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 driver_pages, and
85 DEFINE_SPINLOCK(xen_reservation_lock);
87 #ifdef CONFIG_XEN_DEBUG_FS
91 u32 pgd_update_pinned;
92 u32 pgd_update_batched;
95 u32 pud_update_pinned;
96 u32 pud_update_batched;
99 u32 pmd_update_pinned;
100 u32 pmd_update_batched;
103 u32 pte_update_pinned;
104 u32 pte_update_batched;
107 u32 mmu_update_extended;
108 u32 mmu_update_histo[MMU_UPDATE_HISTO];
111 u32 prot_commit_batched;
114 u32 set_pte_at_batched;
115 u32 set_pte_at_pinned;
116 u32 set_pte_at_current;
117 u32 set_pte_at_kernel;
120 static u8 zero_stats;
122 static inline void check_zero(void)
124 if (unlikely(zero_stats)) {
125 memset(&mmu_stats, 0, sizeof(mmu_stats));
130 #define ADD_STATS(elem, val) \
131 do { check_zero(); mmu_stats.elem += (val); } while(0)
133 #else /* !CONFIG_XEN_DEBUG_FS */
135 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
137 #endif /* CONFIG_XEN_DEBUG_FS */
141 * Identity map, in addition to plain kernel map. This needs to be
142 * large enough to allocate page table pages to allocate the rest.
143 * Each page can map 2MB.
145 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
146 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
149 /* l3 pud for userspace vsyscall mapping */
150 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
151 #endif /* CONFIG_X86_64 */
154 * Note about cr3 (pagetable base) values:
156 * xen_cr3 contains the current logical cr3 value; it contains the
157 * last set cr3. This may not be the current effective cr3, because
158 * its update may be being lazily deferred. However, a vcpu looking
159 * at its own cr3 can use this value knowing that it everything will
160 * be self-consistent.
162 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
163 * hypercall to set the vcpu cr3 is complete (so it may be a little
164 * out of date, but it will never be set early). If one vcpu is
165 * looking at another vcpu's cr3 value, it should use this variable.
167 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
168 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
172 * Just beyond the highest usermode address. STACK_TOP_MAX has a
173 * redzone above it, so round it up to a PGD boundary.
175 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
177 unsigned long arbitrary_virt_to_mfn(void *vaddr)
179 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
181 return PFN_DOWN(maddr.maddr);
184 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
186 unsigned long address = (unsigned long)vaddr;
192 * if the PFN is in the linear mapped vaddr range, we can just use
193 * the (quick) virt_to_machine() p2m lookup
195 if (virt_addr_valid(vaddr))
196 return virt_to_machine(vaddr);
198 /* otherwise we have to do a (slower) full page-table walk */
200 pte = lookup_address(address, &level);
202 offset = address & ~PAGE_MASK;
203 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
205 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
207 void make_lowmem_page_readonly(void *vaddr)
210 unsigned long address = (unsigned long)vaddr;
213 pte = lookup_address(address, &level);
215 return; /* vaddr missing */
217 ptev = pte_wrprotect(*pte);
219 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
223 void make_lowmem_page_readwrite(void *vaddr)
226 unsigned long address = (unsigned long)vaddr;
229 pte = lookup_address(address, &level);
231 return; /* vaddr missing */
233 ptev = pte_mkwrite(*pte);
235 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
240 static bool xen_page_pinned(void *ptr)
242 struct page *page = virt_to_page(ptr);
244 return PagePinned(page);
247 static bool xen_iomap_pte(pte_t pte)
249 return pte_flags(pte) & _PAGE_IOMAP;
252 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
254 struct multicall_space mcs;
255 struct mmu_update *u;
257 mcs = xen_mc_entry(sizeof(*u));
260 /* ptep might be kmapped when using 32-bit HIGHPTE */
261 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
262 u->val = pte_val_ma(pteval);
264 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
266 xen_mc_issue(PARAVIRT_LAZY_MMU);
268 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
270 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
272 xen_set_domain_pte(ptep, pteval, DOMID_IO);
275 static void xen_extend_mmu_update(const struct mmu_update *update)
277 struct multicall_space mcs;
278 struct mmu_update *u;
280 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
282 if (mcs.mc != NULL) {
283 ADD_STATS(mmu_update_extended, 1);
284 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
288 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
289 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
291 ADD_STATS(mmu_update_histo[0], 1);
293 ADD_STATS(mmu_update, 1);
294 mcs = __xen_mc_entry(sizeof(*u));
295 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
296 ADD_STATS(mmu_update_histo[1], 1);
303 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
311 /* ptr may be ioremapped for 64-bit pagetable setup */
312 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
313 u.val = pmd_val_ma(val);
314 xen_extend_mmu_update(&u);
316 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
318 xen_mc_issue(PARAVIRT_LAZY_MMU);
323 void xen_set_pmd(pmd_t *ptr, pmd_t val)
325 ADD_STATS(pmd_update, 1);
327 /* If page is not pinned, we can just update the entry
329 if (!xen_page_pinned(ptr)) {
334 ADD_STATS(pmd_update_pinned, 1);
336 xen_set_pmd_hyper(ptr, val);
340 * Associate a virtual page frame with a given physical page frame
341 * and protection flags for that frame.
343 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
345 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
348 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
349 pte_t *ptep, pte_t pteval)
351 if (xen_iomap_pte(pteval)) {
352 xen_set_iomap_pte(ptep, pteval);
356 ADD_STATS(set_pte_at, 1);
357 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
358 ADD_STATS(set_pte_at_current, mm == current->mm);
359 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
361 if (mm == current->mm || mm == &init_mm) {
362 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
363 struct multicall_space mcs;
364 mcs = xen_mc_entry(0);
366 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
367 ADD_STATS(set_pte_at_batched, 1);
368 xen_mc_issue(PARAVIRT_LAZY_MMU);
371 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
374 xen_set_pte(ptep, pteval);
379 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
380 unsigned long addr, pte_t *ptep)
382 /* Just return the pte as-is. We preserve the bits on commit */
386 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
387 pte_t *ptep, pte_t pte)
393 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
394 u.val = pte_val_ma(pte);
395 xen_extend_mmu_update(&u);
397 ADD_STATS(prot_commit, 1);
398 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
400 xen_mc_issue(PARAVIRT_LAZY_MMU);
403 /* Assume pteval_t is equivalent to all the other *val_t types. */
404 static pteval_t pte_mfn_to_pfn(pteval_t val)
406 if (val & _PAGE_PRESENT) {
407 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
408 pteval_t flags = val & PTE_FLAGS_MASK;
409 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
415 static pteval_t pte_pfn_to_mfn(pteval_t val)
417 if (val & _PAGE_PRESENT) {
418 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
419 pteval_t flags = val & PTE_FLAGS_MASK;
422 if (!xen_feature(XENFEAT_auto_translated_physmap))
423 mfn = get_phys_to_machine(pfn);
427 * If there's no mfn for the pfn, then just create an
428 * empty non-present pte. Unfortunately this loses
429 * information about the original pfn, so
430 * pte_mfn_to_pfn is asymmetric.
432 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
437 * Paramount to do this test _after_ the
438 * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
439 * IDENTITY_FRAME_BIT resolves to true.
441 mfn &= ~FOREIGN_FRAME_BIT;
442 if (mfn & IDENTITY_FRAME_BIT) {
443 mfn &= ~IDENTITY_FRAME_BIT;
444 flags |= _PAGE_IOMAP;
447 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
453 static pteval_t iomap_pte(pteval_t val)
455 if (val & _PAGE_PRESENT) {
456 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
457 pteval_t flags = val & PTE_FLAGS_MASK;
459 /* We assume the pte frame number is a MFN, so
460 just use it as-is. */
461 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
467 pteval_t xen_pte_val(pte_t pte)
469 pteval_t pteval = pte.pte;
471 /* If this is a WC pte, convert back from Xen WC to Linux WC */
472 if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
473 WARN_ON(!pat_enabled);
474 pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
477 if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
480 return pte_mfn_to_pfn(pteval);
482 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
484 pgdval_t xen_pgd_val(pgd_t pgd)
486 return pte_mfn_to_pfn(pgd.pgd);
488 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
491 * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
492 * are reserved for now, to correspond to the Intel-reserved PAT
495 * We expect Linux's PAT set as follows:
497 * Idx PTE flags Linux Xen Default
504 * 6 PAT PCD UC- UC UC-
505 * 7 PAT PCD PWT UC UC UC
508 void xen_set_pat(u64 pat)
510 /* We expect Linux to use a PAT setting of
511 * UC UC- WC WB (ignoring the PAT flag) */
512 WARN_ON(pat != 0x0007010600070106ull);
515 pte_t xen_make_pte(pteval_t pte)
517 phys_addr_t addr = (pte & PTE_PFN_MASK);
519 /* If Linux is trying to set a WC pte, then map to the Xen WC.
520 * If _PAGE_PAT is set, then it probably means it is really
521 * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
522 * things work out OK...
524 * (We should never see kernel mappings with _PAGE_PSE set,
525 * but we could see hugetlbfs mappings, I think.).
527 if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
528 if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
529 pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
533 * Unprivileged domains are allowed to do IOMAPpings for
534 * PCI passthrough, but not map ISA space. The ISA
535 * mappings are just dummy local mappings to keep other
536 * parts of the kernel happy.
538 if (unlikely(pte & _PAGE_IOMAP) &&
539 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
540 pte = iomap_pte(pte);
543 pte = pte_pfn_to_mfn(pte);
546 return native_make_pte(pte);
548 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
550 #ifdef CONFIG_XEN_DEBUG
551 pte_t xen_make_pte_debug(pteval_t pte)
553 phys_addr_t addr = (pte & PTE_PFN_MASK);
554 phys_addr_t other_addr;
555 bool io_page = false;
558 if (pte & _PAGE_IOMAP)
561 _pte = xen_make_pte(pte);
567 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
568 other_addr = pfn_to_mfn(addr >> PAGE_SHIFT) << PAGE_SHIFT;
569 WARN(addr != other_addr,
570 "0x%lx is using VM_IO, but it is 0x%lx!\n",
571 (unsigned long)addr, (unsigned long)other_addr);
573 pteval_t iomap_set = (_pte.pte & PTE_FLAGS_MASK) & _PAGE_IOMAP;
574 other_addr = (_pte.pte & PTE_PFN_MASK);
575 WARN((addr == other_addr) && (!io_page) && (!iomap_set),
576 "0x%lx is missing VM_IO (and wasn't fixed)!\n",
577 (unsigned long)addr);
582 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_debug);
585 pgd_t xen_make_pgd(pgdval_t pgd)
587 pgd = pte_pfn_to_mfn(pgd);
588 return native_make_pgd(pgd);
590 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
592 pmdval_t xen_pmd_val(pmd_t pmd)
594 return pte_mfn_to_pfn(pmd.pmd);
596 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
598 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
606 /* ptr may be ioremapped for 64-bit pagetable setup */
607 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
608 u.val = pud_val_ma(val);
609 xen_extend_mmu_update(&u);
611 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
613 xen_mc_issue(PARAVIRT_LAZY_MMU);
618 void xen_set_pud(pud_t *ptr, pud_t val)
620 ADD_STATS(pud_update, 1);
622 /* If page is not pinned, we can just update the entry
624 if (!xen_page_pinned(ptr)) {
629 ADD_STATS(pud_update_pinned, 1);
631 xen_set_pud_hyper(ptr, val);
634 void xen_set_pte(pte_t *ptep, pte_t pte)
636 if (xen_iomap_pte(pte)) {
637 xen_set_iomap_pte(ptep, pte);
641 ADD_STATS(pte_update, 1);
642 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
643 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
645 #ifdef CONFIG_X86_PAE
646 ptep->pte_high = pte.pte_high;
648 ptep->pte_low = pte.pte_low;
654 #ifdef CONFIG_X86_PAE
655 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
657 if (xen_iomap_pte(pte)) {
658 xen_set_iomap_pte(ptep, pte);
662 set_64bit((u64 *)ptep, native_pte_val(pte));
665 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
668 smp_wmb(); /* make sure low gets written first */
672 void xen_pmd_clear(pmd_t *pmdp)
674 set_pmd(pmdp, __pmd(0));
676 #endif /* CONFIG_X86_PAE */
678 pmd_t xen_make_pmd(pmdval_t pmd)
680 pmd = pte_pfn_to_mfn(pmd);
681 return native_make_pmd(pmd);
683 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
685 #if PAGETABLE_LEVELS == 4
686 pudval_t xen_pud_val(pud_t pud)
688 return pte_mfn_to_pfn(pud.pud);
690 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
692 pud_t xen_make_pud(pudval_t pud)
694 pud = pte_pfn_to_mfn(pud);
696 return native_make_pud(pud);
698 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
700 pgd_t *xen_get_user_pgd(pgd_t *pgd)
702 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
703 unsigned offset = pgd - pgd_page;
704 pgd_t *user_ptr = NULL;
706 if (offset < pgd_index(USER_LIMIT)) {
707 struct page *page = virt_to_page(pgd_page);
708 user_ptr = (pgd_t *)page->private;
716 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
720 u.ptr = virt_to_machine(ptr).maddr;
721 u.val = pgd_val_ma(val);
722 xen_extend_mmu_update(&u);
726 * Raw hypercall-based set_pgd, intended for in early boot before
727 * there's a page structure. This implies:
728 * 1. The only existing pagetable is the kernel's
729 * 2. It is always pinned
730 * 3. It has no user pagetable attached to it
732 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
738 __xen_set_pgd_hyper(ptr, val);
740 xen_mc_issue(PARAVIRT_LAZY_MMU);
745 void xen_set_pgd(pgd_t *ptr, pgd_t val)
747 pgd_t *user_ptr = xen_get_user_pgd(ptr);
749 ADD_STATS(pgd_update, 1);
751 /* If page is not pinned, we can just update the entry
753 if (!xen_page_pinned(ptr)) {
756 WARN_ON(xen_page_pinned(user_ptr));
762 ADD_STATS(pgd_update_pinned, 1);
763 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
765 /* If it's pinned, then we can at least batch the kernel and
766 user updates together. */
769 __xen_set_pgd_hyper(ptr, val);
771 __xen_set_pgd_hyper(user_ptr, val);
773 xen_mc_issue(PARAVIRT_LAZY_MMU);
775 #endif /* PAGETABLE_LEVELS == 4 */
778 * (Yet another) pagetable walker. This one is intended for pinning a
779 * pagetable. This means that it walks a pagetable and calls the
780 * callback function on each page it finds making up the page table,
781 * at every level. It walks the entire pagetable, but it only bothers
782 * pinning pte pages which are below limit. In the normal case this
783 * will be STACK_TOP_MAX, but at boot we need to pin up to
786 * For 32-bit the important bit is that we don't pin beyond there,
787 * because then we start getting into Xen's ptes.
789 * For 64-bit, we must skip the Xen hole in the middle of the address
790 * space, just after the big x86-64 virtual hole.
792 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
793 int (*func)(struct mm_struct *mm, struct page *,
798 unsigned hole_low, hole_high;
799 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
800 unsigned pgdidx, pudidx, pmdidx;
802 /* The limit is the last byte to be touched */
804 BUG_ON(limit >= FIXADDR_TOP);
806 if (xen_feature(XENFEAT_auto_translated_physmap))
810 * 64-bit has a great big hole in the middle of the address
811 * space, which contains the Xen mappings. On 32-bit these
812 * will end up making a zero-sized hole and so is a no-op.
814 hole_low = pgd_index(USER_LIMIT);
815 hole_high = pgd_index(PAGE_OFFSET);
817 pgdidx_limit = pgd_index(limit);
819 pudidx_limit = pud_index(limit);
824 pmdidx_limit = pmd_index(limit);
829 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
832 if (pgdidx >= hole_low && pgdidx < hole_high)
835 if (!pgd_val(pgd[pgdidx]))
838 pud = pud_offset(&pgd[pgdidx], 0);
840 if (PTRS_PER_PUD > 1) /* not folded */
841 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
843 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
846 if (pgdidx == pgdidx_limit &&
847 pudidx > pudidx_limit)
850 if (pud_none(pud[pudidx]))
853 pmd = pmd_offset(&pud[pudidx], 0);
855 if (PTRS_PER_PMD > 1) /* not folded */
856 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
858 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
861 if (pgdidx == pgdidx_limit &&
862 pudidx == pudidx_limit &&
863 pmdidx > pmdidx_limit)
866 if (pmd_none(pmd[pmdidx]))
869 pte = pmd_page(pmd[pmdidx]);
870 flush |= (*func)(mm, pte, PT_PTE);
876 /* Do the top level last, so that the callbacks can use it as
877 a cue to do final things like tlb flushes. */
878 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
883 static int xen_pgd_walk(struct mm_struct *mm,
884 int (*func)(struct mm_struct *mm, struct page *,
888 return __xen_pgd_walk(mm, mm->pgd, func, limit);
891 /* If we're using split pte locks, then take the page's lock and
892 return a pointer to it. Otherwise return NULL. */
893 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
895 spinlock_t *ptl = NULL;
897 #if USE_SPLIT_PTLOCKS
898 ptl = __pte_lockptr(page);
899 spin_lock_nest_lock(ptl, &mm->page_table_lock);
905 static void xen_pte_unlock(void *v)
911 static void xen_do_pin(unsigned level, unsigned long pfn)
913 struct mmuext_op *op;
914 struct multicall_space mcs;
916 mcs = __xen_mc_entry(sizeof(*op));
919 op->arg1.mfn = pfn_to_mfn(pfn);
920 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
923 static int xen_pin_page(struct mm_struct *mm, struct page *page,
926 unsigned pgfl = TestSetPagePinned(page);
930 flush = 0; /* already pinned */
931 else if (PageHighMem(page))
932 /* kmaps need flushing if we found an unpinned
936 void *pt = lowmem_page_address(page);
937 unsigned long pfn = page_to_pfn(page);
938 struct multicall_space mcs = __xen_mc_entry(0);
944 * We need to hold the pagetable lock between the time
945 * we make the pagetable RO and when we actually pin
946 * it. If we don't, then other users may come in and
947 * attempt to update the pagetable by writing it,
948 * which will fail because the memory is RO but not
949 * pinned, so Xen won't do the trap'n'emulate.
951 * If we're using split pte locks, we can't hold the
952 * entire pagetable's worth of locks during the
953 * traverse, because we may wrap the preempt count (8
954 * bits). The solution is to mark RO and pin each PTE
955 * page while holding the lock. This means the number
956 * of locks we end up holding is never more than a
957 * batch size (~32 entries, at present).
959 * If we're not using split pte locks, we needn't pin
960 * the PTE pages independently, because we're
961 * protected by the overall pagetable lock.
965 ptl = xen_pte_lock(page, mm);
967 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
968 pfn_pte(pfn, PAGE_KERNEL_RO),
969 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
972 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
974 /* Queue a deferred unlock for when this batch
976 xen_mc_callback(xen_pte_unlock, ptl);
983 /* This is called just after a mm has been created, but it has not
984 been used yet. We need to make sure that its pagetable is all
985 read-only, and can be pinned. */
986 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
990 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
991 /* re-enable interrupts for flushing */
1001 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1003 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
1006 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
1007 xen_do_pin(MMUEXT_PIN_L4_TABLE,
1008 PFN_DOWN(__pa(user_pgd)));
1011 #else /* CONFIG_X86_32 */
1012 #ifdef CONFIG_X86_PAE
1013 /* Need to make sure unshared kernel PMD is pinnable */
1014 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1017 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
1018 #endif /* CONFIG_X86_64 */
1022 static void xen_pgd_pin(struct mm_struct *mm)
1024 __xen_pgd_pin(mm, mm->pgd);
1028 * On save, we need to pin all pagetables to make sure they get their
1029 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1030 * them (unpinned pgds are not currently in use, probably because the
1031 * process is under construction or destruction).
1033 * Expected to be called in stop_machine() ("equivalent to taking
1034 * every spinlock in the system"), so the locking doesn't really
1035 * matter all that much.
1037 void xen_mm_pin_all(void)
1041 spin_lock(&pgd_lock);
1043 list_for_each_entry(page, &pgd_list, lru) {
1044 if (!PagePinned(page)) {
1045 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1046 SetPageSavePinned(page);
1050 spin_unlock(&pgd_lock);
1054 * The init_mm pagetable is really pinned as soon as its created, but
1055 * that's before we have page structures to store the bits. So do all
1056 * the book-keeping now.
1058 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1059 enum pt_level level)
1061 SetPagePinned(page);
1065 static void __init xen_mark_init_mm_pinned(void)
1067 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1070 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1071 enum pt_level level)
1073 unsigned pgfl = TestClearPagePinned(page);
1075 if (pgfl && !PageHighMem(page)) {
1076 void *pt = lowmem_page_address(page);
1077 unsigned long pfn = page_to_pfn(page);
1078 spinlock_t *ptl = NULL;
1079 struct multicall_space mcs;
1082 * Do the converse to pin_page. If we're using split
1083 * pte locks, we must be holding the lock for while
1084 * the pte page is unpinned but still RO to prevent
1085 * concurrent updates from seeing it in this
1086 * partially-pinned state.
1088 if (level == PT_PTE) {
1089 ptl = xen_pte_lock(page, mm);
1092 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1095 mcs = __xen_mc_entry(0);
1097 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1098 pfn_pte(pfn, PAGE_KERNEL),
1099 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1102 /* unlock when batch completed */
1103 xen_mc_callback(xen_pte_unlock, ptl);
1107 return 0; /* never need to flush on unpin */
1110 /* Release a pagetables pages back as normal RW */
1111 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1115 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1117 #ifdef CONFIG_X86_64
1119 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1122 xen_do_pin(MMUEXT_UNPIN_TABLE,
1123 PFN_DOWN(__pa(user_pgd)));
1124 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1129 #ifdef CONFIG_X86_PAE
1130 /* Need to make sure unshared kernel PMD is unpinned */
1131 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1135 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1140 static void xen_pgd_unpin(struct mm_struct *mm)
1142 __xen_pgd_unpin(mm, mm->pgd);
1146 * On resume, undo any pinning done at save, so that the rest of the
1147 * kernel doesn't see any unexpected pinned pagetables.
1149 void xen_mm_unpin_all(void)
1153 spin_lock(&pgd_lock);
1155 list_for_each_entry(page, &pgd_list, lru) {
1156 if (PageSavePinned(page)) {
1157 BUG_ON(!PagePinned(page));
1158 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1159 ClearPageSavePinned(page);
1163 spin_unlock(&pgd_lock);
1166 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1168 spin_lock(&next->page_table_lock);
1170 spin_unlock(&next->page_table_lock);
1173 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1175 spin_lock(&mm->page_table_lock);
1177 spin_unlock(&mm->page_table_lock);
1182 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1183 we need to repoint it somewhere else before we can unpin it. */
1184 static void drop_other_mm_ref(void *info)
1186 struct mm_struct *mm = info;
1187 struct mm_struct *active_mm;
1189 active_mm = percpu_read(cpu_tlbstate.active_mm);
1191 if (active_mm == mm)
1192 leave_mm(smp_processor_id());
1194 /* If this cpu still has a stale cr3 reference, then make sure
1195 it has been flushed. */
1196 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1197 load_cr3(swapper_pg_dir);
1200 static void xen_drop_mm_ref(struct mm_struct *mm)
1205 if (current->active_mm == mm) {
1206 if (current->mm == mm)
1207 load_cr3(swapper_pg_dir);
1209 leave_mm(smp_processor_id());
1212 /* Get the "official" set of cpus referring to our pagetable. */
1213 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1214 for_each_online_cpu(cpu) {
1215 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1216 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1218 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1222 cpumask_copy(mask, mm_cpumask(mm));
1224 /* It's possible that a vcpu may have a stale reference to our
1225 cr3, because its in lazy mode, and it hasn't yet flushed
1226 its set of pending hypercalls yet. In this case, we can
1227 look at its actual current cr3 value, and force it to flush
1229 for_each_online_cpu(cpu) {
1230 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1231 cpumask_set_cpu(cpu, mask);
1234 if (!cpumask_empty(mask))
1235 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1236 free_cpumask_var(mask);
1239 static void xen_drop_mm_ref(struct mm_struct *mm)
1241 if (current->active_mm == mm)
1242 load_cr3(swapper_pg_dir);
1247 * While a process runs, Xen pins its pagetables, which means that the
1248 * hypervisor forces it to be read-only, and it controls all updates
1249 * to it. This means that all pagetable updates have to go via the
1250 * hypervisor, which is moderately expensive.
1252 * Since we're pulling the pagetable down, we switch to use init_mm,
1253 * unpin old process pagetable and mark it all read-write, which
1254 * allows further operations on it to be simple memory accesses.
1256 * The only subtle point is that another CPU may be still using the
1257 * pagetable because of lazy tlb flushing. This means we need need to
1258 * switch all CPUs off this pagetable before we can unpin it.
1260 void xen_exit_mmap(struct mm_struct *mm)
1262 get_cpu(); /* make sure we don't move around */
1263 xen_drop_mm_ref(mm);
1266 spin_lock(&mm->page_table_lock);
1268 /* pgd may not be pinned in the error exit path of execve */
1269 if (xen_page_pinned(mm->pgd))
1272 spin_unlock(&mm->page_table_lock);
1275 static __init void xen_pagetable_setup_start(pgd_t *base)
1279 static void xen_post_allocator_init(void);
1281 static __init void xen_pagetable_setup_done(pgd_t *base)
1283 xen_setup_shared_info();
1284 xen_post_allocator_init();
1287 static void xen_write_cr2(unsigned long cr2)
1289 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1292 static unsigned long xen_read_cr2(void)
1294 return percpu_read(xen_vcpu)->arch.cr2;
1297 unsigned long xen_read_cr2_direct(void)
1299 return percpu_read(xen_vcpu_info.arch.cr2);
1302 static void xen_flush_tlb(void)
1304 struct mmuext_op *op;
1305 struct multicall_space mcs;
1309 mcs = xen_mc_entry(sizeof(*op));
1312 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1313 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1315 xen_mc_issue(PARAVIRT_LAZY_MMU);
1320 static void xen_flush_tlb_single(unsigned long addr)
1322 struct mmuext_op *op;
1323 struct multicall_space mcs;
1327 mcs = xen_mc_entry(sizeof(*op));
1329 op->cmd = MMUEXT_INVLPG_LOCAL;
1330 op->arg1.linear_addr = addr & PAGE_MASK;
1331 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1333 xen_mc_issue(PARAVIRT_LAZY_MMU);
1338 static void xen_flush_tlb_others(const struct cpumask *cpus,
1339 struct mm_struct *mm, unsigned long va)
1342 struct mmuext_op op;
1343 DECLARE_BITMAP(mask, NR_CPUS);
1345 struct multicall_space mcs;
1347 if (cpumask_empty(cpus))
1348 return; /* nothing to do */
1350 mcs = xen_mc_entry(sizeof(*args));
1352 args->op.arg2.vcpumask = to_cpumask(args->mask);
1354 /* Remove us, and any offline CPUS. */
1355 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1356 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1358 if (va == TLB_FLUSH_ALL) {
1359 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1361 args->op.cmd = MMUEXT_INVLPG_MULTI;
1362 args->op.arg1.linear_addr = va;
1365 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1367 xen_mc_issue(PARAVIRT_LAZY_MMU);
1370 static unsigned long xen_read_cr3(void)
1372 return percpu_read(xen_cr3);
1375 static void set_current_cr3(void *v)
1377 percpu_write(xen_current_cr3, (unsigned long)v);
1380 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1382 struct mmuext_op *op;
1383 struct multicall_space mcs;
1387 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1391 WARN_ON(mfn == 0 && kernel);
1393 mcs = __xen_mc_entry(sizeof(*op));
1396 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1399 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1402 percpu_write(xen_cr3, cr3);
1404 /* Update xen_current_cr3 once the batch has actually
1406 xen_mc_callback(set_current_cr3, (void *)cr3);
1410 static void xen_write_cr3(unsigned long cr3)
1412 BUG_ON(preemptible());
1414 xen_mc_batch(); /* disables interrupts */
1416 /* Update while interrupts are disabled, so its atomic with
1418 percpu_write(xen_cr3, cr3);
1420 __xen_write_cr3(true, cr3);
1422 #ifdef CONFIG_X86_64
1424 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1426 __xen_write_cr3(false, __pa(user_pgd));
1428 __xen_write_cr3(false, 0);
1432 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1435 static int xen_pgd_alloc(struct mm_struct *mm)
1437 pgd_t *pgd = mm->pgd;
1440 BUG_ON(PagePinned(virt_to_page(pgd)));
1442 #ifdef CONFIG_X86_64
1444 struct page *page = virt_to_page(pgd);
1447 BUG_ON(page->private != 0);
1451 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1452 page->private = (unsigned long)user_pgd;
1454 if (user_pgd != NULL) {
1455 user_pgd[pgd_index(VSYSCALL_START)] =
1456 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1460 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1467 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1469 #ifdef CONFIG_X86_64
1470 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1473 free_page((unsigned long)user_pgd);
1477 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1479 unsigned long pfn = pte_pfn(pte);
1481 #ifdef CONFIG_X86_32
1482 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1483 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1484 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1489 * If the new pfn is within the range of the newly allocated
1490 * kernel pagetable, and it isn't being mapped into an
1491 * early_ioremap fixmap slot as a freshly allocated page, make sure
1494 if (((!is_early_ioremap_ptep(ptep) &&
1495 pfn >= pgt_buf_start && pfn < pgt_buf_end)) ||
1496 (is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1497 pte = pte_wrprotect(pte);
1502 /* Init-time set_pte while constructing initial pagetables, which
1503 doesn't allow RO pagetable pages to be remapped RW */
1504 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1506 pte = mask_rw_pte(ptep, pte);
1508 xen_set_pte(ptep, pte);
1511 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1513 struct mmuext_op op;
1515 op.arg1.mfn = pfn_to_mfn(pfn);
1516 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1520 /* Early in boot, while setting up the initial pagetable, assume
1521 everything is pinned. */
1522 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1524 #ifdef CONFIG_FLATMEM
1525 BUG_ON(mem_map); /* should only be used early */
1527 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1528 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1531 /* Used for pmd and pud */
1532 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1534 #ifdef CONFIG_FLATMEM
1535 BUG_ON(mem_map); /* should only be used early */
1537 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1540 /* Early release_pte assumes that all pts are pinned, since there's
1541 only init_mm and anything attached to that is pinned. */
1542 static __init void xen_release_pte_init(unsigned long pfn)
1544 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1545 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1548 static __init void xen_release_pmd_init(unsigned long pfn)
1550 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1553 /* This needs to make sure the new pte page is pinned iff its being
1554 attached to a pinned pagetable. */
1555 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1557 struct page *page = pfn_to_page(pfn);
1559 if (PagePinned(virt_to_page(mm->pgd))) {
1560 SetPagePinned(page);
1562 if (!PageHighMem(page)) {
1563 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1564 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1565 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1567 /* make sure there are no stray mappings of
1569 kmap_flush_unused();
1574 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1576 xen_alloc_ptpage(mm, pfn, PT_PTE);
1579 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1581 xen_alloc_ptpage(mm, pfn, PT_PMD);
1584 /* This should never happen until we're OK to use struct page */
1585 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1587 struct page *page = pfn_to_page(pfn);
1589 if (PagePinned(page)) {
1590 if (!PageHighMem(page)) {
1591 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1592 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1593 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1595 ClearPagePinned(page);
1599 static void xen_release_pte(unsigned long pfn)
1601 xen_release_ptpage(pfn, PT_PTE);
1604 static void xen_release_pmd(unsigned long pfn)
1606 xen_release_ptpage(pfn, PT_PMD);
1609 #if PAGETABLE_LEVELS == 4
1610 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1612 xen_alloc_ptpage(mm, pfn, PT_PUD);
1615 static void xen_release_pud(unsigned long pfn)
1617 xen_release_ptpage(pfn, PT_PUD);
1621 void __init xen_reserve_top(void)
1623 #ifdef CONFIG_X86_32
1624 unsigned long top = HYPERVISOR_VIRT_START;
1625 struct xen_platform_parameters pp;
1627 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1628 top = pp.virt_start;
1630 reserve_top_address(-top);
1631 #endif /* CONFIG_X86_32 */
1635 * Like __va(), but returns address in the kernel mapping (which is
1636 * all we have until the physical memory mapping has been set up.
1638 static void *__ka(phys_addr_t paddr)
1640 #ifdef CONFIG_X86_64
1641 return (void *)(paddr + __START_KERNEL_map);
1647 /* Convert a machine address to physical address */
1648 static unsigned long m2p(phys_addr_t maddr)
1652 maddr &= PTE_PFN_MASK;
1653 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1658 /* Convert a machine address to kernel virtual */
1659 static void *m2v(phys_addr_t maddr)
1661 return __ka(m2p(maddr));
1664 /* Set the page permissions on an identity-mapped pages */
1665 static void set_page_prot(void *addr, pgprot_t prot)
1667 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1668 pte_t pte = pfn_pte(pfn, prot);
1670 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1674 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1676 unsigned pmdidx, pteidx;
1680 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1685 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1688 /* Reuse or allocate a page of ptes */
1689 if (pmd_present(pmd[pmdidx]))
1690 pte_page = m2v(pmd[pmdidx].pmd);
1692 /* Check for free pte pages */
1693 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1696 pte_page = &level1_ident_pgt[ident_pte];
1697 ident_pte += PTRS_PER_PTE;
1699 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1702 /* Install mappings */
1703 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1706 if (!pte_none(pte_page[pteidx]))
1709 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1710 pte_page[pteidx] = pte;
1714 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1715 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1717 set_page_prot(pmd, PAGE_KERNEL_RO);
1720 void __init xen_setup_machphys_mapping(void)
1722 struct xen_machphys_mapping mapping;
1723 unsigned long machine_to_phys_nr_ents;
1725 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1726 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1727 machine_to_phys_nr_ents = mapping.max_mfn + 1;
1729 machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
1731 machine_to_phys_order = fls(machine_to_phys_nr_ents - 1);
1734 #ifdef CONFIG_X86_64
1735 static void convert_pfn_mfn(void *v)
1740 /* All levels are converted the same way, so just treat them
1742 for (i = 0; i < PTRS_PER_PTE; i++)
1743 pte[i] = xen_make_pte(pte[i].pte);
1747 * Set up the initial kernel pagetable.
1749 * We can construct this by grafting the Xen provided pagetable into
1750 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1751 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1752 * means that only the kernel has a physical mapping to start with -
1753 * but that's enough to get __va working. We need to fill in the rest
1754 * of the physical mapping once some sort of allocator has been set
1757 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1758 unsigned long max_pfn)
1763 /* max_pfn_mapped is the last pfn mapped in the initial memory
1764 * mappings. Considering that on Xen after the kernel mappings we
1765 * have the mappings of some pages that don't exist in pfn space, we
1766 * set max_pfn_mapped to the last real pfn mapped. */
1767 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1769 /* Zap identity mapping */
1770 init_level4_pgt[0] = __pgd(0);
1772 /* Pre-constructed entries are in pfn, so convert to mfn */
1773 convert_pfn_mfn(init_level4_pgt);
1774 convert_pfn_mfn(level3_ident_pgt);
1775 convert_pfn_mfn(level3_kernel_pgt);
1777 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1778 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1780 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1781 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1783 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1784 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1785 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1787 /* Set up identity map */
1788 xen_map_identity_early(level2_ident_pgt, max_pfn);
1790 /* Make pagetable pieces RO */
1791 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1792 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1793 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1794 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1795 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1796 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1798 /* Pin down new L4 */
1799 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1800 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1802 /* Unpin Xen-provided one */
1803 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1806 pgd = init_level4_pgt;
1809 * At this stage there can be no user pgd, and no page
1810 * structure to attach it to, so make sure we just set kernel
1814 __xen_write_cr3(true, __pa(pgd));
1815 xen_mc_issue(PARAVIRT_LAZY_CPU);
1817 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1818 __pa(xen_start_info->pt_base +
1819 xen_start_info->nr_pt_frames * PAGE_SIZE),
1824 #else /* !CONFIG_X86_64 */
1825 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1826 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1828 static __init void xen_write_cr3_init(unsigned long cr3)
1830 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1832 BUG_ON(read_cr3() != __pa(initial_page_table));
1833 BUG_ON(cr3 != __pa(swapper_pg_dir));
1836 * We are switching to swapper_pg_dir for the first time (from
1837 * initial_page_table) and therefore need to mark that page
1838 * read-only and then pin it.
1840 * Xen disallows sharing of kernel PMDs for PAE
1841 * guests. Therefore we must copy the kernel PMD from
1842 * initial_page_table into a new kernel PMD to be used in
1845 swapper_kernel_pmd =
1846 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1847 memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1848 sizeof(pmd_t) * PTRS_PER_PMD);
1849 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1850 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1851 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1853 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1855 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1857 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1858 PFN_DOWN(__pa(initial_page_table)));
1859 set_page_prot(initial_page_table, PAGE_KERNEL);
1860 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1862 pv_mmu_ops.write_cr3 = &xen_write_cr3;
1865 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1866 unsigned long max_pfn)
1870 initial_kernel_pmd =
1871 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1873 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1875 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1876 memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1878 xen_map_identity_early(initial_kernel_pmd, max_pfn);
1880 memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1881 initial_page_table[KERNEL_PGD_BOUNDARY] =
1882 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1884 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1885 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1886 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1888 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1890 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1891 PFN_DOWN(__pa(initial_page_table)));
1892 xen_write_cr3(__pa(initial_page_table));
1894 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1895 __pa(xen_start_info->pt_base +
1896 xen_start_info->nr_pt_frames * PAGE_SIZE),
1899 return initial_page_table;
1901 #endif /* CONFIG_X86_64 */
1903 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1905 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1909 phys >>= PAGE_SHIFT;
1912 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1913 #ifdef CONFIG_X86_F00F_BUG
1916 #ifdef CONFIG_X86_32
1919 # ifdef CONFIG_HIGHMEM
1920 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1923 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1925 case FIX_TEXT_POKE0:
1926 case FIX_TEXT_POKE1:
1927 /* All local page mappings */
1928 pte = pfn_pte(phys, prot);
1931 #ifdef CONFIG_X86_LOCAL_APIC
1932 case FIX_APIC_BASE: /* maps dummy local APIC */
1933 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1937 #ifdef CONFIG_X86_IO_APIC
1938 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1940 * We just don't map the IO APIC - all access is via
1941 * hypercalls. Keep the address in the pte for reference.
1943 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1947 case FIX_PARAVIRT_BOOTMAP:
1948 /* This is an MFN, but it isn't an IO mapping from the
1950 pte = mfn_pte(phys, prot);
1954 /* By default, set_fixmap is used for hardware mappings */
1955 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1959 __native_set_fixmap(idx, pte);
1961 #ifdef CONFIG_X86_64
1962 /* Replicate changes to map the vsyscall page into the user
1963 pagetable vsyscall mapping. */
1964 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1965 unsigned long vaddr = __fix_to_virt(idx);
1966 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1971 __init void xen_ident_map_ISA(void)
1976 * If we're dom0, then linear map the ISA machine addresses into
1977 * the kernel's address space.
1979 if (!xen_initial_domain())
1982 xen_raw_printk("Xen: setup ISA identity maps\n");
1984 for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1985 pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1987 if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1994 static __init void xen_post_allocator_init(void)
1996 #ifdef CONFIG_XEN_DEBUG
1997 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte_debug);
1999 pv_mmu_ops.set_pte = xen_set_pte;
2000 pv_mmu_ops.set_pmd = xen_set_pmd;
2001 pv_mmu_ops.set_pud = xen_set_pud;
2002 #if PAGETABLE_LEVELS == 4
2003 pv_mmu_ops.set_pgd = xen_set_pgd;
2006 /* This will work as long as patching hasn't happened yet
2007 (which it hasn't) */
2008 pv_mmu_ops.alloc_pte = xen_alloc_pte;
2009 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2010 pv_mmu_ops.release_pte = xen_release_pte;
2011 pv_mmu_ops.release_pmd = xen_release_pmd;
2012 #if PAGETABLE_LEVELS == 4
2013 pv_mmu_ops.alloc_pud = xen_alloc_pud;
2014 pv_mmu_ops.release_pud = xen_release_pud;
2017 #ifdef CONFIG_X86_64
2018 SetPagePinned(virt_to_page(level3_user_vsyscall));
2020 xen_mark_init_mm_pinned();
2023 static void xen_leave_lazy_mmu(void)
2027 paravirt_leave_lazy_mmu();
2031 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
2032 .read_cr2 = xen_read_cr2,
2033 .write_cr2 = xen_write_cr2,
2035 .read_cr3 = xen_read_cr3,
2036 #ifdef CONFIG_X86_32
2037 .write_cr3 = xen_write_cr3_init,
2039 .write_cr3 = xen_write_cr3,
2042 .flush_tlb_user = xen_flush_tlb,
2043 .flush_tlb_kernel = xen_flush_tlb,
2044 .flush_tlb_single = xen_flush_tlb_single,
2045 .flush_tlb_others = xen_flush_tlb_others,
2047 .pte_update = paravirt_nop,
2048 .pte_update_defer = paravirt_nop,
2050 .pgd_alloc = xen_pgd_alloc,
2051 .pgd_free = xen_pgd_free,
2053 .alloc_pte = xen_alloc_pte_init,
2054 .release_pte = xen_release_pte_init,
2055 .alloc_pmd = xen_alloc_pmd_init,
2056 .release_pmd = xen_release_pmd_init,
2058 .set_pte = xen_set_pte_init,
2059 .set_pte_at = xen_set_pte_at,
2060 .set_pmd = xen_set_pmd_hyper,
2062 .ptep_modify_prot_start = __ptep_modify_prot_start,
2063 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2065 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2066 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2068 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2069 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2071 #ifdef CONFIG_X86_PAE
2072 .set_pte_atomic = xen_set_pte_atomic,
2073 .pte_clear = xen_pte_clear,
2074 .pmd_clear = xen_pmd_clear,
2075 #endif /* CONFIG_X86_PAE */
2076 .set_pud = xen_set_pud_hyper,
2078 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2079 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2081 #if PAGETABLE_LEVELS == 4
2082 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2083 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2084 .set_pgd = xen_set_pgd_hyper,
2086 .alloc_pud = xen_alloc_pmd_init,
2087 .release_pud = xen_release_pmd_init,
2088 #endif /* PAGETABLE_LEVELS == 4 */
2090 .activate_mm = xen_activate_mm,
2091 .dup_mmap = xen_dup_mmap,
2092 .exit_mmap = xen_exit_mmap,
2095 .enter = paravirt_enter_lazy_mmu,
2096 .leave = xen_leave_lazy_mmu,
2099 .set_fixmap = xen_set_fixmap,
2102 void __init xen_init_mmu_ops(void)
2104 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2105 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2106 pv_mmu_ops = xen_mmu_ops;
2108 memset(dummy_mapping, 0xff, PAGE_SIZE);
2111 /* Protected by xen_reservation_lock. */
2112 #define MAX_CONTIG_ORDER 9 /* 2MB */
2113 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2115 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2116 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2117 unsigned long *in_frames,
2118 unsigned long *out_frames)
2121 struct multicall_space mcs;
2124 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2125 mcs = __xen_mc_entry(0);
2128 in_frames[i] = virt_to_mfn(vaddr);
2130 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2131 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2134 out_frames[i] = virt_to_pfn(vaddr);
2140 * Update the pfn-to-mfn mappings for a virtual address range, either to
2141 * point to an array of mfns, or contiguously from a single starting
2144 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2145 unsigned long *mfns,
2146 unsigned long first_mfn)
2153 limit = 1u << order;
2154 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2155 struct multicall_space mcs;
2158 mcs = __xen_mc_entry(0);
2162 mfn = first_mfn + i;
2164 if (i < (limit - 1))
2168 flags = UVMF_INVLPG | UVMF_ALL;
2170 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2173 MULTI_update_va_mapping(mcs.mc, vaddr,
2174 mfn_pte(mfn, PAGE_KERNEL), flags);
2176 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2183 * Perform the hypercall to exchange a region of our pfns to point to
2184 * memory with the required contiguous alignment. Takes the pfns as
2185 * input, and populates mfns as output.
2187 * Returns a success code indicating whether the hypervisor was able to
2188 * satisfy the request or not.
2190 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2191 unsigned long *pfns_in,
2192 unsigned long extents_out,
2193 unsigned int order_out,
2194 unsigned long *mfns_out,
2195 unsigned int address_bits)
2200 struct xen_memory_exchange exchange = {
2202 .nr_extents = extents_in,
2203 .extent_order = order_in,
2204 .extent_start = pfns_in,
2208 .nr_extents = extents_out,
2209 .extent_order = order_out,
2210 .extent_start = mfns_out,
2211 .address_bits = address_bits,
2216 BUG_ON(extents_in << order_in != extents_out << order_out);
2218 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2219 success = (exchange.nr_exchanged == extents_in);
2221 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2222 BUG_ON(success && (rc != 0));
2227 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2228 unsigned int address_bits)
2230 unsigned long *in_frames = discontig_frames, out_frame;
2231 unsigned long flags;
2235 * Currently an auto-translated guest will not perform I/O, nor will
2236 * it require PAE page directories below 4GB. Therefore any calls to
2237 * this function are redundant and can be ignored.
2240 if (xen_feature(XENFEAT_auto_translated_physmap))
2243 if (unlikely(order > MAX_CONTIG_ORDER))
2246 memset((void *) vstart, 0, PAGE_SIZE << order);
2248 spin_lock_irqsave(&xen_reservation_lock, flags);
2250 /* 1. Zap current PTEs, remembering MFNs. */
2251 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2253 /* 2. Get a new contiguous memory extent. */
2254 out_frame = virt_to_pfn(vstart);
2255 success = xen_exchange_memory(1UL << order, 0, in_frames,
2256 1, order, &out_frame,
2259 /* 3. Map the new extent in place of old pages. */
2261 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2263 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2265 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2267 return success ? 0 : -ENOMEM;
2269 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2271 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2273 unsigned long *out_frames = discontig_frames, in_frame;
2274 unsigned long flags;
2277 if (xen_feature(XENFEAT_auto_translated_physmap))
2280 if (unlikely(order > MAX_CONTIG_ORDER))
2283 memset((void *) vstart, 0, PAGE_SIZE << order);
2285 spin_lock_irqsave(&xen_reservation_lock, flags);
2287 /* 1. Find start MFN of contiguous extent. */
2288 in_frame = virt_to_mfn(vstart);
2290 /* 2. Zap current PTEs. */
2291 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2293 /* 3. Do the exchange for non-contiguous MFNs. */
2294 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2297 /* 4. Map new pages in place of old pages. */
2299 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2301 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2303 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2305 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2307 #ifdef CONFIG_XEN_PVHVM
2308 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2310 struct xen_hvm_pagetable_dying a;
2313 a.domid = DOMID_SELF;
2314 a.gpa = __pa(mm->pgd);
2315 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2316 WARN_ON_ONCE(rc < 0);
2319 static int is_pagetable_dying_supported(void)
2321 struct xen_hvm_pagetable_dying a;
2324 a.domid = DOMID_SELF;
2326 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2328 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2334 void __init xen_hvm_init_mmu_ops(void)
2336 if (is_pagetable_dying_supported())
2337 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2341 #define REMAP_BATCH_SIZE 16
2346 struct mmu_update *mmu_update;
2349 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2350 unsigned long addr, void *data)
2352 struct remap_data *rmd = data;
2353 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2355 rmd->mmu_update->ptr = arbitrary_virt_to_machine(ptep).maddr;
2356 rmd->mmu_update->val = pte_val_ma(pte);
2362 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2364 unsigned long mfn, int nr,
2365 pgprot_t prot, unsigned domid)
2367 struct remap_data rmd;
2368 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2370 unsigned long range;
2373 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2375 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2376 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2382 batch = min(REMAP_BATCH_SIZE, nr);
2383 range = (unsigned long)batch << PAGE_SHIFT;
2385 rmd.mmu_update = mmu_update;
2386 err = apply_to_page_range(vma->vm_mm, addr, range,
2387 remap_area_mfn_pte_fn, &rmd);
2392 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2406 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2408 #ifdef CONFIG_XEN_DEBUG_FS
2410 static int p2m_dump_open(struct inode *inode, struct file *filp)
2412 return single_open(filp, p2m_dump_show, NULL);
2415 static const struct file_operations p2m_dump_fops = {
2416 .open = p2m_dump_open,
2418 .llseek = seq_lseek,
2419 .release = single_release,
2422 static struct dentry *d_mmu_debug;
2424 static int __init xen_mmu_debugfs(void)
2426 struct dentry *d_xen = xen_init_debugfs();
2431 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2433 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2435 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2436 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2437 &mmu_stats.pgd_update_pinned);
2438 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2439 &mmu_stats.pgd_update_pinned);
2441 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2442 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2443 &mmu_stats.pud_update_pinned);
2444 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2445 &mmu_stats.pud_update_pinned);
2447 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2448 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2449 &mmu_stats.pmd_update_pinned);
2450 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2451 &mmu_stats.pmd_update_pinned);
2453 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2454 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2455 // &mmu_stats.pte_update_pinned);
2456 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2457 &mmu_stats.pte_update_pinned);
2459 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2460 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2461 &mmu_stats.mmu_update_extended);
2462 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2463 mmu_stats.mmu_update_histo, 20);
2465 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2466 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2467 &mmu_stats.set_pte_at_batched);
2468 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2469 &mmu_stats.set_pte_at_current);
2470 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2471 &mmu_stats.set_pte_at_kernel);
2473 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2474 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2475 &mmu_stats.prot_commit_batched);
2477 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
2480 fs_initcall(xen_mmu_debugfs);
2482 #endif /* CONFIG_XEN_DEBUG_FS */