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>
46 #include <linux/gfp.h>
48 #include <asm/pgtable.h>
49 #include <asm/tlbflush.h>
50 #include <asm/fixmap.h>
51 #include <asm/mmu_context.h>
52 #include <asm/setup.h>
53 #include <asm/paravirt.h>
54 #include <asm/linkage.h>
56 #include <asm/xen/hypercall.h>
57 #include <asm/xen/hypervisor.h>
61 #include <xen/interface/xen.h>
62 #include <xen/interface/version.h>
63 #include <xen/interface/memory.h>
64 #include <xen/hvc-console.h>
66 #include "multicalls.h"
70 #define MMU_UPDATE_HISTO 30
72 #ifdef CONFIG_XEN_DEBUG_FS
76 u32 pgd_update_pinned;
77 u32 pgd_update_batched;
80 u32 pud_update_pinned;
81 u32 pud_update_batched;
84 u32 pmd_update_pinned;
85 u32 pmd_update_batched;
88 u32 pte_update_pinned;
89 u32 pte_update_batched;
92 u32 mmu_update_extended;
93 u32 mmu_update_histo[MMU_UPDATE_HISTO];
96 u32 prot_commit_batched;
99 u32 set_pte_at_batched;
100 u32 set_pte_at_pinned;
101 u32 set_pte_at_current;
102 u32 set_pte_at_kernel;
105 static u8 zero_stats;
107 static inline void check_zero(void)
109 if (unlikely(zero_stats)) {
110 memset(&mmu_stats, 0, sizeof(mmu_stats));
115 #define ADD_STATS(elem, val) \
116 do { check_zero(); mmu_stats.elem += (val); } while(0)
118 #else /* !CONFIG_XEN_DEBUG_FS */
120 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
122 #endif /* CONFIG_XEN_DEBUG_FS */
126 * Identity map, in addition to plain kernel map. This needs to be
127 * large enough to allocate page table pages to allocate the rest.
128 * Each page can map 2MB.
130 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
133 /* l3 pud for userspace vsyscall mapping */
134 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
135 #endif /* CONFIG_X86_64 */
138 * Note about cr3 (pagetable base) values:
140 * xen_cr3 contains the current logical cr3 value; it contains the
141 * last set cr3. This may not be the current effective cr3, because
142 * its update may be being lazily deferred. However, a vcpu looking
143 * at its own cr3 can use this value knowing that it everything will
144 * be self-consistent.
146 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
147 * hypercall to set the vcpu cr3 is complete (so it may be a little
148 * out of date, but it will never be set early). If one vcpu is
149 * looking at another vcpu's cr3 value, it should use this variable.
151 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
152 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
156 * Just beyond the highest usermode address. STACK_TOP_MAX has a
157 * redzone above it, so round it up to a PGD boundary.
159 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
162 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
163 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
165 /* Placeholder for holes in the address space */
166 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
167 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
169 /* Array of pointers to pages containing p2m entries */
170 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
171 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
173 /* Arrays of p2m arrays expressed in mfns used for save/restore */
174 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
176 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
179 static inline unsigned p2m_top_index(unsigned long pfn)
181 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
182 return pfn / P2M_ENTRIES_PER_PAGE;
185 static inline unsigned p2m_index(unsigned long pfn)
187 return pfn % P2M_ENTRIES_PER_PAGE;
190 /* Build the parallel p2m_top_mfn structures */
191 void xen_build_mfn_list_list(void)
195 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
196 unsigned topidx = p2m_top_index(pfn);
198 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
201 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
202 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
203 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
207 void xen_setup_mfn_list_list(void)
209 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
211 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
212 virt_to_mfn(p2m_top_mfn_list);
213 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
216 /* Set up p2m_top to point to the domain-builder provided p2m pages */
217 void __init xen_build_dynamic_phys_to_machine(void)
219 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
220 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
223 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
224 unsigned topidx = p2m_top_index(pfn);
226 p2m_top[topidx] = &mfn_list[pfn];
229 xen_build_mfn_list_list();
232 unsigned long get_phys_to_machine(unsigned long pfn)
234 unsigned topidx, idx;
236 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
237 return INVALID_P2M_ENTRY;
239 topidx = p2m_top_index(pfn);
240 idx = p2m_index(pfn);
241 return p2m_top[topidx][idx];
243 EXPORT_SYMBOL_GPL(get_phys_to_machine);
245 /* install a new p2m_top page */
246 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
248 unsigned topidx = p2m_top_index(pfn);
249 unsigned long **pfnp, *mfnp;
252 pfnp = &p2m_top[topidx];
253 mfnp = &p2m_top_mfn[topidx];
255 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
256 p[i] = INVALID_P2M_ENTRY;
258 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
259 *mfnp = virt_to_mfn(p);
266 static void alloc_p2m(unsigned long pfn)
270 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
273 if (!install_p2mtop_page(pfn, p))
274 free_page((unsigned long)p);
277 /* Try to install p2m mapping; fail if intermediate bits missing */
278 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
280 unsigned topidx, idx;
282 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
283 BUG_ON(mfn != INVALID_P2M_ENTRY);
287 topidx = p2m_top_index(pfn);
288 if (p2m_top[topidx] == p2m_missing) {
289 if (mfn == INVALID_P2M_ENTRY)
294 idx = p2m_index(pfn);
295 p2m_top[topidx][idx] = mfn;
300 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
302 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
303 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
307 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
310 if (!__set_phys_to_machine(pfn, mfn))
315 unsigned long arbitrary_virt_to_mfn(void *vaddr)
317 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
319 return PFN_DOWN(maddr.maddr);
322 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
324 unsigned long address = (unsigned long)vaddr;
330 * if the PFN is in the linear mapped vaddr range, we can just use
331 * the (quick) virt_to_machine() p2m lookup
333 if (virt_addr_valid(vaddr))
334 return virt_to_machine(vaddr);
336 /* otherwise we have to do a (slower) full page-table walk */
338 pte = lookup_address(address, &level);
340 offset = address & ~PAGE_MASK;
341 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
344 void make_lowmem_page_readonly(void *vaddr)
347 unsigned long address = (unsigned long)vaddr;
350 pte = lookup_address(address, &level);
353 ptev = pte_wrprotect(*pte);
355 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
359 void make_lowmem_page_readwrite(void *vaddr)
362 unsigned long address = (unsigned long)vaddr;
365 pte = lookup_address(address, &level);
368 ptev = pte_mkwrite(*pte);
370 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
375 static bool xen_page_pinned(void *ptr)
377 struct page *page = virt_to_page(ptr);
379 return PagePinned(page);
382 static bool xen_iomap_pte(pte_t pte)
384 return xen_initial_domain() && (pte_flags(pte) & _PAGE_IOMAP);
387 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
389 struct multicall_space mcs;
390 struct mmu_update *u;
392 mcs = xen_mc_entry(sizeof(*u));
395 /* ptep might be kmapped when using 32-bit HIGHPTE */
396 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
397 u->val = pte_val_ma(pteval);
399 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_IO);
401 xen_mc_issue(PARAVIRT_LAZY_MMU);
404 static void xen_extend_mmu_update(const struct mmu_update *update)
406 struct multicall_space mcs;
407 struct mmu_update *u;
409 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
411 if (mcs.mc != NULL) {
412 ADD_STATS(mmu_update_extended, 1);
413 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
417 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
418 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
420 ADD_STATS(mmu_update_histo[0], 1);
422 ADD_STATS(mmu_update, 1);
423 mcs = __xen_mc_entry(sizeof(*u));
424 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
425 ADD_STATS(mmu_update_histo[1], 1);
432 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
440 /* ptr may be ioremapped for 64-bit pagetable setup */
441 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
442 u.val = pmd_val_ma(val);
443 xen_extend_mmu_update(&u);
445 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
447 xen_mc_issue(PARAVIRT_LAZY_MMU);
452 void xen_set_pmd(pmd_t *ptr, pmd_t val)
454 ADD_STATS(pmd_update, 1);
456 /* If page is not pinned, we can just update the entry
458 if (!xen_page_pinned(ptr)) {
463 ADD_STATS(pmd_update_pinned, 1);
465 xen_set_pmd_hyper(ptr, val);
469 * Associate a virtual page frame with a given physical page frame
470 * and protection flags for that frame.
472 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
474 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
477 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
478 pte_t *ptep, pte_t pteval)
480 if (xen_iomap_pte(pteval)) {
481 xen_set_iomap_pte(ptep, pteval);
485 ADD_STATS(set_pte_at, 1);
486 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
487 ADD_STATS(set_pte_at_current, mm == current->mm);
488 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
490 if (mm == current->mm || mm == &init_mm) {
491 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
492 struct multicall_space mcs;
493 mcs = xen_mc_entry(0);
495 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
496 ADD_STATS(set_pte_at_batched, 1);
497 xen_mc_issue(PARAVIRT_LAZY_MMU);
500 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
503 xen_set_pte(ptep, pteval);
508 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
509 unsigned long addr, pte_t *ptep)
511 /* Just return the pte as-is. We preserve the bits on commit */
515 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
516 pte_t *ptep, pte_t pte)
522 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
523 u.val = pte_val_ma(pte);
524 xen_extend_mmu_update(&u);
526 ADD_STATS(prot_commit, 1);
527 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
529 xen_mc_issue(PARAVIRT_LAZY_MMU);
532 /* Assume pteval_t is equivalent to all the other *val_t types. */
533 static pteval_t pte_mfn_to_pfn(pteval_t val)
535 if (val & _PAGE_PRESENT) {
536 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
537 pteval_t flags = val & PTE_FLAGS_MASK;
538 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
544 static pteval_t pte_pfn_to_mfn(pteval_t val)
546 if (val & _PAGE_PRESENT) {
547 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
548 pteval_t flags = val & PTE_FLAGS_MASK;
549 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
555 static pteval_t iomap_pte(pteval_t val)
557 if (val & _PAGE_PRESENT) {
558 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
559 pteval_t flags = val & PTE_FLAGS_MASK;
561 /* We assume the pte frame number is a MFN, so
562 just use it as-is. */
563 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
569 pteval_t xen_pte_val(pte_t pte)
571 if (xen_initial_domain() && (pte.pte & _PAGE_IOMAP))
574 return pte_mfn_to_pfn(pte.pte);
576 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
578 pgdval_t xen_pgd_val(pgd_t pgd)
580 return pte_mfn_to_pfn(pgd.pgd);
582 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
584 pte_t xen_make_pte(pteval_t pte)
586 if (unlikely(xen_initial_domain() && (pte & _PAGE_IOMAP)))
587 pte = iomap_pte(pte);
589 pte = pte_pfn_to_mfn(pte);
591 return native_make_pte(pte);
593 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
595 pgd_t xen_make_pgd(pgdval_t pgd)
597 pgd = pte_pfn_to_mfn(pgd);
598 return native_make_pgd(pgd);
600 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
602 pmdval_t xen_pmd_val(pmd_t pmd)
604 return pte_mfn_to_pfn(pmd.pmd);
606 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
608 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
616 /* ptr may be ioremapped for 64-bit pagetable setup */
617 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
618 u.val = pud_val_ma(val);
619 xen_extend_mmu_update(&u);
621 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
623 xen_mc_issue(PARAVIRT_LAZY_MMU);
628 void xen_set_pud(pud_t *ptr, pud_t val)
630 ADD_STATS(pud_update, 1);
632 /* If page is not pinned, we can just update the entry
634 if (!xen_page_pinned(ptr)) {
639 ADD_STATS(pud_update_pinned, 1);
641 xen_set_pud_hyper(ptr, val);
644 void xen_set_pte(pte_t *ptep, pte_t pte)
646 if (xen_iomap_pte(pte)) {
647 xen_set_iomap_pte(ptep, pte);
651 ADD_STATS(pte_update, 1);
652 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
653 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
655 #ifdef CONFIG_X86_PAE
656 ptep->pte_high = pte.pte_high;
658 ptep->pte_low = pte.pte_low;
664 #ifdef CONFIG_X86_PAE
665 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
667 if (xen_iomap_pte(pte)) {
668 xen_set_iomap_pte(ptep, pte);
672 set_64bit((u64 *)ptep, native_pte_val(pte));
675 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
678 smp_wmb(); /* make sure low gets written first */
682 void xen_pmd_clear(pmd_t *pmdp)
684 set_pmd(pmdp, __pmd(0));
686 #endif /* CONFIG_X86_PAE */
688 pmd_t xen_make_pmd(pmdval_t pmd)
690 pmd = pte_pfn_to_mfn(pmd);
691 return native_make_pmd(pmd);
693 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
695 #if PAGETABLE_LEVELS == 4
696 pudval_t xen_pud_val(pud_t pud)
698 return pte_mfn_to_pfn(pud.pud);
700 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
702 pud_t xen_make_pud(pudval_t pud)
704 pud = pte_pfn_to_mfn(pud);
706 return native_make_pud(pud);
708 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
710 pgd_t *xen_get_user_pgd(pgd_t *pgd)
712 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
713 unsigned offset = pgd - pgd_page;
714 pgd_t *user_ptr = NULL;
716 if (offset < pgd_index(USER_LIMIT)) {
717 struct page *page = virt_to_page(pgd_page);
718 user_ptr = (pgd_t *)page->private;
726 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
730 u.ptr = virt_to_machine(ptr).maddr;
731 u.val = pgd_val_ma(val);
732 xen_extend_mmu_update(&u);
736 * Raw hypercall-based set_pgd, intended for in early boot before
737 * there's a page structure. This implies:
738 * 1. The only existing pagetable is the kernel's
739 * 2. It is always pinned
740 * 3. It has no user pagetable attached to it
742 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
748 __xen_set_pgd_hyper(ptr, val);
750 xen_mc_issue(PARAVIRT_LAZY_MMU);
755 void xen_set_pgd(pgd_t *ptr, pgd_t val)
757 pgd_t *user_ptr = xen_get_user_pgd(ptr);
759 ADD_STATS(pgd_update, 1);
761 /* If page is not pinned, we can just update the entry
763 if (!xen_page_pinned(ptr)) {
766 WARN_ON(xen_page_pinned(user_ptr));
772 ADD_STATS(pgd_update_pinned, 1);
773 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
775 /* If it's pinned, then we can at least batch the kernel and
776 user updates together. */
779 __xen_set_pgd_hyper(ptr, val);
781 __xen_set_pgd_hyper(user_ptr, val);
783 xen_mc_issue(PARAVIRT_LAZY_MMU);
785 #endif /* PAGETABLE_LEVELS == 4 */
788 * (Yet another) pagetable walker. This one is intended for pinning a
789 * pagetable. This means that it walks a pagetable and calls the
790 * callback function on each page it finds making up the page table,
791 * at every level. It walks the entire pagetable, but it only bothers
792 * pinning pte pages which are below limit. In the normal case this
793 * will be STACK_TOP_MAX, but at boot we need to pin up to
796 * For 32-bit the important bit is that we don't pin beyond there,
797 * because then we start getting into Xen's ptes.
799 * For 64-bit, we must skip the Xen hole in the middle of the address
800 * space, just after the big x86-64 virtual hole.
802 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
803 int (*func)(struct mm_struct *mm, struct page *,
808 unsigned hole_low, hole_high;
809 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
810 unsigned pgdidx, pudidx, pmdidx;
812 /* The limit is the last byte to be touched */
814 BUG_ON(limit >= FIXADDR_TOP);
816 if (xen_feature(XENFEAT_auto_translated_physmap))
820 * 64-bit has a great big hole in the middle of the address
821 * space, which contains the Xen mappings. On 32-bit these
822 * will end up making a zero-sized hole and so is a no-op.
824 hole_low = pgd_index(USER_LIMIT);
825 hole_high = pgd_index(PAGE_OFFSET);
827 pgdidx_limit = pgd_index(limit);
829 pudidx_limit = pud_index(limit);
834 pmdidx_limit = pmd_index(limit);
839 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
842 if (pgdidx >= hole_low && pgdidx < hole_high)
845 if (!pgd_val(pgd[pgdidx]))
848 pud = pud_offset(&pgd[pgdidx], 0);
850 if (PTRS_PER_PUD > 1) /* not folded */
851 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
853 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
856 if (pgdidx == pgdidx_limit &&
857 pudidx > pudidx_limit)
860 if (pud_none(pud[pudidx]))
863 pmd = pmd_offset(&pud[pudidx], 0);
865 if (PTRS_PER_PMD > 1) /* not folded */
866 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
868 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
871 if (pgdidx == pgdidx_limit &&
872 pudidx == pudidx_limit &&
873 pmdidx > pmdidx_limit)
876 if (pmd_none(pmd[pmdidx]))
879 pte = pmd_page(pmd[pmdidx]);
880 flush |= (*func)(mm, pte, PT_PTE);
886 /* Do the top level last, so that the callbacks can use it as
887 a cue to do final things like tlb flushes. */
888 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
893 static int xen_pgd_walk(struct mm_struct *mm,
894 int (*func)(struct mm_struct *mm, struct page *,
898 return __xen_pgd_walk(mm, mm->pgd, func, limit);
901 /* If we're using split pte locks, then take the page's lock and
902 return a pointer to it. Otherwise return NULL. */
903 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
905 spinlock_t *ptl = NULL;
907 #if USE_SPLIT_PTLOCKS
908 ptl = __pte_lockptr(page);
909 spin_lock_nest_lock(ptl, &mm->page_table_lock);
915 static void xen_pte_unlock(void *v)
921 static void xen_do_pin(unsigned level, unsigned long pfn)
923 struct mmuext_op *op;
924 struct multicall_space mcs;
926 mcs = __xen_mc_entry(sizeof(*op));
929 op->arg1.mfn = pfn_to_mfn(pfn);
930 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
933 static int xen_pin_page(struct mm_struct *mm, struct page *page,
936 unsigned pgfl = TestSetPagePinned(page);
940 flush = 0; /* already pinned */
941 else if (PageHighMem(page))
942 /* kmaps need flushing if we found an unpinned
946 void *pt = lowmem_page_address(page);
947 unsigned long pfn = page_to_pfn(page);
948 struct multicall_space mcs = __xen_mc_entry(0);
954 * We need to hold the pagetable lock between the time
955 * we make the pagetable RO and when we actually pin
956 * it. If we don't, then other users may come in and
957 * attempt to update the pagetable by writing it,
958 * which will fail because the memory is RO but not
959 * pinned, so Xen won't do the trap'n'emulate.
961 * If we're using split pte locks, we can't hold the
962 * entire pagetable's worth of locks during the
963 * traverse, because we may wrap the preempt count (8
964 * bits). The solution is to mark RO and pin each PTE
965 * page while holding the lock. This means the number
966 * of locks we end up holding is never more than a
967 * batch size (~32 entries, at present).
969 * If we're not using split pte locks, we needn't pin
970 * the PTE pages independently, because we're
971 * protected by the overall pagetable lock.
975 ptl = xen_pte_lock(page, mm);
977 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
978 pfn_pte(pfn, PAGE_KERNEL_RO),
979 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
982 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
984 /* Queue a deferred unlock for when this batch
986 xen_mc_callback(xen_pte_unlock, ptl);
993 /* This is called just after a mm has been created, but it has not
994 been used yet. We need to make sure that its pagetable is all
995 read-only, and can be pinned. */
996 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
1002 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
1003 /* re-enable interrupts for flushing */
1006 kmap_flush_unused();
1011 #ifdef CONFIG_X86_64
1013 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1015 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
1018 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
1019 xen_do_pin(MMUEXT_PIN_L4_TABLE,
1020 PFN_DOWN(__pa(user_pgd)));
1023 #else /* CONFIG_X86_32 */
1024 #ifdef CONFIG_X86_PAE
1025 /* Need to make sure unshared kernel PMD is pinnable */
1026 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1029 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
1030 #endif /* CONFIG_X86_64 */
1034 static void xen_pgd_pin(struct mm_struct *mm)
1036 __xen_pgd_pin(mm, mm->pgd);
1040 * On save, we need to pin all pagetables to make sure they get their
1041 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1042 * them (unpinned pgds are not currently in use, probably because the
1043 * process is under construction or destruction).
1045 * Expected to be called in stop_machine() ("equivalent to taking
1046 * every spinlock in the system"), so the locking doesn't really
1047 * matter all that much.
1049 void xen_mm_pin_all(void)
1051 unsigned long flags;
1054 spin_lock_irqsave(&pgd_lock, flags);
1056 list_for_each_entry(page, &pgd_list, lru) {
1057 if (!PagePinned(page)) {
1058 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1059 SetPageSavePinned(page);
1063 spin_unlock_irqrestore(&pgd_lock, flags);
1067 * The init_mm pagetable is really pinned as soon as its created, but
1068 * that's before we have page structures to store the bits. So do all
1069 * the book-keeping now.
1071 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1072 enum pt_level level)
1074 SetPagePinned(page);
1078 static void __init xen_mark_init_mm_pinned(void)
1080 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1083 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1084 enum pt_level level)
1086 unsigned pgfl = TestClearPagePinned(page);
1088 if (pgfl && !PageHighMem(page)) {
1089 void *pt = lowmem_page_address(page);
1090 unsigned long pfn = page_to_pfn(page);
1091 spinlock_t *ptl = NULL;
1092 struct multicall_space mcs;
1095 * Do the converse to pin_page. If we're using split
1096 * pte locks, we must be holding the lock for while
1097 * the pte page is unpinned but still RO to prevent
1098 * concurrent updates from seeing it in this
1099 * partially-pinned state.
1101 if (level == PT_PTE) {
1102 ptl = xen_pte_lock(page, mm);
1105 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1108 mcs = __xen_mc_entry(0);
1110 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1111 pfn_pte(pfn, PAGE_KERNEL),
1112 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1115 /* unlock when batch completed */
1116 xen_mc_callback(xen_pte_unlock, ptl);
1120 return 0; /* never need to flush on unpin */
1123 /* Release a pagetables pages back as normal RW */
1124 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1128 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1130 #ifdef CONFIG_X86_64
1132 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1135 xen_do_pin(MMUEXT_UNPIN_TABLE,
1136 PFN_DOWN(__pa(user_pgd)));
1137 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1142 #ifdef CONFIG_X86_PAE
1143 /* Need to make sure unshared kernel PMD is unpinned */
1144 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1148 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1153 static void xen_pgd_unpin(struct mm_struct *mm)
1155 __xen_pgd_unpin(mm, mm->pgd);
1159 * On resume, undo any pinning done at save, so that the rest of the
1160 * kernel doesn't see any unexpected pinned pagetables.
1162 void xen_mm_unpin_all(void)
1164 unsigned long flags;
1167 spin_lock_irqsave(&pgd_lock, flags);
1169 list_for_each_entry(page, &pgd_list, lru) {
1170 if (PageSavePinned(page)) {
1171 BUG_ON(!PagePinned(page));
1172 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1173 ClearPageSavePinned(page);
1177 spin_unlock_irqrestore(&pgd_lock, flags);
1180 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1182 spin_lock(&next->page_table_lock);
1184 spin_unlock(&next->page_table_lock);
1187 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1189 spin_lock(&mm->page_table_lock);
1191 spin_unlock(&mm->page_table_lock);
1196 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1197 we need to repoint it somewhere else before we can unpin it. */
1198 static void drop_other_mm_ref(void *info)
1200 struct mm_struct *mm = info;
1201 struct mm_struct *active_mm;
1203 active_mm = percpu_read(cpu_tlbstate.active_mm);
1205 if (active_mm == mm)
1206 leave_mm(smp_processor_id());
1208 /* If this cpu still has a stale cr3 reference, then make sure
1209 it has been flushed. */
1210 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1211 load_cr3(swapper_pg_dir);
1214 static void xen_drop_mm_ref(struct mm_struct *mm)
1219 if (current->active_mm == mm) {
1220 if (current->mm == mm)
1221 load_cr3(swapper_pg_dir);
1223 leave_mm(smp_processor_id());
1226 /* Get the "official" set of cpus referring to our pagetable. */
1227 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1228 for_each_online_cpu(cpu) {
1229 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1230 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1232 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1236 cpumask_copy(mask, mm_cpumask(mm));
1238 /* It's possible that a vcpu may have a stale reference to our
1239 cr3, because its in lazy mode, and it hasn't yet flushed
1240 its set of pending hypercalls yet. In this case, we can
1241 look at its actual current cr3 value, and force it to flush
1243 for_each_online_cpu(cpu) {
1244 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1245 cpumask_set_cpu(cpu, mask);
1248 if (!cpumask_empty(mask))
1249 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1250 free_cpumask_var(mask);
1253 static void xen_drop_mm_ref(struct mm_struct *mm)
1255 if (current->active_mm == mm)
1256 load_cr3(swapper_pg_dir);
1261 * While a process runs, Xen pins its pagetables, which means that the
1262 * hypervisor forces it to be read-only, and it controls all updates
1263 * to it. This means that all pagetable updates have to go via the
1264 * hypervisor, which is moderately expensive.
1266 * Since we're pulling the pagetable down, we switch to use init_mm,
1267 * unpin old process pagetable and mark it all read-write, which
1268 * allows further operations on it to be simple memory accesses.
1270 * The only subtle point is that another CPU may be still using the
1271 * pagetable because of lazy tlb flushing. This means we need need to
1272 * switch all CPUs off this pagetable before we can unpin it.
1274 void xen_exit_mmap(struct mm_struct *mm)
1276 get_cpu(); /* make sure we don't move around */
1277 xen_drop_mm_ref(mm);
1280 spin_lock(&mm->page_table_lock);
1282 /* pgd may not be pinned in the error exit path of execve */
1283 if (xen_page_pinned(mm->pgd))
1286 spin_unlock(&mm->page_table_lock);
1289 static __init void xen_pagetable_setup_start(pgd_t *base)
1293 static void xen_post_allocator_init(void);
1295 static __init void xen_pagetable_setup_done(pgd_t *base)
1297 xen_setup_shared_info();
1298 xen_post_allocator_init();
1301 static void xen_write_cr2(unsigned long cr2)
1303 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1306 static unsigned long xen_read_cr2(void)
1308 return percpu_read(xen_vcpu)->arch.cr2;
1311 unsigned long xen_read_cr2_direct(void)
1313 return percpu_read(xen_vcpu_info.arch.cr2);
1316 static void xen_flush_tlb(void)
1318 struct mmuext_op *op;
1319 struct multicall_space mcs;
1323 mcs = xen_mc_entry(sizeof(*op));
1326 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1327 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1329 xen_mc_issue(PARAVIRT_LAZY_MMU);
1334 static void xen_flush_tlb_single(unsigned long addr)
1336 struct mmuext_op *op;
1337 struct multicall_space mcs;
1341 mcs = xen_mc_entry(sizeof(*op));
1343 op->cmd = MMUEXT_INVLPG_LOCAL;
1344 op->arg1.linear_addr = addr & PAGE_MASK;
1345 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1347 xen_mc_issue(PARAVIRT_LAZY_MMU);
1352 static void xen_flush_tlb_others(const struct cpumask *cpus,
1353 struct mm_struct *mm, unsigned long va)
1356 struct mmuext_op op;
1357 DECLARE_BITMAP(mask, NR_CPUS);
1359 struct multicall_space mcs;
1361 if (cpumask_empty(cpus))
1362 return; /* nothing to do */
1364 mcs = xen_mc_entry(sizeof(*args));
1366 args->op.arg2.vcpumask = to_cpumask(args->mask);
1368 /* Remove us, and any offline CPUS. */
1369 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1370 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1372 if (va == TLB_FLUSH_ALL) {
1373 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1375 args->op.cmd = MMUEXT_INVLPG_MULTI;
1376 args->op.arg1.linear_addr = va;
1379 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1381 xen_mc_issue(PARAVIRT_LAZY_MMU);
1384 static unsigned long xen_read_cr3(void)
1386 return percpu_read(xen_cr3);
1389 static void set_current_cr3(void *v)
1391 percpu_write(xen_current_cr3, (unsigned long)v);
1394 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1396 struct mmuext_op *op;
1397 struct multicall_space mcs;
1401 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1405 WARN_ON(mfn == 0 && kernel);
1407 mcs = __xen_mc_entry(sizeof(*op));
1410 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1413 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1416 percpu_write(xen_cr3, cr3);
1418 /* Update xen_current_cr3 once the batch has actually
1420 xen_mc_callback(set_current_cr3, (void *)cr3);
1424 static void xen_write_cr3(unsigned long cr3)
1426 BUG_ON(preemptible());
1428 xen_mc_batch(); /* disables interrupts */
1430 /* Update while interrupts are disabled, so its atomic with
1432 percpu_write(xen_cr3, cr3);
1434 __xen_write_cr3(true, cr3);
1436 #ifdef CONFIG_X86_64
1438 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1440 __xen_write_cr3(false, __pa(user_pgd));
1442 __xen_write_cr3(false, 0);
1446 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1449 static int xen_pgd_alloc(struct mm_struct *mm)
1451 pgd_t *pgd = mm->pgd;
1454 BUG_ON(PagePinned(virt_to_page(pgd)));
1456 #ifdef CONFIG_X86_64
1458 struct page *page = virt_to_page(pgd);
1461 BUG_ON(page->private != 0);
1465 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1466 page->private = (unsigned long)user_pgd;
1468 if (user_pgd != NULL) {
1469 user_pgd[pgd_index(VSYSCALL_START)] =
1470 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1474 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1481 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1483 #ifdef CONFIG_X86_64
1484 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1487 free_page((unsigned long)user_pgd);
1491 #ifdef CONFIG_X86_32
1492 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1494 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1495 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1496 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1502 /* Init-time set_pte while constructing initial pagetables, which
1503 doesn't allow RO pagetable pages to be remapped RW */
1504 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1506 pte = mask_rw_pte(ptep, pte);
1508 xen_set_pte(ptep, pte);
1512 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1514 struct mmuext_op op;
1516 op.arg1.mfn = pfn_to_mfn(pfn);
1517 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1521 /* Early in boot, while setting up the initial pagetable, assume
1522 everything is pinned. */
1523 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1525 #ifdef CONFIG_FLATMEM
1526 BUG_ON(mem_map); /* should only be used early */
1528 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1529 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1532 /* Used for pmd and pud */
1533 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1535 #ifdef CONFIG_FLATMEM
1536 BUG_ON(mem_map); /* should only be used early */
1538 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1541 /* Early release_pte assumes that all pts are pinned, since there's
1542 only init_mm and anything attached to that is pinned. */
1543 static __init void xen_release_pte_init(unsigned long pfn)
1545 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1546 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1549 static __init void xen_release_pmd_init(unsigned long pfn)
1551 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1554 /* This needs to make sure the new pte page is pinned iff its being
1555 attached to a pinned pagetable. */
1556 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1558 struct page *page = pfn_to_page(pfn);
1560 if (PagePinned(virt_to_page(mm->pgd))) {
1561 SetPagePinned(page);
1564 if (!PageHighMem(page)) {
1565 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1566 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1567 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1569 /* make sure there are no stray mappings of
1571 kmap_flush_unused();
1576 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1578 xen_alloc_ptpage(mm, pfn, PT_PTE);
1581 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1583 xen_alloc_ptpage(mm, pfn, PT_PMD);
1586 /* This should never happen until we're OK to use struct page */
1587 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1589 struct page *page = pfn_to_page(pfn);
1591 if (PagePinned(page)) {
1592 if (!PageHighMem(page)) {
1593 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1594 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1595 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1597 ClearPagePinned(page);
1601 static void xen_release_pte(unsigned long pfn)
1603 xen_release_ptpage(pfn, PT_PTE);
1606 static void xen_release_pmd(unsigned long pfn)
1608 xen_release_ptpage(pfn, PT_PMD);
1611 #if PAGETABLE_LEVELS == 4
1612 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1614 xen_alloc_ptpage(mm, pfn, PT_PUD);
1617 static void xen_release_pud(unsigned long pfn)
1619 xen_release_ptpage(pfn, PT_PUD);
1623 void __init xen_reserve_top(void)
1625 #ifdef CONFIG_X86_32
1626 unsigned long top = HYPERVISOR_VIRT_START;
1627 struct xen_platform_parameters pp;
1629 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1630 top = pp.virt_start;
1632 reserve_top_address(-top);
1633 #endif /* CONFIG_X86_32 */
1637 * Like __va(), but returns address in the kernel mapping (which is
1638 * all we have until the physical memory mapping has been set up.
1640 static void *__ka(phys_addr_t paddr)
1642 #ifdef CONFIG_X86_64
1643 return (void *)(paddr + __START_KERNEL_map);
1649 /* Convert a machine address to physical address */
1650 static unsigned long m2p(phys_addr_t maddr)
1654 maddr &= PTE_PFN_MASK;
1655 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1660 /* Convert a machine address to kernel virtual */
1661 static void *m2v(phys_addr_t maddr)
1663 return __ka(m2p(maddr));
1666 static void set_page_prot(void *addr, pgprot_t prot)
1668 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1669 pte_t pte = pfn_pte(pfn, prot);
1671 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1675 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1677 unsigned pmdidx, pteidx;
1683 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1686 /* Reuse or allocate a page of ptes */
1687 if (pmd_present(pmd[pmdidx]))
1688 pte_page = m2v(pmd[pmdidx].pmd);
1690 /* Check for free pte pages */
1691 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1694 pte_page = &level1_ident_pgt[ident_pte];
1695 ident_pte += PTRS_PER_PTE;
1697 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1700 /* Install mappings */
1701 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1704 if (pfn > max_pfn_mapped)
1705 max_pfn_mapped = pfn;
1707 if (!pte_none(pte_page[pteidx]))
1710 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1711 pte_page[pteidx] = pte;
1715 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1716 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1718 set_page_prot(pmd, PAGE_KERNEL_RO);
1721 #ifdef CONFIG_X86_64
1722 static void convert_pfn_mfn(void *v)
1727 /* All levels are converted the same way, so just treat them
1729 for (i = 0; i < PTRS_PER_PTE; i++)
1730 pte[i] = xen_make_pte(pte[i].pte);
1734 * Set up the inital kernel pagetable.
1736 * We can construct this by grafting the Xen provided pagetable into
1737 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1738 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1739 * means that only the kernel has a physical mapping to start with -
1740 * but that's enough to get __va working. We need to fill in the rest
1741 * of the physical mapping once some sort of allocator has been set
1744 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1745 unsigned long max_pfn)
1750 /* Zap identity mapping */
1751 init_level4_pgt[0] = __pgd(0);
1753 /* Pre-constructed entries are in pfn, so convert to mfn */
1754 convert_pfn_mfn(init_level4_pgt);
1755 convert_pfn_mfn(level3_ident_pgt);
1756 convert_pfn_mfn(level3_kernel_pgt);
1758 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1759 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1761 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1762 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1764 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1765 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1766 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1768 /* Set up identity map */
1769 xen_map_identity_early(level2_ident_pgt, max_pfn);
1771 /* Make pagetable pieces RO */
1772 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1773 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1774 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1775 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1776 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1777 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1779 /* Pin down new L4 */
1780 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1781 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1783 /* Unpin Xen-provided one */
1784 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1787 pgd = init_level4_pgt;
1790 * At this stage there can be no user pgd, and no page
1791 * structure to attach it to, so make sure we just set kernel
1795 __xen_write_cr3(true, __pa(pgd));
1796 xen_mc_issue(PARAVIRT_LAZY_CPU);
1798 reserve_early(__pa(xen_start_info->pt_base),
1799 __pa(xen_start_info->pt_base +
1800 xen_start_info->nr_pt_frames * PAGE_SIZE),
1805 #else /* !CONFIG_X86_64 */
1806 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1808 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1809 unsigned long max_pfn)
1813 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1814 xen_start_info->nr_pt_frames * PAGE_SIZE +
1817 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1818 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1820 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1822 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1823 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1824 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1826 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1827 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1828 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1830 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1832 xen_write_cr3(__pa(swapper_pg_dir));
1834 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1836 reserve_early(__pa(xen_start_info->pt_base),
1837 __pa(xen_start_info->pt_base +
1838 xen_start_info->nr_pt_frames * PAGE_SIZE),
1841 return swapper_pg_dir;
1843 #endif /* CONFIG_X86_64 */
1845 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1849 phys >>= PAGE_SHIFT;
1852 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1853 #ifdef CONFIG_X86_F00F_BUG
1856 #ifdef CONFIG_X86_32
1859 # ifdef CONFIG_HIGHMEM
1860 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1863 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1865 #ifdef CONFIG_X86_LOCAL_APIC
1866 case FIX_APIC_BASE: /* maps dummy local APIC */
1868 case FIX_TEXT_POKE0:
1869 case FIX_TEXT_POKE1:
1870 /* All local page mappings */
1871 pte = pfn_pte(phys, prot);
1874 case FIX_PARAVIRT_BOOTMAP:
1875 /* This is an MFN, but it isn't an IO mapping from the
1877 pte = mfn_pte(phys, prot);
1881 /* By default, set_fixmap is used for hardware mappings */
1882 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1886 __native_set_fixmap(idx, pte);
1888 #ifdef CONFIG_X86_64
1889 /* Replicate changes to map the vsyscall page into the user
1890 pagetable vsyscall mapping. */
1891 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1892 unsigned long vaddr = __fix_to_virt(idx);
1893 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1898 static __init void xen_post_allocator_init(void)
1900 pv_mmu_ops.set_pte = xen_set_pte;
1901 pv_mmu_ops.set_pmd = xen_set_pmd;
1902 pv_mmu_ops.set_pud = xen_set_pud;
1903 #if PAGETABLE_LEVELS == 4
1904 pv_mmu_ops.set_pgd = xen_set_pgd;
1907 /* This will work as long as patching hasn't happened yet
1908 (which it hasn't) */
1909 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1910 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1911 pv_mmu_ops.release_pte = xen_release_pte;
1912 pv_mmu_ops.release_pmd = xen_release_pmd;
1913 #if PAGETABLE_LEVELS == 4
1914 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1915 pv_mmu_ops.release_pud = xen_release_pud;
1918 #ifdef CONFIG_X86_64
1919 SetPagePinned(virt_to_page(level3_user_vsyscall));
1921 xen_mark_init_mm_pinned();
1924 static void xen_leave_lazy_mmu(void)
1928 paravirt_leave_lazy_mmu();
1932 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1933 .read_cr2 = xen_read_cr2,
1934 .write_cr2 = xen_write_cr2,
1936 .read_cr3 = xen_read_cr3,
1937 .write_cr3 = xen_write_cr3,
1939 .flush_tlb_user = xen_flush_tlb,
1940 .flush_tlb_kernel = xen_flush_tlb,
1941 .flush_tlb_single = xen_flush_tlb_single,
1942 .flush_tlb_others = xen_flush_tlb_others,
1944 .pte_update = paravirt_nop,
1945 .pte_update_defer = paravirt_nop,
1947 .pgd_alloc = xen_pgd_alloc,
1948 .pgd_free = xen_pgd_free,
1950 .alloc_pte = xen_alloc_pte_init,
1951 .release_pte = xen_release_pte_init,
1952 .alloc_pmd = xen_alloc_pmd_init,
1953 .alloc_pmd_clone = paravirt_nop,
1954 .release_pmd = xen_release_pmd_init,
1956 #ifdef CONFIG_X86_64
1957 .set_pte = xen_set_pte,
1959 .set_pte = xen_set_pte_init,
1961 .set_pte_at = xen_set_pte_at,
1962 .set_pmd = xen_set_pmd_hyper,
1964 .ptep_modify_prot_start = __ptep_modify_prot_start,
1965 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1967 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1968 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1970 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1971 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1973 #ifdef CONFIG_X86_PAE
1974 .set_pte_atomic = xen_set_pte_atomic,
1975 .pte_clear = xen_pte_clear,
1976 .pmd_clear = xen_pmd_clear,
1977 #endif /* CONFIG_X86_PAE */
1978 .set_pud = xen_set_pud_hyper,
1980 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
1981 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
1983 #if PAGETABLE_LEVELS == 4
1984 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
1985 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
1986 .set_pgd = xen_set_pgd_hyper,
1988 .alloc_pud = xen_alloc_pmd_init,
1989 .release_pud = xen_release_pmd_init,
1990 #endif /* PAGETABLE_LEVELS == 4 */
1992 .activate_mm = xen_activate_mm,
1993 .dup_mmap = xen_dup_mmap,
1994 .exit_mmap = xen_exit_mmap,
1997 .enter = paravirt_enter_lazy_mmu,
1998 .leave = xen_leave_lazy_mmu,
2001 .set_fixmap = xen_set_fixmap,
2004 void __init xen_init_mmu_ops(void)
2006 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2007 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2008 pv_mmu_ops = xen_mmu_ops;
2011 #ifdef CONFIG_XEN_DEBUG_FS
2013 static struct dentry *d_mmu_debug;
2015 static int __init xen_mmu_debugfs(void)
2017 struct dentry *d_xen = xen_init_debugfs();
2022 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2024 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2026 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2027 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2028 &mmu_stats.pgd_update_pinned);
2029 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2030 &mmu_stats.pgd_update_pinned);
2032 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2033 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2034 &mmu_stats.pud_update_pinned);
2035 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2036 &mmu_stats.pud_update_pinned);
2038 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2039 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2040 &mmu_stats.pmd_update_pinned);
2041 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2042 &mmu_stats.pmd_update_pinned);
2044 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2045 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2046 // &mmu_stats.pte_update_pinned);
2047 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2048 &mmu_stats.pte_update_pinned);
2050 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2051 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2052 &mmu_stats.mmu_update_extended);
2053 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2054 mmu_stats.mmu_update_histo, 20);
2056 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2057 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2058 &mmu_stats.set_pte_at_batched);
2059 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2060 &mmu_stats.set_pte_at_current);
2061 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2062 &mmu_stats.set_pte_at_kernel);
2064 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2065 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2066 &mmu_stats.prot_commit_batched);
2070 fs_initcall(xen_mmu_debugfs);
2072 #endif /* CONFIG_XEN_DEBUG_FS */