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_extend_mmuext_op(const struct mmuext_op *op)
234 struct multicall_space mcs;
237 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
239 if (mcs.mc != NULL) {
242 mcs = __xen_mc_entry(sizeof(*u));
243 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
250 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
258 /* ptr may be ioremapped for 64-bit pagetable setup */
259 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
260 u.val = pmd_val_ma(val);
261 xen_extend_mmu_update(&u);
263 xen_mc_issue(PARAVIRT_LAZY_MMU);
268 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
270 trace_xen_mmu_set_pmd(ptr, val);
272 /* If page is not pinned, we can just update the entry
274 if (!xen_page_pinned(ptr)) {
279 xen_set_pmd_hyper(ptr, val);
283 * Associate a virtual page frame with a given physical page frame
284 * and protection flags for that frame.
286 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
288 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
291 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
295 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
300 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
301 u.val = pte_val_ma(pteval);
302 xen_extend_mmu_update(&u);
304 xen_mc_issue(PARAVIRT_LAZY_MMU);
309 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
311 if (!xen_batched_set_pte(ptep, pteval)) {
313 * Could call native_set_pte() here and trap and
314 * emulate the PTE write but with 32-bit guests this
315 * needs two traps (one for each of the two 32-bit
316 * words in the PTE) so do one hypercall directly
321 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
322 u.val = pte_val_ma(pteval);
323 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
327 static void xen_set_pte(pte_t *ptep, pte_t pteval)
329 trace_xen_mmu_set_pte(ptep, pteval);
330 __xen_set_pte(ptep, pteval);
333 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
334 pte_t *ptep, pte_t pteval)
336 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
337 __xen_set_pte(ptep, pteval);
340 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
341 unsigned long addr, pte_t *ptep)
343 /* Just return the pte as-is. We preserve the bits on commit */
344 trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
348 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
349 pte_t *ptep, pte_t pte)
353 trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
356 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
357 u.val = pte_val_ma(pte);
358 xen_extend_mmu_update(&u);
360 xen_mc_issue(PARAVIRT_LAZY_MMU);
363 /* Assume pteval_t is equivalent to all the other *val_t types. */
364 static pteval_t pte_mfn_to_pfn(pteval_t val)
366 if (val & _PAGE_PRESENT) {
367 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
368 unsigned long pfn = mfn_to_pfn(mfn);
370 pteval_t flags = val & PTE_FLAGS_MASK;
371 if (unlikely(pfn == ~0))
372 val = flags & ~_PAGE_PRESENT;
374 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
380 static pteval_t pte_pfn_to_mfn(pteval_t val)
382 if (val & _PAGE_PRESENT) {
383 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
384 pteval_t flags = val & PTE_FLAGS_MASK;
387 if (!xen_feature(XENFEAT_auto_translated_physmap))
388 mfn = get_phys_to_machine(pfn);
392 * If there's no mfn for the pfn, then just create an
393 * empty non-present pte. Unfortunately this loses
394 * information about the original pfn, so
395 * pte_mfn_to_pfn is asymmetric.
397 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
402 * Paramount to do this test _after_ the
403 * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
404 * IDENTITY_FRAME_BIT resolves to true.
406 mfn &= ~FOREIGN_FRAME_BIT;
407 if (mfn & IDENTITY_FRAME_BIT) {
408 mfn &= ~IDENTITY_FRAME_BIT;
409 flags |= _PAGE_IOMAP;
412 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
418 static pteval_t iomap_pte(pteval_t val)
420 if (val & _PAGE_PRESENT) {
421 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
422 pteval_t flags = val & PTE_FLAGS_MASK;
424 /* We assume the pte frame number is a MFN, so
425 just use it as-is. */
426 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
432 static pteval_t xen_pte_val(pte_t pte)
434 pteval_t pteval = pte.pte;
436 /* If this is a WC pte, convert back from Xen WC to Linux WC */
437 if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
438 WARN_ON(!pat_enabled);
439 pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
442 if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
445 return pte_mfn_to_pfn(pteval);
447 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
449 static pgdval_t xen_pgd_val(pgd_t pgd)
451 return pte_mfn_to_pfn(pgd.pgd);
453 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
456 * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
457 * are reserved for now, to correspond to the Intel-reserved PAT
460 * We expect Linux's PAT set as follows:
462 * Idx PTE flags Linux Xen Default
469 * 6 PAT PCD UC- UC UC-
470 * 7 PAT PCD PWT UC UC UC
473 void xen_set_pat(u64 pat)
475 /* We expect Linux to use a PAT setting of
476 * UC UC- WC WB (ignoring the PAT flag) */
477 WARN_ON(pat != 0x0007010600070106ull);
480 static pte_t xen_make_pte(pteval_t pte)
482 phys_addr_t addr = (pte & PTE_PFN_MASK);
484 /* If Linux is trying to set a WC pte, then map to the Xen WC.
485 * If _PAGE_PAT is set, then it probably means it is really
486 * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
487 * things work out OK...
489 * (We should never see kernel mappings with _PAGE_PSE set,
490 * but we could see hugetlbfs mappings, I think.).
492 if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
493 if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
494 pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
498 * Unprivileged domains are allowed to do IOMAPpings for
499 * PCI passthrough, but not map ISA space. The ISA
500 * mappings are just dummy local mappings to keep other
501 * parts of the kernel happy.
503 if (unlikely(pte & _PAGE_IOMAP) &&
504 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
505 pte = iomap_pte(pte);
508 pte = pte_pfn_to_mfn(pte);
511 return native_make_pte(pte);
513 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
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)
816 op.arg1.mfn = pfn_to_mfn(pfn);
818 xen_extend_mmuext_op(&op);
821 static int xen_pin_page(struct mm_struct *mm, struct page *page,
824 unsigned pgfl = TestSetPagePinned(page);
828 flush = 0; /* already pinned */
829 else if (PageHighMem(page))
830 /* kmaps need flushing if we found an unpinned
834 void *pt = lowmem_page_address(page);
835 unsigned long pfn = page_to_pfn(page);
836 struct multicall_space mcs = __xen_mc_entry(0);
842 * We need to hold the pagetable lock between the time
843 * we make the pagetable RO and when we actually pin
844 * it. If we don't, then other users may come in and
845 * attempt to update the pagetable by writing it,
846 * which will fail because the memory is RO but not
847 * pinned, so Xen won't do the trap'n'emulate.
849 * If we're using split pte locks, we can't hold the
850 * entire pagetable's worth of locks during the
851 * traverse, because we may wrap the preempt count (8
852 * bits). The solution is to mark RO and pin each PTE
853 * page while holding the lock. This means the number
854 * of locks we end up holding is never more than a
855 * batch size (~32 entries, at present).
857 * If we're not using split pte locks, we needn't pin
858 * the PTE pages independently, because we're
859 * protected by the overall pagetable lock.
863 ptl = xen_pte_lock(page, mm);
865 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
866 pfn_pte(pfn, PAGE_KERNEL_RO),
867 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
870 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
872 /* Queue a deferred unlock for when this batch
874 xen_mc_callback(xen_pte_unlock, ptl);
881 /* This is called just after a mm has been created, but it has not
882 been used yet. We need to make sure that its pagetable is all
883 read-only, and can be pinned. */
884 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
886 trace_xen_mmu_pgd_pin(mm, 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)
1013 trace_xen_mmu_pgd_unpin(mm, pgd);
1017 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1019 #ifdef CONFIG_X86_64
1021 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1024 xen_do_pin(MMUEXT_UNPIN_TABLE,
1025 PFN_DOWN(__pa(user_pgd)));
1026 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1031 #ifdef CONFIG_X86_PAE
1032 /* Need to make sure unshared kernel PMD is unpinned */
1033 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1037 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1042 static void xen_pgd_unpin(struct mm_struct *mm)
1044 __xen_pgd_unpin(mm, mm->pgd);
1048 * On resume, undo any pinning done at save, so that the rest of the
1049 * kernel doesn't see any unexpected pinned pagetables.
1051 void xen_mm_unpin_all(void)
1055 spin_lock(&pgd_lock);
1057 list_for_each_entry(page, &pgd_list, lru) {
1058 if (PageSavePinned(page)) {
1059 BUG_ON(!PagePinned(page));
1060 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1061 ClearPageSavePinned(page);
1065 spin_unlock(&pgd_lock);
1068 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1070 spin_lock(&next->page_table_lock);
1072 spin_unlock(&next->page_table_lock);
1075 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1077 spin_lock(&mm->page_table_lock);
1079 spin_unlock(&mm->page_table_lock);
1084 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1085 we need to repoint it somewhere else before we can unpin it. */
1086 static void drop_other_mm_ref(void *info)
1088 struct mm_struct *mm = info;
1089 struct mm_struct *active_mm;
1091 active_mm = this_cpu_read(cpu_tlbstate.active_mm);
1093 if (active_mm == mm && this_cpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
1094 leave_mm(smp_processor_id());
1096 /* If this cpu still has a stale cr3 reference, then make sure
1097 it has been flushed. */
1098 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
1099 load_cr3(swapper_pg_dir);
1102 static void xen_drop_mm_ref(struct mm_struct *mm)
1107 if (current->active_mm == mm) {
1108 if (current->mm == mm)
1109 load_cr3(swapper_pg_dir);
1111 leave_mm(smp_processor_id());
1114 /* Get the "official" set of cpus referring to our pagetable. */
1115 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1116 for_each_online_cpu(cpu) {
1117 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1118 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1120 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1124 cpumask_copy(mask, mm_cpumask(mm));
1126 /* It's possible that a vcpu may have a stale reference to our
1127 cr3, because its in lazy mode, and it hasn't yet flushed
1128 its set of pending hypercalls yet. In this case, we can
1129 look at its actual current cr3 value, and force it to flush
1131 for_each_online_cpu(cpu) {
1132 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1133 cpumask_set_cpu(cpu, mask);
1136 if (!cpumask_empty(mask))
1137 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1138 free_cpumask_var(mask);
1141 static void xen_drop_mm_ref(struct mm_struct *mm)
1143 if (current->active_mm == mm)
1144 load_cr3(swapper_pg_dir);
1149 * While a process runs, Xen pins its pagetables, which means that the
1150 * hypervisor forces it to be read-only, and it controls all updates
1151 * to it. This means that all pagetable updates have to go via the
1152 * hypervisor, which is moderately expensive.
1154 * Since we're pulling the pagetable down, we switch to use init_mm,
1155 * unpin old process pagetable and mark it all read-write, which
1156 * allows further operations on it to be simple memory accesses.
1158 * The only subtle point is that another CPU may be still using the
1159 * pagetable because of lazy tlb flushing. This means we need need to
1160 * switch all CPUs off this pagetable before we can unpin it.
1162 static void xen_exit_mmap(struct mm_struct *mm)
1164 get_cpu(); /* make sure we don't move around */
1165 xen_drop_mm_ref(mm);
1168 spin_lock(&mm->page_table_lock);
1170 /* pgd may not be pinned in the error exit path of execve */
1171 if (xen_page_pinned(mm->pgd))
1174 spin_unlock(&mm->page_table_lock);
1177 static void xen_post_allocator_init(void);
1179 static void __init xen_pagetable_init(void)
1182 xen_setup_shared_info();
1183 xen_post_allocator_init();
1186 static __init void xen_mapping_pagetable_reserve(u64 start, u64 end)
1188 /* reserve the range used */
1189 native_pagetable_reserve(start, end);
1191 /* set as RW the rest */
1192 printk(KERN_DEBUG "xen: setting RW the range %llx - %llx\n", end,
1193 PFN_PHYS(pgt_buf_top));
1194 while (end < PFN_PHYS(pgt_buf_top)) {
1195 make_lowmem_page_readwrite(__va(end));
1200 static void xen_write_cr2(unsigned long cr2)
1202 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1205 static unsigned long xen_read_cr2(void)
1207 return this_cpu_read(xen_vcpu)->arch.cr2;
1210 unsigned long xen_read_cr2_direct(void)
1212 return this_cpu_read(xen_vcpu_info.arch.cr2);
1215 static void xen_flush_tlb(void)
1217 struct mmuext_op *op;
1218 struct multicall_space mcs;
1220 trace_xen_mmu_flush_tlb(0);
1224 mcs = xen_mc_entry(sizeof(*op));
1227 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1228 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1230 xen_mc_issue(PARAVIRT_LAZY_MMU);
1235 static void xen_flush_tlb_single(unsigned long addr)
1237 struct mmuext_op *op;
1238 struct multicall_space mcs;
1240 trace_xen_mmu_flush_tlb_single(addr);
1244 mcs = xen_mc_entry(sizeof(*op));
1246 op->cmd = MMUEXT_INVLPG_LOCAL;
1247 op->arg1.linear_addr = addr & PAGE_MASK;
1248 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1250 xen_mc_issue(PARAVIRT_LAZY_MMU);
1255 static void xen_flush_tlb_others(const struct cpumask *cpus,
1256 struct mm_struct *mm, unsigned long start,
1260 struct mmuext_op op;
1262 DECLARE_BITMAP(mask, num_processors);
1264 DECLARE_BITMAP(mask, NR_CPUS);
1267 struct multicall_space mcs;
1269 trace_xen_mmu_flush_tlb_others(cpus, mm, start, end);
1271 if (cpumask_empty(cpus))
1272 return; /* nothing to do */
1274 mcs = xen_mc_entry(sizeof(*args));
1276 args->op.arg2.vcpumask = to_cpumask(args->mask);
1278 /* Remove us, and any offline CPUS. */
1279 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1280 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1282 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1283 if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) {
1284 args->op.cmd = MMUEXT_INVLPG_MULTI;
1285 args->op.arg1.linear_addr = start;
1288 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1290 xen_mc_issue(PARAVIRT_LAZY_MMU);
1293 static unsigned long xen_read_cr3(void)
1295 return this_cpu_read(xen_cr3);
1298 static void set_current_cr3(void *v)
1300 this_cpu_write(xen_current_cr3, (unsigned long)v);
1303 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1305 struct mmuext_op op;
1308 trace_xen_mmu_write_cr3(kernel, cr3);
1311 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1315 WARN_ON(mfn == 0 && kernel);
1317 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1320 xen_extend_mmuext_op(&op);
1323 this_cpu_write(xen_cr3, cr3);
1325 /* Update xen_current_cr3 once the batch has actually
1327 xen_mc_callback(set_current_cr3, (void *)cr3);
1331 static void xen_write_cr3(unsigned long cr3)
1333 BUG_ON(preemptible());
1335 xen_mc_batch(); /* disables interrupts */
1337 /* Update while interrupts are disabled, so its atomic with
1339 this_cpu_write(xen_cr3, cr3);
1341 __xen_write_cr3(true, cr3);
1343 #ifdef CONFIG_X86_64
1345 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1347 __xen_write_cr3(false, __pa(user_pgd));
1349 __xen_write_cr3(false, 0);
1353 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1356 static int xen_pgd_alloc(struct mm_struct *mm)
1358 pgd_t *pgd = mm->pgd;
1361 BUG_ON(PagePinned(virt_to_page(pgd)));
1363 #ifdef CONFIG_X86_64
1365 struct page *page = virt_to_page(pgd);
1368 BUG_ON(page->private != 0);
1372 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1373 page->private = (unsigned long)user_pgd;
1375 if (user_pgd != NULL) {
1376 user_pgd[pgd_index(VSYSCALL_START)] =
1377 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1381 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1388 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1390 #ifdef CONFIG_X86_64
1391 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1394 free_page((unsigned long)user_pgd);
1398 #ifdef CONFIG_X86_32
1399 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1401 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1402 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1403 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1408 #else /* CONFIG_X86_64 */
1409 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1411 unsigned long pfn = pte_pfn(pte);
1414 * If the new pfn is within the range of the newly allocated
1415 * kernel pagetable, and it isn't being mapped into an
1416 * early_ioremap fixmap slot as a freshly allocated page, make sure
1419 if (((!is_early_ioremap_ptep(ptep) &&
1420 pfn >= pgt_buf_start && pfn < pgt_buf_top)) ||
1421 (is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1422 pte = pte_wrprotect(pte);
1426 #endif /* CONFIG_X86_64 */
1429 * Init-time set_pte while constructing initial pagetables, which
1430 * doesn't allow RO page table pages to be remapped RW.
1432 * If there is no MFN for this PFN then this page is initially
1433 * ballooned out so clear the PTE (as in decrease_reservation() in
1434 * drivers/xen/balloon.c).
1436 * Many of these PTE updates are done on unpinned and writable pages
1437 * and doing a hypercall for these is unnecessary and expensive. At
1438 * this point it is not possible to tell if a page is pinned or not,
1439 * so always write the PTE directly and rely on Xen trapping and
1440 * emulating any updates as necessary.
1442 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1444 if (pte_mfn(pte) != INVALID_P2M_ENTRY)
1445 pte = mask_rw_pte(ptep, pte);
1449 native_set_pte(ptep, pte);
1452 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1454 struct mmuext_op op;
1456 op.arg1.mfn = pfn_to_mfn(pfn);
1457 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1461 /* Early in boot, while setting up the initial pagetable, assume
1462 everything is pinned. */
1463 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1465 #ifdef CONFIG_FLATMEM
1466 BUG_ON(mem_map); /* should only be used early */
1468 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1469 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1472 /* Used for pmd and pud */
1473 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1475 #ifdef CONFIG_FLATMEM
1476 BUG_ON(mem_map); /* should only be used early */
1478 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1481 /* Early release_pte assumes that all pts are pinned, since there's
1482 only init_mm and anything attached to that is pinned. */
1483 static void __init xen_release_pte_init(unsigned long pfn)
1485 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1486 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1489 static void __init xen_release_pmd_init(unsigned long pfn)
1491 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1494 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1496 struct multicall_space mcs;
1497 struct mmuext_op *op;
1499 mcs = __xen_mc_entry(sizeof(*op));
1502 op->arg1.mfn = pfn_to_mfn(pfn);
1504 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1507 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1509 struct multicall_space mcs;
1510 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1512 mcs = __xen_mc_entry(0);
1513 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1514 pfn_pte(pfn, prot), 0);
1517 /* This needs to make sure the new pte page is pinned iff its being
1518 attached to a pinned pagetable. */
1519 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1522 bool pinned = PagePinned(virt_to_page(mm->pgd));
1524 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1527 struct page *page = pfn_to_page(pfn);
1529 SetPagePinned(page);
1531 if (!PageHighMem(page)) {
1534 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1536 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1537 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1539 xen_mc_issue(PARAVIRT_LAZY_MMU);
1541 /* make sure there are no stray mappings of
1543 kmap_flush_unused();
1548 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1550 xen_alloc_ptpage(mm, pfn, PT_PTE);
1553 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1555 xen_alloc_ptpage(mm, pfn, PT_PMD);
1558 /* This should never happen until we're OK to use struct page */
1559 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1561 struct page *page = pfn_to_page(pfn);
1562 bool pinned = PagePinned(page);
1564 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1567 if (!PageHighMem(page)) {
1570 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1571 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1573 __set_pfn_prot(pfn, PAGE_KERNEL);
1575 xen_mc_issue(PARAVIRT_LAZY_MMU);
1577 ClearPagePinned(page);
1581 static void xen_release_pte(unsigned long pfn)
1583 xen_release_ptpage(pfn, PT_PTE);
1586 static void xen_release_pmd(unsigned long pfn)
1588 xen_release_ptpage(pfn, PT_PMD);
1591 #if PAGETABLE_LEVELS == 4
1592 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1594 xen_alloc_ptpage(mm, pfn, PT_PUD);
1597 static void xen_release_pud(unsigned long pfn)
1599 xen_release_ptpage(pfn, PT_PUD);
1603 void __init xen_reserve_top(void)
1605 #ifdef CONFIG_X86_32
1606 unsigned long top = HYPERVISOR_VIRT_START;
1607 struct xen_platform_parameters pp;
1609 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1610 top = pp.virt_start;
1612 reserve_top_address(-top);
1613 #endif /* CONFIG_X86_32 */
1617 * Like __va(), but returns address in the kernel mapping (which is
1618 * all we have until the physical memory mapping has been set up.
1620 static void *__ka(phys_addr_t paddr)
1622 #ifdef CONFIG_X86_64
1623 return (void *)(paddr + __START_KERNEL_map);
1629 /* Convert a machine address to physical address */
1630 static unsigned long m2p(phys_addr_t maddr)
1634 maddr &= PTE_PFN_MASK;
1635 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1640 /* Convert a machine address to kernel virtual */
1641 static void *m2v(phys_addr_t maddr)
1643 return __ka(m2p(maddr));
1646 /* Set the page permissions on an identity-mapped pages */
1647 static void set_page_prot(void *addr, pgprot_t prot)
1649 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1650 pte_t pte = pfn_pte(pfn, prot);
1652 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1656 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1658 unsigned pmdidx, pteidx;
1662 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1667 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1670 /* Reuse or allocate a page of ptes */
1671 if (pmd_present(pmd[pmdidx]))
1672 pte_page = m2v(pmd[pmdidx].pmd);
1674 /* Check for free pte pages */
1675 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1678 pte_page = &level1_ident_pgt[ident_pte];
1679 ident_pte += PTRS_PER_PTE;
1681 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1684 /* Install mappings */
1685 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1688 #ifdef CONFIG_X86_32
1689 if (pfn > max_pfn_mapped)
1690 max_pfn_mapped = pfn;
1693 if (!pte_none(pte_page[pteidx]))
1696 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1697 pte_page[pteidx] = pte;
1701 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1702 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1704 set_page_prot(pmd, PAGE_KERNEL_RO);
1707 void __init xen_setup_machphys_mapping(void)
1709 struct xen_machphys_mapping mapping;
1711 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1712 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1713 machine_to_phys_nr = mapping.max_mfn + 1;
1715 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1717 #ifdef CONFIG_X86_32
1718 WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1719 < machine_to_phys_mapping);
1723 #ifdef CONFIG_X86_64
1724 static void convert_pfn_mfn(void *v)
1729 /* All levels are converted the same way, so just treat them
1731 for (i = 0; i < PTRS_PER_PTE; i++)
1732 pte[i] = xen_make_pte(pte[i].pte);
1736 * Set up the initial kernel pagetable.
1738 * We can construct this by grafting the Xen provided pagetable into
1739 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1740 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1741 * means that only the kernel has a physical mapping to start with -
1742 * but that's enough to get __va working. We need to fill in the rest
1743 * of the physical mapping once some sort of allocator has been set
1746 pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1747 unsigned long max_pfn)
1752 /* max_pfn_mapped is the last pfn mapped in the initial memory
1753 * mappings. Considering that on Xen after the kernel mappings we
1754 * have the mappings of some pages that don't exist in pfn space, we
1755 * set max_pfn_mapped to the last real pfn mapped. */
1756 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1758 /* Zap identity mapping */
1759 init_level4_pgt[0] = __pgd(0);
1761 /* Pre-constructed entries are in pfn, so convert to mfn */
1762 convert_pfn_mfn(init_level4_pgt);
1763 convert_pfn_mfn(level3_ident_pgt);
1764 convert_pfn_mfn(level3_kernel_pgt);
1766 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1767 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1769 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1770 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1772 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1773 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1774 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1776 /* Set up identity map */
1777 xen_map_identity_early(level2_ident_pgt, max_pfn);
1779 /* Make pagetable pieces RO */
1780 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1781 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1782 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1783 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1784 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1785 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1787 /* Pin down new L4 */
1788 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1789 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1791 /* Unpin Xen-provided one */
1792 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1795 pgd = init_level4_pgt;
1798 * At this stage there can be no user pgd, and no page
1799 * structure to attach it to, so make sure we just set kernel
1803 __xen_write_cr3(true, __pa(pgd));
1804 xen_mc_issue(PARAVIRT_LAZY_CPU);
1806 memblock_reserve(__pa(xen_start_info->pt_base),
1807 xen_start_info->nr_pt_frames * PAGE_SIZE);
1811 #else /* !CONFIG_X86_64 */
1812 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1813 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1815 static void __init xen_write_cr3_init(unsigned long cr3)
1817 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1819 BUG_ON(read_cr3() != __pa(initial_page_table));
1820 BUG_ON(cr3 != __pa(swapper_pg_dir));
1823 * We are switching to swapper_pg_dir for the first time (from
1824 * initial_page_table) and therefore need to mark that page
1825 * read-only and then pin it.
1827 * Xen disallows sharing of kernel PMDs for PAE
1828 * guests. Therefore we must copy the kernel PMD from
1829 * initial_page_table into a new kernel PMD to be used in
1832 swapper_kernel_pmd =
1833 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1834 memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1835 sizeof(pmd_t) * PTRS_PER_PMD);
1836 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1837 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1838 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1840 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1842 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1844 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1845 PFN_DOWN(__pa(initial_page_table)));
1846 set_page_prot(initial_page_table, PAGE_KERNEL);
1847 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1849 pv_mmu_ops.write_cr3 = &xen_write_cr3;
1852 pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1853 unsigned long max_pfn)
1857 initial_kernel_pmd =
1858 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1860 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1861 xen_start_info->nr_pt_frames * PAGE_SIZE +
1864 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1865 memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1867 xen_map_identity_early(initial_kernel_pmd, max_pfn);
1869 memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1870 initial_page_table[KERNEL_PGD_BOUNDARY] =
1871 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1873 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1874 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1875 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1877 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1879 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1880 PFN_DOWN(__pa(initial_page_table)));
1881 xen_write_cr3(__pa(initial_page_table));
1883 memblock_reserve(__pa(xen_start_info->pt_base),
1884 xen_start_info->nr_pt_frames * PAGE_SIZE);
1886 return initial_page_table;
1888 #endif /* CONFIG_X86_64 */
1890 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1892 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1896 phys >>= PAGE_SHIFT;
1899 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1900 #ifdef CONFIG_X86_F00F_BUG
1903 #ifdef CONFIG_X86_32
1906 # ifdef CONFIG_HIGHMEM
1907 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1910 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1913 case FIX_TEXT_POKE0:
1914 case FIX_TEXT_POKE1:
1915 /* All local page mappings */
1916 pte = pfn_pte(phys, prot);
1919 #ifdef CONFIG_X86_LOCAL_APIC
1920 case FIX_APIC_BASE: /* maps dummy local APIC */
1921 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1925 #ifdef CONFIG_X86_IO_APIC
1926 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1928 * We just don't map the IO APIC - all access is via
1929 * hypercalls. Keep the address in the pte for reference.
1931 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1935 case FIX_PARAVIRT_BOOTMAP:
1936 /* This is an MFN, but it isn't an IO mapping from the
1938 pte = mfn_pte(phys, prot);
1942 /* By default, set_fixmap is used for hardware mappings */
1943 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1947 __native_set_fixmap(idx, pte);
1949 #ifdef CONFIG_X86_64
1950 /* Replicate changes to map the vsyscall page into the user
1951 pagetable vsyscall mapping. */
1952 if ((idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) ||
1954 unsigned long vaddr = __fix_to_virt(idx);
1955 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1960 static void __init xen_post_allocator_init(void)
1962 pv_mmu_ops.set_pte = xen_set_pte;
1963 pv_mmu_ops.set_pmd = xen_set_pmd;
1964 pv_mmu_ops.set_pud = xen_set_pud;
1965 #if PAGETABLE_LEVELS == 4
1966 pv_mmu_ops.set_pgd = xen_set_pgd;
1969 /* This will work as long as patching hasn't happened yet
1970 (which it hasn't) */
1971 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1972 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1973 pv_mmu_ops.release_pte = xen_release_pte;
1974 pv_mmu_ops.release_pmd = xen_release_pmd;
1975 #if PAGETABLE_LEVELS == 4
1976 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1977 pv_mmu_ops.release_pud = xen_release_pud;
1980 #ifdef CONFIG_X86_64
1981 SetPagePinned(virt_to_page(level3_user_vsyscall));
1983 xen_mark_init_mm_pinned();
1986 static void xen_leave_lazy_mmu(void)
1990 paravirt_leave_lazy_mmu();
1994 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
1995 .read_cr2 = xen_read_cr2,
1996 .write_cr2 = xen_write_cr2,
1998 .read_cr3 = xen_read_cr3,
1999 #ifdef CONFIG_X86_32
2000 .write_cr3 = xen_write_cr3_init,
2002 .write_cr3 = xen_write_cr3,
2005 .flush_tlb_user = xen_flush_tlb,
2006 .flush_tlb_kernel = xen_flush_tlb,
2007 .flush_tlb_single = xen_flush_tlb_single,
2008 .flush_tlb_others = xen_flush_tlb_others,
2010 .pte_update = paravirt_nop,
2011 .pte_update_defer = paravirt_nop,
2013 .pgd_alloc = xen_pgd_alloc,
2014 .pgd_free = xen_pgd_free,
2016 .alloc_pte = xen_alloc_pte_init,
2017 .release_pte = xen_release_pte_init,
2018 .alloc_pmd = xen_alloc_pmd_init,
2019 .release_pmd = xen_release_pmd_init,
2021 .set_pte = xen_set_pte_init,
2022 .set_pte_at = xen_set_pte_at,
2023 .set_pmd = xen_set_pmd_hyper,
2025 .ptep_modify_prot_start = __ptep_modify_prot_start,
2026 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2028 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2029 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2031 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2032 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2034 #ifdef CONFIG_X86_PAE
2035 .set_pte_atomic = xen_set_pte_atomic,
2036 .pte_clear = xen_pte_clear,
2037 .pmd_clear = xen_pmd_clear,
2038 #endif /* CONFIG_X86_PAE */
2039 .set_pud = xen_set_pud_hyper,
2041 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2042 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2044 #if PAGETABLE_LEVELS == 4
2045 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2046 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2047 .set_pgd = xen_set_pgd_hyper,
2049 .alloc_pud = xen_alloc_pmd_init,
2050 .release_pud = xen_release_pmd_init,
2051 #endif /* PAGETABLE_LEVELS == 4 */
2053 .activate_mm = xen_activate_mm,
2054 .dup_mmap = xen_dup_mmap,
2055 .exit_mmap = xen_exit_mmap,
2058 .enter = paravirt_enter_lazy_mmu,
2059 .leave = xen_leave_lazy_mmu,
2062 .set_fixmap = xen_set_fixmap,
2065 void __init xen_init_mmu_ops(void)
2067 x86_init.mapping.pagetable_reserve = xen_mapping_pagetable_reserve;
2068 x86_init.paging.pagetable_init = xen_pagetable_init;
2069 pv_mmu_ops = xen_mmu_ops;
2071 memset(dummy_mapping, 0xff, PAGE_SIZE);
2074 /* Protected by xen_reservation_lock. */
2075 #define MAX_CONTIG_ORDER 9 /* 2MB */
2076 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2078 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2079 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2080 unsigned long *in_frames,
2081 unsigned long *out_frames)
2084 struct multicall_space mcs;
2087 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2088 mcs = __xen_mc_entry(0);
2091 in_frames[i] = virt_to_mfn(vaddr);
2093 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2094 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2097 out_frames[i] = virt_to_pfn(vaddr);
2103 * Update the pfn-to-mfn mappings for a virtual address range, either to
2104 * point to an array of mfns, or contiguously from a single starting
2107 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2108 unsigned long *mfns,
2109 unsigned long first_mfn)
2116 limit = 1u << order;
2117 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2118 struct multicall_space mcs;
2121 mcs = __xen_mc_entry(0);
2125 mfn = first_mfn + i;
2127 if (i < (limit - 1))
2131 flags = UVMF_INVLPG | UVMF_ALL;
2133 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2136 MULTI_update_va_mapping(mcs.mc, vaddr,
2137 mfn_pte(mfn, PAGE_KERNEL), flags);
2139 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2146 * Perform the hypercall to exchange a region of our pfns to point to
2147 * memory with the required contiguous alignment. Takes the pfns as
2148 * input, and populates mfns as output.
2150 * Returns a success code indicating whether the hypervisor was able to
2151 * satisfy the request or not.
2153 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2154 unsigned long *pfns_in,
2155 unsigned long extents_out,
2156 unsigned int order_out,
2157 unsigned long *mfns_out,
2158 unsigned int address_bits)
2163 struct xen_memory_exchange exchange = {
2165 .nr_extents = extents_in,
2166 .extent_order = order_in,
2167 .extent_start = pfns_in,
2171 .nr_extents = extents_out,
2172 .extent_order = order_out,
2173 .extent_start = mfns_out,
2174 .address_bits = address_bits,
2179 BUG_ON(extents_in << order_in != extents_out << order_out);
2181 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2182 success = (exchange.nr_exchanged == extents_in);
2184 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2185 BUG_ON(success && (rc != 0));
2190 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2191 unsigned int address_bits)
2193 unsigned long *in_frames = discontig_frames, out_frame;
2194 unsigned long flags;
2198 * Currently an auto-translated guest will not perform I/O, nor will
2199 * it require PAE page directories below 4GB. Therefore any calls to
2200 * this function are redundant and can be ignored.
2203 if (xen_feature(XENFEAT_auto_translated_physmap))
2206 if (unlikely(order > MAX_CONTIG_ORDER))
2209 memset((void *) vstart, 0, PAGE_SIZE << order);
2211 spin_lock_irqsave(&xen_reservation_lock, flags);
2213 /* 1. Zap current PTEs, remembering MFNs. */
2214 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2216 /* 2. Get a new contiguous memory extent. */
2217 out_frame = virt_to_pfn(vstart);
2218 success = xen_exchange_memory(1UL << order, 0, in_frames,
2219 1, order, &out_frame,
2222 /* 3. Map the new extent in place of old pages. */
2224 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2226 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2228 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2230 return success ? 0 : -ENOMEM;
2232 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2234 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2236 unsigned long *out_frames = discontig_frames, in_frame;
2237 unsigned long flags;
2240 if (xen_feature(XENFEAT_auto_translated_physmap))
2243 if (unlikely(order > MAX_CONTIG_ORDER))
2246 memset((void *) vstart, 0, PAGE_SIZE << order);
2248 spin_lock_irqsave(&xen_reservation_lock, flags);
2250 /* 1. Find start MFN of contiguous extent. */
2251 in_frame = virt_to_mfn(vstart);
2253 /* 2. Zap current PTEs. */
2254 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2256 /* 3. Do the exchange for non-contiguous MFNs. */
2257 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2260 /* 4. Map new pages in place of old pages. */
2262 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2264 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2266 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2268 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2270 #ifdef CONFIG_XEN_PVHVM
2271 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2273 struct xen_hvm_pagetable_dying a;
2276 a.domid = DOMID_SELF;
2277 a.gpa = __pa(mm->pgd);
2278 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2279 WARN_ON_ONCE(rc < 0);
2282 static int is_pagetable_dying_supported(void)
2284 struct xen_hvm_pagetable_dying a;
2287 a.domid = DOMID_SELF;
2289 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2291 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2297 void __init xen_hvm_init_mmu_ops(void)
2299 if (is_pagetable_dying_supported())
2300 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2304 #define REMAP_BATCH_SIZE 16
2309 struct mmu_update *mmu_update;
2312 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2313 unsigned long addr, void *data)
2315 struct remap_data *rmd = data;
2316 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2318 rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2319 rmd->mmu_update->val = pte_val_ma(pte);
2325 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2327 unsigned long mfn, int nr,
2328 pgprot_t prot, unsigned domid)
2330 struct remap_data rmd;
2331 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2333 unsigned long range;
2336 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2338 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2339 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2345 batch = min(REMAP_BATCH_SIZE, nr);
2346 range = (unsigned long)batch << PAGE_SHIFT;
2348 rmd.mmu_update = mmu_update;
2349 err = apply_to_page_range(vma->vm_mm, addr, range,
2350 remap_area_mfn_pte_fn, &rmd);
2355 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2369 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);