4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/module.h>
47 #include <linux/gfp.h>
48 #include <linux/memblock.h>
49 #include <linux/seq_file.h>
50 #include <linux/crash_dump.h>
52 #include <trace/events/xen.h>
54 #include <asm/pgtable.h>
55 #include <asm/tlbflush.h>
56 #include <asm/fixmap.h>
57 #include <asm/mmu_context.h>
58 #include <asm/setup.h>
59 #include <asm/paravirt.h>
61 #include <asm/linkage.h>
67 #include <asm/xen/hypercall.h>
68 #include <asm/xen/hypervisor.h>
72 #include <xen/interface/xen.h>
73 #include <xen/interface/hvm/hvm_op.h>
74 #include <xen/interface/version.h>
75 #include <xen/interface/memory.h>
76 #include <xen/hvc-console.h>
78 #include "multicalls.h"
83 * Protects atomic reservation decrease/increase against concurrent increases.
84 * Also protects non-atomic updates of current_pages and balloon lists.
86 DEFINE_SPINLOCK(xen_reservation_lock);
90 * Identity map, in addition to plain kernel map. This needs to be
91 * large enough to allocate page table pages to allocate the rest.
92 * Each page can map 2MB.
94 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
95 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
98 /* l3 pud for userspace vsyscall mapping */
99 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
100 #endif /* CONFIG_X86_64 */
103 * Note about cr3 (pagetable base) values:
105 * xen_cr3 contains the current logical cr3 value; it contains the
106 * last set cr3. This may not be the current effective cr3, because
107 * its update may be being lazily deferred. However, a vcpu looking
108 * at its own cr3 can use this value knowing that it everything will
109 * be self-consistent.
111 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
112 * hypercall to set the vcpu cr3 is complete (so it may be a little
113 * out of date, but it will never be set early). If one vcpu is
114 * looking at another vcpu's cr3 value, it should use this variable.
116 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
117 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
119 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
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 unsigned long arbitrary_virt_to_mfn(void *vaddr)
129 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
131 return PFN_DOWN(maddr.maddr);
134 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
136 unsigned long address = (unsigned long)vaddr;
142 * if the PFN is in the linear mapped vaddr range, we can just use
143 * the (quick) virt_to_machine() p2m lookup
145 if (virt_addr_valid(vaddr))
146 return virt_to_machine(vaddr);
148 /* otherwise we have to do a (slower) full page-table walk */
150 pte = lookup_address(address, &level);
152 offset = address & ~PAGE_MASK;
153 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
155 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
157 void make_lowmem_page_readonly(void *vaddr)
160 unsigned long address = (unsigned long)vaddr;
163 pte = lookup_address(address, &level);
165 return; /* vaddr missing */
167 ptev = pte_wrprotect(*pte);
169 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
173 void make_lowmem_page_readwrite(void *vaddr)
176 unsigned long address = (unsigned long)vaddr;
179 pte = lookup_address(address, &level);
181 return; /* vaddr missing */
183 ptev = pte_mkwrite(*pte);
185 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
190 static bool xen_page_pinned(void *ptr)
192 struct page *page = virt_to_page(ptr);
194 return PagePinned(page);
197 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
199 struct multicall_space mcs;
200 struct mmu_update *u;
202 trace_xen_mmu_set_domain_pte(ptep, pteval, domid);
204 mcs = xen_mc_entry(sizeof(*u));
207 /* ptep might be kmapped when using 32-bit HIGHPTE */
208 u->ptr = virt_to_machine(ptep).maddr;
209 u->val = pte_val_ma(pteval);
211 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
213 xen_mc_issue(PARAVIRT_LAZY_MMU);
215 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
217 static void xen_extend_mmu_update(const struct mmu_update *update)
219 struct multicall_space mcs;
220 struct mmu_update *u;
222 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
224 if (mcs.mc != NULL) {
227 mcs = __xen_mc_entry(sizeof(*u));
228 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
235 static void xen_extend_mmuext_op(const struct mmuext_op *op)
237 struct multicall_space mcs;
240 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
242 if (mcs.mc != NULL) {
245 mcs = __xen_mc_entry(sizeof(*u));
246 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
253 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
261 /* ptr may be ioremapped for 64-bit pagetable setup */
262 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
263 u.val = pmd_val_ma(val);
264 xen_extend_mmu_update(&u);
266 xen_mc_issue(PARAVIRT_LAZY_MMU);
271 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
273 trace_xen_mmu_set_pmd(ptr, val);
275 /* If page is not pinned, we can just update the entry
277 if (!xen_page_pinned(ptr)) {
282 xen_set_pmd_hyper(ptr, val);
286 * Associate a virtual page frame with a given physical page frame
287 * and protection flags for that frame.
289 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
291 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
294 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
298 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
303 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
304 u.val = pte_val_ma(pteval);
305 xen_extend_mmu_update(&u);
307 xen_mc_issue(PARAVIRT_LAZY_MMU);
312 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
314 if (!xen_batched_set_pte(ptep, pteval)) {
316 * Could call native_set_pte() here and trap and
317 * emulate the PTE write but with 32-bit guests this
318 * needs two traps (one for each of the two 32-bit
319 * words in the PTE) so do one hypercall directly
324 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
325 u.val = pte_val_ma(pteval);
326 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
330 static void xen_set_pte(pte_t *ptep, pte_t pteval)
332 trace_xen_mmu_set_pte(ptep, pteval);
333 __xen_set_pte(ptep, pteval);
336 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
337 pte_t *ptep, pte_t pteval)
339 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
340 __xen_set_pte(ptep, pteval);
343 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
344 unsigned long addr, pte_t *ptep)
346 /* Just return the pte as-is. We preserve the bits on commit */
347 trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
351 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
352 pte_t *ptep, pte_t pte)
356 trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
359 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
360 u.val = pte_val_ma(pte);
361 xen_extend_mmu_update(&u);
363 xen_mc_issue(PARAVIRT_LAZY_MMU);
366 /* Assume pteval_t is equivalent to all the other *val_t types. */
367 static pteval_t pte_mfn_to_pfn(pteval_t val)
369 if (val & _PAGE_PRESENT) {
370 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
371 unsigned long pfn = mfn_to_pfn(mfn);
373 pteval_t flags = val & PTE_FLAGS_MASK;
374 if (unlikely(pfn == ~0))
375 val = flags & ~_PAGE_PRESENT;
377 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
383 static pteval_t pte_pfn_to_mfn(pteval_t val)
385 if (val & _PAGE_PRESENT) {
386 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
387 pteval_t flags = val & PTE_FLAGS_MASK;
390 if (!xen_feature(XENFEAT_auto_translated_physmap))
391 mfn = __pfn_to_mfn(pfn);
395 * If there's no mfn for the pfn, then just create an
396 * empty non-present pte. Unfortunately this loses
397 * information about the original pfn, so
398 * pte_mfn_to_pfn is asymmetric.
400 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
404 mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
405 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
411 __visible pteval_t xen_pte_val(pte_t pte)
413 pteval_t pteval = pte.pte;
415 return pte_mfn_to_pfn(pteval);
417 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
419 __visible pgdval_t xen_pgd_val(pgd_t pgd)
421 return pte_mfn_to_pfn(pgd.pgd);
423 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
425 __visible pte_t xen_make_pte(pteval_t pte)
427 pte = pte_pfn_to_mfn(pte);
429 return native_make_pte(pte);
431 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
433 __visible pgd_t xen_make_pgd(pgdval_t pgd)
435 pgd = pte_pfn_to_mfn(pgd);
436 return native_make_pgd(pgd);
438 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
440 __visible pmdval_t xen_pmd_val(pmd_t pmd)
442 return pte_mfn_to_pfn(pmd.pmd);
444 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
446 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
454 /* ptr may be ioremapped for 64-bit pagetable setup */
455 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
456 u.val = pud_val_ma(val);
457 xen_extend_mmu_update(&u);
459 xen_mc_issue(PARAVIRT_LAZY_MMU);
464 static void xen_set_pud(pud_t *ptr, pud_t val)
466 trace_xen_mmu_set_pud(ptr, val);
468 /* If page is not pinned, we can just update the entry
470 if (!xen_page_pinned(ptr)) {
475 xen_set_pud_hyper(ptr, val);
478 #ifdef CONFIG_X86_PAE
479 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
481 trace_xen_mmu_set_pte_atomic(ptep, pte);
482 set_64bit((u64 *)ptep, native_pte_val(pte));
485 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
487 trace_xen_mmu_pte_clear(mm, addr, ptep);
488 if (!xen_batched_set_pte(ptep, native_make_pte(0)))
489 native_pte_clear(mm, addr, ptep);
492 static void xen_pmd_clear(pmd_t *pmdp)
494 trace_xen_mmu_pmd_clear(pmdp);
495 set_pmd(pmdp, __pmd(0));
497 #endif /* CONFIG_X86_PAE */
499 __visible pmd_t xen_make_pmd(pmdval_t pmd)
501 pmd = pte_pfn_to_mfn(pmd);
502 return native_make_pmd(pmd);
504 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
506 #if CONFIG_PGTABLE_LEVELS == 4
507 __visible pudval_t xen_pud_val(pud_t pud)
509 return pte_mfn_to_pfn(pud.pud);
511 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
513 __visible pud_t xen_make_pud(pudval_t pud)
515 pud = pte_pfn_to_mfn(pud);
517 return native_make_pud(pud);
519 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
521 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
523 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
524 unsigned offset = pgd - pgd_page;
525 pgd_t *user_ptr = NULL;
527 if (offset < pgd_index(USER_LIMIT)) {
528 struct page *page = virt_to_page(pgd_page);
529 user_ptr = (pgd_t *)page->private;
537 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
541 u.ptr = virt_to_machine(ptr).maddr;
542 u.val = pgd_val_ma(val);
543 xen_extend_mmu_update(&u);
547 * Raw hypercall-based set_pgd, intended for in early boot before
548 * there's a page structure. This implies:
549 * 1. The only existing pagetable is the kernel's
550 * 2. It is always pinned
551 * 3. It has no user pagetable attached to it
553 static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
559 __xen_set_pgd_hyper(ptr, val);
561 xen_mc_issue(PARAVIRT_LAZY_MMU);
566 static void xen_set_pgd(pgd_t *ptr, pgd_t val)
568 pgd_t *user_ptr = xen_get_user_pgd(ptr);
570 trace_xen_mmu_set_pgd(ptr, user_ptr, val);
572 /* If page is not pinned, we can just update the entry
574 if (!xen_page_pinned(ptr)) {
577 WARN_ON(xen_page_pinned(user_ptr));
583 /* If it's pinned, then we can at least batch the kernel and
584 user updates together. */
587 __xen_set_pgd_hyper(ptr, val);
589 __xen_set_pgd_hyper(user_ptr, val);
591 xen_mc_issue(PARAVIRT_LAZY_MMU);
593 #endif /* CONFIG_PGTABLE_LEVELS == 4 */
596 * (Yet another) pagetable walker. This one is intended for pinning a
597 * pagetable. This means that it walks a pagetable and calls the
598 * callback function on each page it finds making up the page table,
599 * at every level. It walks the entire pagetable, but it only bothers
600 * pinning pte pages which are below limit. In the normal case this
601 * will be STACK_TOP_MAX, but at boot we need to pin up to
604 * For 32-bit the important bit is that we don't pin beyond there,
605 * because then we start getting into Xen's ptes.
607 * For 64-bit, we must skip the Xen hole in the middle of the address
608 * space, just after the big x86-64 virtual hole.
610 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
611 int (*func)(struct mm_struct *mm, struct page *,
616 unsigned hole_low, hole_high;
617 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
618 unsigned pgdidx, pudidx, pmdidx;
620 /* The limit is the last byte to be touched */
622 BUG_ON(limit >= FIXADDR_TOP);
624 if (xen_feature(XENFEAT_auto_translated_physmap))
628 * 64-bit has a great big hole in the middle of the address
629 * space, which contains the Xen mappings. On 32-bit these
630 * will end up making a zero-sized hole and so is a no-op.
632 hole_low = pgd_index(USER_LIMIT);
633 hole_high = pgd_index(PAGE_OFFSET);
635 pgdidx_limit = pgd_index(limit);
637 pudidx_limit = pud_index(limit);
642 pmdidx_limit = pmd_index(limit);
647 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
650 if (pgdidx >= hole_low && pgdidx < hole_high)
653 if (!pgd_val(pgd[pgdidx]))
656 pud = pud_offset(&pgd[pgdidx], 0);
658 if (PTRS_PER_PUD > 1) /* not folded */
659 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
661 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
664 if (pgdidx == pgdidx_limit &&
665 pudidx > pudidx_limit)
668 if (pud_none(pud[pudidx]))
671 pmd = pmd_offset(&pud[pudidx], 0);
673 if (PTRS_PER_PMD > 1) /* not folded */
674 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
676 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
679 if (pgdidx == pgdidx_limit &&
680 pudidx == pudidx_limit &&
681 pmdidx > pmdidx_limit)
684 if (pmd_none(pmd[pmdidx]))
687 pte = pmd_page(pmd[pmdidx]);
688 flush |= (*func)(mm, pte, PT_PTE);
694 /* Do the top level last, so that the callbacks can use it as
695 a cue to do final things like tlb flushes. */
696 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
701 static int xen_pgd_walk(struct mm_struct *mm,
702 int (*func)(struct mm_struct *mm, struct page *,
706 return __xen_pgd_walk(mm, mm->pgd, func, limit);
709 /* If we're using split pte locks, then take the page's lock and
710 return a pointer to it. Otherwise return NULL. */
711 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
713 spinlock_t *ptl = NULL;
715 #if USE_SPLIT_PTE_PTLOCKS
716 ptl = ptlock_ptr(page);
717 spin_lock_nest_lock(ptl, &mm->page_table_lock);
723 static void xen_pte_unlock(void *v)
729 static void xen_do_pin(unsigned level, unsigned long pfn)
734 op.arg1.mfn = pfn_to_mfn(pfn);
736 xen_extend_mmuext_op(&op);
739 static int xen_pin_page(struct mm_struct *mm, struct page *page,
742 unsigned pgfl = TestSetPagePinned(page);
746 flush = 0; /* already pinned */
747 else if (PageHighMem(page))
748 /* kmaps need flushing if we found an unpinned
752 void *pt = lowmem_page_address(page);
753 unsigned long pfn = page_to_pfn(page);
754 struct multicall_space mcs = __xen_mc_entry(0);
760 * We need to hold the pagetable lock between the time
761 * we make the pagetable RO and when we actually pin
762 * it. If we don't, then other users may come in and
763 * attempt to update the pagetable by writing it,
764 * which will fail because the memory is RO but not
765 * pinned, so Xen won't do the trap'n'emulate.
767 * If we're using split pte locks, we can't hold the
768 * entire pagetable's worth of locks during the
769 * traverse, because we may wrap the preempt count (8
770 * bits). The solution is to mark RO and pin each PTE
771 * page while holding the lock. This means the number
772 * of locks we end up holding is never more than a
773 * batch size (~32 entries, at present).
775 * If we're not using split pte locks, we needn't pin
776 * the PTE pages independently, because we're
777 * protected by the overall pagetable lock.
781 ptl = xen_pte_lock(page, mm);
783 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
784 pfn_pte(pfn, PAGE_KERNEL_RO),
785 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
788 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
790 /* Queue a deferred unlock for when this batch
792 xen_mc_callback(xen_pte_unlock, ptl);
799 /* This is called just after a mm has been created, but it has not
800 been used yet. We need to make sure that its pagetable is all
801 read-only, and can be pinned. */
802 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
804 trace_xen_mmu_pgd_pin(mm, pgd);
808 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
809 /* re-enable interrupts for flushing */
819 pgd_t *user_pgd = xen_get_user_pgd(pgd);
821 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
824 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
825 xen_do_pin(MMUEXT_PIN_L4_TABLE,
826 PFN_DOWN(__pa(user_pgd)));
829 #else /* CONFIG_X86_32 */
830 #ifdef CONFIG_X86_PAE
831 /* Need to make sure unshared kernel PMD is pinnable */
832 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
835 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
836 #endif /* CONFIG_X86_64 */
840 static void xen_pgd_pin(struct mm_struct *mm)
842 __xen_pgd_pin(mm, mm->pgd);
846 * On save, we need to pin all pagetables to make sure they get their
847 * mfns turned into pfns. Search the list for any unpinned pgds and pin
848 * them (unpinned pgds are not currently in use, probably because the
849 * process is under construction or destruction).
851 * Expected to be called in stop_machine() ("equivalent to taking
852 * every spinlock in the system"), so the locking doesn't really
853 * matter all that much.
855 void xen_mm_pin_all(void)
859 spin_lock(&pgd_lock);
861 list_for_each_entry(page, &pgd_list, lru) {
862 if (!PagePinned(page)) {
863 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
864 SetPageSavePinned(page);
868 spin_unlock(&pgd_lock);
872 * The init_mm pagetable is really pinned as soon as its created, but
873 * that's before we have page structures to store the bits. So do all
874 * the book-keeping now.
876 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
883 static void __init xen_mark_init_mm_pinned(void)
885 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
888 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
891 unsigned pgfl = TestClearPagePinned(page);
893 if (pgfl && !PageHighMem(page)) {
894 void *pt = lowmem_page_address(page);
895 unsigned long pfn = page_to_pfn(page);
896 spinlock_t *ptl = NULL;
897 struct multicall_space mcs;
900 * Do the converse to pin_page. If we're using split
901 * pte locks, we must be holding the lock for while
902 * the pte page is unpinned but still RO to prevent
903 * concurrent updates from seeing it in this
904 * partially-pinned state.
906 if (level == PT_PTE) {
907 ptl = xen_pte_lock(page, mm);
910 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
913 mcs = __xen_mc_entry(0);
915 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
916 pfn_pte(pfn, PAGE_KERNEL),
917 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
920 /* unlock when batch completed */
921 xen_mc_callback(xen_pte_unlock, ptl);
925 return 0; /* never need to flush on unpin */
928 /* Release a pagetables pages back as normal RW */
929 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
931 trace_xen_mmu_pgd_unpin(mm, pgd);
935 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
939 pgd_t *user_pgd = xen_get_user_pgd(pgd);
942 xen_do_pin(MMUEXT_UNPIN_TABLE,
943 PFN_DOWN(__pa(user_pgd)));
944 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
949 #ifdef CONFIG_X86_PAE
950 /* Need to make sure unshared kernel PMD is unpinned */
951 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
955 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
960 static void xen_pgd_unpin(struct mm_struct *mm)
962 __xen_pgd_unpin(mm, mm->pgd);
966 * On resume, undo any pinning done at save, so that the rest of the
967 * kernel doesn't see any unexpected pinned pagetables.
969 void xen_mm_unpin_all(void)
973 spin_lock(&pgd_lock);
975 list_for_each_entry(page, &pgd_list, lru) {
976 if (PageSavePinned(page)) {
977 BUG_ON(!PagePinned(page));
978 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
979 ClearPageSavePinned(page);
983 spin_unlock(&pgd_lock);
986 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
988 spin_lock(&next->page_table_lock);
990 spin_unlock(&next->page_table_lock);
993 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
995 spin_lock(&mm->page_table_lock);
997 spin_unlock(&mm->page_table_lock);
1002 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1003 we need to repoint it somewhere else before we can unpin it. */
1004 static void drop_other_mm_ref(void *info)
1006 struct mm_struct *mm = info;
1007 struct mm_struct *active_mm;
1009 active_mm = this_cpu_read(cpu_tlbstate.active_mm);
1011 if (active_mm == mm && this_cpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
1012 leave_mm(smp_processor_id());
1014 /* If this cpu still has a stale cr3 reference, then make sure
1015 it has been flushed. */
1016 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
1017 load_cr3(swapper_pg_dir);
1020 static void xen_drop_mm_ref(struct mm_struct *mm)
1025 if (current->active_mm == mm) {
1026 if (current->mm == mm)
1027 load_cr3(swapper_pg_dir);
1029 leave_mm(smp_processor_id());
1032 /* Get the "official" set of cpus referring to our pagetable. */
1033 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1034 for_each_online_cpu(cpu) {
1035 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1036 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1038 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1042 cpumask_copy(mask, mm_cpumask(mm));
1044 /* It's possible that a vcpu may have a stale reference to our
1045 cr3, because its in lazy mode, and it hasn't yet flushed
1046 its set of pending hypercalls yet. In this case, we can
1047 look at its actual current cr3 value, and force it to flush
1049 for_each_online_cpu(cpu) {
1050 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1051 cpumask_set_cpu(cpu, mask);
1054 if (!cpumask_empty(mask))
1055 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1056 free_cpumask_var(mask);
1059 static void xen_drop_mm_ref(struct mm_struct *mm)
1061 if (current->active_mm == mm)
1062 load_cr3(swapper_pg_dir);
1067 * While a process runs, Xen pins its pagetables, which means that the
1068 * hypervisor forces it to be read-only, and it controls all updates
1069 * to it. This means that all pagetable updates have to go via the
1070 * hypervisor, which is moderately expensive.
1072 * Since we're pulling the pagetable down, we switch to use init_mm,
1073 * unpin old process pagetable and mark it all read-write, which
1074 * allows further operations on it to be simple memory accesses.
1076 * The only subtle point is that another CPU may be still using the
1077 * pagetable because of lazy tlb flushing. This means we need need to
1078 * switch all CPUs off this pagetable before we can unpin it.
1080 static void xen_exit_mmap(struct mm_struct *mm)
1082 get_cpu(); /* make sure we don't move around */
1083 xen_drop_mm_ref(mm);
1086 spin_lock(&mm->page_table_lock);
1088 /* pgd may not be pinned in the error exit path of execve */
1089 if (xen_page_pinned(mm->pgd))
1092 spin_unlock(&mm->page_table_lock);
1095 static void xen_post_allocator_init(void);
1097 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1099 struct mmuext_op op;
1102 op.arg1.mfn = pfn_to_mfn(pfn);
1103 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1107 #ifdef CONFIG_X86_64
1108 static void __init xen_cleanhighmap(unsigned long vaddr,
1109 unsigned long vaddr_end)
1111 unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1112 pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1114 /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1115 * We include the PMD passed in on _both_ boundaries. */
1116 for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1117 pmd++, vaddr += PMD_SIZE) {
1120 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1121 set_pmd(pmd, __pmd(0));
1123 /* In case we did something silly, we should crash in this function
1124 * instead of somewhere later and be confusing. */
1129 * Make a page range writeable and free it.
1131 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1133 void *vaddr = __va(paddr);
1134 void *vaddr_end = vaddr + size;
1136 for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1137 make_lowmem_page_readwrite(vaddr);
1139 memblock_free(paddr, size);
1142 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1144 unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1147 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1148 ClearPagePinned(virt_to_page(__va(pa)));
1149 xen_free_ro_pages(pa, PAGE_SIZE);
1153 * Since it is well isolated we can (and since it is perhaps large we should)
1154 * also free the page tables mapping the initial P->M table.
1156 static void __init xen_cleanmfnmap(unsigned long vaddr)
1158 unsigned long va = vaddr & PMD_MASK;
1160 pgd_t *pgd = pgd_offset_k(va);
1161 pud_t *pud_page = pud_offset(pgd, 0);
1168 unpin = (vaddr == 2 * PGDIR_SIZE);
1169 set_pgd(pgd, __pgd(0));
1171 pud = pud_page + pud_index(va);
1172 if (pud_none(*pud)) {
1174 } else if (pud_large(*pud)) {
1175 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1176 xen_free_ro_pages(pa, PUD_SIZE);
1179 pmd = pmd_offset(pud, va);
1180 if (pmd_large(*pmd)) {
1181 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1182 xen_free_ro_pages(pa, PMD_SIZE);
1183 } else if (!pmd_none(*pmd)) {
1184 pte = pte_offset_kernel(pmd, va);
1185 set_pmd(pmd, __pmd(0));
1186 for (i = 0; i < PTRS_PER_PTE; ++i) {
1187 if (pte_none(pte[i]))
1189 pa = pte_pfn(pte[i]) << PAGE_SHIFT;
1190 xen_free_ro_pages(pa, PAGE_SIZE);
1192 xen_cleanmfnmap_free_pgtbl(pte, unpin);
1197 set_pud(pud, __pud(0));
1198 xen_cleanmfnmap_free_pgtbl(pmd, unpin);
1201 } while (pud_index(va) || pmd_index(va));
1202 xen_cleanmfnmap_free_pgtbl(pud_page, unpin);
1205 static void __init xen_pagetable_p2m_free(void)
1210 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1212 /* No memory or already called. */
1213 if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1216 /* using __ka address and sticking INVALID_P2M_ENTRY! */
1217 memset((void *)xen_start_info->mfn_list, 0xff, size);
1219 addr = xen_start_info->mfn_list;
1221 * We could be in __ka space.
1222 * We roundup to the PMD, which means that if anybody at this stage is
1223 * using the __ka address of xen_start_info or
1224 * xen_start_info->shared_info they are in going to crash. Fortunatly
1225 * we have already revectored in xen_setup_kernel_pagetable and in
1226 * xen_setup_shared_info.
1228 size = roundup(size, PMD_SIZE);
1230 if (addr >= __START_KERNEL_map) {
1231 xen_cleanhighmap(addr, addr + size);
1232 size = PAGE_ALIGN(xen_start_info->nr_pages *
1233 sizeof(unsigned long));
1234 memblock_free(__pa(addr), size);
1236 xen_cleanmfnmap(addr);
1240 static void __init xen_pagetable_cleanhighmap(void)
1245 /* At this stage, cleanup_highmap has already cleaned __ka space
1246 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1247 * the ramdisk). We continue on, erasing PMD entries that point to page
1248 * tables - do note that they are accessible at this stage via __va.
1249 * For good measure we also round up to the PMD - which means that if
1250 * anybody is using __ka address to the initial boot-stack - and try
1251 * to use it - they are going to crash. The xen_start_info has been
1252 * taken care of already in xen_setup_kernel_pagetable. */
1253 addr = xen_start_info->pt_base;
1254 size = roundup(xen_start_info->nr_pt_frames * PAGE_SIZE, PMD_SIZE);
1256 xen_cleanhighmap(addr, addr + size);
1257 xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1259 /* This is superflous and is not neccessary, but you know what
1260 * lets do it. The MODULES_VADDR -> MODULES_END should be clear of
1261 * anything at this stage. */
1262 xen_cleanhighmap(MODULES_VADDR, roundup(MODULES_VADDR, PUD_SIZE) - 1);
1267 static void __init xen_pagetable_p2m_setup(void)
1269 if (xen_feature(XENFEAT_auto_translated_physmap))
1272 xen_vmalloc_p2m_tree();
1274 #ifdef CONFIG_X86_64
1275 xen_pagetable_p2m_free();
1277 xen_pagetable_cleanhighmap();
1279 /* And revector! Bye bye old array */
1280 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1283 static void __init xen_pagetable_init(void)
1286 xen_post_allocator_init();
1288 xen_pagetable_p2m_setup();
1290 /* Allocate and initialize top and mid mfn levels for p2m structure */
1291 xen_build_mfn_list_list();
1293 /* Remap memory freed due to conflicts with E820 map */
1294 if (!xen_feature(XENFEAT_auto_translated_physmap))
1297 xen_setup_shared_info();
1299 static void xen_write_cr2(unsigned long cr2)
1301 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1304 static unsigned long xen_read_cr2(void)
1306 return this_cpu_read(xen_vcpu)->arch.cr2;
1309 unsigned long xen_read_cr2_direct(void)
1311 return this_cpu_read(xen_vcpu_info.arch.cr2);
1314 void xen_flush_tlb_all(void)
1316 struct mmuext_op *op;
1317 struct multicall_space mcs;
1321 mcs = xen_mc_entry(sizeof(*op));
1324 op->cmd = MMUEXT_TLB_FLUSH_ALL;
1325 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1327 xen_mc_issue(PARAVIRT_LAZY_MMU);
1331 static void xen_flush_tlb(void)
1333 struct mmuext_op *op;
1334 struct multicall_space mcs;
1338 mcs = xen_mc_entry(sizeof(*op));
1341 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1342 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1344 xen_mc_issue(PARAVIRT_LAZY_MMU);
1349 static void xen_flush_tlb_single(unsigned long addr)
1351 struct mmuext_op *op;
1352 struct multicall_space mcs;
1354 trace_xen_mmu_flush_tlb_single(addr);
1358 mcs = xen_mc_entry(sizeof(*op));
1360 op->cmd = MMUEXT_INVLPG_LOCAL;
1361 op->arg1.linear_addr = addr & PAGE_MASK;
1362 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1364 xen_mc_issue(PARAVIRT_LAZY_MMU);
1369 static void xen_flush_tlb_others(const struct cpumask *cpus,
1370 struct mm_struct *mm, unsigned long start,
1374 struct mmuext_op op;
1376 DECLARE_BITMAP(mask, num_processors);
1378 DECLARE_BITMAP(mask, NR_CPUS);
1381 struct multicall_space mcs;
1383 trace_xen_mmu_flush_tlb_others(cpus, mm, start, end);
1385 if (cpumask_empty(cpus))
1386 return; /* nothing to do */
1388 mcs = xen_mc_entry(sizeof(*args));
1390 args->op.arg2.vcpumask = to_cpumask(args->mask);
1392 /* Remove us, and any offline CPUS. */
1393 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1394 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1396 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1397 if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) {
1398 args->op.cmd = MMUEXT_INVLPG_MULTI;
1399 args->op.arg1.linear_addr = start;
1402 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1404 xen_mc_issue(PARAVIRT_LAZY_MMU);
1407 static unsigned long xen_read_cr3(void)
1409 return this_cpu_read(xen_cr3);
1412 static void set_current_cr3(void *v)
1414 this_cpu_write(xen_current_cr3, (unsigned long)v);
1417 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1419 struct mmuext_op op;
1422 trace_xen_mmu_write_cr3(kernel, cr3);
1425 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1429 WARN_ON(mfn == 0 && kernel);
1431 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1434 xen_extend_mmuext_op(&op);
1437 this_cpu_write(xen_cr3, cr3);
1439 /* Update xen_current_cr3 once the batch has actually
1441 xen_mc_callback(set_current_cr3, (void *)cr3);
1444 static void xen_write_cr3(unsigned long cr3)
1446 BUG_ON(preemptible());
1448 xen_mc_batch(); /* disables interrupts */
1450 /* Update while interrupts are disabled, so its atomic with
1452 this_cpu_write(xen_cr3, cr3);
1454 __xen_write_cr3(true, cr3);
1456 #ifdef CONFIG_X86_64
1458 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1460 __xen_write_cr3(false, __pa(user_pgd));
1462 __xen_write_cr3(false, 0);
1466 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1469 #ifdef CONFIG_X86_64
1471 * At the start of the day - when Xen launches a guest, it has already
1472 * built pagetables for the guest. We diligently look over them
1473 * in xen_setup_kernel_pagetable and graft as appropiate them in the
1474 * init_level4_pgt and its friends. Then when we are happy we load
1475 * the new init_level4_pgt - and continue on.
1477 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1478 * up the rest of the pagetables. When it has completed it loads the cr3.
1479 * N.B. that baremetal would start at 'start_kernel' (and the early
1480 * #PF handler would create bootstrap pagetables) - so we are running
1481 * with the same assumptions as what to do when write_cr3 is executed
1484 * Since there are no user-page tables at all, we have two variants
1485 * of xen_write_cr3 - the early bootup (this one), and the late one
1486 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1487 * the Linux kernel and user-space are both in ring 3 while the
1488 * hypervisor is in ring 0.
1490 static void __init xen_write_cr3_init(unsigned long cr3)
1492 BUG_ON(preemptible());
1494 xen_mc_batch(); /* disables interrupts */
1496 /* Update while interrupts are disabled, so its atomic with
1498 this_cpu_write(xen_cr3, cr3);
1500 __xen_write_cr3(true, cr3);
1502 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1506 static int xen_pgd_alloc(struct mm_struct *mm)
1508 pgd_t *pgd = mm->pgd;
1511 BUG_ON(PagePinned(virt_to_page(pgd)));
1513 #ifdef CONFIG_X86_64
1515 struct page *page = virt_to_page(pgd);
1518 BUG_ON(page->private != 0);
1522 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1523 page->private = (unsigned long)user_pgd;
1525 if (user_pgd != NULL) {
1526 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1527 user_pgd[pgd_index(VSYSCALL_ADDR)] =
1528 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1533 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1540 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1542 #ifdef CONFIG_X86_64
1543 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1546 free_page((unsigned long)user_pgd);
1550 #ifdef CONFIG_X86_32
1551 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1553 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1554 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1555 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1560 #else /* CONFIG_X86_64 */
1561 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1565 if (xen_feature(XENFEAT_writable_page_tables) ||
1566 xen_feature(XENFEAT_auto_translated_physmap) ||
1567 xen_start_info->mfn_list >= __START_KERNEL_map)
1571 * Pages belonging to the initial p2m list mapped outside the default
1572 * address range must be mapped read-only. This region contains the
1573 * page tables for mapping the p2m list, too, and page tables MUST be
1577 if (pfn >= xen_start_info->first_p2m_pfn &&
1578 pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1579 pte = __pte_ma(pte_val_ma(pte) & ~_PAGE_RW);
1583 #endif /* CONFIG_X86_64 */
1586 * Init-time set_pte while constructing initial pagetables, which
1587 * doesn't allow RO page table pages to be remapped RW.
1589 * If there is no MFN for this PFN then this page is initially
1590 * ballooned out so clear the PTE (as in decrease_reservation() in
1591 * drivers/xen/balloon.c).
1593 * Many of these PTE updates are done on unpinned and writable pages
1594 * and doing a hypercall for these is unnecessary and expensive. At
1595 * this point it is not possible to tell if a page is pinned or not,
1596 * so always write the PTE directly and rely on Xen trapping and
1597 * emulating any updates as necessary.
1599 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1601 if (pte_mfn(pte) != INVALID_P2M_ENTRY)
1602 pte = mask_rw_pte(ptep, pte);
1606 native_set_pte(ptep, pte);
1609 /* Early in boot, while setting up the initial pagetable, assume
1610 everything is pinned. */
1611 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1613 #ifdef CONFIG_FLATMEM
1614 BUG_ON(mem_map); /* should only be used early */
1616 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1617 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1620 /* Used for pmd and pud */
1621 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1623 #ifdef CONFIG_FLATMEM
1624 BUG_ON(mem_map); /* should only be used early */
1626 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1629 /* Early release_pte assumes that all pts are pinned, since there's
1630 only init_mm and anything attached to that is pinned. */
1631 static void __init xen_release_pte_init(unsigned long pfn)
1633 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1634 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1637 static void __init xen_release_pmd_init(unsigned long pfn)
1639 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1642 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1644 struct multicall_space mcs;
1645 struct mmuext_op *op;
1647 mcs = __xen_mc_entry(sizeof(*op));
1650 op->arg1.mfn = pfn_to_mfn(pfn);
1652 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1655 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1657 struct multicall_space mcs;
1658 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1660 mcs = __xen_mc_entry(0);
1661 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1662 pfn_pte(pfn, prot), 0);
1665 /* This needs to make sure the new pte page is pinned iff its being
1666 attached to a pinned pagetable. */
1667 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1670 bool pinned = PagePinned(virt_to_page(mm->pgd));
1672 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1675 struct page *page = pfn_to_page(pfn);
1677 SetPagePinned(page);
1679 if (!PageHighMem(page)) {
1682 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1684 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1685 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1687 xen_mc_issue(PARAVIRT_LAZY_MMU);
1689 /* make sure there are no stray mappings of
1691 kmap_flush_unused();
1696 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1698 xen_alloc_ptpage(mm, pfn, PT_PTE);
1701 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1703 xen_alloc_ptpage(mm, pfn, PT_PMD);
1706 /* This should never happen until we're OK to use struct page */
1707 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1709 struct page *page = pfn_to_page(pfn);
1710 bool pinned = PagePinned(page);
1712 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1715 if (!PageHighMem(page)) {
1718 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1719 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1721 __set_pfn_prot(pfn, PAGE_KERNEL);
1723 xen_mc_issue(PARAVIRT_LAZY_MMU);
1725 ClearPagePinned(page);
1729 static void xen_release_pte(unsigned long pfn)
1731 xen_release_ptpage(pfn, PT_PTE);
1734 static void xen_release_pmd(unsigned long pfn)
1736 xen_release_ptpage(pfn, PT_PMD);
1739 #if CONFIG_PGTABLE_LEVELS == 4
1740 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1742 xen_alloc_ptpage(mm, pfn, PT_PUD);
1745 static void xen_release_pud(unsigned long pfn)
1747 xen_release_ptpage(pfn, PT_PUD);
1751 void __init xen_reserve_top(void)
1753 #ifdef CONFIG_X86_32
1754 unsigned long top = HYPERVISOR_VIRT_START;
1755 struct xen_platform_parameters pp;
1757 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1758 top = pp.virt_start;
1760 reserve_top_address(-top);
1761 #endif /* CONFIG_X86_32 */
1765 * Like __va(), but returns address in the kernel mapping (which is
1766 * all we have until the physical memory mapping has been set up.
1768 static void * __init __ka(phys_addr_t paddr)
1770 #ifdef CONFIG_X86_64
1771 return (void *)(paddr + __START_KERNEL_map);
1777 /* Convert a machine address to physical address */
1778 static unsigned long __init m2p(phys_addr_t maddr)
1782 maddr &= PTE_PFN_MASK;
1783 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1788 /* Convert a machine address to kernel virtual */
1789 static void * __init m2v(phys_addr_t maddr)
1791 return __ka(m2p(maddr));
1794 /* Set the page permissions on an identity-mapped pages */
1795 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1796 unsigned long flags)
1798 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1799 pte_t pte = pfn_pte(pfn, prot);
1801 /* For PVH no need to set R/O or R/W to pin them or unpin them. */
1802 if (xen_feature(XENFEAT_auto_translated_physmap))
1805 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1808 static void __init set_page_prot(void *addr, pgprot_t prot)
1810 return set_page_prot_flags(addr, prot, UVMF_NONE);
1812 #ifdef CONFIG_X86_32
1813 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1815 unsigned pmdidx, pteidx;
1819 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1824 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1827 /* Reuse or allocate a page of ptes */
1828 if (pmd_present(pmd[pmdidx]))
1829 pte_page = m2v(pmd[pmdidx].pmd);
1831 /* Check for free pte pages */
1832 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1835 pte_page = &level1_ident_pgt[ident_pte];
1836 ident_pte += PTRS_PER_PTE;
1838 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1841 /* Install mappings */
1842 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1845 if (pfn > max_pfn_mapped)
1846 max_pfn_mapped = pfn;
1848 if (!pte_none(pte_page[pteidx]))
1851 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1852 pte_page[pteidx] = pte;
1856 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1857 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1859 set_page_prot(pmd, PAGE_KERNEL_RO);
1862 void __init xen_setup_machphys_mapping(void)
1864 struct xen_machphys_mapping mapping;
1866 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1867 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1868 machine_to_phys_nr = mapping.max_mfn + 1;
1870 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1872 #ifdef CONFIG_X86_32
1873 WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1874 < machine_to_phys_mapping);
1878 #ifdef CONFIG_X86_64
1879 static void __init convert_pfn_mfn(void *v)
1884 /* All levels are converted the same way, so just treat them
1886 for (i = 0; i < PTRS_PER_PTE; i++)
1887 pte[i] = xen_make_pte(pte[i].pte);
1889 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1892 if (*pt_base == PFN_DOWN(__pa(addr))) {
1893 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1894 clear_page((void *)addr);
1897 if (*pt_end == PFN_DOWN(__pa(addr))) {
1898 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1899 clear_page((void *)addr);
1904 * Set up the initial kernel pagetable.
1906 * We can construct this by grafting the Xen provided pagetable into
1907 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1908 * level2_ident_pgt, and level2_kernel_pgt. This means that only the
1909 * kernel has a physical mapping to start with - but that's enough to
1910 * get __va working. We need to fill in the rest of the physical
1911 * mapping once some sort of allocator has been set up. NOTE: for
1912 * PVH, the page tables are native.
1914 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1918 unsigned long addr[3];
1919 unsigned long pt_base, pt_end;
1922 /* max_pfn_mapped is the last pfn mapped in the initial memory
1923 * mappings. Considering that on Xen after the kernel mappings we
1924 * have the mappings of some pages that don't exist in pfn space, we
1925 * set max_pfn_mapped to the last real pfn mapped. */
1926 if (xen_start_info->mfn_list < __START_KERNEL_map)
1927 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1929 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1931 pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1932 pt_end = pt_base + xen_start_info->nr_pt_frames;
1934 /* Zap identity mapping */
1935 init_level4_pgt[0] = __pgd(0);
1937 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1938 /* Pre-constructed entries are in pfn, so convert to mfn */
1939 /* L4[272] -> level3_ident_pgt
1940 * L4[511] -> level3_kernel_pgt */
1941 convert_pfn_mfn(init_level4_pgt);
1943 /* L3_i[0] -> level2_ident_pgt */
1944 convert_pfn_mfn(level3_ident_pgt);
1945 /* L3_k[510] -> level2_kernel_pgt
1946 * L3_k[511] -> level2_fixmap_pgt */
1947 convert_pfn_mfn(level3_kernel_pgt);
1949 /* L3_k[511][506] -> level1_fixmap_pgt */
1950 convert_pfn_mfn(level2_fixmap_pgt);
1952 /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1953 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1954 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1956 addr[0] = (unsigned long)pgd;
1957 addr[1] = (unsigned long)l3;
1958 addr[2] = (unsigned long)l2;
1959 /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1960 * Both L4[272][0] and L4[511][510] have entries that point to the same
1961 * L2 (PMD) tables. Meaning that if you modify it in __va space
1962 * it will be also modified in the __ka space! (But if you just
1963 * modify the PMD table to point to other PTE's or none, then you
1964 * are OK - which is what cleanup_highmap does) */
1965 copy_page(level2_ident_pgt, l2);
1966 /* Graft it onto L4[511][510] */
1967 copy_page(level2_kernel_pgt, l2);
1969 /* Copy the initial P->M table mappings if necessary. */
1970 i = pgd_index(xen_start_info->mfn_list);
1971 if (i && i < pgd_index(__START_KERNEL_map))
1972 init_level4_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1974 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1975 /* Make pagetable pieces RO */
1976 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1977 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1978 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1979 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1980 set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1981 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1982 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1983 set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
1985 /* Pin down new L4 */
1986 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1987 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1989 /* Unpin Xen-provided one */
1990 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1993 * At this stage there can be no user pgd, and no page
1994 * structure to attach it to, so make sure we just set kernel
1998 __xen_write_cr3(true, __pa(init_level4_pgt));
1999 xen_mc_issue(PARAVIRT_LAZY_CPU);
2001 native_write_cr3(__pa(init_level4_pgt));
2003 /* We can't that easily rip out L3 and L2, as the Xen pagetables are
2004 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
2005 * the initial domain. For guests using the toolstack, they are in:
2006 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
2007 * rip out the [L4] (pgd), but for guests we shave off three pages.
2009 for (i = 0; i < ARRAY_SIZE(addr); i++)
2010 check_pt_base(&pt_base, &pt_end, addr[i]);
2012 /* Our (by three pages) smaller Xen pagetable that we are using */
2013 xen_pt_base = PFN_PHYS(pt_base);
2014 xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
2015 memblock_reserve(xen_pt_base, xen_pt_size);
2017 /* Revector the xen_start_info */
2018 xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
2022 * Read a value from a physical address.
2024 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
2026 unsigned long *vaddr;
2029 vaddr = early_memremap_ro(addr, sizeof(val));
2031 early_memunmap(vaddr, sizeof(val));
2036 * Translate a virtual address to a physical one without relying on mapped
2037 * page tables. Don't rely on big pages being aligned in (guest) physical
2040 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
2049 pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
2051 if (!pgd_present(pgd))
2054 pa = pgd_val(pgd) & PTE_PFN_MASK;
2055 pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
2057 if (!pud_present(pud))
2059 pa = pud_val(pud) & PTE_PFN_MASK;
2061 return pa + (vaddr & ~PUD_MASK);
2063 pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
2065 if (!pmd_present(pmd))
2067 pa = pmd_val(pmd) & PTE_PFN_MASK;
2069 return pa + (vaddr & ~PMD_MASK);
2071 pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
2073 if (!pte_present(pte))
2075 pa = pte_pfn(pte) << PAGE_SHIFT;
2077 return pa | (vaddr & ~PAGE_MASK);
2081 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
2084 void __init xen_relocate_p2m(void)
2086 phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
2087 unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
2088 int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
2093 unsigned long *new_p2m;
2095 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
2096 n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
2097 n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
2098 n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
2099 n_pud = roundup(size, PGDIR_SIZE) >> PGDIR_SHIFT;
2100 n_frames = n_pte + n_pt + n_pmd + n_pud;
2102 new_area = xen_find_free_area(PFN_PHYS(n_frames));
2104 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
2109 * Setup the page tables for addressing the new p2m list.
2110 * We have asked the hypervisor to map the p2m list at the user address
2111 * PUD_SIZE. It may have done so, or it may have used a kernel space
2112 * address depending on the Xen version.
2113 * To avoid any possible virtual address collision, just use
2114 * 2 * PUD_SIZE for the new area.
2116 pud_phys = new_area;
2117 pmd_phys = pud_phys + PFN_PHYS(n_pud);
2118 pt_phys = pmd_phys + PFN_PHYS(n_pmd);
2119 p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
2121 pgd = __va(read_cr3());
2122 new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
2123 for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
2124 pud = early_memremap(pud_phys, PAGE_SIZE);
2126 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
2128 pmd = early_memremap(pmd_phys, PAGE_SIZE);
2130 for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
2132 pt = early_memremap(pt_phys, PAGE_SIZE);
2135 idx_pte < min(n_pte, PTRS_PER_PTE);
2137 set_pte(pt + idx_pte,
2138 pfn_pte(p2m_pfn, PAGE_KERNEL));
2141 n_pte -= PTRS_PER_PTE;
2142 early_memunmap(pt, PAGE_SIZE);
2143 make_lowmem_page_readonly(__va(pt_phys));
2144 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2146 set_pmd(pmd + idx_pt,
2147 __pmd(_PAGE_TABLE | pt_phys));
2148 pt_phys += PAGE_SIZE;
2150 n_pt -= PTRS_PER_PMD;
2151 early_memunmap(pmd, PAGE_SIZE);
2152 make_lowmem_page_readonly(__va(pmd_phys));
2153 pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2154 PFN_DOWN(pmd_phys));
2155 set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
2156 pmd_phys += PAGE_SIZE;
2158 n_pmd -= PTRS_PER_PUD;
2159 early_memunmap(pud, PAGE_SIZE);
2160 make_lowmem_page_readonly(__va(pud_phys));
2161 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2162 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2163 pud_phys += PAGE_SIZE;
2166 /* Now copy the old p2m info to the new area. */
2167 memcpy(new_p2m, xen_p2m_addr, size);
2168 xen_p2m_addr = new_p2m;
2170 /* Release the old p2m list and set new list info. */
2171 p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2173 p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2175 if (xen_start_info->mfn_list < __START_KERNEL_map) {
2176 pfn = xen_start_info->first_p2m_pfn;
2177 pfn_end = xen_start_info->first_p2m_pfn +
2178 xen_start_info->nr_p2m_frames;
2179 set_pgd(pgd + 1, __pgd(0));
2182 pfn_end = p2m_pfn_end;
2185 memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2186 while (pfn < pfn_end) {
2187 if (pfn == p2m_pfn) {
2191 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2195 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2196 xen_start_info->first_p2m_pfn = PFN_DOWN(new_area);
2197 xen_start_info->nr_p2m_frames = n_frames;
2200 #else /* !CONFIG_X86_64 */
2201 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
2202 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
2204 static void __init xen_write_cr3_init(unsigned long cr3)
2206 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
2208 BUG_ON(read_cr3() != __pa(initial_page_table));
2209 BUG_ON(cr3 != __pa(swapper_pg_dir));
2212 * We are switching to swapper_pg_dir for the first time (from
2213 * initial_page_table) and therefore need to mark that page
2214 * read-only and then pin it.
2216 * Xen disallows sharing of kernel PMDs for PAE
2217 * guests. Therefore we must copy the kernel PMD from
2218 * initial_page_table into a new kernel PMD to be used in
2221 swapper_kernel_pmd =
2222 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2223 copy_page(swapper_kernel_pmd, initial_kernel_pmd);
2224 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
2225 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
2226 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
2228 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
2230 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
2232 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
2233 PFN_DOWN(__pa(initial_page_table)));
2234 set_page_prot(initial_page_table, PAGE_KERNEL);
2235 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
2237 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2241 * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
2242 * not the first page table in the page table pool.
2243 * Iterate through the initial page tables to find the real page table base.
2245 static phys_addr_t xen_find_pt_base(pmd_t *pmd)
2247 phys_addr_t pt_base, paddr;
2250 pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
2252 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
2253 if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
2254 paddr = m2p(pmd[pmdidx].pmd);
2255 pt_base = min(pt_base, paddr);
2261 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2265 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2267 xen_pt_base = xen_find_pt_base(kernel_pmd);
2268 xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
2270 initial_kernel_pmd =
2271 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2273 max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
2275 copy_page(initial_kernel_pmd, kernel_pmd);
2277 xen_map_identity_early(initial_kernel_pmd, max_pfn);
2279 copy_page(initial_page_table, pgd);
2280 initial_page_table[KERNEL_PGD_BOUNDARY] =
2281 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2283 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2284 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2285 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2287 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2289 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2290 PFN_DOWN(__pa(initial_page_table)));
2291 xen_write_cr3(__pa(initial_page_table));
2293 memblock_reserve(xen_pt_base, xen_pt_size);
2295 #endif /* CONFIG_X86_64 */
2297 void __init xen_reserve_special_pages(void)
2301 memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2302 if (xen_start_info->store_mfn) {
2303 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2304 memblock_reserve(paddr, PAGE_SIZE);
2306 if (!xen_initial_domain()) {
2307 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2308 memblock_reserve(paddr, PAGE_SIZE);
2312 void __init xen_pt_check_e820(void)
2314 if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2315 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2320 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2322 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2326 phys >>= PAGE_SHIFT;
2329 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2331 #ifdef CONFIG_X86_32
2333 # ifdef CONFIG_HIGHMEM
2334 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2336 #elif defined(CONFIG_X86_VSYSCALL_EMULATION)
2339 case FIX_TEXT_POKE0:
2340 case FIX_TEXT_POKE1:
2341 /* All local page mappings */
2342 pte = pfn_pte(phys, prot);
2345 #ifdef CONFIG_X86_LOCAL_APIC
2346 case FIX_APIC_BASE: /* maps dummy local APIC */
2347 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2351 #ifdef CONFIG_X86_IO_APIC
2352 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2354 * We just don't map the IO APIC - all access is via
2355 * hypercalls. Keep the address in the pte for reference.
2357 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2361 case FIX_PARAVIRT_BOOTMAP:
2362 /* This is an MFN, but it isn't an IO mapping from the
2364 pte = mfn_pte(phys, prot);
2368 /* By default, set_fixmap is used for hardware mappings */
2369 pte = mfn_pte(phys, prot);
2373 __native_set_fixmap(idx, pte);
2375 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2376 /* Replicate changes to map the vsyscall page into the user
2377 pagetable vsyscall mapping. */
2378 if (idx == VSYSCALL_PAGE) {
2379 unsigned long vaddr = __fix_to_virt(idx);
2380 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2385 static void __init xen_post_allocator_init(void)
2387 if (xen_feature(XENFEAT_auto_translated_physmap))
2390 pv_mmu_ops.set_pte = xen_set_pte;
2391 pv_mmu_ops.set_pmd = xen_set_pmd;
2392 pv_mmu_ops.set_pud = xen_set_pud;
2393 #if CONFIG_PGTABLE_LEVELS == 4
2394 pv_mmu_ops.set_pgd = xen_set_pgd;
2397 /* This will work as long as patching hasn't happened yet
2398 (which it hasn't) */
2399 pv_mmu_ops.alloc_pte = xen_alloc_pte;
2400 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2401 pv_mmu_ops.release_pte = xen_release_pte;
2402 pv_mmu_ops.release_pmd = xen_release_pmd;
2403 #if CONFIG_PGTABLE_LEVELS == 4
2404 pv_mmu_ops.alloc_pud = xen_alloc_pud;
2405 pv_mmu_ops.release_pud = xen_release_pud;
2408 #ifdef CONFIG_X86_64
2409 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2410 SetPagePinned(virt_to_page(level3_user_vsyscall));
2412 xen_mark_init_mm_pinned();
2415 static void xen_leave_lazy_mmu(void)
2419 paravirt_leave_lazy_mmu();
2423 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2424 .read_cr2 = xen_read_cr2,
2425 .write_cr2 = xen_write_cr2,
2427 .read_cr3 = xen_read_cr3,
2428 .write_cr3 = xen_write_cr3_init,
2430 .flush_tlb_user = xen_flush_tlb,
2431 .flush_tlb_kernel = xen_flush_tlb,
2432 .flush_tlb_single = xen_flush_tlb_single,
2433 .flush_tlb_others = xen_flush_tlb_others,
2435 .pte_update = paravirt_nop,
2436 .pte_update_defer = paravirt_nop,
2438 .pgd_alloc = xen_pgd_alloc,
2439 .pgd_free = xen_pgd_free,
2441 .alloc_pte = xen_alloc_pte_init,
2442 .release_pte = xen_release_pte_init,
2443 .alloc_pmd = xen_alloc_pmd_init,
2444 .release_pmd = xen_release_pmd_init,
2446 .set_pte = xen_set_pte_init,
2447 .set_pte_at = xen_set_pte_at,
2448 .set_pmd = xen_set_pmd_hyper,
2450 .ptep_modify_prot_start = __ptep_modify_prot_start,
2451 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2453 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2454 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2456 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2457 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2459 #ifdef CONFIG_X86_PAE
2460 .set_pte_atomic = xen_set_pte_atomic,
2461 .pte_clear = xen_pte_clear,
2462 .pmd_clear = xen_pmd_clear,
2463 #endif /* CONFIG_X86_PAE */
2464 .set_pud = xen_set_pud_hyper,
2466 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2467 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2469 #if CONFIG_PGTABLE_LEVELS == 4
2470 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2471 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2472 .set_pgd = xen_set_pgd_hyper,
2474 .alloc_pud = xen_alloc_pmd_init,
2475 .release_pud = xen_release_pmd_init,
2476 #endif /* CONFIG_PGTABLE_LEVELS == 4 */
2478 .activate_mm = xen_activate_mm,
2479 .dup_mmap = xen_dup_mmap,
2480 .exit_mmap = xen_exit_mmap,
2483 .enter = paravirt_enter_lazy_mmu,
2484 .leave = xen_leave_lazy_mmu,
2485 .flush = paravirt_flush_lazy_mmu,
2488 .set_fixmap = xen_set_fixmap,
2491 void __init xen_init_mmu_ops(void)
2493 x86_init.paging.pagetable_init = xen_pagetable_init;
2495 if (xen_feature(XENFEAT_auto_translated_physmap))
2498 pv_mmu_ops = xen_mmu_ops;
2500 memset(dummy_mapping, 0xff, PAGE_SIZE);
2503 /* Protected by xen_reservation_lock. */
2504 #define MAX_CONTIG_ORDER 9 /* 2MB */
2505 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2507 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2508 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2509 unsigned long *in_frames,
2510 unsigned long *out_frames)
2513 struct multicall_space mcs;
2516 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2517 mcs = __xen_mc_entry(0);
2520 in_frames[i] = virt_to_mfn(vaddr);
2522 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2523 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2526 out_frames[i] = virt_to_pfn(vaddr);
2532 * Update the pfn-to-mfn mappings for a virtual address range, either to
2533 * point to an array of mfns, or contiguously from a single starting
2536 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2537 unsigned long *mfns,
2538 unsigned long first_mfn)
2545 limit = 1u << order;
2546 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2547 struct multicall_space mcs;
2550 mcs = __xen_mc_entry(0);
2554 mfn = first_mfn + i;
2556 if (i < (limit - 1))
2560 flags = UVMF_INVLPG | UVMF_ALL;
2562 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2565 MULTI_update_va_mapping(mcs.mc, vaddr,
2566 mfn_pte(mfn, PAGE_KERNEL), flags);
2568 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2575 * Perform the hypercall to exchange a region of our pfns to point to
2576 * memory with the required contiguous alignment. Takes the pfns as
2577 * input, and populates mfns as output.
2579 * Returns a success code indicating whether the hypervisor was able to
2580 * satisfy the request or not.
2582 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2583 unsigned long *pfns_in,
2584 unsigned long extents_out,
2585 unsigned int order_out,
2586 unsigned long *mfns_out,
2587 unsigned int address_bits)
2592 struct xen_memory_exchange exchange = {
2594 .nr_extents = extents_in,
2595 .extent_order = order_in,
2596 .extent_start = pfns_in,
2600 .nr_extents = extents_out,
2601 .extent_order = order_out,
2602 .extent_start = mfns_out,
2603 .address_bits = address_bits,
2608 BUG_ON(extents_in << order_in != extents_out << order_out);
2610 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2611 success = (exchange.nr_exchanged == extents_in);
2613 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2614 BUG_ON(success && (rc != 0));
2619 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2620 unsigned int address_bits,
2621 dma_addr_t *dma_handle)
2623 unsigned long *in_frames = discontig_frames, out_frame;
2624 unsigned long flags;
2626 unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2629 * Currently an auto-translated guest will not perform I/O, nor will
2630 * it require PAE page directories below 4GB. Therefore any calls to
2631 * this function are redundant and can be ignored.
2634 if (xen_feature(XENFEAT_auto_translated_physmap))
2637 if (unlikely(order > MAX_CONTIG_ORDER))
2640 memset((void *) vstart, 0, PAGE_SIZE << order);
2642 spin_lock_irqsave(&xen_reservation_lock, flags);
2644 /* 1. Zap current PTEs, remembering MFNs. */
2645 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2647 /* 2. Get a new contiguous memory extent. */
2648 out_frame = virt_to_pfn(vstart);
2649 success = xen_exchange_memory(1UL << order, 0, in_frames,
2650 1, order, &out_frame,
2653 /* 3. Map the new extent in place of old pages. */
2655 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2657 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2659 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2661 *dma_handle = virt_to_machine(vstart).maddr;
2662 return success ? 0 : -ENOMEM;
2664 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2666 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2668 unsigned long *out_frames = discontig_frames, in_frame;
2669 unsigned long flags;
2671 unsigned long vstart;
2673 if (xen_feature(XENFEAT_auto_translated_physmap))
2676 if (unlikely(order > MAX_CONTIG_ORDER))
2679 vstart = (unsigned long)phys_to_virt(pstart);
2680 memset((void *) vstart, 0, PAGE_SIZE << order);
2682 spin_lock_irqsave(&xen_reservation_lock, flags);
2684 /* 1. Find start MFN of contiguous extent. */
2685 in_frame = virt_to_mfn(vstart);
2687 /* 2. Zap current PTEs. */
2688 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2690 /* 3. Do the exchange for non-contiguous MFNs. */
2691 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2694 /* 4. Map new pages in place of old pages. */
2696 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2698 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2700 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2702 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2704 #ifdef CONFIG_XEN_PVHVM
2705 #ifdef CONFIG_PROC_VMCORE
2707 * This function is used in two contexts:
2708 * - the kdump kernel has to check whether a pfn of the crashed kernel
2709 * was a ballooned page. vmcore is using this function to decide
2710 * whether to access a pfn of the crashed kernel.
2711 * - the kexec kernel has to check whether a pfn was ballooned by the
2712 * previous kernel. If the pfn is ballooned, handle it properly.
2713 * Returns 0 if the pfn is not backed by a RAM page, the caller may
2714 * handle the pfn special in this case.
2716 static int xen_oldmem_pfn_is_ram(unsigned long pfn)
2718 struct xen_hvm_get_mem_type a = {
2719 .domid = DOMID_SELF,
2724 if (HYPERVISOR_hvm_op(HVMOP_get_mem_type, &a))
2727 switch (a.mem_type) {
2728 case HVMMEM_mmio_dm:
2742 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2744 struct xen_hvm_pagetable_dying a;
2747 a.domid = DOMID_SELF;
2748 a.gpa = __pa(mm->pgd);
2749 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2750 WARN_ON_ONCE(rc < 0);
2753 static int is_pagetable_dying_supported(void)
2755 struct xen_hvm_pagetable_dying a;
2758 a.domid = DOMID_SELF;
2760 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2762 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2768 void __init xen_hvm_init_mmu_ops(void)
2770 if (is_pagetable_dying_supported())
2771 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2772 #ifdef CONFIG_PROC_VMCORE
2773 register_oldmem_pfn_is_ram(&xen_oldmem_pfn_is_ram);
2778 #define REMAP_BATCH_SIZE 16
2784 struct mmu_update *mmu_update;
2787 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2788 unsigned long addr, void *data)
2790 struct remap_data *rmd = data;
2791 pte_t pte = pte_mkspecial(mfn_pte(*rmd->mfn, rmd->prot));
2793 /* If we have a contigious range, just update the mfn itself,
2794 else update pointer to be "next mfn". */
2795 if (rmd->contiguous)
2800 rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2801 rmd->mmu_update->val = pte_val_ma(pte);
2807 static int do_remap_gfn(struct vm_area_struct *vma,
2809 xen_pfn_t *gfn, int nr,
2810 int *err_ptr, pgprot_t prot,
2812 struct page **pages)
2815 struct remap_data rmd;
2816 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2817 unsigned long range;
2820 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2822 if (xen_feature(XENFEAT_auto_translated_physmap)) {
2823 #ifdef CONFIG_XEN_PVH
2824 /* We need to update the local page tables and the xen HAP */
2825 return xen_xlate_remap_gfn_array(vma, addr, gfn, nr, err_ptr,
2826 prot, domid, pages);
2834 /* We use the err_ptr to indicate if there we are doing a contigious
2835 * mapping or a discontigious mapping. */
2836 rmd.contiguous = !err_ptr;
2841 int batch = min(REMAP_BATCH_SIZE, nr);
2842 int batch_left = batch;
2843 range = (unsigned long)batch << PAGE_SHIFT;
2845 rmd.mmu_update = mmu_update;
2846 err = apply_to_page_range(vma->vm_mm, addr, range,
2847 remap_area_mfn_pte_fn, &rmd);
2851 /* We record the error for each page that gives an error, but
2852 * continue mapping until the whole set is done */
2856 err = HYPERVISOR_mmu_update(&mmu_update[index],
2857 batch_left, &done, domid);
2860 * @err_ptr may be the same buffer as @gfn, so
2861 * only clear it after each chunk of @gfn is
2865 for (i = index; i < index + done; i++)
2872 done++; /* Skip failed frame. */
2877 } while (batch_left);
2887 xen_flush_tlb_all();
2889 return err < 0 ? err : mapped;
2892 int xen_remap_domain_gfn_range(struct vm_area_struct *vma,
2894 xen_pfn_t gfn, int nr,
2895 pgprot_t prot, unsigned domid,
2896 struct page **pages)
2898 return do_remap_gfn(vma, addr, &gfn, nr, NULL, prot, domid, pages);
2900 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range);
2902 int xen_remap_domain_gfn_array(struct vm_area_struct *vma,
2904 xen_pfn_t *gfn, int nr,
2905 int *err_ptr, pgprot_t prot,
2906 unsigned domid, struct page **pages)
2908 /* We BUG_ON because it's a programmer error to pass a NULL err_ptr,
2909 * and the consequences later is quite hard to detect what the actual
2910 * cause of "wrong memory was mapped in".
2912 BUG_ON(err_ptr == NULL);
2913 return do_remap_gfn(vma, addr, gfn, nr, err_ptr, prot, domid, pages);
2915 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array);
2918 /* Returns: 0 success */
2919 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma,
2920 int numpgs, struct page **pages)
2922 if (!pages || !xen_feature(XENFEAT_auto_translated_physmap))
2925 #ifdef CONFIG_XEN_PVH
2926 return xen_xlate_unmap_gfn_range(vma, numpgs, pages);
2931 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range);