4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/module.h>
47 #include <linux/gfp.h>
49 #include <asm/pgtable.h>
50 #include <asm/tlbflush.h>
51 #include <asm/fixmap.h>
52 #include <asm/mmu_context.h>
53 #include <asm/setup.h>
54 #include <asm/paravirt.h>
56 #include <asm/linkage.h>
59 #include <asm/xen/hypercall.h>
60 #include <asm/xen/hypervisor.h>
64 #include <xen/interface/xen.h>
65 #include <xen/interface/version.h>
66 #include <xen/interface/memory.h>
67 #include <xen/hvc-console.h>
69 #include "multicalls.h"
73 #define MMU_UPDATE_HISTO 30
76 * Protects atomic reservation decrease/increase against concurrent increases.
77 * Also protects non-atomic updates of current_pages and driver_pages, and
80 DEFINE_SPINLOCK(xen_reservation_lock);
82 #ifdef CONFIG_XEN_DEBUG_FS
86 u32 pgd_update_pinned;
87 u32 pgd_update_batched;
90 u32 pud_update_pinned;
91 u32 pud_update_batched;
94 u32 pmd_update_pinned;
95 u32 pmd_update_batched;
98 u32 pte_update_pinned;
99 u32 pte_update_batched;
102 u32 mmu_update_extended;
103 u32 mmu_update_histo[MMU_UPDATE_HISTO];
106 u32 prot_commit_batched;
109 u32 set_pte_at_batched;
110 u32 set_pte_at_pinned;
111 u32 set_pte_at_current;
112 u32 set_pte_at_kernel;
115 static u8 zero_stats;
117 static inline void check_zero(void)
119 if (unlikely(zero_stats)) {
120 memset(&mmu_stats, 0, sizeof(mmu_stats));
125 #define ADD_STATS(elem, val) \
126 do { check_zero(); mmu_stats.elem += (val); } while(0)
128 #else /* !CONFIG_XEN_DEBUG_FS */
130 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
132 #endif /* CONFIG_XEN_DEBUG_FS */
136 * Identity map, in addition to plain kernel map. This needs to be
137 * large enough to allocate page table pages to allocate the rest.
138 * Each page can map 2MB.
140 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
143 /* l3 pud for userspace vsyscall mapping */
144 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
145 #endif /* CONFIG_X86_64 */
148 * Note about cr3 (pagetable base) values:
150 * xen_cr3 contains the current logical cr3 value; it contains the
151 * last set cr3. This may not be the current effective cr3, because
152 * its update may be being lazily deferred. However, a vcpu looking
153 * at its own cr3 can use this value knowing that it everything will
154 * be self-consistent.
156 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
157 * hypercall to set the vcpu cr3 is complete (so it may be a little
158 * out of date, but it will never be set early). If one vcpu is
159 * looking at another vcpu's cr3 value, it should use this variable.
161 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
162 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
166 * Just beyond the highest usermode address. STACK_TOP_MAX has a
167 * redzone above it, so round it up to a PGD boundary.
169 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
172 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
173 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
175 /* Placeholder for holes in the address space */
176 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
177 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
179 /* Array of pointers to pages containing p2m entries */
180 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
181 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
183 /* Arrays of p2m arrays expressed in mfns used for save/restore */
184 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
186 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
189 static inline unsigned p2m_top_index(unsigned long pfn)
191 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
192 return pfn / P2M_ENTRIES_PER_PAGE;
195 static inline unsigned p2m_index(unsigned long pfn)
197 return pfn % P2M_ENTRIES_PER_PAGE;
200 /* Build the parallel p2m_top_mfn structures */
201 void xen_build_mfn_list_list(void)
205 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
206 unsigned topidx = p2m_top_index(pfn);
208 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
211 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
212 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
213 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
217 void xen_setup_mfn_list_list(void)
219 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
221 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
222 virt_to_mfn(p2m_top_mfn_list);
223 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
226 /* Set up p2m_top to point to the domain-builder provided p2m pages */
227 void __init xen_build_dynamic_phys_to_machine(void)
229 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
230 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
233 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
234 unsigned topidx = p2m_top_index(pfn);
236 p2m_top[topidx] = &mfn_list[pfn];
239 xen_build_mfn_list_list();
242 unsigned long get_phys_to_machine(unsigned long pfn)
244 unsigned topidx, idx;
246 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
247 return INVALID_P2M_ENTRY;
249 topidx = p2m_top_index(pfn);
250 idx = p2m_index(pfn);
251 return p2m_top[topidx][idx];
253 EXPORT_SYMBOL_GPL(get_phys_to_machine);
255 /* install a new p2m_top page */
256 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
258 unsigned topidx = p2m_top_index(pfn);
259 unsigned long **pfnp, *mfnp;
262 pfnp = &p2m_top[topidx];
263 mfnp = &p2m_top_mfn[topidx];
265 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
266 p[i] = INVALID_P2M_ENTRY;
268 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
269 *mfnp = virt_to_mfn(p);
276 static void alloc_p2m(unsigned long pfn)
280 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
283 if (!install_p2mtop_page(pfn, p))
284 free_page((unsigned long)p);
287 /* Try to install p2m mapping; fail if intermediate bits missing */
288 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
290 unsigned topidx, idx;
292 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
293 BUG_ON(mfn != INVALID_P2M_ENTRY);
297 topidx = p2m_top_index(pfn);
298 if (p2m_top[topidx] == p2m_missing) {
299 if (mfn == INVALID_P2M_ENTRY)
304 idx = p2m_index(pfn);
305 p2m_top[topidx][idx] = mfn;
310 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
312 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
313 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
317 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
320 if (!__set_phys_to_machine(pfn, mfn))
325 unsigned long arbitrary_virt_to_mfn(void *vaddr)
327 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
329 return PFN_DOWN(maddr.maddr);
332 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
334 unsigned long address = (unsigned long)vaddr;
340 * if the PFN is in the linear mapped vaddr range, we can just use
341 * the (quick) virt_to_machine() p2m lookup
343 if (virt_addr_valid(vaddr))
344 return virt_to_machine(vaddr);
346 /* otherwise we have to do a (slower) full page-table walk */
348 pte = lookup_address(address, &level);
350 offset = address & ~PAGE_MASK;
351 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
354 void make_lowmem_page_readonly(void *vaddr)
357 unsigned long address = (unsigned long)vaddr;
360 pte = lookup_address(address, &level);
363 ptev = pte_wrprotect(*pte);
365 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
369 void make_lowmem_page_readwrite(void *vaddr)
372 unsigned long address = (unsigned long)vaddr;
375 pte = lookup_address(address, &level);
378 ptev = pte_mkwrite(*pte);
380 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
385 static bool xen_page_pinned(void *ptr)
387 struct page *page = virt_to_page(ptr);
389 return PagePinned(page);
392 static bool xen_iomap_pte(pte_t pte)
394 return pte_flags(pte) & _PAGE_IOMAP;
397 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
399 struct multicall_space mcs;
400 struct mmu_update *u;
402 mcs = xen_mc_entry(sizeof(*u));
405 /* ptep might be kmapped when using 32-bit HIGHPTE */
406 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
407 u->val = pte_val_ma(pteval);
409 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_IO);
411 xen_mc_issue(PARAVIRT_LAZY_MMU);
414 static void xen_extend_mmu_update(const struct mmu_update *update)
416 struct multicall_space mcs;
417 struct mmu_update *u;
419 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
421 if (mcs.mc != NULL) {
422 ADD_STATS(mmu_update_extended, 1);
423 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
427 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
428 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
430 ADD_STATS(mmu_update_histo[0], 1);
432 ADD_STATS(mmu_update, 1);
433 mcs = __xen_mc_entry(sizeof(*u));
434 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
435 ADD_STATS(mmu_update_histo[1], 1);
442 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
450 /* ptr may be ioremapped for 64-bit pagetable setup */
451 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
452 u.val = pmd_val_ma(val);
453 xen_extend_mmu_update(&u);
455 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
457 xen_mc_issue(PARAVIRT_LAZY_MMU);
462 void xen_set_pmd(pmd_t *ptr, pmd_t val)
464 ADD_STATS(pmd_update, 1);
466 /* If page is not pinned, we can just update the entry
468 if (!xen_page_pinned(ptr)) {
473 ADD_STATS(pmd_update_pinned, 1);
475 xen_set_pmd_hyper(ptr, val);
479 * Associate a virtual page frame with a given physical page frame
480 * and protection flags for that frame.
482 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
484 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
487 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
488 pte_t *ptep, pte_t pteval)
490 if (xen_iomap_pte(pteval)) {
491 xen_set_iomap_pte(ptep, pteval);
495 ADD_STATS(set_pte_at, 1);
496 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
497 ADD_STATS(set_pte_at_current, mm == current->mm);
498 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
500 if (mm == current->mm || mm == &init_mm) {
501 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
502 struct multicall_space mcs;
503 mcs = xen_mc_entry(0);
505 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
506 ADD_STATS(set_pte_at_batched, 1);
507 xen_mc_issue(PARAVIRT_LAZY_MMU);
510 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
513 xen_set_pte(ptep, pteval);
518 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
519 unsigned long addr, pte_t *ptep)
521 /* Just return the pte as-is. We preserve the bits on commit */
525 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
526 pte_t *ptep, pte_t pte)
532 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
533 u.val = pte_val_ma(pte);
534 xen_extend_mmu_update(&u);
536 ADD_STATS(prot_commit, 1);
537 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
539 xen_mc_issue(PARAVIRT_LAZY_MMU);
542 /* Assume pteval_t is equivalent to all the other *val_t types. */
543 static pteval_t pte_mfn_to_pfn(pteval_t val)
545 if (val & _PAGE_PRESENT) {
546 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
547 pteval_t flags = val & PTE_FLAGS_MASK;
548 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
554 static pteval_t pte_pfn_to_mfn(pteval_t val)
556 if (val & _PAGE_PRESENT) {
557 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
558 pteval_t flags = val & PTE_FLAGS_MASK;
559 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
565 static pteval_t iomap_pte(pteval_t val)
567 if (val & _PAGE_PRESENT) {
568 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
569 pteval_t flags = val & PTE_FLAGS_MASK;
571 /* We assume the pte frame number is a MFN, so
572 just use it as-is. */
573 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
579 pteval_t xen_pte_val(pte_t pte)
581 if (xen_initial_domain() && (pte.pte & _PAGE_IOMAP))
584 return pte_mfn_to_pfn(pte.pte);
586 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
588 pgdval_t xen_pgd_val(pgd_t pgd)
590 return pte_mfn_to_pfn(pgd.pgd);
592 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
594 pte_t xen_make_pte(pteval_t pte)
596 phys_addr_t addr = (pte & PTE_PFN_MASK);
599 * Unprivileged domains are allowed to do IOMAPpings for
600 * PCI passthrough, but not map ISA space. The ISA
601 * mappings are just dummy local mappings to keep other
602 * parts of the kernel happy.
604 if (unlikely(pte & _PAGE_IOMAP) &&
605 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
606 pte = iomap_pte(pte);
609 pte = pte_pfn_to_mfn(pte);
612 return native_make_pte(pte);
614 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
616 pgd_t xen_make_pgd(pgdval_t pgd)
618 pgd = pte_pfn_to_mfn(pgd);
619 return native_make_pgd(pgd);
621 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
623 pmdval_t xen_pmd_val(pmd_t pmd)
625 return pte_mfn_to_pfn(pmd.pmd);
627 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
629 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
637 /* ptr may be ioremapped for 64-bit pagetable setup */
638 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
639 u.val = pud_val_ma(val);
640 xen_extend_mmu_update(&u);
642 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
644 xen_mc_issue(PARAVIRT_LAZY_MMU);
649 void xen_set_pud(pud_t *ptr, pud_t val)
651 ADD_STATS(pud_update, 1);
653 /* If page is not pinned, we can just update the entry
655 if (!xen_page_pinned(ptr)) {
660 ADD_STATS(pud_update_pinned, 1);
662 xen_set_pud_hyper(ptr, val);
665 void xen_set_pte(pte_t *ptep, pte_t pte)
667 if (xen_iomap_pte(pte)) {
668 xen_set_iomap_pte(ptep, pte);
672 ADD_STATS(pte_update, 1);
673 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
674 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
676 #ifdef CONFIG_X86_PAE
677 ptep->pte_high = pte.pte_high;
679 ptep->pte_low = pte.pte_low;
685 #ifdef CONFIG_X86_PAE
686 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
688 if (xen_iomap_pte(pte)) {
689 xen_set_iomap_pte(ptep, pte);
693 set_64bit((u64 *)ptep, native_pte_val(pte));
696 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
699 smp_wmb(); /* make sure low gets written first */
703 void xen_pmd_clear(pmd_t *pmdp)
705 set_pmd(pmdp, __pmd(0));
707 #endif /* CONFIG_X86_PAE */
709 pmd_t xen_make_pmd(pmdval_t pmd)
711 pmd = pte_pfn_to_mfn(pmd);
712 return native_make_pmd(pmd);
714 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
716 #if PAGETABLE_LEVELS == 4
717 pudval_t xen_pud_val(pud_t pud)
719 return pte_mfn_to_pfn(pud.pud);
721 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
723 pud_t xen_make_pud(pudval_t pud)
725 pud = pte_pfn_to_mfn(pud);
727 return native_make_pud(pud);
729 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
731 pgd_t *xen_get_user_pgd(pgd_t *pgd)
733 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
734 unsigned offset = pgd - pgd_page;
735 pgd_t *user_ptr = NULL;
737 if (offset < pgd_index(USER_LIMIT)) {
738 struct page *page = virt_to_page(pgd_page);
739 user_ptr = (pgd_t *)page->private;
747 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
751 u.ptr = virt_to_machine(ptr).maddr;
752 u.val = pgd_val_ma(val);
753 xen_extend_mmu_update(&u);
757 * Raw hypercall-based set_pgd, intended for in early boot before
758 * there's a page structure. This implies:
759 * 1. The only existing pagetable is the kernel's
760 * 2. It is always pinned
761 * 3. It has no user pagetable attached to it
763 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
769 __xen_set_pgd_hyper(ptr, val);
771 xen_mc_issue(PARAVIRT_LAZY_MMU);
776 void xen_set_pgd(pgd_t *ptr, pgd_t val)
778 pgd_t *user_ptr = xen_get_user_pgd(ptr);
780 ADD_STATS(pgd_update, 1);
782 /* If page is not pinned, we can just update the entry
784 if (!xen_page_pinned(ptr)) {
787 WARN_ON(xen_page_pinned(user_ptr));
793 ADD_STATS(pgd_update_pinned, 1);
794 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
796 /* If it's pinned, then we can at least batch the kernel and
797 user updates together. */
800 __xen_set_pgd_hyper(ptr, val);
802 __xen_set_pgd_hyper(user_ptr, val);
804 xen_mc_issue(PARAVIRT_LAZY_MMU);
806 #endif /* PAGETABLE_LEVELS == 4 */
809 * (Yet another) pagetable walker. This one is intended for pinning a
810 * pagetable. This means that it walks a pagetable and calls the
811 * callback function on each page it finds making up the page table,
812 * at every level. It walks the entire pagetable, but it only bothers
813 * pinning pte pages which are below limit. In the normal case this
814 * will be STACK_TOP_MAX, but at boot we need to pin up to
817 * For 32-bit the important bit is that we don't pin beyond there,
818 * because then we start getting into Xen's ptes.
820 * For 64-bit, we must skip the Xen hole in the middle of the address
821 * space, just after the big x86-64 virtual hole.
823 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
824 int (*func)(struct mm_struct *mm, struct page *,
829 unsigned hole_low, hole_high;
830 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
831 unsigned pgdidx, pudidx, pmdidx;
833 /* The limit is the last byte to be touched */
835 BUG_ON(limit >= FIXADDR_TOP);
837 if (xen_feature(XENFEAT_auto_translated_physmap))
841 * 64-bit has a great big hole in the middle of the address
842 * space, which contains the Xen mappings. On 32-bit these
843 * will end up making a zero-sized hole and so is a no-op.
845 hole_low = pgd_index(USER_LIMIT);
846 hole_high = pgd_index(PAGE_OFFSET);
848 pgdidx_limit = pgd_index(limit);
850 pudidx_limit = pud_index(limit);
855 pmdidx_limit = pmd_index(limit);
860 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
863 if (pgdidx >= hole_low && pgdidx < hole_high)
866 if (!pgd_val(pgd[pgdidx]))
869 pud = pud_offset(&pgd[pgdidx], 0);
871 if (PTRS_PER_PUD > 1) /* not folded */
872 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
874 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
877 if (pgdidx == pgdidx_limit &&
878 pudidx > pudidx_limit)
881 if (pud_none(pud[pudidx]))
884 pmd = pmd_offset(&pud[pudidx], 0);
886 if (PTRS_PER_PMD > 1) /* not folded */
887 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
889 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
892 if (pgdidx == pgdidx_limit &&
893 pudidx == pudidx_limit &&
894 pmdidx > pmdidx_limit)
897 if (pmd_none(pmd[pmdidx]))
900 pte = pmd_page(pmd[pmdidx]);
901 flush |= (*func)(mm, pte, PT_PTE);
907 /* Do the top level last, so that the callbacks can use it as
908 a cue to do final things like tlb flushes. */
909 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
914 static int xen_pgd_walk(struct mm_struct *mm,
915 int (*func)(struct mm_struct *mm, struct page *,
919 return __xen_pgd_walk(mm, mm->pgd, func, limit);
922 /* If we're using split pte locks, then take the page's lock and
923 return a pointer to it. Otherwise return NULL. */
924 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
926 spinlock_t *ptl = NULL;
928 #if USE_SPLIT_PTLOCKS
929 ptl = __pte_lockptr(page);
930 spin_lock_nest_lock(ptl, &mm->page_table_lock);
936 static void xen_pte_unlock(void *v)
942 static void xen_do_pin(unsigned level, unsigned long pfn)
944 struct mmuext_op *op;
945 struct multicall_space mcs;
947 mcs = __xen_mc_entry(sizeof(*op));
950 op->arg1.mfn = pfn_to_mfn(pfn);
951 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
954 static int xen_pin_page(struct mm_struct *mm, struct page *page,
957 unsigned pgfl = TestSetPagePinned(page);
961 flush = 0; /* already pinned */
962 else if (PageHighMem(page))
963 /* kmaps need flushing if we found an unpinned
967 void *pt = lowmem_page_address(page);
968 unsigned long pfn = page_to_pfn(page);
969 struct multicall_space mcs = __xen_mc_entry(0);
975 * We need to hold the pagetable lock between the time
976 * we make the pagetable RO and when we actually pin
977 * it. If we don't, then other users may come in and
978 * attempt to update the pagetable by writing it,
979 * which will fail because the memory is RO but not
980 * pinned, so Xen won't do the trap'n'emulate.
982 * If we're using split pte locks, we can't hold the
983 * entire pagetable's worth of locks during the
984 * traverse, because we may wrap the preempt count (8
985 * bits). The solution is to mark RO and pin each PTE
986 * page while holding the lock. This means the number
987 * of locks we end up holding is never more than a
988 * batch size (~32 entries, at present).
990 * If we're not using split pte locks, we needn't pin
991 * the PTE pages independently, because we're
992 * protected by the overall pagetable lock.
996 ptl = xen_pte_lock(page, mm);
998 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
999 pfn_pte(pfn, PAGE_KERNEL_RO),
1000 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1003 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
1005 /* Queue a deferred unlock for when this batch
1007 xen_mc_callback(xen_pte_unlock, ptl);
1014 /* This is called just after a mm has been created, but it has not
1015 been used yet. We need to make sure that its pagetable is all
1016 read-only, and can be pinned. */
1017 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
1021 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
1022 /* re-enable interrupts for flushing */
1025 kmap_flush_unused();
1030 #ifdef CONFIG_X86_64
1032 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1034 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
1037 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
1038 xen_do_pin(MMUEXT_PIN_L4_TABLE,
1039 PFN_DOWN(__pa(user_pgd)));
1042 #else /* CONFIG_X86_32 */
1043 #ifdef CONFIG_X86_PAE
1044 /* Need to make sure unshared kernel PMD is pinnable */
1045 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1048 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
1049 #endif /* CONFIG_X86_64 */
1053 static void xen_pgd_pin(struct mm_struct *mm)
1055 __xen_pgd_pin(mm, mm->pgd);
1059 * On save, we need to pin all pagetables to make sure they get their
1060 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1061 * them (unpinned pgds are not currently in use, probably because the
1062 * process is under construction or destruction).
1064 * Expected to be called in stop_machine() ("equivalent to taking
1065 * every spinlock in the system"), so the locking doesn't really
1066 * matter all that much.
1068 void xen_mm_pin_all(void)
1070 unsigned long flags;
1073 spin_lock_irqsave(&pgd_lock, flags);
1075 list_for_each_entry(page, &pgd_list, lru) {
1076 if (!PagePinned(page)) {
1077 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1078 SetPageSavePinned(page);
1082 spin_unlock_irqrestore(&pgd_lock, flags);
1086 * The init_mm pagetable is really pinned as soon as its created, but
1087 * that's before we have page structures to store the bits. So do all
1088 * the book-keeping now.
1090 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1091 enum pt_level level)
1093 SetPagePinned(page);
1097 static void __init xen_mark_init_mm_pinned(void)
1099 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1102 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1103 enum pt_level level)
1105 unsigned pgfl = TestClearPagePinned(page);
1107 if (pgfl && !PageHighMem(page)) {
1108 void *pt = lowmem_page_address(page);
1109 unsigned long pfn = page_to_pfn(page);
1110 spinlock_t *ptl = NULL;
1111 struct multicall_space mcs;
1114 * Do the converse to pin_page. If we're using split
1115 * pte locks, we must be holding the lock for while
1116 * the pte page is unpinned but still RO to prevent
1117 * concurrent updates from seeing it in this
1118 * partially-pinned state.
1120 if (level == PT_PTE) {
1121 ptl = xen_pte_lock(page, mm);
1124 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1127 mcs = __xen_mc_entry(0);
1129 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1130 pfn_pte(pfn, PAGE_KERNEL),
1131 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1134 /* unlock when batch completed */
1135 xen_mc_callback(xen_pte_unlock, ptl);
1139 return 0; /* never need to flush on unpin */
1142 /* Release a pagetables pages back as normal RW */
1143 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1147 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1149 #ifdef CONFIG_X86_64
1151 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1154 xen_do_pin(MMUEXT_UNPIN_TABLE,
1155 PFN_DOWN(__pa(user_pgd)));
1156 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1161 #ifdef CONFIG_X86_PAE
1162 /* Need to make sure unshared kernel PMD is unpinned */
1163 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1167 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1172 static void xen_pgd_unpin(struct mm_struct *mm)
1174 __xen_pgd_unpin(mm, mm->pgd);
1178 * On resume, undo any pinning done at save, so that the rest of the
1179 * kernel doesn't see any unexpected pinned pagetables.
1181 void xen_mm_unpin_all(void)
1183 unsigned long flags;
1186 spin_lock_irqsave(&pgd_lock, flags);
1188 list_for_each_entry(page, &pgd_list, lru) {
1189 if (PageSavePinned(page)) {
1190 BUG_ON(!PagePinned(page));
1191 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1192 ClearPageSavePinned(page);
1196 spin_unlock_irqrestore(&pgd_lock, flags);
1199 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1201 spin_lock(&next->page_table_lock);
1203 spin_unlock(&next->page_table_lock);
1206 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1208 spin_lock(&mm->page_table_lock);
1210 spin_unlock(&mm->page_table_lock);
1215 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1216 we need to repoint it somewhere else before we can unpin it. */
1217 static void drop_other_mm_ref(void *info)
1219 struct mm_struct *mm = info;
1220 struct mm_struct *active_mm;
1222 active_mm = percpu_read(cpu_tlbstate.active_mm);
1224 if (active_mm == mm)
1225 leave_mm(smp_processor_id());
1227 /* If this cpu still has a stale cr3 reference, then make sure
1228 it has been flushed. */
1229 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1230 load_cr3(swapper_pg_dir);
1233 static void xen_drop_mm_ref(struct mm_struct *mm)
1238 if (current->active_mm == mm) {
1239 if (current->mm == mm)
1240 load_cr3(swapper_pg_dir);
1242 leave_mm(smp_processor_id());
1245 /* Get the "official" set of cpus referring to our pagetable. */
1246 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1247 for_each_online_cpu(cpu) {
1248 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1249 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1251 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1255 cpumask_copy(mask, mm_cpumask(mm));
1257 /* It's possible that a vcpu may have a stale reference to our
1258 cr3, because its in lazy mode, and it hasn't yet flushed
1259 its set of pending hypercalls yet. In this case, we can
1260 look at its actual current cr3 value, and force it to flush
1262 for_each_online_cpu(cpu) {
1263 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1264 cpumask_set_cpu(cpu, mask);
1267 if (!cpumask_empty(mask))
1268 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1269 free_cpumask_var(mask);
1272 static void xen_drop_mm_ref(struct mm_struct *mm)
1274 if (current->active_mm == mm)
1275 load_cr3(swapper_pg_dir);
1280 * While a process runs, Xen pins its pagetables, which means that the
1281 * hypervisor forces it to be read-only, and it controls all updates
1282 * to it. This means that all pagetable updates have to go via the
1283 * hypervisor, which is moderately expensive.
1285 * Since we're pulling the pagetable down, we switch to use init_mm,
1286 * unpin old process pagetable and mark it all read-write, which
1287 * allows further operations on it to be simple memory accesses.
1289 * The only subtle point is that another CPU may be still using the
1290 * pagetable because of lazy tlb flushing. This means we need need to
1291 * switch all CPUs off this pagetable before we can unpin it.
1293 void xen_exit_mmap(struct mm_struct *mm)
1295 get_cpu(); /* make sure we don't move around */
1296 xen_drop_mm_ref(mm);
1299 spin_lock(&mm->page_table_lock);
1301 /* pgd may not be pinned in the error exit path of execve */
1302 if (xen_page_pinned(mm->pgd))
1305 spin_unlock(&mm->page_table_lock);
1308 static __init void xen_pagetable_setup_start(pgd_t *base)
1312 static void xen_post_allocator_init(void);
1314 static __init void xen_pagetable_setup_done(pgd_t *base)
1316 xen_setup_shared_info();
1317 xen_post_allocator_init();
1320 static void xen_write_cr2(unsigned long cr2)
1322 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1325 static unsigned long xen_read_cr2(void)
1327 return percpu_read(xen_vcpu)->arch.cr2;
1330 unsigned long xen_read_cr2_direct(void)
1332 return percpu_read(xen_vcpu_info.arch.cr2);
1335 static void xen_flush_tlb(void)
1337 struct mmuext_op *op;
1338 struct multicall_space mcs;
1342 mcs = xen_mc_entry(sizeof(*op));
1345 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1346 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1348 xen_mc_issue(PARAVIRT_LAZY_MMU);
1353 static void xen_flush_tlb_single(unsigned long addr)
1355 struct mmuext_op *op;
1356 struct multicall_space mcs;
1360 mcs = xen_mc_entry(sizeof(*op));
1362 op->cmd = MMUEXT_INVLPG_LOCAL;
1363 op->arg1.linear_addr = addr & PAGE_MASK;
1364 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1366 xen_mc_issue(PARAVIRT_LAZY_MMU);
1371 static void xen_flush_tlb_others(const struct cpumask *cpus,
1372 struct mm_struct *mm, unsigned long va)
1375 struct mmuext_op op;
1376 DECLARE_BITMAP(mask, NR_CPUS);
1378 struct multicall_space mcs;
1380 if (cpumask_empty(cpus))
1381 return; /* nothing to do */
1383 mcs = xen_mc_entry(sizeof(*args));
1385 args->op.arg2.vcpumask = to_cpumask(args->mask);
1387 /* Remove us, and any offline CPUS. */
1388 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1389 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1391 if (va == TLB_FLUSH_ALL) {
1392 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1394 args->op.cmd = MMUEXT_INVLPG_MULTI;
1395 args->op.arg1.linear_addr = va;
1398 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1400 xen_mc_issue(PARAVIRT_LAZY_MMU);
1403 static unsigned long xen_read_cr3(void)
1405 return percpu_read(xen_cr3);
1408 static void set_current_cr3(void *v)
1410 percpu_write(xen_current_cr3, (unsigned long)v);
1413 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1415 struct mmuext_op *op;
1416 struct multicall_space mcs;
1420 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1424 WARN_ON(mfn == 0 && kernel);
1426 mcs = __xen_mc_entry(sizeof(*op));
1429 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1432 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1435 percpu_write(xen_cr3, cr3);
1437 /* Update xen_current_cr3 once the batch has actually
1439 xen_mc_callback(set_current_cr3, (void *)cr3);
1443 static void xen_write_cr3(unsigned long cr3)
1445 BUG_ON(preemptible());
1447 xen_mc_batch(); /* disables interrupts */
1449 /* Update while interrupts are disabled, so its atomic with
1451 percpu_write(xen_cr3, cr3);
1453 __xen_write_cr3(true, cr3);
1455 #ifdef CONFIG_X86_64
1457 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1459 __xen_write_cr3(false, __pa(user_pgd));
1461 __xen_write_cr3(false, 0);
1465 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1468 static int xen_pgd_alloc(struct mm_struct *mm)
1470 pgd_t *pgd = mm->pgd;
1473 BUG_ON(PagePinned(virt_to_page(pgd)));
1475 #ifdef CONFIG_X86_64
1477 struct page *page = virt_to_page(pgd);
1480 BUG_ON(page->private != 0);
1484 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1485 page->private = (unsigned long)user_pgd;
1487 if (user_pgd != NULL) {
1488 user_pgd[pgd_index(VSYSCALL_START)] =
1489 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1493 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1500 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1502 #ifdef CONFIG_X86_64
1503 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1506 free_page((unsigned long)user_pgd);
1510 #ifdef CONFIG_X86_32
1511 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1513 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1514 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1515 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1521 /* Init-time set_pte while constructing initial pagetables, which
1522 doesn't allow RO pagetable pages to be remapped RW */
1523 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1525 pte = mask_rw_pte(ptep, pte);
1527 xen_set_pte(ptep, pte);
1531 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1533 struct mmuext_op op;
1535 op.arg1.mfn = pfn_to_mfn(pfn);
1536 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1540 /* Early in boot, while setting up the initial pagetable, assume
1541 everything is pinned. */
1542 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1544 #ifdef CONFIG_FLATMEM
1545 BUG_ON(mem_map); /* should only be used early */
1547 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1548 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1551 /* Used for pmd and pud */
1552 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1554 #ifdef CONFIG_FLATMEM
1555 BUG_ON(mem_map); /* should only be used early */
1557 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1560 /* Early release_pte assumes that all pts are pinned, since there's
1561 only init_mm and anything attached to that is pinned. */
1562 static __init void xen_release_pte_init(unsigned long pfn)
1564 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1565 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1568 static __init void xen_release_pmd_init(unsigned long pfn)
1570 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1573 /* This needs to make sure the new pte page is pinned iff its being
1574 attached to a pinned pagetable. */
1575 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1577 struct page *page = pfn_to_page(pfn);
1579 if (PagePinned(virt_to_page(mm->pgd))) {
1580 SetPagePinned(page);
1582 if (!PageHighMem(page)) {
1583 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1584 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1585 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1587 /* make sure there are no stray mappings of
1589 kmap_flush_unused();
1594 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1596 xen_alloc_ptpage(mm, pfn, PT_PTE);
1599 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1601 xen_alloc_ptpage(mm, pfn, PT_PMD);
1604 /* This should never happen until we're OK to use struct page */
1605 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1607 struct page *page = pfn_to_page(pfn);
1609 if (PagePinned(page)) {
1610 if (!PageHighMem(page)) {
1611 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1612 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1613 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1615 ClearPagePinned(page);
1619 static void xen_release_pte(unsigned long pfn)
1621 xen_release_ptpage(pfn, PT_PTE);
1624 static void xen_release_pmd(unsigned long pfn)
1626 xen_release_ptpage(pfn, PT_PMD);
1629 #if PAGETABLE_LEVELS == 4
1630 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1632 xen_alloc_ptpage(mm, pfn, PT_PUD);
1635 static void xen_release_pud(unsigned long pfn)
1637 xen_release_ptpage(pfn, PT_PUD);
1641 void __init xen_reserve_top(void)
1643 #ifdef CONFIG_X86_32
1644 unsigned long top = HYPERVISOR_VIRT_START;
1645 struct xen_platform_parameters pp;
1647 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1648 top = pp.virt_start;
1650 reserve_top_address(-top);
1651 #endif /* CONFIG_X86_32 */
1655 * Like __va(), but returns address in the kernel mapping (which is
1656 * all we have until the physical memory mapping has been set up.
1658 static void *__ka(phys_addr_t paddr)
1660 #ifdef CONFIG_X86_64
1661 return (void *)(paddr + __START_KERNEL_map);
1667 /* Convert a machine address to physical address */
1668 static unsigned long m2p(phys_addr_t maddr)
1672 maddr &= PTE_PFN_MASK;
1673 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1678 /* Convert a machine address to kernel virtual */
1679 static void *m2v(phys_addr_t maddr)
1681 return __ka(m2p(maddr));
1684 static void set_page_prot(void *addr, pgprot_t prot)
1686 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1687 pte_t pte = pfn_pte(pfn, prot);
1689 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1693 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1695 unsigned pmdidx, pteidx;
1701 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1704 /* Reuse or allocate a page of ptes */
1705 if (pmd_present(pmd[pmdidx]))
1706 pte_page = m2v(pmd[pmdidx].pmd);
1708 /* Check for free pte pages */
1709 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1712 pte_page = &level1_ident_pgt[ident_pte];
1713 ident_pte += PTRS_PER_PTE;
1715 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1718 /* Install mappings */
1719 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1722 if (pfn > max_pfn_mapped)
1723 max_pfn_mapped = pfn;
1725 if (!pte_none(pte_page[pteidx]))
1728 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1729 pte_page[pteidx] = pte;
1733 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1734 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1736 set_page_prot(pmd, PAGE_KERNEL_RO);
1739 #ifdef CONFIG_X86_64
1740 static void convert_pfn_mfn(void *v)
1745 /* All levels are converted the same way, so just treat them
1747 for (i = 0; i < PTRS_PER_PTE; i++)
1748 pte[i] = xen_make_pte(pte[i].pte);
1752 * Set up the inital kernel pagetable.
1754 * We can construct this by grafting the Xen provided pagetable into
1755 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1756 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1757 * means that only the kernel has a physical mapping to start with -
1758 * but that's enough to get __va working. We need to fill in the rest
1759 * of the physical mapping once some sort of allocator has been set
1762 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1763 unsigned long max_pfn)
1768 /* Zap identity mapping */
1769 init_level4_pgt[0] = __pgd(0);
1771 /* Pre-constructed entries are in pfn, so convert to mfn */
1772 convert_pfn_mfn(init_level4_pgt);
1773 convert_pfn_mfn(level3_ident_pgt);
1774 convert_pfn_mfn(level3_kernel_pgt);
1776 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1777 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1779 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1780 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1782 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1783 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1784 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1786 /* Set up identity map */
1787 xen_map_identity_early(level2_ident_pgt, max_pfn);
1789 /* Make pagetable pieces RO */
1790 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1791 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1792 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1793 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1794 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1795 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1797 /* Pin down new L4 */
1798 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1799 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1801 /* Unpin Xen-provided one */
1802 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1805 pgd = init_level4_pgt;
1808 * At this stage there can be no user pgd, and no page
1809 * structure to attach it to, so make sure we just set kernel
1813 __xen_write_cr3(true, __pa(pgd));
1814 xen_mc_issue(PARAVIRT_LAZY_CPU);
1816 reserve_early(__pa(xen_start_info->pt_base),
1817 __pa(xen_start_info->pt_base +
1818 xen_start_info->nr_pt_frames * PAGE_SIZE),
1823 #else /* !CONFIG_X86_64 */
1824 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1826 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1827 unsigned long max_pfn)
1831 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1832 xen_start_info->nr_pt_frames * PAGE_SIZE +
1835 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1836 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1838 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1840 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1841 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1842 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1844 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1845 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1846 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1848 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1850 xen_write_cr3(__pa(swapper_pg_dir));
1852 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1854 reserve_early(__pa(xen_start_info->pt_base),
1855 __pa(xen_start_info->pt_base +
1856 xen_start_info->nr_pt_frames * PAGE_SIZE),
1859 return swapper_pg_dir;
1861 #endif /* CONFIG_X86_64 */
1863 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1867 phys >>= PAGE_SHIFT;
1870 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1871 #ifdef CONFIG_X86_F00F_BUG
1874 #ifdef CONFIG_X86_32
1877 # ifdef CONFIG_HIGHMEM
1878 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1881 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1883 #ifdef CONFIG_X86_LOCAL_APIC
1884 case FIX_APIC_BASE: /* maps dummy local APIC */
1886 case FIX_TEXT_POKE0:
1887 case FIX_TEXT_POKE1:
1888 /* All local page mappings */
1889 pte = pfn_pte(phys, prot);
1892 case FIX_PARAVIRT_BOOTMAP:
1893 /* This is an MFN, but it isn't an IO mapping from the
1895 pte = mfn_pte(phys, prot);
1899 /* By default, set_fixmap is used for hardware mappings */
1900 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1904 __native_set_fixmap(idx, pte);
1906 #ifdef CONFIG_X86_64
1907 /* Replicate changes to map the vsyscall page into the user
1908 pagetable vsyscall mapping. */
1909 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1910 unsigned long vaddr = __fix_to_virt(idx);
1911 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1916 static __init void xen_post_allocator_init(void)
1918 pv_mmu_ops.set_pte = xen_set_pte;
1919 pv_mmu_ops.set_pmd = xen_set_pmd;
1920 pv_mmu_ops.set_pud = xen_set_pud;
1921 #if PAGETABLE_LEVELS == 4
1922 pv_mmu_ops.set_pgd = xen_set_pgd;
1925 /* This will work as long as patching hasn't happened yet
1926 (which it hasn't) */
1927 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1928 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1929 pv_mmu_ops.release_pte = xen_release_pte;
1930 pv_mmu_ops.release_pmd = xen_release_pmd;
1931 #if PAGETABLE_LEVELS == 4
1932 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1933 pv_mmu_ops.release_pud = xen_release_pud;
1936 #ifdef CONFIG_X86_64
1937 SetPagePinned(virt_to_page(level3_user_vsyscall));
1939 xen_mark_init_mm_pinned();
1942 static void xen_leave_lazy_mmu(void)
1946 paravirt_leave_lazy_mmu();
1950 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1951 .read_cr2 = xen_read_cr2,
1952 .write_cr2 = xen_write_cr2,
1954 .read_cr3 = xen_read_cr3,
1955 .write_cr3 = xen_write_cr3,
1957 .flush_tlb_user = xen_flush_tlb,
1958 .flush_tlb_kernel = xen_flush_tlb,
1959 .flush_tlb_single = xen_flush_tlb_single,
1960 .flush_tlb_others = xen_flush_tlb_others,
1962 .pte_update = paravirt_nop,
1963 .pte_update_defer = paravirt_nop,
1965 .pgd_alloc = xen_pgd_alloc,
1966 .pgd_free = xen_pgd_free,
1968 .alloc_pte = xen_alloc_pte_init,
1969 .release_pte = xen_release_pte_init,
1970 .alloc_pmd = xen_alloc_pmd_init,
1971 .alloc_pmd_clone = paravirt_nop,
1972 .release_pmd = xen_release_pmd_init,
1974 #ifdef CONFIG_X86_64
1975 .set_pte = xen_set_pte,
1977 .set_pte = xen_set_pte_init,
1979 .set_pte_at = xen_set_pte_at,
1980 .set_pmd = xen_set_pmd_hyper,
1982 .ptep_modify_prot_start = __ptep_modify_prot_start,
1983 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1985 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1986 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1988 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1989 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1991 #ifdef CONFIG_X86_PAE
1992 .set_pte_atomic = xen_set_pte_atomic,
1993 .pte_clear = xen_pte_clear,
1994 .pmd_clear = xen_pmd_clear,
1995 #endif /* CONFIG_X86_PAE */
1996 .set_pud = xen_set_pud_hyper,
1998 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
1999 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2001 #if PAGETABLE_LEVELS == 4
2002 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2003 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2004 .set_pgd = xen_set_pgd_hyper,
2006 .alloc_pud = xen_alloc_pmd_init,
2007 .release_pud = xen_release_pmd_init,
2008 #endif /* PAGETABLE_LEVELS == 4 */
2010 .activate_mm = xen_activate_mm,
2011 .dup_mmap = xen_dup_mmap,
2012 .exit_mmap = xen_exit_mmap,
2015 .enter = paravirt_enter_lazy_mmu,
2016 .leave = xen_leave_lazy_mmu,
2019 .set_fixmap = xen_set_fixmap,
2022 void __init xen_init_mmu_ops(void)
2024 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2025 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2026 pv_mmu_ops = xen_mmu_ops;
2028 vmap_lazy_unmap = false;
2031 /* Protected by xen_reservation_lock. */
2032 #define MAX_CONTIG_ORDER 9 /* 2MB */
2033 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2035 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2036 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2037 unsigned long *in_frames,
2038 unsigned long *out_frames)
2041 struct multicall_space mcs;
2044 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2045 mcs = __xen_mc_entry(0);
2048 in_frames[i] = virt_to_mfn(vaddr);
2050 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2051 set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2054 out_frames[i] = virt_to_pfn(vaddr);
2060 * Update the pfn-to-mfn mappings for a virtual address range, either to
2061 * point to an array of mfns, or contiguously from a single starting
2064 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2065 unsigned long *mfns,
2066 unsigned long first_mfn)
2073 limit = 1u << order;
2074 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2075 struct multicall_space mcs;
2078 mcs = __xen_mc_entry(0);
2082 mfn = first_mfn + i;
2084 if (i < (limit - 1))
2088 flags = UVMF_INVLPG | UVMF_ALL;
2090 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2093 MULTI_update_va_mapping(mcs.mc, vaddr,
2094 mfn_pte(mfn, PAGE_KERNEL), flags);
2096 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2103 * Perform the hypercall to exchange a region of our pfns to point to
2104 * memory with the required contiguous alignment. Takes the pfns as
2105 * input, and populates mfns as output.
2107 * Returns a success code indicating whether the hypervisor was able to
2108 * satisfy the request or not.
2110 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2111 unsigned long *pfns_in,
2112 unsigned long extents_out,
2113 unsigned int order_out,
2114 unsigned long *mfns_out,
2115 unsigned int address_bits)
2120 struct xen_memory_exchange exchange = {
2122 .nr_extents = extents_in,
2123 .extent_order = order_in,
2124 .extent_start = pfns_in,
2128 .nr_extents = extents_out,
2129 .extent_order = order_out,
2130 .extent_start = mfns_out,
2131 .address_bits = address_bits,
2136 BUG_ON(extents_in << order_in != extents_out << order_out);
2138 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2139 success = (exchange.nr_exchanged == extents_in);
2141 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2142 BUG_ON(success && (rc != 0));
2147 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2148 unsigned int address_bits)
2150 unsigned long *in_frames = discontig_frames, out_frame;
2151 unsigned long flags;
2155 * Currently an auto-translated guest will not perform I/O, nor will
2156 * it require PAE page directories below 4GB. Therefore any calls to
2157 * this function are redundant and can be ignored.
2160 if (xen_feature(XENFEAT_auto_translated_physmap))
2163 if (unlikely(order > MAX_CONTIG_ORDER))
2166 memset((void *) vstart, 0, PAGE_SIZE << order);
2168 spin_lock_irqsave(&xen_reservation_lock, flags);
2170 /* 1. Zap current PTEs, remembering MFNs. */
2171 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2173 /* 2. Get a new contiguous memory extent. */
2174 out_frame = virt_to_pfn(vstart);
2175 success = xen_exchange_memory(1UL << order, 0, in_frames,
2176 1, order, &out_frame,
2179 /* 3. Map the new extent in place of old pages. */
2181 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2183 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2185 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2187 return success ? 0 : -ENOMEM;
2189 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2191 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2193 unsigned long *out_frames = discontig_frames, in_frame;
2194 unsigned long flags;
2197 if (xen_feature(XENFEAT_auto_translated_physmap))
2200 if (unlikely(order > MAX_CONTIG_ORDER))
2203 memset((void *) vstart, 0, PAGE_SIZE << order);
2205 spin_lock_irqsave(&xen_reservation_lock, flags);
2207 /* 1. Find start MFN of contiguous extent. */
2208 in_frame = virt_to_mfn(vstart);
2210 /* 2. Zap current PTEs. */
2211 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2213 /* 3. Do the exchange for non-contiguous MFNs. */
2214 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2217 /* 4. Map new pages in place of old pages. */
2219 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2221 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2223 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2225 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2227 #ifdef CONFIG_XEN_DEBUG_FS
2229 static struct dentry *d_mmu_debug;
2231 static int __init xen_mmu_debugfs(void)
2233 struct dentry *d_xen = xen_init_debugfs();
2238 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2240 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2242 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2243 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2244 &mmu_stats.pgd_update_pinned);
2245 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2246 &mmu_stats.pgd_update_pinned);
2248 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2249 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2250 &mmu_stats.pud_update_pinned);
2251 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2252 &mmu_stats.pud_update_pinned);
2254 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2255 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2256 &mmu_stats.pmd_update_pinned);
2257 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2258 &mmu_stats.pmd_update_pinned);
2260 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2261 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2262 // &mmu_stats.pte_update_pinned);
2263 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2264 &mmu_stats.pte_update_pinned);
2266 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2267 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2268 &mmu_stats.mmu_update_extended);
2269 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2270 mmu_stats.mmu_update_histo, 20);
2272 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2273 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2274 &mmu_stats.set_pte_at_batched);
2275 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2276 &mmu_stats.set_pte_at_current);
2277 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2278 &mmu_stats.set_pte_at_kernel);
2280 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2281 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2282 &mmu_stats.prot_commit_batched);
2286 fs_initcall(xen_mmu_debugfs);
2288 #endif /* CONFIG_XEN_DEBUG_FS */