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.
41 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
43 #include <linux/sched/mm.h>
44 #include <linux/debugfs.h>
45 #include <linux/bug.h>
46 #include <linux/vmalloc.h>
47 #include <linux/export.h>
48 #include <linux/init.h>
49 #include <linux/gfp.h>
50 #include <linux/memblock.h>
51 #include <linux/seq_file.h>
52 #include <linux/crash_dump.h>
53 #include <linux/pgtable.h>
54 #ifdef CONFIG_KEXEC_CORE
55 #include <linux/kexec.h>
58 #include <trace/events/xen.h>
60 #include <asm/tlbflush.h>
61 #include <asm/fixmap.h>
62 #include <asm/mmu_context.h>
63 #include <asm/setup.h>
64 #include <asm/paravirt.h>
65 #include <asm/e820/api.h>
66 #include <asm/linkage.h>
69 #include <asm/memtype.h>
73 #include <asm/xen/hypercall.h>
74 #include <asm/xen/hypervisor.h>
78 #include <xen/interface/xen.h>
79 #include <xen/interface/hvm/hvm_op.h>
80 #include <xen/interface/version.h>
81 #include <xen/interface/memory.h>
82 #include <xen/hvc-console.h>
83 #include <xen/swiotlb-xen.h>
85 #include "multicalls.h"
90 * Prototypes for functions called via PV_CALLEE_SAVE_REGS_THUNK() in order
91 * to avoid warnings with "-Wmissing-prototypes".
93 pteval_t xen_pte_val(pte_t pte);
94 pgdval_t xen_pgd_val(pgd_t pgd);
95 pmdval_t xen_pmd_val(pmd_t pmd);
96 pudval_t xen_pud_val(pud_t pud);
97 p4dval_t xen_p4d_val(p4d_t p4d);
98 pte_t xen_make_pte(pteval_t pte);
99 pgd_t xen_make_pgd(pgdval_t pgd);
100 pmd_t xen_make_pmd(pmdval_t pmd);
101 pud_t xen_make_pud(pudval_t pud);
102 p4d_t xen_make_p4d(p4dval_t p4d);
103 pte_t xen_make_pte_init(pteval_t pte);
105 #ifdef CONFIG_X86_VSYSCALL_EMULATION
106 /* l3 pud for userspace vsyscall mapping */
107 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
111 * Protects atomic reservation decrease/increase against concurrent increases.
112 * Also protects non-atomic updates of current_pages and balloon lists.
114 static DEFINE_SPINLOCK(xen_reservation_lock);
117 * Note about cr3 (pagetable base) values:
119 * xen_cr3 contains the current logical cr3 value; it contains the
120 * last set cr3. This may not be the current effective cr3, because
121 * its update may be being lazily deferred. However, a vcpu looking
122 * at its own cr3 can use this value knowing that it everything will
123 * be self-consistent.
125 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
126 * hypercall to set the vcpu cr3 is complete (so it may be a little
127 * out of date, but it will never be set early). If one vcpu is
128 * looking at another vcpu's cr3 value, it should use this variable.
130 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
131 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
133 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
135 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
138 * Just beyond the highest usermode address. STACK_TOP_MAX has a
139 * redzone above it, so round it up to a PGD boundary.
141 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
143 void make_lowmem_page_readonly(void *vaddr)
146 unsigned long address = (unsigned long)vaddr;
149 pte = lookup_address(address, &level);
151 return; /* vaddr missing */
153 ptev = pte_wrprotect(*pte);
155 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
159 void make_lowmem_page_readwrite(void *vaddr)
162 unsigned long address = (unsigned long)vaddr;
165 pte = lookup_address(address, &level);
167 return; /* vaddr missing */
169 ptev = pte_mkwrite(*pte);
171 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
177 * During early boot all page table pages are pinned, but we do not have struct
178 * pages, so return true until struct pages are ready.
180 static bool xen_page_pinned(void *ptr)
182 if (static_branch_likely(&xen_struct_pages_ready)) {
183 struct page *page = virt_to_page(ptr);
185 return PagePinned(page);
190 static void xen_extend_mmu_update(const struct mmu_update *update)
192 struct multicall_space mcs;
193 struct mmu_update *u;
195 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
197 if (mcs.mc != NULL) {
200 mcs = __xen_mc_entry(sizeof(*u));
201 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
208 static void xen_extend_mmuext_op(const struct mmuext_op *op)
210 struct multicall_space mcs;
213 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
215 if (mcs.mc != NULL) {
218 mcs = __xen_mc_entry(sizeof(*u));
219 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
226 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
234 /* ptr may be ioremapped for 64-bit pagetable setup */
235 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
236 u.val = pmd_val_ma(val);
237 xen_extend_mmu_update(&u);
239 xen_mc_issue(PARAVIRT_LAZY_MMU);
244 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
246 trace_xen_mmu_set_pmd(ptr, val);
248 /* If page is not pinned, we can just update the entry
250 if (!xen_page_pinned(ptr)) {
255 xen_set_pmd_hyper(ptr, val);
259 * Associate a virtual page frame with a given physical page frame
260 * and protection flags for that frame.
262 void __init set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
264 if (HYPERVISOR_update_va_mapping(vaddr, mfn_pte(mfn, flags),
269 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
273 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
278 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
279 u.val = pte_val_ma(pteval);
280 xen_extend_mmu_update(&u);
282 xen_mc_issue(PARAVIRT_LAZY_MMU);
287 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
289 if (!xen_batched_set_pte(ptep, pteval)) {
291 * Could call native_set_pte() here and trap and
292 * emulate the PTE write, but a hypercall is much cheaper.
296 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
297 u.val = pte_val_ma(pteval);
298 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
302 static void xen_set_pte(pte_t *ptep, pte_t pteval)
304 trace_xen_mmu_set_pte(ptep, pteval);
305 __xen_set_pte(ptep, pteval);
308 pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
309 unsigned long addr, pte_t *ptep)
311 /* Just return the pte as-is. We preserve the bits on commit */
312 trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
316 void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
317 pte_t *ptep, pte_t pte)
321 trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
324 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
325 u.val = pte_val_ma(pte);
326 xen_extend_mmu_update(&u);
328 xen_mc_issue(PARAVIRT_LAZY_MMU);
331 /* Assume pteval_t is equivalent to all the other *val_t types. */
332 static pteval_t pte_mfn_to_pfn(pteval_t val)
334 if (val & _PAGE_PRESENT) {
335 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
336 unsigned long pfn = mfn_to_pfn(mfn);
338 pteval_t flags = val & PTE_FLAGS_MASK;
339 if (unlikely(pfn == ~0))
340 val = flags & ~_PAGE_PRESENT;
342 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
348 static pteval_t pte_pfn_to_mfn(pteval_t val)
350 if (val & _PAGE_PRESENT) {
351 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
352 pteval_t flags = val & PTE_FLAGS_MASK;
355 mfn = __pfn_to_mfn(pfn);
358 * If there's no mfn for the pfn, then just create an
359 * empty non-present pte. Unfortunately this loses
360 * information about the original pfn, so
361 * pte_mfn_to_pfn is asymmetric.
363 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
367 mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
368 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
374 __visible pteval_t xen_pte_val(pte_t pte)
376 pteval_t pteval = pte.pte;
378 return pte_mfn_to_pfn(pteval);
380 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
382 __visible pgdval_t xen_pgd_val(pgd_t pgd)
384 return pte_mfn_to_pfn(pgd.pgd);
386 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
388 __visible pte_t xen_make_pte(pteval_t pte)
390 pte = pte_pfn_to_mfn(pte);
392 return native_make_pte(pte);
394 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
396 __visible pgd_t xen_make_pgd(pgdval_t pgd)
398 pgd = pte_pfn_to_mfn(pgd);
399 return native_make_pgd(pgd);
401 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
403 __visible pmdval_t xen_pmd_val(pmd_t pmd)
405 return pte_mfn_to_pfn(pmd.pmd);
407 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
409 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
417 /* ptr may be ioremapped for 64-bit pagetable setup */
418 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
419 u.val = pud_val_ma(val);
420 xen_extend_mmu_update(&u);
422 xen_mc_issue(PARAVIRT_LAZY_MMU);
427 static void xen_set_pud(pud_t *ptr, pud_t val)
429 trace_xen_mmu_set_pud(ptr, val);
431 /* If page is not pinned, we can just update the entry
433 if (!xen_page_pinned(ptr)) {
438 xen_set_pud_hyper(ptr, val);
441 __visible pmd_t xen_make_pmd(pmdval_t pmd)
443 pmd = pte_pfn_to_mfn(pmd);
444 return native_make_pmd(pmd);
446 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
448 __visible pudval_t xen_pud_val(pud_t pud)
450 return pte_mfn_to_pfn(pud.pud);
452 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
454 __visible pud_t xen_make_pud(pudval_t pud)
456 pud = pte_pfn_to_mfn(pud);
458 return native_make_pud(pud);
460 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
462 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
464 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
465 unsigned offset = pgd - pgd_page;
466 pgd_t *user_ptr = NULL;
468 if (offset < pgd_index(USER_LIMIT)) {
469 struct page *page = virt_to_page(pgd_page);
470 user_ptr = (pgd_t *)page->private;
478 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
482 u.ptr = virt_to_machine(ptr).maddr;
483 u.val = p4d_val_ma(val);
484 xen_extend_mmu_update(&u);
488 * Raw hypercall-based set_p4d, intended for in early boot before
489 * there's a page structure. This implies:
490 * 1. The only existing pagetable is the kernel's
491 * 2. It is always pinned
492 * 3. It has no user pagetable attached to it
494 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
500 __xen_set_p4d_hyper(ptr, val);
502 xen_mc_issue(PARAVIRT_LAZY_MMU);
507 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
509 pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
512 trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
514 /* If page is not pinned, we can just update the entry
516 if (!xen_page_pinned(ptr)) {
519 WARN_ON(xen_page_pinned(user_ptr));
520 pgd_val.pgd = p4d_val_ma(val);
526 /* If it's pinned, then we can at least batch the kernel and
527 user updates together. */
530 __xen_set_p4d_hyper(ptr, val);
532 __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
534 xen_mc_issue(PARAVIRT_LAZY_MMU);
537 #if CONFIG_PGTABLE_LEVELS >= 5
538 __visible p4dval_t xen_p4d_val(p4d_t p4d)
540 return pte_mfn_to_pfn(p4d.p4d);
542 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
544 __visible p4d_t xen_make_p4d(p4dval_t p4d)
546 p4d = pte_pfn_to_mfn(p4d);
548 return native_make_p4d(p4d);
550 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
551 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
553 static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
554 void (*func)(struct mm_struct *mm, struct page *,
556 bool last, unsigned long limit)
560 nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
561 for (i = 0; i < nr; i++) {
562 if (!pmd_none(pmd[i]))
563 (*func)(mm, pmd_page(pmd[i]), PT_PTE);
567 static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
568 void (*func)(struct mm_struct *mm, struct page *,
570 bool last, unsigned long limit)
574 nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
575 for (i = 0; i < nr; i++) {
578 if (pud_none(pud[i]))
581 pmd = pmd_offset(&pud[i], 0);
582 if (PTRS_PER_PMD > 1)
583 (*func)(mm, virt_to_page(pmd), PT_PMD);
584 xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
588 static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
589 void (*func)(struct mm_struct *mm, struct page *,
591 bool last, unsigned long limit)
599 pud = pud_offset(p4d, 0);
600 if (PTRS_PER_PUD > 1)
601 (*func)(mm, virt_to_page(pud), PT_PUD);
602 xen_pud_walk(mm, pud, func, last, limit);
606 * (Yet another) pagetable walker. This one is intended for pinning a
607 * pagetable. This means that it walks a pagetable and calls the
608 * callback function on each page it finds making up the page table,
609 * at every level. It walks the entire pagetable, but it only bothers
610 * pinning pte pages which are below limit. In the normal case this
611 * will be STACK_TOP_MAX, but at boot we need to pin up to
614 * We must skip the Xen hole in the middle of the address space, just after
615 * the big x86-64 virtual hole.
617 static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
618 void (*func)(struct mm_struct *mm, struct page *,
623 unsigned hole_low = 0, hole_high = 0;
625 /* The limit is the last byte to be touched */
627 BUG_ON(limit >= FIXADDR_TOP);
630 * 64-bit has a great big hole in the middle of the address
631 * space, which contains the Xen mappings.
633 hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
634 hole_high = pgd_index(GUARD_HOLE_END_ADDR);
636 nr = pgd_index(limit) + 1;
637 for (i = 0; i < nr; i++) {
640 if (i >= hole_low && i < hole_high)
643 if (pgd_none(pgd[i]))
646 p4d = p4d_offset(&pgd[i], 0);
647 xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
650 /* Do the top level last, so that the callbacks can use it as
651 a cue to do final things like tlb flushes. */
652 (*func)(mm, virt_to_page(pgd), PT_PGD);
655 static void xen_pgd_walk(struct mm_struct *mm,
656 void (*func)(struct mm_struct *mm, struct page *,
660 __xen_pgd_walk(mm, mm->pgd, func, limit);
663 /* If we're using split pte locks, then take the page's lock and
664 return a pointer to it. Otherwise return NULL. */
665 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
667 spinlock_t *ptl = NULL;
669 #if USE_SPLIT_PTE_PTLOCKS
670 ptl = ptlock_ptr(page);
671 spin_lock_nest_lock(ptl, &mm->page_table_lock);
677 static void xen_pte_unlock(void *v)
683 static void xen_do_pin(unsigned level, unsigned long pfn)
688 op.arg1.mfn = pfn_to_mfn(pfn);
690 xen_extend_mmuext_op(&op);
693 static void xen_pin_page(struct mm_struct *mm, struct page *page,
696 unsigned pgfl = TestSetPagePinned(page);
699 void *pt = lowmem_page_address(page);
700 unsigned long pfn = page_to_pfn(page);
701 struct multicall_space mcs = __xen_mc_entry(0);
705 * We need to hold the pagetable lock between the time
706 * we make the pagetable RO and when we actually pin
707 * it. If we don't, then other users may come in and
708 * attempt to update the pagetable by writing it,
709 * which will fail because the memory is RO but not
710 * pinned, so Xen won't do the trap'n'emulate.
712 * If we're using split pte locks, we can't hold the
713 * entire pagetable's worth of locks during the
714 * traverse, because we may wrap the preempt count (8
715 * bits). The solution is to mark RO and pin each PTE
716 * page while holding the lock. This means the number
717 * of locks we end up holding is never more than a
718 * batch size (~32 entries, at present).
720 * If we're not using split pte locks, we needn't pin
721 * the PTE pages independently, because we're
722 * protected by the overall pagetable lock.
726 ptl = xen_pte_lock(page, mm);
728 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
729 pfn_pte(pfn, PAGE_KERNEL_RO),
730 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
733 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
735 /* Queue a deferred unlock for when this batch
737 xen_mc_callback(xen_pte_unlock, ptl);
742 /* This is called just after a mm has been created, but it has not
743 been used yet. We need to make sure that its pagetable is all
744 read-only, and can be pinned. */
745 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
747 pgd_t *user_pgd = xen_get_user_pgd(pgd);
749 trace_xen_mmu_pgd_pin(mm, pgd);
753 __xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
755 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
758 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
759 xen_do_pin(MMUEXT_PIN_L4_TABLE,
760 PFN_DOWN(__pa(user_pgd)));
766 static void xen_pgd_pin(struct mm_struct *mm)
768 __xen_pgd_pin(mm, mm->pgd);
772 * On save, we need to pin all pagetables to make sure they get their
773 * mfns turned into pfns. Search the list for any unpinned pgds and pin
774 * them (unpinned pgds are not currently in use, probably because the
775 * process is under construction or destruction).
777 * Expected to be called in stop_machine() ("equivalent to taking
778 * every spinlock in the system"), so the locking doesn't really
779 * matter all that much.
781 void xen_mm_pin_all(void)
785 spin_lock(&pgd_lock);
787 list_for_each_entry(page, &pgd_list, lru) {
788 if (!PagePinned(page)) {
789 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
790 SetPageSavePinned(page);
794 spin_unlock(&pgd_lock);
797 static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
804 * The init_mm pagetable is really pinned as soon as its created, but
805 * that's before we have page structures to store the bits. So do all
806 * the book-keeping now once struct pages for allocated pages are
807 * initialized. This happens only after memblock_free_all() is called.
809 static void __init xen_after_bootmem(void)
811 static_branch_enable(&xen_struct_pages_ready);
812 #ifdef CONFIG_X86_VSYSCALL_EMULATION
813 SetPagePinned(virt_to_page(level3_user_vsyscall));
815 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
818 static void xen_unpin_page(struct mm_struct *mm, struct page *page,
821 unsigned pgfl = TestClearPagePinned(page);
824 void *pt = lowmem_page_address(page);
825 unsigned long pfn = page_to_pfn(page);
826 spinlock_t *ptl = NULL;
827 struct multicall_space mcs;
830 * Do the converse to pin_page. If we're using split
831 * pte locks, we must be holding the lock for while
832 * the pte page is unpinned but still RO to prevent
833 * concurrent updates from seeing it in this
834 * partially-pinned state.
836 if (level == PT_PTE) {
837 ptl = xen_pte_lock(page, mm);
840 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
843 mcs = __xen_mc_entry(0);
845 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
846 pfn_pte(pfn, PAGE_KERNEL),
847 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
850 /* unlock when batch completed */
851 xen_mc_callback(xen_pte_unlock, ptl);
856 /* Release a pagetables pages back as normal RW */
857 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
859 pgd_t *user_pgd = xen_get_user_pgd(pgd);
861 trace_xen_mmu_pgd_unpin(mm, pgd);
865 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
868 xen_do_pin(MMUEXT_UNPIN_TABLE,
869 PFN_DOWN(__pa(user_pgd)));
870 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
873 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
878 static void xen_pgd_unpin(struct mm_struct *mm)
880 __xen_pgd_unpin(mm, mm->pgd);
884 * On resume, undo any pinning done at save, so that the rest of the
885 * kernel doesn't see any unexpected pinned pagetables.
887 void xen_mm_unpin_all(void)
891 spin_lock(&pgd_lock);
893 list_for_each_entry(page, &pgd_list, lru) {
894 if (PageSavePinned(page)) {
895 BUG_ON(!PagePinned(page));
896 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
897 ClearPageSavePinned(page);
901 spin_unlock(&pgd_lock);
904 static void xen_enter_mmap(struct mm_struct *mm)
906 spin_lock(&mm->page_table_lock);
908 spin_unlock(&mm->page_table_lock);
911 static void drop_mm_ref_this_cpu(void *info)
913 struct mm_struct *mm = info;
915 if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
916 leave_mm(smp_processor_id());
919 * If this cpu still has a stale cr3 reference, then make sure
920 * it has been flushed.
922 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
928 * Another cpu may still have their %cr3 pointing at the pagetable, so
929 * we need to repoint it somewhere else before we can unpin it.
931 static void xen_drop_mm_ref(struct mm_struct *mm)
936 drop_mm_ref_this_cpu(mm);
938 /* Get the "official" set of cpus referring to our pagetable. */
939 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
940 for_each_online_cpu(cpu) {
941 if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
943 smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
949 * It's possible that a vcpu may have a stale reference to our
950 * cr3, because its in lazy mode, and it hasn't yet flushed
951 * its set of pending hypercalls yet. In this case, we can
952 * look at its actual current cr3 value, and force it to flush
956 for_each_online_cpu(cpu) {
957 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
958 cpumask_set_cpu(cpu, mask);
961 smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
962 free_cpumask_var(mask);
965 static void xen_drop_mm_ref(struct mm_struct *mm)
967 drop_mm_ref_this_cpu(mm);
972 * While a process runs, Xen pins its pagetables, which means that the
973 * hypervisor forces it to be read-only, and it controls all updates
974 * to it. This means that all pagetable updates have to go via the
975 * hypervisor, which is moderately expensive.
977 * Since we're pulling the pagetable down, we switch to use init_mm,
978 * unpin old process pagetable and mark it all read-write, which
979 * allows further operations on it to be simple memory accesses.
981 * The only subtle point is that another CPU may be still using the
982 * pagetable because of lazy tlb flushing. This means we need need to
983 * switch all CPUs off this pagetable before we can unpin it.
985 static void xen_exit_mmap(struct mm_struct *mm)
987 get_cpu(); /* make sure we don't move around */
991 spin_lock(&mm->page_table_lock);
993 /* pgd may not be pinned in the error exit path of execve */
994 if (xen_page_pinned(mm->pgd))
997 spin_unlock(&mm->page_table_lock);
1000 static void xen_post_allocator_init(void);
1002 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1004 struct mmuext_op op;
1007 op.arg1.mfn = pfn_to_mfn(pfn);
1008 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1012 static void __init xen_cleanhighmap(unsigned long vaddr,
1013 unsigned long vaddr_end)
1015 unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1016 pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1018 /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1019 * We include the PMD passed in on _both_ boundaries. */
1020 for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1021 pmd++, vaddr += PMD_SIZE) {
1024 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1025 set_pmd(pmd, __pmd(0));
1027 /* In case we did something silly, we should crash in this function
1028 * instead of somewhere later and be confusing. */
1033 * Make a page range writeable and free it.
1035 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1037 void *vaddr = __va(paddr);
1038 void *vaddr_end = vaddr + size;
1040 for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1041 make_lowmem_page_readwrite(vaddr);
1043 memblock_phys_free(paddr, size);
1046 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1048 unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1051 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1052 ClearPagePinned(virt_to_page(__va(pa)));
1053 xen_free_ro_pages(pa, PAGE_SIZE);
1056 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1062 if (pmd_large(*pmd)) {
1063 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1064 xen_free_ro_pages(pa, PMD_SIZE);
1068 pte_tbl = pte_offset_kernel(pmd, 0);
1069 for (i = 0; i < PTRS_PER_PTE; i++) {
1070 if (pte_none(pte_tbl[i]))
1072 pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1073 xen_free_ro_pages(pa, PAGE_SIZE);
1075 set_pmd(pmd, __pmd(0));
1076 xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1079 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1085 if (pud_large(*pud)) {
1086 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1087 xen_free_ro_pages(pa, PUD_SIZE);
1091 pmd_tbl = pmd_offset(pud, 0);
1092 for (i = 0; i < PTRS_PER_PMD; i++) {
1093 if (pmd_none(pmd_tbl[i]))
1095 xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1097 set_pud(pud, __pud(0));
1098 xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1101 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1107 if (p4d_large(*p4d)) {
1108 pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1109 xen_free_ro_pages(pa, P4D_SIZE);
1113 pud_tbl = pud_offset(p4d, 0);
1114 for (i = 0; i < PTRS_PER_PUD; i++) {
1115 if (pud_none(pud_tbl[i]))
1117 xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1119 set_p4d(p4d, __p4d(0));
1120 xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1124 * Since it is well isolated we can (and since it is perhaps large we should)
1125 * also free the page tables mapping the initial P->M table.
1127 static void __init xen_cleanmfnmap(unsigned long vaddr)
1133 unpin = (vaddr == 2 * PGDIR_SIZE);
1135 pgd = pgd_offset_k(vaddr);
1136 p4d = p4d_offset(pgd, 0);
1137 if (!p4d_none(*p4d))
1138 xen_cleanmfnmap_p4d(p4d, unpin);
1141 static void __init xen_pagetable_p2m_free(void)
1146 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1148 /* No memory or already called. */
1149 if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1152 /* using __ka address and sticking INVALID_P2M_ENTRY! */
1153 memset((void *)xen_start_info->mfn_list, 0xff, size);
1155 addr = xen_start_info->mfn_list;
1157 * We could be in __ka space.
1158 * We roundup to the PMD, which means that if anybody at this stage is
1159 * using the __ka address of xen_start_info or
1160 * xen_start_info->shared_info they are in going to crash. Fortunately
1161 * we have already revectored in xen_setup_kernel_pagetable.
1163 size = roundup(size, PMD_SIZE);
1165 if (addr >= __START_KERNEL_map) {
1166 xen_cleanhighmap(addr, addr + size);
1167 size = PAGE_ALIGN(xen_start_info->nr_pages *
1168 sizeof(unsigned long));
1169 memblock_free((void *)addr, size);
1171 xen_cleanmfnmap(addr);
1175 static void __init xen_pagetable_cleanhighmap(void)
1180 /* At this stage, cleanup_highmap has already cleaned __ka space
1181 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1182 * the ramdisk). We continue on, erasing PMD entries that point to page
1183 * tables - do note that they are accessible at this stage via __va.
1184 * As Xen is aligning the memory end to a 4MB boundary, for good
1185 * measure we also round up to PMD_SIZE * 2 - which means that if
1186 * anybody is using __ka address to the initial boot-stack - and try
1187 * to use it - they are going to crash. The xen_start_info has been
1188 * taken care of already in xen_setup_kernel_pagetable. */
1189 addr = xen_start_info->pt_base;
1190 size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1192 xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1193 xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1196 static void __init xen_pagetable_p2m_setup(void)
1198 xen_vmalloc_p2m_tree();
1200 xen_pagetable_p2m_free();
1202 xen_pagetable_cleanhighmap();
1204 /* And revector! Bye bye old array */
1205 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1208 static void __init xen_pagetable_init(void)
1211 * The majority of further PTE writes is to pagetables already
1212 * announced as such to Xen. Hence it is more efficient to use
1213 * hypercalls for these updates.
1215 pv_ops.mmu.set_pte = __xen_set_pte;
1218 xen_post_allocator_init();
1220 xen_pagetable_p2m_setup();
1222 /* Allocate and initialize top and mid mfn levels for p2m structure */
1223 xen_build_mfn_list_list();
1225 /* Remap memory freed due to conflicts with E820 map */
1227 xen_setup_mfn_list_list();
1230 static noinstr void xen_write_cr2(unsigned long cr2)
1232 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1235 static noinline void xen_flush_tlb(void)
1237 struct mmuext_op *op;
1238 struct multicall_space mcs;
1242 mcs = xen_mc_entry(sizeof(*op));
1245 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1246 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1248 xen_mc_issue(PARAVIRT_LAZY_MMU);
1253 static void xen_flush_tlb_one_user(unsigned long addr)
1255 struct mmuext_op *op;
1256 struct multicall_space mcs;
1258 trace_xen_mmu_flush_tlb_one_user(addr);
1262 mcs = xen_mc_entry(sizeof(*op));
1264 op->cmd = MMUEXT_INVLPG_LOCAL;
1265 op->arg1.linear_addr = addr & PAGE_MASK;
1266 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1268 xen_mc_issue(PARAVIRT_LAZY_MMU);
1273 static void xen_flush_tlb_multi(const struct cpumask *cpus,
1274 const struct flush_tlb_info *info)
1277 struct mmuext_op op;
1278 DECLARE_BITMAP(mask, NR_CPUS);
1280 struct multicall_space mcs;
1281 const size_t mc_entry_size = sizeof(args->op) +
1282 sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1284 trace_xen_mmu_flush_tlb_multi(cpus, info->mm, info->start, info->end);
1286 if (cpumask_empty(cpus))
1287 return; /* nothing to do */
1289 mcs = xen_mc_entry(mc_entry_size);
1291 args->op.arg2.vcpumask = to_cpumask(args->mask);
1293 /* Remove any offline CPUs */
1294 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1296 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1297 if (info->end != TLB_FLUSH_ALL &&
1298 (info->end - info->start) <= PAGE_SIZE) {
1299 args->op.cmd = MMUEXT_INVLPG_MULTI;
1300 args->op.arg1.linear_addr = info->start;
1303 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1305 xen_mc_issue(PARAVIRT_LAZY_MMU);
1308 static unsigned long xen_read_cr3(void)
1310 return this_cpu_read(xen_cr3);
1313 static void set_current_cr3(void *v)
1315 this_cpu_write(xen_current_cr3, (unsigned long)v);
1318 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1320 struct mmuext_op op;
1323 trace_xen_mmu_write_cr3(kernel, cr3);
1326 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1330 WARN_ON(mfn == 0 && kernel);
1332 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1335 xen_extend_mmuext_op(&op);
1338 this_cpu_write(xen_cr3, cr3);
1340 /* Update xen_current_cr3 once the batch has actually
1342 xen_mc_callback(set_current_cr3, (void *)cr3);
1345 static void xen_write_cr3(unsigned long cr3)
1347 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1349 BUG_ON(preemptible());
1351 xen_mc_batch(); /* disables interrupts */
1353 /* Update while interrupts are disabled, so its atomic with
1355 this_cpu_write(xen_cr3, cr3);
1357 __xen_write_cr3(true, cr3);
1360 __xen_write_cr3(false, __pa(user_pgd));
1362 __xen_write_cr3(false, 0);
1364 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1368 * At the start of the day - when Xen launches a guest, it has already
1369 * built pagetables for the guest. We diligently look over them
1370 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1371 * init_top_pgt and its friends. Then when we are happy we load
1372 * the new init_top_pgt - and continue on.
1374 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1375 * up the rest of the pagetables. When it has completed it loads the cr3.
1376 * N.B. that baremetal would start at 'start_kernel' (and the early
1377 * #PF handler would create bootstrap pagetables) - so we are running
1378 * with the same assumptions as what to do when write_cr3 is executed
1381 * Since there are no user-page tables at all, we have two variants
1382 * of xen_write_cr3 - the early bootup (this one), and the late one
1383 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1384 * the Linux kernel and user-space are both in ring 3 while the
1385 * hypervisor is in ring 0.
1387 static void __init xen_write_cr3_init(unsigned long cr3)
1389 BUG_ON(preemptible());
1391 xen_mc_batch(); /* disables interrupts */
1393 /* Update while interrupts are disabled, so its atomic with
1395 this_cpu_write(xen_cr3, cr3);
1397 __xen_write_cr3(true, cr3);
1399 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1402 static int xen_pgd_alloc(struct mm_struct *mm)
1404 pgd_t *pgd = mm->pgd;
1405 struct page *page = virt_to_page(pgd);
1409 BUG_ON(PagePinned(virt_to_page(pgd)));
1410 BUG_ON(page->private != 0);
1412 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1413 page->private = (unsigned long)user_pgd;
1415 if (user_pgd != NULL) {
1416 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1417 user_pgd[pgd_index(VSYSCALL_ADDR)] =
1418 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1423 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1428 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1430 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1433 free_page((unsigned long)user_pgd);
1437 * Init-time set_pte while constructing initial pagetables, which
1438 * doesn't allow RO page table pages to be remapped RW.
1440 * If there is no MFN for this PFN then this page is initially
1441 * ballooned out so clear the PTE (as in decrease_reservation() in
1442 * drivers/xen/balloon.c).
1444 * Many of these PTE updates are done on unpinned and writable pages
1445 * and doing a hypercall for these is unnecessary and expensive. At
1446 * this point it is rarely possible to tell if a page is pinned, so
1447 * mostly write the PTE directly and rely on Xen trapping and
1448 * emulating any updates as necessary.
1450 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1452 if (unlikely(is_early_ioremap_ptep(ptep)))
1453 __xen_set_pte(ptep, pte);
1455 native_set_pte(ptep, pte);
1458 __visible pte_t xen_make_pte_init(pteval_t pte)
1463 * Pages belonging to the initial p2m list mapped outside the default
1464 * address range must be mapped read-only. This region contains the
1465 * page tables for mapping the p2m list, too, and page tables MUST be
1468 pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1469 if (xen_start_info->mfn_list < __START_KERNEL_map &&
1470 pfn >= xen_start_info->first_p2m_pfn &&
1471 pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1474 pte = pte_pfn_to_mfn(pte);
1475 return native_make_pte(pte);
1477 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1479 /* Early in boot, while setting up the initial pagetable, assume
1480 everything is pinned. */
1481 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1483 #ifdef CONFIG_FLATMEM
1484 BUG_ON(mem_map); /* should only be used early */
1486 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1487 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1490 /* Used for pmd and pud */
1491 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1493 #ifdef CONFIG_FLATMEM
1494 BUG_ON(mem_map); /* should only be used early */
1496 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1499 /* Early release_pte assumes that all pts are pinned, since there's
1500 only init_mm and anything attached to that is pinned. */
1501 static void __init xen_release_pte_init(unsigned long pfn)
1503 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1504 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1507 static void __init xen_release_pmd_init(unsigned long pfn)
1509 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1512 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1514 struct multicall_space mcs;
1515 struct mmuext_op *op;
1517 mcs = __xen_mc_entry(sizeof(*op));
1520 op->arg1.mfn = pfn_to_mfn(pfn);
1522 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1525 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1527 struct multicall_space mcs;
1528 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1530 mcs = __xen_mc_entry(0);
1531 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1532 pfn_pte(pfn, prot), 0);
1535 /* This needs to make sure the new pte page is pinned iff its being
1536 attached to a pinned pagetable. */
1537 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1540 bool pinned = xen_page_pinned(mm->pgd);
1542 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1545 struct page *page = pfn_to_page(pfn);
1548 if (static_branch_likely(&xen_struct_pages_ready)) {
1549 pinned = PagePinned(page);
1550 SetPagePinned(page);
1555 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1557 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS && !pinned)
1558 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1560 xen_mc_issue(PARAVIRT_LAZY_MMU);
1564 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1566 xen_alloc_ptpage(mm, pfn, PT_PTE);
1569 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1571 xen_alloc_ptpage(mm, pfn, PT_PMD);
1574 /* This should never happen until we're OK to use struct page */
1575 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1577 struct page *page = pfn_to_page(pfn);
1578 bool pinned = PagePinned(page);
1580 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1585 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1586 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1588 __set_pfn_prot(pfn, PAGE_KERNEL);
1590 xen_mc_issue(PARAVIRT_LAZY_MMU);
1592 ClearPagePinned(page);
1596 static void xen_release_pte(unsigned long pfn)
1598 xen_release_ptpage(pfn, PT_PTE);
1601 static void xen_release_pmd(unsigned long pfn)
1603 xen_release_ptpage(pfn, PT_PMD);
1606 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1608 xen_alloc_ptpage(mm, pfn, PT_PUD);
1611 static void xen_release_pud(unsigned long pfn)
1613 xen_release_ptpage(pfn, PT_PUD);
1617 * Like __va(), but returns address in the kernel mapping (which is
1618 * all we have until the physical memory mapping has been set up.
1620 static void * __init __ka(phys_addr_t paddr)
1622 return (void *)(paddr + __START_KERNEL_map);
1625 /* Convert a machine address to physical address */
1626 static unsigned long __init m2p(phys_addr_t maddr)
1630 maddr &= XEN_PTE_MFN_MASK;
1631 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1636 /* Convert a machine address to kernel virtual */
1637 static void * __init m2v(phys_addr_t maddr)
1639 return __ka(m2p(maddr));
1642 /* Set the page permissions on an identity-mapped pages */
1643 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1644 unsigned long flags)
1646 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1647 pte_t pte = pfn_pte(pfn, prot);
1649 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1652 static void __init set_page_prot(void *addr, pgprot_t prot)
1654 return set_page_prot_flags(addr, prot, UVMF_NONE);
1657 void __init xen_setup_machphys_mapping(void)
1659 struct xen_machphys_mapping mapping;
1661 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1662 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1663 machine_to_phys_nr = mapping.max_mfn + 1;
1665 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1669 static void __init convert_pfn_mfn(void *v)
1674 /* All levels are converted the same way, so just treat them
1676 for (i = 0; i < PTRS_PER_PTE; i++)
1677 pte[i] = xen_make_pte(pte[i].pte);
1679 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1682 if (*pt_base == PFN_DOWN(__pa(addr))) {
1683 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1684 clear_page((void *)addr);
1687 if (*pt_end == PFN_DOWN(__pa(addr))) {
1688 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1689 clear_page((void *)addr);
1694 * Set up the initial kernel pagetable.
1696 * We can construct this by grafting the Xen provided pagetable into
1697 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1698 * level2_ident_pgt, and level2_kernel_pgt. This means that only the
1699 * kernel has a physical mapping to start with - but that's enough to
1700 * get __va working. We need to fill in the rest of the physical
1701 * mapping once some sort of allocator has been set up.
1703 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1707 unsigned long addr[3];
1708 unsigned long pt_base, pt_end;
1711 /* max_pfn_mapped is the last pfn mapped in the initial memory
1712 * mappings. Considering that on Xen after the kernel mappings we
1713 * have the mappings of some pages that don't exist in pfn space, we
1714 * set max_pfn_mapped to the last real pfn mapped. */
1715 if (xen_start_info->mfn_list < __START_KERNEL_map)
1716 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1718 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1720 pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1721 pt_end = pt_base + xen_start_info->nr_pt_frames;
1723 /* Zap identity mapping */
1724 init_top_pgt[0] = __pgd(0);
1726 /* Pre-constructed entries are in pfn, so convert to mfn */
1727 /* L4[273] -> level3_ident_pgt */
1728 /* L4[511] -> level3_kernel_pgt */
1729 convert_pfn_mfn(init_top_pgt);
1731 /* L3_i[0] -> level2_ident_pgt */
1732 convert_pfn_mfn(level3_ident_pgt);
1733 /* L3_k[510] -> level2_kernel_pgt */
1734 /* L3_k[511] -> level2_fixmap_pgt */
1735 convert_pfn_mfn(level3_kernel_pgt);
1737 /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1738 convert_pfn_mfn(level2_fixmap_pgt);
1740 /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1741 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1742 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1744 addr[0] = (unsigned long)pgd;
1745 addr[1] = (unsigned long)l3;
1746 addr[2] = (unsigned long)l2;
1747 /* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1748 * Both L4[273][0] and L4[511][510] have entries that point to the same
1749 * L2 (PMD) tables. Meaning that if you modify it in __va space
1750 * it will be also modified in the __ka space! (But if you just
1751 * modify the PMD table to point to other PTE's or none, then you
1752 * are OK - which is what cleanup_highmap does) */
1753 copy_page(level2_ident_pgt, l2);
1754 /* Graft it onto L4[511][510] */
1755 copy_page(level2_kernel_pgt, l2);
1758 * Zap execute permission from the ident map. Due to the sharing of
1759 * L1 entries we need to do this in the L2.
1761 if (__supported_pte_mask & _PAGE_NX) {
1762 for (i = 0; i < PTRS_PER_PMD; ++i) {
1763 if (pmd_none(level2_ident_pgt[i]))
1765 level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1769 /* Copy the initial P->M table mappings if necessary. */
1770 i = pgd_index(xen_start_info->mfn_list);
1771 if (i && i < pgd_index(__START_KERNEL_map))
1772 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1774 /* Make pagetable pieces RO */
1775 set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1776 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1777 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1778 set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1779 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1780 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1782 for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1783 set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1787 /* Pin down new L4 */
1788 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1789 PFN_DOWN(__pa_symbol(init_top_pgt)));
1791 /* Unpin Xen-provided one */
1792 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1794 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1795 /* Pin user vsyscall L3 */
1796 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1797 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1798 PFN_DOWN(__pa_symbol(level3_user_vsyscall)));
1802 * At this stage there can be no user pgd, and no page structure to
1803 * attach it to, so make sure we just set kernel pgd.
1806 __xen_write_cr3(true, __pa(init_top_pgt));
1807 xen_mc_issue(PARAVIRT_LAZY_CPU);
1809 /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1810 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
1811 * the initial domain. For guests using the toolstack, they are in:
1812 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1813 * rip out the [L4] (pgd), but for guests we shave off three pages.
1815 for (i = 0; i < ARRAY_SIZE(addr); i++)
1816 check_pt_base(&pt_base, &pt_end, addr[i]);
1818 /* Our (by three pages) smaller Xen pagetable that we are using */
1819 xen_pt_base = PFN_PHYS(pt_base);
1820 xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1821 memblock_reserve(xen_pt_base, xen_pt_size);
1823 /* Revector the xen_start_info */
1824 xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1828 * Read a value from a physical address.
1830 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1832 unsigned long *vaddr;
1835 vaddr = early_memremap_ro(addr, sizeof(val));
1837 early_memunmap(vaddr, sizeof(val));
1842 * Translate a virtual address to a physical one without relying on mapped
1843 * page tables. Don't rely on big pages being aligned in (guest) physical
1846 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1855 pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1857 if (!pgd_present(pgd))
1860 pa = pgd_val(pgd) & PTE_PFN_MASK;
1861 pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1863 if (!pud_present(pud))
1865 pa = pud_val(pud) & PTE_PFN_MASK;
1867 return pa + (vaddr & ~PUD_MASK);
1869 pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1871 if (!pmd_present(pmd))
1873 pa = pmd_val(pmd) & PTE_PFN_MASK;
1875 return pa + (vaddr & ~PMD_MASK);
1877 pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1879 if (!pte_present(pte))
1881 pa = pte_pfn(pte) << PAGE_SHIFT;
1883 return pa | (vaddr & ~PAGE_MASK);
1887 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1890 void __init xen_relocate_p2m(void)
1892 phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1893 unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1894 int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1899 unsigned long *new_p2m;
1901 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1902 n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1903 n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1904 n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1905 n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1906 n_frames = n_pte + n_pt + n_pmd + n_pud;
1908 new_area = xen_find_free_area(PFN_PHYS(n_frames));
1910 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1915 * Setup the page tables for addressing the new p2m list.
1916 * We have asked the hypervisor to map the p2m list at the user address
1917 * PUD_SIZE. It may have done so, or it may have used a kernel space
1918 * address depending on the Xen version.
1919 * To avoid any possible virtual address collision, just use
1920 * 2 * PUD_SIZE for the new area.
1922 pud_phys = new_area;
1923 pmd_phys = pud_phys + PFN_PHYS(n_pud);
1924 pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1925 p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1927 pgd = __va(read_cr3_pa());
1928 new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1929 for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1930 pud = early_memremap(pud_phys, PAGE_SIZE);
1932 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1934 pmd = early_memremap(pmd_phys, PAGE_SIZE);
1936 for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1938 pt = early_memremap(pt_phys, PAGE_SIZE);
1941 idx_pte < min(n_pte, PTRS_PER_PTE);
1943 pt[idx_pte] = pfn_pte(p2m_pfn,
1947 n_pte -= PTRS_PER_PTE;
1948 early_memunmap(pt, PAGE_SIZE);
1949 make_lowmem_page_readonly(__va(pt_phys));
1950 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
1952 pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
1953 pt_phys += PAGE_SIZE;
1955 n_pt -= PTRS_PER_PMD;
1956 early_memunmap(pmd, PAGE_SIZE);
1957 make_lowmem_page_readonly(__va(pmd_phys));
1958 pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
1959 PFN_DOWN(pmd_phys));
1960 pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
1961 pmd_phys += PAGE_SIZE;
1963 n_pmd -= PTRS_PER_PUD;
1964 early_memunmap(pud, PAGE_SIZE);
1965 make_lowmem_page_readonly(__va(pud_phys));
1966 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
1967 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
1968 pud_phys += PAGE_SIZE;
1971 /* Now copy the old p2m info to the new area. */
1972 memcpy(new_p2m, xen_p2m_addr, size);
1973 xen_p2m_addr = new_p2m;
1975 /* Release the old p2m list and set new list info. */
1976 p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
1978 p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
1980 if (xen_start_info->mfn_list < __START_KERNEL_map) {
1981 pfn = xen_start_info->first_p2m_pfn;
1982 pfn_end = xen_start_info->first_p2m_pfn +
1983 xen_start_info->nr_p2m_frames;
1984 set_pgd(pgd + 1, __pgd(0));
1987 pfn_end = p2m_pfn_end;
1990 memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
1991 while (pfn < pfn_end) {
1992 if (pfn == p2m_pfn) {
1996 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2000 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2001 xen_start_info->first_p2m_pfn = PFN_DOWN(new_area);
2002 xen_start_info->nr_p2m_frames = n_frames;
2005 void __init xen_reserve_special_pages(void)
2009 memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2010 if (xen_start_info->store_mfn) {
2011 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2012 memblock_reserve(paddr, PAGE_SIZE);
2014 if (!xen_initial_domain()) {
2015 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2016 memblock_reserve(paddr, PAGE_SIZE);
2020 void __init xen_pt_check_e820(void)
2022 if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2023 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2028 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2030 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2033 unsigned long vaddr;
2035 phys >>= PAGE_SHIFT;
2038 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2039 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2042 /* All local page mappings */
2043 pte = pfn_pte(phys, prot);
2046 #ifdef CONFIG_X86_LOCAL_APIC
2047 case FIX_APIC_BASE: /* maps dummy local APIC */
2048 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2052 #ifdef CONFIG_X86_IO_APIC
2053 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2055 * We just don't map the IO APIC - all access is via
2056 * hypercalls. Keep the address in the pte for reference.
2058 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2062 case FIX_PARAVIRT_BOOTMAP:
2063 /* This is an MFN, but it isn't an IO mapping from the
2065 pte = mfn_pte(phys, prot);
2069 /* By default, set_fixmap is used for hardware mappings */
2070 pte = mfn_pte(phys, prot);
2074 vaddr = __fix_to_virt(idx);
2075 if (HYPERVISOR_update_va_mapping(vaddr, pte, UVMF_INVLPG))
2078 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2079 /* Replicate changes to map the vsyscall page into the user
2080 pagetable vsyscall mapping. */
2081 if (idx == VSYSCALL_PAGE)
2082 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2086 static void __init xen_post_allocator_init(void)
2088 pv_ops.mmu.set_pte = xen_set_pte;
2089 pv_ops.mmu.set_pmd = xen_set_pmd;
2090 pv_ops.mmu.set_pud = xen_set_pud;
2091 pv_ops.mmu.set_p4d = xen_set_p4d;
2093 /* This will work as long as patching hasn't happened yet
2094 (which it hasn't) */
2095 pv_ops.mmu.alloc_pte = xen_alloc_pte;
2096 pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2097 pv_ops.mmu.release_pte = xen_release_pte;
2098 pv_ops.mmu.release_pmd = xen_release_pmd;
2099 pv_ops.mmu.alloc_pud = xen_alloc_pud;
2100 pv_ops.mmu.release_pud = xen_release_pud;
2101 pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2103 pv_ops.mmu.write_cr3 = &xen_write_cr3;
2106 static void xen_leave_lazy_mmu(void)
2110 paravirt_leave_lazy_mmu();
2114 static const typeof(pv_ops) xen_mmu_ops __initconst = {
2116 .read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2117 .write_cr2 = xen_write_cr2,
2119 .read_cr3 = xen_read_cr3,
2120 .write_cr3 = xen_write_cr3_init,
2122 .flush_tlb_user = xen_flush_tlb,
2123 .flush_tlb_kernel = xen_flush_tlb,
2124 .flush_tlb_one_user = xen_flush_tlb_one_user,
2125 .flush_tlb_multi = xen_flush_tlb_multi,
2126 .tlb_remove_table = tlb_remove_table,
2128 .pgd_alloc = xen_pgd_alloc,
2129 .pgd_free = xen_pgd_free,
2131 .alloc_pte = xen_alloc_pte_init,
2132 .release_pte = xen_release_pte_init,
2133 .alloc_pmd = xen_alloc_pmd_init,
2134 .release_pmd = xen_release_pmd_init,
2136 .set_pte = xen_set_pte_init,
2137 .set_pmd = xen_set_pmd_hyper,
2139 .ptep_modify_prot_start = xen_ptep_modify_prot_start,
2140 .ptep_modify_prot_commit = xen_ptep_modify_prot_commit,
2142 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2143 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2145 .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2146 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2148 .set_pud = xen_set_pud_hyper,
2150 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2151 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2153 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2154 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2155 .set_p4d = xen_set_p4d_hyper,
2157 .alloc_pud = xen_alloc_pmd_init,
2158 .release_pud = xen_release_pmd_init,
2160 #if CONFIG_PGTABLE_LEVELS >= 5
2161 .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2162 .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2165 .enter_mmap = xen_enter_mmap,
2166 .exit_mmap = xen_exit_mmap,
2169 .enter = paravirt_enter_lazy_mmu,
2170 .leave = xen_leave_lazy_mmu,
2171 .flush = paravirt_flush_lazy_mmu,
2174 .set_fixmap = xen_set_fixmap,
2178 void __init xen_init_mmu_ops(void)
2180 x86_init.paging.pagetable_init = xen_pagetable_init;
2181 x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2183 pv_ops.mmu = xen_mmu_ops.mmu;
2185 memset(dummy_mapping, 0xff, PAGE_SIZE);
2188 /* Protected by xen_reservation_lock. */
2189 #define MAX_CONTIG_ORDER 9 /* 2MB */
2190 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2192 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2193 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2194 unsigned long *in_frames,
2195 unsigned long *out_frames)
2198 struct multicall_space mcs;
2201 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2202 mcs = __xen_mc_entry(0);
2205 in_frames[i] = virt_to_mfn(vaddr);
2207 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2208 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2211 out_frames[i] = virt_to_pfn(vaddr);
2217 * Update the pfn-to-mfn mappings for a virtual address range, either to
2218 * point to an array of mfns, or contiguously from a single starting
2221 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2222 unsigned long *mfns,
2223 unsigned long first_mfn)
2230 limit = 1u << order;
2231 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2232 struct multicall_space mcs;
2235 mcs = __xen_mc_entry(0);
2239 mfn = first_mfn + i;
2241 if (i < (limit - 1))
2245 flags = UVMF_INVLPG | UVMF_ALL;
2247 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2250 MULTI_update_va_mapping(mcs.mc, vaddr,
2251 mfn_pte(mfn, PAGE_KERNEL), flags);
2253 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2260 * Perform the hypercall to exchange a region of our pfns to point to
2261 * memory with the required contiguous alignment. Takes the pfns as
2262 * input, and populates mfns as output.
2264 * Returns a success code indicating whether the hypervisor was able to
2265 * satisfy the request or not.
2267 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2268 unsigned long *pfns_in,
2269 unsigned long extents_out,
2270 unsigned int order_out,
2271 unsigned long *mfns_out,
2272 unsigned int address_bits)
2277 struct xen_memory_exchange exchange = {
2279 .nr_extents = extents_in,
2280 .extent_order = order_in,
2281 .extent_start = pfns_in,
2285 .nr_extents = extents_out,
2286 .extent_order = order_out,
2287 .extent_start = mfns_out,
2288 .address_bits = address_bits,
2293 BUG_ON(extents_in << order_in != extents_out << order_out);
2295 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2296 success = (exchange.nr_exchanged == extents_in);
2298 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2299 BUG_ON(success && (rc != 0));
2304 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2305 unsigned int address_bits,
2306 dma_addr_t *dma_handle)
2308 unsigned long *in_frames = discontig_frames, out_frame;
2309 unsigned long flags;
2311 unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2313 if (unlikely(order > MAX_CONTIG_ORDER))
2316 memset((void *) vstart, 0, PAGE_SIZE << order);
2318 spin_lock_irqsave(&xen_reservation_lock, flags);
2320 /* 1. Zap current PTEs, remembering MFNs. */
2321 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2323 /* 2. Get a new contiguous memory extent. */
2324 out_frame = virt_to_pfn(vstart);
2325 success = xen_exchange_memory(1UL << order, 0, in_frames,
2326 1, order, &out_frame,
2329 /* 3. Map the new extent in place of old pages. */
2331 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2333 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2335 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2337 *dma_handle = virt_to_machine(vstart).maddr;
2338 return success ? 0 : -ENOMEM;
2341 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2343 unsigned long *out_frames = discontig_frames, in_frame;
2344 unsigned long flags;
2346 unsigned long vstart;
2348 if (unlikely(order > MAX_CONTIG_ORDER))
2351 vstart = (unsigned long)phys_to_virt(pstart);
2352 memset((void *) vstart, 0, PAGE_SIZE << order);
2354 spin_lock_irqsave(&xen_reservation_lock, flags);
2356 /* 1. Find start MFN of contiguous extent. */
2357 in_frame = virt_to_mfn(vstart);
2359 /* 2. Zap current PTEs. */
2360 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2362 /* 3. Do the exchange for non-contiguous MFNs. */
2363 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2366 /* 4. Map new pages in place of old pages. */
2368 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2370 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2372 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2375 static noinline void xen_flush_tlb_all(void)
2377 struct mmuext_op *op;
2378 struct multicall_space mcs;
2382 mcs = xen_mc_entry(sizeof(*op));
2385 op->cmd = MMUEXT_TLB_FLUSH_ALL;
2386 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2388 xen_mc_issue(PARAVIRT_LAZY_MMU);
2393 #define REMAP_BATCH_SIZE 16
2400 struct mmu_update *mmu_update;
2403 static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2405 struct remap_data *rmd = data;
2406 pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2409 * If we have a contiguous range, just update the pfn itself,
2410 * else update pointer to be "next pfn".
2412 if (rmd->contiguous)
2417 rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2418 rmd->mmu_update->ptr |= rmd->no_translate ?
2419 MMU_PT_UPDATE_NO_TRANSLATE :
2420 MMU_NORMAL_PT_UPDATE;
2421 rmd->mmu_update->val = pte_val_ma(pte);
2427 int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2428 xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2429 unsigned int domid, bool no_translate)
2432 struct remap_data rmd;
2433 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2434 unsigned long range;
2437 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2442 * We use the err_ptr to indicate if there we are doing a contiguous
2443 * mapping or a discontiguous mapping.
2445 rmd.contiguous = !err_ptr;
2446 rmd.no_translate = no_translate;
2451 int batch = min(REMAP_BATCH_SIZE, nr);
2452 int batch_left = batch;
2454 range = (unsigned long)batch << PAGE_SHIFT;
2456 rmd.mmu_update = mmu_update;
2457 err = apply_to_page_range(vma->vm_mm, addr, range,
2458 remap_area_pfn_pte_fn, &rmd);
2463 * We record the error for each page that gives an error, but
2464 * continue mapping until the whole set is done
2469 err = HYPERVISOR_mmu_update(&mmu_update[index],
2470 batch_left, &done, domid);
2473 * @err_ptr may be the same buffer as @gfn, so
2474 * only clear it after each chunk of @gfn is
2478 for (i = index; i < index + done; i++)
2485 done++; /* Skip failed frame. */
2490 } while (batch_left);
2500 xen_flush_tlb_all();
2502 return err < 0 ? err : mapped;
2504 EXPORT_SYMBOL_GPL(xen_remap_pfn);
2506 #ifdef CONFIG_KEXEC_CORE
2507 phys_addr_t paddr_vmcoreinfo_note(void)
2509 if (xen_pv_domain())
2510 return virt_to_machine(vmcoreinfo_note).maddr;
2512 return __pa(vmcoreinfo_note);
2514 #endif /* CONFIG_KEXEC_CORE */