1 // SPDX-License-Identifier: GPL-2.0
6 * This file contains the various mmu fetch and update operations.
7 * The most important job they must perform is the mapping between the
8 * domain's pfn and the overall machine mfns.
10 * Xen allows guests to directly update the pagetable, in a controlled
11 * fashion. In other words, the guest modifies the same pagetable
12 * that the CPU actually uses, which eliminates the overhead of having
13 * a separate shadow pagetable.
15 * In order to allow this, it falls on the guest domain to map its
16 * notion of a "physical" pfn - which is just a domain-local linear
17 * address - into a real "machine address" which the CPU's MMU can
20 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
21 * inserted directly into the pagetable. When creating a new
22 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
23 * when reading the content back with __(pgd|pmd|pte)_val, it converts
24 * the mfn back into a pfn.
26 * The other constraint is that all pages which make up a pagetable
27 * must be mapped read-only in the guest. This prevents uncontrolled
28 * guest updates to the pagetable. Xen strictly enforces this, and
29 * will disallow any pagetable update which will end up mapping a
30 * pagetable page RW, and will disallow using any writable page as a
33 * Naively, when loading %cr3 with the base of a new pagetable, Xen
34 * would need to validate the whole pagetable before going on.
35 * Naturally, this is quite slow. The solution is to "pin" a
36 * pagetable, which enforces all the constraints on the pagetable even
37 * when it is not actively in use. This menas that Xen can be assured
38 * that it is still valid when you do load it into %cr3, and doesn't
39 * need to revalidate it.
43 #include <linux/sched/mm.h>
44 #include <linux/highmem.h>
45 #include <linux/debugfs.h>
46 #include <linux/bug.h>
47 #include <linux/vmalloc.h>
48 #include <linux/export.h>
49 #include <linux/init.h>
50 #include <linux/gfp.h>
51 #include <linux/memblock.h>
52 #include <linux/seq_file.h>
53 #include <linux/crash_dump.h>
54 #include <linux/pgtable.h>
55 #ifdef CONFIG_KEXEC_CORE
56 #include <linux/kexec.h>
59 #include <trace/events/xen.h>
61 #include <asm/tlbflush.h>
62 #include <asm/fixmap.h>
63 #include <asm/mmu_context.h>
64 #include <asm/setup.h>
65 #include <asm/paravirt.h>
66 #include <asm/e820/api.h>
67 #include <asm/linkage.h>
70 #include <asm/memtype.h>
74 #include <asm/xen/hypercall.h>
75 #include <asm/xen/hypervisor.h>
79 #include <xen/interface/xen.h>
80 #include <xen/interface/hvm/hvm_op.h>
81 #include <xen/interface/version.h>
82 #include <xen/interface/memory.h>
83 #include <xen/hvc-console.h>
85 #include "multicalls.h"
89 /* l3 pud for userspace vsyscall mapping */
90 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
93 * Protects atomic reservation decrease/increase against concurrent increases.
94 * Also protects non-atomic updates of current_pages and balloon lists.
96 static DEFINE_SPINLOCK(xen_reservation_lock);
99 * Note about cr3 (pagetable base) values:
101 * xen_cr3 contains the current logical cr3 value; it contains the
102 * last set cr3. This may not be the current effective cr3, because
103 * its update may be being lazily deferred. However, a vcpu looking
104 * at its own cr3 can use this value knowing that it everything will
105 * be self-consistent.
107 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
108 * hypercall to set the vcpu cr3 is complete (so it may be a little
109 * out of date, but it will never be set early). If one vcpu is
110 * looking at another vcpu's cr3 value, it should use this variable.
112 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
113 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
115 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
117 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
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))
159 * During early boot all page table pages are pinned, but we do not have struct
160 * pages, so return true until struct pages are ready.
162 static bool xen_page_pinned(void *ptr)
164 if (static_branch_likely(&xen_struct_pages_ready)) {
165 struct page *page = virt_to_page(ptr);
167 return PagePinned(page);
172 static void xen_extend_mmu_update(const struct mmu_update *update)
174 struct multicall_space mcs;
175 struct mmu_update *u;
177 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
179 if (mcs.mc != NULL) {
182 mcs = __xen_mc_entry(sizeof(*u));
183 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
190 static void xen_extend_mmuext_op(const struct mmuext_op *op)
192 struct multicall_space mcs;
195 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
197 if (mcs.mc != NULL) {
200 mcs = __xen_mc_entry(sizeof(*u));
201 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
208 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
216 /* ptr may be ioremapped for 64-bit pagetable setup */
217 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
218 u.val = pmd_val_ma(val);
219 xen_extend_mmu_update(&u);
221 xen_mc_issue(PARAVIRT_LAZY_MMU);
226 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
228 trace_xen_mmu_set_pmd(ptr, val);
230 /* If page is not pinned, we can just update the entry
232 if (!xen_page_pinned(ptr)) {
237 xen_set_pmd_hyper(ptr, val);
241 * Associate a virtual page frame with a given physical page frame
242 * and protection flags for that frame.
244 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
246 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
249 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
253 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
258 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
259 u.val = pte_val_ma(pteval);
260 xen_extend_mmu_update(&u);
262 xen_mc_issue(PARAVIRT_LAZY_MMU);
267 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
269 if (!xen_batched_set_pte(ptep, pteval)) {
271 * Could call native_set_pte() here and trap and
272 * emulate the PTE write, but a hypercall is much cheaper.
276 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
277 u.val = pte_val_ma(pteval);
278 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
282 static void xen_set_pte(pte_t *ptep, pte_t pteval)
284 trace_xen_mmu_set_pte(ptep, pteval);
285 __xen_set_pte(ptep, pteval);
288 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
289 pte_t *ptep, pte_t pteval)
291 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
292 __xen_set_pte(ptep, pteval);
295 pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
296 unsigned long addr, pte_t *ptep)
298 /* Just return the pte as-is. We preserve the bits on commit */
299 trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
303 void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
304 pte_t *ptep, pte_t pte)
308 trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
311 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
312 u.val = pte_val_ma(pte);
313 xen_extend_mmu_update(&u);
315 xen_mc_issue(PARAVIRT_LAZY_MMU);
318 /* Assume pteval_t is equivalent to all the other *val_t types. */
319 static pteval_t pte_mfn_to_pfn(pteval_t val)
321 if (val & _PAGE_PRESENT) {
322 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
323 unsigned long pfn = mfn_to_pfn(mfn);
325 pteval_t flags = val & PTE_FLAGS_MASK;
326 if (unlikely(pfn == ~0))
327 val = flags & ~_PAGE_PRESENT;
329 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
335 static pteval_t pte_pfn_to_mfn(pteval_t val)
337 if (val & _PAGE_PRESENT) {
338 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
339 pteval_t flags = val & PTE_FLAGS_MASK;
342 mfn = __pfn_to_mfn(pfn);
345 * If there's no mfn for the pfn, then just create an
346 * empty non-present pte. Unfortunately this loses
347 * information about the original pfn, so
348 * pte_mfn_to_pfn is asymmetric.
350 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
354 mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
355 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
361 __visible pteval_t xen_pte_val(pte_t pte)
363 pteval_t pteval = pte.pte;
365 return pte_mfn_to_pfn(pteval);
367 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
369 __visible pgdval_t xen_pgd_val(pgd_t pgd)
371 return pte_mfn_to_pfn(pgd.pgd);
373 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
375 __visible pte_t xen_make_pte(pteval_t pte)
377 pte = pte_pfn_to_mfn(pte);
379 return native_make_pte(pte);
381 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
383 __visible pgd_t xen_make_pgd(pgdval_t pgd)
385 pgd = pte_pfn_to_mfn(pgd);
386 return native_make_pgd(pgd);
388 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
390 __visible pmdval_t xen_pmd_val(pmd_t pmd)
392 return pte_mfn_to_pfn(pmd.pmd);
394 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
396 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
404 /* ptr may be ioremapped for 64-bit pagetable setup */
405 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
406 u.val = pud_val_ma(val);
407 xen_extend_mmu_update(&u);
409 xen_mc_issue(PARAVIRT_LAZY_MMU);
414 static void xen_set_pud(pud_t *ptr, pud_t val)
416 trace_xen_mmu_set_pud(ptr, val);
418 /* If page is not pinned, we can just update the entry
420 if (!xen_page_pinned(ptr)) {
425 xen_set_pud_hyper(ptr, val);
428 __visible pmd_t xen_make_pmd(pmdval_t pmd)
430 pmd = pte_pfn_to_mfn(pmd);
431 return native_make_pmd(pmd);
433 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
435 __visible pudval_t xen_pud_val(pud_t pud)
437 return pte_mfn_to_pfn(pud.pud);
439 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
441 __visible pud_t xen_make_pud(pudval_t pud)
443 pud = pte_pfn_to_mfn(pud);
445 return native_make_pud(pud);
447 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
449 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
451 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
452 unsigned offset = pgd - pgd_page;
453 pgd_t *user_ptr = NULL;
455 if (offset < pgd_index(USER_LIMIT)) {
456 struct page *page = virt_to_page(pgd_page);
457 user_ptr = (pgd_t *)page->private;
465 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
469 u.ptr = virt_to_machine(ptr).maddr;
470 u.val = p4d_val_ma(val);
471 xen_extend_mmu_update(&u);
475 * Raw hypercall-based set_p4d, intended for in early boot before
476 * there's a page structure. This implies:
477 * 1. The only existing pagetable is the kernel's
478 * 2. It is always pinned
479 * 3. It has no user pagetable attached to it
481 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
487 __xen_set_p4d_hyper(ptr, val);
489 xen_mc_issue(PARAVIRT_LAZY_MMU);
494 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
496 pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
499 trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
501 /* If page is not pinned, we can just update the entry
503 if (!xen_page_pinned(ptr)) {
506 WARN_ON(xen_page_pinned(user_ptr));
507 pgd_val.pgd = p4d_val_ma(val);
513 /* If it's pinned, then we can at least batch the kernel and
514 user updates together. */
517 __xen_set_p4d_hyper(ptr, val);
519 __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
521 xen_mc_issue(PARAVIRT_LAZY_MMU);
524 #if CONFIG_PGTABLE_LEVELS >= 5
525 __visible p4dval_t xen_p4d_val(p4d_t p4d)
527 return pte_mfn_to_pfn(p4d.p4d);
529 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
531 __visible p4d_t xen_make_p4d(p4dval_t p4d)
533 p4d = pte_pfn_to_mfn(p4d);
535 return native_make_p4d(p4d);
537 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
538 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
540 static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
541 void (*func)(struct mm_struct *mm, struct page *,
543 bool last, unsigned long limit)
547 nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
548 for (i = 0; i < nr; i++) {
549 if (!pmd_none(pmd[i]))
550 (*func)(mm, pmd_page(pmd[i]), PT_PTE);
554 static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
555 void (*func)(struct mm_struct *mm, struct page *,
557 bool last, unsigned long limit)
561 nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
562 for (i = 0; i < nr; i++) {
565 if (pud_none(pud[i]))
568 pmd = pmd_offset(&pud[i], 0);
569 if (PTRS_PER_PMD > 1)
570 (*func)(mm, virt_to_page(pmd), PT_PMD);
571 xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
575 static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
576 void (*func)(struct mm_struct *mm, struct page *,
578 bool last, unsigned long limit)
586 pud = pud_offset(p4d, 0);
587 if (PTRS_PER_PUD > 1)
588 (*func)(mm, virt_to_page(pud), PT_PUD);
589 xen_pud_walk(mm, pud, func, last, limit);
593 * (Yet another) pagetable walker. This one is intended for pinning a
594 * pagetable. This means that it walks a pagetable and calls the
595 * callback function on each page it finds making up the page table,
596 * at every level. It walks the entire pagetable, but it only bothers
597 * pinning pte pages which are below limit. In the normal case this
598 * will be STACK_TOP_MAX, but at boot we need to pin up to
601 * We must skip the Xen hole in the middle of the address space, just after
602 * the big x86-64 virtual hole.
604 static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
605 void (*func)(struct mm_struct *mm, struct page *,
610 unsigned hole_low = 0, hole_high = 0;
612 /* The limit is the last byte to be touched */
614 BUG_ON(limit >= FIXADDR_TOP);
617 * 64-bit has a great big hole in the middle of the address
618 * space, which contains the Xen mappings.
620 hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
621 hole_high = pgd_index(GUARD_HOLE_END_ADDR);
623 nr = pgd_index(limit) + 1;
624 for (i = 0; i < nr; i++) {
627 if (i >= hole_low && i < hole_high)
630 if (pgd_none(pgd[i]))
633 p4d = p4d_offset(&pgd[i], 0);
634 xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
637 /* Do the top level last, so that the callbacks can use it as
638 a cue to do final things like tlb flushes. */
639 (*func)(mm, virt_to_page(pgd), PT_PGD);
642 static void xen_pgd_walk(struct mm_struct *mm,
643 void (*func)(struct mm_struct *mm, struct page *,
647 __xen_pgd_walk(mm, mm->pgd, func, limit);
650 /* If we're using split pte locks, then take the page's lock and
651 return a pointer to it. Otherwise return NULL. */
652 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
654 spinlock_t *ptl = NULL;
656 #if USE_SPLIT_PTE_PTLOCKS
657 ptl = ptlock_ptr(page);
658 spin_lock_nest_lock(ptl, &mm->page_table_lock);
664 static void xen_pte_unlock(void *v)
670 static void xen_do_pin(unsigned level, unsigned long pfn)
675 op.arg1.mfn = pfn_to_mfn(pfn);
677 xen_extend_mmuext_op(&op);
680 static void xen_pin_page(struct mm_struct *mm, struct page *page,
683 unsigned pgfl = TestSetPagePinned(page);
686 void *pt = lowmem_page_address(page);
687 unsigned long pfn = page_to_pfn(page);
688 struct multicall_space mcs = __xen_mc_entry(0);
692 * We need to hold the pagetable lock between the time
693 * we make the pagetable RO and when we actually pin
694 * it. If we don't, then other users may come in and
695 * attempt to update the pagetable by writing it,
696 * which will fail because the memory is RO but not
697 * pinned, so Xen won't do the trap'n'emulate.
699 * If we're using split pte locks, we can't hold the
700 * entire pagetable's worth of locks during the
701 * traverse, because we may wrap the preempt count (8
702 * bits). The solution is to mark RO and pin each PTE
703 * page while holding the lock. This means the number
704 * of locks we end up holding is never more than a
705 * batch size (~32 entries, at present).
707 * If we're not using split pte locks, we needn't pin
708 * the PTE pages independently, because we're
709 * protected by the overall pagetable lock.
713 ptl = xen_pte_lock(page, mm);
715 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
716 pfn_pte(pfn, PAGE_KERNEL_RO),
717 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
720 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
722 /* Queue a deferred unlock for when this batch
724 xen_mc_callback(xen_pte_unlock, ptl);
729 /* This is called just after a mm has been created, but it has not
730 been used yet. We need to make sure that its pagetable is all
731 read-only, and can be pinned. */
732 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
734 pgd_t *user_pgd = xen_get_user_pgd(pgd);
736 trace_xen_mmu_pgd_pin(mm, pgd);
740 __xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
742 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
745 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
746 xen_do_pin(MMUEXT_PIN_L4_TABLE,
747 PFN_DOWN(__pa(user_pgd)));
753 static void xen_pgd_pin(struct mm_struct *mm)
755 __xen_pgd_pin(mm, mm->pgd);
759 * On save, we need to pin all pagetables to make sure they get their
760 * mfns turned into pfns. Search the list for any unpinned pgds and pin
761 * them (unpinned pgds are not currently in use, probably because the
762 * process is under construction or destruction).
764 * Expected to be called in stop_machine() ("equivalent to taking
765 * every spinlock in the system"), so the locking doesn't really
766 * matter all that much.
768 void xen_mm_pin_all(void)
772 spin_lock(&pgd_lock);
774 list_for_each_entry(page, &pgd_list, lru) {
775 if (!PagePinned(page)) {
776 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
777 SetPageSavePinned(page);
781 spin_unlock(&pgd_lock);
784 static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
791 * The init_mm pagetable is really pinned as soon as its created, but
792 * that's before we have page structures to store the bits. So do all
793 * the book-keeping now once struct pages for allocated pages are
794 * initialized. This happens only after memblock_free_all() is called.
796 static void __init xen_after_bootmem(void)
798 static_branch_enable(&xen_struct_pages_ready);
799 SetPagePinned(virt_to_page(level3_user_vsyscall));
800 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
803 static void xen_unpin_page(struct mm_struct *mm, struct page *page,
806 unsigned pgfl = TestClearPagePinned(page);
809 void *pt = lowmem_page_address(page);
810 unsigned long pfn = page_to_pfn(page);
811 spinlock_t *ptl = NULL;
812 struct multicall_space mcs;
815 * Do the converse to pin_page. If we're using split
816 * pte locks, we must be holding the lock for while
817 * the pte page is unpinned but still RO to prevent
818 * concurrent updates from seeing it in this
819 * partially-pinned state.
821 if (level == PT_PTE) {
822 ptl = xen_pte_lock(page, mm);
825 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
828 mcs = __xen_mc_entry(0);
830 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
831 pfn_pte(pfn, PAGE_KERNEL),
832 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
835 /* unlock when batch completed */
836 xen_mc_callback(xen_pte_unlock, ptl);
841 /* Release a pagetables pages back as normal RW */
842 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
844 pgd_t *user_pgd = xen_get_user_pgd(pgd);
846 trace_xen_mmu_pgd_unpin(mm, pgd);
850 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
853 xen_do_pin(MMUEXT_UNPIN_TABLE,
854 PFN_DOWN(__pa(user_pgd)));
855 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
858 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
863 static void xen_pgd_unpin(struct mm_struct *mm)
865 __xen_pgd_unpin(mm, mm->pgd);
869 * On resume, undo any pinning done at save, so that the rest of the
870 * kernel doesn't see any unexpected pinned pagetables.
872 void xen_mm_unpin_all(void)
876 spin_lock(&pgd_lock);
878 list_for_each_entry(page, &pgd_list, lru) {
879 if (PageSavePinned(page)) {
880 BUG_ON(!PagePinned(page));
881 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
882 ClearPageSavePinned(page);
886 spin_unlock(&pgd_lock);
889 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
891 spin_lock(&next->page_table_lock);
893 spin_unlock(&next->page_table_lock);
896 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
898 spin_lock(&mm->page_table_lock);
900 spin_unlock(&mm->page_table_lock);
903 static void drop_mm_ref_this_cpu(void *info)
905 struct mm_struct *mm = info;
907 if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
908 leave_mm(smp_processor_id());
911 * If this cpu still has a stale cr3 reference, then make sure
912 * it has been flushed.
914 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
920 * Another cpu may still have their %cr3 pointing at the pagetable, so
921 * we need to repoint it somewhere else before we can unpin it.
923 static void xen_drop_mm_ref(struct mm_struct *mm)
928 drop_mm_ref_this_cpu(mm);
930 /* Get the "official" set of cpus referring to our pagetable. */
931 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
932 for_each_online_cpu(cpu) {
933 if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
935 smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
941 * It's possible that a vcpu may have a stale reference to our
942 * cr3, because its in lazy mode, and it hasn't yet flushed
943 * its set of pending hypercalls yet. In this case, we can
944 * look at its actual current cr3 value, and force it to flush
948 for_each_online_cpu(cpu) {
949 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
950 cpumask_set_cpu(cpu, mask);
953 smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
954 free_cpumask_var(mask);
957 static void xen_drop_mm_ref(struct mm_struct *mm)
959 drop_mm_ref_this_cpu(mm);
964 * While a process runs, Xen pins its pagetables, which means that the
965 * hypervisor forces it to be read-only, and it controls all updates
966 * to it. This means that all pagetable updates have to go via the
967 * hypervisor, which is moderately expensive.
969 * Since we're pulling the pagetable down, we switch to use init_mm,
970 * unpin old process pagetable and mark it all read-write, which
971 * allows further operations on it to be simple memory accesses.
973 * The only subtle point is that another CPU may be still using the
974 * pagetable because of lazy tlb flushing. This means we need need to
975 * switch all CPUs off this pagetable before we can unpin it.
977 static void xen_exit_mmap(struct mm_struct *mm)
979 get_cpu(); /* make sure we don't move around */
983 spin_lock(&mm->page_table_lock);
985 /* pgd may not be pinned in the error exit path of execve */
986 if (xen_page_pinned(mm->pgd))
989 spin_unlock(&mm->page_table_lock);
992 static void xen_post_allocator_init(void);
994 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
999 op.arg1.mfn = pfn_to_mfn(pfn);
1000 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1004 static void __init xen_cleanhighmap(unsigned long vaddr,
1005 unsigned long vaddr_end)
1007 unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1008 pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1010 /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1011 * We include the PMD passed in on _both_ boundaries. */
1012 for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1013 pmd++, vaddr += PMD_SIZE) {
1016 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1017 set_pmd(pmd, __pmd(0));
1019 /* In case we did something silly, we should crash in this function
1020 * instead of somewhere later and be confusing. */
1025 * Make a page range writeable and free it.
1027 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1029 void *vaddr = __va(paddr);
1030 void *vaddr_end = vaddr + size;
1032 for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1033 make_lowmem_page_readwrite(vaddr);
1035 memblock_free(paddr, size);
1038 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1040 unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1043 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1044 ClearPagePinned(virt_to_page(__va(pa)));
1045 xen_free_ro_pages(pa, PAGE_SIZE);
1048 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1054 if (pmd_large(*pmd)) {
1055 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1056 xen_free_ro_pages(pa, PMD_SIZE);
1060 pte_tbl = pte_offset_kernel(pmd, 0);
1061 for (i = 0; i < PTRS_PER_PTE; i++) {
1062 if (pte_none(pte_tbl[i]))
1064 pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1065 xen_free_ro_pages(pa, PAGE_SIZE);
1067 set_pmd(pmd, __pmd(0));
1068 xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1071 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1077 if (pud_large(*pud)) {
1078 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1079 xen_free_ro_pages(pa, PUD_SIZE);
1083 pmd_tbl = pmd_offset(pud, 0);
1084 for (i = 0; i < PTRS_PER_PMD; i++) {
1085 if (pmd_none(pmd_tbl[i]))
1087 xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1089 set_pud(pud, __pud(0));
1090 xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1093 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1099 if (p4d_large(*p4d)) {
1100 pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1101 xen_free_ro_pages(pa, P4D_SIZE);
1105 pud_tbl = pud_offset(p4d, 0);
1106 for (i = 0; i < PTRS_PER_PUD; i++) {
1107 if (pud_none(pud_tbl[i]))
1109 xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1111 set_p4d(p4d, __p4d(0));
1112 xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1116 * Since it is well isolated we can (and since it is perhaps large we should)
1117 * also free the page tables mapping the initial P->M table.
1119 static void __init xen_cleanmfnmap(unsigned long vaddr)
1125 unpin = (vaddr == 2 * PGDIR_SIZE);
1127 pgd = pgd_offset_k(vaddr);
1128 p4d = p4d_offset(pgd, 0);
1129 if (!p4d_none(*p4d))
1130 xen_cleanmfnmap_p4d(p4d, unpin);
1133 static void __init xen_pagetable_p2m_free(void)
1138 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1140 /* No memory or already called. */
1141 if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1144 /* using __ka address and sticking INVALID_P2M_ENTRY! */
1145 memset((void *)xen_start_info->mfn_list, 0xff, size);
1147 addr = xen_start_info->mfn_list;
1149 * We could be in __ka space.
1150 * We roundup to the PMD, which means that if anybody at this stage is
1151 * using the __ka address of xen_start_info or
1152 * xen_start_info->shared_info they are in going to crash. Fortunatly
1153 * we have already revectored in xen_setup_kernel_pagetable.
1155 size = roundup(size, PMD_SIZE);
1157 if (addr >= __START_KERNEL_map) {
1158 xen_cleanhighmap(addr, addr + size);
1159 size = PAGE_ALIGN(xen_start_info->nr_pages *
1160 sizeof(unsigned long));
1161 memblock_free(__pa(addr), size);
1163 xen_cleanmfnmap(addr);
1167 static void __init xen_pagetable_cleanhighmap(void)
1172 /* At this stage, cleanup_highmap has already cleaned __ka space
1173 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1174 * the ramdisk). We continue on, erasing PMD entries that point to page
1175 * tables - do note that they are accessible at this stage via __va.
1176 * As Xen is aligning the memory end to a 4MB boundary, for good
1177 * measure we also round up to PMD_SIZE * 2 - which means that if
1178 * anybody is using __ka address to the initial boot-stack - and try
1179 * to use it - they are going to crash. The xen_start_info has been
1180 * taken care of already in xen_setup_kernel_pagetable. */
1181 addr = xen_start_info->pt_base;
1182 size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1184 xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1185 xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1188 static void __init xen_pagetable_p2m_setup(void)
1190 xen_vmalloc_p2m_tree();
1192 xen_pagetable_p2m_free();
1194 xen_pagetable_cleanhighmap();
1196 /* And revector! Bye bye old array */
1197 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1200 static void __init xen_pagetable_init(void)
1203 xen_post_allocator_init();
1205 xen_pagetable_p2m_setup();
1207 /* Allocate and initialize top and mid mfn levels for p2m structure */
1208 xen_build_mfn_list_list();
1210 /* Remap memory freed due to conflicts with E820 map */
1212 xen_setup_mfn_list_list();
1214 static void xen_write_cr2(unsigned long cr2)
1216 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1219 static noinline void xen_flush_tlb(void)
1221 struct mmuext_op *op;
1222 struct multicall_space mcs;
1226 mcs = xen_mc_entry(sizeof(*op));
1229 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1230 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1232 xen_mc_issue(PARAVIRT_LAZY_MMU);
1237 static void xen_flush_tlb_one_user(unsigned long addr)
1239 struct mmuext_op *op;
1240 struct multicall_space mcs;
1242 trace_xen_mmu_flush_tlb_one_user(addr);
1246 mcs = xen_mc_entry(sizeof(*op));
1248 op->cmd = MMUEXT_INVLPG_LOCAL;
1249 op->arg1.linear_addr = addr & PAGE_MASK;
1250 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1252 xen_mc_issue(PARAVIRT_LAZY_MMU);
1257 static void xen_flush_tlb_others(const struct cpumask *cpus,
1258 const struct flush_tlb_info *info)
1261 struct mmuext_op op;
1262 DECLARE_BITMAP(mask, NR_CPUS);
1264 struct multicall_space mcs;
1265 const size_t mc_entry_size = sizeof(args->op) +
1266 sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1268 trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end);
1270 if (cpumask_empty(cpus))
1271 return; /* nothing to do */
1273 mcs = xen_mc_entry(mc_entry_size);
1275 args->op.arg2.vcpumask = to_cpumask(args->mask);
1277 /* Remove us, and any offline CPUS. */
1278 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1279 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1281 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1282 if (info->end != TLB_FLUSH_ALL &&
1283 (info->end - info->start) <= PAGE_SIZE) {
1284 args->op.cmd = MMUEXT_INVLPG_MULTI;
1285 args->op.arg1.linear_addr = info->start;
1288 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1290 xen_mc_issue(PARAVIRT_LAZY_MMU);
1293 static unsigned long xen_read_cr3(void)
1295 return this_cpu_read(xen_cr3);
1298 static void set_current_cr3(void *v)
1300 this_cpu_write(xen_current_cr3, (unsigned long)v);
1303 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1305 struct mmuext_op op;
1308 trace_xen_mmu_write_cr3(kernel, cr3);
1311 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1315 WARN_ON(mfn == 0 && kernel);
1317 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1320 xen_extend_mmuext_op(&op);
1323 this_cpu_write(xen_cr3, cr3);
1325 /* Update xen_current_cr3 once the batch has actually
1327 xen_mc_callback(set_current_cr3, (void *)cr3);
1330 static void xen_write_cr3(unsigned long cr3)
1332 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1334 BUG_ON(preemptible());
1336 xen_mc_batch(); /* disables interrupts */
1338 /* Update while interrupts are disabled, so its atomic with
1340 this_cpu_write(xen_cr3, cr3);
1342 __xen_write_cr3(true, cr3);
1345 __xen_write_cr3(false, __pa(user_pgd));
1347 __xen_write_cr3(false, 0);
1349 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1353 * At the start of the day - when Xen launches a guest, it has already
1354 * built pagetables for the guest. We diligently look over them
1355 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1356 * init_top_pgt and its friends. Then when we are happy we load
1357 * the new init_top_pgt - and continue on.
1359 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1360 * up the rest of the pagetables. When it has completed it loads the cr3.
1361 * N.B. that baremetal would start at 'start_kernel' (and the early
1362 * #PF handler would create bootstrap pagetables) - so we are running
1363 * with the same assumptions as what to do when write_cr3 is executed
1366 * Since there are no user-page tables at all, we have two variants
1367 * of xen_write_cr3 - the early bootup (this one), and the late one
1368 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1369 * the Linux kernel and user-space are both in ring 3 while the
1370 * hypervisor is in ring 0.
1372 static void __init xen_write_cr3_init(unsigned long cr3)
1374 BUG_ON(preemptible());
1376 xen_mc_batch(); /* disables interrupts */
1378 /* Update while interrupts are disabled, so its atomic with
1380 this_cpu_write(xen_cr3, cr3);
1382 __xen_write_cr3(true, cr3);
1384 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1387 static int xen_pgd_alloc(struct mm_struct *mm)
1389 pgd_t *pgd = mm->pgd;
1390 struct page *page = virt_to_page(pgd);
1394 BUG_ON(PagePinned(virt_to_page(pgd)));
1395 BUG_ON(page->private != 0);
1397 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1398 page->private = (unsigned long)user_pgd;
1400 if (user_pgd != NULL) {
1401 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1402 user_pgd[pgd_index(VSYSCALL_ADDR)] =
1403 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1408 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1413 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1415 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1418 free_page((unsigned long)user_pgd);
1422 * Init-time set_pte while constructing initial pagetables, which
1423 * doesn't allow RO page table pages to be remapped RW.
1425 * If there is no MFN for this PFN then this page is initially
1426 * ballooned out so clear the PTE (as in decrease_reservation() in
1427 * drivers/xen/balloon.c).
1429 * Many of these PTE updates are done on unpinned and writable pages
1430 * and doing a hypercall for these is unnecessary and expensive. At
1431 * this point it is not possible to tell if a page is pinned or not,
1432 * so always write the PTE directly and rely on Xen trapping and
1433 * emulating any updates as necessary.
1435 __visible pte_t xen_make_pte_init(pteval_t pte)
1440 * Pages belonging to the initial p2m list mapped outside the default
1441 * address range must be mapped read-only. This region contains the
1442 * page tables for mapping the p2m list, too, and page tables MUST be
1445 pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1446 if (xen_start_info->mfn_list < __START_KERNEL_map &&
1447 pfn >= xen_start_info->first_p2m_pfn &&
1448 pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1451 pte = pte_pfn_to_mfn(pte);
1452 return native_make_pte(pte);
1454 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1456 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1458 __xen_set_pte(ptep, pte);
1461 /* Early in boot, while setting up the initial pagetable, assume
1462 everything is pinned. */
1463 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1465 #ifdef CONFIG_FLATMEM
1466 BUG_ON(mem_map); /* should only be used early */
1468 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1469 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1472 /* Used for pmd and pud */
1473 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1475 #ifdef CONFIG_FLATMEM
1476 BUG_ON(mem_map); /* should only be used early */
1478 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1481 /* Early release_pte assumes that all pts are pinned, since there's
1482 only init_mm and anything attached to that is pinned. */
1483 static void __init xen_release_pte_init(unsigned long pfn)
1485 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1486 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1489 static void __init xen_release_pmd_init(unsigned long pfn)
1491 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1494 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1496 struct multicall_space mcs;
1497 struct mmuext_op *op;
1499 mcs = __xen_mc_entry(sizeof(*op));
1502 op->arg1.mfn = pfn_to_mfn(pfn);
1504 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1507 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1509 struct multicall_space mcs;
1510 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1512 mcs = __xen_mc_entry(0);
1513 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1514 pfn_pte(pfn, prot), 0);
1517 /* This needs to make sure the new pte page is pinned iff its being
1518 attached to a pinned pagetable. */
1519 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1522 bool pinned = xen_page_pinned(mm->pgd);
1524 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1527 struct page *page = pfn_to_page(pfn);
1529 if (static_branch_likely(&xen_struct_pages_ready))
1530 SetPagePinned(page);
1534 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1536 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1537 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1539 xen_mc_issue(PARAVIRT_LAZY_MMU);
1543 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1545 xen_alloc_ptpage(mm, pfn, PT_PTE);
1548 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1550 xen_alloc_ptpage(mm, pfn, PT_PMD);
1553 /* This should never happen until we're OK to use struct page */
1554 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1556 struct page *page = pfn_to_page(pfn);
1557 bool pinned = PagePinned(page);
1559 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1564 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1565 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1567 __set_pfn_prot(pfn, PAGE_KERNEL);
1569 xen_mc_issue(PARAVIRT_LAZY_MMU);
1571 ClearPagePinned(page);
1575 static void xen_release_pte(unsigned long pfn)
1577 xen_release_ptpage(pfn, PT_PTE);
1580 static void xen_release_pmd(unsigned long pfn)
1582 xen_release_ptpage(pfn, PT_PMD);
1585 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1587 xen_alloc_ptpage(mm, pfn, PT_PUD);
1590 static void xen_release_pud(unsigned long pfn)
1592 xen_release_ptpage(pfn, PT_PUD);
1596 * Like __va(), but returns address in the kernel mapping (which is
1597 * all we have until the physical memory mapping has been set up.
1599 static void * __init __ka(phys_addr_t paddr)
1601 return (void *)(paddr + __START_KERNEL_map);
1604 /* Convert a machine address to physical address */
1605 static unsigned long __init m2p(phys_addr_t maddr)
1609 maddr &= XEN_PTE_MFN_MASK;
1610 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1615 /* Convert a machine address to kernel virtual */
1616 static void * __init m2v(phys_addr_t maddr)
1618 return __ka(m2p(maddr));
1621 /* Set the page permissions on an identity-mapped pages */
1622 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1623 unsigned long flags)
1625 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1626 pte_t pte = pfn_pte(pfn, prot);
1628 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1631 static void __init set_page_prot(void *addr, pgprot_t prot)
1633 return set_page_prot_flags(addr, prot, UVMF_NONE);
1636 void __init xen_setup_machphys_mapping(void)
1638 struct xen_machphys_mapping mapping;
1640 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1641 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1642 machine_to_phys_nr = mapping.max_mfn + 1;
1644 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1648 static void __init convert_pfn_mfn(void *v)
1653 /* All levels are converted the same way, so just treat them
1655 for (i = 0; i < PTRS_PER_PTE; i++)
1656 pte[i] = xen_make_pte(pte[i].pte);
1658 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1661 if (*pt_base == PFN_DOWN(__pa(addr))) {
1662 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1663 clear_page((void *)addr);
1666 if (*pt_end == PFN_DOWN(__pa(addr))) {
1667 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1668 clear_page((void *)addr);
1673 * Set up the initial kernel pagetable.
1675 * We can construct this by grafting the Xen provided pagetable into
1676 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1677 * level2_ident_pgt, and level2_kernel_pgt. This means that only the
1678 * kernel has a physical mapping to start with - but that's enough to
1679 * get __va working. We need to fill in the rest of the physical
1680 * mapping once some sort of allocator has been set up.
1682 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1686 unsigned long addr[3];
1687 unsigned long pt_base, pt_end;
1690 /* max_pfn_mapped is the last pfn mapped in the initial memory
1691 * mappings. Considering that on Xen after the kernel mappings we
1692 * have the mappings of some pages that don't exist in pfn space, we
1693 * set max_pfn_mapped to the last real pfn mapped. */
1694 if (xen_start_info->mfn_list < __START_KERNEL_map)
1695 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1697 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1699 pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1700 pt_end = pt_base + xen_start_info->nr_pt_frames;
1702 /* Zap identity mapping */
1703 init_top_pgt[0] = __pgd(0);
1705 /* Pre-constructed entries are in pfn, so convert to mfn */
1706 /* L4[273] -> level3_ident_pgt */
1707 /* L4[511] -> level3_kernel_pgt */
1708 convert_pfn_mfn(init_top_pgt);
1710 /* L3_i[0] -> level2_ident_pgt */
1711 convert_pfn_mfn(level3_ident_pgt);
1712 /* L3_k[510] -> level2_kernel_pgt */
1713 /* L3_k[511] -> level2_fixmap_pgt */
1714 convert_pfn_mfn(level3_kernel_pgt);
1716 /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1717 convert_pfn_mfn(level2_fixmap_pgt);
1719 /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1720 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1721 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1723 addr[0] = (unsigned long)pgd;
1724 addr[1] = (unsigned long)l3;
1725 addr[2] = (unsigned long)l2;
1726 /* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1727 * Both L4[273][0] and L4[511][510] have entries that point to the same
1728 * L2 (PMD) tables. Meaning that if you modify it in __va space
1729 * it will be also modified in the __ka space! (But if you just
1730 * modify the PMD table to point to other PTE's or none, then you
1731 * are OK - which is what cleanup_highmap does) */
1732 copy_page(level2_ident_pgt, l2);
1733 /* Graft it onto L4[511][510] */
1734 copy_page(level2_kernel_pgt, l2);
1737 * Zap execute permission from the ident map. Due to the sharing of
1738 * L1 entries we need to do this in the L2.
1740 if (__supported_pte_mask & _PAGE_NX) {
1741 for (i = 0; i < PTRS_PER_PMD; ++i) {
1742 if (pmd_none(level2_ident_pgt[i]))
1744 level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1748 /* Copy the initial P->M table mappings if necessary. */
1749 i = pgd_index(xen_start_info->mfn_list);
1750 if (i && i < pgd_index(__START_KERNEL_map))
1751 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1753 /* Make pagetable pieces RO */
1754 set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1755 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1756 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1757 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1758 set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1759 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1760 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1762 for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1763 set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1767 /* Pin down new L4 */
1768 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1769 PFN_DOWN(__pa_symbol(init_top_pgt)));
1771 /* Unpin Xen-provided one */
1772 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1775 * At this stage there can be no user pgd, and no page structure to
1776 * attach it to, so make sure we just set kernel pgd.
1779 __xen_write_cr3(true, __pa(init_top_pgt));
1780 xen_mc_issue(PARAVIRT_LAZY_CPU);
1782 /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1783 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
1784 * the initial domain. For guests using the toolstack, they are in:
1785 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1786 * rip out the [L4] (pgd), but for guests we shave off three pages.
1788 for (i = 0; i < ARRAY_SIZE(addr); i++)
1789 check_pt_base(&pt_base, &pt_end, addr[i]);
1791 /* Our (by three pages) smaller Xen pagetable that we are using */
1792 xen_pt_base = PFN_PHYS(pt_base);
1793 xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1794 memblock_reserve(xen_pt_base, xen_pt_size);
1796 /* Revector the xen_start_info */
1797 xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1801 * Read a value from a physical address.
1803 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1805 unsigned long *vaddr;
1808 vaddr = early_memremap_ro(addr, sizeof(val));
1810 early_memunmap(vaddr, sizeof(val));
1815 * Translate a virtual address to a physical one without relying on mapped
1816 * page tables. Don't rely on big pages being aligned in (guest) physical
1819 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1828 pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1830 if (!pgd_present(pgd))
1833 pa = pgd_val(pgd) & PTE_PFN_MASK;
1834 pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1836 if (!pud_present(pud))
1838 pa = pud_val(pud) & PTE_PFN_MASK;
1840 return pa + (vaddr & ~PUD_MASK);
1842 pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1844 if (!pmd_present(pmd))
1846 pa = pmd_val(pmd) & PTE_PFN_MASK;
1848 return pa + (vaddr & ~PMD_MASK);
1850 pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1852 if (!pte_present(pte))
1854 pa = pte_pfn(pte) << PAGE_SHIFT;
1856 return pa | (vaddr & ~PAGE_MASK);
1860 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1863 void __init xen_relocate_p2m(void)
1865 phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1866 unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1867 int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1872 unsigned long *new_p2m;
1874 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1875 n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1876 n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1877 n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1878 n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1879 n_frames = n_pte + n_pt + n_pmd + n_pud;
1881 new_area = xen_find_free_area(PFN_PHYS(n_frames));
1883 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1888 * Setup the page tables for addressing the new p2m list.
1889 * We have asked the hypervisor to map the p2m list at the user address
1890 * PUD_SIZE. It may have done so, or it may have used a kernel space
1891 * address depending on the Xen version.
1892 * To avoid any possible virtual address collision, just use
1893 * 2 * PUD_SIZE for the new area.
1895 pud_phys = new_area;
1896 pmd_phys = pud_phys + PFN_PHYS(n_pud);
1897 pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1898 p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1900 pgd = __va(read_cr3_pa());
1901 new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1902 for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1903 pud = early_memremap(pud_phys, PAGE_SIZE);
1905 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1907 pmd = early_memremap(pmd_phys, PAGE_SIZE);
1909 for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1911 pt = early_memremap(pt_phys, PAGE_SIZE);
1914 idx_pte < min(n_pte, PTRS_PER_PTE);
1916 pt[idx_pte] = pfn_pte(p2m_pfn,
1920 n_pte -= PTRS_PER_PTE;
1921 early_memunmap(pt, PAGE_SIZE);
1922 make_lowmem_page_readonly(__va(pt_phys));
1923 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
1925 pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
1926 pt_phys += PAGE_SIZE;
1928 n_pt -= PTRS_PER_PMD;
1929 early_memunmap(pmd, PAGE_SIZE);
1930 make_lowmem_page_readonly(__va(pmd_phys));
1931 pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
1932 PFN_DOWN(pmd_phys));
1933 pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
1934 pmd_phys += PAGE_SIZE;
1936 n_pmd -= PTRS_PER_PUD;
1937 early_memunmap(pud, PAGE_SIZE);
1938 make_lowmem_page_readonly(__va(pud_phys));
1939 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
1940 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
1941 pud_phys += PAGE_SIZE;
1944 /* Now copy the old p2m info to the new area. */
1945 memcpy(new_p2m, xen_p2m_addr, size);
1946 xen_p2m_addr = new_p2m;
1948 /* Release the old p2m list and set new list info. */
1949 p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
1951 p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
1953 if (xen_start_info->mfn_list < __START_KERNEL_map) {
1954 pfn = xen_start_info->first_p2m_pfn;
1955 pfn_end = xen_start_info->first_p2m_pfn +
1956 xen_start_info->nr_p2m_frames;
1957 set_pgd(pgd + 1, __pgd(0));
1960 pfn_end = p2m_pfn_end;
1963 memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
1964 while (pfn < pfn_end) {
1965 if (pfn == p2m_pfn) {
1969 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1973 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1974 xen_start_info->first_p2m_pfn = PFN_DOWN(new_area);
1975 xen_start_info->nr_p2m_frames = n_frames;
1978 void __init xen_reserve_special_pages(void)
1982 memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
1983 if (xen_start_info->store_mfn) {
1984 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
1985 memblock_reserve(paddr, PAGE_SIZE);
1987 if (!xen_initial_domain()) {
1988 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
1989 memblock_reserve(paddr, PAGE_SIZE);
1993 void __init xen_pt_check_e820(void)
1995 if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
1996 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2001 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2003 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2007 phys >>= PAGE_SHIFT;
2010 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2011 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2014 /* All local page mappings */
2015 pte = pfn_pte(phys, prot);
2018 #ifdef CONFIG_X86_LOCAL_APIC
2019 case FIX_APIC_BASE: /* maps dummy local APIC */
2020 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2024 #ifdef CONFIG_X86_IO_APIC
2025 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2027 * We just don't map the IO APIC - all access is via
2028 * hypercalls. Keep the address in the pte for reference.
2030 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2034 case FIX_PARAVIRT_BOOTMAP:
2035 /* This is an MFN, but it isn't an IO mapping from the
2037 pte = mfn_pte(phys, prot);
2041 /* By default, set_fixmap is used for hardware mappings */
2042 pte = mfn_pte(phys, prot);
2046 __native_set_fixmap(idx, pte);
2048 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2049 /* Replicate changes to map the vsyscall page into the user
2050 pagetable vsyscall mapping. */
2051 if (idx == VSYSCALL_PAGE) {
2052 unsigned long vaddr = __fix_to_virt(idx);
2053 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2058 static void __init xen_post_allocator_init(void)
2060 pv_ops.mmu.set_pte = xen_set_pte;
2061 pv_ops.mmu.set_pmd = xen_set_pmd;
2062 pv_ops.mmu.set_pud = xen_set_pud;
2063 pv_ops.mmu.set_p4d = xen_set_p4d;
2065 /* This will work as long as patching hasn't happened yet
2066 (which it hasn't) */
2067 pv_ops.mmu.alloc_pte = xen_alloc_pte;
2068 pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2069 pv_ops.mmu.release_pte = xen_release_pte;
2070 pv_ops.mmu.release_pmd = xen_release_pmd;
2071 pv_ops.mmu.alloc_pud = xen_alloc_pud;
2072 pv_ops.mmu.release_pud = xen_release_pud;
2073 pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2075 pv_ops.mmu.write_cr3 = &xen_write_cr3;
2078 static void xen_leave_lazy_mmu(void)
2082 paravirt_leave_lazy_mmu();
2086 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2087 .read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2088 .write_cr2 = xen_write_cr2,
2090 .read_cr3 = xen_read_cr3,
2091 .write_cr3 = xen_write_cr3_init,
2093 .flush_tlb_user = xen_flush_tlb,
2094 .flush_tlb_kernel = xen_flush_tlb,
2095 .flush_tlb_one_user = xen_flush_tlb_one_user,
2096 .flush_tlb_others = xen_flush_tlb_others,
2097 .tlb_remove_table = tlb_remove_table,
2099 .pgd_alloc = xen_pgd_alloc,
2100 .pgd_free = xen_pgd_free,
2102 .alloc_pte = xen_alloc_pte_init,
2103 .release_pte = xen_release_pte_init,
2104 .alloc_pmd = xen_alloc_pmd_init,
2105 .release_pmd = xen_release_pmd_init,
2107 .set_pte = xen_set_pte_init,
2108 .set_pte_at = xen_set_pte_at,
2109 .set_pmd = xen_set_pmd_hyper,
2111 .ptep_modify_prot_start = __ptep_modify_prot_start,
2112 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2114 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2115 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2117 .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2118 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2120 .set_pud = xen_set_pud_hyper,
2122 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2123 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2125 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2126 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2127 .set_p4d = xen_set_p4d_hyper,
2129 .alloc_pud = xen_alloc_pmd_init,
2130 .release_pud = xen_release_pmd_init,
2132 #if CONFIG_PGTABLE_LEVELS >= 5
2133 .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2134 .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2137 .activate_mm = xen_activate_mm,
2138 .dup_mmap = xen_dup_mmap,
2139 .exit_mmap = xen_exit_mmap,
2142 .enter = paravirt_enter_lazy_mmu,
2143 .leave = xen_leave_lazy_mmu,
2144 .flush = paravirt_flush_lazy_mmu,
2147 .set_fixmap = xen_set_fixmap,
2150 void __init xen_init_mmu_ops(void)
2152 x86_init.paging.pagetable_init = xen_pagetable_init;
2153 x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2155 pv_ops.mmu = xen_mmu_ops;
2157 memset(dummy_mapping, 0xff, PAGE_SIZE);
2160 /* Protected by xen_reservation_lock. */
2161 #define MAX_CONTIG_ORDER 9 /* 2MB */
2162 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2164 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2165 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2166 unsigned long *in_frames,
2167 unsigned long *out_frames)
2170 struct multicall_space mcs;
2173 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2174 mcs = __xen_mc_entry(0);
2177 in_frames[i] = virt_to_mfn(vaddr);
2179 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2180 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2183 out_frames[i] = virt_to_pfn(vaddr);
2189 * Update the pfn-to-mfn mappings for a virtual address range, either to
2190 * point to an array of mfns, or contiguously from a single starting
2193 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2194 unsigned long *mfns,
2195 unsigned long first_mfn)
2202 limit = 1u << order;
2203 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2204 struct multicall_space mcs;
2207 mcs = __xen_mc_entry(0);
2211 mfn = first_mfn + i;
2213 if (i < (limit - 1))
2217 flags = UVMF_INVLPG | UVMF_ALL;
2219 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2222 MULTI_update_va_mapping(mcs.mc, vaddr,
2223 mfn_pte(mfn, PAGE_KERNEL), flags);
2225 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2232 * Perform the hypercall to exchange a region of our pfns to point to
2233 * memory with the required contiguous alignment. Takes the pfns as
2234 * input, and populates mfns as output.
2236 * Returns a success code indicating whether the hypervisor was able to
2237 * satisfy the request or not.
2239 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2240 unsigned long *pfns_in,
2241 unsigned long extents_out,
2242 unsigned int order_out,
2243 unsigned long *mfns_out,
2244 unsigned int address_bits)
2249 struct xen_memory_exchange exchange = {
2251 .nr_extents = extents_in,
2252 .extent_order = order_in,
2253 .extent_start = pfns_in,
2257 .nr_extents = extents_out,
2258 .extent_order = order_out,
2259 .extent_start = mfns_out,
2260 .address_bits = address_bits,
2265 BUG_ON(extents_in << order_in != extents_out << order_out);
2267 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2268 success = (exchange.nr_exchanged == extents_in);
2270 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2271 BUG_ON(success && (rc != 0));
2276 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2277 unsigned int address_bits,
2278 dma_addr_t *dma_handle)
2280 unsigned long *in_frames = discontig_frames, out_frame;
2281 unsigned long flags;
2283 unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2286 * Currently an auto-translated guest will not perform I/O, nor will
2287 * it require PAE page directories below 4GB. Therefore any calls to
2288 * this function are redundant and can be ignored.
2291 if (unlikely(order > MAX_CONTIG_ORDER))
2294 memset((void *) vstart, 0, PAGE_SIZE << order);
2296 spin_lock_irqsave(&xen_reservation_lock, flags);
2298 /* 1. Zap current PTEs, remembering MFNs. */
2299 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2301 /* 2. Get a new contiguous memory extent. */
2302 out_frame = virt_to_pfn(vstart);
2303 success = xen_exchange_memory(1UL << order, 0, in_frames,
2304 1, order, &out_frame,
2307 /* 3. Map the new extent in place of old pages. */
2309 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2311 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2313 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2315 *dma_handle = virt_to_machine(vstart).maddr;
2316 return success ? 0 : -ENOMEM;
2319 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2321 unsigned long *out_frames = discontig_frames, in_frame;
2322 unsigned long flags;
2324 unsigned long vstart;
2326 if (unlikely(order > MAX_CONTIG_ORDER))
2329 vstart = (unsigned long)phys_to_virt(pstart);
2330 memset((void *) vstart, 0, PAGE_SIZE << order);
2332 spin_lock_irqsave(&xen_reservation_lock, flags);
2334 /* 1. Find start MFN of contiguous extent. */
2335 in_frame = virt_to_mfn(vstart);
2337 /* 2. Zap current PTEs. */
2338 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2340 /* 3. Do the exchange for non-contiguous MFNs. */
2341 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2344 /* 4. Map new pages in place of old pages. */
2346 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2348 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2350 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2353 static noinline void xen_flush_tlb_all(void)
2355 struct mmuext_op *op;
2356 struct multicall_space mcs;
2360 mcs = xen_mc_entry(sizeof(*op));
2363 op->cmd = MMUEXT_TLB_FLUSH_ALL;
2364 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2366 xen_mc_issue(PARAVIRT_LAZY_MMU);
2371 #define REMAP_BATCH_SIZE 16
2378 struct mmu_update *mmu_update;
2381 static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2383 struct remap_data *rmd = data;
2384 pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2387 * If we have a contiguous range, just update the pfn itself,
2388 * else update pointer to be "next pfn".
2390 if (rmd->contiguous)
2395 rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2396 rmd->mmu_update->ptr |= rmd->no_translate ?
2397 MMU_PT_UPDATE_NO_TRANSLATE :
2398 MMU_NORMAL_PT_UPDATE;
2399 rmd->mmu_update->val = pte_val_ma(pte);
2405 int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2406 xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2407 unsigned int domid, bool no_translate, struct page **pages)
2410 struct remap_data rmd;
2411 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2412 unsigned long range;
2415 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2420 * We use the err_ptr to indicate if there we are doing a contiguous
2421 * mapping or a discontigious mapping.
2423 rmd.contiguous = !err_ptr;
2424 rmd.no_translate = no_translate;
2429 int batch = min(REMAP_BATCH_SIZE, nr);
2430 int batch_left = batch;
2432 range = (unsigned long)batch << PAGE_SHIFT;
2434 rmd.mmu_update = mmu_update;
2435 err = apply_to_page_range(vma->vm_mm, addr, range,
2436 remap_area_pfn_pte_fn, &rmd);
2441 * We record the error for each page that gives an error, but
2442 * continue mapping until the whole set is done
2447 err = HYPERVISOR_mmu_update(&mmu_update[index],
2448 batch_left, &done, domid);
2451 * @err_ptr may be the same buffer as @gfn, so
2452 * only clear it after each chunk of @gfn is
2456 for (i = index; i < index + done; i++)
2463 done++; /* Skip failed frame. */
2468 } while (batch_left);
2478 xen_flush_tlb_all();
2480 return err < 0 ? err : mapped;
2482 EXPORT_SYMBOL_GPL(xen_remap_pfn);
2484 #ifdef CONFIG_KEXEC_CORE
2485 phys_addr_t paddr_vmcoreinfo_note(void)
2487 if (xen_pv_domain())
2488 return virt_to_machine(vmcoreinfo_note).maddr;
2490 return __pa(vmcoreinfo_note);
2492 #endif /* CONFIG_KEXEC_CORE */