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 <asm/pgtable.h>
52 #include <asm/tlbflush.h>
53 #include <asm/fixmap.h>
54 #include <asm/mmu_context.h>
55 #include <asm/setup.h>
56 #include <asm/paravirt.h>
58 #include <asm/linkage.h>
63 #include <asm/xen/hypercall.h>
64 #include <asm/xen/hypervisor.h>
68 #include <xen/interface/xen.h>
69 #include <xen/interface/hvm/hvm_op.h>
70 #include <xen/interface/version.h>
71 #include <xen/interface/memory.h>
72 #include <xen/hvc-console.h>
74 #include "multicalls.h"
78 #define MMU_UPDATE_HISTO 30
81 * Protects atomic reservation decrease/increase against concurrent increases.
82 * Also protects non-atomic updates of current_pages and driver_pages, and
85 DEFINE_SPINLOCK(xen_reservation_lock);
87 #ifdef CONFIG_XEN_DEBUG_FS
91 u32 pgd_update_pinned;
92 u32 pgd_update_batched;
95 u32 pud_update_pinned;
96 u32 pud_update_batched;
99 u32 pmd_update_pinned;
100 u32 pmd_update_batched;
103 u32 pte_update_pinned;
104 u32 pte_update_batched;
107 u32 mmu_update_extended;
108 u32 mmu_update_histo[MMU_UPDATE_HISTO];
111 u32 prot_commit_batched;
114 u32 set_pte_at_batched;
115 u32 set_pte_at_pinned;
116 u32 set_pte_at_current;
117 u32 set_pte_at_kernel;
120 static u8 zero_stats;
122 static inline void check_zero(void)
124 if (unlikely(zero_stats)) {
125 memset(&mmu_stats, 0, sizeof(mmu_stats));
130 #define ADD_STATS(elem, val) \
131 do { check_zero(); mmu_stats.elem += (val); } while(0)
133 #else /* !CONFIG_XEN_DEBUG_FS */
135 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
137 #endif /* CONFIG_XEN_DEBUG_FS */
141 * Identity map, in addition to plain kernel map. This needs to be
142 * large enough to allocate page table pages to allocate the rest.
143 * Each page can map 2MB.
145 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
146 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
149 /* l3 pud for userspace vsyscall mapping */
150 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
151 #endif /* CONFIG_X86_64 */
154 * Note about cr3 (pagetable base) values:
156 * xen_cr3 contains the current logical cr3 value; it contains the
157 * last set cr3. This may not be the current effective cr3, because
158 * its update may be being lazily deferred. However, a vcpu looking
159 * at its own cr3 can use this value knowing that it everything will
160 * be self-consistent.
162 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
163 * hypercall to set the vcpu cr3 is complete (so it may be a little
164 * out of date, but it will never be set early). If one vcpu is
165 * looking at another vcpu's cr3 value, it should use this variable.
167 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
168 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
172 * Just beyond the highest usermode address. STACK_TOP_MAX has a
173 * redzone above it, so round it up to a PGD boundary.
175 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
177 unsigned long arbitrary_virt_to_mfn(void *vaddr)
179 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
181 return PFN_DOWN(maddr.maddr);
184 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
186 unsigned long address = (unsigned long)vaddr;
192 * if the PFN is in the linear mapped vaddr range, we can just use
193 * the (quick) virt_to_machine() p2m lookup
195 if (virt_addr_valid(vaddr))
196 return virt_to_machine(vaddr);
198 /* otherwise we have to do a (slower) full page-table walk */
200 pte = lookup_address(address, &level);
202 offset = address & ~PAGE_MASK;
203 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
205 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
207 void make_lowmem_page_readonly(void *vaddr)
210 unsigned long address = (unsigned long)vaddr;
213 pte = lookup_address(address, &level);
215 return; /* vaddr missing */
217 ptev = pte_wrprotect(*pte);
219 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
223 void make_lowmem_page_readwrite(void *vaddr)
226 unsigned long address = (unsigned long)vaddr;
229 pte = lookup_address(address, &level);
231 return; /* vaddr missing */
233 ptev = pte_mkwrite(*pte);
235 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
240 static bool xen_page_pinned(void *ptr)
242 struct page *page = virt_to_page(ptr);
244 return PagePinned(page);
247 static bool xen_iomap_pte(pte_t pte)
249 return pte_flags(pte) & _PAGE_IOMAP;
252 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
254 struct multicall_space mcs;
255 struct mmu_update *u;
257 mcs = xen_mc_entry(sizeof(*u));
260 /* ptep might be kmapped when using 32-bit HIGHPTE */
261 u->ptr = arbitrary_virt_to_machine(ptep).maddr;
262 u->val = pte_val_ma(pteval);
264 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
266 xen_mc_issue(PARAVIRT_LAZY_MMU);
268 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
270 static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval)
272 xen_set_domain_pte(ptep, pteval, DOMID_IO);
275 static void xen_extend_mmu_update(const struct mmu_update *update)
277 struct multicall_space mcs;
278 struct mmu_update *u;
280 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
282 if (mcs.mc != NULL) {
283 ADD_STATS(mmu_update_extended, 1);
284 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
288 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
289 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
291 ADD_STATS(mmu_update_histo[0], 1);
293 ADD_STATS(mmu_update, 1);
294 mcs = __xen_mc_entry(sizeof(*u));
295 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
296 ADD_STATS(mmu_update_histo[1], 1);
303 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
311 /* ptr may be ioremapped for 64-bit pagetable setup */
312 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
313 u.val = pmd_val_ma(val);
314 xen_extend_mmu_update(&u);
316 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
318 xen_mc_issue(PARAVIRT_LAZY_MMU);
323 void xen_set_pmd(pmd_t *ptr, pmd_t val)
325 ADD_STATS(pmd_update, 1);
327 /* If page is not pinned, we can just update the entry
329 if (!xen_page_pinned(ptr)) {
334 ADD_STATS(pmd_update_pinned, 1);
336 xen_set_pmd_hyper(ptr, val);
340 * Associate a virtual page frame with a given physical page frame
341 * and protection flags for that frame.
343 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
345 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
348 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
349 pte_t *ptep, pte_t pteval)
351 if (xen_iomap_pte(pteval)) {
352 xen_set_iomap_pte(ptep, pteval);
356 ADD_STATS(set_pte_at, 1);
357 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
358 ADD_STATS(set_pte_at_current, mm == current->mm);
359 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
361 if (mm == current->mm || mm == &init_mm) {
362 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
363 struct multicall_space mcs;
364 mcs = xen_mc_entry(0);
366 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
367 ADD_STATS(set_pte_at_batched, 1);
368 xen_mc_issue(PARAVIRT_LAZY_MMU);
371 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
374 xen_set_pte(ptep, pteval);
379 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
380 unsigned long addr, pte_t *ptep)
382 /* Just return the pte as-is. We preserve the bits on commit */
386 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
387 pte_t *ptep, pte_t pte)
393 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
394 u.val = pte_val_ma(pte);
395 xen_extend_mmu_update(&u);
397 ADD_STATS(prot_commit, 1);
398 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
400 xen_mc_issue(PARAVIRT_LAZY_MMU);
403 /* Assume pteval_t is equivalent to all the other *val_t types. */
404 static pteval_t pte_mfn_to_pfn(pteval_t val)
406 if (val & _PAGE_PRESENT) {
407 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
408 pteval_t flags = val & PTE_FLAGS_MASK;
409 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
415 static pteval_t pte_pfn_to_mfn(pteval_t val)
417 if (val & _PAGE_PRESENT) {
418 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
419 pteval_t flags = val & PTE_FLAGS_MASK;
422 if (!xen_feature(XENFEAT_auto_translated_physmap))
423 mfn = get_phys_to_machine(pfn);
427 * If there's no mfn for the pfn, then just create an
428 * empty non-present pte. Unfortunately this loses
429 * information about the original pfn, so
430 * pte_mfn_to_pfn is asymmetric.
432 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
437 * Paramount to do this test _after_ the
438 * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
439 * IDENTITY_FRAME_BIT resolves to true.
441 mfn &= ~FOREIGN_FRAME_BIT;
442 if (mfn & IDENTITY_FRAME_BIT) {
443 mfn &= ~IDENTITY_FRAME_BIT;
444 flags |= _PAGE_IOMAP;
447 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
453 static pteval_t iomap_pte(pteval_t val)
455 if (val & _PAGE_PRESENT) {
456 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
457 pteval_t flags = val & PTE_FLAGS_MASK;
459 /* We assume the pte frame number is a MFN, so
460 just use it as-is. */
461 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
467 pteval_t xen_pte_val(pte_t pte)
469 pteval_t pteval = pte.pte;
471 /* If this is a WC pte, convert back from Xen WC to Linux WC */
472 if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
473 WARN_ON(!pat_enabled);
474 pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
477 if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
480 return pte_mfn_to_pfn(pteval);
482 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
484 pgdval_t xen_pgd_val(pgd_t pgd)
486 return pte_mfn_to_pfn(pgd.pgd);
488 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
491 * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
492 * are reserved for now, to correspond to the Intel-reserved PAT
495 * We expect Linux's PAT set as follows:
497 * Idx PTE flags Linux Xen Default
504 * 6 PAT PCD UC- UC UC-
505 * 7 PAT PCD PWT UC UC UC
508 void xen_set_pat(u64 pat)
510 /* We expect Linux to use a PAT setting of
511 * UC UC- WC WB (ignoring the PAT flag) */
512 WARN_ON(pat != 0x0007010600070106ull);
515 pte_t xen_make_pte(pteval_t pte)
517 phys_addr_t addr = (pte & PTE_PFN_MASK);
519 /* If Linux is trying to set a WC pte, then map to the Xen WC.
520 * If _PAGE_PAT is set, then it probably means it is really
521 * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
522 * things work out OK...
524 * (We should never see kernel mappings with _PAGE_PSE set,
525 * but we could see hugetlbfs mappings, I think.).
527 if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
528 if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
529 pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
533 * Unprivileged domains are allowed to do IOMAPpings for
534 * PCI passthrough, but not map ISA space. The ISA
535 * mappings are just dummy local mappings to keep other
536 * parts of the kernel happy.
538 if (unlikely(pte & _PAGE_IOMAP) &&
539 (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
540 pte = iomap_pte(pte);
543 pte = pte_pfn_to_mfn(pte);
546 return native_make_pte(pte);
548 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
550 pgd_t xen_make_pgd(pgdval_t pgd)
552 pgd = pte_pfn_to_mfn(pgd);
553 return native_make_pgd(pgd);
555 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
557 pmdval_t xen_pmd_val(pmd_t pmd)
559 return pte_mfn_to_pfn(pmd.pmd);
561 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
563 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
571 /* ptr may be ioremapped for 64-bit pagetable setup */
572 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
573 u.val = pud_val_ma(val);
574 xen_extend_mmu_update(&u);
576 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
578 xen_mc_issue(PARAVIRT_LAZY_MMU);
583 void xen_set_pud(pud_t *ptr, pud_t val)
585 ADD_STATS(pud_update, 1);
587 /* If page is not pinned, we can just update the entry
589 if (!xen_page_pinned(ptr)) {
594 ADD_STATS(pud_update_pinned, 1);
596 xen_set_pud_hyper(ptr, val);
599 void xen_set_pte(pte_t *ptep, pte_t pte)
601 if (xen_iomap_pte(pte)) {
602 xen_set_iomap_pte(ptep, pte);
606 ADD_STATS(pte_update, 1);
607 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
608 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
610 #ifdef CONFIG_X86_PAE
611 ptep->pte_high = pte.pte_high;
613 ptep->pte_low = pte.pte_low;
619 #ifdef CONFIG_X86_PAE
620 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
622 if (xen_iomap_pte(pte)) {
623 xen_set_iomap_pte(ptep, pte);
627 set_64bit((u64 *)ptep, native_pte_val(pte));
630 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
633 smp_wmb(); /* make sure low gets written first */
637 void xen_pmd_clear(pmd_t *pmdp)
639 set_pmd(pmdp, __pmd(0));
641 #endif /* CONFIG_X86_PAE */
643 pmd_t xen_make_pmd(pmdval_t pmd)
645 pmd = pte_pfn_to_mfn(pmd);
646 return native_make_pmd(pmd);
648 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
650 #if PAGETABLE_LEVELS == 4
651 pudval_t xen_pud_val(pud_t pud)
653 return pte_mfn_to_pfn(pud.pud);
655 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
657 pud_t xen_make_pud(pudval_t pud)
659 pud = pte_pfn_to_mfn(pud);
661 return native_make_pud(pud);
663 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
665 pgd_t *xen_get_user_pgd(pgd_t *pgd)
667 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
668 unsigned offset = pgd - pgd_page;
669 pgd_t *user_ptr = NULL;
671 if (offset < pgd_index(USER_LIMIT)) {
672 struct page *page = virt_to_page(pgd_page);
673 user_ptr = (pgd_t *)page->private;
681 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
685 u.ptr = virt_to_machine(ptr).maddr;
686 u.val = pgd_val_ma(val);
687 xen_extend_mmu_update(&u);
691 * Raw hypercall-based set_pgd, intended for in early boot before
692 * there's a page structure. This implies:
693 * 1. The only existing pagetable is the kernel's
694 * 2. It is always pinned
695 * 3. It has no user pagetable attached to it
697 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
703 __xen_set_pgd_hyper(ptr, val);
705 xen_mc_issue(PARAVIRT_LAZY_MMU);
710 void xen_set_pgd(pgd_t *ptr, pgd_t val)
712 pgd_t *user_ptr = xen_get_user_pgd(ptr);
714 ADD_STATS(pgd_update, 1);
716 /* If page is not pinned, we can just update the entry
718 if (!xen_page_pinned(ptr)) {
721 WARN_ON(xen_page_pinned(user_ptr));
727 ADD_STATS(pgd_update_pinned, 1);
728 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
730 /* If it's pinned, then we can at least batch the kernel and
731 user updates together. */
734 __xen_set_pgd_hyper(ptr, val);
736 __xen_set_pgd_hyper(user_ptr, val);
738 xen_mc_issue(PARAVIRT_LAZY_MMU);
740 #endif /* PAGETABLE_LEVELS == 4 */
743 * (Yet another) pagetable walker. This one is intended for pinning a
744 * pagetable. This means that it walks a pagetable and calls the
745 * callback function on each page it finds making up the page table,
746 * at every level. It walks the entire pagetable, but it only bothers
747 * pinning pte pages which are below limit. In the normal case this
748 * will be STACK_TOP_MAX, but at boot we need to pin up to
751 * For 32-bit the important bit is that we don't pin beyond there,
752 * because then we start getting into Xen's ptes.
754 * For 64-bit, we must skip the Xen hole in the middle of the address
755 * space, just after the big x86-64 virtual hole.
757 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
758 int (*func)(struct mm_struct *mm, struct page *,
763 unsigned hole_low, hole_high;
764 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
765 unsigned pgdidx, pudidx, pmdidx;
767 /* The limit is the last byte to be touched */
769 BUG_ON(limit >= FIXADDR_TOP);
771 if (xen_feature(XENFEAT_auto_translated_physmap))
775 * 64-bit has a great big hole in the middle of the address
776 * space, which contains the Xen mappings. On 32-bit these
777 * will end up making a zero-sized hole and so is a no-op.
779 hole_low = pgd_index(USER_LIMIT);
780 hole_high = pgd_index(PAGE_OFFSET);
782 pgdidx_limit = pgd_index(limit);
784 pudidx_limit = pud_index(limit);
789 pmdidx_limit = pmd_index(limit);
794 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
797 if (pgdidx >= hole_low && pgdidx < hole_high)
800 if (!pgd_val(pgd[pgdidx]))
803 pud = pud_offset(&pgd[pgdidx], 0);
805 if (PTRS_PER_PUD > 1) /* not folded */
806 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
808 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
811 if (pgdidx == pgdidx_limit &&
812 pudidx > pudidx_limit)
815 if (pud_none(pud[pudidx]))
818 pmd = pmd_offset(&pud[pudidx], 0);
820 if (PTRS_PER_PMD > 1) /* not folded */
821 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
823 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
826 if (pgdidx == pgdidx_limit &&
827 pudidx == pudidx_limit &&
828 pmdidx > pmdidx_limit)
831 if (pmd_none(pmd[pmdidx]))
834 pte = pmd_page(pmd[pmdidx]);
835 flush |= (*func)(mm, pte, PT_PTE);
841 /* Do the top level last, so that the callbacks can use it as
842 a cue to do final things like tlb flushes. */
843 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
848 static int xen_pgd_walk(struct mm_struct *mm,
849 int (*func)(struct mm_struct *mm, struct page *,
853 return __xen_pgd_walk(mm, mm->pgd, func, limit);
856 /* If we're using split pte locks, then take the page's lock and
857 return a pointer to it. Otherwise return NULL. */
858 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
860 spinlock_t *ptl = NULL;
862 #if USE_SPLIT_PTLOCKS
863 ptl = __pte_lockptr(page);
864 spin_lock_nest_lock(ptl, &mm->page_table_lock);
870 static void xen_pte_unlock(void *v)
876 static void xen_do_pin(unsigned level, unsigned long pfn)
878 struct mmuext_op *op;
879 struct multicall_space mcs;
881 mcs = __xen_mc_entry(sizeof(*op));
884 op->arg1.mfn = pfn_to_mfn(pfn);
885 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
888 static int xen_pin_page(struct mm_struct *mm, struct page *page,
891 unsigned pgfl = TestSetPagePinned(page);
895 flush = 0; /* already pinned */
896 else if (PageHighMem(page))
897 /* kmaps need flushing if we found an unpinned
901 void *pt = lowmem_page_address(page);
902 unsigned long pfn = page_to_pfn(page);
903 struct multicall_space mcs = __xen_mc_entry(0);
909 * We need to hold the pagetable lock between the time
910 * we make the pagetable RO and when we actually pin
911 * it. If we don't, then other users may come in and
912 * attempt to update the pagetable by writing it,
913 * which will fail because the memory is RO but not
914 * pinned, so Xen won't do the trap'n'emulate.
916 * If we're using split pte locks, we can't hold the
917 * entire pagetable's worth of locks during the
918 * traverse, because we may wrap the preempt count (8
919 * bits). The solution is to mark RO and pin each PTE
920 * page while holding the lock. This means the number
921 * of locks we end up holding is never more than a
922 * batch size (~32 entries, at present).
924 * If we're not using split pte locks, we needn't pin
925 * the PTE pages independently, because we're
926 * protected by the overall pagetable lock.
930 ptl = xen_pte_lock(page, mm);
932 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
933 pfn_pte(pfn, PAGE_KERNEL_RO),
934 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
937 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
939 /* Queue a deferred unlock for when this batch
941 xen_mc_callback(xen_pte_unlock, ptl);
948 /* This is called just after a mm has been created, but it has not
949 been used yet. We need to make sure that its pagetable is all
950 read-only, and can be pinned. */
951 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
955 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
956 /* re-enable interrupts for flushing */
966 pgd_t *user_pgd = xen_get_user_pgd(pgd);
968 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
971 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
972 xen_do_pin(MMUEXT_PIN_L4_TABLE,
973 PFN_DOWN(__pa(user_pgd)));
976 #else /* CONFIG_X86_32 */
977 #ifdef CONFIG_X86_PAE
978 /* Need to make sure unshared kernel PMD is pinnable */
979 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
982 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
983 #endif /* CONFIG_X86_64 */
987 static void xen_pgd_pin(struct mm_struct *mm)
989 __xen_pgd_pin(mm, mm->pgd);
993 * On save, we need to pin all pagetables to make sure they get their
994 * mfns turned into pfns. Search the list for any unpinned pgds and pin
995 * them (unpinned pgds are not currently in use, probably because the
996 * process is under construction or destruction).
998 * Expected to be called in stop_machine() ("equivalent to taking
999 * every spinlock in the system"), so the locking doesn't really
1000 * matter all that much.
1002 void xen_mm_pin_all(void)
1004 unsigned long flags;
1007 spin_lock_irqsave(&pgd_lock, flags);
1009 list_for_each_entry(page, &pgd_list, lru) {
1010 if (!PagePinned(page)) {
1011 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1012 SetPageSavePinned(page);
1016 spin_unlock_irqrestore(&pgd_lock, flags);
1020 * The init_mm pagetable is really pinned as soon as its created, but
1021 * that's before we have page structures to store the bits. So do all
1022 * the book-keeping now.
1024 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1025 enum pt_level level)
1027 SetPagePinned(page);
1031 static void __init xen_mark_init_mm_pinned(void)
1033 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1036 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1037 enum pt_level level)
1039 unsigned pgfl = TestClearPagePinned(page);
1041 if (pgfl && !PageHighMem(page)) {
1042 void *pt = lowmem_page_address(page);
1043 unsigned long pfn = page_to_pfn(page);
1044 spinlock_t *ptl = NULL;
1045 struct multicall_space mcs;
1048 * Do the converse to pin_page. If we're using split
1049 * pte locks, we must be holding the lock for while
1050 * the pte page is unpinned but still RO to prevent
1051 * concurrent updates from seeing it in this
1052 * partially-pinned state.
1054 if (level == PT_PTE) {
1055 ptl = xen_pte_lock(page, mm);
1058 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1061 mcs = __xen_mc_entry(0);
1063 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1064 pfn_pte(pfn, PAGE_KERNEL),
1065 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1068 /* unlock when batch completed */
1069 xen_mc_callback(xen_pte_unlock, ptl);
1073 return 0; /* never need to flush on unpin */
1076 /* Release a pagetables pages back as normal RW */
1077 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1081 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1083 #ifdef CONFIG_X86_64
1085 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1088 xen_do_pin(MMUEXT_UNPIN_TABLE,
1089 PFN_DOWN(__pa(user_pgd)));
1090 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1095 #ifdef CONFIG_X86_PAE
1096 /* Need to make sure unshared kernel PMD is unpinned */
1097 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1101 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1106 static void xen_pgd_unpin(struct mm_struct *mm)
1108 __xen_pgd_unpin(mm, mm->pgd);
1112 * On resume, undo any pinning done at save, so that the rest of the
1113 * kernel doesn't see any unexpected pinned pagetables.
1115 void xen_mm_unpin_all(void)
1117 unsigned long flags;
1120 spin_lock_irqsave(&pgd_lock, flags);
1122 list_for_each_entry(page, &pgd_list, lru) {
1123 if (PageSavePinned(page)) {
1124 BUG_ON(!PagePinned(page));
1125 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1126 ClearPageSavePinned(page);
1130 spin_unlock_irqrestore(&pgd_lock, flags);
1133 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1135 spin_lock(&next->page_table_lock);
1137 spin_unlock(&next->page_table_lock);
1140 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1142 spin_lock(&mm->page_table_lock);
1144 spin_unlock(&mm->page_table_lock);
1149 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1150 we need to repoint it somewhere else before we can unpin it. */
1151 static void drop_other_mm_ref(void *info)
1153 struct mm_struct *mm = info;
1154 struct mm_struct *active_mm;
1156 active_mm = percpu_read(cpu_tlbstate.active_mm);
1158 if (active_mm == mm)
1159 leave_mm(smp_processor_id());
1161 /* If this cpu still has a stale cr3 reference, then make sure
1162 it has been flushed. */
1163 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1164 load_cr3(swapper_pg_dir);
1167 static void xen_drop_mm_ref(struct mm_struct *mm)
1172 if (current->active_mm == mm) {
1173 if (current->mm == mm)
1174 load_cr3(swapper_pg_dir);
1176 leave_mm(smp_processor_id());
1179 /* Get the "official" set of cpus referring to our pagetable. */
1180 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1181 for_each_online_cpu(cpu) {
1182 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1183 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1185 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1189 cpumask_copy(mask, mm_cpumask(mm));
1191 /* It's possible that a vcpu may have a stale reference to our
1192 cr3, because its in lazy mode, and it hasn't yet flushed
1193 its set of pending hypercalls yet. In this case, we can
1194 look at its actual current cr3 value, and force it to flush
1196 for_each_online_cpu(cpu) {
1197 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1198 cpumask_set_cpu(cpu, mask);
1201 if (!cpumask_empty(mask))
1202 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1203 free_cpumask_var(mask);
1206 static void xen_drop_mm_ref(struct mm_struct *mm)
1208 if (current->active_mm == mm)
1209 load_cr3(swapper_pg_dir);
1214 * While a process runs, Xen pins its pagetables, which means that the
1215 * hypervisor forces it to be read-only, and it controls all updates
1216 * to it. This means that all pagetable updates have to go via the
1217 * hypervisor, which is moderately expensive.
1219 * Since we're pulling the pagetable down, we switch to use init_mm,
1220 * unpin old process pagetable and mark it all read-write, which
1221 * allows further operations on it to be simple memory accesses.
1223 * The only subtle point is that another CPU may be still using the
1224 * pagetable because of lazy tlb flushing. This means we need need to
1225 * switch all CPUs off this pagetable before we can unpin it.
1227 void xen_exit_mmap(struct mm_struct *mm)
1229 get_cpu(); /* make sure we don't move around */
1230 xen_drop_mm_ref(mm);
1233 spin_lock(&mm->page_table_lock);
1235 /* pgd may not be pinned in the error exit path of execve */
1236 if (xen_page_pinned(mm->pgd))
1239 spin_unlock(&mm->page_table_lock);
1242 static __init void xen_pagetable_setup_start(pgd_t *base)
1246 static void xen_post_allocator_init(void);
1248 static __init void xen_pagetable_setup_done(pgd_t *base)
1250 xen_setup_shared_info();
1251 xen_post_allocator_init();
1254 static void xen_write_cr2(unsigned long cr2)
1256 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1259 static unsigned long xen_read_cr2(void)
1261 return percpu_read(xen_vcpu)->arch.cr2;
1264 unsigned long xen_read_cr2_direct(void)
1266 return percpu_read(xen_vcpu_info.arch.cr2);
1269 static void xen_flush_tlb(void)
1271 struct mmuext_op *op;
1272 struct multicall_space mcs;
1276 mcs = xen_mc_entry(sizeof(*op));
1279 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1280 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1282 xen_mc_issue(PARAVIRT_LAZY_MMU);
1287 static void xen_flush_tlb_single(unsigned long addr)
1289 struct mmuext_op *op;
1290 struct multicall_space mcs;
1294 mcs = xen_mc_entry(sizeof(*op));
1296 op->cmd = MMUEXT_INVLPG_LOCAL;
1297 op->arg1.linear_addr = addr & PAGE_MASK;
1298 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1300 xen_mc_issue(PARAVIRT_LAZY_MMU);
1305 static void xen_flush_tlb_others(const struct cpumask *cpus,
1306 struct mm_struct *mm, unsigned long va)
1309 struct mmuext_op op;
1310 DECLARE_BITMAP(mask, NR_CPUS);
1312 struct multicall_space mcs;
1314 if (cpumask_empty(cpus))
1315 return; /* nothing to do */
1317 mcs = xen_mc_entry(sizeof(*args));
1319 args->op.arg2.vcpumask = to_cpumask(args->mask);
1321 /* Remove us, and any offline CPUS. */
1322 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1323 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1325 if (va == TLB_FLUSH_ALL) {
1326 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1328 args->op.cmd = MMUEXT_INVLPG_MULTI;
1329 args->op.arg1.linear_addr = va;
1332 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1334 xen_mc_issue(PARAVIRT_LAZY_MMU);
1337 static unsigned long xen_read_cr3(void)
1339 return percpu_read(xen_cr3);
1342 static void set_current_cr3(void *v)
1344 percpu_write(xen_current_cr3, (unsigned long)v);
1347 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1349 struct mmuext_op *op;
1350 struct multicall_space mcs;
1354 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1358 WARN_ON(mfn == 0 && kernel);
1360 mcs = __xen_mc_entry(sizeof(*op));
1363 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1366 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1369 percpu_write(xen_cr3, cr3);
1371 /* Update xen_current_cr3 once the batch has actually
1373 xen_mc_callback(set_current_cr3, (void *)cr3);
1377 static void xen_write_cr3(unsigned long cr3)
1379 BUG_ON(preemptible());
1381 xen_mc_batch(); /* disables interrupts */
1383 /* Update while interrupts are disabled, so its atomic with
1385 percpu_write(xen_cr3, cr3);
1387 __xen_write_cr3(true, cr3);
1389 #ifdef CONFIG_X86_64
1391 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1393 __xen_write_cr3(false, __pa(user_pgd));
1395 __xen_write_cr3(false, 0);
1399 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1402 static int xen_pgd_alloc(struct mm_struct *mm)
1404 pgd_t *pgd = mm->pgd;
1407 BUG_ON(PagePinned(virt_to_page(pgd)));
1409 #ifdef CONFIG_X86_64
1411 struct page *page = virt_to_page(pgd);
1414 BUG_ON(page->private != 0);
1418 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1419 page->private = (unsigned long)user_pgd;
1421 if (user_pgd != NULL) {
1422 user_pgd[pgd_index(VSYSCALL_START)] =
1423 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1427 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1434 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1436 #ifdef CONFIG_X86_64
1437 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1440 free_page((unsigned long)user_pgd);
1444 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1446 unsigned long pfn = pte_pfn(pte);
1448 #ifdef CONFIG_X86_32
1449 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1450 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1451 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1456 * If the new pfn is within the range of the newly allocated
1457 * kernel pagetable, and it isn't being mapped into an
1458 * early_ioremap fixmap slot, make sure it is RO.
1460 if (!is_early_ioremap_ptep(ptep) &&
1461 pfn >= e820_table_start && pfn < e820_table_end)
1462 pte = pte_wrprotect(pte);
1467 /* Init-time set_pte while constructing initial pagetables, which
1468 doesn't allow RO pagetable pages to be remapped RW */
1469 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1471 pte = mask_rw_pte(ptep, pte);
1473 xen_set_pte(ptep, pte);
1476 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1478 struct mmuext_op op;
1480 op.arg1.mfn = pfn_to_mfn(pfn);
1481 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1485 /* Early in boot, while setting up the initial pagetable, assume
1486 everything is pinned. */
1487 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1489 #ifdef CONFIG_FLATMEM
1490 BUG_ON(mem_map); /* should only be used early */
1492 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1493 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1496 /* Used for pmd and pud */
1497 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1499 #ifdef CONFIG_FLATMEM
1500 BUG_ON(mem_map); /* should only be used early */
1502 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1505 /* Early release_pte assumes that all pts are pinned, since there's
1506 only init_mm and anything attached to that is pinned. */
1507 static __init void xen_release_pte_init(unsigned long pfn)
1509 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1510 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1513 static __init void xen_release_pmd_init(unsigned long pfn)
1515 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1518 /* This needs to make sure the new pte page is pinned iff its being
1519 attached to a pinned pagetable. */
1520 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1522 struct page *page = pfn_to_page(pfn);
1524 if (PagePinned(virt_to_page(mm->pgd))) {
1525 SetPagePinned(page);
1527 if (!PageHighMem(page)) {
1528 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1529 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1530 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1532 /* make sure there are no stray mappings of
1534 kmap_flush_unused();
1539 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1541 xen_alloc_ptpage(mm, pfn, PT_PTE);
1544 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1546 xen_alloc_ptpage(mm, pfn, PT_PMD);
1549 /* This should never happen until we're OK to use struct page */
1550 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1552 struct page *page = pfn_to_page(pfn);
1554 if (PagePinned(page)) {
1555 if (!PageHighMem(page)) {
1556 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1557 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1558 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1560 ClearPagePinned(page);
1564 static void xen_release_pte(unsigned long pfn)
1566 xen_release_ptpage(pfn, PT_PTE);
1569 static void xen_release_pmd(unsigned long pfn)
1571 xen_release_ptpage(pfn, PT_PMD);
1574 #if PAGETABLE_LEVELS == 4
1575 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1577 xen_alloc_ptpage(mm, pfn, PT_PUD);
1580 static void xen_release_pud(unsigned long pfn)
1582 xen_release_ptpage(pfn, PT_PUD);
1586 void __init xen_reserve_top(void)
1588 #ifdef CONFIG_X86_32
1589 unsigned long top = HYPERVISOR_VIRT_START;
1590 struct xen_platform_parameters pp;
1592 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1593 top = pp.virt_start;
1595 reserve_top_address(-top);
1596 #endif /* CONFIG_X86_32 */
1600 * Like __va(), but returns address in the kernel mapping (which is
1601 * all we have until the physical memory mapping has been set up.
1603 static void *__ka(phys_addr_t paddr)
1605 #ifdef CONFIG_X86_64
1606 return (void *)(paddr + __START_KERNEL_map);
1612 /* Convert a machine address to physical address */
1613 static unsigned long m2p(phys_addr_t maddr)
1617 maddr &= PTE_PFN_MASK;
1618 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1623 /* Convert a machine address to kernel virtual */
1624 static void *m2v(phys_addr_t maddr)
1626 return __ka(m2p(maddr));
1629 /* Set the page permissions on an identity-mapped pages */
1630 static void set_page_prot(void *addr, pgprot_t prot)
1632 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1633 pte_t pte = pfn_pte(pfn, prot);
1635 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1639 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1641 unsigned pmdidx, pteidx;
1645 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1650 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1653 /* Reuse or allocate a page of ptes */
1654 if (pmd_present(pmd[pmdidx]))
1655 pte_page = m2v(pmd[pmdidx].pmd);
1657 /* Check for free pte pages */
1658 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1661 pte_page = &level1_ident_pgt[ident_pte];
1662 ident_pte += PTRS_PER_PTE;
1664 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1667 /* Install mappings */
1668 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1671 if (pfn > max_pfn_mapped)
1672 max_pfn_mapped = pfn;
1674 if (!pte_none(pte_page[pteidx]))
1677 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1678 pte_page[pteidx] = pte;
1682 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1683 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1685 set_page_prot(pmd, PAGE_KERNEL_RO);
1688 void __init xen_setup_machphys_mapping(void)
1690 struct xen_machphys_mapping mapping;
1691 unsigned long machine_to_phys_nr_ents;
1693 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1694 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1695 machine_to_phys_nr_ents = mapping.max_mfn + 1;
1697 machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
1699 machine_to_phys_order = fls(machine_to_phys_nr_ents - 1);
1702 #ifdef CONFIG_X86_64
1703 static void convert_pfn_mfn(void *v)
1708 /* All levels are converted the same way, so just treat them
1710 for (i = 0; i < PTRS_PER_PTE; i++)
1711 pte[i] = xen_make_pte(pte[i].pte);
1715 * Set up the inital kernel pagetable.
1717 * We can construct this by grafting the Xen provided pagetable into
1718 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1719 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1720 * means that only the kernel has a physical mapping to start with -
1721 * but that's enough to get __va working. We need to fill in the rest
1722 * of the physical mapping once some sort of allocator has been set
1725 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1726 unsigned long max_pfn)
1731 /* Zap identity mapping */
1732 init_level4_pgt[0] = __pgd(0);
1734 /* Pre-constructed entries are in pfn, so convert to mfn */
1735 convert_pfn_mfn(init_level4_pgt);
1736 convert_pfn_mfn(level3_ident_pgt);
1737 convert_pfn_mfn(level3_kernel_pgt);
1739 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1740 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1742 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1743 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1745 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1746 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1747 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1749 /* Set up identity map */
1750 xen_map_identity_early(level2_ident_pgt, max_pfn);
1752 /* Make pagetable pieces RO */
1753 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1754 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1755 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1756 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1757 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1758 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1760 /* Pin down new L4 */
1761 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1762 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1764 /* Unpin Xen-provided one */
1765 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1768 pgd = init_level4_pgt;
1771 * At this stage there can be no user pgd, and no page
1772 * structure to attach it to, so make sure we just set kernel
1776 __xen_write_cr3(true, __pa(pgd));
1777 xen_mc_issue(PARAVIRT_LAZY_CPU);
1779 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1780 __pa(xen_start_info->pt_base +
1781 xen_start_info->nr_pt_frames * PAGE_SIZE),
1786 #else /* !CONFIG_X86_64 */
1787 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1788 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1790 static __init void xen_write_cr3_init(unsigned long cr3)
1792 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1794 BUG_ON(read_cr3() != __pa(initial_page_table));
1795 BUG_ON(cr3 != __pa(swapper_pg_dir));
1798 * We are switching to swapper_pg_dir for the first time (from
1799 * initial_page_table) and therefore need to mark that page
1800 * read-only and then pin it.
1802 * Xen disallows sharing of kernel PMDs for PAE
1803 * guests. Therefore we must copy the kernel PMD from
1804 * initial_page_table into a new kernel PMD to be used in
1807 swapper_kernel_pmd =
1808 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1809 memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1810 sizeof(pmd_t) * PTRS_PER_PMD);
1811 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1812 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1813 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1815 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1817 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1819 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1820 PFN_DOWN(__pa(initial_page_table)));
1821 set_page_prot(initial_page_table, PAGE_KERNEL);
1822 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1824 pv_mmu_ops.write_cr3 = &xen_write_cr3;
1827 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1828 unsigned long max_pfn)
1832 initial_kernel_pmd =
1833 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1835 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1836 xen_start_info->nr_pt_frames * PAGE_SIZE +
1839 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1840 memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1842 xen_map_identity_early(initial_kernel_pmd, max_pfn);
1844 memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1845 initial_page_table[KERNEL_PGD_BOUNDARY] =
1846 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1848 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1849 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1850 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1852 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1854 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1855 PFN_DOWN(__pa(initial_page_table)));
1856 xen_write_cr3(__pa(initial_page_table));
1858 memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1859 __pa(xen_start_info->pt_base +
1860 xen_start_info->nr_pt_frames * PAGE_SIZE),
1863 return initial_page_table;
1865 #endif /* CONFIG_X86_64 */
1867 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1869 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1873 phys >>= PAGE_SHIFT;
1876 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1877 #ifdef CONFIG_X86_F00F_BUG
1880 #ifdef CONFIG_X86_32
1883 # ifdef CONFIG_HIGHMEM
1884 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1887 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1889 case FIX_TEXT_POKE0:
1890 case FIX_TEXT_POKE1:
1891 /* All local page mappings */
1892 pte = pfn_pte(phys, prot);
1895 #ifdef CONFIG_X86_LOCAL_APIC
1896 case FIX_APIC_BASE: /* maps dummy local APIC */
1897 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1901 #ifdef CONFIG_X86_IO_APIC
1902 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1904 * We just don't map the IO APIC - all access is via
1905 * hypercalls. Keep the address in the pte for reference.
1907 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1911 case FIX_PARAVIRT_BOOTMAP:
1912 /* This is an MFN, but it isn't an IO mapping from the
1914 pte = mfn_pte(phys, prot);
1918 /* By default, set_fixmap is used for hardware mappings */
1919 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1923 __native_set_fixmap(idx, pte);
1925 #ifdef CONFIG_X86_64
1926 /* Replicate changes to map the vsyscall page into the user
1927 pagetable vsyscall mapping. */
1928 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1929 unsigned long vaddr = __fix_to_virt(idx);
1930 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1935 __init void xen_ident_map_ISA(void)
1940 * If we're dom0, then linear map the ISA machine addresses into
1941 * the kernel's address space.
1943 if (!xen_initial_domain())
1946 xen_raw_printk("Xen: setup ISA identity maps\n");
1948 for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1949 pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1951 if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1958 static __init void xen_post_allocator_init(void)
1960 pv_mmu_ops.set_pte = xen_set_pte;
1961 pv_mmu_ops.set_pmd = xen_set_pmd;
1962 pv_mmu_ops.set_pud = xen_set_pud;
1963 #if PAGETABLE_LEVELS == 4
1964 pv_mmu_ops.set_pgd = xen_set_pgd;
1967 /* This will work as long as patching hasn't happened yet
1968 (which it hasn't) */
1969 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1970 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1971 pv_mmu_ops.release_pte = xen_release_pte;
1972 pv_mmu_ops.release_pmd = xen_release_pmd;
1973 #if PAGETABLE_LEVELS == 4
1974 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1975 pv_mmu_ops.release_pud = xen_release_pud;
1978 #ifdef CONFIG_X86_64
1979 SetPagePinned(virt_to_page(level3_user_vsyscall));
1981 xen_mark_init_mm_pinned();
1984 static void xen_leave_lazy_mmu(void)
1988 paravirt_leave_lazy_mmu();
1992 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1993 .read_cr2 = xen_read_cr2,
1994 .write_cr2 = xen_write_cr2,
1996 .read_cr3 = xen_read_cr3,
1997 #ifdef CONFIG_X86_32
1998 .write_cr3 = xen_write_cr3_init,
2000 .write_cr3 = xen_write_cr3,
2003 .flush_tlb_user = xen_flush_tlb,
2004 .flush_tlb_kernel = xen_flush_tlb,
2005 .flush_tlb_single = xen_flush_tlb_single,
2006 .flush_tlb_others = xen_flush_tlb_others,
2008 .pte_update = paravirt_nop,
2009 .pte_update_defer = paravirt_nop,
2011 .pgd_alloc = xen_pgd_alloc,
2012 .pgd_free = xen_pgd_free,
2014 .alloc_pte = xen_alloc_pte_init,
2015 .release_pte = xen_release_pte_init,
2016 .alloc_pmd = xen_alloc_pmd_init,
2017 .release_pmd = xen_release_pmd_init,
2019 .set_pte = xen_set_pte_init,
2020 .set_pte_at = xen_set_pte_at,
2021 .set_pmd = xen_set_pmd_hyper,
2023 .ptep_modify_prot_start = __ptep_modify_prot_start,
2024 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2026 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2027 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2029 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2030 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2032 #ifdef CONFIG_X86_PAE
2033 .set_pte_atomic = xen_set_pte_atomic,
2034 .pte_clear = xen_pte_clear,
2035 .pmd_clear = xen_pmd_clear,
2036 #endif /* CONFIG_X86_PAE */
2037 .set_pud = xen_set_pud_hyper,
2039 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2040 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2042 #if PAGETABLE_LEVELS == 4
2043 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2044 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2045 .set_pgd = xen_set_pgd_hyper,
2047 .alloc_pud = xen_alloc_pmd_init,
2048 .release_pud = xen_release_pmd_init,
2049 #endif /* PAGETABLE_LEVELS == 4 */
2051 .activate_mm = xen_activate_mm,
2052 .dup_mmap = xen_dup_mmap,
2053 .exit_mmap = xen_exit_mmap,
2056 .enter = paravirt_enter_lazy_mmu,
2057 .leave = xen_leave_lazy_mmu,
2060 .set_fixmap = xen_set_fixmap,
2063 void __init xen_init_mmu_ops(void)
2065 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2066 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2067 pv_mmu_ops = xen_mmu_ops;
2069 memset(dummy_mapping, 0xff, PAGE_SIZE);
2072 /* Protected by xen_reservation_lock. */
2073 #define MAX_CONTIG_ORDER 9 /* 2MB */
2074 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2076 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2077 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2078 unsigned long *in_frames,
2079 unsigned long *out_frames)
2082 struct multicall_space mcs;
2085 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2086 mcs = __xen_mc_entry(0);
2089 in_frames[i] = virt_to_mfn(vaddr);
2091 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2092 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2095 out_frames[i] = virt_to_pfn(vaddr);
2101 * Update the pfn-to-mfn mappings for a virtual address range, either to
2102 * point to an array of mfns, or contiguously from a single starting
2105 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2106 unsigned long *mfns,
2107 unsigned long first_mfn)
2114 limit = 1u << order;
2115 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2116 struct multicall_space mcs;
2119 mcs = __xen_mc_entry(0);
2123 mfn = first_mfn + i;
2125 if (i < (limit - 1))
2129 flags = UVMF_INVLPG | UVMF_ALL;
2131 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2134 MULTI_update_va_mapping(mcs.mc, vaddr,
2135 mfn_pte(mfn, PAGE_KERNEL), flags);
2137 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2144 * Perform the hypercall to exchange a region of our pfns to point to
2145 * memory with the required contiguous alignment. Takes the pfns as
2146 * input, and populates mfns as output.
2148 * Returns a success code indicating whether the hypervisor was able to
2149 * satisfy the request or not.
2151 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2152 unsigned long *pfns_in,
2153 unsigned long extents_out,
2154 unsigned int order_out,
2155 unsigned long *mfns_out,
2156 unsigned int address_bits)
2161 struct xen_memory_exchange exchange = {
2163 .nr_extents = extents_in,
2164 .extent_order = order_in,
2165 .extent_start = pfns_in,
2169 .nr_extents = extents_out,
2170 .extent_order = order_out,
2171 .extent_start = mfns_out,
2172 .address_bits = address_bits,
2177 BUG_ON(extents_in << order_in != extents_out << order_out);
2179 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2180 success = (exchange.nr_exchanged == extents_in);
2182 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2183 BUG_ON(success && (rc != 0));
2188 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2189 unsigned int address_bits)
2191 unsigned long *in_frames = discontig_frames, out_frame;
2192 unsigned long flags;
2196 * Currently an auto-translated guest will not perform I/O, nor will
2197 * it require PAE page directories below 4GB. Therefore any calls to
2198 * this function are redundant and can be ignored.
2201 if (xen_feature(XENFEAT_auto_translated_physmap))
2204 if (unlikely(order > MAX_CONTIG_ORDER))
2207 memset((void *) vstart, 0, PAGE_SIZE << order);
2209 spin_lock_irqsave(&xen_reservation_lock, flags);
2211 /* 1. Zap current PTEs, remembering MFNs. */
2212 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2214 /* 2. Get a new contiguous memory extent. */
2215 out_frame = virt_to_pfn(vstart);
2216 success = xen_exchange_memory(1UL << order, 0, in_frames,
2217 1, order, &out_frame,
2220 /* 3. Map the new extent in place of old pages. */
2222 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2224 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2226 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2228 return success ? 0 : -ENOMEM;
2230 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2232 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2234 unsigned long *out_frames = discontig_frames, in_frame;
2235 unsigned long flags;
2238 if (xen_feature(XENFEAT_auto_translated_physmap))
2241 if (unlikely(order > MAX_CONTIG_ORDER))
2244 memset((void *) vstart, 0, PAGE_SIZE << order);
2246 spin_lock_irqsave(&xen_reservation_lock, flags);
2248 /* 1. Find start MFN of contiguous extent. */
2249 in_frame = virt_to_mfn(vstart);
2251 /* 2. Zap current PTEs. */
2252 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2254 /* 3. Do the exchange for non-contiguous MFNs. */
2255 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2258 /* 4. Map new pages in place of old pages. */
2260 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2262 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2264 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2266 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2268 #ifdef CONFIG_XEN_PVHVM
2269 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2271 struct xen_hvm_pagetable_dying a;
2274 a.domid = DOMID_SELF;
2275 a.gpa = __pa(mm->pgd);
2276 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2277 WARN_ON_ONCE(rc < 0);
2280 static int is_pagetable_dying_supported(void)
2282 struct xen_hvm_pagetable_dying a;
2285 a.domid = DOMID_SELF;
2287 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2289 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2295 void __init xen_hvm_init_mmu_ops(void)
2297 if (is_pagetable_dying_supported())
2298 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2302 #define REMAP_BATCH_SIZE 16
2307 struct mmu_update *mmu_update;
2310 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2311 unsigned long addr, void *data)
2313 struct remap_data *rmd = data;
2314 pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2316 rmd->mmu_update->ptr = arbitrary_virt_to_machine(ptep).maddr;
2317 rmd->mmu_update->val = pte_val_ma(pte);
2323 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2325 unsigned long mfn, int nr,
2326 pgprot_t prot, unsigned domid)
2328 struct remap_data rmd;
2329 struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2331 unsigned long range;
2334 prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2336 BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2337 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2343 batch = min(REMAP_BATCH_SIZE, nr);
2344 range = (unsigned long)batch << PAGE_SHIFT;
2346 rmd.mmu_update = mmu_update;
2347 err = apply_to_page_range(vma->vm_mm, addr, range,
2348 remap_area_mfn_pte_fn, &rmd);
2353 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2367 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2369 #ifdef CONFIG_XEN_DEBUG_FS
2371 static int p2m_dump_open(struct inode *inode, struct file *filp)
2373 return single_open(filp, p2m_dump_show, NULL);
2376 static const struct file_operations p2m_dump_fops = {
2377 .open = p2m_dump_open,
2379 .llseek = seq_lseek,
2380 .release = single_release,
2383 static struct dentry *d_mmu_debug;
2385 static int __init xen_mmu_debugfs(void)
2387 struct dentry *d_xen = xen_init_debugfs();
2392 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
2394 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
2396 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
2397 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
2398 &mmu_stats.pgd_update_pinned);
2399 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
2400 &mmu_stats.pgd_update_pinned);
2402 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2403 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2404 &mmu_stats.pud_update_pinned);
2405 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2406 &mmu_stats.pud_update_pinned);
2408 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2409 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2410 &mmu_stats.pmd_update_pinned);
2411 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2412 &mmu_stats.pmd_update_pinned);
2414 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2415 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2416 // &mmu_stats.pte_update_pinned);
2417 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2418 &mmu_stats.pte_update_pinned);
2420 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2421 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2422 &mmu_stats.mmu_update_extended);
2423 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2424 mmu_stats.mmu_update_histo, 20);
2426 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2427 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2428 &mmu_stats.set_pte_at_batched);
2429 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2430 &mmu_stats.set_pte_at_current);
2431 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2432 &mmu_stats.set_pte_at_kernel);
2434 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2435 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2436 &mmu_stats.prot_commit_batched);
2438 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
2441 fs_initcall(xen_mmu_debugfs);
2443 #endif /* CONFIG_XEN_DEBUG_FS */