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>
51 #include <trace/events/xen.h>
53 #include <asm/pgtable.h>
54 #include <asm/tlbflush.h>
55 #include <asm/fixmap.h>
56 #include <asm/mmu_context.h>
57 #include <asm/setup.h>
58 #include <asm/paravirt.h>
60 #include <asm/linkage.h>
66 #include <asm/xen/hypercall.h>
67 #include <asm/xen/hypervisor.h>
71 #include <xen/interface/xen.h>
72 #include <xen/interface/hvm/hvm_op.h>
73 #include <xen/interface/version.h>
74 #include <xen/interface/memory.h>
75 #include <xen/hvc-console.h>
77 #include "multicalls.h"
82 * Protects atomic reservation decrease/increase against concurrent increases.
83 * Also protects non-atomic updates of current_pages and balloon lists.
85 DEFINE_SPINLOCK(xen_reservation_lock);
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 */
119 * Just beyond the highest usermode address. STACK_TOP_MAX has a
120 * redzone above it, so round it up to a PGD boundary.
122 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
124 unsigned long arbitrary_virt_to_mfn(void *vaddr)
126 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
128 return PFN_DOWN(maddr.maddr);
131 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
133 unsigned long address = (unsigned long)vaddr;
139 * if the PFN is in the linear mapped vaddr range, we can just use
140 * the (quick) virt_to_machine() p2m lookup
142 if (virt_addr_valid(vaddr))
143 return virt_to_machine(vaddr);
145 /* otherwise we have to do a (slower) full page-table walk */
147 pte = lookup_address(address, &level);
149 offset = address & ~PAGE_MASK;
150 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
152 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
154 void make_lowmem_page_readonly(void *vaddr)
157 unsigned long address = (unsigned long)vaddr;
160 pte = lookup_address(address, &level);
162 return; /* vaddr missing */
164 ptev = pte_wrprotect(*pte);
166 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
170 void make_lowmem_page_readwrite(void *vaddr)
173 unsigned long address = (unsigned long)vaddr;
176 pte = lookup_address(address, &level);
178 return; /* vaddr missing */
180 ptev = pte_mkwrite(*pte);
182 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
187 static bool xen_page_pinned(void *ptr)
189 struct page *page = virt_to_page(ptr);
191 return PagePinned(page);
194 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
196 struct multicall_space mcs;
197 struct mmu_update *u;
199 trace_xen_mmu_set_domain_pte(ptep, pteval, domid);
201 mcs = xen_mc_entry(sizeof(*u));
204 /* ptep might be kmapped when using 32-bit HIGHPTE */
205 u->ptr = virt_to_machine(ptep).maddr;
206 u->val = pte_val_ma(pteval);
208 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
210 xen_mc_issue(PARAVIRT_LAZY_MMU);
212 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
214 static void xen_extend_mmu_update(const struct mmu_update *update)
216 struct multicall_space mcs;
217 struct mmu_update *u;
219 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
221 if (mcs.mc != NULL) {
224 mcs = __xen_mc_entry(sizeof(*u));
225 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
232 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
240 /* ptr may be ioremapped for 64-bit pagetable setup */
241 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
242 u.val = pmd_val_ma(val);
243 xen_extend_mmu_update(&u);
245 xen_mc_issue(PARAVIRT_LAZY_MMU);
250 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
252 trace_xen_mmu_set_pmd(ptr, val);
254 /* If page is not pinned, we can just update the entry
256 if (!xen_page_pinned(ptr)) {
261 xen_set_pmd_hyper(ptr, val);
265 * Associate a virtual page frame with a given physical page frame
266 * and protection flags for that frame.
268 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
270 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
273 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
277 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
282 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
283 u.val = pte_val_ma(pteval);
284 xen_extend_mmu_update(&u);
286 xen_mc_issue(PARAVIRT_LAZY_MMU);
291 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
293 if (!xen_batched_set_pte(ptep, pteval))
294 native_set_pte(ptep, pteval);
297 static void xen_set_pte(pte_t *ptep, pte_t pteval)
299 trace_xen_mmu_set_pte(ptep, pteval);
300 __xen_set_pte(ptep, pteval);
303 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
304 pte_t *ptep, pte_t pteval)
306 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
307 __xen_set_pte(ptep, pteval);
310 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
311 unsigned long addr, pte_t *ptep)
313 /* Just return the pte as-is. We preserve the bits on commit */
314 trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
318 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
319 pte_t *ptep, pte_t pte)
323 trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
326 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
327 u.val = pte_val_ma(pte);
328 xen_extend_mmu_update(&u);
330 xen_mc_issue(PARAVIRT_LAZY_MMU);
333 /* Assume pteval_t is equivalent to all the other *val_t types. */
334 static pteval_t pte_mfn_to_pfn(pteval_t val)
336 if (val & _PAGE_PRESENT) {
337 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
338 pteval_t flags = val & PTE_FLAGS_MASK;
339 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
345 static pteval_t pte_pfn_to_mfn(pteval_t val)
347 if (val & _PAGE_PRESENT) {
348 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
349 pteval_t flags = val & PTE_FLAGS_MASK;
352 if (!xen_feature(XENFEAT_auto_translated_physmap))
353 mfn = get_phys_to_machine(pfn);
357 * If there's no mfn for the pfn, then just create an
358 * empty non-present pte. Unfortunately this loses
359 * information about the original pfn, so
360 * pte_mfn_to_pfn is asymmetric.
362 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
367 * Paramount to do this test _after_ the
368 * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
369 * IDENTITY_FRAME_BIT resolves to true.
371 mfn &= ~FOREIGN_FRAME_BIT;
372 if (mfn & IDENTITY_FRAME_BIT) {
373 mfn &= ~IDENTITY_FRAME_BIT;
374 flags |= _PAGE_IOMAP;
377 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
383 static pteval_t iomap_pte(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;
389 /* We assume the pte frame number is a MFN, so
390 just use it as-is. */
391 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
397 static pteval_t xen_pte_val(pte_t pte)
399 pteval_t pteval = pte.pte;
401 /* If this is a WC pte, convert back from Xen WC to Linux WC */
402 if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
403 WARN_ON(!pat_enabled);
404 pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
407 if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
410 return pte_mfn_to_pfn(pteval);
412 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
414 static pgdval_t xen_pgd_val(pgd_t pgd)
416 return pte_mfn_to_pfn(pgd.pgd);
418 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
421 * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
422 * are reserved for now, to correspond to the Intel-reserved PAT
425 * We expect Linux's PAT set as follows:
427 * Idx PTE flags Linux Xen Default
434 * 6 PAT PCD UC- UC UC-
435 * 7 PAT PCD PWT UC UC UC
438 void xen_set_pat(u64 pat)
440 /* We expect Linux to use a PAT setting of
441 * UC UC- WC WB (ignoring the PAT flag) */
442 WARN_ON(pat != 0x0007010600070106ull);
445 static pte_t xen_make_pte(pteval_t pte)
447 phys_addr_t addr = (pte & PTE_PFN_MASK);
449 /* If Linux is trying to set a WC pte, then map to the Xen WC.
450 * If _PAGE_PAT is set, then it probably means it is really
451 * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
452 * things work out OK...
454 * (We should never see kernel mappings with _PAGE_PSE set,
455 * but we could see hugetlbfs mappings, I think.).
457 if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
458 if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
459 pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
463 * Unprivileged domains are allowed to do IOMAPpings for
464 * PCI passthrough, but not map ISA space. The ISA
465 * mappings are just dummy local mappings to keep other
466 * parts of the kernel happy.
468 if (unlikely(pte & _PAGE_IOMAP) &&
469 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
470 pte = iomap_pte(pte);
473 pte = pte_pfn_to_mfn(pte);
476 return native_make_pte(pte);
478 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
480 #ifdef CONFIG_XEN_DEBUG
481 pte_t xen_make_pte_debug(pteval_t pte)
483 phys_addr_t addr = (pte & PTE_PFN_MASK);
484 phys_addr_t other_addr;
485 bool io_page = false;
488 if (pte & _PAGE_IOMAP)
491 _pte = xen_make_pte(pte);
497 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
498 other_addr = pfn_to_mfn(addr >> PAGE_SHIFT) << PAGE_SHIFT;
499 WARN_ONCE(addr != other_addr,
500 "0x%lx is using VM_IO, but it is 0x%lx!\n",
501 (unsigned long)addr, (unsigned long)other_addr);
503 pteval_t iomap_set = (_pte.pte & PTE_FLAGS_MASK) & _PAGE_IOMAP;
504 other_addr = (_pte.pte & PTE_PFN_MASK);
505 WARN_ONCE((addr == other_addr) && (!io_page) && (!iomap_set),
506 "0x%lx is missing VM_IO (and wasn't fixed)!\n",
507 (unsigned long)addr);
512 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_debug);
515 static pgd_t xen_make_pgd(pgdval_t pgd)
517 pgd = pte_pfn_to_mfn(pgd);
518 return native_make_pgd(pgd);
520 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
522 static pmdval_t xen_pmd_val(pmd_t pmd)
524 return pte_mfn_to_pfn(pmd.pmd);
526 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
528 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
536 /* ptr may be ioremapped for 64-bit pagetable setup */
537 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
538 u.val = pud_val_ma(val);
539 xen_extend_mmu_update(&u);
541 xen_mc_issue(PARAVIRT_LAZY_MMU);
546 static void xen_set_pud(pud_t *ptr, pud_t val)
548 trace_xen_mmu_set_pud(ptr, val);
550 /* If page is not pinned, we can just update the entry
552 if (!xen_page_pinned(ptr)) {
557 xen_set_pud_hyper(ptr, val);
560 #ifdef CONFIG_X86_PAE
561 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
563 trace_xen_mmu_set_pte_atomic(ptep, pte);
564 set_64bit((u64 *)ptep, native_pte_val(pte));
567 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
569 trace_xen_mmu_pte_clear(mm, addr, ptep);
570 if (!xen_batched_set_pte(ptep, native_make_pte(0)))
571 native_pte_clear(mm, addr, ptep);
574 static void xen_pmd_clear(pmd_t *pmdp)
576 trace_xen_mmu_pmd_clear(pmdp);
577 set_pmd(pmdp, __pmd(0));
579 #endif /* CONFIG_X86_PAE */
581 static pmd_t xen_make_pmd(pmdval_t pmd)
583 pmd = pte_pfn_to_mfn(pmd);
584 return native_make_pmd(pmd);
586 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
588 #if PAGETABLE_LEVELS == 4
589 static pudval_t xen_pud_val(pud_t pud)
591 return pte_mfn_to_pfn(pud.pud);
593 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
595 static pud_t xen_make_pud(pudval_t pud)
597 pud = pte_pfn_to_mfn(pud);
599 return native_make_pud(pud);
601 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
603 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
605 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
606 unsigned offset = pgd - pgd_page;
607 pgd_t *user_ptr = NULL;
609 if (offset < pgd_index(USER_LIMIT)) {
610 struct page *page = virt_to_page(pgd_page);
611 user_ptr = (pgd_t *)page->private;
619 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
623 u.ptr = virt_to_machine(ptr).maddr;
624 u.val = pgd_val_ma(val);
625 xen_extend_mmu_update(&u);
629 * Raw hypercall-based set_pgd, intended for in early boot before
630 * there's a page structure. This implies:
631 * 1. The only existing pagetable is the kernel's
632 * 2. It is always pinned
633 * 3. It has no user pagetable attached to it
635 static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
641 __xen_set_pgd_hyper(ptr, val);
643 xen_mc_issue(PARAVIRT_LAZY_MMU);
648 static void xen_set_pgd(pgd_t *ptr, pgd_t val)
650 pgd_t *user_ptr = xen_get_user_pgd(ptr);
652 trace_xen_mmu_set_pgd(ptr, user_ptr, val);
654 /* If page is not pinned, we can just update the entry
656 if (!xen_page_pinned(ptr)) {
659 WARN_ON(xen_page_pinned(user_ptr));
665 /* If it's pinned, then we can at least batch the kernel and
666 user updates together. */
669 __xen_set_pgd_hyper(ptr, val);
671 __xen_set_pgd_hyper(user_ptr, val);
673 xen_mc_issue(PARAVIRT_LAZY_MMU);
675 #endif /* PAGETABLE_LEVELS == 4 */
678 * (Yet another) pagetable walker. This one is intended for pinning a
679 * pagetable. This means that it walks a pagetable and calls the
680 * callback function on each page it finds making up the page table,
681 * at every level. It walks the entire pagetable, but it only bothers
682 * pinning pte pages which are below limit. In the normal case this
683 * will be STACK_TOP_MAX, but at boot we need to pin up to
686 * For 32-bit the important bit is that we don't pin beyond there,
687 * because then we start getting into Xen's ptes.
689 * For 64-bit, we must skip the Xen hole in the middle of the address
690 * space, just after the big x86-64 virtual hole.
692 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
693 int (*func)(struct mm_struct *mm, struct page *,
698 unsigned hole_low, hole_high;
699 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
700 unsigned pgdidx, pudidx, pmdidx;
702 /* The limit is the last byte to be touched */
704 BUG_ON(limit >= FIXADDR_TOP);
706 if (xen_feature(XENFEAT_auto_translated_physmap))
710 * 64-bit has a great big hole in the middle of the address
711 * space, which contains the Xen mappings. On 32-bit these
712 * will end up making a zero-sized hole and so is a no-op.
714 hole_low = pgd_index(USER_LIMIT);
715 hole_high = pgd_index(PAGE_OFFSET);
717 pgdidx_limit = pgd_index(limit);
719 pudidx_limit = pud_index(limit);
724 pmdidx_limit = pmd_index(limit);
729 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
732 if (pgdidx >= hole_low && pgdidx < hole_high)
735 if (!pgd_val(pgd[pgdidx]))
738 pud = pud_offset(&pgd[pgdidx], 0);
740 if (PTRS_PER_PUD > 1) /* not folded */
741 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
743 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
746 if (pgdidx == pgdidx_limit &&
747 pudidx > pudidx_limit)
750 if (pud_none(pud[pudidx]))
753 pmd = pmd_offset(&pud[pudidx], 0);
755 if (PTRS_PER_PMD > 1) /* not folded */
756 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
758 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
761 if (pgdidx == pgdidx_limit &&
762 pudidx == pudidx_limit &&
763 pmdidx > pmdidx_limit)
766 if (pmd_none(pmd[pmdidx]))
769 pte = pmd_page(pmd[pmdidx]);
770 flush |= (*func)(mm, pte, PT_PTE);
776 /* Do the top level last, so that the callbacks can use it as
777 a cue to do final things like tlb flushes. */
778 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
783 static int xen_pgd_walk(struct mm_struct *mm,
784 int (*func)(struct mm_struct *mm, struct page *,
788 return __xen_pgd_walk(mm, mm->pgd, func, limit);
791 /* If we're using split pte locks, then take the page's lock and
792 return a pointer to it. Otherwise return NULL. */
793 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
795 spinlock_t *ptl = NULL;
797 #if USE_SPLIT_PTLOCKS
798 ptl = __pte_lockptr(page);
799 spin_lock_nest_lock(ptl, &mm->page_table_lock);
805 static void xen_pte_unlock(void *v)
811 static void xen_do_pin(unsigned level, unsigned long pfn)
813 struct mmuext_op *op;
814 struct multicall_space mcs;
816 mcs = __xen_mc_entry(sizeof(*op));
819 op->arg1.mfn = pfn_to_mfn(pfn);
820 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
823 static int xen_pin_page(struct mm_struct *mm, struct page *page,
826 unsigned pgfl = TestSetPagePinned(page);
830 flush = 0; /* already pinned */
831 else if (PageHighMem(page))
832 /* kmaps need flushing if we found an unpinned
836 void *pt = lowmem_page_address(page);
837 unsigned long pfn = page_to_pfn(page);
838 struct multicall_space mcs = __xen_mc_entry(0);
844 * We need to hold the pagetable lock between the time
845 * we make the pagetable RO and when we actually pin
846 * it. If we don't, then other users may come in and
847 * attempt to update the pagetable by writing it,
848 * which will fail because the memory is RO but not
849 * pinned, so Xen won't do the trap'n'emulate.
851 * If we're using split pte locks, we can't hold the
852 * entire pagetable's worth of locks during the
853 * traverse, because we may wrap the preempt count (8
854 * bits). The solution is to mark RO and pin each PTE
855 * page while holding the lock. This means the number
856 * of locks we end up holding is never more than a
857 * batch size (~32 entries, at present).
859 * If we're not using split pte locks, we needn't pin
860 * the PTE pages independently, because we're
861 * protected by the overall pagetable lock.
865 ptl = xen_pte_lock(page, mm);
867 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
868 pfn_pte(pfn, PAGE_KERNEL_RO),
869 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
872 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
874 /* Queue a deferred unlock for when this batch
876 xen_mc_callback(xen_pte_unlock, ptl);
883 /* This is called just after a mm has been created, but it has not
884 been used yet. We need to make sure that its pagetable is all
885 read-only, and can be pinned. */
886 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
890 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
891 /* re-enable interrupts for flushing */
901 pgd_t *user_pgd = xen_get_user_pgd(pgd);
903 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
906 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
907 xen_do_pin(MMUEXT_PIN_L4_TABLE,
908 PFN_DOWN(__pa(user_pgd)));
911 #else /* CONFIG_X86_32 */
912 #ifdef CONFIG_X86_PAE
913 /* Need to make sure unshared kernel PMD is pinnable */
914 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
917 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
918 #endif /* CONFIG_X86_64 */
922 static void xen_pgd_pin(struct mm_struct *mm)
924 __xen_pgd_pin(mm, mm->pgd);
928 * On save, we need to pin all pagetables to make sure they get their
929 * mfns turned into pfns. Search the list for any unpinned pgds and pin
930 * them (unpinned pgds are not currently in use, probably because the
931 * process is under construction or destruction).
933 * Expected to be called in stop_machine() ("equivalent to taking
934 * every spinlock in the system"), so the locking doesn't really
935 * matter all that much.
937 void xen_mm_pin_all(void)
941 spin_lock(&pgd_lock);
943 list_for_each_entry(page, &pgd_list, lru) {
944 if (!PagePinned(page)) {
945 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
946 SetPageSavePinned(page);
950 spin_unlock(&pgd_lock);
954 * The init_mm pagetable is really pinned as soon as its created, but
955 * that's before we have page structures to store the bits. So do all
956 * the book-keeping now.
958 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
965 static void __init xen_mark_init_mm_pinned(void)
967 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
970 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
973 unsigned pgfl = TestClearPagePinned(page);
975 if (pgfl && !PageHighMem(page)) {
976 void *pt = lowmem_page_address(page);
977 unsigned long pfn = page_to_pfn(page);
978 spinlock_t *ptl = NULL;
979 struct multicall_space mcs;
982 * Do the converse to pin_page. If we're using split
983 * pte locks, we must be holding the lock for while
984 * the pte page is unpinned but still RO to prevent
985 * concurrent updates from seeing it in this
986 * partially-pinned state.
988 if (level == PT_PTE) {
989 ptl = xen_pte_lock(page, mm);
992 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
995 mcs = __xen_mc_entry(0);
997 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
998 pfn_pte(pfn, PAGE_KERNEL),
999 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1002 /* unlock when batch completed */
1003 xen_mc_callback(xen_pte_unlock, ptl);
1007 return 0; /* never need to flush on unpin */
1010 /* Release a pagetables pages back as normal RW */
1011 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1015 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1017 #ifdef CONFIG_X86_64
1019 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1022 xen_do_pin(MMUEXT_UNPIN_TABLE,
1023 PFN_DOWN(__pa(user_pgd)));
1024 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1029 #ifdef CONFIG_X86_PAE
1030 /* Need to make sure unshared kernel PMD is unpinned */
1031 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1035 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1040 static void xen_pgd_unpin(struct mm_struct *mm)
1042 __xen_pgd_unpin(mm, mm->pgd);
1046 * On resume, undo any pinning done at save, so that the rest of the
1047 * kernel doesn't see any unexpected pinned pagetables.
1049 void xen_mm_unpin_all(void)
1053 spin_lock(&pgd_lock);
1055 list_for_each_entry(page, &pgd_list, lru) {
1056 if (PageSavePinned(page)) {
1057 BUG_ON(!PagePinned(page));
1058 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1059 ClearPageSavePinned(page);
1063 spin_unlock(&pgd_lock);
1066 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1068 spin_lock(&next->page_table_lock);
1070 spin_unlock(&next->page_table_lock);
1073 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1075 spin_lock(&mm->page_table_lock);
1077 spin_unlock(&mm->page_table_lock);
1082 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1083 we need to repoint it somewhere else before we can unpin it. */
1084 static void drop_other_mm_ref(void *info)
1086 struct mm_struct *mm = info;
1087 struct mm_struct *active_mm;
1089 active_mm = percpu_read(cpu_tlbstate.active_mm);
1091 if (active_mm == mm && percpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
1092 leave_mm(smp_processor_id());
1094 /* If this cpu still has a stale cr3 reference, then make sure
1095 it has been flushed. */
1096 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1097 load_cr3(swapper_pg_dir);
1100 static void xen_drop_mm_ref(struct mm_struct *mm)
1105 if (current->active_mm == mm) {
1106 if (current->mm == mm)
1107 load_cr3(swapper_pg_dir);
1109 leave_mm(smp_processor_id());
1112 /* Get the "official" set of cpus referring to our pagetable. */
1113 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1114 for_each_online_cpu(cpu) {
1115 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1116 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1118 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1122 cpumask_copy(mask, mm_cpumask(mm));
1124 /* It's possible that a vcpu may have a stale reference to our
1125 cr3, because its in lazy mode, and it hasn't yet flushed
1126 its set of pending hypercalls yet. In this case, we can
1127 look at its actual current cr3 value, and force it to flush
1129 for_each_online_cpu(cpu) {
1130 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1131 cpumask_set_cpu(cpu, mask);
1134 if (!cpumask_empty(mask))
1135 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1136 free_cpumask_var(mask);
1139 static void xen_drop_mm_ref(struct mm_struct *mm)
1141 if (current->active_mm == mm)
1142 load_cr3(swapper_pg_dir);
1147 * While a process runs, Xen pins its pagetables, which means that the
1148 * hypervisor forces it to be read-only, and it controls all updates
1149 * to it. This means that all pagetable updates have to go via the
1150 * hypervisor, which is moderately expensive.
1152 * Since we're pulling the pagetable down, we switch to use init_mm,
1153 * unpin old process pagetable and mark it all read-write, which
1154 * allows further operations on it to be simple memory accesses.
1156 * The only subtle point is that another CPU may be still using the
1157 * pagetable because of lazy tlb flushing. This means we need need to
1158 * switch all CPUs off this pagetable before we can unpin it.
1160 static void xen_exit_mmap(struct mm_struct *mm)
1162 get_cpu(); /* make sure we don't move around */
1163 xen_drop_mm_ref(mm);
1166 spin_lock(&mm->page_table_lock);
1168 /* pgd may not be pinned in the error exit path of execve */
1169 if (xen_page_pinned(mm->pgd))
1172 spin_unlock(&mm->page_table_lock);
1175 static void __init xen_pagetable_setup_start(pgd_t *base)
1179 static __init void xen_mapping_pagetable_reserve(u64 start, u64 end)
1181 /* reserve the range used */
1182 native_pagetable_reserve(start, end);
1184 /* set as RW the rest */
1185 printk(KERN_DEBUG "xen: setting RW the range %llx - %llx\n", end,
1186 PFN_PHYS(pgt_buf_top));
1187 while (end < PFN_PHYS(pgt_buf_top)) {
1188 make_lowmem_page_readwrite(__va(end));
1193 static void xen_post_allocator_init(void);
1195 static void __init xen_pagetable_setup_done(pgd_t *base)
1197 xen_setup_shared_info();
1198 xen_post_allocator_init();
1201 static void xen_write_cr2(unsigned long cr2)
1203 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1206 static unsigned long xen_read_cr2(void)
1208 return percpu_read(xen_vcpu)->arch.cr2;
1211 unsigned long xen_read_cr2_direct(void)
1213 return percpu_read(xen_vcpu_info.arch.cr2);
1216 static void xen_flush_tlb(void)
1218 struct mmuext_op *op;
1219 struct multicall_space mcs;
1223 mcs = xen_mc_entry(sizeof(*op));
1226 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1227 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1229 xen_mc_issue(PARAVIRT_LAZY_MMU);
1234 static void xen_flush_tlb_single(unsigned long addr)
1236 struct mmuext_op *op;
1237 struct multicall_space mcs;
1241 mcs = xen_mc_entry(sizeof(*op));
1243 op->cmd = MMUEXT_INVLPG_LOCAL;
1244 op->arg1.linear_addr = addr & PAGE_MASK;
1245 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1247 xen_mc_issue(PARAVIRT_LAZY_MMU);
1252 static void xen_flush_tlb_others(const struct cpumask *cpus,
1253 struct mm_struct *mm, unsigned long va)
1256 struct mmuext_op op;
1258 DECLARE_BITMAP(mask, num_processors);
1260 DECLARE_BITMAP(mask, NR_CPUS);
1263 struct multicall_space mcs;
1265 if (cpumask_empty(cpus))
1266 return; /* nothing to do */
1268 mcs = xen_mc_entry(sizeof(*args));
1270 args->op.arg2.vcpumask = to_cpumask(args->mask);
1272 /* Remove us, and any offline CPUS. */
1273 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1274 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1276 if (va == TLB_FLUSH_ALL) {
1277 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1279 args->op.cmd = MMUEXT_INVLPG_MULTI;
1280 args->op.arg1.linear_addr = va;
1283 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1285 xen_mc_issue(PARAVIRT_LAZY_MMU);
1288 static unsigned long xen_read_cr3(void)
1290 return percpu_read(xen_cr3);
1293 static void set_current_cr3(void *v)
1295 percpu_write(xen_current_cr3, (unsigned long)v);
1298 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1300 struct mmuext_op *op;
1301 struct multicall_space mcs;
1305 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1309 WARN_ON(mfn == 0 && kernel);
1311 mcs = __xen_mc_entry(sizeof(*op));
1314 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1317 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1320 percpu_write(xen_cr3, cr3);
1322 /* Update xen_current_cr3 once the batch has actually
1324 xen_mc_callback(set_current_cr3, (void *)cr3);
1328 static void xen_write_cr3(unsigned long cr3)
1330 BUG_ON(preemptible());
1332 xen_mc_batch(); /* disables interrupts */
1334 /* Update while interrupts are disabled, so its atomic with
1336 percpu_write(xen_cr3, cr3);
1338 __xen_write_cr3(true, cr3);
1340 #ifdef CONFIG_X86_64
1342 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1344 __xen_write_cr3(false, __pa(user_pgd));
1346 __xen_write_cr3(false, 0);
1350 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1353 static int xen_pgd_alloc(struct mm_struct *mm)
1355 pgd_t *pgd = mm->pgd;
1358 BUG_ON(PagePinned(virt_to_page(pgd)));
1360 #ifdef CONFIG_X86_64
1362 struct page *page = virt_to_page(pgd);
1365 BUG_ON(page->private != 0);
1369 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1370 page->private = (unsigned long)user_pgd;
1372 if (user_pgd != NULL) {
1373 user_pgd[pgd_index(VSYSCALL_START)] =
1374 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1378 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1385 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1387 #ifdef CONFIG_X86_64
1388 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1391 free_page((unsigned long)user_pgd);
1395 #ifdef CONFIG_X86_32
1396 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1398 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1399 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1400 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1405 #else /* CONFIG_X86_64 */
1406 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1408 unsigned long pfn = pte_pfn(pte);
1411 * If the new pfn is within the range of the newly allocated
1412 * kernel pagetable, and it isn't being mapped into an
1413 * early_ioremap fixmap slot as a freshly allocated page, make sure
1416 if (((!is_early_ioremap_ptep(ptep) &&
1417 pfn >= pgt_buf_start && pfn < pgt_buf_top)) ||
1418 (is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1419 pte = pte_wrprotect(pte);
1423 #endif /* CONFIG_X86_64 */
1425 /* Init-time set_pte while constructing initial pagetables, which
1426 doesn't allow RO pagetable pages to be remapped RW */
1427 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1429 pte = mask_rw_pte(ptep, pte);
1431 xen_set_pte(ptep, pte);
1434 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1436 struct mmuext_op op;
1438 op.arg1.mfn = pfn_to_mfn(pfn);
1439 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1443 /* Early in boot, while setting up the initial pagetable, assume
1444 everything is pinned. */
1445 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1447 #ifdef CONFIG_FLATMEM
1448 BUG_ON(mem_map); /* should only be used early */
1450 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1451 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1454 /* Used for pmd and pud */
1455 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1457 #ifdef CONFIG_FLATMEM
1458 BUG_ON(mem_map); /* should only be used early */
1460 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1463 /* Early release_pte assumes that all pts are pinned, since there's
1464 only init_mm and anything attached to that is pinned. */
1465 static void __init xen_release_pte_init(unsigned long pfn)
1467 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1468 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1471 static void __init xen_release_pmd_init(unsigned long pfn)
1473 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1476 /* This needs to make sure the new pte page is pinned iff its being
1477 attached to a pinned pagetable. */
1478 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1480 struct page *page = pfn_to_page(pfn);
1481 int pinned = PagePinned(virt_to_page(mm->pgd));
1483 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1486 SetPagePinned(page);
1488 if (!PageHighMem(page)) {
1489 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1490 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1491 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1493 /* make sure there are no stray mappings of
1495 kmap_flush_unused();
1500 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1502 xen_alloc_ptpage(mm, pfn, PT_PTE);
1505 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1507 xen_alloc_ptpage(mm, pfn, PT_PMD);
1510 /* This should never happen until we're OK to use struct page */
1511 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1513 struct page *page = pfn_to_page(pfn);
1514 bool pinned = PagePinned(page);
1516 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1519 if (!PageHighMem(page)) {
1520 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1521 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1522 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1524 ClearPagePinned(page);
1528 static void xen_release_pte(unsigned long pfn)
1530 xen_release_ptpage(pfn, PT_PTE);
1533 static void xen_release_pmd(unsigned long pfn)
1535 xen_release_ptpage(pfn, PT_PMD);
1538 #if PAGETABLE_LEVELS == 4
1539 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1541 xen_alloc_ptpage(mm, pfn, PT_PUD);
1544 static void xen_release_pud(unsigned long pfn)
1546 xen_release_ptpage(pfn, PT_PUD);
1550 void __init xen_reserve_top(void)
1552 #ifdef CONFIG_X86_32
1553 unsigned long top = HYPERVISOR_VIRT_START;
1554 struct xen_platform_parameters pp;
1556 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1557 top = pp.virt_start;
1559 reserve_top_address(-top);
1560 #endif /* CONFIG_X86_32 */
1564 * Like __va(), but returns address in the kernel mapping (which is
1565 * all we have until the physical memory mapping has been set up.
1567 static void *__ka(phys_addr_t paddr)
1569 #ifdef CONFIG_X86_64
1570 return (void *)(paddr + __START_KERNEL_map);
1576 /* Convert a machine address to physical address */
1577 static unsigned long m2p(phys_addr_t maddr)
1581 maddr &= PTE_PFN_MASK;
1582 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1587 /* Convert a machine address to kernel virtual */
1588 static void *m2v(phys_addr_t maddr)
1590 return __ka(m2p(maddr));
1593 /* Set the page permissions on an identity-mapped pages */
1594 static void set_page_prot(void *addr, pgprot_t prot)
1596 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1597 pte_t pte = pfn_pte(pfn, prot);
1599 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1603 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1605 unsigned pmdidx, pteidx;
1609 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1614 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1617 /* Reuse or allocate a page of ptes */
1618 if (pmd_present(pmd[pmdidx]))
1619 pte_page = m2v(pmd[pmdidx].pmd);
1621 /* Check for free pte pages */
1622 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1625 pte_page = &level1_ident_pgt[ident_pte];
1626 ident_pte += PTRS_PER_PTE;
1628 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1631 /* Install mappings */
1632 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1635 #ifdef CONFIG_X86_32
1636 if (pfn > max_pfn_mapped)
1637 max_pfn_mapped = pfn;
1640 if (!pte_none(pte_page[pteidx]))
1643 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1644 pte_page[pteidx] = pte;
1648 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1649 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1651 set_page_prot(pmd, PAGE_KERNEL_RO);
1654 void __init xen_setup_machphys_mapping(void)
1656 struct xen_machphys_mapping mapping;
1657 unsigned long machine_to_phys_nr_ents;
1659 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1660 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1661 machine_to_phys_nr_ents = mapping.max_mfn + 1;
1663 machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
1665 machine_to_phys_order = fls(machine_to_phys_nr_ents - 1);
1668 #ifdef CONFIG_X86_64
1669 static void convert_pfn_mfn(void *v)
1674 /* All levels are converted the same way, so just treat them
1676 for (i = 0; i < PTRS_PER_PTE; i++)
1677 pte[i] = xen_make_pte(pte[i].pte);
1681 * Set up the initial kernel pagetable.
1683 * We can construct this by grafting the Xen provided pagetable into
1684 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1685 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1686 * means that only the kernel has a physical mapping to start with -
1687 * but that's enough to get __va working. We need to fill in the rest
1688 * of the physical mapping once some sort of allocator has been set
1691 pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1692 unsigned long max_pfn)
1697 /* max_pfn_mapped is the last pfn mapped in the initial memory
1698 * mappings. Considering that on Xen after the kernel mappings we
1699 * have the mappings of some pages that don't exist in pfn space, we
1700 * set max_pfn_mapped to the last real pfn mapped. */
1701 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1703 /* Zap identity mapping */
1704 init_level4_pgt[0] = __pgd(0);
1706 /* Pre-constructed entries are in pfn, so convert to mfn */
1707 convert_pfn_mfn(init_level4_pgt);
1708 convert_pfn_mfn(level3_ident_pgt);
1709 convert_pfn_mfn(level3_kernel_pgt);
1711 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1712 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1714 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1715 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1717 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1718 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1719 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1721 /* Set up identity map */
1722 xen_map_identity_early(level2_ident_pgt, max_pfn);
1724 /* Make pagetable pieces RO */
1725 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1726 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1727 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1728 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1729 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1730 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1732 /* Pin down new L4 */
1733 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1734 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1736 /* Unpin Xen-provided one */
1737 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1740 pgd = init_level4_pgt;
1743 * At this stage there can be no user pgd, and no page
1744 * structure to attach it to, so make sure we just set kernel
1748 __xen_write_cr3(true, __pa(pgd));
1749 xen_mc_issue(PARAVIRT_LAZY_CPU);
1751 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1752 __pa(xen_start_info->pt_base +
1753 xen_start_info->nr_pt_frames * PAGE_SIZE),
1758 #else /* !CONFIG_X86_64 */
1759 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1760 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1762 static void __init xen_write_cr3_init(unsigned long cr3)
1764 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1766 BUG_ON(read_cr3() != __pa(initial_page_table));
1767 BUG_ON(cr3 != __pa(swapper_pg_dir));
1770 * We are switching to swapper_pg_dir for the first time (from
1771 * initial_page_table) and therefore need to mark that page
1772 * read-only and then pin it.
1774 * Xen disallows sharing of kernel PMDs for PAE
1775 * guests. Therefore we must copy the kernel PMD from
1776 * initial_page_table into a new kernel PMD to be used in
1779 swapper_kernel_pmd =
1780 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1781 memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1782 sizeof(pmd_t) * PTRS_PER_PMD);
1783 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1784 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1785 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1787 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1789 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1791 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1792 PFN_DOWN(__pa(initial_page_table)));
1793 set_page_prot(initial_page_table, PAGE_KERNEL);
1794 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1796 pv_mmu_ops.write_cr3 = &xen_write_cr3;
1799 pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1800 unsigned long max_pfn)
1804 initial_kernel_pmd =
1805 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1807 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1808 xen_start_info->nr_pt_frames * PAGE_SIZE +
1811 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1812 memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1814 xen_map_identity_early(initial_kernel_pmd, max_pfn);
1816 memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1817 initial_page_table[KERNEL_PGD_BOUNDARY] =
1818 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1820 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1821 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1822 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1824 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1826 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1827 PFN_DOWN(__pa(initial_page_table)));
1828 xen_write_cr3(__pa(initial_page_table));
1830 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1831 __pa(xen_start_info->pt_base +
1832 xen_start_info->nr_pt_frames * PAGE_SIZE),
1835 return initial_page_table;
1837 #endif /* CONFIG_X86_64 */
1839 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1841 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1845 phys >>= PAGE_SHIFT;
1848 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1849 #ifdef CONFIG_X86_F00F_BUG
1852 #ifdef CONFIG_X86_32
1855 # ifdef CONFIG_HIGHMEM
1856 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1859 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1861 case FIX_TEXT_POKE0:
1862 case FIX_TEXT_POKE1:
1863 /* All local page mappings */
1864 pte = pfn_pte(phys, prot);
1867 #ifdef CONFIG_X86_LOCAL_APIC
1868 case FIX_APIC_BASE: /* maps dummy local APIC */
1869 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1873 #ifdef CONFIG_X86_IO_APIC
1874 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1876 * We just don't map the IO APIC - all access is via
1877 * hypercalls. Keep the address in the pte for reference.
1879 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1883 case FIX_PARAVIRT_BOOTMAP:
1884 /* This is an MFN, but it isn't an IO mapping from the
1886 pte = mfn_pte(phys, prot);
1890 /* By default, set_fixmap is used for hardware mappings */
1891 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1895 __native_set_fixmap(idx, pte);
1897 #ifdef CONFIG_X86_64
1898 /* Replicate changes to map the vsyscall page into the user
1899 pagetable vsyscall mapping. */
1900 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1901 unsigned long vaddr = __fix_to_virt(idx);
1902 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1907 void __init xen_ident_map_ISA(void)
1912 * If we're dom0, then linear map the ISA machine addresses into
1913 * the kernel's address space.
1915 if (!xen_initial_domain())
1918 xen_raw_printk("Xen: setup ISA identity maps\n");
1920 for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1921 pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1923 if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1930 static void __init xen_post_allocator_init(void)
1932 #ifdef CONFIG_XEN_DEBUG
1933 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte_debug);
1935 pv_mmu_ops.set_pte = xen_set_pte;
1936 pv_mmu_ops.set_pmd = xen_set_pmd;
1937 pv_mmu_ops.set_pud = xen_set_pud;
1938 #if PAGETABLE_LEVELS == 4
1939 pv_mmu_ops.set_pgd = xen_set_pgd;
1942 /* This will work as long as patching hasn't happened yet
1943 (which it hasn't) */
1944 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1945 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1946 pv_mmu_ops.release_pte = xen_release_pte;
1947 pv_mmu_ops.release_pmd = xen_release_pmd;
1948 #if PAGETABLE_LEVELS == 4
1949 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1950 pv_mmu_ops.release_pud = xen_release_pud;
1953 #ifdef CONFIG_X86_64
1954 SetPagePinned(virt_to_page(level3_user_vsyscall));
1956 xen_mark_init_mm_pinned();
1959 static void xen_leave_lazy_mmu(void)
1963 paravirt_leave_lazy_mmu();
1967 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
1968 .read_cr2 = xen_read_cr2,
1969 .write_cr2 = xen_write_cr2,
1971 .read_cr3 = xen_read_cr3,
1972 #ifdef CONFIG_X86_32
1973 .write_cr3 = xen_write_cr3_init,
1975 .write_cr3 = xen_write_cr3,
1978 .flush_tlb_user = xen_flush_tlb,
1979 .flush_tlb_kernel = xen_flush_tlb,
1980 .flush_tlb_single = xen_flush_tlb_single,
1981 .flush_tlb_others = xen_flush_tlb_others,
1983 .pte_update = paravirt_nop,
1984 .pte_update_defer = paravirt_nop,
1986 .pgd_alloc = xen_pgd_alloc,
1987 .pgd_free = xen_pgd_free,
1989 .alloc_pte = xen_alloc_pte_init,
1990 .release_pte = xen_release_pte_init,
1991 .alloc_pmd = xen_alloc_pmd_init,
1992 .release_pmd = xen_release_pmd_init,
1994 .set_pte = xen_set_pte_init,
1995 .set_pte_at = xen_set_pte_at,
1996 .set_pmd = xen_set_pmd_hyper,
1998 .ptep_modify_prot_start = __ptep_modify_prot_start,
1999 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2001 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2002 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2004 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2005 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2007 #ifdef CONFIG_X86_PAE
2008 .set_pte_atomic = xen_set_pte_atomic,
2009 .pte_clear = xen_pte_clear,
2010 .pmd_clear = xen_pmd_clear,
2011 #endif /* CONFIG_X86_PAE */
2012 .set_pud = xen_set_pud_hyper,
2014 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2015 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2017 #if PAGETABLE_LEVELS == 4
2018 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2019 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2020 .set_pgd = xen_set_pgd_hyper,
2022 .alloc_pud = xen_alloc_pmd_init,
2023 .release_pud = xen_release_pmd_init,
2024 #endif /* PAGETABLE_LEVELS == 4 */
2026 .activate_mm = xen_activate_mm,
2027 .dup_mmap = xen_dup_mmap,
2028 .exit_mmap = xen_exit_mmap,
2031 .enter = paravirt_enter_lazy_mmu,
2032 .leave = xen_leave_lazy_mmu,
2035 .set_fixmap = xen_set_fixmap,
2038 void __init xen_init_mmu_ops(void)
2040 x86_init.mapping.pagetable_reserve = xen_mapping_pagetable_reserve;
2041 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2042 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2043 pv_mmu_ops = xen_mmu_ops;
2045 memset(dummy_mapping, 0xff, PAGE_SIZE);
2048 /* Protected by xen_reservation_lock. */
2049 #define MAX_CONTIG_ORDER 9 /* 2MB */
2050 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2052 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2053 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2054 unsigned long *in_frames,
2055 unsigned long *out_frames)
2058 struct multicall_space mcs;
2061 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2062 mcs = __xen_mc_entry(0);
2065 in_frames[i] = virt_to_mfn(vaddr);
2067 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2068 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2071 out_frames[i] = virt_to_pfn(vaddr);
2077 * Update the pfn-to-mfn mappings for a virtual address range, either to
2078 * point to an array of mfns, or contiguously from a single starting
2081 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2082 unsigned long *mfns,
2083 unsigned long first_mfn)
2090 limit = 1u << order;
2091 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2092 struct multicall_space mcs;
2095 mcs = __xen_mc_entry(0);
2099 mfn = first_mfn + i;
2101 if (i < (limit - 1))
2105 flags = UVMF_INVLPG | UVMF_ALL;
2107 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2110 MULTI_update_va_mapping(mcs.mc, vaddr,
2111 mfn_pte(mfn, PAGE_KERNEL), flags);
2113 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2120 * Perform the hypercall to exchange a region of our pfns to point to
2121 * memory with the required contiguous alignment. Takes the pfns as
2122 * input, and populates mfns as output.
2124 * Returns a success code indicating whether the hypervisor was able to
2125 * satisfy the request or not.
2127 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2128 unsigned long *pfns_in,
2129 unsigned long extents_out,
2130 unsigned int order_out,
2131 unsigned long *mfns_out,
2132 unsigned int address_bits)
2137 struct xen_memory_exchange exchange = {
2139 .nr_extents = extents_in,
2140 .extent_order = order_in,
2141 .extent_start = pfns_in,
2145 .nr_extents = extents_out,
2146 .extent_order = order_out,
2147 .extent_start = mfns_out,
2148 .address_bits = address_bits,
2153 BUG_ON(extents_in << order_in != extents_out << order_out);
2155 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2156 success = (exchange.nr_exchanged == extents_in);
2158 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2159 BUG_ON(success && (rc != 0));
2164 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2165 unsigned int address_bits)
2167 unsigned long *in_frames = discontig_frames, out_frame;
2168 unsigned long flags;
2172 * Currently an auto-translated guest will not perform I/O, nor will
2173 * it require PAE page directories below 4GB. Therefore any calls to
2174 * this function are redundant and can be ignored.
2177 if (xen_feature(XENFEAT_auto_translated_physmap))
2180 if (unlikely(order > MAX_CONTIG_ORDER))
2183 memset((void *) vstart, 0, PAGE_SIZE << order);
2185 spin_lock_irqsave(&xen_reservation_lock, flags);
2187 /* 1. Zap current PTEs, remembering MFNs. */
2188 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2190 /* 2. Get a new contiguous memory extent. */
2191 out_frame = virt_to_pfn(vstart);
2192 success = xen_exchange_memory(1UL << order, 0, in_frames,
2193 1, order, &out_frame,
2196 /* 3. Map the new extent in place of old pages. */
2198 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2200 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2202 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2204 return success ? 0 : -ENOMEM;
2206 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2208 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2210 unsigned long *out_frames = discontig_frames, in_frame;
2211 unsigned long flags;
2214 if (xen_feature(XENFEAT_auto_translated_physmap))
2217 if (unlikely(order > MAX_CONTIG_ORDER))
2220 memset((void *) vstart, 0, PAGE_SIZE << order);
2222 spin_lock_irqsave(&xen_reservation_lock, flags);
2224 /* 1. Find start MFN of contiguous extent. */
2225 in_frame = virt_to_mfn(vstart);
2227 /* 2. Zap current PTEs. */
2228 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2230 /* 3. Do the exchange for non-contiguous MFNs. */
2231 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2234 /* 4. Map new pages in place of old pages. */
2236 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2238 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2240 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2242 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2244 #ifdef CONFIG_XEN_PVHVM
2245 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2247 struct xen_hvm_pagetable_dying a;
2250 a.domid = DOMID_SELF;
2251 a.gpa = __pa(mm->pgd);
2252 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2253 WARN_ON_ONCE(rc < 0);
2256 static int is_pagetable_dying_supported(void)
2258 struct xen_hvm_pagetable_dying a;
2261 a.domid = DOMID_SELF;
2263 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2265 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2271 void __init xen_hvm_init_mmu_ops(void)
2273 if (is_pagetable_dying_supported())
2274 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2278 #define REMAP_BATCH_SIZE 16
2283 struct mmu_update *mmu_update;
2286 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2287 unsigned long addr, void *data)
2289 struct remap_data *rmd = data;
2290 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2292 rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2293 rmd->mmu_update->val = pte_val_ma(pte);
2299 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2301 unsigned long mfn, int nr,
2302 pgprot_t prot, unsigned domid)
2304 struct remap_data rmd;
2305 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2307 unsigned long range;
2310 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2312 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2313 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2319 batch = min(REMAP_BATCH_SIZE, nr);
2320 range = (unsigned long)batch << PAGE_SHIFT;
2322 rmd.mmu_update = mmu_update;
2323 err = apply_to_page_range(vma->vm_mm, addr, range,
2324 remap_area_mfn_pte_fn, &rmd);
2329 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2343 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2345 #ifdef CONFIG_XEN_DEBUG_FS
2346 static int p2m_dump_open(struct inode *inode, struct file *filp)
2348 return single_open(filp, p2m_dump_show, NULL);
2351 static const struct file_operations p2m_dump_fops = {
2352 .open = p2m_dump_open,
2354 .llseek = seq_lseek,
2355 .release = single_release,
2357 #endif /* CONFIG_XEN_DEBUG_FS */