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>
46 #include <asm/pgtable.h>
47 #include <asm/tlbflush.h>
48 #include <asm/fixmap.h>
49 #include <asm/mmu_context.h>
50 #include <asm/setup.h>
51 #include <asm/paravirt.h>
52 #include <asm/linkage.h>
54 #include <asm/xen/hypercall.h>
55 #include <asm/xen/hypervisor.h>
58 #include <xen/interface/xen.h>
59 #include <xen/interface/version.h>
60 #include <xen/hvc-console.h>
62 #include "multicalls.h"
66 #define MMU_UPDATE_HISTO 30
68 #ifdef CONFIG_XEN_DEBUG_FS
72 u32 pgd_update_pinned;
73 u32 pgd_update_batched;
76 u32 pud_update_pinned;
77 u32 pud_update_batched;
80 u32 pmd_update_pinned;
81 u32 pmd_update_batched;
84 u32 pte_update_pinned;
85 u32 pte_update_batched;
88 u32 mmu_update_extended;
89 u32 mmu_update_histo[MMU_UPDATE_HISTO];
92 u32 prot_commit_batched;
95 u32 set_pte_at_batched;
96 u32 set_pte_at_pinned;
97 u32 set_pte_at_current;
98 u32 set_pte_at_kernel;
101 static u8 zero_stats;
103 static inline void check_zero(void)
105 if (unlikely(zero_stats)) {
106 memset(&mmu_stats, 0, sizeof(mmu_stats));
111 #define ADD_STATS(elem, val) \
112 do { check_zero(); mmu_stats.elem += (val); } while(0)
114 #else /* !CONFIG_XEN_DEBUG_FS */
116 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
118 #endif /* CONFIG_XEN_DEBUG_FS */
122 * Identity map, in addition to plain kernel map. This needs to be
123 * large enough to allocate page table pages to allocate the rest.
124 * Each page can map 2MB.
126 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
129 /* l3 pud for userspace vsyscall mapping */
130 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
131 #endif /* CONFIG_X86_64 */
134 * Note about cr3 (pagetable base) values:
136 * xen_cr3 contains the current logical cr3 value; it contains the
137 * last set cr3. This may not be the current effective cr3, because
138 * its update may be being lazily deferred. However, a vcpu looking
139 * at its own cr3 can use this value knowing that it everything will
140 * be self-consistent.
142 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
143 * hypercall to set the vcpu cr3 is complete (so it may be a little
144 * out of date, but it will never be set early). If one vcpu is
145 * looking at another vcpu's cr3 value, it should use this variable.
147 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
148 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
152 * Just beyond the highest usermode address. STACK_TOP_MAX has a
153 * redzone above it, so round it up to a PGD boundary.
155 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
158 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
159 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
161 /* Placeholder for holes in the address space */
162 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
163 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
165 /* Array of pointers to pages containing p2m entries */
166 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
167 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
169 /* Arrays of p2m arrays expressed in mfns used for save/restore */
170 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
172 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
175 static inline unsigned p2m_top_index(unsigned long pfn)
177 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
178 return pfn / P2M_ENTRIES_PER_PAGE;
181 static inline unsigned p2m_index(unsigned long pfn)
183 return pfn % P2M_ENTRIES_PER_PAGE;
186 /* Build the parallel p2m_top_mfn structures */
187 static void __init xen_build_mfn_list_list(void)
191 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
192 unsigned topidx = p2m_top_index(pfn);
194 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
197 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
198 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
199 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
203 void xen_setup_mfn_list_list(void)
205 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
207 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
208 virt_to_mfn(p2m_top_mfn_list);
209 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
212 /* Set up p2m_top to point to the domain-builder provided p2m pages */
213 void __init xen_build_dynamic_phys_to_machine(void)
215 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
216 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
219 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
220 unsigned topidx = p2m_top_index(pfn);
222 p2m_top[topidx] = &mfn_list[pfn];
225 xen_build_mfn_list_list();
228 unsigned long get_phys_to_machine(unsigned long pfn)
230 unsigned topidx, idx;
232 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
233 return INVALID_P2M_ENTRY;
235 topidx = p2m_top_index(pfn);
236 idx = p2m_index(pfn);
237 return p2m_top[topidx][idx];
239 EXPORT_SYMBOL_GPL(get_phys_to_machine);
241 /* install a new p2m_top page */
242 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
244 unsigned topidx = p2m_top_index(pfn);
245 unsigned long **pfnp, *mfnp;
248 pfnp = &p2m_top[topidx];
249 mfnp = &p2m_top_mfn[topidx];
251 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
252 p[i] = INVALID_P2M_ENTRY;
254 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
255 *mfnp = virt_to_mfn(p);
262 static void alloc_p2m(unsigned long pfn)
266 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
269 if (!install_p2mtop_page(pfn, p))
270 free_page((unsigned long)p);
273 /* Try to install p2m mapping; fail if intermediate bits missing */
274 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
276 unsigned topidx, idx;
278 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
279 BUG_ON(mfn != INVALID_P2M_ENTRY);
283 topidx = p2m_top_index(pfn);
284 if (p2m_top[topidx] == p2m_missing) {
285 if (mfn == INVALID_P2M_ENTRY)
290 idx = p2m_index(pfn);
291 p2m_top[topidx][idx] = mfn;
296 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
298 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
299 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
303 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
306 if (!__set_phys_to_machine(pfn, mfn))
311 unsigned long arbitrary_virt_to_mfn(void *vaddr)
313 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
315 return PFN_DOWN(maddr.maddr);
318 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
320 unsigned long address = (unsigned long)vaddr;
326 * if the PFN is in the linear mapped vaddr range, we can just use
327 * the (quick) virt_to_machine() p2m lookup
329 if (virt_addr_valid(vaddr))
330 return virt_to_machine(vaddr);
332 /* otherwise we have to do a (slower) full page-table walk */
334 pte = lookup_address(address, &level);
336 offset = address & ~PAGE_MASK;
337 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
340 void make_lowmem_page_readonly(void *vaddr)
343 unsigned long address = (unsigned long)vaddr;
346 pte = lookup_address(address, &level);
349 ptev = pte_wrprotect(*pte);
351 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
355 void make_lowmem_page_readwrite(void *vaddr)
358 unsigned long address = (unsigned long)vaddr;
361 pte = lookup_address(address, &level);
364 ptev = pte_mkwrite(*pte);
366 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
371 static bool xen_page_pinned(void *ptr)
373 struct page *page = virt_to_page(ptr);
375 return PagePinned(page);
378 static void xen_extend_mmu_update(const struct mmu_update *update)
380 struct multicall_space mcs;
381 struct mmu_update *u;
383 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
385 if (mcs.mc != NULL) {
386 ADD_STATS(mmu_update_extended, 1);
387 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
391 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
392 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
394 ADD_STATS(mmu_update_histo[0], 1);
396 ADD_STATS(mmu_update, 1);
397 mcs = __xen_mc_entry(sizeof(*u));
398 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
399 ADD_STATS(mmu_update_histo[1], 1);
406 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
414 /* ptr may be ioremapped for 64-bit pagetable setup */
415 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
416 u.val = pmd_val_ma(val);
417 xen_extend_mmu_update(&u);
419 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
421 xen_mc_issue(PARAVIRT_LAZY_MMU);
426 void xen_set_pmd(pmd_t *ptr, pmd_t val)
428 ADD_STATS(pmd_update, 1);
430 /* If page is not pinned, we can just update the entry
432 if (!xen_page_pinned(ptr)) {
437 ADD_STATS(pmd_update_pinned, 1);
439 xen_set_pmd_hyper(ptr, val);
443 * Associate a virtual page frame with a given physical page frame
444 * and protection flags for that frame.
446 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
448 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
451 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
452 pte_t *ptep, pte_t pteval)
454 ADD_STATS(set_pte_at, 1);
455 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
456 ADD_STATS(set_pte_at_current, mm == current->mm);
457 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
459 if (mm == current->mm || mm == &init_mm) {
460 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
461 struct multicall_space mcs;
462 mcs = xen_mc_entry(0);
464 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
465 ADD_STATS(set_pte_at_batched, 1);
466 xen_mc_issue(PARAVIRT_LAZY_MMU);
469 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
472 xen_set_pte(ptep, pteval);
477 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
478 unsigned long addr, pte_t *ptep)
480 /* Just return the pte as-is. We preserve the bits on commit */
484 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
485 pte_t *ptep, pte_t pte)
491 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
492 u.val = pte_val_ma(pte);
493 xen_extend_mmu_update(&u);
495 ADD_STATS(prot_commit, 1);
496 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
498 xen_mc_issue(PARAVIRT_LAZY_MMU);
501 /* Assume pteval_t is equivalent to all the other *val_t types. */
502 static pteval_t pte_mfn_to_pfn(pteval_t val)
504 if (val & _PAGE_PRESENT) {
505 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
506 pteval_t flags = val & PTE_FLAGS_MASK;
507 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
513 static pteval_t pte_pfn_to_mfn(pteval_t val)
515 if (val & _PAGE_PRESENT) {
516 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
517 pteval_t flags = val & PTE_FLAGS_MASK;
518 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
524 pteval_t xen_pte_val(pte_t pte)
526 return pte_mfn_to_pfn(pte.pte);
528 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
530 pgdval_t xen_pgd_val(pgd_t pgd)
532 return pte_mfn_to_pfn(pgd.pgd);
534 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
536 pte_t xen_make_pte(pteval_t pte)
538 pte = pte_pfn_to_mfn(pte);
539 return native_make_pte(pte);
541 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
543 pgd_t xen_make_pgd(pgdval_t pgd)
545 pgd = pte_pfn_to_mfn(pgd);
546 return native_make_pgd(pgd);
548 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
550 pmdval_t xen_pmd_val(pmd_t pmd)
552 return pte_mfn_to_pfn(pmd.pmd);
554 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
556 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
564 /* ptr may be ioremapped for 64-bit pagetable setup */
565 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
566 u.val = pud_val_ma(val);
567 xen_extend_mmu_update(&u);
569 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
571 xen_mc_issue(PARAVIRT_LAZY_MMU);
576 void xen_set_pud(pud_t *ptr, pud_t val)
578 ADD_STATS(pud_update, 1);
580 /* If page is not pinned, we can just update the entry
582 if (!xen_page_pinned(ptr)) {
587 ADD_STATS(pud_update_pinned, 1);
589 xen_set_pud_hyper(ptr, val);
592 void xen_set_pte(pte_t *ptep, pte_t pte)
594 ADD_STATS(pte_update, 1);
595 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
596 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
598 #ifdef CONFIG_X86_PAE
599 ptep->pte_high = pte.pte_high;
601 ptep->pte_low = pte.pte_low;
607 #ifdef CONFIG_X86_PAE
608 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
610 set_64bit((u64 *)ptep, native_pte_val(pte));
613 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
616 smp_wmb(); /* make sure low gets written first */
620 void xen_pmd_clear(pmd_t *pmdp)
622 set_pmd(pmdp, __pmd(0));
624 #endif /* CONFIG_X86_PAE */
626 pmd_t xen_make_pmd(pmdval_t pmd)
628 pmd = pte_pfn_to_mfn(pmd);
629 return native_make_pmd(pmd);
631 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
633 #if PAGETABLE_LEVELS == 4
634 pudval_t xen_pud_val(pud_t pud)
636 return pte_mfn_to_pfn(pud.pud);
638 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
640 pud_t xen_make_pud(pudval_t pud)
642 pud = pte_pfn_to_mfn(pud);
644 return native_make_pud(pud);
646 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
648 pgd_t *xen_get_user_pgd(pgd_t *pgd)
650 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
651 unsigned offset = pgd - pgd_page;
652 pgd_t *user_ptr = NULL;
654 if (offset < pgd_index(USER_LIMIT)) {
655 struct page *page = virt_to_page(pgd_page);
656 user_ptr = (pgd_t *)page->private;
664 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
668 u.ptr = virt_to_machine(ptr).maddr;
669 u.val = pgd_val_ma(val);
670 xen_extend_mmu_update(&u);
674 * Raw hypercall-based set_pgd, intended for in early boot before
675 * there's a page structure. This implies:
676 * 1. The only existing pagetable is the kernel's
677 * 2. It is always pinned
678 * 3. It has no user pagetable attached to it
680 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
686 __xen_set_pgd_hyper(ptr, val);
688 xen_mc_issue(PARAVIRT_LAZY_MMU);
693 void xen_set_pgd(pgd_t *ptr, pgd_t val)
695 pgd_t *user_ptr = xen_get_user_pgd(ptr);
697 ADD_STATS(pgd_update, 1);
699 /* If page is not pinned, we can just update the entry
701 if (!xen_page_pinned(ptr)) {
704 WARN_ON(xen_page_pinned(user_ptr));
710 ADD_STATS(pgd_update_pinned, 1);
711 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
713 /* If it's pinned, then we can at least batch the kernel and
714 user updates together. */
717 __xen_set_pgd_hyper(ptr, val);
719 __xen_set_pgd_hyper(user_ptr, val);
721 xen_mc_issue(PARAVIRT_LAZY_MMU);
723 #endif /* PAGETABLE_LEVELS == 4 */
726 * (Yet another) pagetable walker. This one is intended for pinning a
727 * pagetable. This means that it walks a pagetable and calls the
728 * callback function on each page it finds making up the page table,
729 * at every level. It walks the entire pagetable, but it only bothers
730 * pinning pte pages which are below limit. In the normal case this
731 * will be STACK_TOP_MAX, but at boot we need to pin up to
734 * For 32-bit the important bit is that we don't pin beyond there,
735 * because then we start getting into Xen's ptes.
737 * For 64-bit, we must skip the Xen hole in the middle of the address
738 * space, just after the big x86-64 virtual hole.
740 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
741 int (*func)(struct mm_struct *mm, struct page *,
746 unsigned hole_low, hole_high;
747 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
748 unsigned pgdidx, pudidx, pmdidx;
750 /* The limit is the last byte to be touched */
752 BUG_ON(limit >= FIXADDR_TOP);
754 if (xen_feature(XENFEAT_auto_translated_physmap))
758 * 64-bit has a great big hole in the middle of the address
759 * space, which contains the Xen mappings. On 32-bit these
760 * will end up making a zero-sized hole and so is a no-op.
762 hole_low = pgd_index(USER_LIMIT);
763 hole_high = pgd_index(PAGE_OFFSET);
765 pgdidx_limit = pgd_index(limit);
767 pudidx_limit = pud_index(limit);
772 pmdidx_limit = pmd_index(limit);
777 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
780 if (pgdidx >= hole_low && pgdidx < hole_high)
783 if (!pgd_val(pgd[pgdidx]))
786 pud = pud_offset(&pgd[pgdidx], 0);
788 if (PTRS_PER_PUD > 1) /* not folded */
789 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
791 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
794 if (pgdidx == pgdidx_limit &&
795 pudidx > pudidx_limit)
798 if (pud_none(pud[pudidx]))
801 pmd = pmd_offset(&pud[pudidx], 0);
803 if (PTRS_PER_PMD > 1) /* not folded */
804 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
806 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
809 if (pgdidx == pgdidx_limit &&
810 pudidx == pudidx_limit &&
811 pmdidx > pmdidx_limit)
814 if (pmd_none(pmd[pmdidx]))
817 pte = pmd_page(pmd[pmdidx]);
818 flush |= (*func)(mm, pte, PT_PTE);
824 /* Do the top level last, so that the callbacks can use it as
825 a cue to do final things like tlb flushes. */
826 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
831 static int xen_pgd_walk(struct mm_struct *mm,
832 int (*func)(struct mm_struct *mm, struct page *,
836 return __xen_pgd_walk(mm, mm->pgd, func, limit);
839 /* If we're using split pte locks, then take the page's lock and
840 return a pointer to it. Otherwise return NULL. */
841 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
843 spinlock_t *ptl = NULL;
845 #if USE_SPLIT_PTLOCKS
846 ptl = __pte_lockptr(page);
847 spin_lock_nest_lock(ptl, &mm->page_table_lock);
853 static void xen_pte_unlock(void *v)
859 static void xen_do_pin(unsigned level, unsigned long pfn)
861 struct mmuext_op *op;
862 struct multicall_space mcs;
864 mcs = __xen_mc_entry(sizeof(*op));
867 op->arg1.mfn = pfn_to_mfn(pfn);
868 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
871 static int xen_pin_page(struct mm_struct *mm, struct page *page,
874 unsigned pgfl = TestSetPagePinned(page);
878 flush = 0; /* already pinned */
879 else if (PageHighMem(page))
880 /* kmaps need flushing if we found an unpinned
884 void *pt = lowmem_page_address(page);
885 unsigned long pfn = page_to_pfn(page);
886 struct multicall_space mcs = __xen_mc_entry(0);
892 * We need to hold the pagetable lock between the time
893 * we make the pagetable RO and when we actually pin
894 * it. If we don't, then other users may come in and
895 * attempt to update the pagetable by writing it,
896 * which will fail because the memory is RO but not
897 * pinned, so Xen won't do the trap'n'emulate.
899 * If we're using split pte locks, we can't hold the
900 * entire pagetable's worth of locks during the
901 * traverse, because we may wrap the preempt count (8
902 * bits). The solution is to mark RO and pin each PTE
903 * page while holding the lock. This means the number
904 * of locks we end up holding is never more than a
905 * batch size (~32 entries, at present).
907 * If we're not using split pte locks, we needn't pin
908 * the PTE pages independently, because we're
909 * protected by the overall pagetable lock.
913 ptl = xen_pte_lock(page, mm);
915 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
916 pfn_pte(pfn, PAGE_KERNEL_RO),
917 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
920 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
922 /* Queue a deferred unlock for when this batch
924 xen_mc_callback(xen_pte_unlock, ptl);
931 /* This is called just after a mm has been created, but it has not
932 been used yet. We need to make sure that its pagetable is all
933 read-only, and can be pinned. */
934 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
940 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
941 /* re-enable interrupts for flushing */
951 pgd_t *user_pgd = xen_get_user_pgd(pgd);
953 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
956 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
957 xen_do_pin(MMUEXT_PIN_L4_TABLE,
958 PFN_DOWN(__pa(user_pgd)));
961 #else /* CONFIG_X86_32 */
962 #ifdef CONFIG_X86_PAE
963 /* Need to make sure unshared kernel PMD is pinnable */
964 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
967 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
968 #endif /* CONFIG_X86_64 */
972 static void xen_pgd_pin(struct mm_struct *mm)
974 __xen_pgd_pin(mm, mm->pgd);
978 * On save, we need to pin all pagetables to make sure they get their
979 * mfns turned into pfns. Search the list for any unpinned pgds and pin
980 * them (unpinned pgds are not currently in use, probably because the
981 * process is under construction or destruction).
983 * Expected to be called in stop_machine() ("equivalent to taking
984 * every spinlock in the system"), so the locking doesn't really
985 * matter all that much.
987 void xen_mm_pin_all(void)
992 spin_lock_irqsave(&pgd_lock, flags);
994 list_for_each_entry(page, &pgd_list, lru) {
995 if (!PagePinned(page)) {
996 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
997 SetPageSavePinned(page);
1001 spin_unlock_irqrestore(&pgd_lock, flags);
1005 * The init_mm pagetable is really pinned as soon as its created, but
1006 * that's before we have page structures to store the bits. So do all
1007 * the book-keeping now.
1009 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1010 enum pt_level level)
1012 SetPagePinned(page);
1016 static void __init xen_mark_init_mm_pinned(void)
1018 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1021 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1022 enum pt_level level)
1024 unsigned pgfl = TestClearPagePinned(page);
1026 if (pgfl && !PageHighMem(page)) {
1027 void *pt = lowmem_page_address(page);
1028 unsigned long pfn = page_to_pfn(page);
1029 spinlock_t *ptl = NULL;
1030 struct multicall_space mcs;
1033 * Do the converse to pin_page. If we're using split
1034 * pte locks, we must be holding the lock for while
1035 * the pte page is unpinned but still RO to prevent
1036 * concurrent updates from seeing it in this
1037 * partially-pinned state.
1039 if (level == PT_PTE) {
1040 ptl = xen_pte_lock(page, mm);
1043 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1046 mcs = __xen_mc_entry(0);
1048 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1049 pfn_pte(pfn, PAGE_KERNEL),
1050 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1053 /* unlock when batch completed */
1054 xen_mc_callback(xen_pte_unlock, ptl);
1058 return 0; /* never need to flush on unpin */
1061 /* Release a pagetables pages back as normal RW */
1062 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1066 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1068 #ifdef CONFIG_X86_64
1070 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1073 xen_do_pin(MMUEXT_UNPIN_TABLE,
1074 PFN_DOWN(__pa(user_pgd)));
1075 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1080 #ifdef CONFIG_X86_PAE
1081 /* Need to make sure unshared kernel PMD is unpinned */
1082 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1086 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1091 static void xen_pgd_unpin(struct mm_struct *mm)
1093 __xen_pgd_unpin(mm, mm->pgd);
1097 * On resume, undo any pinning done at save, so that the rest of the
1098 * kernel doesn't see any unexpected pinned pagetables.
1100 void xen_mm_unpin_all(void)
1102 unsigned long flags;
1105 spin_lock_irqsave(&pgd_lock, flags);
1107 list_for_each_entry(page, &pgd_list, lru) {
1108 if (PageSavePinned(page)) {
1109 BUG_ON(!PagePinned(page));
1110 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1111 ClearPageSavePinned(page);
1115 spin_unlock_irqrestore(&pgd_lock, flags);
1118 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1120 spin_lock(&next->page_table_lock);
1122 spin_unlock(&next->page_table_lock);
1125 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1127 spin_lock(&mm->page_table_lock);
1129 spin_unlock(&mm->page_table_lock);
1134 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1135 we need to repoint it somewhere else before we can unpin it. */
1136 static void drop_other_mm_ref(void *info)
1138 struct mm_struct *mm = info;
1139 struct mm_struct *active_mm;
1141 active_mm = percpu_read(cpu_tlbstate.active_mm);
1143 if (active_mm == mm)
1144 leave_mm(smp_processor_id());
1146 /* If this cpu still has a stale cr3 reference, then make sure
1147 it has been flushed. */
1148 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1149 load_cr3(swapper_pg_dir);
1152 static void xen_drop_mm_ref(struct mm_struct *mm)
1157 if (current->active_mm == mm) {
1158 if (current->mm == mm)
1159 load_cr3(swapper_pg_dir);
1161 leave_mm(smp_processor_id());
1164 /* Get the "official" set of cpus referring to our pagetable. */
1165 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1166 for_each_online_cpu(cpu) {
1167 if (!cpumask_test_cpu(cpu, &mm->cpu_vm_mask)
1168 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1170 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1174 cpumask_copy(mask, &mm->cpu_vm_mask);
1176 /* It's possible that a vcpu may have a stale reference to our
1177 cr3, because its in lazy mode, and it hasn't yet flushed
1178 its set of pending hypercalls yet. In this case, we can
1179 look at its actual current cr3 value, and force it to flush
1181 for_each_online_cpu(cpu) {
1182 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1183 cpumask_set_cpu(cpu, mask);
1186 if (!cpumask_empty(mask))
1187 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1188 free_cpumask_var(mask);
1191 static void xen_drop_mm_ref(struct mm_struct *mm)
1193 if (current->active_mm == mm)
1194 load_cr3(swapper_pg_dir);
1199 * While a process runs, Xen pins its pagetables, which means that the
1200 * hypervisor forces it to be read-only, and it controls all updates
1201 * to it. This means that all pagetable updates have to go via the
1202 * hypervisor, which is moderately expensive.
1204 * Since we're pulling the pagetable down, we switch to use init_mm,
1205 * unpin old process pagetable and mark it all read-write, which
1206 * allows further operations on it to be simple memory accesses.
1208 * The only subtle point is that another CPU may be still using the
1209 * pagetable because of lazy tlb flushing. This means we need need to
1210 * switch all CPUs off this pagetable before we can unpin it.
1212 void xen_exit_mmap(struct mm_struct *mm)
1214 get_cpu(); /* make sure we don't move around */
1215 xen_drop_mm_ref(mm);
1218 spin_lock(&mm->page_table_lock);
1220 /* pgd may not be pinned in the error exit path of execve */
1221 if (xen_page_pinned(mm->pgd))
1224 spin_unlock(&mm->page_table_lock);
1227 static __init void xen_pagetable_setup_start(pgd_t *base)
1231 static __init void xen_pagetable_setup_done(pgd_t *base)
1233 xen_setup_shared_info();
1236 static void xen_write_cr2(unsigned long cr2)
1238 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1241 static unsigned long xen_read_cr2(void)
1243 return percpu_read(xen_vcpu)->arch.cr2;
1246 unsigned long xen_read_cr2_direct(void)
1248 return percpu_read(xen_vcpu_info.arch.cr2);
1251 static void xen_flush_tlb(void)
1253 struct mmuext_op *op;
1254 struct multicall_space mcs;
1258 mcs = xen_mc_entry(sizeof(*op));
1261 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1262 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1264 xen_mc_issue(PARAVIRT_LAZY_MMU);
1269 static void xen_flush_tlb_single(unsigned long addr)
1271 struct mmuext_op *op;
1272 struct multicall_space mcs;
1276 mcs = xen_mc_entry(sizeof(*op));
1278 op->cmd = MMUEXT_INVLPG_LOCAL;
1279 op->arg1.linear_addr = addr & PAGE_MASK;
1280 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1282 xen_mc_issue(PARAVIRT_LAZY_MMU);
1287 static void xen_flush_tlb_others(const struct cpumask *cpus,
1288 struct mm_struct *mm, unsigned long va)
1291 struct mmuext_op op;
1292 DECLARE_BITMAP(mask, NR_CPUS);
1294 struct multicall_space mcs;
1296 if (cpumask_empty(cpus))
1297 return; /* nothing to do */
1299 mcs = xen_mc_entry(sizeof(*args));
1301 args->op.arg2.vcpumask = to_cpumask(args->mask);
1303 /* Remove us, and any offline CPUS. */
1304 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1305 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1307 if (va == TLB_FLUSH_ALL) {
1308 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1310 args->op.cmd = MMUEXT_INVLPG_MULTI;
1311 args->op.arg1.linear_addr = va;
1314 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1316 xen_mc_issue(PARAVIRT_LAZY_MMU);
1319 static unsigned long xen_read_cr3(void)
1321 return percpu_read(xen_cr3);
1324 static void set_current_cr3(void *v)
1326 percpu_write(xen_current_cr3, (unsigned long)v);
1329 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1331 struct mmuext_op *op;
1332 struct multicall_space mcs;
1336 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1340 WARN_ON(mfn == 0 && kernel);
1342 mcs = __xen_mc_entry(sizeof(*op));
1345 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1348 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1351 percpu_write(xen_cr3, cr3);
1353 /* Update xen_current_cr3 once the batch has actually
1355 xen_mc_callback(set_current_cr3, (void *)cr3);
1359 static void xen_write_cr3(unsigned long cr3)
1361 BUG_ON(preemptible());
1363 xen_mc_batch(); /* disables interrupts */
1365 /* Update while interrupts are disabled, so its atomic with
1367 percpu_write(xen_cr3, cr3);
1369 __xen_write_cr3(true, cr3);
1371 #ifdef CONFIG_X86_64
1373 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1375 __xen_write_cr3(false, __pa(user_pgd));
1377 __xen_write_cr3(false, 0);
1381 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1384 static int xen_pgd_alloc(struct mm_struct *mm)
1386 pgd_t *pgd = mm->pgd;
1389 BUG_ON(PagePinned(virt_to_page(pgd)));
1391 #ifdef CONFIG_X86_64
1393 struct page *page = virt_to_page(pgd);
1396 BUG_ON(page->private != 0);
1400 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1401 page->private = (unsigned long)user_pgd;
1403 if (user_pgd != NULL) {
1404 user_pgd[pgd_index(VSYSCALL_START)] =
1405 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1409 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1416 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1418 #ifdef CONFIG_X86_64
1419 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1422 free_page((unsigned long)user_pgd);
1426 #ifdef CONFIG_HIGHPTE
1427 static void *xen_kmap_atomic_pte(struct page *page, enum km_type type)
1429 pgprot_t prot = PAGE_KERNEL;
1431 if (PagePinned(page))
1432 prot = PAGE_KERNEL_RO;
1434 if (0 && PageHighMem(page))
1435 printk("mapping highpte %lx type %d prot %s\n",
1436 page_to_pfn(page), type,
1437 (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ");
1439 return kmap_atomic_prot(page, type, prot);
1443 #ifdef CONFIG_X86_32
1444 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1446 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1447 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1448 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1454 /* Init-time set_pte while constructing initial pagetables, which
1455 doesn't allow RO pagetable pages to be remapped RW */
1456 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1458 pte = mask_rw_pte(ptep, pte);
1460 xen_set_pte(ptep, pte);
1464 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1466 struct mmuext_op op;
1468 op.arg1.mfn = pfn_to_mfn(pfn);
1469 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1473 /* Early in boot, while setting up the initial pagetable, assume
1474 everything is pinned. */
1475 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1477 #ifdef CONFIG_FLATMEM
1478 BUG_ON(mem_map); /* should only be used early */
1480 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1481 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1484 /* Used for pmd and pud */
1485 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1487 #ifdef CONFIG_FLATMEM
1488 BUG_ON(mem_map); /* should only be used early */
1490 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1493 /* Early release_pte assumes that all pts are pinned, since there's
1494 only init_mm and anything attached to that is pinned. */
1495 static __init void xen_release_pte_init(unsigned long pfn)
1497 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1498 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1501 static __init void xen_release_pmd_init(unsigned long pfn)
1503 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1506 /* This needs to make sure the new pte page is pinned iff its being
1507 attached to a pinned pagetable. */
1508 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1510 struct page *page = pfn_to_page(pfn);
1512 if (PagePinned(virt_to_page(mm->pgd))) {
1513 SetPagePinned(page);
1516 if (!PageHighMem(page)) {
1517 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1518 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1519 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1521 /* make sure there are no stray mappings of
1523 kmap_flush_unused();
1528 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1530 xen_alloc_ptpage(mm, pfn, PT_PTE);
1533 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1535 xen_alloc_ptpage(mm, pfn, PT_PMD);
1538 /* This should never happen until we're OK to use struct page */
1539 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1541 struct page *page = pfn_to_page(pfn);
1543 if (PagePinned(page)) {
1544 if (!PageHighMem(page)) {
1545 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1546 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1547 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1549 ClearPagePinned(page);
1553 static void xen_release_pte(unsigned long pfn)
1555 xen_release_ptpage(pfn, PT_PTE);
1558 static void xen_release_pmd(unsigned long pfn)
1560 xen_release_ptpage(pfn, PT_PMD);
1563 #if PAGETABLE_LEVELS == 4
1564 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1566 xen_alloc_ptpage(mm, pfn, PT_PUD);
1569 static void xen_release_pud(unsigned long pfn)
1571 xen_release_ptpage(pfn, PT_PUD);
1575 void __init xen_reserve_top(void)
1577 #ifdef CONFIG_X86_32
1578 unsigned long top = HYPERVISOR_VIRT_START;
1579 struct xen_platform_parameters pp;
1581 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1582 top = pp.virt_start;
1584 reserve_top_address(-top);
1585 #endif /* CONFIG_X86_32 */
1589 * Like __va(), but returns address in the kernel mapping (which is
1590 * all we have until the physical memory mapping has been set up.
1592 static void *__ka(phys_addr_t paddr)
1594 #ifdef CONFIG_X86_64
1595 return (void *)(paddr + __START_KERNEL_map);
1601 /* Convert a machine address to physical address */
1602 static unsigned long m2p(phys_addr_t maddr)
1606 maddr &= PTE_PFN_MASK;
1607 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1612 /* Convert a machine address to kernel virtual */
1613 static void *m2v(phys_addr_t maddr)
1615 return __ka(m2p(maddr));
1618 static void set_page_prot(void *addr, pgprot_t prot)
1620 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1621 pte_t pte = pfn_pte(pfn, prot);
1623 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1627 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1629 unsigned pmdidx, pteidx;
1635 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1638 /* Reuse or allocate a page of ptes */
1639 if (pmd_present(pmd[pmdidx]))
1640 pte_page = m2v(pmd[pmdidx].pmd);
1642 /* Check for free pte pages */
1643 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1646 pte_page = &level1_ident_pgt[ident_pte];
1647 ident_pte += PTRS_PER_PTE;
1649 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1652 /* Install mappings */
1653 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1656 if (pfn > max_pfn_mapped)
1657 max_pfn_mapped = pfn;
1659 if (!pte_none(pte_page[pteidx]))
1662 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1663 pte_page[pteidx] = pte;
1667 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1668 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1670 set_page_prot(pmd, PAGE_KERNEL_RO);
1673 #ifdef CONFIG_X86_64
1674 static void convert_pfn_mfn(void *v)
1679 /* All levels are converted the same way, so just treat them
1681 for (i = 0; i < PTRS_PER_PTE; i++)
1682 pte[i] = xen_make_pte(pte[i].pte);
1686 * Set up the inital kernel pagetable.
1688 * We can construct this by grafting the Xen provided pagetable into
1689 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1690 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1691 * means that only the kernel has a physical mapping to start with -
1692 * but that's enough to get __va working. We need to fill in the rest
1693 * of the physical mapping once some sort of allocator has been set
1696 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1697 unsigned long max_pfn)
1702 /* Zap identity mapping */
1703 init_level4_pgt[0] = __pgd(0);
1705 /* Pre-constructed entries are in pfn, so convert to mfn */
1706 convert_pfn_mfn(init_level4_pgt);
1707 convert_pfn_mfn(level3_ident_pgt);
1708 convert_pfn_mfn(level3_kernel_pgt);
1710 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1711 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1713 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1714 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1716 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1717 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1718 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1720 /* Set up identity map */
1721 xen_map_identity_early(level2_ident_pgt, max_pfn);
1723 /* Make pagetable pieces RO */
1724 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1725 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1726 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1727 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1728 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1729 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1731 /* Pin down new L4 */
1732 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1733 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1735 /* Unpin Xen-provided one */
1736 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1739 pgd = init_level4_pgt;
1742 * At this stage there can be no user pgd, and no page
1743 * structure to attach it to, so make sure we just set kernel
1747 __xen_write_cr3(true, __pa(pgd));
1748 xen_mc_issue(PARAVIRT_LAZY_CPU);
1750 reserve_early(__pa(xen_start_info->pt_base),
1751 __pa(xen_start_info->pt_base +
1752 xen_start_info->nr_pt_frames * PAGE_SIZE),
1757 #else /* !CONFIG_X86_64 */
1758 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1760 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1761 unsigned long max_pfn)
1765 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1766 xen_start_info->nr_pt_frames * PAGE_SIZE +
1769 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1770 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1772 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1774 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1775 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1776 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1778 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1779 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1780 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1782 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1784 xen_write_cr3(__pa(swapper_pg_dir));
1786 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1788 reserve_early(__pa(xen_start_info->pt_base),
1789 __pa(xen_start_info->pt_base +
1790 xen_start_info->nr_pt_frames * PAGE_SIZE),
1793 return swapper_pg_dir;
1795 #endif /* CONFIG_X86_64 */
1797 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1801 phys >>= PAGE_SHIFT;
1804 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1805 #ifdef CONFIG_X86_F00F_BUG
1808 #ifdef CONFIG_X86_32
1811 # ifdef CONFIG_HIGHMEM
1812 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1815 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1817 #ifdef CONFIG_X86_LOCAL_APIC
1818 case FIX_APIC_BASE: /* maps dummy local APIC */
1820 case FIX_TEXT_POKE0:
1821 case FIX_TEXT_POKE1:
1822 /* All local page mappings */
1823 pte = pfn_pte(phys, prot);
1827 pte = mfn_pte(phys, prot);
1831 __native_set_fixmap(idx, pte);
1833 #ifdef CONFIG_X86_64
1834 /* Replicate changes to map the vsyscall page into the user
1835 pagetable vsyscall mapping. */
1836 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1837 unsigned long vaddr = __fix_to_virt(idx);
1838 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1843 __init void xen_post_allocator_init(void)
1845 pv_mmu_ops.set_pte = xen_set_pte;
1846 pv_mmu_ops.set_pmd = xen_set_pmd;
1847 pv_mmu_ops.set_pud = xen_set_pud;
1848 #if PAGETABLE_LEVELS == 4
1849 pv_mmu_ops.set_pgd = xen_set_pgd;
1852 /* This will work as long as patching hasn't happened yet
1853 (which it hasn't) */
1854 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1855 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1856 pv_mmu_ops.release_pte = xen_release_pte;
1857 pv_mmu_ops.release_pmd = xen_release_pmd;
1858 #if PAGETABLE_LEVELS == 4
1859 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1860 pv_mmu_ops.release_pud = xen_release_pud;
1863 #ifdef CONFIG_X86_64
1864 SetPagePinned(virt_to_page(level3_user_vsyscall));
1866 xen_mark_init_mm_pinned();
1869 static void xen_leave_lazy_mmu(void)
1873 paravirt_leave_lazy_mmu();
1877 const struct pv_mmu_ops xen_mmu_ops __initdata = {
1878 .pagetable_setup_start = xen_pagetable_setup_start,
1879 .pagetable_setup_done = xen_pagetable_setup_done,
1881 .read_cr2 = xen_read_cr2,
1882 .write_cr2 = xen_write_cr2,
1884 .read_cr3 = xen_read_cr3,
1885 .write_cr3 = xen_write_cr3,
1887 .flush_tlb_user = xen_flush_tlb,
1888 .flush_tlb_kernel = xen_flush_tlb,
1889 .flush_tlb_single = xen_flush_tlb_single,
1890 .flush_tlb_others = xen_flush_tlb_others,
1892 .pte_update = paravirt_nop,
1893 .pte_update_defer = paravirt_nop,
1895 .pgd_alloc = xen_pgd_alloc,
1896 .pgd_free = xen_pgd_free,
1898 .alloc_pte = xen_alloc_pte_init,
1899 .release_pte = xen_release_pte_init,
1900 .alloc_pmd = xen_alloc_pmd_init,
1901 .alloc_pmd_clone = paravirt_nop,
1902 .release_pmd = xen_release_pmd_init,
1904 #ifdef CONFIG_HIGHPTE
1905 .kmap_atomic_pte = xen_kmap_atomic_pte,
1908 #ifdef CONFIG_X86_64
1909 .set_pte = xen_set_pte,
1911 .set_pte = xen_set_pte_init,
1913 .set_pte_at = xen_set_pte_at,
1914 .set_pmd = xen_set_pmd_hyper,
1916 .ptep_modify_prot_start = __ptep_modify_prot_start,
1917 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1919 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1920 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1922 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1923 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1925 #ifdef CONFIG_X86_PAE
1926 .set_pte_atomic = xen_set_pte_atomic,
1927 .pte_clear = xen_pte_clear,
1928 .pmd_clear = xen_pmd_clear,
1929 #endif /* CONFIG_X86_PAE */
1930 .set_pud = xen_set_pud_hyper,
1932 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
1933 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
1935 #if PAGETABLE_LEVELS == 4
1936 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
1937 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
1938 .set_pgd = xen_set_pgd_hyper,
1940 .alloc_pud = xen_alloc_pmd_init,
1941 .release_pud = xen_release_pmd_init,
1942 #endif /* PAGETABLE_LEVELS == 4 */
1944 .activate_mm = xen_activate_mm,
1945 .dup_mmap = xen_dup_mmap,
1946 .exit_mmap = xen_exit_mmap,
1949 .enter = paravirt_enter_lazy_mmu,
1950 .leave = xen_leave_lazy_mmu,
1953 .set_fixmap = xen_set_fixmap,
1957 #ifdef CONFIG_XEN_DEBUG_FS
1959 static struct dentry *d_mmu_debug;
1961 static int __init xen_mmu_debugfs(void)
1963 struct dentry *d_xen = xen_init_debugfs();
1968 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1970 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
1972 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
1973 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
1974 &mmu_stats.pgd_update_pinned);
1975 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
1976 &mmu_stats.pgd_update_pinned);
1978 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
1979 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
1980 &mmu_stats.pud_update_pinned);
1981 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
1982 &mmu_stats.pud_update_pinned);
1984 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
1985 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
1986 &mmu_stats.pmd_update_pinned);
1987 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
1988 &mmu_stats.pmd_update_pinned);
1990 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
1991 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
1992 // &mmu_stats.pte_update_pinned);
1993 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
1994 &mmu_stats.pte_update_pinned);
1996 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
1997 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
1998 &mmu_stats.mmu_update_extended);
1999 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2000 mmu_stats.mmu_update_histo, 20);
2002 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2003 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2004 &mmu_stats.set_pte_at_batched);
2005 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2006 &mmu_stats.set_pte_at_current);
2007 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2008 &mmu_stats.set_pte_at_kernel);
2010 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2011 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2012 &mmu_stats.prot_commit_batched);
2016 fs_initcall(xen_mmu_debugfs);
2018 #endif /* CONFIG_XEN_DEBUG_FS */