4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/module.h>
47 #include <linux/gfp.h>
48 #include <linux/memblock.h>
50 #include <asm/pgtable.h>
51 #include <asm/tlbflush.h>
52 #include <asm/fixmap.h>
53 #include <asm/mmu_context.h>
54 #include <asm/setup.h>
55 #include <asm/paravirt.h>
57 #include <asm/linkage.h>
61 #include <asm/xen/hypercall.h>
62 #include <asm/xen/hypervisor.h>
66 #include <xen/interface/xen.h>
67 #include <xen/interface/hvm/hvm_op.h>
68 #include <xen/interface/version.h>
69 #include <xen/interface/memory.h>
70 #include <xen/hvc-console.h>
72 #include "multicalls.h"
76 #define MMU_UPDATE_HISTO 30
79 * Protects atomic reservation decrease/increase against concurrent increases.
80 * Also protects non-atomic updates of current_pages and driver_pages, and
83 DEFINE_SPINLOCK(xen_reservation_lock);
85 #ifdef CONFIG_XEN_DEBUG_FS
89 u32 pgd_update_pinned;
90 u32 pgd_update_batched;
93 u32 pud_update_pinned;
94 u32 pud_update_batched;
97 u32 pmd_update_pinned;
98 u32 pmd_update_batched;
101 u32 pte_update_pinned;
102 u32 pte_update_batched;
105 u32 mmu_update_extended;
106 u32 mmu_update_histo[MMU_UPDATE_HISTO];
109 u32 prot_commit_batched;
112 u32 set_pte_at_batched;
113 u32 set_pte_at_pinned;
114 u32 set_pte_at_current;
115 u32 set_pte_at_kernel;
118 static u8 zero_stats;
120 static inline void check_zero(void)
122 if (unlikely(zero_stats)) {
123 memset(&mmu_stats, 0, sizeof(mmu_stats));
128 #define ADD_STATS(elem, val) \
129 do { check_zero(); mmu_stats.elem += (val); } while(0)
131 #else /* !CONFIG_XEN_DEBUG_FS */
133 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
135 #endif /* CONFIG_XEN_DEBUG_FS */
139 * Identity map, in addition to plain kernel map. This needs to be
140 * large enough to allocate page table pages to allocate the rest.
141 * Each page can map 2MB.
143 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
146 /* l3 pud for userspace vsyscall mapping */
147 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
148 #endif /* CONFIG_X86_64 */
151 * Note about cr3 (pagetable base) values:
153 * xen_cr3 contains the current logical cr3 value; it contains the
154 * last set cr3. This may not be the current effective cr3, because
155 * its update may be being lazily deferred. However, a vcpu looking
156 * at its own cr3 can use this value knowing that it everything will
157 * be self-consistent.
159 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
160 * hypercall to set the vcpu cr3 is complete (so it may be a little
161 * out of date, but it will never be set early). If one vcpu is
162 * looking at another vcpu's cr3 value, it should use this variable.
164 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
165 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
169 * Just beyond the highest usermode address. STACK_TOP_MAX has a
170 * redzone above it, so round it up to a PGD boundary.
172 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
175 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
176 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
178 /* Placeholder for holes in the address space */
179 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
180 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
182 /* Array of pointers to pages containing p2m entries */
183 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
184 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
186 /* Arrays of p2m arrays expressed in mfns used for save/restore */
187 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
189 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
192 static inline unsigned p2m_top_index(unsigned long pfn)
194 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
195 return pfn / P2M_ENTRIES_PER_PAGE;
198 static inline unsigned p2m_index(unsigned long pfn)
200 return pfn % P2M_ENTRIES_PER_PAGE;
203 /* Build the parallel p2m_top_mfn structures */
204 void xen_build_mfn_list_list(void)
208 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
209 unsigned topidx = p2m_top_index(pfn);
211 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
214 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
215 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
216 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
220 void xen_setup_mfn_list_list(void)
222 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
224 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
225 virt_to_mfn(p2m_top_mfn_list);
226 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
229 /* Set up p2m_top to point to the domain-builder provided p2m pages */
230 void __init xen_build_dynamic_phys_to_machine(void)
232 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
233 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
236 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
237 unsigned topidx = p2m_top_index(pfn);
239 p2m_top[topidx] = &mfn_list[pfn];
242 xen_build_mfn_list_list();
245 unsigned long get_phys_to_machine(unsigned long pfn)
247 unsigned topidx, idx;
249 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
250 return INVALID_P2M_ENTRY;
252 topidx = p2m_top_index(pfn);
253 idx = p2m_index(pfn);
254 return p2m_top[topidx][idx];
256 EXPORT_SYMBOL_GPL(get_phys_to_machine);
258 /* install a new p2m_top page */
259 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
261 unsigned topidx = p2m_top_index(pfn);
262 unsigned long **pfnp, *mfnp;
265 pfnp = &p2m_top[topidx];
266 mfnp = &p2m_top_mfn[topidx];
268 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
269 p[i] = INVALID_P2M_ENTRY;
271 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
272 *mfnp = virt_to_mfn(p);
279 static void alloc_p2m(unsigned long pfn)
283 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
286 if (!install_p2mtop_page(pfn, p))
287 free_page((unsigned long)p);
290 /* Try to install p2m mapping; fail if intermediate bits missing */
291 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
293 unsigned topidx, idx;
295 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
296 BUG_ON(mfn != INVALID_P2M_ENTRY);
300 topidx = p2m_top_index(pfn);
301 if (p2m_top[topidx] == p2m_missing) {
302 if (mfn == INVALID_P2M_ENTRY)
307 idx = p2m_index(pfn);
308 p2m_top[topidx][idx] = mfn;
313 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
315 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
316 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
320 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
323 if (!__set_phys_to_machine(pfn, mfn))
328 unsigned long arbitrary_virt_to_mfn(void *vaddr)
330 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
332 return PFN_DOWN(maddr.maddr);
335 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
337 unsigned long address = (unsigned long)vaddr;
343 * if the PFN is in the linear mapped vaddr range, we can just use
344 * the (quick) virt_to_machine() p2m lookup
346 if (virt_addr_valid(vaddr))
347 return virt_to_machine(vaddr);
349 /* otherwise we have to do a (slower) full page-table walk */
351 pte = lookup_address(address, &level);
353 offset = address & ~PAGE_MASK;
354 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
357 void make_lowmem_page_readonly(void *vaddr)
360 unsigned long address = (unsigned long)vaddr;
363 pte = lookup_address(address, &level);
365 return; /* vaddr missing */
367 ptev = pte_wrprotect(*pte);
369 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
373 void make_lowmem_page_readwrite(void *vaddr)
376 unsigned long address = (unsigned long)vaddr;
379 pte = lookup_address(address, &level);
381 return; /* vaddr missing */
383 ptev = pte_mkwrite(*pte);
385 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
390 static bool xen_page_pinned(void *ptr)
392 struct page *page = virt_to_page(ptr);
394 return PagePinned(page);
397 static bool xen_iomap_pte(pte_t pte)
399 return pte_flags(pte) & _PAGE_IOMAP;
402 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
404 struct multicall_space mcs;
405 struct mmu_update *u;
407 mcs = xen_mc_entry(sizeof(*u));
410 /* ptep might be kmapped when using 32-bit HIGHPTE */
411 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
412 u->val = pte_val_ma(pteval);
414 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_IO);
416 xen_mc_issue(PARAVIRT_LAZY_MMU);
419 static void xen_extend_mmu_update(const struct mmu_update *update)
421 struct multicall_space mcs;
422 struct mmu_update *u;
424 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
426 if (mcs.mc != NULL) {
427 ADD_STATS(mmu_update_extended, 1);
428 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
432 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
433 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
435 ADD_STATS(mmu_update_histo[0], 1);
437 ADD_STATS(mmu_update, 1);
438 mcs = __xen_mc_entry(sizeof(*u));
439 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
440 ADD_STATS(mmu_update_histo[1], 1);
447 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
455 /* ptr may be ioremapped for 64-bit pagetable setup */
456 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
457 u.val = pmd_val_ma(val);
458 xen_extend_mmu_update(&u);
460 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
462 xen_mc_issue(PARAVIRT_LAZY_MMU);
467 void xen_set_pmd(pmd_t *ptr, pmd_t val)
469 ADD_STATS(pmd_update, 1);
471 /* If page is not pinned, we can just update the entry
473 if (!xen_page_pinned(ptr)) {
478 ADD_STATS(pmd_update_pinned, 1);
480 xen_set_pmd_hyper(ptr, val);
484 * Associate a virtual page frame with a given physical page frame
485 * and protection flags for that frame.
487 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
489 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
492 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
493 pte_t *ptep, pte_t pteval)
495 if (xen_iomap_pte(pteval)) {
496 xen_set_iomap_pte(ptep, pteval);
500 ADD_STATS(set_pte_at, 1);
501 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
502 ADD_STATS(set_pte_at_current, mm == current->mm);
503 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
505 if (mm == current->mm || mm == &init_mm) {
506 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
507 struct multicall_space mcs;
508 mcs = xen_mc_entry(0);
510 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
511 ADD_STATS(set_pte_at_batched, 1);
512 xen_mc_issue(PARAVIRT_LAZY_MMU);
515 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
518 xen_set_pte(ptep, pteval);
523 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
524 unsigned long addr, pte_t *ptep)
526 /* Just return the pte as-is. We preserve the bits on commit */
530 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
531 pte_t *ptep, pte_t pte)
537 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
538 u.val = pte_val_ma(pte);
539 xen_extend_mmu_update(&u);
541 ADD_STATS(prot_commit, 1);
542 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
544 xen_mc_issue(PARAVIRT_LAZY_MMU);
547 /* Assume pteval_t is equivalent to all the other *val_t types. */
548 static pteval_t pte_mfn_to_pfn(pteval_t val)
550 if (val & _PAGE_PRESENT) {
551 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
552 pteval_t flags = val & PTE_FLAGS_MASK;
553 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
559 static pteval_t pte_pfn_to_mfn(pteval_t val)
561 if (val & _PAGE_PRESENT) {
562 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
563 pteval_t flags = val & PTE_FLAGS_MASK;
564 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
570 static pteval_t iomap_pte(pteval_t val)
572 if (val & _PAGE_PRESENT) {
573 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
574 pteval_t flags = val & PTE_FLAGS_MASK;
576 /* We assume the pte frame number is a MFN, so
577 just use it as-is. */
578 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
584 pteval_t xen_pte_val(pte_t pte)
586 if (xen_initial_domain() && (pte.pte & _PAGE_IOMAP))
589 return pte_mfn_to_pfn(pte.pte);
591 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
593 pgdval_t xen_pgd_val(pgd_t pgd)
595 return pte_mfn_to_pfn(pgd.pgd);
597 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
599 pte_t xen_make_pte(pteval_t pte)
601 phys_addr_t addr = (pte & PTE_PFN_MASK);
604 * Unprivileged domains are allowed to do IOMAPpings for
605 * PCI passthrough, but not map ISA space. The ISA
606 * mappings are just dummy local mappings to keep other
607 * parts of the kernel happy.
609 if (unlikely(pte & _PAGE_IOMAP) &&
610 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
611 pte = iomap_pte(pte);
614 pte = pte_pfn_to_mfn(pte);
617 return native_make_pte(pte);
619 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
621 pgd_t xen_make_pgd(pgdval_t pgd)
623 pgd = pte_pfn_to_mfn(pgd);
624 return native_make_pgd(pgd);
626 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
628 pmdval_t xen_pmd_val(pmd_t pmd)
630 return pte_mfn_to_pfn(pmd.pmd);
632 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
634 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
642 /* ptr may be ioremapped for 64-bit pagetable setup */
643 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
644 u.val = pud_val_ma(val);
645 xen_extend_mmu_update(&u);
647 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
649 xen_mc_issue(PARAVIRT_LAZY_MMU);
654 void xen_set_pud(pud_t *ptr, pud_t val)
656 ADD_STATS(pud_update, 1);
658 /* If page is not pinned, we can just update the entry
660 if (!xen_page_pinned(ptr)) {
665 ADD_STATS(pud_update_pinned, 1);
667 xen_set_pud_hyper(ptr, val);
670 void xen_set_pte(pte_t *ptep, pte_t pte)
672 if (xen_iomap_pte(pte)) {
673 xen_set_iomap_pte(ptep, pte);
677 ADD_STATS(pte_update, 1);
678 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
679 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
681 #ifdef CONFIG_X86_PAE
682 ptep->pte_high = pte.pte_high;
684 ptep->pte_low = pte.pte_low;
690 #ifdef CONFIG_X86_PAE
691 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
693 if (xen_iomap_pte(pte)) {
694 xen_set_iomap_pte(ptep, pte);
698 set_64bit((u64 *)ptep, native_pte_val(pte));
701 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
704 smp_wmb(); /* make sure low gets written first */
708 void xen_pmd_clear(pmd_t *pmdp)
710 set_pmd(pmdp, __pmd(0));
712 #endif /* CONFIG_X86_PAE */
714 pmd_t xen_make_pmd(pmdval_t pmd)
716 pmd = pte_pfn_to_mfn(pmd);
717 return native_make_pmd(pmd);
719 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
721 #if PAGETABLE_LEVELS == 4
722 pudval_t xen_pud_val(pud_t pud)
724 return pte_mfn_to_pfn(pud.pud);
726 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
728 pud_t xen_make_pud(pudval_t pud)
730 pud = pte_pfn_to_mfn(pud);
732 return native_make_pud(pud);
734 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
736 pgd_t *xen_get_user_pgd(pgd_t *pgd)
738 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
739 unsigned offset = pgd - pgd_page;
740 pgd_t *user_ptr = NULL;
742 if (offset < pgd_index(USER_LIMIT)) {
743 struct page *page = virt_to_page(pgd_page);
744 user_ptr = (pgd_t *)page->private;
752 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
756 u.ptr = virt_to_machine(ptr).maddr;
757 u.val = pgd_val_ma(val);
758 xen_extend_mmu_update(&u);
762 * Raw hypercall-based set_pgd, intended for in early boot before
763 * there's a page structure. This implies:
764 * 1. The only existing pagetable is the kernel's
765 * 2. It is always pinned
766 * 3. It has no user pagetable attached to it
768 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
774 __xen_set_pgd_hyper(ptr, val);
776 xen_mc_issue(PARAVIRT_LAZY_MMU);
781 void xen_set_pgd(pgd_t *ptr, pgd_t val)
783 pgd_t *user_ptr = xen_get_user_pgd(ptr);
785 ADD_STATS(pgd_update, 1);
787 /* If page is not pinned, we can just update the entry
789 if (!xen_page_pinned(ptr)) {
792 WARN_ON(xen_page_pinned(user_ptr));
798 ADD_STATS(pgd_update_pinned, 1);
799 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
801 /* If it's pinned, then we can at least batch the kernel and
802 user updates together. */
805 __xen_set_pgd_hyper(ptr, val);
807 __xen_set_pgd_hyper(user_ptr, val);
809 xen_mc_issue(PARAVIRT_LAZY_MMU);
811 #endif /* PAGETABLE_LEVELS == 4 */
814 * (Yet another) pagetable walker. This one is intended for pinning a
815 * pagetable. This means that it walks a pagetable and calls the
816 * callback function on each page it finds making up the page table,
817 * at every level. It walks the entire pagetable, but it only bothers
818 * pinning pte pages which are below limit. In the normal case this
819 * will be STACK_TOP_MAX, but at boot we need to pin up to
822 * For 32-bit the important bit is that we don't pin beyond there,
823 * because then we start getting into Xen's ptes.
825 * For 64-bit, we must skip the Xen hole in the middle of the address
826 * space, just after the big x86-64 virtual hole.
828 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
829 int (*func)(struct mm_struct *mm, struct page *,
834 unsigned hole_low, hole_high;
835 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
836 unsigned pgdidx, pudidx, pmdidx;
838 /* The limit is the last byte to be touched */
840 BUG_ON(limit >= FIXADDR_TOP);
842 if (xen_feature(XENFEAT_auto_translated_physmap))
846 * 64-bit has a great big hole in the middle of the address
847 * space, which contains the Xen mappings. On 32-bit these
848 * will end up making a zero-sized hole and so is a no-op.
850 hole_low = pgd_index(USER_LIMIT);
851 hole_high = pgd_index(PAGE_OFFSET);
853 pgdidx_limit = pgd_index(limit);
855 pudidx_limit = pud_index(limit);
860 pmdidx_limit = pmd_index(limit);
865 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
868 if (pgdidx >= hole_low && pgdidx < hole_high)
871 if (!pgd_val(pgd[pgdidx]))
874 pud = pud_offset(&pgd[pgdidx], 0);
876 if (PTRS_PER_PUD > 1) /* not folded */
877 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
879 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
882 if (pgdidx == pgdidx_limit &&
883 pudidx > pudidx_limit)
886 if (pud_none(pud[pudidx]))
889 pmd = pmd_offset(&pud[pudidx], 0);
891 if (PTRS_PER_PMD > 1) /* not folded */
892 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
894 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
897 if (pgdidx == pgdidx_limit &&
898 pudidx == pudidx_limit &&
899 pmdidx > pmdidx_limit)
902 if (pmd_none(pmd[pmdidx]))
905 pte = pmd_page(pmd[pmdidx]);
906 flush |= (*func)(mm, pte, PT_PTE);
912 /* Do the top level last, so that the callbacks can use it as
913 a cue to do final things like tlb flushes. */
914 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
919 static int xen_pgd_walk(struct mm_struct *mm,
920 int (*func)(struct mm_struct *mm, struct page *,
924 return __xen_pgd_walk(mm, mm->pgd, func, limit);
927 /* If we're using split pte locks, then take the page's lock and
928 return a pointer to it. Otherwise return NULL. */
929 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
931 spinlock_t *ptl = NULL;
933 #if USE_SPLIT_PTLOCKS
934 ptl = __pte_lockptr(page);
935 spin_lock_nest_lock(ptl, &mm->page_table_lock);
941 static void xen_pte_unlock(void *v)
947 static void xen_do_pin(unsigned level, unsigned long pfn)
949 struct mmuext_op *op;
950 struct multicall_space mcs;
952 mcs = __xen_mc_entry(sizeof(*op));
955 op->arg1.mfn = pfn_to_mfn(pfn);
956 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
959 static int xen_pin_page(struct mm_struct *mm, struct page *page,
962 unsigned pgfl = TestSetPagePinned(page);
966 flush = 0; /* already pinned */
967 else if (PageHighMem(page))
968 /* kmaps need flushing if we found an unpinned
972 void *pt = lowmem_page_address(page);
973 unsigned long pfn = page_to_pfn(page);
974 struct multicall_space mcs = __xen_mc_entry(0);
980 * We need to hold the pagetable lock between the time
981 * we make the pagetable RO and when we actually pin
982 * it. If we don't, then other users may come in and
983 * attempt to update the pagetable by writing it,
984 * which will fail because the memory is RO but not
985 * pinned, so Xen won't do the trap'n'emulate.
987 * If we're using split pte locks, we can't hold the
988 * entire pagetable's worth of locks during the
989 * traverse, because we may wrap the preempt count (8
990 * bits). The solution is to mark RO and pin each PTE
991 * page while holding the lock. This means the number
992 * of locks we end up holding is never more than a
993 * batch size (~32 entries, at present).
995 * If we're not using split pte locks, we needn't pin
996 * the PTE pages independently, because we're
997 * protected by the overall pagetable lock.
1000 if (level == PT_PTE)
1001 ptl = xen_pte_lock(page, mm);
1003 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1004 pfn_pte(pfn, PAGE_KERNEL_RO),
1005 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1008 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
1010 /* Queue a deferred unlock for when this batch
1012 xen_mc_callback(xen_pte_unlock, ptl);
1019 /* This is called just after a mm has been created, but it has not
1020 been used yet. We need to make sure that its pagetable is all
1021 read-only, and can be pinned. */
1022 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
1026 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
1027 /* re-enable interrupts for flushing */
1030 kmap_flush_unused();
1035 #ifdef CONFIG_X86_64
1037 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1039 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
1042 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
1043 xen_do_pin(MMUEXT_PIN_L4_TABLE,
1044 PFN_DOWN(__pa(user_pgd)));
1047 #else /* CONFIG_X86_32 */
1048 #ifdef CONFIG_X86_PAE
1049 /* Need to make sure unshared kernel PMD is pinnable */
1050 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1053 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
1054 #endif /* CONFIG_X86_64 */
1058 static void xen_pgd_pin(struct mm_struct *mm)
1060 __xen_pgd_pin(mm, mm->pgd);
1064 * On save, we need to pin all pagetables to make sure they get their
1065 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1066 * them (unpinned pgds are not currently in use, probably because the
1067 * process is under construction or destruction).
1069 * Expected to be called in stop_machine() ("equivalent to taking
1070 * every spinlock in the system"), so the locking doesn't really
1071 * matter all that much.
1073 void xen_mm_pin_all(void)
1075 unsigned long flags;
1078 spin_lock_irqsave(&pgd_lock, flags);
1080 list_for_each_entry(page, &pgd_list, lru) {
1081 if (!PagePinned(page)) {
1082 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1083 SetPageSavePinned(page);
1087 spin_unlock_irqrestore(&pgd_lock, flags);
1091 * The init_mm pagetable is really pinned as soon as its created, but
1092 * that's before we have page structures to store the bits. So do all
1093 * the book-keeping now.
1095 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1096 enum pt_level level)
1098 SetPagePinned(page);
1102 static void __init xen_mark_init_mm_pinned(void)
1104 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1107 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1108 enum pt_level level)
1110 unsigned pgfl = TestClearPagePinned(page);
1112 if (pgfl && !PageHighMem(page)) {
1113 void *pt = lowmem_page_address(page);
1114 unsigned long pfn = page_to_pfn(page);
1115 spinlock_t *ptl = NULL;
1116 struct multicall_space mcs;
1119 * Do the converse to pin_page. If we're using split
1120 * pte locks, we must be holding the lock for while
1121 * the pte page is unpinned but still RO to prevent
1122 * concurrent updates from seeing it in this
1123 * partially-pinned state.
1125 if (level == PT_PTE) {
1126 ptl = xen_pte_lock(page, mm);
1129 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1132 mcs = __xen_mc_entry(0);
1134 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1135 pfn_pte(pfn, PAGE_KERNEL),
1136 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1139 /* unlock when batch completed */
1140 xen_mc_callback(xen_pte_unlock, ptl);
1144 return 0; /* never need to flush on unpin */
1147 /* Release a pagetables pages back as normal RW */
1148 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1152 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1154 #ifdef CONFIG_X86_64
1156 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1159 xen_do_pin(MMUEXT_UNPIN_TABLE,
1160 PFN_DOWN(__pa(user_pgd)));
1161 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1166 #ifdef CONFIG_X86_PAE
1167 /* Need to make sure unshared kernel PMD is unpinned */
1168 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1172 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1177 static void xen_pgd_unpin(struct mm_struct *mm)
1179 __xen_pgd_unpin(mm, mm->pgd);
1183 * On resume, undo any pinning done at save, so that the rest of the
1184 * kernel doesn't see any unexpected pinned pagetables.
1186 void xen_mm_unpin_all(void)
1188 unsigned long flags;
1191 spin_lock_irqsave(&pgd_lock, flags);
1193 list_for_each_entry(page, &pgd_list, lru) {
1194 if (PageSavePinned(page)) {
1195 BUG_ON(!PagePinned(page));
1196 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1197 ClearPageSavePinned(page);
1201 spin_unlock_irqrestore(&pgd_lock, flags);
1204 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1206 spin_lock(&next->page_table_lock);
1208 spin_unlock(&next->page_table_lock);
1211 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1213 spin_lock(&mm->page_table_lock);
1215 spin_unlock(&mm->page_table_lock);
1220 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1221 we need to repoint it somewhere else before we can unpin it. */
1222 static void drop_other_mm_ref(void *info)
1224 struct mm_struct *mm = info;
1225 struct mm_struct *active_mm;
1227 active_mm = percpu_read(cpu_tlbstate.active_mm);
1229 if (active_mm == mm)
1230 leave_mm(smp_processor_id());
1232 /* If this cpu still has a stale cr3 reference, then make sure
1233 it has been flushed. */
1234 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1235 load_cr3(swapper_pg_dir);
1238 static void xen_drop_mm_ref(struct mm_struct *mm)
1243 if (current->active_mm == mm) {
1244 if (current->mm == mm)
1245 load_cr3(swapper_pg_dir);
1247 leave_mm(smp_processor_id());
1250 /* Get the "official" set of cpus referring to our pagetable. */
1251 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1252 for_each_online_cpu(cpu) {
1253 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1254 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1256 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1260 cpumask_copy(mask, mm_cpumask(mm));
1262 /* It's possible that a vcpu may have a stale reference to our
1263 cr3, because its in lazy mode, and it hasn't yet flushed
1264 its set of pending hypercalls yet. In this case, we can
1265 look at its actual current cr3 value, and force it to flush
1267 for_each_online_cpu(cpu) {
1268 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1269 cpumask_set_cpu(cpu, mask);
1272 if (!cpumask_empty(mask))
1273 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1274 free_cpumask_var(mask);
1277 static void xen_drop_mm_ref(struct mm_struct *mm)
1279 if (current->active_mm == mm)
1280 load_cr3(swapper_pg_dir);
1285 * While a process runs, Xen pins its pagetables, which means that the
1286 * hypervisor forces it to be read-only, and it controls all updates
1287 * to it. This means that all pagetable updates have to go via the
1288 * hypervisor, which is moderately expensive.
1290 * Since we're pulling the pagetable down, we switch to use init_mm,
1291 * unpin old process pagetable and mark it all read-write, which
1292 * allows further operations on it to be simple memory accesses.
1294 * The only subtle point is that another CPU may be still using the
1295 * pagetable because of lazy tlb flushing. This means we need need to
1296 * switch all CPUs off this pagetable before we can unpin it.
1298 void xen_exit_mmap(struct mm_struct *mm)
1300 get_cpu(); /* make sure we don't move around */
1301 xen_drop_mm_ref(mm);
1304 spin_lock(&mm->page_table_lock);
1306 /* pgd may not be pinned in the error exit path of execve */
1307 if (xen_page_pinned(mm->pgd))
1310 spin_unlock(&mm->page_table_lock);
1313 static __init void xen_pagetable_setup_start(pgd_t *base)
1317 static void xen_post_allocator_init(void);
1319 static __init void xen_pagetable_setup_done(pgd_t *base)
1321 xen_setup_shared_info();
1322 xen_post_allocator_init();
1325 static void xen_write_cr2(unsigned long cr2)
1327 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1330 static unsigned long xen_read_cr2(void)
1332 return percpu_read(xen_vcpu)->arch.cr2;
1335 unsigned long xen_read_cr2_direct(void)
1337 return percpu_read(xen_vcpu_info.arch.cr2);
1340 static void xen_flush_tlb(void)
1342 struct mmuext_op *op;
1343 struct multicall_space mcs;
1347 mcs = xen_mc_entry(sizeof(*op));
1350 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1351 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1353 xen_mc_issue(PARAVIRT_LAZY_MMU);
1358 static void xen_flush_tlb_single(unsigned long addr)
1360 struct mmuext_op *op;
1361 struct multicall_space mcs;
1365 mcs = xen_mc_entry(sizeof(*op));
1367 op->cmd = MMUEXT_INVLPG_LOCAL;
1368 op->arg1.linear_addr = addr & PAGE_MASK;
1369 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1371 xen_mc_issue(PARAVIRT_LAZY_MMU);
1376 static void xen_flush_tlb_others(const struct cpumask *cpus,
1377 struct mm_struct *mm, unsigned long va)
1380 struct mmuext_op op;
1381 DECLARE_BITMAP(mask, NR_CPUS);
1383 struct multicall_space mcs;
1385 if (cpumask_empty(cpus))
1386 return; /* nothing to do */
1388 mcs = xen_mc_entry(sizeof(*args));
1390 args->op.arg2.vcpumask = to_cpumask(args->mask);
1392 /* Remove us, and any offline CPUS. */
1393 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1394 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1396 if (va == TLB_FLUSH_ALL) {
1397 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1399 args->op.cmd = MMUEXT_INVLPG_MULTI;
1400 args->op.arg1.linear_addr = va;
1403 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1405 xen_mc_issue(PARAVIRT_LAZY_MMU);
1408 static unsigned long xen_read_cr3(void)
1410 return percpu_read(xen_cr3);
1413 static void set_current_cr3(void *v)
1415 percpu_write(xen_current_cr3, (unsigned long)v);
1418 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1420 struct mmuext_op *op;
1421 struct multicall_space mcs;
1425 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1429 WARN_ON(mfn == 0 && kernel);
1431 mcs = __xen_mc_entry(sizeof(*op));
1434 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1437 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1440 percpu_write(xen_cr3, cr3);
1442 /* Update xen_current_cr3 once the batch has actually
1444 xen_mc_callback(set_current_cr3, (void *)cr3);
1448 static void xen_write_cr3(unsigned long cr3)
1450 BUG_ON(preemptible());
1452 xen_mc_batch(); /* disables interrupts */
1454 /* Update while interrupts are disabled, so its atomic with
1456 percpu_write(xen_cr3, cr3);
1458 __xen_write_cr3(true, cr3);
1460 #ifdef CONFIG_X86_64
1462 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1464 __xen_write_cr3(false, __pa(user_pgd));
1466 __xen_write_cr3(false, 0);
1470 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1473 static int xen_pgd_alloc(struct mm_struct *mm)
1475 pgd_t *pgd = mm->pgd;
1478 BUG_ON(PagePinned(virt_to_page(pgd)));
1480 #ifdef CONFIG_X86_64
1482 struct page *page = virt_to_page(pgd);
1485 BUG_ON(page->private != 0);
1489 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1490 page->private = (unsigned long)user_pgd;
1492 if (user_pgd != NULL) {
1493 user_pgd[pgd_index(VSYSCALL_START)] =
1494 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1498 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1505 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1507 #ifdef CONFIG_X86_64
1508 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1511 free_page((unsigned long)user_pgd);
1515 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1517 unsigned long pfn = pte_pfn(pte);
1519 #ifdef CONFIG_X86_32
1520 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1521 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1522 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1527 * If the new pfn is within the range of the newly allocated
1528 * kernel pagetable, and it isn't being mapped into an
1529 * early_ioremap fixmap slot, make sure it is RO.
1531 if (!is_early_ioremap_ptep(ptep) &&
1532 pfn >= e820_table_start && pfn < e820_table_end)
1533 pte = pte_wrprotect(pte);
1538 /* Init-time set_pte while constructing initial pagetables, which
1539 doesn't allow RO pagetable pages to be remapped RW */
1540 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1542 pte = mask_rw_pte(ptep, pte);
1544 xen_set_pte(ptep, pte);
1547 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1549 struct mmuext_op op;
1551 op.arg1.mfn = pfn_to_mfn(pfn);
1552 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1556 /* Early in boot, while setting up the initial pagetable, assume
1557 everything is pinned. */
1558 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1560 #ifdef CONFIG_FLATMEM
1561 BUG_ON(mem_map); /* should only be used early */
1563 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1564 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1567 /* Used for pmd and pud */
1568 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1570 #ifdef CONFIG_FLATMEM
1571 BUG_ON(mem_map); /* should only be used early */
1573 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1576 /* Early release_pte assumes that all pts are pinned, since there's
1577 only init_mm and anything attached to that is pinned. */
1578 static __init void xen_release_pte_init(unsigned long pfn)
1580 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1581 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1584 static __init void xen_release_pmd_init(unsigned long pfn)
1586 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1589 /* This needs to make sure the new pte page is pinned iff its being
1590 attached to a pinned pagetable. */
1591 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1593 struct page *page = pfn_to_page(pfn);
1595 if (PagePinned(virt_to_page(mm->pgd))) {
1596 SetPagePinned(page);
1598 if (!PageHighMem(page)) {
1599 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1600 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1601 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1603 /* make sure there are no stray mappings of
1605 kmap_flush_unused();
1610 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1612 xen_alloc_ptpage(mm, pfn, PT_PTE);
1615 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1617 xen_alloc_ptpage(mm, pfn, PT_PMD);
1620 /* This should never happen until we're OK to use struct page */
1621 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1623 struct page *page = pfn_to_page(pfn);
1625 if (PagePinned(page)) {
1626 if (!PageHighMem(page)) {
1627 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1628 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1629 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1631 ClearPagePinned(page);
1635 static void xen_release_pte(unsigned long pfn)
1637 xen_release_ptpage(pfn, PT_PTE);
1640 static void xen_release_pmd(unsigned long pfn)
1642 xen_release_ptpage(pfn, PT_PMD);
1645 #if PAGETABLE_LEVELS == 4
1646 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1648 xen_alloc_ptpage(mm, pfn, PT_PUD);
1651 static void xen_release_pud(unsigned long pfn)
1653 xen_release_ptpage(pfn, PT_PUD);
1657 void __init xen_reserve_top(void)
1659 #ifdef CONFIG_X86_32
1660 unsigned long top = HYPERVISOR_VIRT_START;
1661 struct xen_platform_parameters pp;
1663 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1664 top = pp.virt_start;
1666 reserve_top_address(-top);
1667 #endif /* CONFIG_X86_32 */
1671 * Like __va(), but returns address in the kernel mapping (which is
1672 * all we have until the physical memory mapping has been set up.
1674 static void *__ka(phys_addr_t paddr)
1676 #ifdef CONFIG_X86_64
1677 return (void *)(paddr + __START_KERNEL_map);
1683 /* Convert a machine address to physical address */
1684 static unsigned long m2p(phys_addr_t maddr)
1688 maddr &= PTE_PFN_MASK;
1689 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1694 /* Convert a machine address to kernel virtual */
1695 static void *m2v(phys_addr_t maddr)
1697 return __ka(m2p(maddr));
1700 static void set_page_prot(void *addr, pgprot_t prot)
1702 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1703 pte_t pte = pfn_pte(pfn, prot);
1705 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1709 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1711 unsigned pmdidx, pteidx;
1717 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1720 /* Reuse or allocate a page of ptes */
1721 if (pmd_present(pmd[pmdidx]))
1722 pte_page = m2v(pmd[pmdidx].pmd);
1724 /* Check for free pte pages */
1725 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1728 pte_page = &level1_ident_pgt[ident_pte];
1729 ident_pte += PTRS_PER_PTE;
1731 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1734 /* Install mappings */
1735 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1738 if (pfn > max_pfn_mapped)
1739 max_pfn_mapped = pfn;
1741 if (!pte_none(pte_page[pteidx]))
1744 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1745 pte_page[pteidx] = pte;
1749 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1750 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1752 set_page_prot(pmd, PAGE_KERNEL_RO);
1755 #ifdef CONFIG_X86_64
1756 static void convert_pfn_mfn(void *v)
1761 /* All levels are converted the same way, so just treat them
1763 for (i = 0; i < PTRS_PER_PTE; i++)
1764 pte[i] = xen_make_pte(pte[i].pte);
1768 * Set up the inital kernel pagetable.
1770 * We can construct this by grafting the Xen provided pagetable into
1771 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1772 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1773 * means that only the kernel has a physical mapping to start with -
1774 * but that's enough to get __va working. We need to fill in the rest
1775 * of the physical mapping once some sort of allocator has been set
1778 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1779 unsigned long max_pfn)
1784 /* Zap identity mapping */
1785 init_level4_pgt[0] = __pgd(0);
1787 /* Pre-constructed entries are in pfn, so convert to mfn */
1788 convert_pfn_mfn(init_level4_pgt);
1789 convert_pfn_mfn(level3_ident_pgt);
1790 convert_pfn_mfn(level3_kernel_pgt);
1792 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1793 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1795 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1796 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1798 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1799 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1800 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1802 /* Set up identity map */
1803 xen_map_identity_early(level2_ident_pgt, max_pfn);
1805 /* Make pagetable pieces RO */
1806 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1807 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1808 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1809 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1810 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1811 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1813 /* Pin down new L4 */
1814 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1815 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1817 /* Unpin Xen-provided one */
1818 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1821 pgd = init_level4_pgt;
1824 * At this stage there can be no user pgd, and no page
1825 * structure to attach it to, so make sure we just set kernel
1829 __xen_write_cr3(true, __pa(pgd));
1830 xen_mc_issue(PARAVIRT_LAZY_CPU);
1832 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1833 __pa(xen_start_info->pt_base +
1834 xen_start_info->nr_pt_frames * PAGE_SIZE),
1839 #else /* !CONFIG_X86_64 */
1840 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1842 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1843 unsigned long max_pfn)
1847 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1848 xen_start_info->nr_pt_frames * PAGE_SIZE +
1851 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1852 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1854 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1856 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1857 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1858 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1860 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1861 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1862 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1864 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1866 xen_write_cr3(__pa(swapper_pg_dir));
1868 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1870 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1871 __pa(xen_start_info->pt_base +
1872 xen_start_info->nr_pt_frames * PAGE_SIZE),
1875 return swapper_pg_dir;
1877 #endif /* CONFIG_X86_64 */
1879 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1883 phys >>= PAGE_SHIFT;
1886 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1887 #ifdef CONFIG_X86_F00F_BUG
1890 #ifdef CONFIG_X86_32
1893 # ifdef CONFIG_HIGHMEM
1894 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1897 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1899 #ifdef CONFIG_X86_LOCAL_APIC
1900 case FIX_APIC_BASE: /* maps dummy local APIC */
1902 case FIX_TEXT_POKE0:
1903 case FIX_TEXT_POKE1:
1904 /* All local page mappings */
1905 pte = pfn_pte(phys, prot);
1908 case FIX_PARAVIRT_BOOTMAP:
1909 /* This is an MFN, but it isn't an IO mapping from the
1911 pte = mfn_pte(phys, prot);
1915 /* By default, set_fixmap is used for hardware mappings */
1916 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1920 __native_set_fixmap(idx, pte);
1922 #ifdef CONFIG_X86_64
1923 /* Replicate changes to map the vsyscall page into the user
1924 pagetable vsyscall mapping. */
1925 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1926 unsigned long vaddr = __fix_to_virt(idx);
1927 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1932 static __init void xen_post_allocator_init(void)
1934 pv_mmu_ops.set_pte = xen_set_pte;
1935 pv_mmu_ops.set_pmd = xen_set_pmd;
1936 pv_mmu_ops.set_pud = xen_set_pud;
1937 #if PAGETABLE_LEVELS == 4
1938 pv_mmu_ops.set_pgd = xen_set_pgd;
1941 /* This will work as long as patching hasn't happened yet
1942 (which it hasn't) */
1943 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1944 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1945 pv_mmu_ops.release_pte = xen_release_pte;
1946 pv_mmu_ops.release_pmd = xen_release_pmd;
1947 #if PAGETABLE_LEVELS == 4
1948 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1949 pv_mmu_ops.release_pud = xen_release_pud;
1952 #ifdef CONFIG_X86_64
1953 SetPagePinned(virt_to_page(level3_user_vsyscall));
1955 xen_mark_init_mm_pinned();
1958 static void xen_leave_lazy_mmu(void)
1962 paravirt_leave_lazy_mmu();
1966 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1967 .read_cr2 = xen_read_cr2,
1968 .write_cr2 = xen_write_cr2,
1970 .read_cr3 = xen_read_cr3,
1971 .write_cr3 = xen_write_cr3,
1973 .flush_tlb_user = xen_flush_tlb,
1974 .flush_tlb_kernel = xen_flush_tlb,
1975 .flush_tlb_single = xen_flush_tlb_single,
1976 .flush_tlb_others = xen_flush_tlb_others,
1978 .pte_update = paravirt_nop,
1979 .pte_update_defer = paravirt_nop,
1981 .pgd_alloc = xen_pgd_alloc,
1982 .pgd_free = xen_pgd_free,
1984 .alloc_pte = xen_alloc_pte_init,
1985 .release_pte = xen_release_pte_init,
1986 .alloc_pmd = xen_alloc_pmd_init,
1987 .release_pmd = xen_release_pmd_init,
1989 .set_pte = xen_set_pte_init,
1990 .set_pte_at = xen_set_pte_at,
1991 .set_pmd = xen_set_pmd_hyper,
1993 .ptep_modify_prot_start = __ptep_modify_prot_start,
1994 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1996 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1997 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1999 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2000 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2002 #ifdef CONFIG_X86_PAE
2003 .set_pte_atomic = xen_set_pte_atomic,
2004 .pte_clear = xen_pte_clear,
2005 .pmd_clear = xen_pmd_clear,
2006 #endif /* CONFIG_X86_PAE */
2007 .set_pud = xen_set_pud_hyper,
2009 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2010 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2012 #if PAGETABLE_LEVELS == 4
2013 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2014 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2015 .set_pgd = xen_set_pgd_hyper,
2017 .alloc_pud = xen_alloc_pmd_init,
2018 .release_pud = xen_release_pmd_init,
2019 #endif /* PAGETABLE_LEVELS == 4 */
2021 .activate_mm = xen_activate_mm,
2022 .dup_mmap = xen_dup_mmap,
2023 .exit_mmap = xen_exit_mmap,
2026 .enter = paravirt_enter_lazy_mmu,
2027 .leave = xen_leave_lazy_mmu,
2030 .set_fixmap = xen_set_fixmap,
2033 void __init xen_init_mmu_ops(void)
2035 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2036 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2037 pv_mmu_ops = xen_mmu_ops;
2039 vmap_lazy_unmap = false;
2042 /* Protected by xen_reservation_lock. */
2043 #define MAX_CONTIG_ORDER 9 /* 2MB */
2044 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2046 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2047 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2048 unsigned long *in_frames,
2049 unsigned long *out_frames)
2052 struct multicall_space mcs;
2055 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2056 mcs = __xen_mc_entry(0);
2059 in_frames[i] = virt_to_mfn(vaddr);
2061 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2062 set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2065 out_frames[i] = virt_to_pfn(vaddr);
2071 * Update the pfn-to-mfn mappings for a virtual address range, either to
2072 * point to an array of mfns, or contiguously from a single starting
2075 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2076 unsigned long *mfns,
2077 unsigned long first_mfn)
2084 limit = 1u << order;
2085 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2086 struct multicall_space mcs;
2089 mcs = __xen_mc_entry(0);
2093 mfn = first_mfn + i;
2095 if (i < (limit - 1))
2099 flags = UVMF_INVLPG | UVMF_ALL;
2101 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2104 MULTI_update_va_mapping(mcs.mc, vaddr,
2105 mfn_pte(mfn, PAGE_KERNEL), flags);
2107 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2114 * Perform the hypercall to exchange a region of our pfns to point to
2115 * memory with the required contiguous alignment. Takes the pfns as
2116 * input, and populates mfns as output.
2118 * Returns a success code indicating whether the hypervisor was able to
2119 * satisfy the request or not.
2121 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2122 unsigned long *pfns_in,
2123 unsigned long extents_out,
2124 unsigned int order_out,
2125 unsigned long *mfns_out,
2126 unsigned int address_bits)
2131 struct xen_memory_exchange exchange = {
2133 .nr_extents = extents_in,
2134 .extent_order = order_in,
2135 .extent_start = pfns_in,
2139 .nr_extents = extents_out,
2140 .extent_order = order_out,
2141 .extent_start = mfns_out,
2142 .address_bits = address_bits,
2147 BUG_ON(extents_in << order_in != extents_out << order_out);
2149 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2150 success = (exchange.nr_exchanged == extents_in);
2152 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2153 BUG_ON(success && (rc != 0));
2158 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2159 unsigned int address_bits)
2161 unsigned long *in_frames = discontig_frames, out_frame;
2162 unsigned long flags;
2166 * Currently an auto-translated guest will not perform I/O, nor will
2167 * it require PAE page directories below 4GB. Therefore any calls to
2168 * this function are redundant and can be ignored.
2171 if (xen_feature(XENFEAT_auto_translated_physmap))
2174 if (unlikely(order > MAX_CONTIG_ORDER))
2177 memset((void *) vstart, 0, PAGE_SIZE << order);
2179 spin_lock_irqsave(&xen_reservation_lock, flags);
2181 /* 1. Zap current PTEs, remembering MFNs. */
2182 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2184 /* 2. Get a new contiguous memory extent. */
2185 out_frame = virt_to_pfn(vstart);
2186 success = xen_exchange_memory(1UL << order, 0, in_frames,
2187 1, order, &out_frame,
2190 /* 3. Map the new extent in place of old pages. */
2192 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2194 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2196 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2198 return success ? 0 : -ENOMEM;
2200 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2202 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2204 unsigned long *out_frames = discontig_frames, in_frame;
2205 unsigned long flags;
2208 if (xen_feature(XENFEAT_auto_translated_physmap))
2211 if (unlikely(order > MAX_CONTIG_ORDER))
2214 memset((void *) vstart, 0, PAGE_SIZE << order);
2216 spin_lock_irqsave(&xen_reservation_lock, flags);
2218 /* 1. Find start MFN of contiguous extent. */
2219 in_frame = virt_to_mfn(vstart);
2221 /* 2. Zap current PTEs. */
2222 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2224 /* 3. Do the exchange for non-contiguous MFNs. */
2225 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2228 /* 4. Map new pages in place of old pages. */
2230 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2232 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2234 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2236 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2238 #ifdef CONFIG_XEN_PVHVM
2239 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2241 struct xen_hvm_pagetable_dying a;
2244 a.domid = DOMID_SELF;
2245 a.gpa = __pa(mm->pgd);
2246 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2247 WARN_ON_ONCE(rc < 0);
2250 static int is_pagetable_dying_supported(void)
2252 struct xen_hvm_pagetable_dying a;
2255 a.domid = DOMID_SELF;
2257 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2259 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2265 void __init xen_hvm_init_mmu_ops(void)
2267 if (is_pagetable_dying_supported())
2268 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2272 #ifdef CONFIG_XEN_DEBUG_FS
2274 static struct dentry *d_mmu_debug;
2276 static int __init xen_mmu_debugfs(void)
2278 struct dentry *d_xen = xen_init_debugfs();
2283 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2285 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2287 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2288 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2289 &mmu_stats.pgd_update_pinned);
2290 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2291 &mmu_stats.pgd_update_pinned);
2293 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2294 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2295 &mmu_stats.pud_update_pinned);
2296 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2297 &mmu_stats.pud_update_pinned);
2299 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2300 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2301 &mmu_stats.pmd_update_pinned);
2302 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2303 &mmu_stats.pmd_update_pinned);
2305 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2306 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2307 // &mmu_stats.pte_update_pinned);
2308 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2309 &mmu_stats.pte_update_pinned);
2311 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2312 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2313 &mmu_stats.mmu_update_extended);
2314 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2315 mmu_stats.mmu_update_histo, 20);
2317 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2318 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2319 &mmu_stats.set_pte_at_batched);
2320 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2321 &mmu_stats.set_pte_at_current);
2322 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2323 &mmu_stats.set_pte_at_kernel);
2325 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2326 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2327 &mmu_stats.prot_commit_batched);
2331 fs_initcall(xen_mmu_debugfs);
2333 #endif /* CONFIG_XEN_DEBUG_FS */