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>
48 #include <linux/memblock.h>
49 #include <linux/seq_file.h>
51 #include <asm/pgtable.h>
52 #include <asm/tlbflush.h>
53 #include <asm/fixmap.h>
54 #include <asm/mmu_context.h>
55 #include <asm/setup.h>
56 #include <asm/paravirt.h>
58 #include <asm/linkage.h>
63 #include <asm/xen/hypercall.h>
64 #include <asm/xen/hypervisor.h>
68 #include <xen/interface/xen.h>
69 #include <xen/interface/hvm/hvm_op.h>
70 #include <xen/interface/version.h>
71 #include <xen/interface/memory.h>
72 #include <xen/hvc-console.h>
74 #include "multicalls.h"
78 #define MMU_UPDATE_HISTO 30
81 * Protects atomic reservation decrease/increase against concurrent increases.
82 * Also protects non-atomic updates of current_pages and balloon lists.
84 DEFINE_SPINLOCK(xen_reservation_lock);
86 #ifdef CONFIG_XEN_DEBUG_FS
90 u32 pgd_update_pinned;
91 u32 pgd_update_batched;
94 u32 pud_update_pinned;
95 u32 pud_update_batched;
98 u32 pmd_update_pinned;
99 u32 pmd_update_batched;
102 u32 pte_update_pinned;
103 u32 pte_update_batched;
106 u32 mmu_update_extended;
107 u32 mmu_update_histo[MMU_UPDATE_HISTO];
110 u32 prot_commit_batched;
113 static u8 zero_stats;
115 static inline void check_zero(void)
117 if (unlikely(zero_stats)) {
118 memset(&mmu_stats, 0, sizeof(mmu_stats));
123 #define ADD_STATS(elem, val) \
124 do { check_zero(); mmu_stats.elem += (val); } while(0)
126 #else /* !CONFIG_XEN_DEBUG_FS */
128 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
130 #endif /* CONFIG_XEN_DEBUG_FS */
134 * Identity map, in addition to plain kernel map. This needs to be
135 * large enough to allocate page table pages to allocate the rest.
136 * Each page can map 2MB.
138 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
139 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
142 /* l3 pud for userspace vsyscall mapping */
143 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
144 #endif /* CONFIG_X86_64 */
147 * Note about cr3 (pagetable base) values:
149 * xen_cr3 contains the current logical cr3 value; it contains the
150 * last set cr3. This may not be the current effective cr3, because
151 * its update may be being lazily deferred. However, a vcpu looking
152 * at its own cr3 can use this value knowing that it everything will
153 * be self-consistent.
155 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
156 * hypercall to set the vcpu cr3 is complete (so it may be a little
157 * out of date, but it will never be set early). If one vcpu is
158 * looking at another vcpu's cr3 value, it should use this variable.
160 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
161 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
165 * Just beyond the highest usermode address. STACK_TOP_MAX has a
166 * redzone above it, so round it up to a PGD boundary.
168 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
170 unsigned long arbitrary_virt_to_mfn(void *vaddr)
172 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
174 return PFN_DOWN(maddr.maddr);
177 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
179 unsigned long address = (unsigned long)vaddr;
185 * if the PFN is in the linear mapped vaddr range, we can just use
186 * the (quick) virt_to_machine() p2m lookup
188 if (virt_addr_valid(vaddr))
189 return virt_to_machine(vaddr);
191 /* otherwise we have to do a (slower) full page-table walk */
193 pte = lookup_address(address, &level);
195 offset = address & ~PAGE_MASK;
196 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
198 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
200 void make_lowmem_page_readonly(void *vaddr)
203 unsigned long address = (unsigned long)vaddr;
206 pte = lookup_address(address, &level);
208 return; /* vaddr missing */
210 ptev = pte_wrprotect(*pte);
212 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
216 void make_lowmem_page_readwrite(void *vaddr)
219 unsigned long address = (unsigned long)vaddr;
222 pte = lookup_address(address, &level);
224 return; /* vaddr missing */
226 ptev = pte_mkwrite(*pte);
228 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
233 static bool xen_page_pinned(void *ptr)
235 struct page *page = virt_to_page(ptr);
237 return PagePinned(page);
240 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
242 struct multicall_space mcs;
243 struct mmu_update *u;
245 mcs = xen_mc_entry(sizeof(*u));
248 /* ptep might be kmapped when using 32-bit HIGHPTE */
249 u->ptr = virt_to_machine(ptep).maddr;
250 u->val = pte_val_ma(pteval);
252 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
254 xen_mc_issue(PARAVIRT_LAZY_MMU);
256 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
258 static void xen_extend_mmu_update(const struct mmu_update *update)
260 struct multicall_space mcs;
261 struct mmu_update *u;
263 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
265 if (mcs.mc != NULL) {
266 ADD_STATS(mmu_update_extended, 1);
267 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
271 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
272 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
274 ADD_STATS(mmu_update_histo[0], 1);
276 ADD_STATS(mmu_update, 1);
277 mcs = __xen_mc_entry(sizeof(*u));
278 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
279 ADD_STATS(mmu_update_histo[1], 1);
286 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
294 /* ptr may be ioremapped for 64-bit pagetable setup */
295 u.ptr = virt_to_machine(ptr).maddr;
296 u.val = pmd_val_ma(val);
297 xen_extend_mmu_update(&u);
299 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
301 xen_mc_issue(PARAVIRT_LAZY_MMU);
306 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
308 ADD_STATS(pmd_update, 1);
310 /* If page is not pinned, we can just update the entry
312 if (!xen_page_pinned(ptr)) {
317 ADD_STATS(pmd_update_pinned, 1);
319 xen_set_pmd_hyper(ptr, val);
323 * Associate a virtual page frame with a given physical page frame
324 * and protection flags for that frame.
326 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
328 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
331 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
335 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
340 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
341 u.val = pte_val_ma(pteval);
342 xen_extend_mmu_update(&u);
344 xen_mc_issue(PARAVIRT_LAZY_MMU);
349 static void xen_set_pte(pte_t *ptep, pte_t pteval)
351 ADD_STATS(pte_update, 1);
352 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
354 if (!xen_batched_set_pte(ptep, pteval))
355 native_set_pte(ptep, pteval);
358 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
359 pte_t *ptep, pte_t pteval)
361 xen_set_pte(ptep, pteval);
364 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
365 unsigned long addr, pte_t *ptep)
367 /* Just return the pte as-is. We preserve the bits on commit */
371 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
372 pte_t *ptep, pte_t pte)
378 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
379 u.val = pte_val_ma(pte);
380 xen_extend_mmu_update(&u);
382 ADD_STATS(prot_commit, 1);
383 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
385 xen_mc_issue(PARAVIRT_LAZY_MMU);
388 /* Assume pteval_t is equivalent to all the other *val_t types. */
389 static pteval_t pte_mfn_to_pfn(pteval_t val)
391 if (val & _PAGE_PRESENT) {
392 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
393 pteval_t flags = val & PTE_FLAGS_MASK;
394 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
400 static pteval_t pte_pfn_to_mfn(pteval_t val)
402 if (val & _PAGE_PRESENT) {
403 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
404 pteval_t flags = val & PTE_FLAGS_MASK;
407 if (!xen_feature(XENFEAT_auto_translated_physmap))
408 mfn = get_phys_to_machine(pfn);
412 * If there's no mfn for the pfn, then just create an
413 * empty non-present pte. Unfortunately this loses
414 * information about the original pfn, so
415 * pte_mfn_to_pfn is asymmetric.
417 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
422 * Paramount to do this test _after_ the
423 * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
424 * IDENTITY_FRAME_BIT resolves to true.
426 mfn &= ~FOREIGN_FRAME_BIT;
427 if (mfn & IDENTITY_FRAME_BIT) {
428 mfn &= ~IDENTITY_FRAME_BIT;
429 flags |= _PAGE_IOMAP;
432 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
438 static pteval_t iomap_pte(pteval_t val)
440 if (val & _PAGE_PRESENT) {
441 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
442 pteval_t flags = val & PTE_FLAGS_MASK;
444 /* We assume the pte frame number is a MFN, so
445 just use it as-is. */
446 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
452 static pteval_t xen_pte_val(pte_t pte)
454 pteval_t pteval = pte.pte;
456 /* If this is a WC pte, convert back from Xen WC to Linux WC */
457 if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
458 WARN_ON(!pat_enabled);
459 pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
462 if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
465 return pte_mfn_to_pfn(pteval);
467 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
469 static pgdval_t xen_pgd_val(pgd_t pgd)
471 return pte_mfn_to_pfn(pgd.pgd);
473 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
476 * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
477 * are reserved for now, to correspond to the Intel-reserved PAT
480 * We expect Linux's PAT set as follows:
482 * Idx PTE flags Linux Xen Default
489 * 6 PAT PCD UC- UC UC-
490 * 7 PAT PCD PWT UC UC UC
493 void xen_set_pat(u64 pat)
495 /* We expect Linux to use a PAT setting of
496 * UC UC- WC WB (ignoring the PAT flag) */
497 WARN_ON(pat != 0x0007010600070106ull);
500 static pte_t xen_make_pte(pteval_t pte)
502 phys_addr_t addr = (pte & PTE_PFN_MASK);
504 /* If Linux is trying to set a WC pte, then map to the Xen WC.
505 * If _PAGE_PAT is set, then it probably means it is really
506 * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
507 * things work out OK...
509 * (We should never see kernel mappings with _PAGE_PSE set,
510 * but we could see hugetlbfs mappings, I think.).
512 if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
513 if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
514 pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
518 * Unprivileged domains are allowed to do IOMAPpings for
519 * PCI passthrough, but not map ISA space. The ISA
520 * mappings are just dummy local mappings to keep other
521 * parts of the kernel happy.
523 if (unlikely(pte & _PAGE_IOMAP) &&
524 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
525 pte = iomap_pte(pte);
528 pte = pte_pfn_to_mfn(pte);
531 return native_make_pte(pte);
533 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
535 #ifdef CONFIG_XEN_DEBUG
536 pte_t xen_make_pte_debug(pteval_t pte)
538 phys_addr_t addr = (pte & PTE_PFN_MASK);
539 phys_addr_t other_addr;
540 bool io_page = false;
543 if (pte & _PAGE_IOMAP)
546 _pte = xen_make_pte(pte);
552 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
553 other_addr = pfn_to_mfn(addr >> PAGE_SHIFT) << PAGE_SHIFT;
554 WARN_ONCE(addr != other_addr,
555 "0x%lx is using VM_IO, but it is 0x%lx!\n",
556 (unsigned long)addr, (unsigned long)other_addr);
558 pteval_t iomap_set = (_pte.pte & PTE_FLAGS_MASK) & _PAGE_IOMAP;
559 other_addr = (_pte.pte & PTE_PFN_MASK);
560 WARN_ONCE((addr == other_addr) && (!io_page) && (!iomap_set),
561 "0x%lx is missing VM_IO (and wasn't fixed)!\n",
562 (unsigned long)addr);
567 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_debug);
570 static pgd_t xen_make_pgd(pgdval_t pgd)
572 pgd = pte_pfn_to_mfn(pgd);
573 return native_make_pgd(pgd);
575 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
577 static pmdval_t xen_pmd_val(pmd_t pmd)
579 return pte_mfn_to_pfn(pmd.pmd);
581 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
583 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
591 /* ptr may be ioremapped for 64-bit pagetable setup */
592 u.ptr = virt_to_machine(ptr).maddr;
593 u.val = pud_val_ma(val);
594 xen_extend_mmu_update(&u);
596 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
598 xen_mc_issue(PARAVIRT_LAZY_MMU);
603 static void xen_set_pud(pud_t *ptr, pud_t val)
605 ADD_STATS(pud_update, 1);
607 /* If page is not pinned, we can just update the entry
609 if (!xen_page_pinned(ptr)) {
614 ADD_STATS(pud_update_pinned, 1);
616 xen_set_pud_hyper(ptr, val);
619 #ifdef CONFIG_X86_PAE
620 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
622 set_64bit((u64 *)ptep, native_pte_val(pte));
625 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
627 if (!xen_batched_set_pte(ptep, native_make_pte(0)))
628 native_pte_clear(mm, addr, ptep);
631 static void xen_pmd_clear(pmd_t *pmdp)
633 set_pmd(pmdp, __pmd(0));
635 #endif /* CONFIG_X86_PAE */
637 static pmd_t xen_make_pmd(pmdval_t pmd)
639 pmd = pte_pfn_to_mfn(pmd);
640 return native_make_pmd(pmd);
642 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
644 #if PAGETABLE_LEVELS == 4
645 static pudval_t xen_pud_val(pud_t pud)
647 return pte_mfn_to_pfn(pud.pud);
649 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
651 static pud_t xen_make_pud(pudval_t pud)
653 pud = pte_pfn_to_mfn(pud);
655 return native_make_pud(pud);
657 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
659 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
661 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
662 unsigned offset = pgd - pgd_page;
663 pgd_t *user_ptr = NULL;
665 if (offset < pgd_index(USER_LIMIT)) {
666 struct page *page = virt_to_page(pgd_page);
667 user_ptr = (pgd_t *)page->private;
675 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
679 u.ptr = virt_to_machine(ptr).maddr;
680 u.val = pgd_val_ma(val);
681 xen_extend_mmu_update(&u);
685 * Raw hypercall-based set_pgd, intended for in early boot before
686 * there's a page structure. This implies:
687 * 1. The only existing pagetable is the kernel's
688 * 2. It is always pinned
689 * 3. It has no user pagetable attached to it
691 static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
697 __xen_set_pgd_hyper(ptr, val);
699 xen_mc_issue(PARAVIRT_LAZY_MMU);
704 static void xen_set_pgd(pgd_t *ptr, pgd_t val)
706 pgd_t *user_ptr = xen_get_user_pgd(ptr);
708 ADD_STATS(pgd_update, 1);
710 /* If page is not pinned, we can just update the entry
712 if (!xen_page_pinned(ptr)) {
715 WARN_ON(xen_page_pinned(user_ptr));
721 ADD_STATS(pgd_update_pinned, 1);
722 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
724 /* If it's pinned, then we can at least batch the kernel and
725 user updates together. */
728 __xen_set_pgd_hyper(ptr, val);
730 __xen_set_pgd_hyper(user_ptr, val);
732 xen_mc_issue(PARAVIRT_LAZY_MMU);
734 #endif /* PAGETABLE_LEVELS == 4 */
737 * (Yet another) pagetable walker. This one is intended for pinning a
738 * pagetable. This means that it walks a pagetable and calls the
739 * callback function on each page it finds making up the page table,
740 * at every level. It walks the entire pagetable, but it only bothers
741 * pinning pte pages which are below limit. In the normal case this
742 * will be STACK_TOP_MAX, but at boot we need to pin up to
745 * For 32-bit the important bit is that we don't pin beyond there,
746 * because then we start getting into Xen's ptes.
748 * For 64-bit, we must skip the Xen hole in the middle of the address
749 * space, just after the big x86-64 virtual hole.
751 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
752 int (*func)(struct mm_struct *mm, struct page *,
757 unsigned hole_low, hole_high;
758 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
759 unsigned pgdidx, pudidx, pmdidx;
761 /* The limit is the last byte to be touched */
763 BUG_ON(limit >= FIXADDR_TOP);
765 if (xen_feature(XENFEAT_auto_translated_physmap))
769 * 64-bit has a great big hole in the middle of the address
770 * space, which contains the Xen mappings. On 32-bit these
771 * will end up making a zero-sized hole and so is a no-op.
773 hole_low = pgd_index(USER_LIMIT);
774 hole_high = pgd_index(PAGE_OFFSET);
776 pgdidx_limit = pgd_index(limit);
778 pudidx_limit = pud_index(limit);
783 pmdidx_limit = pmd_index(limit);
788 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
791 if (pgdidx >= hole_low && pgdidx < hole_high)
794 if (!pgd_val(pgd[pgdidx]))
797 pud = pud_offset(&pgd[pgdidx], 0);
799 if (PTRS_PER_PUD > 1) /* not folded */
800 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
802 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
805 if (pgdidx == pgdidx_limit &&
806 pudidx > pudidx_limit)
809 if (pud_none(pud[pudidx]))
812 pmd = pmd_offset(&pud[pudidx], 0);
814 if (PTRS_PER_PMD > 1) /* not folded */
815 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
817 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
820 if (pgdidx == pgdidx_limit &&
821 pudidx == pudidx_limit &&
822 pmdidx > pmdidx_limit)
825 if (pmd_none(pmd[pmdidx]))
828 pte = pmd_page(pmd[pmdidx]);
829 flush |= (*func)(mm, pte, PT_PTE);
835 /* Do the top level last, so that the callbacks can use it as
836 a cue to do final things like tlb flushes. */
837 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
842 static int xen_pgd_walk(struct mm_struct *mm,
843 int (*func)(struct mm_struct *mm, struct page *,
847 return __xen_pgd_walk(mm, mm->pgd, func, limit);
850 /* If we're using split pte locks, then take the page's lock and
851 return a pointer to it. Otherwise return NULL. */
852 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
854 spinlock_t *ptl = NULL;
856 #if USE_SPLIT_PTLOCKS
857 ptl = __pte_lockptr(page);
858 spin_lock_nest_lock(ptl, &mm->page_table_lock);
864 static void xen_pte_unlock(void *v)
870 static void xen_do_pin(unsigned level, unsigned long pfn)
872 struct mmuext_op *op;
873 struct multicall_space mcs;
875 mcs = __xen_mc_entry(sizeof(*op));
878 op->arg1.mfn = pfn_to_mfn(pfn);
879 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
882 static int xen_pin_page(struct mm_struct *mm, struct page *page,
885 unsigned pgfl = TestSetPagePinned(page);
889 flush = 0; /* already pinned */
890 else if (PageHighMem(page))
891 /* kmaps need flushing if we found an unpinned
895 void *pt = lowmem_page_address(page);
896 unsigned long pfn = page_to_pfn(page);
897 struct multicall_space mcs = __xen_mc_entry(0);
903 * We need to hold the pagetable lock between the time
904 * we make the pagetable RO and when we actually pin
905 * it. If we don't, then other users may come in and
906 * attempt to update the pagetable by writing it,
907 * which will fail because the memory is RO but not
908 * pinned, so Xen won't do the trap'n'emulate.
910 * If we're using split pte locks, we can't hold the
911 * entire pagetable's worth of locks during the
912 * traverse, because we may wrap the preempt count (8
913 * bits). The solution is to mark RO and pin each PTE
914 * page while holding the lock. This means the number
915 * of locks we end up holding is never more than a
916 * batch size (~32 entries, at present).
918 * If we're not using split pte locks, we needn't pin
919 * the PTE pages independently, because we're
920 * protected by the overall pagetable lock.
924 ptl = xen_pte_lock(page, mm);
926 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
927 pfn_pte(pfn, PAGE_KERNEL_RO),
928 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
931 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
933 /* Queue a deferred unlock for when this batch
935 xen_mc_callback(xen_pte_unlock, ptl);
942 /* This is called just after a mm has been created, but it has not
943 been used yet. We need to make sure that its pagetable is all
944 read-only, and can be pinned. */
945 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
949 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
950 /* re-enable interrupts for flushing */
960 pgd_t *user_pgd = xen_get_user_pgd(pgd);
962 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
965 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
966 xen_do_pin(MMUEXT_PIN_L4_TABLE,
967 PFN_DOWN(__pa(user_pgd)));
970 #else /* CONFIG_X86_32 */
971 #ifdef CONFIG_X86_PAE
972 /* Need to make sure unshared kernel PMD is pinnable */
973 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
976 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
977 #endif /* CONFIG_X86_64 */
981 static void xen_pgd_pin(struct mm_struct *mm)
983 __xen_pgd_pin(mm, mm->pgd);
987 * On save, we need to pin all pagetables to make sure they get their
988 * mfns turned into pfns. Search the list for any unpinned pgds and pin
989 * them (unpinned pgds are not currently in use, probably because the
990 * process is under construction or destruction).
992 * Expected to be called in stop_machine() ("equivalent to taking
993 * every spinlock in the system"), so the locking doesn't really
994 * matter all that much.
996 void xen_mm_pin_all(void)
1000 spin_lock(&pgd_lock);
1002 list_for_each_entry(page, &pgd_list, lru) {
1003 if (!PagePinned(page)) {
1004 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1005 SetPageSavePinned(page);
1009 spin_unlock(&pgd_lock);
1013 * The init_mm pagetable is really pinned as soon as its created, but
1014 * that's before we have page structures to store the bits. So do all
1015 * the book-keeping now.
1017 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1018 enum pt_level level)
1020 SetPagePinned(page);
1024 static void __init xen_mark_init_mm_pinned(void)
1026 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1029 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1030 enum pt_level level)
1032 unsigned pgfl = TestClearPagePinned(page);
1034 if (pgfl && !PageHighMem(page)) {
1035 void *pt = lowmem_page_address(page);
1036 unsigned long pfn = page_to_pfn(page);
1037 spinlock_t *ptl = NULL;
1038 struct multicall_space mcs;
1041 * Do the converse to pin_page. If we're using split
1042 * pte locks, we must be holding the lock for while
1043 * the pte page is unpinned but still RO to prevent
1044 * concurrent updates from seeing it in this
1045 * partially-pinned state.
1047 if (level == PT_PTE) {
1048 ptl = xen_pte_lock(page, mm);
1051 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1054 mcs = __xen_mc_entry(0);
1056 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1057 pfn_pte(pfn, PAGE_KERNEL),
1058 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1061 /* unlock when batch completed */
1062 xen_mc_callback(xen_pte_unlock, ptl);
1066 return 0; /* never need to flush on unpin */
1069 /* Release a pagetables pages back as normal RW */
1070 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1074 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1076 #ifdef CONFIG_X86_64
1078 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1081 xen_do_pin(MMUEXT_UNPIN_TABLE,
1082 PFN_DOWN(__pa(user_pgd)));
1083 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1088 #ifdef CONFIG_X86_PAE
1089 /* Need to make sure unshared kernel PMD is unpinned */
1090 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1094 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1099 static void xen_pgd_unpin(struct mm_struct *mm)
1101 __xen_pgd_unpin(mm, mm->pgd);
1105 * On resume, undo any pinning done at save, so that the rest of the
1106 * kernel doesn't see any unexpected pinned pagetables.
1108 void xen_mm_unpin_all(void)
1112 spin_lock(&pgd_lock);
1114 list_for_each_entry(page, &pgd_list, lru) {
1115 if (PageSavePinned(page)) {
1116 BUG_ON(!PagePinned(page));
1117 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1118 ClearPageSavePinned(page);
1122 spin_unlock(&pgd_lock);
1125 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1127 spin_lock(&next->page_table_lock);
1129 spin_unlock(&next->page_table_lock);
1132 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1134 spin_lock(&mm->page_table_lock);
1136 spin_unlock(&mm->page_table_lock);
1141 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1142 we need to repoint it somewhere else before we can unpin it. */
1143 static void drop_other_mm_ref(void *info)
1145 struct mm_struct *mm = info;
1146 struct mm_struct *active_mm;
1148 active_mm = percpu_read(cpu_tlbstate.active_mm);
1150 if (active_mm == mm)
1151 leave_mm(smp_processor_id());
1153 /* If this cpu still has a stale cr3 reference, then make sure
1154 it has been flushed. */
1155 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1156 load_cr3(swapper_pg_dir);
1159 static void xen_drop_mm_ref(struct mm_struct *mm)
1164 if (current->active_mm == mm) {
1165 if (current->mm == mm)
1166 load_cr3(swapper_pg_dir);
1168 leave_mm(smp_processor_id());
1171 /* Get the "official" set of cpus referring to our pagetable. */
1172 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1173 for_each_online_cpu(cpu) {
1174 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1175 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1177 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1181 cpumask_copy(mask, mm_cpumask(mm));
1183 /* It's possible that a vcpu may have a stale reference to our
1184 cr3, because its in lazy mode, and it hasn't yet flushed
1185 its set of pending hypercalls yet. In this case, we can
1186 look at its actual current cr3 value, and force it to flush
1188 for_each_online_cpu(cpu) {
1189 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1190 cpumask_set_cpu(cpu, mask);
1193 if (!cpumask_empty(mask))
1194 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1195 free_cpumask_var(mask);
1198 static void xen_drop_mm_ref(struct mm_struct *mm)
1200 if (current->active_mm == mm)
1201 load_cr3(swapper_pg_dir);
1206 * While a process runs, Xen pins its pagetables, which means that the
1207 * hypervisor forces it to be read-only, and it controls all updates
1208 * to it. This means that all pagetable updates have to go via the
1209 * hypervisor, which is moderately expensive.
1211 * Since we're pulling the pagetable down, we switch to use init_mm,
1212 * unpin old process pagetable and mark it all read-write, which
1213 * allows further operations on it to be simple memory accesses.
1215 * The only subtle point is that another CPU may be still using the
1216 * pagetable because of lazy tlb flushing. This means we need need to
1217 * switch all CPUs off this pagetable before we can unpin it.
1219 static void xen_exit_mmap(struct mm_struct *mm)
1221 get_cpu(); /* make sure we don't move around */
1222 xen_drop_mm_ref(mm);
1225 spin_lock(&mm->page_table_lock);
1227 /* pgd may not be pinned in the error exit path of execve */
1228 if (xen_page_pinned(mm->pgd))
1231 spin_unlock(&mm->page_table_lock);
1234 static __init void xen_pagetable_setup_start(pgd_t *base)
1238 static __init void xen_mapping_pagetable_reserve(u64 start, u64 end)
1240 /* reserve the range used */
1241 native_pagetable_reserve(start, end);
1243 /* set as RW the rest */
1244 printk(KERN_DEBUG "xen: setting RW the range %llx - %llx\n", end,
1245 PFN_PHYS(pgt_buf_top));
1246 while (end < PFN_PHYS(pgt_buf_top)) {
1247 make_lowmem_page_readwrite(__va(end));
1252 static void xen_post_allocator_init(void);
1254 static __init void xen_pagetable_setup_done(pgd_t *base)
1256 xen_setup_shared_info();
1257 xen_post_allocator_init();
1260 static void xen_write_cr2(unsigned long cr2)
1262 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1265 static unsigned long xen_read_cr2(void)
1267 return percpu_read(xen_vcpu)->arch.cr2;
1270 unsigned long xen_read_cr2_direct(void)
1272 return percpu_read(xen_vcpu_info.arch.cr2);
1275 static void xen_flush_tlb(void)
1277 struct mmuext_op *op;
1278 struct multicall_space mcs;
1282 mcs = xen_mc_entry(sizeof(*op));
1285 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1286 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1288 xen_mc_issue(PARAVIRT_LAZY_MMU);
1293 static void xen_flush_tlb_single(unsigned long addr)
1295 struct mmuext_op *op;
1296 struct multicall_space mcs;
1300 mcs = xen_mc_entry(sizeof(*op));
1302 op->cmd = MMUEXT_INVLPG_LOCAL;
1303 op->arg1.linear_addr = addr & PAGE_MASK;
1304 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1306 xen_mc_issue(PARAVIRT_LAZY_MMU);
1311 static void xen_flush_tlb_others(const struct cpumask *cpus,
1312 struct mm_struct *mm, unsigned long va)
1315 struct mmuext_op op;
1316 DECLARE_BITMAP(mask, NR_CPUS);
1318 struct multicall_space mcs;
1320 if (cpumask_empty(cpus))
1321 return; /* nothing to do */
1323 mcs = xen_mc_entry(sizeof(*args));
1325 args->op.arg2.vcpumask = to_cpumask(args->mask);
1327 /* Remove us, and any offline CPUS. */
1328 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1329 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1331 if (va == TLB_FLUSH_ALL) {
1332 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1334 args->op.cmd = MMUEXT_INVLPG_MULTI;
1335 args->op.arg1.linear_addr = va;
1338 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1340 xen_mc_issue(PARAVIRT_LAZY_MMU);
1343 static unsigned long xen_read_cr3(void)
1345 return percpu_read(xen_cr3);
1348 static void set_current_cr3(void *v)
1350 percpu_write(xen_current_cr3, (unsigned long)v);
1353 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1355 struct mmuext_op *op;
1356 struct multicall_space mcs;
1360 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1364 WARN_ON(mfn == 0 && kernel);
1366 mcs = __xen_mc_entry(sizeof(*op));
1369 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1372 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1375 percpu_write(xen_cr3, cr3);
1377 /* Update xen_current_cr3 once the batch has actually
1379 xen_mc_callback(set_current_cr3, (void *)cr3);
1383 static void xen_write_cr3(unsigned long cr3)
1385 BUG_ON(preemptible());
1387 xen_mc_batch(); /* disables interrupts */
1389 /* Update while interrupts are disabled, so its atomic with
1391 percpu_write(xen_cr3, cr3);
1393 __xen_write_cr3(true, cr3);
1395 #ifdef CONFIG_X86_64
1397 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1399 __xen_write_cr3(false, __pa(user_pgd));
1401 __xen_write_cr3(false, 0);
1405 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1408 static int xen_pgd_alloc(struct mm_struct *mm)
1410 pgd_t *pgd = mm->pgd;
1413 BUG_ON(PagePinned(virt_to_page(pgd)));
1415 #ifdef CONFIG_X86_64
1417 struct page *page = virt_to_page(pgd);
1420 BUG_ON(page->private != 0);
1424 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1425 page->private = (unsigned long)user_pgd;
1427 if (user_pgd != NULL) {
1428 user_pgd[pgd_index(VSYSCALL_START)] =
1429 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1433 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1440 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1442 #ifdef CONFIG_X86_64
1443 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1446 free_page((unsigned long)user_pgd);
1450 #ifdef CONFIG_X86_32
1451 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1453 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1454 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1455 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1460 #else /* CONFIG_X86_64 */
1461 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1463 unsigned long pfn = pte_pfn(pte);
1466 * If the new pfn is within the range of the newly allocated
1467 * kernel pagetable, and it isn't being mapped into an
1468 * early_ioremap fixmap slot as a freshly allocated page, make sure
1471 if (((!is_early_ioremap_ptep(ptep) &&
1472 pfn >= pgt_buf_start && pfn < pgt_buf_top)) ||
1473 (is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1474 pte = pte_wrprotect(pte);
1478 #endif /* CONFIG_X86_64 */
1480 /* Init-time set_pte while constructing initial pagetables, which
1481 doesn't allow RO pagetable pages to be remapped RW */
1482 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1484 pte = mask_rw_pte(ptep, pte);
1486 xen_set_pte(ptep, pte);
1489 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1491 struct mmuext_op op;
1493 op.arg1.mfn = pfn_to_mfn(pfn);
1494 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1498 /* Early in boot, while setting up the initial pagetable, assume
1499 everything is pinned. */
1500 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1502 #ifdef CONFIG_FLATMEM
1503 BUG_ON(mem_map); /* should only be used early */
1505 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1506 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1509 /* Used for pmd and pud */
1510 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1512 #ifdef CONFIG_FLATMEM
1513 BUG_ON(mem_map); /* should only be used early */
1515 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1518 /* Early release_pte assumes that all pts are pinned, since there's
1519 only init_mm and anything attached to that is pinned. */
1520 static __init void xen_release_pte_init(unsigned long pfn)
1522 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1523 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1526 static __init void xen_release_pmd_init(unsigned long pfn)
1528 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1531 /* This needs to make sure the new pte page is pinned iff its being
1532 attached to a pinned pagetable. */
1533 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1535 struct page *page = pfn_to_page(pfn);
1537 if (PagePinned(virt_to_page(mm->pgd))) {
1538 SetPagePinned(page);
1540 if (!PageHighMem(page)) {
1541 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1542 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1543 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1545 /* make sure there are no stray mappings of
1547 kmap_flush_unused();
1552 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1554 xen_alloc_ptpage(mm, pfn, PT_PTE);
1557 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1559 xen_alloc_ptpage(mm, pfn, PT_PMD);
1562 /* This should never happen until we're OK to use struct page */
1563 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1565 struct page *page = pfn_to_page(pfn);
1567 if (PagePinned(page)) {
1568 if (!PageHighMem(page)) {
1569 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1570 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1571 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1573 ClearPagePinned(page);
1577 static void xen_release_pte(unsigned long pfn)
1579 xen_release_ptpage(pfn, PT_PTE);
1582 static void xen_release_pmd(unsigned long pfn)
1584 xen_release_ptpage(pfn, PT_PMD);
1587 #if PAGETABLE_LEVELS == 4
1588 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1590 xen_alloc_ptpage(mm, pfn, PT_PUD);
1593 static void xen_release_pud(unsigned long pfn)
1595 xen_release_ptpage(pfn, PT_PUD);
1599 void __init xen_reserve_top(void)
1601 #ifdef CONFIG_X86_32
1602 unsigned long top = HYPERVISOR_VIRT_START;
1603 struct xen_platform_parameters pp;
1605 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1606 top = pp.virt_start;
1608 reserve_top_address(-top);
1609 #endif /* CONFIG_X86_32 */
1613 * Like __va(), but returns address in the kernel mapping (which is
1614 * all we have until the physical memory mapping has been set up.
1616 static void *__ka(phys_addr_t paddr)
1618 #ifdef CONFIG_X86_64
1619 return (void *)(paddr + __START_KERNEL_map);
1625 /* Convert a machine address to physical address */
1626 static unsigned long m2p(phys_addr_t maddr)
1630 maddr &= PTE_PFN_MASK;
1631 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1636 /* Convert a machine address to kernel virtual */
1637 static void *m2v(phys_addr_t maddr)
1639 return __ka(m2p(maddr));
1642 /* Set the page permissions on an identity-mapped pages */
1643 static void set_page_prot(void *addr, pgprot_t prot)
1645 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1646 pte_t pte = pfn_pte(pfn, prot);
1648 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1652 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1654 unsigned pmdidx, pteidx;
1658 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1663 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1666 /* Reuse or allocate a page of ptes */
1667 if (pmd_present(pmd[pmdidx]))
1668 pte_page = m2v(pmd[pmdidx].pmd);
1670 /* Check for free pte pages */
1671 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1674 pte_page = &level1_ident_pgt[ident_pte];
1675 ident_pte += PTRS_PER_PTE;
1677 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1680 /* Install mappings */
1681 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1684 if (!pte_none(pte_page[pteidx]))
1687 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1688 pte_page[pteidx] = pte;
1692 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1693 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1695 set_page_prot(pmd, PAGE_KERNEL_RO);
1698 void __init xen_setup_machphys_mapping(void)
1700 struct xen_machphys_mapping mapping;
1701 unsigned long machine_to_phys_nr_ents;
1703 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1704 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1705 machine_to_phys_nr_ents = mapping.max_mfn + 1;
1707 machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
1709 machine_to_phys_order = fls(machine_to_phys_nr_ents - 1);
1712 #ifdef CONFIG_X86_64
1713 static void convert_pfn_mfn(void *v)
1718 /* All levels are converted the same way, so just treat them
1720 for (i = 0; i < PTRS_PER_PTE; i++)
1721 pte[i] = xen_make_pte(pte[i].pte);
1725 * Set up the initial kernel pagetable.
1727 * We can construct this by grafting the Xen provided pagetable into
1728 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1729 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1730 * means that only the kernel has a physical mapping to start with -
1731 * but that's enough to get __va working. We need to fill in the rest
1732 * of the physical mapping once some sort of allocator has been set
1735 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1736 unsigned long max_pfn)
1741 /* max_pfn_mapped is the last pfn mapped in the initial memory
1742 * mappings. Considering that on Xen after the kernel mappings we
1743 * have the mappings of some pages that don't exist in pfn space, we
1744 * set max_pfn_mapped to the last real pfn mapped. */
1745 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1747 /* Zap identity mapping */
1748 init_level4_pgt[0] = __pgd(0);
1750 /* Pre-constructed entries are in pfn, so convert to mfn */
1751 convert_pfn_mfn(init_level4_pgt);
1752 convert_pfn_mfn(level3_ident_pgt);
1753 convert_pfn_mfn(level3_kernel_pgt);
1755 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1756 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1758 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1759 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1761 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1762 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1763 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1765 /* Set up identity map */
1766 xen_map_identity_early(level2_ident_pgt, max_pfn);
1768 /* Make pagetable pieces RO */
1769 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1770 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1771 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1772 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1773 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1774 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1776 /* Pin down new L4 */
1777 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1778 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1780 /* Unpin Xen-provided one */
1781 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1784 pgd = init_level4_pgt;
1787 * At this stage there can be no user pgd, and no page
1788 * structure to attach it to, so make sure we just set kernel
1792 __xen_write_cr3(true, __pa(pgd));
1793 xen_mc_issue(PARAVIRT_LAZY_CPU);
1795 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1796 __pa(xen_start_info->pt_base +
1797 xen_start_info->nr_pt_frames * PAGE_SIZE),
1802 #else /* !CONFIG_X86_64 */
1803 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1804 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1806 static __init void xen_write_cr3_init(unsigned long cr3)
1808 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1810 BUG_ON(read_cr3() != __pa(initial_page_table));
1811 BUG_ON(cr3 != __pa(swapper_pg_dir));
1814 * We are switching to swapper_pg_dir for the first time (from
1815 * initial_page_table) and therefore need to mark that page
1816 * read-only and then pin it.
1818 * Xen disallows sharing of kernel PMDs for PAE
1819 * guests. Therefore we must copy the kernel PMD from
1820 * initial_page_table into a new kernel PMD to be used in
1823 swapper_kernel_pmd =
1824 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1825 memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1826 sizeof(pmd_t) * PTRS_PER_PMD);
1827 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1828 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1829 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1831 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1833 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1835 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1836 PFN_DOWN(__pa(initial_page_table)));
1837 set_page_prot(initial_page_table, PAGE_KERNEL);
1838 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1840 pv_mmu_ops.write_cr3 = &xen_write_cr3;
1843 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1844 unsigned long max_pfn)
1848 initial_kernel_pmd =
1849 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1851 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1853 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1854 memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1856 xen_map_identity_early(initial_kernel_pmd, max_pfn);
1858 memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1859 initial_page_table[KERNEL_PGD_BOUNDARY] =
1860 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1862 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1863 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1864 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1866 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1868 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1869 PFN_DOWN(__pa(initial_page_table)));
1870 xen_write_cr3(__pa(initial_page_table));
1872 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1873 __pa(xen_start_info->pt_base +
1874 xen_start_info->nr_pt_frames * PAGE_SIZE),
1877 return initial_page_table;
1879 #endif /* CONFIG_X86_64 */
1881 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1883 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1887 phys >>= PAGE_SHIFT;
1890 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1891 #ifdef CONFIG_X86_F00F_BUG
1894 #ifdef CONFIG_X86_32
1897 # ifdef CONFIG_HIGHMEM
1898 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1901 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1903 case FIX_TEXT_POKE0:
1904 case FIX_TEXT_POKE1:
1905 /* All local page mappings */
1906 pte = pfn_pte(phys, prot);
1909 #ifdef CONFIG_X86_LOCAL_APIC
1910 case FIX_APIC_BASE: /* maps dummy local APIC */
1911 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1915 #ifdef CONFIG_X86_IO_APIC
1916 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1918 * We just don't map the IO APIC - all access is via
1919 * hypercalls. Keep the address in the pte for reference.
1921 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1925 case FIX_PARAVIRT_BOOTMAP:
1926 /* This is an MFN, but it isn't an IO mapping from the
1928 pte = mfn_pte(phys, prot);
1932 /* By default, set_fixmap is used for hardware mappings */
1933 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1937 __native_set_fixmap(idx, pte);
1939 #ifdef CONFIG_X86_64
1940 /* Replicate changes to map the vsyscall page into the user
1941 pagetable vsyscall mapping. */
1942 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1943 unsigned long vaddr = __fix_to_virt(idx);
1944 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1949 __init void xen_ident_map_ISA(void)
1954 * If we're dom0, then linear map the ISA machine addresses into
1955 * the kernel's address space.
1957 if (!xen_initial_domain())
1960 xen_raw_printk("Xen: setup ISA identity maps\n");
1962 for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1963 pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1965 if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1972 static __init void xen_post_allocator_init(void)
1974 #ifdef CONFIG_XEN_DEBUG
1975 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte_debug);
1977 pv_mmu_ops.set_pte = xen_set_pte;
1978 pv_mmu_ops.set_pmd = xen_set_pmd;
1979 pv_mmu_ops.set_pud = xen_set_pud;
1980 #if PAGETABLE_LEVELS == 4
1981 pv_mmu_ops.set_pgd = xen_set_pgd;
1984 /* This will work as long as patching hasn't happened yet
1985 (which it hasn't) */
1986 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1987 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1988 pv_mmu_ops.release_pte = xen_release_pte;
1989 pv_mmu_ops.release_pmd = xen_release_pmd;
1990 #if PAGETABLE_LEVELS == 4
1991 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1992 pv_mmu_ops.release_pud = xen_release_pud;
1995 #ifdef CONFIG_X86_64
1996 SetPagePinned(virt_to_page(level3_user_vsyscall));
1998 xen_mark_init_mm_pinned();
2001 static void xen_leave_lazy_mmu(void)
2005 paravirt_leave_lazy_mmu();
2009 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
2010 .read_cr2 = xen_read_cr2,
2011 .write_cr2 = xen_write_cr2,
2013 .read_cr3 = xen_read_cr3,
2014 #ifdef CONFIG_X86_32
2015 .write_cr3 = xen_write_cr3_init,
2017 .write_cr3 = xen_write_cr3,
2020 .flush_tlb_user = xen_flush_tlb,
2021 .flush_tlb_kernel = xen_flush_tlb,
2022 .flush_tlb_single = xen_flush_tlb_single,
2023 .flush_tlb_others = xen_flush_tlb_others,
2025 .pte_update = paravirt_nop,
2026 .pte_update_defer = paravirt_nop,
2028 .pgd_alloc = xen_pgd_alloc,
2029 .pgd_free = xen_pgd_free,
2031 .alloc_pte = xen_alloc_pte_init,
2032 .release_pte = xen_release_pte_init,
2033 .alloc_pmd = xen_alloc_pmd_init,
2034 .release_pmd = xen_release_pmd_init,
2036 .set_pte = xen_set_pte_init,
2037 .set_pte_at = xen_set_pte_at,
2038 .set_pmd = xen_set_pmd_hyper,
2040 .ptep_modify_prot_start = __ptep_modify_prot_start,
2041 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2043 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2044 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2046 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2047 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2049 #ifdef CONFIG_X86_PAE
2050 .set_pte_atomic = xen_set_pte_atomic,
2051 .pte_clear = xen_pte_clear,
2052 .pmd_clear = xen_pmd_clear,
2053 #endif /* CONFIG_X86_PAE */
2054 .set_pud = xen_set_pud_hyper,
2056 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2057 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2059 #if PAGETABLE_LEVELS == 4
2060 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2061 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2062 .set_pgd = xen_set_pgd_hyper,
2064 .alloc_pud = xen_alloc_pmd_init,
2065 .release_pud = xen_release_pmd_init,
2066 #endif /* PAGETABLE_LEVELS == 4 */
2068 .activate_mm = xen_activate_mm,
2069 .dup_mmap = xen_dup_mmap,
2070 .exit_mmap = xen_exit_mmap,
2073 .enter = paravirt_enter_lazy_mmu,
2074 .leave = xen_leave_lazy_mmu,
2077 .set_fixmap = xen_set_fixmap,
2080 void __init xen_init_mmu_ops(void)
2082 x86_init.mapping.pagetable_reserve = xen_mapping_pagetable_reserve;
2083 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2084 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2085 pv_mmu_ops = xen_mmu_ops;
2087 memset(dummy_mapping, 0xff, PAGE_SIZE);
2090 /* Protected by xen_reservation_lock. */
2091 #define MAX_CONTIG_ORDER 9 /* 2MB */
2092 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2094 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2095 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2096 unsigned long *in_frames,
2097 unsigned long *out_frames)
2100 struct multicall_space mcs;
2103 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2104 mcs = __xen_mc_entry(0);
2107 in_frames[i] = virt_to_mfn(vaddr);
2109 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2110 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2113 out_frames[i] = virt_to_pfn(vaddr);
2119 * Update the pfn-to-mfn mappings for a virtual address range, either to
2120 * point to an array of mfns, or contiguously from a single starting
2123 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2124 unsigned long *mfns,
2125 unsigned long first_mfn)
2132 limit = 1u << order;
2133 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2134 struct multicall_space mcs;
2137 mcs = __xen_mc_entry(0);
2141 mfn = first_mfn + i;
2143 if (i < (limit - 1))
2147 flags = UVMF_INVLPG | UVMF_ALL;
2149 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2152 MULTI_update_va_mapping(mcs.mc, vaddr,
2153 mfn_pte(mfn, PAGE_KERNEL), flags);
2155 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2162 * Perform the hypercall to exchange a region of our pfns to point to
2163 * memory with the required contiguous alignment. Takes the pfns as
2164 * input, and populates mfns as output.
2166 * Returns a success code indicating whether the hypervisor was able to
2167 * satisfy the request or not.
2169 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2170 unsigned long *pfns_in,
2171 unsigned long extents_out,
2172 unsigned int order_out,
2173 unsigned long *mfns_out,
2174 unsigned int address_bits)
2179 struct xen_memory_exchange exchange = {
2181 .nr_extents = extents_in,
2182 .extent_order = order_in,
2183 .extent_start = pfns_in,
2187 .nr_extents = extents_out,
2188 .extent_order = order_out,
2189 .extent_start = mfns_out,
2190 .address_bits = address_bits,
2195 BUG_ON(extents_in << order_in != extents_out << order_out);
2197 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2198 success = (exchange.nr_exchanged == extents_in);
2200 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2201 BUG_ON(success && (rc != 0));
2206 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2207 unsigned int address_bits)
2209 unsigned long *in_frames = discontig_frames, out_frame;
2210 unsigned long flags;
2214 * Currently an auto-translated guest will not perform I/O, nor will
2215 * it require PAE page directories below 4GB. Therefore any calls to
2216 * this function are redundant and can be ignored.
2219 if (xen_feature(XENFEAT_auto_translated_physmap))
2222 if (unlikely(order > MAX_CONTIG_ORDER))
2225 memset((void *) vstart, 0, PAGE_SIZE << order);
2227 spin_lock_irqsave(&xen_reservation_lock, flags);
2229 /* 1. Zap current PTEs, remembering MFNs. */
2230 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2232 /* 2. Get a new contiguous memory extent. */
2233 out_frame = virt_to_pfn(vstart);
2234 success = xen_exchange_memory(1UL << order, 0, in_frames,
2235 1, order, &out_frame,
2238 /* 3. Map the new extent in place of old pages. */
2240 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2242 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2244 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2246 return success ? 0 : -ENOMEM;
2248 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2250 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2252 unsigned long *out_frames = discontig_frames, in_frame;
2253 unsigned long flags;
2256 if (xen_feature(XENFEAT_auto_translated_physmap))
2259 if (unlikely(order > MAX_CONTIG_ORDER))
2262 memset((void *) vstart, 0, PAGE_SIZE << order);
2264 spin_lock_irqsave(&xen_reservation_lock, flags);
2266 /* 1. Find start MFN of contiguous extent. */
2267 in_frame = virt_to_mfn(vstart);
2269 /* 2. Zap current PTEs. */
2270 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2272 /* 3. Do the exchange for non-contiguous MFNs. */
2273 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2276 /* 4. Map new pages in place of old pages. */
2278 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2280 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2282 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2284 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2286 #ifdef CONFIG_XEN_PVHVM
2287 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2289 struct xen_hvm_pagetable_dying a;
2292 a.domid = DOMID_SELF;
2293 a.gpa = __pa(mm->pgd);
2294 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2295 WARN_ON_ONCE(rc < 0);
2298 static int is_pagetable_dying_supported(void)
2300 struct xen_hvm_pagetable_dying a;
2303 a.domid = DOMID_SELF;
2305 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2307 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2313 void __init xen_hvm_init_mmu_ops(void)
2315 if (is_pagetable_dying_supported())
2316 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2320 #define REMAP_BATCH_SIZE 16
2325 struct mmu_update *mmu_update;
2328 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2329 unsigned long addr, void *data)
2331 struct remap_data *rmd = data;
2332 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2334 rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2335 rmd->mmu_update->val = pte_val_ma(pte);
2341 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2343 unsigned long mfn, int nr,
2344 pgprot_t prot, unsigned domid)
2346 struct remap_data rmd;
2347 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2349 unsigned long range;
2352 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2354 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2355 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2361 batch = min(REMAP_BATCH_SIZE, nr);
2362 range = (unsigned long)batch << PAGE_SHIFT;
2364 rmd.mmu_update = mmu_update;
2365 err = apply_to_page_range(vma->vm_mm, addr, range,
2366 remap_area_mfn_pte_fn, &rmd);
2371 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2385 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2387 #ifdef CONFIG_XEN_DEBUG_FS
2389 static int p2m_dump_open(struct inode *inode, struct file *filp)
2391 return single_open(filp, p2m_dump_show, NULL);
2394 static const struct file_operations p2m_dump_fops = {
2395 .open = p2m_dump_open,
2397 .llseek = seq_lseek,
2398 .release = single_release,
2401 static struct dentry *d_mmu_debug;
2403 static int __init xen_mmu_debugfs(void)
2405 struct dentry *d_xen = xen_init_debugfs();
2410 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2412 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2414 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2415 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2416 &mmu_stats.pgd_update_pinned);
2417 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2418 &mmu_stats.pgd_update_pinned);
2420 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2421 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2422 &mmu_stats.pud_update_pinned);
2423 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2424 &mmu_stats.pud_update_pinned);
2426 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2427 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2428 &mmu_stats.pmd_update_pinned);
2429 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2430 &mmu_stats.pmd_update_pinned);
2432 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2433 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2434 // &mmu_stats.pte_update_pinned);
2435 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2436 &mmu_stats.pte_update_pinned);
2438 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2439 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2440 &mmu_stats.mmu_update_extended);
2441 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2442 mmu_stats.mmu_update_histo, 20);
2444 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2445 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2446 &mmu_stats.prot_commit_batched);
2448 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
2451 fs_initcall(xen_mmu_debugfs);
2453 #endif /* CONFIG_XEN_DEBUG_FS */