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/module.h>
47 #include <asm/pgtable.h>
48 #include <asm/tlbflush.h>
49 #include <asm/fixmap.h>
50 #include <asm/mmu_context.h>
51 #include <asm/setup.h>
52 #include <asm/paravirt.h>
53 #include <asm/linkage.h>
55 #include <asm/xen/hypercall.h>
56 #include <asm/xen/hypervisor.h>
59 #include <xen/interface/xen.h>
60 #include <xen/interface/version.h>
61 #include <xen/hvc-console.h>
63 #include "multicalls.h"
67 #define MMU_UPDATE_HISTO 30
69 #ifdef CONFIG_XEN_DEBUG_FS
73 u32 pgd_update_pinned;
74 u32 pgd_update_batched;
77 u32 pud_update_pinned;
78 u32 pud_update_batched;
81 u32 pmd_update_pinned;
82 u32 pmd_update_batched;
85 u32 pte_update_pinned;
86 u32 pte_update_batched;
89 u32 mmu_update_extended;
90 u32 mmu_update_histo[MMU_UPDATE_HISTO];
93 u32 prot_commit_batched;
96 u32 set_pte_at_batched;
97 u32 set_pte_at_pinned;
98 u32 set_pte_at_current;
99 u32 set_pte_at_kernel;
102 static u8 zero_stats;
104 static inline void check_zero(void)
106 if (unlikely(zero_stats)) {
107 memset(&mmu_stats, 0, sizeof(mmu_stats));
112 #define ADD_STATS(elem, val) \
113 do { check_zero(); mmu_stats.elem += (val); } while(0)
115 #else /* !CONFIG_XEN_DEBUG_FS */
117 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
119 #endif /* CONFIG_XEN_DEBUG_FS */
123 * Identity map, in addition to plain kernel map. This needs to be
124 * large enough to allocate page table pages to allocate the rest.
125 * Each page can map 2MB.
127 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
130 /* l3 pud for userspace vsyscall mapping */
131 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
132 #endif /* CONFIG_X86_64 */
135 * Note about cr3 (pagetable base) values:
137 * xen_cr3 contains the current logical cr3 value; it contains the
138 * last set cr3. This may not be the current effective cr3, because
139 * its update may be being lazily deferred. However, a vcpu looking
140 * at its own cr3 can use this value knowing that it everything will
141 * be self-consistent.
143 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
144 * hypercall to set the vcpu cr3 is complete (so it may be a little
145 * out of date, but it will never be set early). If one vcpu is
146 * looking at another vcpu's cr3 value, it should use this variable.
148 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
149 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
153 * Just beyond the highest usermode address. STACK_TOP_MAX has a
154 * redzone above it, so round it up to a PGD boundary.
156 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
159 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
160 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
162 /* Placeholder for holes in the address space */
163 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
164 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
166 /* Array of pointers to pages containing p2m entries */
167 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
168 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
170 /* Arrays of p2m arrays expressed in mfns used for save/restore */
171 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
173 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
176 static inline unsigned p2m_top_index(unsigned long pfn)
178 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
179 return pfn / P2M_ENTRIES_PER_PAGE;
182 static inline unsigned p2m_index(unsigned long pfn)
184 return pfn % P2M_ENTRIES_PER_PAGE;
187 /* Build the parallel p2m_top_mfn structures */
188 void xen_build_mfn_list_list(void)
192 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
193 unsigned topidx = p2m_top_index(pfn);
195 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
198 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
199 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
200 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
204 void xen_setup_mfn_list_list(void)
206 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
208 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
209 virt_to_mfn(p2m_top_mfn_list);
210 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
213 /* Set up p2m_top to point to the domain-builder provided p2m pages */
214 void __init xen_build_dynamic_phys_to_machine(void)
216 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
217 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
220 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
221 unsigned topidx = p2m_top_index(pfn);
223 p2m_top[topidx] = &mfn_list[pfn];
226 xen_build_mfn_list_list();
229 unsigned long get_phys_to_machine(unsigned long pfn)
231 unsigned topidx, idx;
233 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
234 return INVALID_P2M_ENTRY;
236 topidx = p2m_top_index(pfn);
237 idx = p2m_index(pfn);
238 return p2m_top[topidx][idx];
240 EXPORT_SYMBOL_GPL(get_phys_to_machine);
242 /* install a new p2m_top page */
243 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
245 unsigned topidx = p2m_top_index(pfn);
246 unsigned long **pfnp, *mfnp;
249 pfnp = &p2m_top[topidx];
250 mfnp = &p2m_top_mfn[topidx];
252 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
253 p[i] = INVALID_P2M_ENTRY;
255 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
256 *mfnp = virt_to_mfn(p);
263 static void alloc_p2m(unsigned long pfn)
267 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
270 if (!install_p2mtop_page(pfn, p))
271 free_page((unsigned long)p);
274 /* Try to install p2m mapping; fail if intermediate bits missing */
275 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
277 unsigned topidx, idx;
279 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
280 BUG_ON(mfn != INVALID_P2M_ENTRY);
284 topidx = p2m_top_index(pfn);
285 if (p2m_top[topidx] == p2m_missing) {
286 if (mfn == INVALID_P2M_ENTRY)
291 idx = p2m_index(pfn);
292 p2m_top[topidx][idx] = mfn;
297 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
299 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
300 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
304 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
307 if (!__set_phys_to_machine(pfn, mfn))
312 unsigned long arbitrary_virt_to_mfn(void *vaddr)
314 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
316 return PFN_DOWN(maddr.maddr);
319 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
321 unsigned long address = (unsigned long)vaddr;
327 * if the PFN is in the linear mapped vaddr range, we can just use
328 * the (quick) virt_to_machine() p2m lookup
330 if (virt_addr_valid(vaddr))
331 return virt_to_machine(vaddr);
333 /* otherwise we have to do a (slower) full page-table walk */
335 pte = lookup_address(address, &level);
337 offset = address & ~PAGE_MASK;
338 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
341 void make_lowmem_page_readonly(void *vaddr)
344 unsigned long address = (unsigned long)vaddr;
347 pte = lookup_address(address, &level);
350 ptev = pte_wrprotect(*pte);
352 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
356 void make_lowmem_page_readwrite(void *vaddr)
359 unsigned long address = (unsigned long)vaddr;
362 pte = lookup_address(address, &level);
365 ptev = pte_mkwrite(*pte);
367 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
372 static bool xen_page_pinned(void *ptr)
374 struct page *page = virt_to_page(ptr);
376 return PagePinned(page);
379 static void xen_extend_mmu_update(const struct mmu_update *update)
381 struct multicall_space mcs;
382 struct mmu_update *u;
384 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
386 if (mcs.mc != NULL) {
387 ADD_STATS(mmu_update_extended, 1);
388 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
392 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
393 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
395 ADD_STATS(mmu_update_histo[0], 1);
397 ADD_STATS(mmu_update, 1);
398 mcs = __xen_mc_entry(sizeof(*u));
399 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
400 ADD_STATS(mmu_update_histo[1], 1);
407 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
415 /* ptr may be ioremapped for 64-bit pagetable setup */
416 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
417 u.val = pmd_val_ma(val);
418 xen_extend_mmu_update(&u);
420 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
422 xen_mc_issue(PARAVIRT_LAZY_MMU);
427 void xen_set_pmd(pmd_t *ptr, pmd_t val)
429 ADD_STATS(pmd_update, 1);
431 /* If page is not pinned, we can just update the entry
433 if (!xen_page_pinned(ptr)) {
438 ADD_STATS(pmd_update_pinned, 1);
440 xen_set_pmd_hyper(ptr, val);
444 * Associate a virtual page frame with a given physical page frame
445 * and protection flags for that frame.
447 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
449 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
452 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
453 pte_t *ptep, pte_t pteval)
455 ADD_STATS(set_pte_at, 1);
456 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
457 ADD_STATS(set_pte_at_current, mm == current->mm);
458 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
460 if (mm == current->mm || mm == &init_mm) {
461 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
462 struct multicall_space mcs;
463 mcs = xen_mc_entry(0);
465 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
466 ADD_STATS(set_pte_at_batched, 1);
467 xen_mc_issue(PARAVIRT_LAZY_MMU);
470 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
473 xen_set_pte(ptep, pteval);
478 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
479 unsigned long addr, pte_t *ptep)
481 /* Just return the pte as-is. We preserve the bits on commit */
485 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
486 pte_t *ptep, pte_t pte)
492 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
493 u.val = pte_val_ma(pte);
494 xen_extend_mmu_update(&u);
496 ADD_STATS(prot_commit, 1);
497 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
499 xen_mc_issue(PARAVIRT_LAZY_MMU);
502 /* Assume pteval_t is equivalent to all the other *val_t types. */
503 static pteval_t pte_mfn_to_pfn(pteval_t val)
505 if (val & _PAGE_PRESENT) {
506 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
507 pteval_t flags = val & PTE_FLAGS_MASK;
508 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
514 static pteval_t pte_pfn_to_mfn(pteval_t val)
516 if (val & _PAGE_PRESENT) {
517 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
518 pteval_t flags = val & PTE_FLAGS_MASK;
519 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
525 pteval_t xen_pte_val(pte_t pte)
527 return pte_mfn_to_pfn(pte.pte);
529 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
531 pgdval_t xen_pgd_val(pgd_t pgd)
533 return pte_mfn_to_pfn(pgd.pgd);
535 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
537 pte_t xen_make_pte(pteval_t pte)
539 pte = pte_pfn_to_mfn(pte);
540 return native_make_pte(pte);
542 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
544 pgd_t xen_make_pgd(pgdval_t pgd)
546 pgd = pte_pfn_to_mfn(pgd);
547 return native_make_pgd(pgd);
549 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
551 pmdval_t xen_pmd_val(pmd_t pmd)
553 return pte_mfn_to_pfn(pmd.pmd);
555 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
557 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
565 /* ptr may be ioremapped for 64-bit pagetable setup */
566 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
567 u.val = pud_val_ma(val);
568 xen_extend_mmu_update(&u);
570 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
572 xen_mc_issue(PARAVIRT_LAZY_MMU);
577 void xen_set_pud(pud_t *ptr, pud_t val)
579 ADD_STATS(pud_update, 1);
581 /* If page is not pinned, we can just update the entry
583 if (!xen_page_pinned(ptr)) {
588 ADD_STATS(pud_update_pinned, 1);
590 xen_set_pud_hyper(ptr, val);
593 void xen_set_pte(pte_t *ptep, pte_t pte)
595 ADD_STATS(pte_update, 1);
596 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
597 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
599 #ifdef CONFIG_X86_PAE
600 ptep->pte_high = pte.pte_high;
602 ptep->pte_low = pte.pte_low;
608 #ifdef CONFIG_X86_PAE
609 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
611 set_64bit((u64 *)ptep, native_pte_val(pte));
614 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
617 smp_wmb(); /* make sure low gets written first */
621 void xen_pmd_clear(pmd_t *pmdp)
623 set_pmd(pmdp, __pmd(0));
625 #endif /* CONFIG_X86_PAE */
627 pmd_t xen_make_pmd(pmdval_t pmd)
629 pmd = pte_pfn_to_mfn(pmd);
630 return native_make_pmd(pmd);
632 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
634 #if PAGETABLE_LEVELS == 4
635 pudval_t xen_pud_val(pud_t pud)
637 return pte_mfn_to_pfn(pud.pud);
639 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
641 pud_t xen_make_pud(pudval_t pud)
643 pud = pte_pfn_to_mfn(pud);
645 return native_make_pud(pud);
647 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
649 pgd_t *xen_get_user_pgd(pgd_t *pgd)
651 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
652 unsigned offset = pgd - pgd_page;
653 pgd_t *user_ptr = NULL;
655 if (offset < pgd_index(USER_LIMIT)) {
656 struct page *page = virt_to_page(pgd_page);
657 user_ptr = (pgd_t *)page->private;
665 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
669 u.ptr = virt_to_machine(ptr).maddr;
670 u.val = pgd_val_ma(val);
671 xen_extend_mmu_update(&u);
675 * Raw hypercall-based set_pgd, intended for in early boot before
676 * there's a page structure. This implies:
677 * 1. The only existing pagetable is the kernel's
678 * 2. It is always pinned
679 * 3. It has no user pagetable attached to it
681 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
687 __xen_set_pgd_hyper(ptr, val);
689 xen_mc_issue(PARAVIRT_LAZY_MMU);
694 void xen_set_pgd(pgd_t *ptr, pgd_t val)
696 pgd_t *user_ptr = xen_get_user_pgd(ptr);
698 ADD_STATS(pgd_update, 1);
700 /* If page is not pinned, we can just update the entry
702 if (!xen_page_pinned(ptr)) {
705 WARN_ON(xen_page_pinned(user_ptr));
711 ADD_STATS(pgd_update_pinned, 1);
712 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
714 /* If it's pinned, then we can at least batch the kernel and
715 user updates together. */
718 __xen_set_pgd_hyper(ptr, val);
720 __xen_set_pgd_hyper(user_ptr, val);
722 xen_mc_issue(PARAVIRT_LAZY_MMU);
724 #endif /* PAGETABLE_LEVELS == 4 */
727 * (Yet another) pagetable walker. This one is intended for pinning a
728 * pagetable. This means that it walks a pagetable and calls the
729 * callback function on each page it finds making up the page table,
730 * at every level. It walks the entire pagetable, but it only bothers
731 * pinning pte pages which are below limit. In the normal case this
732 * will be STACK_TOP_MAX, but at boot we need to pin up to
735 * For 32-bit the important bit is that we don't pin beyond there,
736 * because then we start getting into Xen's ptes.
738 * For 64-bit, we must skip the Xen hole in the middle of the address
739 * space, just after the big x86-64 virtual hole.
741 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
742 int (*func)(struct mm_struct *mm, struct page *,
747 unsigned hole_low, hole_high;
748 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
749 unsigned pgdidx, pudidx, pmdidx;
751 /* The limit is the last byte to be touched */
753 BUG_ON(limit >= FIXADDR_TOP);
755 if (xen_feature(XENFEAT_auto_translated_physmap))
759 * 64-bit has a great big hole in the middle of the address
760 * space, which contains the Xen mappings. On 32-bit these
761 * will end up making a zero-sized hole and so is a no-op.
763 hole_low = pgd_index(USER_LIMIT);
764 hole_high = pgd_index(PAGE_OFFSET);
766 pgdidx_limit = pgd_index(limit);
768 pudidx_limit = pud_index(limit);
773 pmdidx_limit = pmd_index(limit);
778 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
781 if (pgdidx >= hole_low && pgdidx < hole_high)
784 if (!pgd_val(pgd[pgdidx]))
787 pud = pud_offset(&pgd[pgdidx], 0);
789 if (PTRS_PER_PUD > 1) /* not folded */
790 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
792 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
795 if (pgdidx == pgdidx_limit &&
796 pudidx > pudidx_limit)
799 if (pud_none(pud[pudidx]))
802 pmd = pmd_offset(&pud[pudidx], 0);
804 if (PTRS_PER_PMD > 1) /* not folded */
805 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
807 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
810 if (pgdidx == pgdidx_limit &&
811 pudidx == pudidx_limit &&
812 pmdidx > pmdidx_limit)
815 if (pmd_none(pmd[pmdidx]))
818 pte = pmd_page(pmd[pmdidx]);
819 flush |= (*func)(mm, pte, PT_PTE);
825 /* Do the top level last, so that the callbacks can use it as
826 a cue to do final things like tlb flushes. */
827 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
832 static int xen_pgd_walk(struct mm_struct *mm,
833 int (*func)(struct mm_struct *mm, struct page *,
837 return __xen_pgd_walk(mm, mm->pgd, func, limit);
840 /* If we're using split pte locks, then take the page's lock and
841 return a pointer to it. Otherwise return NULL. */
842 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
844 spinlock_t *ptl = NULL;
846 #if USE_SPLIT_PTLOCKS
847 ptl = __pte_lockptr(page);
848 spin_lock_nest_lock(ptl, &mm->page_table_lock);
854 static void xen_pte_unlock(void *v)
860 static void xen_do_pin(unsigned level, unsigned long pfn)
862 struct mmuext_op *op;
863 struct multicall_space mcs;
865 mcs = __xen_mc_entry(sizeof(*op));
868 op->arg1.mfn = pfn_to_mfn(pfn);
869 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
872 static int xen_pin_page(struct mm_struct *mm, struct page *page,
875 unsigned pgfl = TestSetPagePinned(page);
879 flush = 0; /* already pinned */
880 else if (PageHighMem(page))
881 /* kmaps need flushing if we found an unpinned
885 void *pt = lowmem_page_address(page);
886 unsigned long pfn = page_to_pfn(page);
887 struct multicall_space mcs = __xen_mc_entry(0);
893 * We need to hold the pagetable lock between the time
894 * we make the pagetable RO and when we actually pin
895 * it. If we don't, then other users may come in and
896 * attempt to update the pagetable by writing it,
897 * which will fail because the memory is RO but not
898 * pinned, so Xen won't do the trap'n'emulate.
900 * If we're using split pte locks, we can't hold the
901 * entire pagetable's worth of locks during the
902 * traverse, because we may wrap the preempt count (8
903 * bits). The solution is to mark RO and pin each PTE
904 * page while holding the lock. This means the number
905 * of locks we end up holding is never more than a
906 * batch size (~32 entries, at present).
908 * If we're not using split pte locks, we needn't pin
909 * the PTE pages independently, because we're
910 * protected by the overall pagetable lock.
914 ptl = xen_pte_lock(page, mm);
916 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
917 pfn_pte(pfn, PAGE_KERNEL_RO),
918 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
921 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
923 /* Queue a deferred unlock for when this batch
925 xen_mc_callback(xen_pte_unlock, ptl);
932 /* This is called just after a mm has been created, but it has not
933 been used yet. We need to make sure that its pagetable is all
934 read-only, and can be pinned. */
935 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
941 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
942 /* re-enable interrupts for flushing */
952 pgd_t *user_pgd = xen_get_user_pgd(pgd);
954 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
957 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
958 xen_do_pin(MMUEXT_PIN_L4_TABLE,
959 PFN_DOWN(__pa(user_pgd)));
962 #else /* CONFIG_X86_32 */
963 #ifdef CONFIG_X86_PAE
964 /* Need to make sure unshared kernel PMD is pinnable */
965 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
968 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
969 #endif /* CONFIG_X86_64 */
973 static void xen_pgd_pin(struct mm_struct *mm)
975 __xen_pgd_pin(mm, mm->pgd);
979 * On save, we need to pin all pagetables to make sure they get their
980 * mfns turned into pfns. Search the list for any unpinned pgds and pin
981 * them (unpinned pgds are not currently in use, probably because the
982 * process is under construction or destruction).
984 * Expected to be called in stop_machine() ("equivalent to taking
985 * every spinlock in the system"), so the locking doesn't really
986 * matter all that much.
988 void xen_mm_pin_all(void)
993 spin_lock_irqsave(&pgd_lock, flags);
995 list_for_each_entry(page, &pgd_list, lru) {
996 if (!PagePinned(page)) {
997 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
998 SetPageSavePinned(page);
1002 spin_unlock_irqrestore(&pgd_lock, flags);
1006 * The init_mm pagetable is really pinned as soon as its created, but
1007 * that's before we have page structures to store the bits. So do all
1008 * the book-keeping now.
1010 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1011 enum pt_level level)
1013 SetPagePinned(page);
1017 static void __init xen_mark_init_mm_pinned(void)
1019 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1022 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1023 enum pt_level level)
1025 unsigned pgfl = TestClearPagePinned(page);
1027 if (pgfl && !PageHighMem(page)) {
1028 void *pt = lowmem_page_address(page);
1029 unsigned long pfn = page_to_pfn(page);
1030 spinlock_t *ptl = NULL;
1031 struct multicall_space mcs;
1034 * Do the converse to pin_page. If we're using split
1035 * pte locks, we must be holding the lock for while
1036 * the pte page is unpinned but still RO to prevent
1037 * concurrent updates from seeing it in this
1038 * partially-pinned state.
1040 if (level == PT_PTE) {
1041 ptl = xen_pte_lock(page, mm);
1044 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1047 mcs = __xen_mc_entry(0);
1049 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1050 pfn_pte(pfn, PAGE_KERNEL),
1051 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1054 /* unlock when batch completed */
1055 xen_mc_callback(xen_pte_unlock, ptl);
1059 return 0; /* never need to flush on unpin */
1062 /* Release a pagetables pages back as normal RW */
1063 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1067 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1069 #ifdef CONFIG_X86_64
1071 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1074 xen_do_pin(MMUEXT_UNPIN_TABLE,
1075 PFN_DOWN(__pa(user_pgd)));
1076 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1081 #ifdef CONFIG_X86_PAE
1082 /* Need to make sure unshared kernel PMD is unpinned */
1083 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1087 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1092 static void xen_pgd_unpin(struct mm_struct *mm)
1094 __xen_pgd_unpin(mm, mm->pgd);
1098 * On resume, undo any pinning done at save, so that the rest of the
1099 * kernel doesn't see any unexpected pinned pagetables.
1101 void xen_mm_unpin_all(void)
1103 unsigned long flags;
1106 spin_lock_irqsave(&pgd_lock, flags);
1108 list_for_each_entry(page, &pgd_list, lru) {
1109 if (PageSavePinned(page)) {
1110 BUG_ON(!PagePinned(page));
1111 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1112 ClearPageSavePinned(page);
1116 spin_unlock_irqrestore(&pgd_lock, flags);
1119 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1121 spin_lock(&next->page_table_lock);
1123 spin_unlock(&next->page_table_lock);
1126 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1128 spin_lock(&mm->page_table_lock);
1130 spin_unlock(&mm->page_table_lock);
1135 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1136 we need to repoint it somewhere else before we can unpin it. */
1137 static void drop_other_mm_ref(void *info)
1139 struct mm_struct *mm = info;
1140 struct mm_struct *active_mm;
1142 active_mm = percpu_read(cpu_tlbstate.active_mm);
1144 if (active_mm == mm)
1145 leave_mm(smp_processor_id());
1147 /* If this cpu still has a stale cr3 reference, then make sure
1148 it has been flushed. */
1149 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1150 load_cr3(swapper_pg_dir);
1153 static void xen_drop_mm_ref(struct mm_struct *mm)
1158 if (current->active_mm == mm) {
1159 if (current->mm == mm)
1160 load_cr3(swapper_pg_dir);
1162 leave_mm(smp_processor_id());
1165 /* Get the "official" set of cpus referring to our pagetable. */
1166 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1167 for_each_online_cpu(cpu) {
1168 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1169 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1171 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1175 cpumask_copy(mask, mm_cpumask(mm));
1177 /* It's possible that a vcpu may have a stale reference to our
1178 cr3, because its in lazy mode, and it hasn't yet flushed
1179 its set of pending hypercalls yet. In this case, we can
1180 look at its actual current cr3 value, and force it to flush
1182 for_each_online_cpu(cpu) {
1183 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1184 cpumask_set_cpu(cpu, mask);
1187 if (!cpumask_empty(mask))
1188 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1189 free_cpumask_var(mask);
1192 static void xen_drop_mm_ref(struct mm_struct *mm)
1194 if (current->active_mm == mm)
1195 load_cr3(swapper_pg_dir);
1200 * While a process runs, Xen pins its pagetables, which means that the
1201 * hypervisor forces it to be read-only, and it controls all updates
1202 * to it. This means that all pagetable updates have to go via the
1203 * hypervisor, which is moderately expensive.
1205 * Since we're pulling the pagetable down, we switch to use init_mm,
1206 * unpin old process pagetable and mark it all read-write, which
1207 * allows further operations on it to be simple memory accesses.
1209 * The only subtle point is that another CPU may be still using the
1210 * pagetable because of lazy tlb flushing. This means we need need to
1211 * switch all CPUs off this pagetable before we can unpin it.
1213 void xen_exit_mmap(struct mm_struct *mm)
1215 get_cpu(); /* make sure we don't move around */
1216 xen_drop_mm_ref(mm);
1219 spin_lock(&mm->page_table_lock);
1221 /* pgd may not be pinned in the error exit path of execve */
1222 if (xen_page_pinned(mm->pgd))
1225 spin_unlock(&mm->page_table_lock);
1228 static __init void xen_pagetable_setup_start(pgd_t *base)
1232 static void xen_post_allocator_init(void);
1234 static __init void xen_pagetable_setup_done(pgd_t *base)
1236 xen_setup_shared_info();
1237 xen_post_allocator_init();
1240 static void xen_write_cr2(unsigned long cr2)
1242 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1245 static unsigned long xen_read_cr2(void)
1247 return percpu_read(xen_vcpu)->arch.cr2;
1250 unsigned long xen_read_cr2_direct(void)
1252 return percpu_read(xen_vcpu_info.arch.cr2);
1255 static void xen_flush_tlb(void)
1257 struct mmuext_op *op;
1258 struct multicall_space mcs;
1262 mcs = xen_mc_entry(sizeof(*op));
1265 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1266 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1268 xen_mc_issue(PARAVIRT_LAZY_MMU);
1273 static void xen_flush_tlb_single(unsigned long addr)
1275 struct mmuext_op *op;
1276 struct multicall_space mcs;
1280 mcs = xen_mc_entry(sizeof(*op));
1282 op->cmd = MMUEXT_INVLPG_LOCAL;
1283 op->arg1.linear_addr = addr & PAGE_MASK;
1284 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1286 xen_mc_issue(PARAVIRT_LAZY_MMU);
1291 static void xen_flush_tlb_others(const struct cpumask *cpus,
1292 struct mm_struct *mm, unsigned long va)
1295 struct mmuext_op op;
1296 DECLARE_BITMAP(mask, NR_CPUS);
1298 struct multicall_space mcs;
1300 if (cpumask_empty(cpus))
1301 return; /* nothing to do */
1303 mcs = xen_mc_entry(sizeof(*args));
1305 args->op.arg2.vcpumask = to_cpumask(args->mask);
1307 /* Remove us, and any offline CPUS. */
1308 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1309 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1311 if (va == TLB_FLUSH_ALL) {
1312 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1314 args->op.cmd = MMUEXT_INVLPG_MULTI;
1315 args->op.arg1.linear_addr = va;
1318 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1320 xen_mc_issue(PARAVIRT_LAZY_MMU);
1323 static unsigned long xen_read_cr3(void)
1325 return percpu_read(xen_cr3);
1328 static void set_current_cr3(void *v)
1330 percpu_write(xen_current_cr3, (unsigned long)v);
1333 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1335 struct mmuext_op *op;
1336 struct multicall_space mcs;
1340 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1344 WARN_ON(mfn == 0 && kernel);
1346 mcs = __xen_mc_entry(sizeof(*op));
1349 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1352 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1355 percpu_write(xen_cr3, cr3);
1357 /* Update xen_current_cr3 once the batch has actually
1359 xen_mc_callback(set_current_cr3, (void *)cr3);
1363 static void xen_write_cr3(unsigned long cr3)
1365 BUG_ON(preemptible());
1367 xen_mc_batch(); /* disables interrupts */
1369 /* Update while interrupts are disabled, so its atomic with
1371 percpu_write(xen_cr3, cr3);
1373 __xen_write_cr3(true, cr3);
1375 #ifdef CONFIG_X86_64
1377 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1379 __xen_write_cr3(false, __pa(user_pgd));
1381 __xen_write_cr3(false, 0);
1385 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1388 static int xen_pgd_alloc(struct mm_struct *mm)
1390 pgd_t *pgd = mm->pgd;
1393 BUG_ON(PagePinned(virt_to_page(pgd)));
1395 #ifdef CONFIG_X86_64
1397 struct page *page = virt_to_page(pgd);
1400 BUG_ON(page->private != 0);
1404 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1405 page->private = (unsigned long)user_pgd;
1407 if (user_pgd != NULL) {
1408 user_pgd[pgd_index(VSYSCALL_START)] =
1409 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1413 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1420 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1422 #ifdef CONFIG_X86_64
1423 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1426 free_page((unsigned long)user_pgd);
1430 #ifdef CONFIG_X86_32
1431 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1433 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1434 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1435 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1441 /* Init-time set_pte while constructing initial pagetables, which
1442 doesn't allow RO pagetable pages to be remapped RW */
1443 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1445 pte = mask_rw_pte(ptep, pte);
1447 xen_set_pte(ptep, pte);
1451 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1453 struct mmuext_op op;
1455 op.arg1.mfn = pfn_to_mfn(pfn);
1456 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1460 /* Early in boot, while setting up the initial pagetable, assume
1461 everything is pinned. */
1462 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1464 #ifdef CONFIG_FLATMEM
1465 BUG_ON(mem_map); /* should only be used early */
1467 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1468 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1471 /* Used for pmd and pud */
1472 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1474 #ifdef CONFIG_FLATMEM
1475 BUG_ON(mem_map); /* should only be used early */
1477 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1480 /* Early release_pte assumes that all pts are pinned, since there's
1481 only init_mm and anything attached to that is pinned. */
1482 static __init void xen_release_pte_init(unsigned long pfn)
1484 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1485 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1488 static __init void xen_release_pmd_init(unsigned long pfn)
1490 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1493 /* This needs to make sure the new pte page is pinned iff its being
1494 attached to a pinned pagetable. */
1495 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1497 struct page *page = pfn_to_page(pfn);
1499 if (PagePinned(virt_to_page(mm->pgd))) {
1500 SetPagePinned(page);
1503 if (!PageHighMem(page)) {
1504 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1505 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1506 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1508 /* make sure there are no stray mappings of
1510 kmap_flush_unused();
1515 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1517 xen_alloc_ptpage(mm, pfn, PT_PTE);
1520 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1522 xen_alloc_ptpage(mm, pfn, PT_PMD);
1525 /* This should never happen until we're OK to use struct page */
1526 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1528 struct page *page = pfn_to_page(pfn);
1530 if (PagePinned(page)) {
1531 if (!PageHighMem(page)) {
1532 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1533 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1534 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1536 ClearPagePinned(page);
1540 static void xen_release_pte(unsigned long pfn)
1542 xen_release_ptpage(pfn, PT_PTE);
1545 static void xen_release_pmd(unsigned long pfn)
1547 xen_release_ptpage(pfn, PT_PMD);
1550 #if PAGETABLE_LEVELS == 4
1551 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1553 xen_alloc_ptpage(mm, pfn, PT_PUD);
1556 static void xen_release_pud(unsigned long pfn)
1558 xen_release_ptpage(pfn, PT_PUD);
1562 void __init xen_reserve_top(void)
1564 #ifdef CONFIG_X86_32
1565 unsigned long top = HYPERVISOR_VIRT_START;
1566 struct xen_platform_parameters pp;
1568 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1569 top = pp.virt_start;
1571 reserve_top_address(-top);
1572 #endif /* CONFIG_X86_32 */
1576 * Like __va(), but returns address in the kernel mapping (which is
1577 * all we have until the physical memory mapping has been set up.
1579 static void *__ka(phys_addr_t paddr)
1581 #ifdef CONFIG_X86_64
1582 return (void *)(paddr + __START_KERNEL_map);
1588 /* Convert a machine address to physical address */
1589 static unsigned long m2p(phys_addr_t maddr)
1593 maddr &= PTE_PFN_MASK;
1594 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1599 /* Convert a machine address to kernel virtual */
1600 static void *m2v(phys_addr_t maddr)
1602 return __ka(m2p(maddr));
1605 static void set_page_prot(void *addr, pgprot_t prot)
1607 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1608 pte_t pte = pfn_pte(pfn, prot);
1610 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1614 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1616 unsigned pmdidx, pteidx;
1622 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1625 /* Reuse or allocate a page of ptes */
1626 if (pmd_present(pmd[pmdidx]))
1627 pte_page = m2v(pmd[pmdidx].pmd);
1629 /* Check for free pte pages */
1630 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1633 pte_page = &level1_ident_pgt[ident_pte];
1634 ident_pte += PTRS_PER_PTE;
1636 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1639 /* Install mappings */
1640 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1643 if (pfn > max_pfn_mapped)
1644 max_pfn_mapped = pfn;
1646 if (!pte_none(pte_page[pteidx]))
1649 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1650 pte_page[pteidx] = pte;
1654 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1655 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1657 set_page_prot(pmd, PAGE_KERNEL_RO);
1660 #ifdef CONFIG_X86_64
1661 static void convert_pfn_mfn(void *v)
1666 /* All levels are converted the same way, so just treat them
1668 for (i = 0; i < PTRS_PER_PTE; i++)
1669 pte[i] = xen_make_pte(pte[i].pte);
1673 * Set up the inital kernel pagetable.
1675 * We can construct this by grafting the Xen provided pagetable into
1676 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1677 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1678 * means that only the kernel has a physical mapping to start with -
1679 * but that's enough to get __va working. We need to fill in the rest
1680 * of the physical mapping once some sort of allocator has been set
1683 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1684 unsigned long max_pfn)
1689 /* Zap identity mapping */
1690 init_level4_pgt[0] = __pgd(0);
1692 /* Pre-constructed entries are in pfn, so convert to mfn */
1693 convert_pfn_mfn(init_level4_pgt);
1694 convert_pfn_mfn(level3_ident_pgt);
1695 convert_pfn_mfn(level3_kernel_pgt);
1697 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1698 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1700 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1701 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1703 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1704 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1705 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1707 /* Set up identity map */
1708 xen_map_identity_early(level2_ident_pgt, max_pfn);
1710 /* Make pagetable pieces RO */
1711 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1712 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1713 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1714 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1715 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1716 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1718 /* Pin down new L4 */
1719 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1720 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1722 /* Unpin Xen-provided one */
1723 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1726 pgd = init_level4_pgt;
1729 * At this stage there can be no user pgd, and no page
1730 * structure to attach it to, so make sure we just set kernel
1734 __xen_write_cr3(true, __pa(pgd));
1735 xen_mc_issue(PARAVIRT_LAZY_CPU);
1737 reserve_early(__pa(xen_start_info->pt_base),
1738 __pa(xen_start_info->pt_base +
1739 xen_start_info->nr_pt_frames * PAGE_SIZE),
1744 #else /* !CONFIG_X86_64 */
1745 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1747 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1748 unsigned long max_pfn)
1752 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1753 xen_start_info->nr_pt_frames * PAGE_SIZE +
1756 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1757 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1759 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1761 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1762 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1763 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1765 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1766 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1767 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1769 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1771 xen_write_cr3(__pa(swapper_pg_dir));
1773 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1775 reserve_early(__pa(xen_start_info->pt_base),
1776 __pa(xen_start_info->pt_base +
1777 xen_start_info->nr_pt_frames * PAGE_SIZE),
1780 return swapper_pg_dir;
1782 #endif /* CONFIG_X86_64 */
1784 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1788 phys >>= PAGE_SHIFT;
1791 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1792 #ifdef CONFIG_X86_F00F_BUG
1795 #ifdef CONFIG_X86_32
1798 # ifdef CONFIG_HIGHMEM
1799 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1802 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1804 #ifdef CONFIG_X86_LOCAL_APIC
1805 case FIX_APIC_BASE: /* maps dummy local APIC */
1807 case FIX_TEXT_POKE0:
1808 case FIX_TEXT_POKE1:
1809 /* All local page mappings */
1810 pte = pfn_pte(phys, prot);
1814 pte = mfn_pte(phys, prot);
1818 __native_set_fixmap(idx, pte);
1820 #ifdef CONFIG_X86_64
1821 /* Replicate changes to map the vsyscall page into the user
1822 pagetable vsyscall mapping. */
1823 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1824 unsigned long vaddr = __fix_to_virt(idx);
1825 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1830 static __init void xen_post_allocator_init(void)
1832 pv_mmu_ops.set_pte = xen_set_pte;
1833 pv_mmu_ops.set_pmd = xen_set_pmd;
1834 pv_mmu_ops.set_pud = xen_set_pud;
1835 #if PAGETABLE_LEVELS == 4
1836 pv_mmu_ops.set_pgd = xen_set_pgd;
1839 /* This will work as long as patching hasn't happened yet
1840 (which it hasn't) */
1841 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1842 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1843 pv_mmu_ops.release_pte = xen_release_pte;
1844 pv_mmu_ops.release_pmd = xen_release_pmd;
1845 #if PAGETABLE_LEVELS == 4
1846 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1847 pv_mmu_ops.release_pud = xen_release_pud;
1850 #ifdef CONFIG_X86_64
1851 SetPagePinned(virt_to_page(level3_user_vsyscall));
1853 xen_mark_init_mm_pinned();
1856 static void xen_leave_lazy_mmu(void)
1860 paravirt_leave_lazy_mmu();
1864 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1865 .read_cr2 = xen_read_cr2,
1866 .write_cr2 = xen_write_cr2,
1868 .read_cr3 = xen_read_cr3,
1869 .write_cr3 = xen_write_cr3,
1871 .flush_tlb_user = xen_flush_tlb,
1872 .flush_tlb_kernel = xen_flush_tlb,
1873 .flush_tlb_single = xen_flush_tlb_single,
1874 .flush_tlb_others = xen_flush_tlb_others,
1876 .pte_update = paravirt_nop,
1877 .pte_update_defer = paravirt_nop,
1879 .pgd_alloc = xen_pgd_alloc,
1880 .pgd_free = xen_pgd_free,
1882 .alloc_pte = xen_alloc_pte_init,
1883 .release_pte = xen_release_pte_init,
1884 .alloc_pmd = xen_alloc_pmd_init,
1885 .alloc_pmd_clone = paravirt_nop,
1886 .release_pmd = xen_release_pmd_init,
1888 #ifdef CONFIG_X86_64
1889 .set_pte = xen_set_pte,
1891 .set_pte = xen_set_pte_init,
1893 .set_pte_at = xen_set_pte_at,
1894 .set_pmd = xen_set_pmd_hyper,
1896 .ptep_modify_prot_start = __ptep_modify_prot_start,
1897 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1899 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1900 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1902 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1903 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1905 #ifdef CONFIG_X86_PAE
1906 .set_pte_atomic = xen_set_pte_atomic,
1907 .pte_clear = xen_pte_clear,
1908 .pmd_clear = xen_pmd_clear,
1909 #endif /* CONFIG_X86_PAE */
1910 .set_pud = xen_set_pud_hyper,
1912 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
1913 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
1915 #if PAGETABLE_LEVELS == 4
1916 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
1917 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
1918 .set_pgd = xen_set_pgd_hyper,
1920 .alloc_pud = xen_alloc_pmd_init,
1921 .release_pud = xen_release_pmd_init,
1922 #endif /* PAGETABLE_LEVELS == 4 */
1924 .activate_mm = xen_activate_mm,
1925 .dup_mmap = xen_dup_mmap,
1926 .exit_mmap = xen_exit_mmap,
1929 .enter = paravirt_enter_lazy_mmu,
1930 .leave = xen_leave_lazy_mmu,
1933 .set_fixmap = xen_set_fixmap,
1936 void __init xen_init_mmu_ops(void)
1938 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
1939 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
1940 pv_mmu_ops = xen_mmu_ops;
1943 #ifdef CONFIG_XEN_DEBUG_FS
1945 static struct dentry *d_mmu_debug;
1947 static int __init xen_mmu_debugfs(void)
1949 struct dentry *d_xen = xen_init_debugfs();
1954 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1956 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
1958 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
1959 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
1960 &mmu_stats.pgd_update_pinned);
1961 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
1962 &mmu_stats.pgd_update_pinned);
1964 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
1965 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
1966 &mmu_stats.pud_update_pinned);
1967 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
1968 &mmu_stats.pud_update_pinned);
1970 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
1971 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
1972 &mmu_stats.pmd_update_pinned);
1973 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
1974 &mmu_stats.pmd_update_pinned);
1976 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
1977 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
1978 // &mmu_stats.pte_update_pinned);
1979 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
1980 &mmu_stats.pte_update_pinned);
1982 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
1983 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
1984 &mmu_stats.mmu_update_extended);
1985 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
1986 mmu_stats.mmu_update_histo, 20);
1988 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
1989 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
1990 &mmu_stats.set_pte_at_batched);
1991 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
1992 &mmu_stats.set_pte_at_current);
1993 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
1994 &mmu_stats.set_pte_at_kernel);
1996 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
1997 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
1998 &mmu_stats.prot_commit_batched);
2002 fs_initcall(xen_mmu_debugfs);
2004 #endif /* CONFIG_XEN_DEBUG_FS */