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/mm.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/export.h>
47 #include <linux/init.h>
48 #include <linux/gfp.h>
49 #include <linux/memblock.h>
50 #include <linux/seq_file.h>
51 #include <linux/crash_dump.h>
52 #ifdef CONFIG_KEXEC_CORE
53 #include <linux/kexec.h>
56 #include <trace/events/xen.h>
58 #include <asm/pgtable.h>
59 #include <asm/tlbflush.h>
60 #include <asm/fixmap.h>
61 #include <asm/mmu_context.h>
62 #include <asm/setup.h>
63 #include <asm/paravirt.h>
64 #include <asm/e820/api.h>
65 #include <asm/linkage.h>
71 #include <asm/xen/hypercall.h>
72 #include <asm/xen/hypervisor.h>
76 #include <xen/interface/xen.h>
77 #include <xen/interface/hvm/hvm_op.h>
78 #include <xen/interface/version.h>
79 #include <xen/interface/memory.h>
80 #include <xen/hvc-console.h>
82 #include "multicalls.h"
88 * Identity map, in addition to plain kernel map. This needs to be
89 * large enough to allocate page table pages to allocate the rest.
90 * Each page can map 2MB.
92 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
93 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
96 /* l3 pud for userspace vsyscall mapping */
97 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
98 #endif /* CONFIG_X86_64 */
101 * Note about cr3 (pagetable base) values:
103 * xen_cr3 contains the current logical cr3 value; it contains the
104 * last set cr3. This may not be the current effective cr3, because
105 * its update may be being lazily deferred. However, a vcpu looking
106 * at its own cr3 can use this value knowing that it everything will
107 * be self-consistent.
109 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
110 * hypercall to set the vcpu cr3 is complete (so it may be a little
111 * out of date, but it will never be set early). If one vcpu is
112 * looking at another vcpu's cr3 value, it should use this variable.
114 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
115 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
117 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
120 * Just beyond the highest usermode address. STACK_TOP_MAX has a
121 * redzone above it, so round it up to a PGD boundary.
123 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
125 void make_lowmem_page_readonly(void *vaddr)
128 unsigned long address = (unsigned long)vaddr;
131 pte = lookup_address(address, &level);
133 return; /* vaddr missing */
135 ptev = pte_wrprotect(*pte);
137 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
141 void make_lowmem_page_readwrite(void *vaddr)
144 unsigned long address = (unsigned long)vaddr;
147 pte = lookup_address(address, &level);
149 return; /* vaddr missing */
151 ptev = pte_mkwrite(*pte);
153 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
158 static bool xen_page_pinned(void *ptr)
160 struct page *page = virt_to_page(ptr);
162 return PagePinned(page);
165 static void xen_extend_mmu_update(const struct mmu_update *update)
167 struct multicall_space mcs;
168 struct mmu_update *u;
170 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
172 if (mcs.mc != NULL) {
175 mcs = __xen_mc_entry(sizeof(*u));
176 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
183 static void xen_extend_mmuext_op(const struct mmuext_op *op)
185 struct multicall_space mcs;
188 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
190 if (mcs.mc != NULL) {
193 mcs = __xen_mc_entry(sizeof(*u));
194 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
201 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
209 /* ptr may be ioremapped for 64-bit pagetable setup */
210 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
211 u.val = pmd_val_ma(val);
212 xen_extend_mmu_update(&u);
214 xen_mc_issue(PARAVIRT_LAZY_MMU);
219 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
221 trace_xen_mmu_set_pmd(ptr, val);
223 /* If page is not pinned, we can just update the entry
225 if (!xen_page_pinned(ptr)) {
230 xen_set_pmd_hyper(ptr, val);
234 * Associate a virtual page frame with a given physical page frame
235 * and protection flags for that frame.
237 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
239 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
242 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
246 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
251 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
252 u.val = pte_val_ma(pteval);
253 xen_extend_mmu_update(&u);
255 xen_mc_issue(PARAVIRT_LAZY_MMU);
260 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
262 if (!xen_batched_set_pte(ptep, pteval)) {
264 * Could call native_set_pte() here and trap and
265 * emulate the PTE write but with 32-bit guests this
266 * needs two traps (one for each of the two 32-bit
267 * words in the PTE) so do one hypercall directly
272 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
273 u.val = pte_val_ma(pteval);
274 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
278 static void xen_set_pte(pte_t *ptep, pte_t pteval)
280 trace_xen_mmu_set_pte(ptep, pteval);
281 __xen_set_pte(ptep, pteval);
284 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
285 pte_t *ptep, pte_t pteval)
287 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
288 __xen_set_pte(ptep, pteval);
291 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
292 unsigned long addr, pte_t *ptep)
294 /* Just return the pte as-is. We preserve the bits on commit */
295 trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
299 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
300 pte_t *ptep, pte_t pte)
304 trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
307 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
308 u.val = pte_val_ma(pte);
309 xen_extend_mmu_update(&u);
311 xen_mc_issue(PARAVIRT_LAZY_MMU);
314 /* Assume pteval_t is equivalent to all the other *val_t types. */
315 static pteval_t pte_mfn_to_pfn(pteval_t val)
317 if (val & _PAGE_PRESENT) {
318 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
319 unsigned long pfn = mfn_to_pfn(mfn);
321 pteval_t flags = val & PTE_FLAGS_MASK;
322 if (unlikely(pfn == ~0))
323 val = flags & ~_PAGE_PRESENT;
325 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
331 static pteval_t pte_pfn_to_mfn(pteval_t val)
333 if (val & _PAGE_PRESENT) {
334 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
335 pteval_t flags = val & PTE_FLAGS_MASK;
338 mfn = __pfn_to_mfn(pfn);
341 * If there's no mfn for the pfn, then just create an
342 * empty non-present pte. Unfortunately this loses
343 * information about the original pfn, so
344 * pte_mfn_to_pfn is asymmetric.
346 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
350 mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
351 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
357 __visible pteval_t xen_pte_val(pte_t pte)
359 pteval_t pteval = pte.pte;
361 return pte_mfn_to_pfn(pteval);
363 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
365 __visible pgdval_t xen_pgd_val(pgd_t pgd)
367 return pte_mfn_to_pfn(pgd.pgd);
369 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
371 __visible pte_t xen_make_pte(pteval_t pte)
373 pte = pte_pfn_to_mfn(pte);
375 return native_make_pte(pte);
377 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
379 __visible pgd_t xen_make_pgd(pgdval_t pgd)
381 pgd = pte_pfn_to_mfn(pgd);
382 return native_make_pgd(pgd);
384 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
386 __visible pmdval_t xen_pmd_val(pmd_t pmd)
388 return pte_mfn_to_pfn(pmd.pmd);
390 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
392 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
400 /* ptr may be ioremapped for 64-bit pagetable setup */
401 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
402 u.val = pud_val_ma(val);
403 xen_extend_mmu_update(&u);
405 xen_mc_issue(PARAVIRT_LAZY_MMU);
410 static void xen_set_pud(pud_t *ptr, pud_t val)
412 trace_xen_mmu_set_pud(ptr, val);
414 /* If page is not pinned, we can just update the entry
416 if (!xen_page_pinned(ptr)) {
421 xen_set_pud_hyper(ptr, val);
424 #ifdef CONFIG_X86_PAE
425 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
427 trace_xen_mmu_set_pte_atomic(ptep, pte);
428 set_64bit((u64 *)ptep, native_pte_val(pte));
431 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
433 trace_xen_mmu_pte_clear(mm, addr, ptep);
434 if (!xen_batched_set_pte(ptep, native_make_pte(0)))
435 native_pte_clear(mm, addr, ptep);
438 static void xen_pmd_clear(pmd_t *pmdp)
440 trace_xen_mmu_pmd_clear(pmdp);
441 set_pmd(pmdp, __pmd(0));
443 #endif /* CONFIG_X86_PAE */
445 __visible pmd_t xen_make_pmd(pmdval_t pmd)
447 pmd = pte_pfn_to_mfn(pmd);
448 return native_make_pmd(pmd);
450 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
453 __visible pudval_t xen_pud_val(pud_t pud)
455 return pte_mfn_to_pfn(pud.pud);
457 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
459 __visible pud_t xen_make_pud(pudval_t pud)
461 pud = pte_pfn_to_mfn(pud);
463 return native_make_pud(pud);
465 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
467 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
469 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
470 unsigned offset = pgd - pgd_page;
471 pgd_t *user_ptr = NULL;
473 if (offset < pgd_index(USER_LIMIT)) {
474 struct page *page = virt_to_page(pgd_page);
475 user_ptr = (pgd_t *)page->private;
483 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
487 u.ptr = virt_to_machine(ptr).maddr;
488 u.val = p4d_val_ma(val);
489 xen_extend_mmu_update(&u);
493 * Raw hypercall-based set_p4d, intended for in early boot before
494 * there's a page structure. This implies:
495 * 1. The only existing pagetable is the kernel's
496 * 2. It is always pinned
497 * 3. It has no user pagetable attached to it
499 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
505 __xen_set_p4d_hyper(ptr, val);
507 xen_mc_issue(PARAVIRT_LAZY_MMU);
512 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
514 pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
517 trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
519 /* If page is not pinned, we can just update the entry
521 if (!xen_page_pinned(ptr)) {
524 WARN_ON(xen_page_pinned(user_ptr));
525 pgd_val.pgd = p4d_val_ma(val);
531 /* If it's pinned, then we can at least batch the kernel and
532 user updates together. */
535 __xen_set_p4d_hyper(ptr, val);
537 __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
539 xen_mc_issue(PARAVIRT_LAZY_MMU);
541 #endif /* CONFIG_X86_64 */
543 static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
544 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
545 bool last, unsigned long limit)
547 int i, nr, flush = 0;
549 nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
550 for (i = 0; i < nr; i++) {
551 if (!pmd_none(pmd[i]))
552 flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE);
557 static int xen_pud_walk(struct mm_struct *mm, pud_t *pud,
558 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
559 bool last, unsigned long limit)
561 int i, nr, flush = 0;
563 nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
564 for (i = 0; i < nr; i++) {
567 if (pud_none(pud[i]))
570 pmd = pmd_offset(&pud[i], 0);
571 if (PTRS_PER_PMD > 1)
572 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
573 flush |= xen_pmd_walk(mm, pmd, func,
574 last && i == nr - 1, limit);
579 static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
580 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
581 bool last, unsigned long limit)
590 pud = pud_offset(p4d, 0);
591 if (PTRS_PER_PUD > 1)
592 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
593 flush |= xen_pud_walk(mm, pud, func, last, limit);
598 * (Yet another) pagetable walker. This one is intended for pinning a
599 * pagetable. This means that it walks a pagetable and calls the
600 * callback function on each page it finds making up the page table,
601 * at every level. It walks the entire pagetable, but it only bothers
602 * pinning pte pages which are below limit. In the normal case this
603 * will be STACK_TOP_MAX, but at boot we need to pin up to
606 * For 32-bit the important bit is that we don't pin beyond there,
607 * because then we start getting into Xen's ptes.
609 * For 64-bit, we must skip the Xen hole in the middle of the address
610 * space, just after the big x86-64 virtual hole.
612 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
613 int (*func)(struct mm_struct *mm, struct page *,
617 int i, nr, flush = 0;
618 unsigned hole_low, hole_high;
620 /* The limit is the last byte to be touched */
622 BUG_ON(limit >= FIXADDR_TOP);
625 * 64-bit has a great big hole in the middle of the address
626 * space, which contains the Xen mappings. On 32-bit these
627 * will end up making a zero-sized hole and so is a no-op.
629 hole_low = pgd_index(USER_LIMIT);
630 hole_high = pgd_index(PAGE_OFFSET);
632 nr = pgd_index(limit) + 1;
633 for (i = 0; i < nr; i++) {
636 if (i >= hole_low && i < hole_high)
639 if (pgd_none(pgd[i]))
642 p4d = p4d_offset(&pgd[i], 0);
643 flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
646 /* Do the top level last, so that the callbacks can use it as
647 a cue to do final things like tlb flushes. */
648 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
653 static int xen_pgd_walk(struct mm_struct *mm,
654 int (*func)(struct mm_struct *mm, struct page *,
658 return __xen_pgd_walk(mm, mm->pgd, func, limit);
661 /* If we're using split pte locks, then take the page's lock and
662 return a pointer to it. Otherwise return NULL. */
663 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
665 spinlock_t *ptl = NULL;
667 #if USE_SPLIT_PTE_PTLOCKS
668 ptl = ptlock_ptr(page);
669 spin_lock_nest_lock(ptl, &mm->page_table_lock);
675 static void xen_pte_unlock(void *v)
681 static void xen_do_pin(unsigned level, unsigned long pfn)
686 op.arg1.mfn = pfn_to_mfn(pfn);
688 xen_extend_mmuext_op(&op);
691 static int xen_pin_page(struct mm_struct *mm, struct page *page,
694 unsigned pgfl = TestSetPagePinned(page);
698 flush = 0; /* already pinned */
699 else if (PageHighMem(page))
700 /* kmaps need flushing if we found an unpinned
704 void *pt = lowmem_page_address(page);
705 unsigned long pfn = page_to_pfn(page);
706 struct multicall_space mcs = __xen_mc_entry(0);
712 * We need to hold the pagetable lock between the time
713 * we make the pagetable RO and when we actually pin
714 * it. If we don't, then other users may come in and
715 * attempt to update the pagetable by writing it,
716 * which will fail because the memory is RO but not
717 * pinned, so Xen won't do the trap'n'emulate.
719 * If we're using split pte locks, we can't hold the
720 * entire pagetable's worth of locks during the
721 * traverse, because we may wrap the preempt count (8
722 * bits). The solution is to mark RO and pin each PTE
723 * page while holding the lock. This means the number
724 * of locks we end up holding is never more than a
725 * batch size (~32 entries, at present).
727 * If we're not using split pte locks, we needn't pin
728 * the PTE pages independently, because we're
729 * protected by the overall pagetable lock.
733 ptl = xen_pte_lock(page, mm);
735 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
736 pfn_pte(pfn, PAGE_KERNEL_RO),
737 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
740 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
742 /* Queue a deferred unlock for when this batch
744 xen_mc_callback(xen_pte_unlock, ptl);
751 /* This is called just after a mm has been created, but it has not
752 been used yet. We need to make sure that its pagetable is all
753 read-only, and can be pinned. */
754 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
756 trace_xen_mmu_pgd_pin(mm, pgd);
760 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
761 /* re-enable interrupts for flushing */
771 pgd_t *user_pgd = xen_get_user_pgd(pgd);
773 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
776 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
777 xen_do_pin(MMUEXT_PIN_L4_TABLE,
778 PFN_DOWN(__pa(user_pgd)));
781 #else /* CONFIG_X86_32 */
782 #ifdef CONFIG_X86_PAE
783 /* Need to make sure unshared kernel PMD is pinnable */
784 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
787 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
788 #endif /* CONFIG_X86_64 */
792 static void xen_pgd_pin(struct mm_struct *mm)
794 __xen_pgd_pin(mm, mm->pgd);
798 * On save, we need to pin all pagetables to make sure they get their
799 * mfns turned into pfns. Search the list for any unpinned pgds and pin
800 * them (unpinned pgds are not currently in use, probably because the
801 * process is under construction or destruction).
803 * Expected to be called in stop_machine() ("equivalent to taking
804 * every spinlock in the system"), so the locking doesn't really
805 * matter all that much.
807 void xen_mm_pin_all(void)
811 spin_lock(&pgd_lock);
813 list_for_each_entry(page, &pgd_list, lru) {
814 if (!PagePinned(page)) {
815 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
816 SetPageSavePinned(page);
820 spin_unlock(&pgd_lock);
824 * The init_mm pagetable is really pinned as soon as its created, but
825 * that's before we have page structures to store the bits. So do all
826 * the book-keeping now.
828 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
835 static void __init xen_mark_init_mm_pinned(void)
837 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
840 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
843 unsigned pgfl = TestClearPagePinned(page);
845 if (pgfl && !PageHighMem(page)) {
846 void *pt = lowmem_page_address(page);
847 unsigned long pfn = page_to_pfn(page);
848 spinlock_t *ptl = NULL;
849 struct multicall_space mcs;
852 * Do the converse to pin_page. If we're using split
853 * pte locks, we must be holding the lock for while
854 * the pte page is unpinned but still RO to prevent
855 * concurrent updates from seeing it in this
856 * partially-pinned state.
858 if (level == PT_PTE) {
859 ptl = xen_pte_lock(page, mm);
862 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
865 mcs = __xen_mc_entry(0);
867 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
868 pfn_pte(pfn, PAGE_KERNEL),
869 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
872 /* unlock when batch completed */
873 xen_mc_callback(xen_pte_unlock, ptl);
877 return 0; /* never need to flush on unpin */
880 /* Release a pagetables pages back as normal RW */
881 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
883 trace_xen_mmu_pgd_unpin(mm, pgd);
887 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
891 pgd_t *user_pgd = xen_get_user_pgd(pgd);
894 xen_do_pin(MMUEXT_UNPIN_TABLE,
895 PFN_DOWN(__pa(user_pgd)));
896 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
901 #ifdef CONFIG_X86_PAE
902 /* Need to make sure unshared kernel PMD is unpinned */
903 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
907 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
912 static void xen_pgd_unpin(struct mm_struct *mm)
914 __xen_pgd_unpin(mm, mm->pgd);
918 * On resume, undo any pinning done at save, so that the rest of the
919 * kernel doesn't see any unexpected pinned pagetables.
921 void xen_mm_unpin_all(void)
925 spin_lock(&pgd_lock);
927 list_for_each_entry(page, &pgd_list, lru) {
928 if (PageSavePinned(page)) {
929 BUG_ON(!PagePinned(page));
930 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
931 ClearPageSavePinned(page);
935 spin_unlock(&pgd_lock);
938 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
940 spin_lock(&next->page_table_lock);
942 spin_unlock(&next->page_table_lock);
945 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
947 spin_lock(&mm->page_table_lock);
949 spin_unlock(&mm->page_table_lock);
952 static void drop_mm_ref_this_cpu(void *info)
954 struct mm_struct *mm = info;
956 if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
957 leave_mm(smp_processor_id());
960 * If this cpu still has a stale cr3 reference, then make sure
961 * it has been flushed.
963 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
969 * Another cpu may still have their %cr3 pointing at the pagetable, so
970 * we need to repoint it somewhere else before we can unpin it.
972 static void xen_drop_mm_ref(struct mm_struct *mm)
977 drop_mm_ref_this_cpu(mm);
979 /* Get the "official" set of cpus referring to our pagetable. */
980 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
981 for_each_online_cpu(cpu) {
982 if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
984 smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
990 * It's possible that a vcpu may have a stale reference to our
991 * cr3, because its in lazy mode, and it hasn't yet flushed
992 * its set of pending hypercalls yet. In this case, we can
993 * look at its actual current cr3 value, and force it to flush
997 for_each_online_cpu(cpu) {
998 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
999 cpumask_set_cpu(cpu, mask);
1002 smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
1003 free_cpumask_var(mask);
1006 static void xen_drop_mm_ref(struct mm_struct *mm)
1008 drop_mm_ref_this_cpu(mm);
1013 * While a process runs, Xen pins its pagetables, which means that the
1014 * hypervisor forces it to be read-only, and it controls all updates
1015 * to it. This means that all pagetable updates have to go via the
1016 * hypervisor, which is moderately expensive.
1018 * Since we're pulling the pagetable down, we switch to use init_mm,
1019 * unpin old process pagetable and mark it all read-write, which
1020 * allows further operations on it to be simple memory accesses.
1022 * The only subtle point is that another CPU may be still using the
1023 * pagetable because of lazy tlb flushing. This means we need need to
1024 * switch all CPUs off this pagetable before we can unpin it.
1026 static void xen_exit_mmap(struct mm_struct *mm)
1028 get_cpu(); /* make sure we don't move around */
1029 xen_drop_mm_ref(mm);
1032 spin_lock(&mm->page_table_lock);
1034 /* pgd may not be pinned in the error exit path of execve */
1035 if (xen_page_pinned(mm->pgd))
1038 spin_unlock(&mm->page_table_lock);
1041 static void xen_post_allocator_init(void);
1043 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1045 struct mmuext_op op;
1048 op.arg1.mfn = pfn_to_mfn(pfn);
1049 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1053 #ifdef CONFIG_X86_64
1054 static void __init xen_cleanhighmap(unsigned long vaddr,
1055 unsigned long vaddr_end)
1057 unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1058 pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1060 /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1061 * We include the PMD passed in on _both_ boundaries. */
1062 for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1063 pmd++, vaddr += PMD_SIZE) {
1066 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1067 set_pmd(pmd, __pmd(0));
1069 /* In case we did something silly, we should crash in this function
1070 * instead of somewhere later and be confusing. */
1075 * Make a page range writeable and free it.
1077 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1079 void *vaddr = __va(paddr);
1080 void *vaddr_end = vaddr + size;
1082 for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1083 make_lowmem_page_readwrite(vaddr);
1085 memblock_free(paddr, size);
1088 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1090 unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1093 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1094 ClearPagePinned(virt_to_page(__va(pa)));
1095 xen_free_ro_pages(pa, PAGE_SIZE);
1098 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1104 if (pmd_large(*pmd)) {
1105 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1106 xen_free_ro_pages(pa, PMD_SIZE);
1110 pte_tbl = pte_offset_kernel(pmd, 0);
1111 for (i = 0; i < PTRS_PER_PTE; i++) {
1112 if (pte_none(pte_tbl[i]))
1114 pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1115 xen_free_ro_pages(pa, PAGE_SIZE);
1117 set_pmd(pmd, __pmd(0));
1118 xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1121 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1127 if (pud_large(*pud)) {
1128 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1129 xen_free_ro_pages(pa, PUD_SIZE);
1133 pmd_tbl = pmd_offset(pud, 0);
1134 for (i = 0; i < PTRS_PER_PMD; i++) {
1135 if (pmd_none(pmd_tbl[i]))
1137 xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1139 set_pud(pud, __pud(0));
1140 xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1143 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1149 if (p4d_large(*p4d)) {
1150 pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1151 xen_free_ro_pages(pa, P4D_SIZE);
1155 pud_tbl = pud_offset(p4d, 0);
1156 for (i = 0; i < PTRS_PER_PUD; i++) {
1157 if (pud_none(pud_tbl[i]))
1159 xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1161 set_p4d(p4d, __p4d(0));
1162 xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1166 * Since it is well isolated we can (and since it is perhaps large we should)
1167 * also free the page tables mapping the initial P->M table.
1169 static void __init xen_cleanmfnmap(unsigned long vaddr)
1175 unpin = (vaddr == 2 * PGDIR_SIZE);
1177 pgd = pgd_offset_k(vaddr);
1178 p4d = p4d_offset(pgd, 0);
1179 if (!p4d_none(*p4d))
1180 xen_cleanmfnmap_p4d(p4d, unpin);
1183 static void __init xen_pagetable_p2m_free(void)
1188 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1190 /* No memory or already called. */
1191 if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1194 /* using __ka address and sticking INVALID_P2M_ENTRY! */
1195 memset((void *)xen_start_info->mfn_list, 0xff, size);
1197 addr = xen_start_info->mfn_list;
1199 * We could be in __ka space.
1200 * We roundup to the PMD, which means that if anybody at this stage is
1201 * using the __ka address of xen_start_info or
1202 * xen_start_info->shared_info they are in going to crash. Fortunatly
1203 * we have already revectored in xen_setup_kernel_pagetable and in
1204 * xen_setup_shared_info.
1206 size = roundup(size, PMD_SIZE);
1208 if (addr >= __START_KERNEL_map) {
1209 xen_cleanhighmap(addr, addr + size);
1210 size = PAGE_ALIGN(xen_start_info->nr_pages *
1211 sizeof(unsigned long));
1212 memblock_free(__pa(addr), size);
1214 xen_cleanmfnmap(addr);
1218 static void __init xen_pagetable_cleanhighmap(void)
1223 /* At this stage, cleanup_highmap has already cleaned __ka space
1224 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1225 * the ramdisk). We continue on, erasing PMD entries that point to page
1226 * tables - do note that they are accessible at this stage via __va.
1227 * As Xen is aligning the memory end to a 4MB boundary, for good
1228 * measure we also round up to PMD_SIZE * 2 - which means that if
1229 * anybody is using __ka address to the initial boot-stack - and try
1230 * to use it - they are going to crash. The xen_start_info has been
1231 * taken care of already in xen_setup_kernel_pagetable. */
1232 addr = xen_start_info->pt_base;
1233 size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1235 xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1236 xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1240 static void __init xen_pagetable_p2m_setup(void)
1242 xen_vmalloc_p2m_tree();
1244 #ifdef CONFIG_X86_64
1245 xen_pagetable_p2m_free();
1247 xen_pagetable_cleanhighmap();
1249 /* And revector! Bye bye old array */
1250 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1253 static void __init xen_pagetable_init(void)
1256 xen_post_allocator_init();
1258 xen_pagetable_p2m_setup();
1260 /* Allocate and initialize top and mid mfn levels for p2m structure */
1261 xen_build_mfn_list_list();
1263 /* Remap memory freed due to conflicts with E820 map */
1266 xen_setup_shared_info();
1268 static void xen_write_cr2(unsigned long cr2)
1270 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1273 static unsigned long xen_read_cr2(void)
1275 return this_cpu_read(xen_vcpu)->arch.cr2;
1278 unsigned long xen_read_cr2_direct(void)
1280 return this_cpu_read(xen_vcpu_info.arch.cr2);
1283 static void xen_flush_tlb(void)
1285 struct mmuext_op *op;
1286 struct multicall_space mcs;
1288 trace_xen_mmu_flush_tlb(0);
1292 mcs = xen_mc_entry(sizeof(*op));
1295 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1296 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1298 xen_mc_issue(PARAVIRT_LAZY_MMU);
1303 static void xen_flush_tlb_single(unsigned long addr)
1305 struct mmuext_op *op;
1306 struct multicall_space mcs;
1308 trace_xen_mmu_flush_tlb_single(addr);
1312 mcs = xen_mc_entry(sizeof(*op));
1314 op->cmd = MMUEXT_INVLPG_LOCAL;
1315 op->arg1.linear_addr = addr & PAGE_MASK;
1316 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1318 xen_mc_issue(PARAVIRT_LAZY_MMU);
1323 static void xen_flush_tlb_others(const struct cpumask *cpus,
1324 const struct flush_tlb_info *info)
1327 struct mmuext_op op;
1329 DECLARE_BITMAP(mask, num_processors);
1331 DECLARE_BITMAP(mask, NR_CPUS);
1334 struct multicall_space mcs;
1336 trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end);
1338 if (cpumask_empty(cpus))
1339 return; /* nothing to do */
1341 mcs = xen_mc_entry(sizeof(*args));
1343 args->op.arg2.vcpumask = to_cpumask(args->mask);
1345 /* Remove us, and any offline CPUS. */
1346 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1347 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1349 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1350 if (info->end != TLB_FLUSH_ALL &&
1351 (info->end - info->start) <= PAGE_SIZE) {
1352 args->op.cmd = MMUEXT_INVLPG_MULTI;
1353 args->op.arg1.linear_addr = info->start;
1356 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1358 xen_mc_issue(PARAVIRT_LAZY_MMU);
1361 static unsigned long xen_read_cr3(void)
1363 return this_cpu_read(xen_cr3);
1366 static void set_current_cr3(void *v)
1368 this_cpu_write(xen_current_cr3, (unsigned long)v);
1371 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1373 struct mmuext_op op;
1376 trace_xen_mmu_write_cr3(kernel, cr3);
1379 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1383 WARN_ON(mfn == 0 && kernel);
1385 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1388 xen_extend_mmuext_op(&op);
1391 this_cpu_write(xen_cr3, cr3);
1393 /* Update xen_current_cr3 once the batch has actually
1395 xen_mc_callback(set_current_cr3, (void *)cr3);
1398 static void xen_write_cr3(unsigned long cr3)
1400 BUG_ON(preemptible());
1402 xen_mc_batch(); /* disables interrupts */
1404 /* Update while interrupts are disabled, so its atomic with
1406 this_cpu_write(xen_cr3, cr3);
1408 __xen_write_cr3(true, cr3);
1410 #ifdef CONFIG_X86_64
1412 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1414 __xen_write_cr3(false, __pa(user_pgd));
1416 __xen_write_cr3(false, 0);
1420 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1423 #ifdef CONFIG_X86_64
1425 * At the start of the day - when Xen launches a guest, it has already
1426 * built pagetables for the guest. We diligently look over them
1427 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1428 * init_top_pgt and its friends. Then when we are happy we load
1429 * the new init_top_pgt - and continue on.
1431 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1432 * up the rest of the pagetables. When it has completed it loads the cr3.
1433 * N.B. that baremetal would start at 'start_kernel' (and the early
1434 * #PF handler would create bootstrap pagetables) - so we are running
1435 * with the same assumptions as what to do when write_cr3 is executed
1438 * Since there are no user-page tables at all, we have two variants
1439 * of xen_write_cr3 - the early bootup (this one), and the late one
1440 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1441 * the Linux kernel and user-space are both in ring 3 while the
1442 * hypervisor is in ring 0.
1444 static void __init xen_write_cr3_init(unsigned long cr3)
1446 BUG_ON(preemptible());
1448 xen_mc_batch(); /* disables interrupts */
1450 /* Update while interrupts are disabled, so its atomic with
1452 this_cpu_write(xen_cr3, cr3);
1454 __xen_write_cr3(true, cr3);
1456 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1460 static int xen_pgd_alloc(struct mm_struct *mm)
1462 pgd_t *pgd = mm->pgd;
1465 BUG_ON(PagePinned(virt_to_page(pgd)));
1467 #ifdef CONFIG_X86_64
1469 struct page *page = virt_to_page(pgd);
1472 BUG_ON(page->private != 0);
1476 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1477 page->private = (unsigned long)user_pgd;
1479 if (user_pgd != NULL) {
1480 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1481 user_pgd[pgd_index(VSYSCALL_ADDR)] =
1482 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1487 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1493 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1495 #ifdef CONFIG_X86_64
1496 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1499 free_page((unsigned long)user_pgd);
1504 * Init-time set_pte while constructing initial pagetables, which
1505 * doesn't allow RO page table pages to be remapped RW.
1507 * If there is no MFN for this PFN then this page is initially
1508 * ballooned out so clear the PTE (as in decrease_reservation() in
1509 * drivers/xen/balloon.c).
1511 * Many of these PTE updates are done on unpinned and writable pages
1512 * and doing a hypercall for these is unnecessary and expensive. At
1513 * this point it is not possible to tell if a page is pinned or not,
1514 * so always write the PTE directly and rely on Xen trapping and
1515 * emulating any updates as necessary.
1517 __visible pte_t xen_make_pte_init(pteval_t pte)
1519 #ifdef CONFIG_X86_64
1523 * Pages belonging to the initial p2m list mapped outside the default
1524 * address range must be mapped read-only. This region contains the
1525 * page tables for mapping the p2m list, too, and page tables MUST be
1528 pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1529 if (xen_start_info->mfn_list < __START_KERNEL_map &&
1530 pfn >= xen_start_info->first_p2m_pfn &&
1531 pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1534 pte = pte_pfn_to_mfn(pte);
1535 return native_make_pte(pte);
1537 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1539 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1541 #ifdef CONFIG_X86_32
1542 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1543 if (pte_mfn(pte) != INVALID_P2M_ENTRY
1544 && pte_val_ma(*ptep) & _PAGE_PRESENT)
1545 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1548 native_set_pte(ptep, pte);
1551 /* Early in boot, while setting up the initial pagetable, assume
1552 everything is pinned. */
1553 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1555 #ifdef CONFIG_FLATMEM
1556 BUG_ON(mem_map); /* should only be used early */
1558 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1559 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1562 /* Used for pmd and pud */
1563 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1565 #ifdef CONFIG_FLATMEM
1566 BUG_ON(mem_map); /* should only be used early */
1568 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1571 /* Early release_pte assumes that all pts are pinned, since there's
1572 only init_mm and anything attached to that is pinned. */
1573 static void __init xen_release_pte_init(unsigned long pfn)
1575 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1576 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1579 static void __init xen_release_pmd_init(unsigned long pfn)
1581 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1584 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1586 struct multicall_space mcs;
1587 struct mmuext_op *op;
1589 mcs = __xen_mc_entry(sizeof(*op));
1592 op->arg1.mfn = pfn_to_mfn(pfn);
1594 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1597 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1599 struct multicall_space mcs;
1600 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1602 mcs = __xen_mc_entry(0);
1603 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1604 pfn_pte(pfn, prot), 0);
1607 /* This needs to make sure the new pte page is pinned iff its being
1608 attached to a pinned pagetable. */
1609 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1612 bool pinned = PagePinned(virt_to_page(mm->pgd));
1614 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1617 struct page *page = pfn_to_page(pfn);
1619 SetPagePinned(page);
1621 if (!PageHighMem(page)) {
1624 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1626 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1627 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1629 xen_mc_issue(PARAVIRT_LAZY_MMU);
1631 /* make sure there are no stray mappings of
1633 kmap_flush_unused();
1638 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1640 xen_alloc_ptpage(mm, pfn, PT_PTE);
1643 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1645 xen_alloc_ptpage(mm, pfn, PT_PMD);
1648 /* This should never happen until we're OK to use struct page */
1649 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1651 struct page *page = pfn_to_page(pfn);
1652 bool pinned = PagePinned(page);
1654 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1657 if (!PageHighMem(page)) {
1660 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1661 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1663 __set_pfn_prot(pfn, PAGE_KERNEL);
1665 xen_mc_issue(PARAVIRT_LAZY_MMU);
1667 ClearPagePinned(page);
1671 static void xen_release_pte(unsigned long pfn)
1673 xen_release_ptpage(pfn, PT_PTE);
1676 static void xen_release_pmd(unsigned long pfn)
1678 xen_release_ptpage(pfn, PT_PMD);
1681 #ifdef CONFIG_X86_64
1682 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1684 xen_alloc_ptpage(mm, pfn, PT_PUD);
1687 static void xen_release_pud(unsigned long pfn)
1689 xen_release_ptpage(pfn, PT_PUD);
1693 void __init xen_reserve_top(void)
1695 #ifdef CONFIG_X86_32
1696 unsigned long top = HYPERVISOR_VIRT_START;
1697 struct xen_platform_parameters pp;
1699 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1700 top = pp.virt_start;
1702 reserve_top_address(-top);
1703 #endif /* CONFIG_X86_32 */
1707 * Like __va(), but returns address in the kernel mapping (which is
1708 * all we have until the physical memory mapping has been set up.
1710 static void * __init __ka(phys_addr_t paddr)
1712 #ifdef CONFIG_X86_64
1713 return (void *)(paddr + __START_KERNEL_map);
1719 /* Convert a machine address to physical address */
1720 static unsigned long __init m2p(phys_addr_t maddr)
1724 maddr &= XEN_PTE_MFN_MASK;
1725 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1730 /* Convert a machine address to kernel virtual */
1731 static void * __init m2v(phys_addr_t maddr)
1733 return __ka(m2p(maddr));
1736 /* Set the page permissions on an identity-mapped pages */
1737 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1738 unsigned long flags)
1740 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1741 pte_t pte = pfn_pte(pfn, prot);
1743 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1746 static void __init set_page_prot(void *addr, pgprot_t prot)
1748 return set_page_prot_flags(addr, prot, UVMF_NONE);
1750 #ifdef CONFIG_X86_32
1751 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1753 unsigned pmdidx, pteidx;
1757 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1762 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1765 /* Reuse or allocate a page of ptes */
1766 if (pmd_present(pmd[pmdidx]))
1767 pte_page = m2v(pmd[pmdidx].pmd);
1769 /* Check for free pte pages */
1770 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1773 pte_page = &level1_ident_pgt[ident_pte];
1774 ident_pte += PTRS_PER_PTE;
1776 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1779 /* Install mappings */
1780 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1783 if (pfn > max_pfn_mapped)
1784 max_pfn_mapped = pfn;
1786 if (!pte_none(pte_page[pteidx]))
1789 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1790 pte_page[pteidx] = pte;
1794 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1795 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1797 set_page_prot(pmd, PAGE_KERNEL_RO);
1800 void __init xen_setup_machphys_mapping(void)
1802 struct xen_machphys_mapping mapping;
1804 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1805 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1806 machine_to_phys_nr = mapping.max_mfn + 1;
1808 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1810 #ifdef CONFIG_X86_32
1811 WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1812 < machine_to_phys_mapping);
1816 #ifdef CONFIG_X86_64
1817 static void __init convert_pfn_mfn(void *v)
1822 /* All levels are converted the same way, so just treat them
1824 for (i = 0; i < PTRS_PER_PTE; i++)
1825 pte[i] = xen_make_pte(pte[i].pte);
1827 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1830 if (*pt_base == PFN_DOWN(__pa(addr))) {
1831 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1832 clear_page((void *)addr);
1835 if (*pt_end == PFN_DOWN(__pa(addr))) {
1836 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1837 clear_page((void *)addr);
1842 * Set up the initial kernel pagetable.
1844 * We can construct this by grafting the Xen provided pagetable into
1845 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1846 * level2_ident_pgt, and level2_kernel_pgt. This means that only the
1847 * kernel has a physical mapping to start with - but that's enough to
1848 * get __va working. We need to fill in the rest of the physical
1849 * mapping once some sort of allocator has been set up.
1851 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1855 unsigned long addr[3];
1856 unsigned long pt_base, pt_end;
1859 /* max_pfn_mapped is the last pfn mapped in the initial memory
1860 * mappings. Considering that on Xen after the kernel mappings we
1861 * have the mappings of some pages that don't exist in pfn space, we
1862 * set max_pfn_mapped to the last real pfn mapped. */
1863 if (xen_start_info->mfn_list < __START_KERNEL_map)
1864 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1866 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1868 pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1869 pt_end = pt_base + xen_start_info->nr_pt_frames;
1871 /* Zap identity mapping */
1872 init_top_pgt[0] = __pgd(0);
1874 /* Pre-constructed entries are in pfn, so convert to mfn */
1875 /* L4[272] -> level3_ident_pgt */
1876 /* L4[511] -> level3_kernel_pgt */
1877 convert_pfn_mfn(init_top_pgt);
1879 /* L3_i[0] -> level2_ident_pgt */
1880 convert_pfn_mfn(level3_ident_pgt);
1881 /* L3_k[510] -> level2_kernel_pgt */
1882 /* L3_k[511] -> level2_fixmap_pgt */
1883 convert_pfn_mfn(level3_kernel_pgt);
1885 /* L3_k[511][506] -> level1_fixmap_pgt */
1886 convert_pfn_mfn(level2_fixmap_pgt);
1888 /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1889 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1890 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1892 addr[0] = (unsigned long)pgd;
1893 addr[1] = (unsigned long)l3;
1894 addr[2] = (unsigned long)l2;
1895 /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1896 * Both L4[272][0] and L4[511][510] have entries that point to the same
1897 * L2 (PMD) tables. Meaning that if you modify it in __va space
1898 * it will be also modified in the __ka space! (But if you just
1899 * modify the PMD table to point to other PTE's or none, then you
1900 * are OK - which is what cleanup_highmap does) */
1901 copy_page(level2_ident_pgt, l2);
1902 /* Graft it onto L4[511][510] */
1903 copy_page(level2_kernel_pgt, l2);
1905 /* Copy the initial P->M table mappings if necessary. */
1906 i = pgd_index(xen_start_info->mfn_list);
1907 if (i && i < pgd_index(__START_KERNEL_map))
1908 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1910 /* Make pagetable pieces RO */
1911 set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1912 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1913 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1914 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1915 set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1916 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1917 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1918 set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
1920 /* Pin down new L4 */
1921 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1922 PFN_DOWN(__pa_symbol(init_top_pgt)));
1924 /* Unpin Xen-provided one */
1925 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1928 * At this stage there can be no user pgd, and no page structure to
1929 * attach it to, so make sure we just set kernel pgd.
1932 __xen_write_cr3(true, __pa(init_top_pgt));
1933 xen_mc_issue(PARAVIRT_LAZY_CPU);
1935 /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1936 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
1937 * the initial domain. For guests using the toolstack, they are in:
1938 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1939 * rip out the [L4] (pgd), but for guests we shave off three pages.
1941 for (i = 0; i < ARRAY_SIZE(addr); i++)
1942 check_pt_base(&pt_base, &pt_end, addr[i]);
1944 /* Our (by three pages) smaller Xen pagetable that we are using */
1945 xen_pt_base = PFN_PHYS(pt_base);
1946 xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1947 memblock_reserve(xen_pt_base, xen_pt_size);
1949 /* Revector the xen_start_info */
1950 xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1954 * Read a value from a physical address.
1956 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1958 unsigned long *vaddr;
1961 vaddr = early_memremap_ro(addr, sizeof(val));
1963 early_memunmap(vaddr, sizeof(val));
1968 * Translate a virtual address to a physical one without relying on mapped
1969 * page tables. Don't rely on big pages being aligned in (guest) physical
1972 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1981 pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1983 if (!pgd_present(pgd))
1986 pa = pgd_val(pgd) & PTE_PFN_MASK;
1987 pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1989 if (!pud_present(pud))
1991 pa = pud_val(pud) & PTE_PFN_MASK;
1993 return pa + (vaddr & ~PUD_MASK);
1995 pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1997 if (!pmd_present(pmd))
1999 pa = pmd_val(pmd) & PTE_PFN_MASK;
2001 return pa + (vaddr & ~PMD_MASK);
2003 pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
2005 if (!pte_present(pte))
2007 pa = pte_pfn(pte) << PAGE_SHIFT;
2009 return pa | (vaddr & ~PAGE_MASK);
2013 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
2016 void __init xen_relocate_p2m(void)
2018 phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
2019 unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
2020 int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
2025 unsigned long *new_p2m;
2028 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
2029 n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
2030 n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
2031 n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
2032 n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
2033 n_frames = n_pte + n_pt + n_pmd + n_pud;
2035 new_area = xen_find_free_area(PFN_PHYS(n_frames));
2037 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
2042 * Setup the page tables for addressing the new p2m list.
2043 * We have asked the hypervisor to map the p2m list at the user address
2044 * PUD_SIZE. It may have done so, or it may have used a kernel space
2045 * address depending on the Xen version.
2046 * To avoid any possible virtual address collision, just use
2047 * 2 * PUD_SIZE for the new area.
2049 pud_phys = new_area;
2050 pmd_phys = pud_phys + PFN_PHYS(n_pud);
2051 pt_phys = pmd_phys + PFN_PHYS(n_pmd);
2052 p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
2054 pgd = __va(read_cr3_pa());
2055 new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
2057 for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
2058 pud = early_memremap(pud_phys, PAGE_SIZE);
2060 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
2062 pmd = early_memremap(pmd_phys, PAGE_SIZE);
2064 for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
2066 pt = early_memremap(pt_phys, PAGE_SIZE);
2069 idx_pte < min(n_pte, PTRS_PER_PTE);
2071 set_pte(pt + idx_pte,
2072 pfn_pte(p2m_pfn, PAGE_KERNEL));
2075 n_pte -= PTRS_PER_PTE;
2076 early_memunmap(pt, PAGE_SIZE);
2077 make_lowmem_page_readonly(__va(pt_phys));
2078 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2080 set_pmd(pmd + idx_pt,
2081 __pmd(_PAGE_TABLE | pt_phys));
2082 pt_phys += PAGE_SIZE;
2084 n_pt -= PTRS_PER_PMD;
2085 early_memunmap(pmd, PAGE_SIZE);
2086 make_lowmem_page_readonly(__va(pmd_phys));
2087 pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2088 PFN_DOWN(pmd_phys));
2089 set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
2090 pmd_phys += PAGE_SIZE;
2092 n_pmd -= PTRS_PER_PUD;
2093 early_memunmap(pud, PAGE_SIZE);
2094 make_lowmem_page_readonly(__va(pud_phys));
2095 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2096 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2097 pud_phys += PAGE_SIZE;
2100 /* Now copy the old p2m info to the new area. */
2101 memcpy(new_p2m, xen_p2m_addr, size);
2102 xen_p2m_addr = new_p2m;
2104 /* Release the old p2m list and set new list info. */
2105 p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2107 p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2109 if (xen_start_info->mfn_list < __START_KERNEL_map) {
2110 pfn = xen_start_info->first_p2m_pfn;
2111 pfn_end = xen_start_info->first_p2m_pfn +
2112 xen_start_info->nr_p2m_frames;
2113 set_pgd(pgd + 1, __pgd(0));
2116 pfn_end = p2m_pfn_end;
2119 memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2120 while (pfn < pfn_end) {
2121 if (pfn == p2m_pfn) {
2125 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2129 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2130 xen_start_info->first_p2m_pfn = PFN_DOWN(new_area);
2131 xen_start_info->nr_p2m_frames = n_frames;
2134 #else /* !CONFIG_X86_64 */
2135 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
2136 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
2138 static void __init xen_write_cr3_init(unsigned long cr3)
2140 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
2142 BUG_ON(read_cr3_pa() != __pa(initial_page_table));
2143 BUG_ON(cr3 != __pa(swapper_pg_dir));
2146 * We are switching to swapper_pg_dir for the first time (from
2147 * initial_page_table) and therefore need to mark that page
2148 * read-only and then pin it.
2150 * Xen disallows sharing of kernel PMDs for PAE
2151 * guests. Therefore we must copy the kernel PMD from
2152 * initial_page_table into a new kernel PMD to be used in
2155 swapper_kernel_pmd =
2156 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2157 copy_page(swapper_kernel_pmd, initial_kernel_pmd);
2158 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
2159 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
2160 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
2162 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
2164 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
2166 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
2167 PFN_DOWN(__pa(initial_page_table)));
2168 set_page_prot(initial_page_table, PAGE_KERNEL);
2169 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
2171 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2175 * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
2176 * not the first page table in the page table pool.
2177 * Iterate through the initial page tables to find the real page table base.
2179 static phys_addr_t __init xen_find_pt_base(pmd_t *pmd)
2181 phys_addr_t pt_base, paddr;
2184 pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
2186 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
2187 if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
2188 paddr = m2p(pmd[pmdidx].pmd);
2189 pt_base = min(pt_base, paddr);
2195 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2199 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2201 xen_pt_base = xen_find_pt_base(kernel_pmd);
2202 xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
2204 initial_kernel_pmd =
2205 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2207 max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
2209 copy_page(initial_kernel_pmd, kernel_pmd);
2211 xen_map_identity_early(initial_kernel_pmd, max_pfn);
2213 copy_page(initial_page_table, pgd);
2214 initial_page_table[KERNEL_PGD_BOUNDARY] =
2215 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2217 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2218 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2219 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2221 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2223 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2224 PFN_DOWN(__pa(initial_page_table)));
2225 xen_write_cr3(__pa(initial_page_table));
2227 memblock_reserve(xen_pt_base, xen_pt_size);
2229 #endif /* CONFIG_X86_64 */
2231 void __init xen_reserve_special_pages(void)
2235 memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2236 if (xen_start_info->store_mfn) {
2237 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2238 memblock_reserve(paddr, PAGE_SIZE);
2240 if (!xen_initial_domain()) {
2241 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2242 memblock_reserve(paddr, PAGE_SIZE);
2246 void __init xen_pt_check_e820(void)
2248 if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2249 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2254 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2256 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2260 phys >>= PAGE_SHIFT;
2263 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2265 #ifdef CONFIG_X86_32
2267 # ifdef CONFIG_HIGHMEM
2268 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2270 #elif defined(CONFIG_X86_VSYSCALL_EMULATION)
2273 case FIX_TEXT_POKE0:
2274 case FIX_TEXT_POKE1:
2275 case FIX_GDT_REMAP_BEGIN ... FIX_GDT_REMAP_END:
2276 /* All local page mappings */
2277 pte = pfn_pte(phys, prot);
2280 #ifdef CONFIG_X86_LOCAL_APIC
2281 case FIX_APIC_BASE: /* maps dummy local APIC */
2282 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2286 #ifdef CONFIG_X86_IO_APIC
2287 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2289 * We just don't map the IO APIC - all access is via
2290 * hypercalls. Keep the address in the pte for reference.
2292 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2296 case FIX_PARAVIRT_BOOTMAP:
2297 /* This is an MFN, but it isn't an IO mapping from the
2299 pte = mfn_pte(phys, prot);
2303 /* By default, set_fixmap is used for hardware mappings */
2304 pte = mfn_pte(phys, prot);
2308 __native_set_fixmap(idx, pte);
2310 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2311 /* Replicate changes to map the vsyscall page into the user
2312 pagetable vsyscall mapping. */
2313 if (idx == VSYSCALL_PAGE) {
2314 unsigned long vaddr = __fix_to_virt(idx);
2315 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2320 static void __init xen_post_allocator_init(void)
2322 pv_mmu_ops.set_pte = xen_set_pte;
2323 pv_mmu_ops.set_pmd = xen_set_pmd;
2324 pv_mmu_ops.set_pud = xen_set_pud;
2325 #ifdef CONFIG_X86_64
2326 pv_mmu_ops.set_p4d = xen_set_p4d;
2329 /* This will work as long as patching hasn't happened yet
2330 (which it hasn't) */
2331 pv_mmu_ops.alloc_pte = xen_alloc_pte;
2332 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2333 pv_mmu_ops.release_pte = xen_release_pte;
2334 pv_mmu_ops.release_pmd = xen_release_pmd;
2335 #ifdef CONFIG_X86_64
2336 pv_mmu_ops.alloc_pud = xen_alloc_pud;
2337 pv_mmu_ops.release_pud = xen_release_pud;
2339 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2341 #ifdef CONFIG_X86_64
2342 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2343 SetPagePinned(virt_to_page(level3_user_vsyscall));
2345 xen_mark_init_mm_pinned();
2348 static void xen_leave_lazy_mmu(void)
2352 paravirt_leave_lazy_mmu();
2356 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2357 .read_cr2 = xen_read_cr2,
2358 .write_cr2 = xen_write_cr2,
2360 .read_cr3 = xen_read_cr3,
2361 .write_cr3 = xen_write_cr3_init,
2363 .flush_tlb_user = xen_flush_tlb,
2364 .flush_tlb_kernel = xen_flush_tlb,
2365 .flush_tlb_single = xen_flush_tlb_single,
2366 .flush_tlb_others = xen_flush_tlb_others,
2368 .pgd_alloc = xen_pgd_alloc,
2369 .pgd_free = xen_pgd_free,
2371 .alloc_pte = xen_alloc_pte_init,
2372 .release_pte = xen_release_pte_init,
2373 .alloc_pmd = xen_alloc_pmd_init,
2374 .release_pmd = xen_release_pmd_init,
2376 .set_pte = xen_set_pte_init,
2377 .set_pte_at = xen_set_pte_at,
2378 .set_pmd = xen_set_pmd_hyper,
2380 .ptep_modify_prot_start = __ptep_modify_prot_start,
2381 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2383 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2384 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2386 .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2387 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2389 #ifdef CONFIG_X86_PAE
2390 .set_pte_atomic = xen_set_pte_atomic,
2391 .pte_clear = xen_pte_clear,
2392 .pmd_clear = xen_pmd_clear,
2393 #endif /* CONFIG_X86_PAE */
2394 .set_pud = xen_set_pud_hyper,
2396 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2397 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2399 #ifdef CONFIG_X86_64
2400 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2401 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2402 .set_p4d = xen_set_p4d_hyper,
2404 .alloc_pud = xen_alloc_pmd_init,
2405 .release_pud = xen_release_pmd_init,
2406 #endif /* CONFIG_X86_64 */
2408 .activate_mm = xen_activate_mm,
2409 .dup_mmap = xen_dup_mmap,
2410 .exit_mmap = xen_exit_mmap,
2413 .enter = paravirt_enter_lazy_mmu,
2414 .leave = xen_leave_lazy_mmu,
2415 .flush = paravirt_flush_lazy_mmu,
2418 .set_fixmap = xen_set_fixmap,
2421 void __init xen_init_mmu_ops(void)
2423 x86_init.paging.pagetable_init = xen_pagetable_init;
2425 pv_mmu_ops = xen_mmu_ops;
2427 memset(dummy_mapping, 0xff, PAGE_SIZE);
2430 /* Protected by xen_reservation_lock. */
2431 #define MAX_CONTIG_ORDER 9 /* 2MB */
2432 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2434 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2435 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2436 unsigned long *in_frames,
2437 unsigned long *out_frames)
2440 struct multicall_space mcs;
2443 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2444 mcs = __xen_mc_entry(0);
2447 in_frames[i] = virt_to_mfn(vaddr);
2449 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2450 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2453 out_frames[i] = virt_to_pfn(vaddr);
2459 * Update the pfn-to-mfn mappings for a virtual address range, either to
2460 * point to an array of mfns, or contiguously from a single starting
2463 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2464 unsigned long *mfns,
2465 unsigned long first_mfn)
2472 limit = 1u << order;
2473 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2474 struct multicall_space mcs;
2477 mcs = __xen_mc_entry(0);
2481 mfn = first_mfn + i;
2483 if (i < (limit - 1))
2487 flags = UVMF_INVLPG | UVMF_ALL;
2489 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2492 MULTI_update_va_mapping(mcs.mc, vaddr,
2493 mfn_pte(mfn, PAGE_KERNEL), flags);
2495 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2502 * Perform the hypercall to exchange a region of our pfns to point to
2503 * memory with the required contiguous alignment. Takes the pfns as
2504 * input, and populates mfns as output.
2506 * Returns a success code indicating whether the hypervisor was able to
2507 * satisfy the request or not.
2509 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2510 unsigned long *pfns_in,
2511 unsigned long extents_out,
2512 unsigned int order_out,
2513 unsigned long *mfns_out,
2514 unsigned int address_bits)
2519 struct xen_memory_exchange exchange = {
2521 .nr_extents = extents_in,
2522 .extent_order = order_in,
2523 .extent_start = pfns_in,
2527 .nr_extents = extents_out,
2528 .extent_order = order_out,
2529 .extent_start = mfns_out,
2530 .address_bits = address_bits,
2535 BUG_ON(extents_in << order_in != extents_out << order_out);
2537 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2538 success = (exchange.nr_exchanged == extents_in);
2540 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2541 BUG_ON(success && (rc != 0));
2546 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2547 unsigned int address_bits,
2548 dma_addr_t *dma_handle)
2550 unsigned long *in_frames = discontig_frames, out_frame;
2551 unsigned long flags;
2553 unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2556 * Currently an auto-translated guest will not perform I/O, nor will
2557 * it require PAE page directories below 4GB. Therefore any calls to
2558 * this function are redundant and can be ignored.
2561 if (unlikely(order > MAX_CONTIG_ORDER))
2564 memset((void *) vstart, 0, PAGE_SIZE << order);
2566 spin_lock_irqsave(&xen_reservation_lock, flags);
2568 /* 1. Zap current PTEs, remembering MFNs. */
2569 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2571 /* 2. Get a new contiguous memory extent. */
2572 out_frame = virt_to_pfn(vstart);
2573 success = xen_exchange_memory(1UL << order, 0, in_frames,
2574 1, order, &out_frame,
2577 /* 3. Map the new extent in place of old pages. */
2579 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2581 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2583 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2585 *dma_handle = virt_to_machine(vstart).maddr;
2586 return success ? 0 : -ENOMEM;
2588 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2590 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2592 unsigned long *out_frames = discontig_frames, in_frame;
2593 unsigned long flags;
2595 unsigned long vstart;
2597 if (unlikely(order > MAX_CONTIG_ORDER))
2600 vstart = (unsigned long)phys_to_virt(pstart);
2601 memset((void *) vstart, 0, PAGE_SIZE << order);
2603 spin_lock_irqsave(&xen_reservation_lock, flags);
2605 /* 1. Find start MFN of contiguous extent. */
2606 in_frame = virt_to_mfn(vstart);
2608 /* 2. Zap current PTEs. */
2609 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2611 /* 3. Do the exchange for non-contiguous MFNs. */
2612 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2615 /* 4. Map new pages in place of old pages. */
2617 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2619 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2621 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2623 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2625 #ifdef CONFIG_KEXEC_CORE
2626 phys_addr_t paddr_vmcoreinfo_note(void)
2628 if (xen_pv_domain())
2629 return virt_to_machine(vmcoreinfo_note).maddr;
2631 return __pa(vmcoreinfo_note);
2633 #endif /* CONFIG_KEXEC_CORE */