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)
1039 unsigned long flags;
1042 spin_lock_irqsave(&pgd_lock, flags);
1044 list_for_each_entry(page, &pgd_list, lru) {
1045 if (!PagePinned(page)) {
1046 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1047 SetPageSavePinned(page);
1051 spin_unlock_irqrestore(&pgd_lock, flags);
1055 * The init_mm pagetable is really pinned as soon as its created, but
1056 * that's before we have page structures to store the bits. So do all
1057 * the book-keeping now.
1059 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1060 enum pt_level level)
1062 SetPagePinned(page);
1066 static void __init xen_mark_init_mm_pinned(void)
1068 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1071 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1072 enum pt_level level)
1074 unsigned pgfl = TestClearPagePinned(page);
1076 if (pgfl && !PageHighMem(page)) {
1077 void *pt = lowmem_page_address(page);
1078 unsigned long pfn = page_to_pfn(page);
1079 spinlock_t *ptl = NULL;
1080 struct multicall_space mcs;
1083 * Do the converse to pin_page. If we're using split
1084 * pte locks, we must be holding the lock for while
1085 * the pte page is unpinned but still RO to prevent
1086 * concurrent updates from seeing it in this
1087 * partially-pinned state.
1089 if (level == PT_PTE) {
1090 ptl = xen_pte_lock(page, mm);
1093 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1096 mcs = __xen_mc_entry(0);
1098 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1099 pfn_pte(pfn, PAGE_KERNEL),
1100 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1103 /* unlock when batch completed */
1104 xen_mc_callback(xen_pte_unlock, ptl);
1108 return 0; /* never need to flush on unpin */
1111 /* Release a pagetables pages back as normal RW */
1112 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1116 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1118 #ifdef CONFIG_X86_64
1120 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1123 xen_do_pin(MMUEXT_UNPIN_TABLE,
1124 PFN_DOWN(__pa(user_pgd)));
1125 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1130 #ifdef CONFIG_X86_PAE
1131 /* Need to make sure unshared kernel PMD is unpinned */
1132 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1136 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1141 static void xen_pgd_unpin(struct mm_struct *mm)
1143 __xen_pgd_unpin(mm, mm->pgd);
1147 * On resume, undo any pinning done at save, so that the rest of the
1148 * kernel doesn't see any unexpected pinned pagetables.
1150 void xen_mm_unpin_all(void)
1152 unsigned long flags;
1155 spin_lock_irqsave(&pgd_lock, flags);
1157 list_for_each_entry(page, &pgd_list, lru) {
1158 if (PageSavePinned(page)) {
1159 BUG_ON(!PagePinned(page));
1160 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1161 ClearPageSavePinned(page);
1165 spin_unlock_irqrestore(&pgd_lock, flags);
1168 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1170 spin_lock(&next->page_table_lock);
1172 spin_unlock(&next->page_table_lock);
1175 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1177 spin_lock(&mm->page_table_lock);
1179 spin_unlock(&mm->page_table_lock);
1184 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1185 we need to repoint it somewhere else before we can unpin it. */
1186 static void drop_other_mm_ref(void *info)
1188 struct mm_struct *mm = info;
1189 struct mm_struct *active_mm;
1191 active_mm = percpu_read(cpu_tlbstate.active_mm);
1193 if (active_mm == mm)
1194 leave_mm(smp_processor_id());
1196 /* If this cpu still has a stale cr3 reference, then make sure
1197 it has been flushed. */
1198 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1199 load_cr3(swapper_pg_dir);
1202 static void xen_drop_mm_ref(struct mm_struct *mm)
1207 if (current->active_mm == mm) {
1208 if (current->mm == mm)
1209 load_cr3(swapper_pg_dir);
1211 leave_mm(smp_processor_id());
1214 /* Get the "official" set of cpus referring to our pagetable. */
1215 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1216 for_each_online_cpu(cpu) {
1217 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1218 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1220 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1224 cpumask_copy(mask, mm_cpumask(mm));
1226 /* It's possible that a vcpu may have a stale reference to our
1227 cr3, because its in lazy mode, and it hasn't yet flushed
1228 its set of pending hypercalls yet. In this case, we can
1229 look at its actual current cr3 value, and force it to flush
1231 for_each_online_cpu(cpu) {
1232 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1233 cpumask_set_cpu(cpu, mask);
1236 if (!cpumask_empty(mask))
1237 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1238 free_cpumask_var(mask);
1241 static void xen_drop_mm_ref(struct mm_struct *mm)
1243 if (current->active_mm == mm)
1244 load_cr3(swapper_pg_dir);
1249 * While a process runs, Xen pins its pagetables, which means that the
1250 * hypervisor forces it to be read-only, and it controls all updates
1251 * to it. This means that all pagetable updates have to go via the
1252 * hypervisor, which is moderately expensive.
1254 * Since we're pulling the pagetable down, we switch to use init_mm,
1255 * unpin old process pagetable and mark it all read-write, which
1256 * allows further operations on it to be simple memory accesses.
1258 * The only subtle point is that another CPU may be still using the
1259 * pagetable because of lazy tlb flushing. This means we need need to
1260 * switch all CPUs off this pagetable before we can unpin it.
1262 void xen_exit_mmap(struct mm_struct *mm)
1264 get_cpu(); /* make sure we don't move around */
1265 xen_drop_mm_ref(mm);
1268 spin_lock(&mm->page_table_lock);
1270 /* pgd may not be pinned in the error exit path of execve */
1271 if (xen_page_pinned(mm->pgd))
1274 spin_unlock(&mm->page_table_lock);
1277 static __init void xen_pagetable_setup_start(pgd_t *base)
1281 static void xen_post_allocator_init(void);
1283 static __init void xen_pagetable_setup_done(pgd_t *base)
1285 xen_setup_shared_info();
1286 xen_post_allocator_init();
1289 static void xen_write_cr2(unsigned long cr2)
1291 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1294 static unsigned long xen_read_cr2(void)
1296 return percpu_read(xen_vcpu)->arch.cr2;
1299 unsigned long xen_read_cr2_direct(void)
1301 return percpu_read(xen_vcpu_info.arch.cr2);
1304 static void xen_flush_tlb(void)
1306 struct mmuext_op *op;
1307 struct multicall_space mcs;
1311 mcs = xen_mc_entry(sizeof(*op));
1314 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1315 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1317 xen_mc_issue(PARAVIRT_LAZY_MMU);
1322 static void xen_flush_tlb_single(unsigned long addr)
1324 struct mmuext_op *op;
1325 struct multicall_space mcs;
1329 mcs = xen_mc_entry(sizeof(*op));
1331 op->cmd = MMUEXT_INVLPG_LOCAL;
1332 op->arg1.linear_addr = addr & PAGE_MASK;
1333 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1335 xen_mc_issue(PARAVIRT_LAZY_MMU);
1340 static void xen_flush_tlb_others(const struct cpumask *cpus,
1341 struct mm_struct *mm, unsigned long va)
1344 struct mmuext_op op;
1345 DECLARE_BITMAP(mask, NR_CPUS);
1347 struct multicall_space mcs;
1349 if (cpumask_empty(cpus))
1350 return; /* nothing to do */
1352 mcs = xen_mc_entry(sizeof(*args));
1354 args->op.arg2.vcpumask = to_cpumask(args->mask);
1356 /* Remove us, and any offline CPUS. */
1357 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1358 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1360 if (va == TLB_FLUSH_ALL) {
1361 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1363 args->op.cmd = MMUEXT_INVLPG_MULTI;
1364 args->op.arg1.linear_addr = va;
1367 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1369 xen_mc_issue(PARAVIRT_LAZY_MMU);
1372 static unsigned long xen_read_cr3(void)
1374 return percpu_read(xen_cr3);
1377 static void set_current_cr3(void *v)
1379 percpu_write(xen_current_cr3, (unsigned long)v);
1382 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1384 struct mmuext_op *op;
1385 struct multicall_space mcs;
1389 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1393 WARN_ON(mfn == 0 && kernel);
1395 mcs = __xen_mc_entry(sizeof(*op));
1398 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1401 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1404 percpu_write(xen_cr3, cr3);
1406 /* Update xen_current_cr3 once the batch has actually
1408 xen_mc_callback(set_current_cr3, (void *)cr3);
1412 static void xen_write_cr3(unsigned long cr3)
1414 BUG_ON(preemptible());
1416 xen_mc_batch(); /* disables interrupts */
1418 /* Update while interrupts are disabled, so its atomic with
1420 percpu_write(xen_cr3, cr3);
1422 __xen_write_cr3(true, cr3);
1424 #ifdef CONFIG_X86_64
1426 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1428 __xen_write_cr3(false, __pa(user_pgd));
1430 __xen_write_cr3(false, 0);
1434 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1437 static int xen_pgd_alloc(struct mm_struct *mm)
1439 pgd_t *pgd = mm->pgd;
1442 BUG_ON(PagePinned(virt_to_page(pgd)));
1444 #ifdef CONFIG_X86_64
1446 struct page *page = virt_to_page(pgd);
1449 BUG_ON(page->private != 0);
1453 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1454 page->private = (unsigned long)user_pgd;
1456 if (user_pgd != NULL) {
1457 user_pgd[pgd_index(VSYSCALL_START)] =
1458 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1462 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1469 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1471 #ifdef CONFIG_X86_64
1472 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1475 free_page((unsigned long)user_pgd);
1479 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1481 unsigned long pfn = pte_pfn(pte);
1483 #ifdef CONFIG_X86_32
1484 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1485 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1486 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1491 * If the new pfn is within the range of the newly allocated
1492 * kernel pagetable, and it isn't being mapped into an
1493 * early_ioremap fixmap slot, make sure it is RO.
1495 if (!is_early_ioremap_ptep(ptep) &&
1496 pfn >= e820_table_start && pfn < e820_table_end)
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 (pfn > max_pfn_mapped)
1707 max_pfn_mapped = pfn;
1709 if (!pte_none(pte_page[pteidx]))
1712 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1713 pte_page[pteidx] = pte;
1717 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1718 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1720 set_page_prot(pmd, PAGE_KERNEL_RO);
1723 void __init xen_setup_machphys_mapping(void)
1725 struct xen_machphys_mapping mapping;
1726 unsigned long machine_to_phys_nr_ents;
1728 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1729 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1730 machine_to_phys_nr_ents = mapping.max_mfn + 1;
1732 machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
1734 machine_to_phys_order = fls(machine_to_phys_nr_ents - 1);
1737 #ifdef CONFIG_X86_64
1738 static void convert_pfn_mfn(void *v)
1743 /* All levels are converted the same way, so just treat them
1745 for (i = 0; i < PTRS_PER_PTE; i++)
1746 pte[i] = xen_make_pte(pte[i].pte);
1750 * Set up the inital kernel pagetable.
1752 * We can construct this by grafting the Xen provided pagetable into
1753 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1754 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1755 * means that only the kernel has a physical mapping to start with -
1756 * but that's enough to get __va working. We need to fill in the rest
1757 * of the physical mapping once some sort of allocator has been set
1760 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1761 unsigned long max_pfn)
1766 /* Zap identity mapping */
1767 init_level4_pgt[0] = __pgd(0);
1769 /* Pre-constructed entries are in pfn, so convert to mfn */
1770 convert_pfn_mfn(init_level4_pgt);
1771 convert_pfn_mfn(level3_ident_pgt);
1772 convert_pfn_mfn(level3_kernel_pgt);
1774 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1775 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1777 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1778 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1780 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1781 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1782 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1784 /* Set up identity map */
1785 xen_map_identity_early(level2_ident_pgt, max_pfn);
1787 /* Make pagetable pieces RO */
1788 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1789 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1790 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1791 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1792 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1793 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1795 /* Pin down new L4 */
1796 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1797 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1799 /* Unpin Xen-provided one */
1800 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1803 pgd = init_level4_pgt;
1806 * At this stage there can be no user pgd, and no page
1807 * structure to attach it to, so make sure we just set kernel
1811 __xen_write_cr3(true, __pa(pgd));
1812 xen_mc_issue(PARAVIRT_LAZY_CPU);
1814 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1815 __pa(xen_start_info->pt_base +
1816 xen_start_info->nr_pt_frames * PAGE_SIZE),
1821 #else /* !CONFIG_X86_64 */
1822 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1823 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1825 static __init void xen_write_cr3_init(unsigned long cr3)
1827 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1829 BUG_ON(read_cr3() != __pa(initial_page_table));
1830 BUG_ON(cr3 != __pa(swapper_pg_dir));
1833 * We are switching to swapper_pg_dir for the first time (from
1834 * initial_page_table) and therefore need to mark that page
1835 * read-only and then pin it.
1837 * Xen disallows sharing of kernel PMDs for PAE
1838 * guests. Therefore we must copy the kernel PMD from
1839 * initial_page_table into a new kernel PMD to be used in
1842 swapper_kernel_pmd =
1843 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1844 memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1845 sizeof(pmd_t) * PTRS_PER_PMD);
1846 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1847 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1848 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1850 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1852 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1854 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1855 PFN_DOWN(__pa(initial_page_table)));
1856 set_page_prot(initial_page_table, PAGE_KERNEL);
1857 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1859 pv_mmu_ops.write_cr3 = &xen_write_cr3;
1862 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1863 unsigned long max_pfn)
1867 initial_kernel_pmd =
1868 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1870 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1871 xen_start_info->nr_pt_frames * PAGE_SIZE +
1874 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1875 memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1877 xen_map_identity_early(initial_kernel_pmd, max_pfn);
1879 memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1880 initial_page_table[KERNEL_PGD_BOUNDARY] =
1881 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1883 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1884 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1885 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1887 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1889 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1890 PFN_DOWN(__pa(initial_page_table)));
1891 xen_write_cr3(__pa(initial_page_table));
1893 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1894 __pa(xen_start_info->pt_base +
1895 xen_start_info->nr_pt_frames * PAGE_SIZE),
1898 return initial_page_table;
1900 #endif /* CONFIG_X86_64 */
1902 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1904 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1908 phys >>= PAGE_SHIFT;
1911 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1912 #ifdef CONFIG_X86_F00F_BUG
1915 #ifdef CONFIG_X86_32
1918 # ifdef CONFIG_HIGHMEM
1919 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1922 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1924 case FIX_TEXT_POKE0:
1925 case FIX_TEXT_POKE1:
1926 /* All local page mappings */
1927 pte = pfn_pte(phys, prot);
1930 #ifdef CONFIG_X86_LOCAL_APIC
1931 case FIX_APIC_BASE: /* maps dummy local APIC */
1932 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1936 #ifdef CONFIG_X86_IO_APIC
1937 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1939 * We just don't map the IO APIC - all access is via
1940 * hypercalls. Keep the address in the pte for reference.
1942 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1946 case FIX_PARAVIRT_BOOTMAP:
1947 /* This is an MFN, but it isn't an IO mapping from the
1949 pte = mfn_pte(phys, prot);
1953 /* By default, set_fixmap is used for hardware mappings */
1954 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1958 __native_set_fixmap(idx, pte);
1960 #ifdef CONFIG_X86_64
1961 /* Replicate changes to map the vsyscall page into the user
1962 pagetable vsyscall mapping. */
1963 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1964 unsigned long vaddr = __fix_to_virt(idx);
1965 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1970 __init void xen_ident_map_ISA(void)
1975 * If we're dom0, then linear map the ISA machine addresses into
1976 * the kernel's address space.
1978 if (!xen_initial_domain())
1981 xen_raw_printk("Xen: setup ISA identity maps\n");
1983 for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1984 pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1986 if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1993 static __init void xen_post_allocator_init(void)
1995 #ifdef CONFIG_XEN_DEBUG
1996 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte_debug);
1998 pv_mmu_ops.set_pte = xen_set_pte;
1999 pv_mmu_ops.set_pmd = xen_set_pmd;
2000 pv_mmu_ops.set_pud = xen_set_pud;
2001 #if PAGETABLE_LEVELS == 4
2002 pv_mmu_ops.set_pgd = xen_set_pgd;
2005 /* This will work as long as patching hasn't happened yet
2006 (which it hasn't) */
2007 pv_mmu_ops.alloc_pte = xen_alloc_pte;
2008 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2009 pv_mmu_ops.release_pte = xen_release_pte;
2010 pv_mmu_ops.release_pmd = xen_release_pmd;
2011 #if PAGETABLE_LEVELS == 4
2012 pv_mmu_ops.alloc_pud = xen_alloc_pud;
2013 pv_mmu_ops.release_pud = xen_release_pud;
2016 #ifdef CONFIG_X86_64
2017 SetPagePinned(virt_to_page(level3_user_vsyscall));
2019 xen_mark_init_mm_pinned();
2022 static void xen_leave_lazy_mmu(void)
2026 paravirt_leave_lazy_mmu();
2030 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
2031 .read_cr2 = xen_read_cr2,
2032 .write_cr2 = xen_write_cr2,
2034 .read_cr3 = xen_read_cr3,
2035 #ifdef CONFIG_X86_32
2036 .write_cr3 = xen_write_cr3_init,
2038 .write_cr3 = xen_write_cr3,
2041 .flush_tlb_user = xen_flush_tlb,
2042 .flush_tlb_kernel = xen_flush_tlb,
2043 .flush_tlb_single = xen_flush_tlb_single,
2044 .flush_tlb_others = xen_flush_tlb_others,
2046 .pte_update = paravirt_nop,
2047 .pte_update_defer = paravirt_nop,
2049 .pgd_alloc = xen_pgd_alloc,
2050 .pgd_free = xen_pgd_free,
2052 .alloc_pte = xen_alloc_pte_init,
2053 .release_pte = xen_release_pte_init,
2054 .alloc_pmd = xen_alloc_pmd_init,
2055 .release_pmd = xen_release_pmd_init,
2057 .set_pte = xen_set_pte_init,
2058 .set_pte_at = xen_set_pte_at,
2059 .set_pmd = xen_set_pmd_hyper,
2061 .ptep_modify_prot_start = __ptep_modify_prot_start,
2062 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2064 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2065 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2067 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2068 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2070 #ifdef CONFIG_X86_PAE
2071 .set_pte_atomic = xen_set_pte_atomic,
2072 .pte_clear = xen_pte_clear,
2073 .pmd_clear = xen_pmd_clear,
2074 #endif /* CONFIG_X86_PAE */
2075 .set_pud = xen_set_pud_hyper,
2077 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2078 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2080 #if PAGETABLE_LEVELS == 4
2081 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2082 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2083 .set_pgd = xen_set_pgd_hyper,
2085 .alloc_pud = xen_alloc_pmd_init,
2086 .release_pud = xen_release_pmd_init,
2087 #endif /* PAGETABLE_LEVELS == 4 */
2089 .activate_mm = xen_activate_mm,
2090 .dup_mmap = xen_dup_mmap,
2091 .exit_mmap = xen_exit_mmap,
2094 .enter = paravirt_enter_lazy_mmu,
2095 .leave = xen_leave_lazy_mmu,
2098 .set_fixmap = xen_set_fixmap,
2101 void __init xen_init_mmu_ops(void)
2103 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2104 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2105 pv_mmu_ops = xen_mmu_ops;
2107 memset(dummy_mapping, 0xff, PAGE_SIZE);
2110 /* Protected by xen_reservation_lock. */
2111 #define MAX_CONTIG_ORDER 9 /* 2MB */
2112 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2114 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2115 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2116 unsigned long *in_frames,
2117 unsigned long *out_frames)
2120 struct multicall_space mcs;
2123 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2124 mcs = __xen_mc_entry(0);
2127 in_frames[i] = virt_to_mfn(vaddr);
2129 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2130 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2133 out_frames[i] = virt_to_pfn(vaddr);
2139 * Update the pfn-to-mfn mappings for a virtual address range, either to
2140 * point to an array of mfns, or contiguously from a single starting
2143 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2144 unsigned long *mfns,
2145 unsigned long first_mfn)
2152 limit = 1u << order;
2153 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2154 struct multicall_space mcs;
2157 mcs = __xen_mc_entry(0);
2161 mfn = first_mfn + i;
2163 if (i < (limit - 1))
2167 flags = UVMF_INVLPG | UVMF_ALL;
2169 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2172 MULTI_update_va_mapping(mcs.mc, vaddr,
2173 mfn_pte(mfn, PAGE_KERNEL), flags);
2175 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2182 * Perform the hypercall to exchange a region of our pfns to point to
2183 * memory with the required contiguous alignment. Takes the pfns as
2184 * input, and populates mfns as output.
2186 * Returns a success code indicating whether the hypervisor was able to
2187 * satisfy the request or not.
2189 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2190 unsigned long *pfns_in,
2191 unsigned long extents_out,
2192 unsigned int order_out,
2193 unsigned long *mfns_out,
2194 unsigned int address_bits)
2199 struct xen_memory_exchange exchange = {
2201 .nr_extents = extents_in,
2202 .extent_order = order_in,
2203 .extent_start = pfns_in,
2207 .nr_extents = extents_out,
2208 .extent_order = order_out,
2209 .extent_start = mfns_out,
2210 .address_bits = address_bits,
2215 BUG_ON(extents_in << order_in != extents_out << order_out);
2217 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2218 success = (exchange.nr_exchanged == extents_in);
2220 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2221 BUG_ON(success && (rc != 0));
2226 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2227 unsigned int address_bits)
2229 unsigned long *in_frames = discontig_frames, out_frame;
2230 unsigned long flags;
2234 * Currently an auto-translated guest will not perform I/O, nor will
2235 * it require PAE page directories below 4GB. Therefore any calls to
2236 * this function are redundant and can be ignored.
2239 if (xen_feature(XENFEAT_auto_translated_physmap))
2242 if (unlikely(order > MAX_CONTIG_ORDER))
2245 memset((void *) vstart, 0, PAGE_SIZE << order);
2247 spin_lock_irqsave(&xen_reservation_lock, flags);
2249 /* 1. Zap current PTEs, remembering MFNs. */
2250 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2252 /* 2. Get a new contiguous memory extent. */
2253 out_frame = virt_to_pfn(vstart);
2254 success = xen_exchange_memory(1UL << order, 0, in_frames,
2255 1, order, &out_frame,
2258 /* 3. Map the new extent in place of old pages. */
2260 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2262 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2264 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2266 return success ? 0 : -ENOMEM;
2268 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2270 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2272 unsigned long *out_frames = discontig_frames, in_frame;
2273 unsigned long flags;
2276 if (xen_feature(XENFEAT_auto_translated_physmap))
2279 if (unlikely(order > MAX_CONTIG_ORDER))
2282 memset((void *) vstart, 0, PAGE_SIZE << order);
2284 spin_lock_irqsave(&xen_reservation_lock, flags);
2286 /* 1. Find start MFN of contiguous extent. */
2287 in_frame = virt_to_mfn(vstart);
2289 /* 2. Zap current PTEs. */
2290 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2292 /* 3. Do the exchange for non-contiguous MFNs. */
2293 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2296 /* 4. Map new pages in place of old pages. */
2298 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2300 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2302 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2304 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2306 #ifdef CONFIG_XEN_PVHVM
2307 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2309 struct xen_hvm_pagetable_dying a;
2312 a.domid = DOMID_SELF;
2313 a.gpa = __pa(mm->pgd);
2314 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2315 WARN_ON_ONCE(rc < 0);
2318 static int is_pagetable_dying_supported(void)
2320 struct xen_hvm_pagetable_dying a;
2323 a.domid = DOMID_SELF;
2325 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2327 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2333 void __init xen_hvm_init_mmu_ops(void)
2335 if (is_pagetable_dying_supported())
2336 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2340 #define REMAP_BATCH_SIZE 16
2345 struct mmu_update *mmu_update;
2348 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2349 unsigned long addr, void *data)
2351 struct remap_data *rmd = data;
2352 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2354 rmd->mmu_update->ptr = arbitrary_virt_to_machine(ptep).maddr;
2355 rmd->mmu_update->val = pte_val_ma(pte);
2361 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2363 unsigned long mfn, int nr,
2364 pgprot_t prot, unsigned domid)
2366 struct remap_data rmd;
2367 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2369 unsigned long range;
2372 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2374 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2375 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2381 batch = min(REMAP_BATCH_SIZE, nr);
2382 range = (unsigned long)batch << PAGE_SHIFT;
2384 rmd.mmu_update = mmu_update;
2385 err = apply_to_page_range(vma->vm_mm, addr, range,
2386 remap_area_mfn_pte_fn, &rmd);
2391 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2405 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2407 #ifdef CONFIG_XEN_DEBUG_FS
2409 static int p2m_dump_open(struct inode *inode, struct file *filp)
2411 return single_open(filp, p2m_dump_show, NULL);
2414 static const struct file_operations p2m_dump_fops = {
2415 .open = p2m_dump_open,
2417 .llseek = seq_lseek,
2418 .release = single_release,
2421 static struct dentry *d_mmu_debug;
2423 static int __init xen_mmu_debugfs(void)
2425 struct dentry *d_xen = xen_init_debugfs();
2430 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2432 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2434 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2435 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2436 &mmu_stats.pgd_update_pinned);
2437 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2438 &mmu_stats.pgd_update_pinned);
2440 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2441 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2442 &mmu_stats.pud_update_pinned);
2443 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2444 &mmu_stats.pud_update_pinned);
2446 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2447 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2448 &mmu_stats.pmd_update_pinned);
2449 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2450 &mmu_stats.pmd_update_pinned);
2452 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2453 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2454 // &mmu_stats.pte_update_pinned);
2455 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2456 &mmu_stats.pte_update_pinned);
2458 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2459 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2460 &mmu_stats.mmu_update_extended);
2461 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2462 mmu_stats.mmu_update_histo, 20);
2464 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2465 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2466 &mmu_stats.set_pte_at_batched);
2467 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2468 &mmu_stats.set_pte_at_current);
2469 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2470 &mmu_stats.set_pte_at_kernel);
2472 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2473 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2474 &mmu_stats.prot_commit_batched);
2476 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
2479 fs_initcall(xen_mmu_debugfs);
2481 #endif /* CONFIG_XEN_DEBUG_FS */