4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched/mm.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/export.h>
47 #include <linux/init.h>
48 #include <linux/gfp.h>
49 #include <linux/memblock.h>
50 #include <linux/seq_file.h>
51 #include <linux/crash_dump.h>
52 #ifdef CONFIG_KEXEC_CORE
53 #include <linux/kexec.h>
56 #include <trace/events/xen.h>
58 #include <asm/pgtable.h>
59 #include <asm/tlbflush.h>
60 #include <asm/fixmap.h>
61 #include <asm/mmu_context.h>
62 #include <asm/setup.h>
63 #include <asm/paravirt.h>
64 #include <asm/e820/api.h>
65 #include <asm/linkage.h>
71 #include <asm/xen/hypercall.h>
72 #include <asm/xen/hypervisor.h>
76 #include <xen/interface/xen.h>
77 #include <xen/interface/hvm/hvm_op.h>
78 #include <xen/interface/version.h>
79 #include <xen/interface/memory.h>
80 #include <xen/hvc-console.h>
82 #include "multicalls.h"
88 * Identity map, in addition to plain kernel map. This needs to be
89 * large enough to allocate page table pages to allocate the rest.
90 * Each page can map 2MB.
92 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
93 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
96 /* l3 pud for userspace vsyscall mapping */
97 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
98 #endif /* CONFIG_X86_64 */
101 * Note about cr3 (pagetable base) values:
103 * xen_cr3 contains the current logical cr3 value; it contains the
104 * last set cr3. This may not be the current effective cr3, because
105 * its update may be being lazily deferred. However, a vcpu looking
106 * at its own cr3 can use this value knowing that it everything will
107 * be self-consistent.
109 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
110 * hypercall to set the vcpu cr3 is complete (so it may be a little
111 * out of date, but it will never be set early). If one vcpu is
112 * looking at another vcpu's cr3 value, it should use this variable.
114 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
115 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
117 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
119 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
122 * Just beyond the highest usermode address. STACK_TOP_MAX has a
123 * redzone above it, so round it up to a PGD boundary.
125 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
127 void make_lowmem_page_readonly(void *vaddr)
130 unsigned long address = (unsigned long)vaddr;
133 pte = lookup_address(address, &level);
135 return; /* vaddr missing */
137 ptev = pte_wrprotect(*pte);
139 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
143 void make_lowmem_page_readwrite(void *vaddr)
146 unsigned long address = (unsigned long)vaddr;
149 pte = lookup_address(address, &level);
151 return; /* vaddr missing */
153 ptev = pte_mkwrite(*pte);
155 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
161 * During early boot all page table pages are pinned, but we do not have struct
162 * pages, so return true until struct pages are ready.
164 static bool xen_page_pinned(void *ptr)
166 if (static_branch_likely(&xen_struct_pages_ready)) {
167 struct page *page = virt_to_page(ptr);
169 return PagePinned(page);
174 static void xen_extend_mmu_update(const struct mmu_update *update)
176 struct multicall_space mcs;
177 struct mmu_update *u;
179 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
181 if (mcs.mc != NULL) {
184 mcs = __xen_mc_entry(sizeof(*u));
185 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
192 static void xen_extend_mmuext_op(const struct mmuext_op *op)
194 struct multicall_space mcs;
197 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
199 if (mcs.mc != NULL) {
202 mcs = __xen_mc_entry(sizeof(*u));
203 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
210 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
218 /* ptr may be ioremapped for 64-bit pagetable setup */
219 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
220 u.val = pmd_val_ma(val);
221 xen_extend_mmu_update(&u);
223 xen_mc_issue(PARAVIRT_LAZY_MMU);
228 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
230 trace_xen_mmu_set_pmd(ptr, val);
232 /* If page is not pinned, we can just update the entry
234 if (!xen_page_pinned(ptr)) {
239 xen_set_pmd_hyper(ptr, val);
243 * Associate a virtual page frame with a given physical page frame
244 * and protection flags for that frame.
246 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
248 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
251 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
255 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
260 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
261 u.val = pte_val_ma(pteval);
262 xen_extend_mmu_update(&u);
264 xen_mc_issue(PARAVIRT_LAZY_MMU);
269 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
271 if (!xen_batched_set_pte(ptep, pteval)) {
273 * Could call native_set_pte() here and trap and
274 * emulate the PTE write but with 32-bit guests this
275 * needs two traps (one for each of the two 32-bit
276 * words in the PTE) so do one hypercall directly
281 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
282 u.val = pte_val_ma(pteval);
283 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
287 static void xen_set_pte(pte_t *ptep, pte_t pteval)
289 trace_xen_mmu_set_pte(ptep, pteval);
290 __xen_set_pte(ptep, pteval);
293 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
294 pte_t *ptep, pte_t pteval)
296 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
297 __xen_set_pte(ptep, pteval);
300 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
301 unsigned long addr, pte_t *ptep)
303 /* Just return the pte as-is. We preserve the bits on commit */
304 trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
308 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
309 pte_t *ptep, pte_t pte)
313 trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
316 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
317 u.val = pte_val_ma(pte);
318 xen_extend_mmu_update(&u);
320 xen_mc_issue(PARAVIRT_LAZY_MMU);
323 /* Assume pteval_t is equivalent to all the other *val_t types. */
324 static pteval_t pte_mfn_to_pfn(pteval_t val)
326 if (val & _PAGE_PRESENT) {
327 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
328 unsigned long pfn = mfn_to_pfn(mfn);
330 pteval_t flags = val & PTE_FLAGS_MASK;
331 if (unlikely(pfn == ~0))
332 val = flags & ~_PAGE_PRESENT;
334 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
340 static pteval_t pte_pfn_to_mfn(pteval_t val)
342 if (val & _PAGE_PRESENT) {
343 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
344 pteval_t flags = val & PTE_FLAGS_MASK;
347 mfn = __pfn_to_mfn(pfn);
350 * If there's no mfn for the pfn, then just create an
351 * empty non-present pte. Unfortunately this loses
352 * information about the original pfn, so
353 * pte_mfn_to_pfn is asymmetric.
355 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
359 mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
360 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
366 __visible pteval_t xen_pte_val(pte_t pte)
368 pteval_t pteval = pte.pte;
370 return pte_mfn_to_pfn(pteval);
372 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
374 __visible pgdval_t xen_pgd_val(pgd_t pgd)
376 return pte_mfn_to_pfn(pgd.pgd);
378 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
380 __visible pte_t xen_make_pte(pteval_t pte)
382 pte = pte_pfn_to_mfn(pte);
384 return native_make_pte(pte);
386 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
388 __visible pgd_t xen_make_pgd(pgdval_t pgd)
390 pgd = pte_pfn_to_mfn(pgd);
391 return native_make_pgd(pgd);
393 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
395 __visible pmdval_t xen_pmd_val(pmd_t pmd)
397 return pte_mfn_to_pfn(pmd.pmd);
399 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
401 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
409 /* ptr may be ioremapped for 64-bit pagetable setup */
410 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
411 u.val = pud_val_ma(val);
412 xen_extend_mmu_update(&u);
414 xen_mc_issue(PARAVIRT_LAZY_MMU);
419 static void xen_set_pud(pud_t *ptr, pud_t val)
421 trace_xen_mmu_set_pud(ptr, val);
423 /* If page is not pinned, we can just update the entry
425 if (!xen_page_pinned(ptr)) {
430 xen_set_pud_hyper(ptr, val);
433 #ifdef CONFIG_X86_PAE
434 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
436 trace_xen_mmu_set_pte_atomic(ptep, pte);
437 set_64bit((u64 *)ptep, native_pte_val(pte));
440 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
442 trace_xen_mmu_pte_clear(mm, addr, ptep);
443 if (!xen_batched_set_pte(ptep, native_make_pte(0)))
444 native_pte_clear(mm, addr, ptep);
447 static void xen_pmd_clear(pmd_t *pmdp)
449 trace_xen_mmu_pmd_clear(pmdp);
450 set_pmd(pmdp, __pmd(0));
452 #endif /* CONFIG_X86_PAE */
454 __visible pmd_t xen_make_pmd(pmdval_t pmd)
456 pmd = pte_pfn_to_mfn(pmd);
457 return native_make_pmd(pmd);
459 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
462 __visible pudval_t xen_pud_val(pud_t pud)
464 return pte_mfn_to_pfn(pud.pud);
466 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
468 __visible pud_t xen_make_pud(pudval_t pud)
470 pud = pte_pfn_to_mfn(pud);
472 return native_make_pud(pud);
474 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
476 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
478 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
479 unsigned offset = pgd - pgd_page;
480 pgd_t *user_ptr = NULL;
482 if (offset < pgd_index(USER_LIMIT)) {
483 struct page *page = virt_to_page(pgd_page);
484 user_ptr = (pgd_t *)page->private;
492 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
496 u.ptr = virt_to_machine(ptr).maddr;
497 u.val = p4d_val_ma(val);
498 xen_extend_mmu_update(&u);
502 * Raw hypercall-based set_p4d, intended for in early boot before
503 * there's a page structure. This implies:
504 * 1. The only existing pagetable is the kernel's
505 * 2. It is always pinned
506 * 3. It has no user pagetable attached to it
508 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
514 __xen_set_p4d_hyper(ptr, val);
516 xen_mc_issue(PARAVIRT_LAZY_MMU);
521 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
523 pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
526 trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
528 /* If page is not pinned, we can just update the entry
530 if (!xen_page_pinned(ptr)) {
533 WARN_ON(xen_page_pinned(user_ptr));
534 pgd_val.pgd = p4d_val_ma(val);
540 /* If it's pinned, then we can at least batch the kernel and
541 user updates together. */
544 __xen_set_p4d_hyper(ptr, val);
546 __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
548 xen_mc_issue(PARAVIRT_LAZY_MMU);
551 #if CONFIG_PGTABLE_LEVELS >= 5
552 __visible p4dval_t xen_p4d_val(p4d_t p4d)
554 return pte_mfn_to_pfn(p4d.p4d);
556 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
558 __visible p4d_t xen_make_p4d(p4dval_t p4d)
560 p4d = pte_pfn_to_mfn(p4d);
562 return native_make_p4d(p4d);
564 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
565 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
566 #endif /* CONFIG_X86_64 */
568 static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
569 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
570 bool last, unsigned long limit)
572 int i, nr, flush = 0;
574 nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
575 for (i = 0; i < nr; i++) {
576 if (!pmd_none(pmd[i]))
577 flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE);
582 static int xen_pud_walk(struct mm_struct *mm, pud_t *pud,
583 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
584 bool last, unsigned long limit)
586 int i, nr, flush = 0;
588 nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
589 for (i = 0; i < nr; i++) {
592 if (pud_none(pud[i]))
595 pmd = pmd_offset(&pud[i], 0);
596 if (PTRS_PER_PMD > 1)
597 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
598 flush |= xen_pmd_walk(mm, pmd, func,
599 last && i == nr - 1, limit);
604 static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
605 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
606 bool last, unsigned long limit)
615 pud = pud_offset(p4d, 0);
616 if (PTRS_PER_PUD > 1)
617 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
618 flush |= xen_pud_walk(mm, pud, func, last, limit);
623 * (Yet another) pagetable walker. This one is intended for pinning a
624 * pagetable. This means that it walks a pagetable and calls the
625 * callback function on each page it finds making up the page table,
626 * at every level. It walks the entire pagetable, but it only bothers
627 * pinning pte pages which are below limit. In the normal case this
628 * will be STACK_TOP_MAX, but at boot we need to pin up to
631 * For 32-bit the important bit is that we don't pin beyond there,
632 * because then we start getting into Xen's ptes.
634 * For 64-bit, we must skip the Xen hole in the middle of the address
635 * space, just after the big x86-64 virtual hole.
637 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
638 int (*func)(struct mm_struct *mm, struct page *,
642 int i, nr, flush = 0;
643 unsigned hole_low, hole_high;
645 /* The limit is the last byte to be touched */
647 BUG_ON(limit >= FIXADDR_TOP);
650 * 64-bit has a great big hole in the middle of the address
651 * space, which contains the Xen mappings. On 32-bit these
652 * will end up making a zero-sized hole and so is a no-op.
654 hole_low = pgd_index(USER_LIMIT);
655 hole_high = pgd_index(PAGE_OFFSET);
657 nr = pgd_index(limit) + 1;
658 for (i = 0; i < nr; i++) {
661 if (i >= hole_low && i < hole_high)
664 if (pgd_none(pgd[i]))
667 p4d = p4d_offset(&pgd[i], 0);
668 flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
671 /* Do the top level last, so that the callbacks can use it as
672 a cue to do final things like tlb flushes. */
673 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
678 static int xen_pgd_walk(struct mm_struct *mm,
679 int (*func)(struct mm_struct *mm, struct page *,
683 return __xen_pgd_walk(mm, mm->pgd, func, limit);
686 /* If we're using split pte locks, then take the page's lock and
687 return a pointer to it. Otherwise return NULL. */
688 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
690 spinlock_t *ptl = NULL;
692 #if USE_SPLIT_PTE_PTLOCKS
693 ptl = ptlock_ptr(page);
694 spin_lock_nest_lock(ptl, &mm->page_table_lock);
700 static void xen_pte_unlock(void *v)
706 static void xen_do_pin(unsigned level, unsigned long pfn)
711 op.arg1.mfn = pfn_to_mfn(pfn);
713 xen_extend_mmuext_op(&op);
716 static int xen_pin_page(struct mm_struct *mm, struct page *page,
719 unsigned pgfl = TestSetPagePinned(page);
723 flush = 0; /* already pinned */
724 else if (PageHighMem(page))
725 /* kmaps need flushing if we found an unpinned
729 void *pt = lowmem_page_address(page);
730 unsigned long pfn = page_to_pfn(page);
731 struct multicall_space mcs = __xen_mc_entry(0);
737 * We need to hold the pagetable lock between the time
738 * we make the pagetable RO and when we actually pin
739 * it. If we don't, then other users may come in and
740 * attempt to update the pagetable by writing it,
741 * which will fail because the memory is RO but not
742 * pinned, so Xen won't do the trap'n'emulate.
744 * If we're using split pte locks, we can't hold the
745 * entire pagetable's worth of locks during the
746 * traverse, because we may wrap the preempt count (8
747 * bits). The solution is to mark RO and pin each PTE
748 * page while holding the lock. This means the number
749 * of locks we end up holding is never more than a
750 * batch size (~32 entries, at present).
752 * If we're not using split pte locks, we needn't pin
753 * the PTE pages independently, because we're
754 * protected by the overall pagetable lock.
758 ptl = xen_pte_lock(page, mm);
760 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
761 pfn_pte(pfn, PAGE_KERNEL_RO),
762 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
765 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
767 /* Queue a deferred unlock for when this batch
769 xen_mc_callback(xen_pte_unlock, ptl);
776 /* This is called just after a mm has been created, but it has not
777 been used yet. We need to make sure that its pagetable is all
778 read-only, and can be pinned. */
779 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
781 trace_xen_mmu_pgd_pin(mm, pgd);
785 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
786 /* re-enable interrupts for flushing */
796 pgd_t *user_pgd = xen_get_user_pgd(pgd);
798 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
801 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
802 xen_do_pin(MMUEXT_PIN_L4_TABLE,
803 PFN_DOWN(__pa(user_pgd)));
806 #else /* CONFIG_X86_32 */
807 #ifdef CONFIG_X86_PAE
808 /* Need to make sure unshared kernel PMD is pinnable */
809 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
812 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
813 #endif /* CONFIG_X86_64 */
817 static void xen_pgd_pin(struct mm_struct *mm)
819 __xen_pgd_pin(mm, mm->pgd);
823 * On save, we need to pin all pagetables to make sure they get their
824 * mfns turned into pfns. Search the list for any unpinned pgds and pin
825 * them (unpinned pgds are not currently in use, probably because the
826 * process is under construction or destruction).
828 * Expected to be called in stop_machine() ("equivalent to taking
829 * every spinlock in the system"), so the locking doesn't really
830 * matter all that much.
832 void xen_mm_pin_all(void)
836 spin_lock(&pgd_lock);
838 list_for_each_entry(page, &pgd_list, lru) {
839 if (!PagePinned(page)) {
840 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
841 SetPageSavePinned(page);
845 spin_unlock(&pgd_lock);
848 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
856 * The init_mm pagetable is really pinned as soon as its created, but
857 * that's before we have page structures to store the bits. So do all
858 * the book-keeping now once struct pages for allocated pages are
859 * initialized. This happens only after free_all_bootmem() is called.
861 static void __init xen_after_bootmem(void)
863 static_branch_enable(&xen_struct_pages_ready);
865 SetPagePinned(virt_to_page(level3_user_vsyscall));
867 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
870 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
873 unsigned pgfl = TestClearPagePinned(page);
875 if (pgfl && !PageHighMem(page)) {
876 void *pt = lowmem_page_address(page);
877 unsigned long pfn = page_to_pfn(page);
878 spinlock_t *ptl = NULL;
879 struct multicall_space mcs;
882 * Do the converse to pin_page. If we're using split
883 * pte locks, we must be holding the lock for while
884 * the pte page is unpinned but still RO to prevent
885 * concurrent updates from seeing it in this
886 * partially-pinned state.
888 if (level == PT_PTE) {
889 ptl = xen_pte_lock(page, mm);
892 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
895 mcs = __xen_mc_entry(0);
897 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
898 pfn_pte(pfn, PAGE_KERNEL),
899 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
902 /* unlock when batch completed */
903 xen_mc_callback(xen_pte_unlock, ptl);
907 return 0; /* never need to flush on unpin */
910 /* Release a pagetables pages back as normal RW */
911 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
913 trace_xen_mmu_pgd_unpin(mm, pgd);
917 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
921 pgd_t *user_pgd = xen_get_user_pgd(pgd);
924 xen_do_pin(MMUEXT_UNPIN_TABLE,
925 PFN_DOWN(__pa(user_pgd)));
926 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
931 #ifdef CONFIG_X86_PAE
932 /* Need to make sure unshared kernel PMD is unpinned */
933 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
937 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
942 static void xen_pgd_unpin(struct mm_struct *mm)
944 __xen_pgd_unpin(mm, mm->pgd);
948 * On resume, undo any pinning done at save, so that the rest of the
949 * kernel doesn't see any unexpected pinned pagetables.
951 void xen_mm_unpin_all(void)
955 spin_lock(&pgd_lock);
957 list_for_each_entry(page, &pgd_list, lru) {
958 if (PageSavePinned(page)) {
959 BUG_ON(!PagePinned(page));
960 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
961 ClearPageSavePinned(page);
965 spin_unlock(&pgd_lock);
968 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
970 spin_lock(&next->page_table_lock);
972 spin_unlock(&next->page_table_lock);
975 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
977 spin_lock(&mm->page_table_lock);
979 spin_unlock(&mm->page_table_lock);
982 static void drop_mm_ref_this_cpu(void *info)
984 struct mm_struct *mm = info;
986 if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
987 leave_mm(smp_processor_id());
990 * If this cpu still has a stale cr3 reference, then make sure
991 * it has been flushed.
993 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
999 * Another cpu may still have their %cr3 pointing at the pagetable, so
1000 * we need to repoint it somewhere else before we can unpin it.
1002 static void xen_drop_mm_ref(struct mm_struct *mm)
1007 drop_mm_ref_this_cpu(mm);
1009 /* Get the "official" set of cpus referring to our pagetable. */
1010 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1011 for_each_online_cpu(cpu) {
1012 if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1014 smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
1020 * It's possible that a vcpu may have a stale reference to our
1021 * cr3, because its in lazy mode, and it hasn't yet flushed
1022 * its set of pending hypercalls yet. In this case, we can
1023 * look at its actual current cr3 value, and force it to flush
1026 cpumask_clear(mask);
1027 for_each_online_cpu(cpu) {
1028 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1029 cpumask_set_cpu(cpu, mask);
1032 smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
1033 free_cpumask_var(mask);
1036 static void xen_drop_mm_ref(struct mm_struct *mm)
1038 drop_mm_ref_this_cpu(mm);
1043 * While a process runs, Xen pins its pagetables, which means that the
1044 * hypervisor forces it to be read-only, and it controls all updates
1045 * to it. This means that all pagetable updates have to go via the
1046 * hypervisor, which is moderately expensive.
1048 * Since we're pulling the pagetable down, we switch to use init_mm,
1049 * unpin old process pagetable and mark it all read-write, which
1050 * allows further operations on it to be simple memory accesses.
1052 * The only subtle point is that another CPU may be still using the
1053 * pagetable because of lazy tlb flushing. This means we need need to
1054 * switch all CPUs off this pagetable before we can unpin it.
1056 static void xen_exit_mmap(struct mm_struct *mm)
1058 get_cpu(); /* make sure we don't move around */
1059 xen_drop_mm_ref(mm);
1062 spin_lock(&mm->page_table_lock);
1064 /* pgd may not be pinned in the error exit path of execve */
1065 if (xen_page_pinned(mm->pgd))
1068 spin_unlock(&mm->page_table_lock);
1071 static void xen_post_allocator_init(void);
1073 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1075 struct mmuext_op op;
1078 op.arg1.mfn = pfn_to_mfn(pfn);
1079 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1083 #ifdef CONFIG_X86_64
1084 static void __init xen_cleanhighmap(unsigned long vaddr,
1085 unsigned long vaddr_end)
1087 unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1088 pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1090 /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1091 * We include the PMD passed in on _both_ boundaries. */
1092 for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1093 pmd++, vaddr += PMD_SIZE) {
1096 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1097 set_pmd(pmd, __pmd(0));
1099 /* In case we did something silly, we should crash in this function
1100 * instead of somewhere later and be confusing. */
1105 * Make a page range writeable and free it.
1107 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1109 void *vaddr = __va(paddr);
1110 void *vaddr_end = vaddr + size;
1112 for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1113 make_lowmem_page_readwrite(vaddr);
1115 memblock_free(paddr, size);
1118 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1120 unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1123 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1124 ClearPagePinned(virt_to_page(__va(pa)));
1125 xen_free_ro_pages(pa, PAGE_SIZE);
1128 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1134 if (pmd_large(*pmd)) {
1135 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1136 xen_free_ro_pages(pa, PMD_SIZE);
1140 pte_tbl = pte_offset_kernel(pmd, 0);
1141 for (i = 0; i < PTRS_PER_PTE; i++) {
1142 if (pte_none(pte_tbl[i]))
1144 pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1145 xen_free_ro_pages(pa, PAGE_SIZE);
1147 set_pmd(pmd, __pmd(0));
1148 xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1151 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1157 if (pud_large(*pud)) {
1158 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1159 xen_free_ro_pages(pa, PUD_SIZE);
1163 pmd_tbl = pmd_offset(pud, 0);
1164 for (i = 0; i < PTRS_PER_PMD; i++) {
1165 if (pmd_none(pmd_tbl[i]))
1167 xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1169 set_pud(pud, __pud(0));
1170 xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1173 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1179 if (p4d_large(*p4d)) {
1180 pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1181 xen_free_ro_pages(pa, P4D_SIZE);
1185 pud_tbl = pud_offset(p4d, 0);
1186 for (i = 0; i < PTRS_PER_PUD; i++) {
1187 if (pud_none(pud_tbl[i]))
1189 xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1191 set_p4d(p4d, __p4d(0));
1192 xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1196 * Since it is well isolated we can (and since it is perhaps large we should)
1197 * also free the page tables mapping the initial P->M table.
1199 static void __init xen_cleanmfnmap(unsigned long vaddr)
1205 unpin = (vaddr == 2 * PGDIR_SIZE);
1207 pgd = pgd_offset_k(vaddr);
1208 p4d = p4d_offset(pgd, 0);
1209 if (!p4d_none(*p4d))
1210 xen_cleanmfnmap_p4d(p4d, unpin);
1213 static void __init xen_pagetable_p2m_free(void)
1218 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1220 /* No memory or already called. */
1221 if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1224 /* using __ka address and sticking INVALID_P2M_ENTRY! */
1225 memset((void *)xen_start_info->mfn_list, 0xff, size);
1227 addr = xen_start_info->mfn_list;
1229 * We could be in __ka space.
1230 * We roundup to the PMD, which means that if anybody at this stage is
1231 * using the __ka address of xen_start_info or
1232 * xen_start_info->shared_info they are in going to crash. Fortunatly
1233 * we have already revectored in xen_setup_kernel_pagetable.
1235 size = roundup(size, PMD_SIZE);
1237 if (addr >= __START_KERNEL_map) {
1238 xen_cleanhighmap(addr, addr + size);
1239 size = PAGE_ALIGN(xen_start_info->nr_pages *
1240 sizeof(unsigned long));
1241 memblock_free(__pa(addr), size);
1243 xen_cleanmfnmap(addr);
1247 static void __init xen_pagetable_cleanhighmap(void)
1252 /* At this stage, cleanup_highmap has already cleaned __ka space
1253 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1254 * the ramdisk). We continue on, erasing PMD entries that point to page
1255 * tables - do note that they are accessible at this stage via __va.
1256 * As Xen is aligning the memory end to a 4MB boundary, for good
1257 * measure we also round up to PMD_SIZE * 2 - which means that if
1258 * anybody is using __ka address to the initial boot-stack - and try
1259 * to use it - they are going to crash. The xen_start_info has been
1260 * taken care of already in xen_setup_kernel_pagetable. */
1261 addr = xen_start_info->pt_base;
1262 size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1264 xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1265 xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1269 static void __init xen_pagetable_p2m_setup(void)
1271 xen_vmalloc_p2m_tree();
1273 #ifdef CONFIG_X86_64
1274 xen_pagetable_p2m_free();
1276 xen_pagetable_cleanhighmap();
1278 /* And revector! Bye bye old array */
1279 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1282 static void __init xen_pagetable_init(void)
1285 xen_post_allocator_init();
1287 xen_pagetable_p2m_setup();
1289 /* Allocate and initialize top and mid mfn levels for p2m structure */
1290 xen_build_mfn_list_list();
1292 /* Remap memory freed due to conflicts with E820 map */
1294 xen_setup_mfn_list_list();
1296 static void xen_write_cr2(unsigned long cr2)
1298 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1301 static unsigned long xen_read_cr2(void)
1303 return this_cpu_read(xen_vcpu)->arch.cr2;
1306 unsigned long xen_read_cr2_direct(void)
1308 return this_cpu_read(xen_vcpu_info.arch.cr2);
1311 static noinline void xen_flush_tlb(void)
1313 struct mmuext_op *op;
1314 struct multicall_space mcs;
1318 mcs = xen_mc_entry(sizeof(*op));
1321 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1322 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1324 xen_mc_issue(PARAVIRT_LAZY_MMU);
1329 static void xen_flush_tlb_one_user(unsigned long addr)
1331 struct mmuext_op *op;
1332 struct multicall_space mcs;
1334 trace_xen_mmu_flush_tlb_one_user(addr);
1338 mcs = xen_mc_entry(sizeof(*op));
1340 op->cmd = MMUEXT_INVLPG_LOCAL;
1341 op->arg1.linear_addr = addr & PAGE_MASK;
1342 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1344 xen_mc_issue(PARAVIRT_LAZY_MMU);
1349 static void xen_flush_tlb_others(const struct cpumask *cpus,
1350 const struct flush_tlb_info *info)
1353 struct mmuext_op op;
1354 DECLARE_BITMAP(mask, NR_CPUS);
1356 struct multicall_space mcs;
1357 const size_t mc_entry_size = sizeof(args->op) +
1358 sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1360 trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end);
1362 if (cpumask_empty(cpus))
1363 return; /* nothing to do */
1365 mcs = xen_mc_entry(mc_entry_size);
1367 args->op.arg2.vcpumask = to_cpumask(args->mask);
1369 /* Remove us, and any offline CPUS. */
1370 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1371 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1373 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1374 if (info->end != TLB_FLUSH_ALL &&
1375 (info->end - info->start) <= PAGE_SIZE) {
1376 args->op.cmd = MMUEXT_INVLPG_MULTI;
1377 args->op.arg1.linear_addr = info->start;
1380 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1382 xen_mc_issue(PARAVIRT_LAZY_MMU);
1385 static unsigned long xen_read_cr3(void)
1387 return this_cpu_read(xen_cr3);
1390 static void set_current_cr3(void *v)
1392 this_cpu_write(xen_current_cr3, (unsigned long)v);
1395 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1397 struct mmuext_op op;
1400 trace_xen_mmu_write_cr3(kernel, cr3);
1403 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1407 WARN_ON(mfn == 0 && kernel);
1409 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1412 xen_extend_mmuext_op(&op);
1415 this_cpu_write(xen_cr3, cr3);
1417 /* Update xen_current_cr3 once the batch has actually
1419 xen_mc_callback(set_current_cr3, (void *)cr3);
1422 static void xen_write_cr3(unsigned long cr3)
1424 BUG_ON(preemptible());
1426 xen_mc_batch(); /* disables interrupts */
1428 /* Update while interrupts are disabled, so its atomic with
1430 this_cpu_write(xen_cr3, cr3);
1432 __xen_write_cr3(true, cr3);
1434 #ifdef CONFIG_X86_64
1436 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1438 __xen_write_cr3(false, __pa(user_pgd));
1440 __xen_write_cr3(false, 0);
1444 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1447 #ifdef CONFIG_X86_64
1449 * At the start of the day - when Xen launches a guest, it has already
1450 * built pagetables for the guest. We diligently look over them
1451 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1452 * init_top_pgt and its friends. Then when we are happy we load
1453 * the new init_top_pgt - and continue on.
1455 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1456 * up the rest of the pagetables. When it has completed it loads the cr3.
1457 * N.B. that baremetal would start at 'start_kernel' (and the early
1458 * #PF handler would create bootstrap pagetables) - so we are running
1459 * with the same assumptions as what to do when write_cr3 is executed
1462 * Since there are no user-page tables at all, we have two variants
1463 * of xen_write_cr3 - the early bootup (this one), and the late one
1464 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1465 * the Linux kernel and user-space are both in ring 3 while the
1466 * hypervisor is in ring 0.
1468 static void __init xen_write_cr3_init(unsigned long cr3)
1470 BUG_ON(preemptible());
1472 xen_mc_batch(); /* disables interrupts */
1474 /* Update while interrupts are disabled, so its atomic with
1476 this_cpu_write(xen_cr3, cr3);
1478 __xen_write_cr3(true, cr3);
1480 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1484 static int xen_pgd_alloc(struct mm_struct *mm)
1486 pgd_t *pgd = mm->pgd;
1489 BUG_ON(PagePinned(virt_to_page(pgd)));
1491 #ifdef CONFIG_X86_64
1493 struct page *page = virt_to_page(pgd);
1496 BUG_ON(page->private != 0);
1500 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1501 page->private = (unsigned long)user_pgd;
1503 if (user_pgd != NULL) {
1504 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1505 user_pgd[pgd_index(VSYSCALL_ADDR)] =
1506 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1511 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1517 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1519 #ifdef CONFIG_X86_64
1520 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1523 free_page((unsigned long)user_pgd);
1528 * Init-time set_pte while constructing initial pagetables, which
1529 * doesn't allow RO page table pages to be remapped RW.
1531 * If there is no MFN for this PFN then this page is initially
1532 * ballooned out so clear the PTE (as in decrease_reservation() in
1533 * drivers/xen/balloon.c).
1535 * Many of these PTE updates are done on unpinned and writable pages
1536 * and doing a hypercall for these is unnecessary and expensive. At
1537 * this point it is not possible to tell if a page is pinned or not,
1538 * so always write the PTE directly and rely on Xen trapping and
1539 * emulating any updates as necessary.
1541 __visible pte_t xen_make_pte_init(pteval_t pte)
1543 #ifdef CONFIG_X86_64
1547 * Pages belonging to the initial p2m list mapped outside the default
1548 * address range must be mapped read-only. This region contains the
1549 * page tables for mapping the p2m list, too, and page tables MUST be
1552 pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1553 if (xen_start_info->mfn_list < __START_KERNEL_map &&
1554 pfn >= xen_start_info->first_p2m_pfn &&
1555 pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1558 pte = pte_pfn_to_mfn(pte);
1559 return native_make_pte(pte);
1561 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1563 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1565 #ifdef CONFIG_X86_32
1566 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1567 if (pte_mfn(pte) != INVALID_P2M_ENTRY
1568 && pte_val_ma(*ptep) & _PAGE_PRESENT)
1569 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1572 native_set_pte(ptep, pte);
1575 /* Early in boot, while setting up the initial pagetable, assume
1576 everything is pinned. */
1577 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1579 #ifdef CONFIG_FLATMEM
1580 BUG_ON(mem_map); /* should only be used early */
1582 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1583 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1586 /* Used for pmd and pud */
1587 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1589 #ifdef CONFIG_FLATMEM
1590 BUG_ON(mem_map); /* should only be used early */
1592 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1595 /* Early release_pte assumes that all pts are pinned, since there's
1596 only init_mm and anything attached to that is pinned. */
1597 static void __init xen_release_pte_init(unsigned long pfn)
1599 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1600 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1603 static void __init xen_release_pmd_init(unsigned long pfn)
1605 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1608 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1610 struct multicall_space mcs;
1611 struct mmuext_op *op;
1613 mcs = __xen_mc_entry(sizeof(*op));
1616 op->arg1.mfn = pfn_to_mfn(pfn);
1618 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1621 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1623 struct multicall_space mcs;
1624 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1626 mcs = __xen_mc_entry(0);
1627 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1628 pfn_pte(pfn, prot), 0);
1631 /* This needs to make sure the new pte page is pinned iff its being
1632 attached to a pinned pagetable. */
1633 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1636 bool pinned = xen_page_pinned(mm->pgd);
1638 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1641 struct page *page = pfn_to_page(pfn);
1643 if (static_branch_likely(&xen_struct_pages_ready))
1644 SetPagePinned(page);
1646 if (!PageHighMem(page)) {
1649 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1651 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1652 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1654 xen_mc_issue(PARAVIRT_LAZY_MMU);
1656 /* make sure there are no stray mappings of
1658 kmap_flush_unused();
1663 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1665 xen_alloc_ptpage(mm, pfn, PT_PTE);
1668 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1670 xen_alloc_ptpage(mm, pfn, PT_PMD);
1673 /* This should never happen until we're OK to use struct page */
1674 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1676 struct page *page = pfn_to_page(pfn);
1677 bool pinned = PagePinned(page);
1679 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1682 if (!PageHighMem(page)) {
1685 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1686 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1688 __set_pfn_prot(pfn, PAGE_KERNEL);
1690 xen_mc_issue(PARAVIRT_LAZY_MMU);
1692 ClearPagePinned(page);
1696 static void xen_release_pte(unsigned long pfn)
1698 xen_release_ptpage(pfn, PT_PTE);
1701 static void xen_release_pmd(unsigned long pfn)
1703 xen_release_ptpage(pfn, PT_PMD);
1706 #ifdef CONFIG_X86_64
1707 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1709 xen_alloc_ptpage(mm, pfn, PT_PUD);
1712 static void xen_release_pud(unsigned long pfn)
1714 xen_release_ptpage(pfn, PT_PUD);
1718 void __init xen_reserve_top(void)
1720 #ifdef CONFIG_X86_32
1721 unsigned long top = HYPERVISOR_VIRT_START;
1722 struct xen_platform_parameters pp;
1724 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1725 top = pp.virt_start;
1727 reserve_top_address(-top);
1728 #endif /* CONFIG_X86_32 */
1732 * Like __va(), but returns address in the kernel mapping (which is
1733 * all we have until the physical memory mapping has been set up.
1735 static void * __init __ka(phys_addr_t paddr)
1737 #ifdef CONFIG_X86_64
1738 return (void *)(paddr + __START_KERNEL_map);
1744 /* Convert a machine address to physical address */
1745 static unsigned long __init m2p(phys_addr_t maddr)
1749 maddr &= XEN_PTE_MFN_MASK;
1750 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1755 /* Convert a machine address to kernel virtual */
1756 static void * __init m2v(phys_addr_t maddr)
1758 return __ka(m2p(maddr));
1761 /* Set the page permissions on an identity-mapped pages */
1762 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1763 unsigned long flags)
1765 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1766 pte_t pte = pfn_pte(pfn, prot);
1768 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1771 static void __init set_page_prot(void *addr, pgprot_t prot)
1773 return set_page_prot_flags(addr, prot, UVMF_NONE);
1775 #ifdef CONFIG_X86_32
1776 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1778 unsigned pmdidx, pteidx;
1782 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1787 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1790 /* Reuse or allocate a page of ptes */
1791 if (pmd_present(pmd[pmdidx]))
1792 pte_page = m2v(pmd[pmdidx].pmd);
1794 /* Check for free pte pages */
1795 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1798 pte_page = &level1_ident_pgt[ident_pte];
1799 ident_pte += PTRS_PER_PTE;
1801 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1804 /* Install mappings */
1805 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1808 if (pfn > max_pfn_mapped)
1809 max_pfn_mapped = pfn;
1811 if (!pte_none(pte_page[pteidx]))
1814 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1815 pte_page[pteidx] = pte;
1819 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1820 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1822 set_page_prot(pmd, PAGE_KERNEL_RO);
1825 void __init xen_setup_machphys_mapping(void)
1827 struct xen_machphys_mapping mapping;
1829 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1830 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1831 machine_to_phys_nr = mapping.max_mfn + 1;
1833 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1835 #ifdef CONFIG_X86_32
1836 WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1837 < machine_to_phys_mapping);
1841 #ifdef CONFIG_X86_64
1842 static void __init convert_pfn_mfn(void *v)
1847 /* All levels are converted the same way, so just treat them
1849 for (i = 0; i < PTRS_PER_PTE; i++)
1850 pte[i] = xen_make_pte(pte[i].pte);
1852 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1855 if (*pt_base == PFN_DOWN(__pa(addr))) {
1856 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1857 clear_page((void *)addr);
1860 if (*pt_end == PFN_DOWN(__pa(addr))) {
1861 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1862 clear_page((void *)addr);
1867 * Set up the initial kernel pagetable.
1869 * We can construct this by grafting the Xen provided pagetable into
1870 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1871 * level2_ident_pgt, and level2_kernel_pgt. This means that only the
1872 * kernel has a physical mapping to start with - but that's enough to
1873 * get __va working. We need to fill in the rest of the physical
1874 * mapping once some sort of allocator has been set up.
1876 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1880 unsigned long addr[3];
1881 unsigned long pt_base, pt_end;
1884 /* max_pfn_mapped is the last pfn mapped in the initial memory
1885 * mappings. Considering that on Xen after the kernel mappings we
1886 * have the mappings of some pages that don't exist in pfn space, we
1887 * set max_pfn_mapped to the last real pfn mapped. */
1888 if (xen_start_info->mfn_list < __START_KERNEL_map)
1889 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1891 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1893 pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1894 pt_end = pt_base + xen_start_info->nr_pt_frames;
1896 /* Zap identity mapping */
1897 init_top_pgt[0] = __pgd(0);
1899 /* Pre-constructed entries are in pfn, so convert to mfn */
1900 /* L4[272] -> level3_ident_pgt */
1901 /* L4[511] -> level3_kernel_pgt */
1902 convert_pfn_mfn(init_top_pgt);
1904 /* L3_i[0] -> level2_ident_pgt */
1905 convert_pfn_mfn(level3_ident_pgt);
1906 /* L3_k[510] -> level2_kernel_pgt */
1907 /* L3_k[511] -> level2_fixmap_pgt */
1908 convert_pfn_mfn(level3_kernel_pgt);
1910 /* L3_k[511][506] -> level1_fixmap_pgt */
1911 convert_pfn_mfn(level2_fixmap_pgt);
1913 /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1914 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1915 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1917 addr[0] = (unsigned long)pgd;
1918 addr[1] = (unsigned long)l3;
1919 addr[2] = (unsigned long)l2;
1920 /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1921 * Both L4[272][0] and L4[511][510] have entries that point to the same
1922 * L2 (PMD) tables. Meaning that if you modify it in __va space
1923 * it will be also modified in the __ka space! (But if you just
1924 * modify the PMD table to point to other PTE's or none, then you
1925 * are OK - which is what cleanup_highmap does) */
1926 copy_page(level2_ident_pgt, l2);
1927 /* Graft it onto L4[511][510] */
1928 copy_page(level2_kernel_pgt, l2);
1931 * Zap execute permission from the ident map. Due to the sharing of
1932 * L1 entries we need to do this in the L2.
1934 if (__supported_pte_mask & _PAGE_NX) {
1935 for (i = 0; i < PTRS_PER_PMD; ++i) {
1936 if (pmd_none(level2_ident_pgt[i]))
1938 level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1942 /* Copy the initial P->M table mappings if necessary. */
1943 i = pgd_index(xen_start_info->mfn_list);
1944 if (i && i < pgd_index(__START_KERNEL_map))
1945 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1947 /* Make pagetable pieces RO */
1948 set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1949 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1950 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1951 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1952 set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1953 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1954 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1955 set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
1957 /* Pin down new L4 */
1958 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1959 PFN_DOWN(__pa_symbol(init_top_pgt)));
1961 /* Unpin Xen-provided one */
1962 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1965 * At this stage there can be no user pgd, and no page structure to
1966 * attach it to, so make sure we just set kernel pgd.
1969 __xen_write_cr3(true, __pa(init_top_pgt));
1970 xen_mc_issue(PARAVIRT_LAZY_CPU);
1972 /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1973 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
1974 * the initial domain. For guests using the toolstack, they are in:
1975 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1976 * rip out the [L4] (pgd), but for guests we shave off three pages.
1978 for (i = 0; i < ARRAY_SIZE(addr); i++)
1979 check_pt_base(&pt_base, &pt_end, addr[i]);
1981 /* Our (by three pages) smaller Xen pagetable that we are using */
1982 xen_pt_base = PFN_PHYS(pt_base);
1983 xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1984 memblock_reserve(xen_pt_base, xen_pt_size);
1986 /* Revector the xen_start_info */
1987 xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1991 * Read a value from a physical address.
1993 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1995 unsigned long *vaddr;
1998 vaddr = early_memremap_ro(addr, sizeof(val));
2000 early_memunmap(vaddr, sizeof(val));
2005 * Translate a virtual address to a physical one without relying on mapped
2006 * page tables. Don't rely on big pages being aligned in (guest) physical
2009 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
2018 pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
2020 if (!pgd_present(pgd))
2023 pa = pgd_val(pgd) & PTE_PFN_MASK;
2024 pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
2026 if (!pud_present(pud))
2028 pa = pud_val(pud) & PTE_PFN_MASK;
2030 return pa + (vaddr & ~PUD_MASK);
2032 pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
2034 if (!pmd_present(pmd))
2036 pa = pmd_val(pmd) & PTE_PFN_MASK;
2038 return pa + (vaddr & ~PMD_MASK);
2040 pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
2042 if (!pte_present(pte))
2044 pa = pte_pfn(pte) << PAGE_SHIFT;
2046 return pa | (vaddr & ~PAGE_MASK);
2050 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
2053 void __init xen_relocate_p2m(void)
2055 phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
2056 unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
2057 int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
2062 unsigned long *new_p2m;
2065 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
2066 n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
2067 n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
2068 n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
2069 n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
2070 n_frames = n_pte + n_pt + n_pmd + n_pud;
2072 new_area = xen_find_free_area(PFN_PHYS(n_frames));
2074 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
2079 * Setup the page tables for addressing the new p2m list.
2080 * We have asked the hypervisor to map the p2m list at the user address
2081 * PUD_SIZE. It may have done so, or it may have used a kernel space
2082 * address depending on the Xen version.
2083 * To avoid any possible virtual address collision, just use
2084 * 2 * PUD_SIZE for the new area.
2086 pud_phys = new_area;
2087 pmd_phys = pud_phys + PFN_PHYS(n_pud);
2088 pt_phys = pmd_phys + PFN_PHYS(n_pmd);
2089 p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
2091 pgd = __va(read_cr3_pa());
2092 new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
2094 for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
2095 pud = early_memremap(pud_phys, PAGE_SIZE);
2097 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
2099 pmd = early_memremap(pmd_phys, PAGE_SIZE);
2101 for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
2103 pt = early_memremap(pt_phys, PAGE_SIZE);
2106 idx_pte < min(n_pte, PTRS_PER_PTE);
2108 set_pte(pt + idx_pte,
2109 pfn_pte(p2m_pfn, PAGE_KERNEL));
2112 n_pte -= PTRS_PER_PTE;
2113 early_memunmap(pt, PAGE_SIZE);
2114 make_lowmem_page_readonly(__va(pt_phys));
2115 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2117 set_pmd(pmd + idx_pt,
2118 __pmd(_PAGE_TABLE | pt_phys));
2119 pt_phys += PAGE_SIZE;
2121 n_pt -= PTRS_PER_PMD;
2122 early_memunmap(pmd, PAGE_SIZE);
2123 make_lowmem_page_readonly(__va(pmd_phys));
2124 pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2125 PFN_DOWN(pmd_phys));
2126 set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
2127 pmd_phys += PAGE_SIZE;
2129 n_pmd -= PTRS_PER_PUD;
2130 early_memunmap(pud, PAGE_SIZE);
2131 make_lowmem_page_readonly(__va(pud_phys));
2132 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2133 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2134 pud_phys += PAGE_SIZE;
2137 /* Now copy the old p2m info to the new area. */
2138 memcpy(new_p2m, xen_p2m_addr, size);
2139 xen_p2m_addr = new_p2m;
2141 /* Release the old p2m list and set new list info. */
2142 p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2144 p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2146 if (xen_start_info->mfn_list < __START_KERNEL_map) {
2147 pfn = xen_start_info->first_p2m_pfn;
2148 pfn_end = xen_start_info->first_p2m_pfn +
2149 xen_start_info->nr_p2m_frames;
2150 set_pgd(pgd + 1, __pgd(0));
2153 pfn_end = p2m_pfn_end;
2156 memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2157 while (pfn < pfn_end) {
2158 if (pfn == p2m_pfn) {
2162 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2166 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2167 xen_start_info->first_p2m_pfn = PFN_DOWN(new_area);
2168 xen_start_info->nr_p2m_frames = n_frames;
2171 #else /* !CONFIG_X86_64 */
2172 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
2173 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
2175 static void __init xen_write_cr3_init(unsigned long cr3)
2177 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
2179 BUG_ON(read_cr3_pa() != __pa(initial_page_table));
2180 BUG_ON(cr3 != __pa(swapper_pg_dir));
2183 * We are switching to swapper_pg_dir for the first time (from
2184 * initial_page_table) and therefore need to mark that page
2185 * read-only and then pin it.
2187 * Xen disallows sharing of kernel PMDs for PAE
2188 * guests. Therefore we must copy the kernel PMD from
2189 * initial_page_table into a new kernel PMD to be used in
2192 swapper_kernel_pmd =
2193 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2194 copy_page(swapper_kernel_pmd, initial_kernel_pmd);
2195 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
2196 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
2197 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
2199 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
2201 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
2203 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
2204 PFN_DOWN(__pa(initial_page_table)));
2205 set_page_prot(initial_page_table, PAGE_KERNEL);
2206 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
2208 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2212 * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
2213 * not the first page table in the page table pool.
2214 * Iterate through the initial page tables to find the real page table base.
2216 static phys_addr_t __init xen_find_pt_base(pmd_t *pmd)
2218 phys_addr_t pt_base, paddr;
2221 pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
2223 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
2224 if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
2225 paddr = m2p(pmd[pmdidx].pmd);
2226 pt_base = min(pt_base, paddr);
2232 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2236 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2238 xen_pt_base = xen_find_pt_base(kernel_pmd);
2239 xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
2241 initial_kernel_pmd =
2242 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2244 max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
2246 copy_page(initial_kernel_pmd, kernel_pmd);
2248 xen_map_identity_early(initial_kernel_pmd, max_pfn);
2250 copy_page(initial_page_table, pgd);
2251 initial_page_table[KERNEL_PGD_BOUNDARY] =
2252 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2254 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2255 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2256 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2258 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2260 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2261 PFN_DOWN(__pa(initial_page_table)));
2262 xen_write_cr3(__pa(initial_page_table));
2264 memblock_reserve(xen_pt_base, xen_pt_size);
2266 #endif /* CONFIG_X86_64 */
2268 void __init xen_reserve_special_pages(void)
2272 memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2273 if (xen_start_info->store_mfn) {
2274 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2275 memblock_reserve(paddr, PAGE_SIZE);
2277 if (!xen_initial_domain()) {
2278 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2279 memblock_reserve(paddr, PAGE_SIZE);
2283 void __init xen_pt_check_e820(void)
2285 if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2286 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2291 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2293 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2297 phys >>= PAGE_SHIFT;
2300 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2301 #ifdef CONFIG_X86_32
2303 # ifdef CONFIG_HIGHMEM
2304 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2306 #elif defined(CONFIG_X86_VSYSCALL_EMULATION)
2309 case FIX_TEXT_POKE0:
2310 case FIX_TEXT_POKE1:
2311 /* All local page mappings */
2312 pte = pfn_pte(phys, prot);
2315 #ifdef CONFIG_X86_LOCAL_APIC
2316 case FIX_APIC_BASE: /* maps dummy local APIC */
2317 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2321 #ifdef CONFIG_X86_IO_APIC
2322 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2324 * We just don't map the IO APIC - all access is via
2325 * hypercalls. Keep the address in the pte for reference.
2327 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2331 case FIX_PARAVIRT_BOOTMAP:
2332 /* This is an MFN, but it isn't an IO mapping from the
2334 pte = mfn_pte(phys, prot);
2338 /* By default, set_fixmap is used for hardware mappings */
2339 pte = mfn_pte(phys, prot);
2343 __native_set_fixmap(idx, pte);
2345 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2346 /* Replicate changes to map the vsyscall page into the user
2347 pagetable vsyscall mapping. */
2348 if (idx == VSYSCALL_PAGE) {
2349 unsigned long vaddr = __fix_to_virt(idx);
2350 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2355 static void __init xen_post_allocator_init(void)
2357 pv_mmu_ops.set_pte = xen_set_pte;
2358 pv_mmu_ops.set_pmd = xen_set_pmd;
2359 pv_mmu_ops.set_pud = xen_set_pud;
2360 #ifdef CONFIG_X86_64
2361 pv_mmu_ops.set_p4d = xen_set_p4d;
2364 /* This will work as long as patching hasn't happened yet
2365 (which it hasn't) */
2366 pv_mmu_ops.alloc_pte = xen_alloc_pte;
2367 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2368 pv_mmu_ops.release_pte = xen_release_pte;
2369 pv_mmu_ops.release_pmd = xen_release_pmd;
2370 #ifdef CONFIG_X86_64
2371 pv_mmu_ops.alloc_pud = xen_alloc_pud;
2372 pv_mmu_ops.release_pud = xen_release_pud;
2374 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2376 #ifdef CONFIG_X86_64
2377 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2381 static void xen_leave_lazy_mmu(void)
2385 paravirt_leave_lazy_mmu();
2389 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2390 .read_cr2 = xen_read_cr2,
2391 .write_cr2 = xen_write_cr2,
2393 .read_cr3 = xen_read_cr3,
2394 .write_cr3 = xen_write_cr3_init,
2396 .flush_tlb_user = xen_flush_tlb,
2397 .flush_tlb_kernel = xen_flush_tlb,
2398 .flush_tlb_one_user = xen_flush_tlb_one_user,
2399 .flush_tlb_others = xen_flush_tlb_others,
2401 .pgd_alloc = xen_pgd_alloc,
2402 .pgd_free = xen_pgd_free,
2404 .alloc_pte = xen_alloc_pte_init,
2405 .release_pte = xen_release_pte_init,
2406 .alloc_pmd = xen_alloc_pmd_init,
2407 .release_pmd = xen_release_pmd_init,
2409 .set_pte = xen_set_pte_init,
2410 .set_pte_at = xen_set_pte_at,
2411 .set_pmd = xen_set_pmd_hyper,
2413 .ptep_modify_prot_start = __ptep_modify_prot_start,
2414 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2416 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2417 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2419 .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2420 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2422 #ifdef CONFIG_X86_PAE
2423 .set_pte_atomic = xen_set_pte_atomic,
2424 .pte_clear = xen_pte_clear,
2425 .pmd_clear = xen_pmd_clear,
2426 #endif /* CONFIG_X86_PAE */
2427 .set_pud = xen_set_pud_hyper,
2429 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2430 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2432 #ifdef CONFIG_X86_64
2433 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2434 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2435 .set_p4d = xen_set_p4d_hyper,
2437 .alloc_pud = xen_alloc_pmd_init,
2438 .release_pud = xen_release_pmd_init,
2440 #if CONFIG_PGTABLE_LEVELS >= 5
2441 .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2442 .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2444 #endif /* CONFIG_X86_64 */
2446 .activate_mm = xen_activate_mm,
2447 .dup_mmap = xen_dup_mmap,
2448 .exit_mmap = xen_exit_mmap,
2451 .enter = paravirt_enter_lazy_mmu,
2452 .leave = xen_leave_lazy_mmu,
2453 .flush = paravirt_flush_lazy_mmu,
2456 .set_fixmap = xen_set_fixmap,
2459 void __init xen_init_mmu_ops(void)
2461 x86_init.paging.pagetable_init = xen_pagetable_init;
2462 x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2464 pv_mmu_ops = xen_mmu_ops;
2466 memset(dummy_mapping, 0xff, PAGE_SIZE);
2469 /* Protected by xen_reservation_lock. */
2470 #define MAX_CONTIG_ORDER 9 /* 2MB */
2471 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2473 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2474 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2475 unsigned long *in_frames,
2476 unsigned long *out_frames)
2479 struct multicall_space mcs;
2482 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2483 mcs = __xen_mc_entry(0);
2486 in_frames[i] = virt_to_mfn(vaddr);
2488 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2489 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2492 out_frames[i] = virt_to_pfn(vaddr);
2498 * Update the pfn-to-mfn mappings for a virtual address range, either to
2499 * point to an array of mfns, or contiguously from a single starting
2502 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2503 unsigned long *mfns,
2504 unsigned long first_mfn)
2511 limit = 1u << order;
2512 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2513 struct multicall_space mcs;
2516 mcs = __xen_mc_entry(0);
2520 mfn = first_mfn + i;
2522 if (i < (limit - 1))
2526 flags = UVMF_INVLPG | UVMF_ALL;
2528 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2531 MULTI_update_va_mapping(mcs.mc, vaddr,
2532 mfn_pte(mfn, PAGE_KERNEL), flags);
2534 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2541 * Perform the hypercall to exchange a region of our pfns to point to
2542 * memory with the required contiguous alignment. Takes the pfns as
2543 * input, and populates mfns as output.
2545 * Returns a success code indicating whether the hypervisor was able to
2546 * satisfy the request or not.
2548 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2549 unsigned long *pfns_in,
2550 unsigned long extents_out,
2551 unsigned int order_out,
2552 unsigned long *mfns_out,
2553 unsigned int address_bits)
2558 struct xen_memory_exchange exchange = {
2560 .nr_extents = extents_in,
2561 .extent_order = order_in,
2562 .extent_start = pfns_in,
2566 .nr_extents = extents_out,
2567 .extent_order = order_out,
2568 .extent_start = mfns_out,
2569 .address_bits = address_bits,
2574 BUG_ON(extents_in << order_in != extents_out << order_out);
2576 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2577 success = (exchange.nr_exchanged == extents_in);
2579 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2580 BUG_ON(success && (rc != 0));
2585 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2586 unsigned int address_bits,
2587 dma_addr_t *dma_handle)
2589 unsigned long *in_frames = discontig_frames, out_frame;
2590 unsigned long flags;
2592 unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2595 * Currently an auto-translated guest will not perform I/O, nor will
2596 * it require PAE page directories below 4GB. Therefore any calls to
2597 * this function are redundant and can be ignored.
2600 if (unlikely(order > MAX_CONTIG_ORDER))
2603 memset((void *) vstart, 0, PAGE_SIZE << order);
2605 spin_lock_irqsave(&xen_reservation_lock, flags);
2607 /* 1. Zap current PTEs, remembering MFNs. */
2608 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2610 /* 2. Get a new contiguous memory extent. */
2611 out_frame = virt_to_pfn(vstart);
2612 success = xen_exchange_memory(1UL << order, 0, in_frames,
2613 1, order, &out_frame,
2616 /* 3. Map the new extent in place of old pages. */
2618 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2620 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2622 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2624 *dma_handle = virt_to_machine(vstart).maddr;
2625 return success ? 0 : -ENOMEM;
2627 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2629 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2631 unsigned long *out_frames = discontig_frames, in_frame;
2632 unsigned long flags;
2634 unsigned long vstart;
2636 if (unlikely(order > MAX_CONTIG_ORDER))
2639 vstart = (unsigned long)phys_to_virt(pstart);
2640 memset((void *) vstart, 0, PAGE_SIZE << order);
2642 spin_lock_irqsave(&xen_reservation_lock, flags);
2644 /* 1. Find start MFN of contiguous extent. */
2645 in_frame = virt_to_mfn(vstart);
2647 /* 2. Zap current PTEs. */
2648 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2650 /* 3. Do the exchange for non-contiguous MFNs. */
2651 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2654 /* 4. Map new pages in place of old pages. */
2656 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2658 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2660 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2662 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2664 #ifdef CONFIG_KEXEC_CORE
2665 phys_addr_t paddr_vmcoreinfo_note(void)
2667 if (xen_pv_domain())
2668 return virt_to_machine(vmcoreinfo_note).maddr;
2670 return __pa(vmcoreinfo_note);
2672 #endif /* CONFIG_KEXEC_CORE */