book-keeping now.
1097 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1098 enum pt_level level)
1100 SetPagePinned(page);
1104 static void __init xen_mark_init_mm_pinned(void)
1106 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1109 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1110 enum pt_level level)
1112 unsigned pgfl = TestClearPagePinned(page);
1114 if (pgfl && !PageHighMem(page)) {
1115 void *pt = lowmem_page_address(page);
1116 unsigned long pfn = page_to_pfn(page);
1117 spinlock_t *ptl = NULL;
1118 struct multicall_space mcs;
1121 * Do the converse to pin_page. If we're using split
1122 * pte locks, we must be holding the lock for while
1123 * the pte page is unpinned but still RO to prevent
1124 * concurrent updates from seeing it in this
1125 * partially-pinned state.
1127 if (level == PT_PTE) {
1128 ptl = xen_pte_lock(page, mm);
1131 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1134 mcs = __xen_mc_entry(0);
1136 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1137 pfn_pte(pfn, PAGE_KERNEL),
1138 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1141 /* unlock when batch completed */
1142 xen_mc_callback(xen_pte_unlock, ptl);
1146 return 0; /* never need to flush on unpin */
1149 /* Release a pagetables pages back as normal RW */
1150 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1154 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1156 #ifdef CONFIG_X86_64
1158 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1161 xen_do_pin(MMUEXT_UNPIN_TABLE,
1162 PFN_DOWN(__pa(user_pgd)));
1163 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1168 #ifdef CONFIG_X86_PAE
1169 /* Need to make sure unshared kernel PMD is unpinned */
1170 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1174 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1179 static void xen_pgd_unpin(struct mm_struct *mm)
1181 __xen_pgd_unpin(mm, mm->pgd);
1185 * On resume, undo any pinning done at save, so that the rest of the
1186 * kernel doesn't see any unexpected pinned pagetables.
1188 void xen_mm_unpin_all(void)
1190 unsigned long flags;
1193 spin_lock_irqsave(&pgd_lock, flags);
1195 list_for_each_entry(page, &pgd_list, lru) {
1196 if (PageSavePinned(page)) {
1197 BUG_ON(!PagePinned(page));
1198 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1199 ClearPageSavePinned(page);
1203 spin_unlock_irqrestore(&pgd_lock, flags);
1206 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1208 spin_lock(&next->page_table_lock);
1210 spin_unlock(&next->page_table_lock);
1213 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1215 spin_lock(&mm->page_table_lock);
1217 spin_unlock(&mm->page_table_lock);
1222 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1223 we need to repoint it somewhere else before we can unpin it. */
1224 static void drop_other_mm_ref(void *info)
1226 struct mm_struct *mm = info;
1227 struct mm_struct *active_mm;
1229 active_mm = percpu_read(cpu_tlbstate.active_mm);
1231 if (active_mm == mm)
1232 leave_mm(smp_processor_id());
1234 /* If this cpu still has a stale cr3 reference, then make sure
1235 it has been flushed. */
1236 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1237 load_cr3(swapper_pg_dir);
1240 static void xen_drop_mm_ref(struct mm_struct *mm)
1245 if (current->active_mm == mm) {
1246 if (current->mm == mm)
1247 load_cr3(swapper_pg_dir);
1249 leave_mm(smp_processor_id());
1252 /* Get the "official" set of cpus referring to our pagetable. */
1253 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1254 for_each_online_cpu(cpu) {
1255 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1256 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1258 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1262 cpumask_copy(mask, mm_cpumask(mm));
1264 /* It's possible that a vcpu may have a stale reference to our
1265 cr3, because its in lazy mode, and it hasn't yet flushed
1266 its set of pending hypercalls yet. In this case, we can
1267 look at its actual current cr3 value, and force it to flush
1269 for_each_online_cpu(cpu) {
1270 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1271 cpumask_set_cpu(cpu, mask);
1274 if (!cpumask_empty(mask))
1275 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1276 free_cpumask_var(mask);
1279 static void xen_drop_mm_ref(struct mm_struct *mm)
1281 if (current->active_mm == mm)
1282 load_cr3(swapper_pg_dir);
1287 * While a process runs, Xen pins its pagetables, which means that the
1288 * hypervisor forces it to be read-only, and it controls all updates
1289 * to it. This means that all pagetable updates have to go via the
1290 * hypervisor, which is moderately expensive.
1292 * Since we're pulling the pagetable down, we switch to use init_mm,
1293 * unpin old process pagetable and mark it all read-write, which
1294 * allows further operations on it to be simple memory accesses.
1296 * The only subtle point is that another CPU may be still using the
1297 * pagetable because of lazy tlb flushing. This means we need need to
1298 * switch all CPUs off this pagetable before we can unpin it.
1300 void xen_exit_mmap(struct mm_struct *mm)
1302 get_cpu(); /* make sure we don't move around */
1303 xen_drop_mm_ref(mm);
1306 spin_lock(&mm->page_table_lock);
1308 /* pgd may not be pinned in the error exit path of execve */
1309 if (xen_page_pinned(mm->pgd))
1312 spin_unlock(&mm->page_table_lock);
1315 static __init void xen_pagetable_setup_start(pgd_t *base)
1319 static void xen_post_allocator_init(void);
1321 static __init void xen_pagetable_setup_done(pgd_t *base)
1323 xen_setup_shared_info();
1324 xen_post_allocator_init();
1327 static void xen_write_cr2(unsigned long cr2)
1329 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1332 static unsigned long xen_read_cr2(void)
1334 return percpu_read(xen_vcpu)->arch.cr2;
1337 unsigned long xen_read_cr2_direct(void)
1339 return percpu_read(xen_vcpu_info.arch.cr2);
1342 static void xen_flush_tlb(void)
1344 struct mmuext_op *op;
1345 struct multicall_space mcs;
1349 mcs = xen_mc_entry(sizeof(*op));
1352 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1353 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1355 xen_mc_issue(PARAVIRT_LAZY_MMU);
1360 static void xen_flush_tlb_single(unsigned long addr)
1362 struct mmuext_op *op;
1363 struct multicall_space mcs;
1367 mcs = xen_mc_entry(sizeof(*op));
1369 op->cmd = MMUEXT_INVLPG_LOCAL;
1370 op->arg1.linear_addr = addr & PAGE_MASK;
1371 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1373 xen_mc_issue(PARAVIRT_LAZY_MMU);
1378 static void xen_flush_tlb_others(const struct cpumask *cpus,
1379 struct mm_struct *mm, unsigned long va)
1382 struct mmuext_op op;
1383 DECLARE_BITMAP(mask, NR_CPUS);
1385 struct multicall_space mcs;
1387 if (cpumask_empty(cpus))
1388 return; /* nothing to do */
1390 mcs = xen_mc_entry(sizeof(*args));
1392 args->op.arg2.vcpumask = to_cpumask(args->mask);
1394 /* Remove us, and any offline CPUS. */
1395 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1396 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1398 if (va == TLB_FLUSH_ALL) {
1399 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1401 args->op.cmd = MMUEXT_INVLPG_MULTI;
1402 args->op.arg1.linear_addr = va;
1405 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1407 xen_mc_issue(PARAVIRT_LAZY_MMU);
1410 static unsigned long xen_read_cr3(void)
1412 return percpu_read(xen_cr3);
1415 static void set_current_cr3(void *v)
1417 percpu_write(xen_current_cr3, (unsigned long)v);
1420 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1422 struct mmuext_op *op;
1423 struct multicall_space mcs;
1427 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1431 WARN_ON(mfn == 0 && kernel);
1433 mcs = __xen_mc_entry(sizeof(*op));
1436 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1439 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1442 percpu_write(xen_cr3, cr3);
1444 /* Update xen_current_cr3 once the batch has actually
1446 xen_mc_callback(set_current_cr3, (void *)cr3);
1450 static void xen_write_cr3(unsigned long cr3)
1452 BUG_ON(preemptible());
1454 xen_mc_batch(); /* disables interrupts */
1456 /* Update while interrupts are disabled, so its atomic with
1458 percpu_write(xen_cr3, cr3);
1460 __xen_write_cr3(true, cr3);
1462 #ifdef CONFIG_X86_64
1464 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1466 __xen_write_cr3(false, __pa(user_pgd));
1468 __xen_write_cr3(false, 0);
1472 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1475 static int xen_pgd_alloc(struct mm_struct *mm)
1477 pgd_t *pgd = mm->pgd;
1480 BUG_ON(PagePinned(virt_to_page(pgd)));
1482 #ifdef CONFIG_X86_64
1484 struct page *page = virt_to_page(pgd);
1487 BUG_ON(page->private != 0);
1491 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1492 page->private = (unsigned long)user_pgd;
1494 if (user_pgd != NULL) {
1495 user_pgd[pgd_index(VSYSCALL_START)] =
1496 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1500 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1507 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1509 #ifdef CONFIG_X86_64
1510 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1513 free_page((unsigned long)user_pgd);
1517 #ifdef CONFIG_X86_32
1518 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1520 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1521 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1522 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1528 /* Init-time set_pte while constructing initial pagetables, which
1529 doesn't allow RO pagetable pages to be remapped RW */
1530 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1532 pte = mask_rw_pte(ptep, pte);
1534 xen_set_pte(ptep, pte);
1538 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1540 struct mmuext_op op;
1542 op.arg1.mfn = pfn_to_mfn(pfn);
1543 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1547 /* Early in boot, while setting up the initial pagetable, assume
1548 everything is pinned. */
1549 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1551 #ifdef CONFIG_FLATMEM
1552 BUG_ON(mem_map); /* should only be used early */
1554 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1555 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1558 /* Used for pmd and pud */
1559 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1561 #ifdef CONFIG_FLATMEM
1562 BUG_ON(mem_map); /* should only be used early */
1564 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1567 /* Early release_pte assumes that all pts are pinned, since there's
1568 only init_mm and anything attached to that is pinned. */
1569 static __init void xen_release_pte_init(unsigned long pfn)
1571 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1572 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1575 static __init void xen_release_pmd_init(unsigned long pfn)
1577 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1580 /* This needs to make sure the new pte page is pinned iff its being
1581 attached to a pinned pagetable. */
1582 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1584 struct page *page = pfn_to_page(pfn);
1586 if (PagePinned(virt_to_page(mm->pgd))) {
1587 SetPagePinned(page);
1589 if (!PageHighMem(page)) {
1590 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1591 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1592 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1594 /* make sure there are no stray mappings of
1596 kmap_flush_unused();
1601 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1603 xen_alloc_ptpage(mm, pfn, PT_PTE);
1606 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1608 xen_alloc_ptpage(mm, pfn, PT_PMD);
1611 /* This should never happen until we're OK to use struct page */
1612 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1614 struct page *page = pfn_to_page(pfn);
1616 if (PagePinned(page)) {
1617 if (!PageHighMem(page)) {
1618 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1619 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1620 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1622 ClearPagePinned(page);
1626 static void xen_release_pte(unsigned long pfn)
1628 xen_release_ptpage(pfn, PT_PTE);
1631 static void xen_release_pmd(unsigned long pfn)
1633 xen_release_ptpage(pfn, PT_PMD);
1636 #if PAGETABLE_LEVELS == 4
1637 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1639 xen_alloc_ptpage(mm, pfn, PT_PUD);
1642 static void xen_release_pud(unsigned long pfn)
1644 xen_release_ptpage(pfn, PT_PUD);
1648 void __init xen_reserve_top(void)
1650 #ifdef CONFIG_X86_32
1651 unsigned long top = HYPERVISOR_VIRT_START;
1652 struct xen_platform_parameters pp;
1654 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1655 top = pp.virt_start;
1657 reserve_top_address(-top);
1658 #endif /* CONFIG_X86_32 */
1662 * Like __va(), but returns address in the kernel mapping (which is
1663 * all we have until the physical memory mapping has been set up.
1665 static void *__ka(phys_addr_t paddr)
1667 #ifdef CONFIG_X86_64
1668 return (void *)(paddr + __START_KERNEL_map);
1674 /* Convert a machine address to physical address */
1675 static unsigned long m2p(phys_addr_t maddr)
1679 maddr &= PTE_PFN_MASK;
1680 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1685 /* Convert a machine address to kernel virtual */
1686 static void *m2v(phys_addr_t maddr)
1688 return __ka(m2p(maddr));
1691 static void set_page_prot(void *addr, pgprot_t prot)
1693 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1694 pte_t pte = pfn_pte(pfn, prot);
1696 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1700 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1702 unsigned pmdidx, pteidx;
1708 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1711 /* Reuse or allocate a page of ptes */
1712 if (pmd_present(pmd[pmdidx]))
1713 pte_page = m2v(pmd[pmdidx].pmd);
1715 /* Check for free pte pages */
1716 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1719 pte_page = &level1_ident_pgt[ident_pte];
1720 ident_pte += PTRS_PER_PTE;
1722 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1725 /* Install mappings */
1726 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1729 if (pfn > max_pfn_mapped)
1730 max_pfn_mapped = pfn;
1732 if (!pte_none(pte_page[pteidx]))
1735 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1736 pte_page[pteidx] = pte;
1740 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1741 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1743 set_page_prot(pmd, PAGE_KERNEL_RO);
1746 #ifdef CONFIG_X86_64
1747 static void convert_pfn_mfn(void *v)
1752 /* All levels are converted the same way, so just treat them
1754 for (i = 0; i < PTRS_PER_PTE; i++)
1755 pte[i] = xen_make_pte(pte[i].pte);
1759 * Set up the inital kernel pagetable.
1761 * We can construct this by grafting the Xen provided pagetable into
1762 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1763 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1764 * means that only the kernel has a physical mapping to start with -
1765 * but that's enough to get __va working. We need to fill in the rest
1766 * of the physical mapping once some sort of allocator has been set
1769 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1770 unsigned long max_pfn)
1775 /* Zap identity mapping */
1776 init_level4_pgt[0] = __pgd(0);
1778 /* Pre-constructed entries are in pfn, so convert to mfn */
1779 convert_pfn_mfn(init_level4_pgt);
1780 convert_pfn_mfn(level3_ident_pgt);
1781 convert_pfn_mfn(level3_kernel_pgt);
1783 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1784 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1786 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1787 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1789 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1790 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1791 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1793 /* Set up identity map */
1794 xen_map_identity_early(level2_ident_pgt, max_pfn);
1796 /* Make pagetable pieces RO */
1797 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1798 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1799 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1800 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1801 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1802 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1804 /* Pin down new L4 */
1805 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1806 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1808 /* Unpin Xen-provided one */
1809 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1812 pgd = init_level4_pgt;
1815 * At this stage there can be no user pgd, and no page
1816 * structure to attach it to, so make sure we just set kernel
1820 __xen_write_cr3(true, __pa(pgd));
1821 xen_mc_issue(PARAVIRT_LAZY_CPU);
1823 reserve_early(__pa(xen_start_info->pt_base),
1824 __pa(xen_start_info->pt_base +
1825 xen_start_info->nr_pt_frames * PAGE_SIZE),
1830 #else /* !CONFIG_X86_64 */
1831 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1833 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1834 unsigned long max_pfn)
1838 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1839 xen_start_info->nr_pt_frames * PAGE_SIZE +
1842 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1843 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1845 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1847 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1848 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1849 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1851 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1852 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1853 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1855 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1857 xen_write_cr3(__pa(swapper_pg_dir));
1859 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1861 reserve_early(__pa(xen_start_info->pt_base),
1862 __pa(xen_start_info->pt_base +
1863 xen_start_info->nr_pt_frames * PAGE_SIZE),
1866 return swapper_pg_dir;
1868 #endif /* CONFIG_X86_64 */
1870 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1874 phys >>= PAGE_SHIFT;
1877 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1878 #ifdef CONFIG_X86_F00F_BUG
1881 #ifdef CONFIG_X86_32
1884 # ifdef CONFIG_HIGHMEM
1885 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1888 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1890 #ifdef CONFIG_X86_LOCAL_APIC
1891 case FIX_APIC_BASE: /* maps dummy local APIC */
1893 case FIX_TEXT_POKE0:
1894 case FIX_TEXT_POKE1:
1895 /* All local page mappings */
1896 pte = pfn_pte(phys, prot);
1899 case FIX_PARAVIRT_BOOTMAP:
1900 /* This is an MFN, but it isn't an IO mapping from the
1902 pte = mfn_pte(phys, prot);
1906 /* By default, set_fixmap is used for hardware mappings */
1907 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1911 __native_set_fixmap(idx, pte);
1913 #ifdef CONFIG_X86_64
1914 /* Replicate changes to map the vsyscall page into the user
1915 pagetable vsyscall mapping. */
1916 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1917 unsigned long vaddr = __fix_to_virt(idx);
1918 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1923 static __init void xen_post_allocator_init(void)
1925 pv_mmu_ops.set_pte = xen_set_pte;
1926 pv_mmu_ops.set_pmd = xen_set_pmd;
1927 pv_mmu_ops.set_pud = xen_set_pud;
1928 #if PAGETABLE_LEVELS == 4
1929 pv_mmu_ops.set_pgd = xen_set_pgd;
1932 /* This will work as long as patching hasn't happened yet
1933 (which it hasn't) */
1934 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1935 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1936 pv_mmu_ops.release_pte = xen_release_pte;
1937 pv_mmu_ops.release_pmd = xen_release_pmd;
1938 #if PAGETABLE_LEVELS == 4
1939 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1940 pv_mmu_ops.release_pud = xen_release_pud;
1943 #ifdef CONFIG_X86_64
1944 SetPagePinned(virt_to_page(level3_user_vsyscall));
1946 xen_mark_init_mm_pinned();
1949 static void xen_leave_lazy_mmu(void)
1953 paravirt_leave_lazy_mmu();
1957 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1958 .read_cr2 = xen_read_cr2,
1959 .write_cr2 = xen_write_cr2,
1961 .read_cr3 = xen_read_cr3,
1962 .write_cr3 = xen_write_cr3,
1964 .flush_tlb_user = xen_flush_tlb,
1965 .flush_tlb_kernel = xen_flush_tlb,
1966 .flush_tlb_single = xen_flush_tlb_single,
1967 .flush_tlb_others = xen_flush_tlb_others,
1969 .pte_update = paravirt_nop,
1970 .pte_update_defer = paravirt_nop,
1972 .pgd_alloc = xen_pgd_alloc,
1973 .pgd_free = xen_pgd_free,
1975 .alloc_pte = xen_alloc_pte_init,
1976 .release_pte = xen_release_pte_init,
1977 .alloc_pmd = xen_alloc_pmd_init,
1978 .alloc_pmd_clone = paravirt_nop,
1979 .release_pmd = xen_release_pmd_init,
1981 #ifdef CONFIG_X86_64
1982 .set_pte = xen_set_pte,
1984 .set_pte = xen_set_pte_init,
1986 .set_pte_at = xen_set_pte_at,
1987 .set_pmd = xen_set_pmd_hyper,
1989 .ptep_modify_prot_start = __ptep_modify_prot_start,
1990 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1992 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1993 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1995 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1996 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1998 #ifdef CONFIG_X86_PAE
1999 .set_pte_atomic = xen_set_pte_atomic,
2000 .pte_clear = xen_pte_clear,
2001 .pmd_clear = xen_pmd_clear,
2002 #endif /* CONFIG_X86_PAE */
2003 .set_pud = xen_set_pud_hyper,
2005 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2006 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2008 #if PAGETABLE_LEVELS == 4
2009 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2010 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2011 .set_pgd = xen_set_pgd_hyper,
2013 .alloc_pud = xen_alloc_pmd_init,
2014 .release_pud = xen_release_pmd_init,
2015 #endif /* PAGETABLE_LEVELS == 4 */
2017 .activate_mm = xen_activate_mm,
2018 .dup_mmap = xen_dup_mmap,
2019 .exit_mmap = xen_exit_mmap,
2022 .enter = paravirt_enter_lazy_mmu,
2023 .leave = xen_leave_lazy_mmu,
2026 .set_fixmap = xen_set_fixmap,
2029 void __init xen_init_mmu_ops(void)
2031 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2032 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2033 pv_mmu_ops = xen_mmu_ops;
2035 vmap_lazy_unmap = false;
2038 /* Protected by xen_reservation_lock. */
2039 #define MAX_CONTIG_ORDER 9 /* 2MB */
2040 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2042 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2043 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2044 unsigned long *in_frames,
2045 unsigned long *out_frames)
2048 struct multicall_space mcs;
2051 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2052 mcs = __xen_mc_entry(0);
2055 in_frames[i] = virt_to_mfn(vaddr);
2057 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2058 set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2061 out_frames[i] = virt_to_pfn(vaddr);
2067 * Update the pfn-to-mfn mappings for a virtual address range, either to
2068 * point to an array of mfns, or contiguously from a single starting
2071 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2072 unsigned long *mfns,
2073 unsigned long first_mfn)
2080 limit = 1u << order;
2081 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2082 struct multicall_space mcs;
2085 mcs = __xen_mc_entry(0);
2089 mfn = first_mfn + i;
2091 if (i < (limit - 1))
2095 flags = UVMF_INVLPG | UVMF_ALL;
2097 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2100 MULTI_update_va_mapping(mcs.mc, vaddr,
2101 mfn_pte(mfn, PAGE_KERNEL), flags);
2103 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2110 * Perform the hypercall to exchange a region of our pfns to point to
2111 * memory with the required contiguous alignment. Takes the pfns as
2112 * input, and populates mfns as output.
2114 * Returns a success code indicating whether the hypervisor was able to
2115 * satisfy the request or not.
2117 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2118 unsigned long *pfns_in,
2119 unsigned long extents_out,
2120 unsigned int order_out,
2121 unsigned long *mfns_out,
2122 unsigned int address_bits)
2127 struct xen_memory_exchange exchange = {
2129 .nr_extents = extents_in,
2130 .extent_order = order_in,
2131 .extent_start = pfns_in,
2135 .nr_extents = extents_out,
2136 .extent_order = order_out,
2137 .extent_start = mfns_out,
2138 .address_bits = address_bits,
2143 BUG_ON(extents_in << order_in != extents_out << order_out);
2145 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2146 success = (exchange.nr_exchanged == extents_in);
2148 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2149 BUG_ON(success && (rc != 0));
2154 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2155 unsigned int address_bits)
2157 unsigned long *in_frames = discontig_frames, out_frame;
2158 unsigned long flags;
2162 * Currently an auto-translated guest will not perform I/O, nor will
2163 * it require PAE page directories below 4GB. Therefore any calls to
2164 * this function are redundant and can be ignored.
2167 if (xen_feature(XENFEAT_auto_translated_physmap))
2170 if (unlikely(order > MAX_CONTIG_ORDER))
2173 memset((void *) vstart, 0, PAGE_SIZE << order);
2175 spin_lock_irqsave(&xen_reservation_lock, flags);
2177 /* 1. Zap current PTEs, remembering MFNs. */
2178 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2180 /* 2. Get a new contiguous memory extent. */
2181 out_frame = virt_to_pfn(vstart);
2182 success = xen_exchange_memory(1UL << order, 0, in_frames,
2183 1, order, &out_frame,
2186 /* 3. Map the new extent in place of old pages. */
2188 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2190 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2192 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2194 return success ? 0 : -ENOMEM;
2196 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2198 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2200 unsigned long *out_frames = discontig_frames, in_frame;
2201 unsigned long flags;
2204 if (xen_feature(XENFEAT_auto_translated_physmap))
2207 if (unlikely(order > MAX_CONTIG_ORDER))
2210 memset((void *) vstart, 0, PAGE_SIZE << order);
2212 spin_lock_irqsave(&xen_reservation_lock, flags);
2214 /* 1. Find start MFN of contiguous extent. */
2215 in_frame = virt_to_mfn(vstart);
2217 /* 2. Zap current PTEs. */
2218 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2220 /* 3. Do the exchange for non-contiguous MFNs. */
2221 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2224 /* 4. Map new pages in place of old pages. */
2226 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2228 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2230 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2232 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2234 #ifdef CONFIG_XEN_PVHVM
2235 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2237 struct xen_hvm_pagetable_dying a;
2240 a.domid = DOMID_SELF;
2241 a.gpa = __pa(mm->pgd);
2242 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2243 WARN_ON_ONCE(rc < 0);
2246 static int is_pagetable_dying_supported(void)
2248 struct xen_hvm_pagetable_dying a;
2251 a.domid = DOMID_SELF;
2253 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2255 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2261 void __init xen_hvm_init_mmu_ops(void)
2263 if (is_pagetable_dying_supported())
2264 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2268 #define REMAP_BATCH_SIZE 16
2273 struct mmu_update *mmu_update;
2276 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2277 unsigned long addr, void *data)
2279 struct remap_data *rmd = data;
2280 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2282 rmd->mmu_update->ptr = arbitrary_virt_to_machine(ptep).maddr;
2283 rmd->mmu_update->val = pte_val_ma(pte);
2289 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2291 unsigned long mfn, int nr,
2292 pgprot_t prot, unsigned domid)
2294 struct remap_data rmd;
2295 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2297 unsigned long range;
2300 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2302 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
2308 batch = min(REMAP_BATCH_SIZE, nr);
2309 range = (unsigned long)batch << PAGE_SHIFT;
2311 rmd.mmu_update = mmu_update;
2312 err = apply_to_page_range(vma->vm_mm, addr, range,
2313 remap_area_mfn_pte_fn, &rmd);
2318 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2332 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2334 #ifdef CONFIG_XEN_DEBUG_FS
2336 static struct dentry *d_mmu_debug;
2338 static int __init xen_mmu_debugfs(void)
2340 struct dentry *d_xen = xen_init_debugfs();
2345 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2347 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2349 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2350 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2351 &mmu_stats.pgd_update_pinned);
2352 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2353 &mmu_stats.pgd_update_pinned);
2355 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2356 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2357 &mmu_stats.pud_update_pinned);
2358 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2359 &mmu_stats.pud_update_pinned);
2361 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2362 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2363 &mmu_stats.pmd_update_pinned);
2364 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2365 &mmu_stats.pmd_update_pinned);
2367 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2368 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2369 // &mmu_stats.pte_update_pinned);
2370 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2371 &mmu_stats.pte_update_pinned);
2373 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2374 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2375 &mmu_stats.mmu_update_extended);
2376 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2377 mmu_stats.mmu_update_histo, 20);
2379 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2380 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2381 &mmu_stats.set_pte_at_batched);
2382 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2383 &mmu_stats.set_pte_at_current);
2384 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2385 &mmu_stats.set_pte_at_kernel);
2387 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2388 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2389 &mmu_stats.prot_commit_batched);
2393 fs_initcall(xen_mmu_debugfs);
2395 #endif /* CONFIG_XEN_DEBUG_FS */