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/module.h>
46 #include <linux/gfp.h>
48 #include <asm/pgtable.h>
49 #include <asm/tlbflush.h>
50 #include <asm/fixmap.h>
51 #include <asm/mmu_context.h>
52 #include <asm/setup.h>
53 #include <asm/paravirt.h>
54 #include <asm/linkage.h>
56 #include <asm/xen/hypercall.h>
57 #include <asm/xen/hypervisor.h>
60 #include <xen/interface/xen.h>
61 #include <xen/interface/hvm/hvm_op.h>
62 #include <xen/interface/version.h>
63 #include <xen/hvc-console.h>
65 #include "multicalls.h"
69 #define MMU_UPDATE_HISTO 30
71 #ifdef CONFIG_XEN_DEBUG_FS
75 u32 pgd_update_pinned;
76 u32 pgd_update_batched;
79 u32 pud_update_pinned;
80 u32 pud_update_batched;
83 u32 pmd_update_pinned;
84 u32 pmd_update_batched;
87 u32 pte_update_pinned;
88 u32 pte_update_batched;
91 u32 mmu_update_extended;
92 u32 mmu_update_histo[MMU_UPDATE_HISTO];
95 u32 prot_commit_batched;
98 u32 set_pte_at_batched;
99 u32 set_pte_at_pinned;
100 u32 set_pte_at_current;
101 u32 set_pte_at_kernel;
104 static u8 zero_stats;
106 static inline void check_zero(void)
108 if (unlikely(zero_stats)) {
109 memset(&mmu_stats, 0, sizeof(mmu_stats));
114 #define ADD_STATS(elem, val) \
115 do { check_zero(); mmu_stats.elem += (val); } while(0)
117 #else /* !CONFIG_XEN_DEBUG_FS */
119 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
121 #endif /* CONFIG_XEN_DEBUG_FS */
125 * Identity map, in addition to plain kernel map. This needs to be
126 * large enough to allocate page table pages to allocate the rest.
127 * Each page can map 2MB.
129 static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss;
132 /* l3 pud for userspace vsyscall mapping */
133 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
134 #endif /* CONFIG_X86_64 */
137 * Note about cr3 (pagetable base) values:
139 * xen_cr3 contains the current logical cr3 value; it contains the
140 * last set cr3. This may not be the current effective cr3, because
141 * its update may be being lazily deferred. However, a vcpu looking
142 * at its own cr3 can use this value knowing that it everything will
143 * be self-consistent.
145 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
146 * hypercall to set the vcpu cr3 is complete (so it may be a little
147 * out of date, but it will never be set early). If one vcpu is
148 * looking at another vcpu's cr3 value, it should use this variable.
150 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
151 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
155 * Just beyond the highest usermode address. STACK_TOP_MAX has a
156 * redzone above it, so round it up to a PGD boundary.
158 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
161 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
162 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
164 /* Placeholder for holes in the address space */
165 static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data =
166 { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL };
168 /* Array of pointers to pages containing p2m entries */
169 static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data =
170 { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] };
172 /* Arrays of p2m arrays expressed in mfns used for save/restore */
173 static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss;
175 static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE]
178 static inline unsigned p2m_top_index(unsigned long pfn)
180 BUG_ON(pfn >= MAX_DOMAIN_PAGES);
181 return pfn / P2M_ENTRIES_PER_PAGE;
184 static inline unsigned p2m_index(unsigned long pfn)
186 return pfn % P2M_ENTRIES_PER_PAGE;
189 /* Build the parallel p2m_top_mfn structures */
190 void xen_build_mfn_list_list(void)
194 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) {
195 unsigned topidx = p2m_top_index(pfn);
197 p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]);
200 for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) {
201 unsigned topidx = idx * P2M_ENTRIES_PER_PAGE;
202 p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]);
206 void xen_setup_mfn_list_list(void)
208 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
210 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
211 virt_to_mfn(p2m_top_mfn_list);
212 HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages;
215 /* Set up p2m_top to point to the domain-builder provided p2m pages */
216 void __init xen_build_dynamic_phys_to_machine(void)
218 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
219 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
222 for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) {
223 unsigned topidx = p2m_top_index(pfn);
225 p2m_top[topidx] = &mfn_list[pfn];
228 xen_build_mfn_list_list();
231 unsigned long get_phys_to_machine(unsigned long pfn)
233 unsigned topidx, idx;
235 if (unlikely(pfn >= MAX_DOMAIN_PAGES))
236 return INVALID_P2M_ENTRY;
238 topidx = p2m_top_index(pfn);
239 idx = p2m_index(pfn);
240 return p2m_top[topidx][idx];
242 EXPORT_SYMBOL_GPL(get_phys_to_machine);
244 /* install a new p2m_top page */
245 bool install_p2mtop_page(unsigned long pfn, unsigned long *p)
247 unsigned topidx = p2m_top_index(pfn);
248 unsigned long **pfnp, *mfnp;
251 pfnp = &p2m_top[topidx];
252 mfnp = &p2m_top_mfn[topidx];
254 for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++)
255 p[i] = INVALID_P2M_ENTRY;
257 if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) {
258 *mfnp = virt_to_mfn(p);
265 static void alloc_p2m(unsigned long pfn)
269 p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL);
272 if (!install_p2mtop_page(pfn, p))
273 free_page((unsigned long)p);
276 /* Try to install p2m mapping; fail if intermediate bits missing */
277 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
279 unsigned topidx, idx;
281 if (unlikely(pfn >= MAX_DOMAIN_PAGES)) {
282 BUG_ON(mfn != INVALID_P2M_ENTRY);
286 topidx = p2m_top_index(pfn);
287 if (p2m_top[topidx] == p2m_missing) {
288 if (mfn == INVALID_P2M_ENTRY)
293 idx = p2m_index(pfn);
294 p2m_top[topidx][idx] = mfn;
299 void set_phys_to_machine(unsigned long pfn, unsigned long mfn)
301 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
302 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
306 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
309 if (!__set_phys_to_machine(pfn, mfn))
314 unsigned long arbitrary_virt_to_mfn(void *vaddr)
316 xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
318 return PFN_DOWN(maddr.maddr);
321 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
323 unsigned long address = (unsigned long)vaddr;
329 * if the PFN is in the linear mapped vaddr range, we can just use
330 * the (quick) virt_to_machine() p2m lookup
332 if (virt_addr_valid(vaddr))
333 return virt_to_machine(vaddr);
335 /* otherwise we have to do a (slower) full page-table walk */
337 pte = lookup_address(address, &level);
339 offset = address & ~PAGE_MASK;
340 return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
343 void make_lowmem_page_readonly(void *vaddr)
346 unsigned long address = (unsigned long)vaddr;
349 pte = lookup_address(address, &level);
352 ptev = pte_wrprotect(*pte);
354 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
358 void make_lowmem_page_readwrite(void *vaddr)
361 unsigned long address = (unsigned long)vaddr;
364 pte = lookup_address(address, &level);
367 ptev = pte_mkwrite(*pte);
369 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
374 static bool xen_page_pinned(void *ptr)
376 struct page *page = virt_to_page(ptr);
378 return PagePinned(page);
381 static void xen_extend_mmu_update(const struct mmu_update *update)
383 struct multicall_space mcs;
384 struct mmu_update *u;
386 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
388 if (mcs.mc != NULL) {
389 ADD_STATS(mmu_update_extended, 1);
390 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1);
394 if (mcs.mc->args[1] < MMU_UPDATE_HISTO)
395 ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1);
397 ADD_STATS(mmu_update_histo[0], 1);
399 ADD_STATS(mmu_update, 1);
400 mcs = __xen_mc_entry(sizeof(*u));
401 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
402 ADD_STATS(mmu_update_histo[1], 1);
409 void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
417 /* ptr may be ioremapped for 64-bit pagetable setup */
418 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
419 u.val = pmd_val_ma(val);
420 xen_extend_mmu_update(&u);
422 ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
424 xen_mc_issue(PARAVIRT_LAZY_MMU);
429 void xen_set_pmd(pmd_t *ptr, pmd_t val)
431 ADD_STATS(pmd_update, 1);
433 /* If page is not pinned, we can just update the entry
435 if (!xen_page_pinned(ptr)) {
440 ADD_STATS(pmd_update_pinned, 1);
442 xen_set_pmd_hyper(ptr, val);
446 * Associate a virtual page frame with a given physical page frame
447 * and protection flags for that frame.
449 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
451 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
454 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
455 pte_t *ptep, pte_t pteval)
457 ADD_STATS(set_pte_at, 1);
458 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
459 ADD_STATS(set_pte_at_current, mm == current->mm);
460 ADD_STATS(set_pte_at_kernel, mm == &init_mm);
462 if (mm == current->mm || mm == &init_mm) {
463 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
464 struct multicall_space mcs;
465 mcs = xen_mc_entry(0);
467 MULTI_update_va_mapping(mcs.mc, addr, pteval, 0);
468 ADD_STATS(set_pte_at_batched, 1);
469 xen_mc_issue(PARAVIRT_LAZY_MMU);
472 if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0)
475 xen_set_pte(ptep, pteval);
480 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
481 unsigned long addr, pte_t *ptep)
483 /* Just return the pte as-is. We preserve the bits on commit */
487 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
488 pte_t *ptep, pte_t pte)
494 u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
495 u.val = pte_val_ma(pte);
496 xen_extend_mmu_update(&u);
498 ADD_STATS(prot_commit, 1);
499 ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
501 xen_mc_issue(PARAVIRT_LAZY_MMU);
504 /* Assume pteval_t is equivalent to all the other *val_t types. */
505 static pteval_t pte_mfn_to_pfn(pteval_t val)
507 if (val & _PAGE_PRESENT) {
508 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
509 pteval_t flags = val & PTE_FLAGS_MASK;
510 val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
516 static pteval_t pte_pfn_to_mfn(pteval_t val)
518 if (val & _PAGE_PRESENT) {
519 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
520 pteval_t flags = val & PTE_FLAGS_MASK;
521 val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags;
527 pteval_t xen_pte_val(pte_t pte)
529 return pte_mfn_to_pfn(pte.pte);
531 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
533 pgdval_t xen_pgd_val(pgd_t pgd)
535 return pte_mfn_to_pfn(pgd.pgd);
537 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
539 pte_t xen_make_pte(pteval_t pte)
541 pte = pte_pfn_to_mfn(pte);
542 return native_make_pte(pte);
544 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
546 pgd_t xen_make_pgd(pgdval_t pgd)
548 pgd = pte_pfn_to_mfn(pgd);
549 return native_make_pgd(pgd);
551 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
553 pmdval_t xen_pmd_val(pmd_t pmd)
555 return pte_mfn_to_pfn(pmd.pmd);
557 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
559 void xen_set_pud_hyper(pud_t *ptr, pud_t val)
567 /* ptr may be ioremapped for 64-bit pagetable setup */
568 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
569 u.val = pud_val_ma(val);
570 xen_extend_mmu_update(&u);
572 ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
574 xen_mc_issue(PARAVIRT_LAZY_MMU);
579 void xen_set_pud(pud_t *ptr, pud_t val)
581 ADD_STATS(pud_update, 1);
583 /* If page is not pinned, we can just update the entry
585 if (!xen_page_pinned(ptr)) {
590 ADD_STATS(pud_update_pinned, 1);
592 xen_set_pud_hyper(ptr, val);
595 void xen_set_pte(pte_t *ptep, pte_t pte)
597 ADD_STATS(pte_update, 1);
598 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
599 ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
601 #ifdef CONFIG_X86_PAE
602 ptep->pte_high = pte.pte_high;
604 ptep->pte_low = pte.pte_low;
610 #ifdef CONFIG_X86_PAE
611 void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
613 set_64bit((u64 *)ptep, native_pte_val(pte));
616 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
619 smp_wmb(); /* make sure low gets written first */
623 void xen_pmd_clear(pmd_t *pmdp)
625 set_pmd(pmdp, __pmd(0));
627 #endif /* CONFIG_X86_PAE */
629 pmd_t xen_make_pmd(pmdval_t pmd)
631 pmd = pte_pfn_to_mfn(pmd);
632 return native_make_pmd(pmd);
634 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
636 #if PAGETABLE_LEVELS == 4
637 pudval_t xen_pud_val(pud_t pud)
639 return pte_mfn_to_pfn(pud.pud);
641 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
643 pud_t xen_make_pud(pudval_t pud)
645 pud = pte_pfn_to_mfn(pud);
647 return native_make_pud(pud);
649 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
651 pgd_t *xen_get_user_pgd(pgd_t *pgd)
653 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
654 unsigned offset = pgd - pgd_page;
655 pgd_t *user_ptr = NULL;
657 if (offset < pgd_index(USER_LIMIT)) {
658 struct page *page = virt_to_page(pgd_page);
659 user_ptr = (pgd_t *)page->private;
667 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
671 u.ptr = virt_to_machine(ptr).maddr;
672 u.val = pgd_val_ma(val);
673 xen_extend_mmu_update(&u);
677 * Raw hypercall-based set_pgd, intended for in early boot before
678 * there's a page structure. This implies:
679 * 1. The only existing pagetable is the kernel's
680 * 2. It is always pinned
681 * 3. It has no user pagetable attached to it
683 void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
689 __xen_set_pgd_hyper(ptr, val);
691 xen_mc_issue(PARAVIRT_LAZY_MMU);
696 void xen_set_pgd(pgd_t *ptr, pgd_t val)
698 pgd_t *user_ptr = xen_get_user_pgd(ptr);
700 ADD_STATS(pgd_update, 1);
702 /* If page is not pinned, we can just update the entry
704 if (!xen_page_pinned(ptr)) {
707 WARN_ON(xen_page_pinned(user_ptr));
713 ADD_STATS(pgd_update_pinned, 1);
714 ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU);
716 /* If it's pinned, then we can at least batch the kernel and
717 user updates together. */
720 __xen_set_pgd_hyper(ptr, val);
722 __xen_set_pgd_hyper(user_ptr, val);
724 xen_mc_issue(PARAVIRT_LAZY_MMU);
726 #endif /* PAGETABLE_LEVELS == 4 */
729 * (Yet another) pagetable walker. This one is intended for pinning a
730 * pagetable. This means that it walks a pagetable and calls the
731 * callback function on each page it finds making up the page table,
732 * at every level. It walks the entire pagetable, but it only bothers
733 * pinning pte pages which are below limit. In the normal case this
734 * will be STACK_TOP_MAX, but at boot we need to pin up to
737 * For 32-bit the important bit is that we don't pin beyond there,
738 * because then we start getting into Xen's ptes.
740 * For 64-bit, we must skip the Xen hole in the middle of the address
741 * space, just after the big x86-64 virtual hole.
743 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
744 int (*func)(struct mm_struct *mm, struct page *,
749 unsigned hole_low, hole_high;
750 unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
751 unsigned pgdidx, pudidx, pmdidx;
753 /* The limit is the last byte to be touched */
755 BUG_ON(limit >= FIXADDR_TOP);
757 if (xen_feature(XENFEAT_auto_translated_physmap))
761 * 64-bit has a great big hole in the middle of the address
762 * space, which contains the Xen mappings. On 32-bit these
763 * will end up making a zero-sized hole and so is a no-op.
765 hole_low = pgd_index(USER_LIMIT);
766 hole_high = pgd_index(PAGE_OFFSET);
768 pgdidx_limit = pgd_index(limit);
770 pudidx_limit = pud_index(limit);
775 pmdidx_limit = pmd_index(limit);
780 for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
783 if (pgdidx >= hole_low && pgdidx < hole_high)
786 if (!pgd_val(pgd[pgdidx]))
789 pud = pud_offset(&pgd[pgdidx], 0);
791 if (PTRS_PER_PUD > 1) /* not folded */
792 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
794 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
797 if (pgdidx == pgdidx_limit &&
798 pudidx > pudidx_limit)
801 if (pud_none(pud[pudidx]))
804 pmd = pmd_offset(&pud[pudidx], 0);
806 if (PTRS_PER_PMD > 1) /* not folded */
807 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
809 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
812 if (pgdidx == pgdidx_limit &&
813 pudidx == pudidx_limit &&
814 pmdidx > pmdidx_limit)
817 if (pmd_none(pmd[pmdidx]))
820 pte = pmd_page(pmd[pmdidx]);
821 flush |= (*func)(mm, pte, PT_PTE);
827 /* Do the top level last, so that the callbacks can use it as
828 a cue to do final things like tlb flushes. */
829 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
834 static int xen_pgd_walk(struct mm_struct *mm,
835 int (*func)(struct mm_struct *mm, struct page *,
839 return __xen_pgd_walk(mm, mm->pgd, func, limit);
842 /* If we're using split pte locks, then take the page's lock and
843 return a pointer to it. Otherwise return NULL. */
844 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
846 spinlock_t *ptl = NULL;
848 #if USE_SPLIT_PTLOCKS
849 ptl = __pte_lockptr(page);
850 spin_lock_nest_lock(ptl, &mm->page_table_lock);
856 static void xen_pte_unlock(void *v)
862 static void xen_do_pin(unsigned level, unsigned long pfn)
864 struct mmuext_op *op;
865 struct multicall_space mcs;
867 mcs = __xen_mc_entry(sizeof(*op));
870 op->arg1.mfn = pfn_to_mfn(pfn);
871 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
874 static int xen_pin_page(struct mm_struct *mm, struct page *page,
877 unsigned pgfl = TestSetPagePinned(page);
881 flush = 0; /* already pinned */
882 else if (PageHighMem(page))
883 /* kmaps need flushing if we found an unpinned
887 void *pt = lowmem_page_address(page);
888 unsigned long pfn = page_to_pfn(page);
889 struct multicall_space mcs = __xen_mc_entry(0);
895 * We need to hold the pagetable lock between the time
896 * we make the pagetable RO and when we actually pin
897 * it. If we don't, then other users may come in and
898 * attempt to update the pagetable by writing it,
899 * which will fail because the memory is RO but not
900 * pinned, so Xen won't do the trap'n'emulate.
902 * If we're using split pte locks, we can't hold the
903 * entire pagetable's worth of locks during the
904 * traverse, because we may wrap the preempt count (8
905 * bits). The solution is to mark RO and pin each PTE
906 * page while holding the lock. This means the number
907 * of locks we end up holding is never more than a
908 * batch size (~32 entries, at present).
910 * If we're not using split pte locks, we needn't pin
911 * the PTE pages independently, because we're
912 * protected by the overall pagetable lock.
916 ptl = xen_pte_lock(page, mm);
918 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
919 pfn_pte(pfn, PAGE_KERNEL_RO),
920 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
923 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
925 /* Queue a deferred unlock for when this batch
927 xen_mc_callback(xen_pte_unlock, ptl);
934 /* This is called just after a mm has been created, but it has not
935 been used yet. We need to make sure that its pagetable is all
936 read-only, and can be pinned. */
937 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
943 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
944 /* re-enable interrupts for flushing */
954 pgd_t *user_pgd = xen_get_user_pgd(pgd);
956 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
959 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
960 xen_do_pin(MMUEXT_PIN_L4_TABLE,
961 PFN_DOWN(__pa(user_pgd)));
964 #else /* CONFIG_X86_32 */
965 #ifdef CONFIG_X86_PAE
966 /* Need to make sure unshared kernel PMD is pinnable */
967 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
970 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
971 #endif /* CONFIG_X86_64 */
975 static void xen_pgd_pin(struct mm_struct *mm)
977 __xen_pgd_pin(mm, mm->pgd);
981 * On save, we need to pin all pagetables to make sure they get their
982 * mfns turned into pfns. Search the list for any unpinned pgds and pin
983 * them (unpinned pgds are not currently in use, probably because the
984 * process is under construction or destruction).
986 * Expected to be called in stop_machine() ("equivalent to taking
987 * every spinlock in the system"), so the locking doesn't really
988 * matter all that much.
990 void xen_mm_pin_all(void)
995 spin_lock_irqsave(&pgd_lock, flags);
997 list_for_each_entry(page, &pgd_list, lru) {
998 if (!PagePinned(page)) {
999 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
1000 SetPageSavePinned(page);
1004 spin_unlock_irqrestore(&pgd_lock, flags);
1008 * The init_mm pagetable is really pinned as soon as its created, but
1009 * that's before we have page structures to store the bits. So do all
1010 * the book-keeping now.
1012 static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page,
1013 enum pt_level level)
1015 SetPagePinned(page);
1019 static void __init xen_mark_init_mm_pinned(void)
1021 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
1024 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
1025 enum pt_level level)
1027 unsigned pgfl = TestClearPagePinned(page);
1029 if (pgfl && !PageHighMem(page)) {
1030 void *pt = lowmem_page_address(page);
1031 unsigned long pfn = page_to_pfn(page);
1032 spinlock_t *ptl = NULL;
1033 struct multicall_space mcs;
1036 * Do the converse to pin_page. If we're using split
1037 * pte locks, we must be holding the lock for while
1038 * the pte page is unpinned but still RO to prevent
1039 * concurrent updates from seeing it in this
1040 * partially-pinned state.
1042 if (level == PT_PTE) {
1043 ptl = xen_pte_lock(page, mm);
1046 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
1049 mcs = __xen_mc_entry(0);
1051 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1052 pfn_pte(pfn, PAGE_KERNEL),
1053 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1056 /* unlock when batch completed */
1057 xen_mc_callback(xen_pte_unlock, ptl);
1061 return 0; /* never need to flush on unpin */
1064 /* Release a pagetables pages back as normal RW */
1065 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1069 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1071 #ifdef CONFIG_X86_64
1073 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1076 xen_do_pin(MMUEXT_UNPIN_TABLE,
1077 PFN_DOWN(__pa(user_pgd)));
1078 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1083 #ifdef CONFIG_X86_PAE
1084 /* Need to make sure unshared kernel PMD is unpinned */
1085 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1089 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1094 static void xen_pgd_unpin(struct mm_struct *mm)
1096 __xen_pgd_unpin(mm, mm->pgd);
1100 * On resume, undo any pinning done at save, so that the rest of the
1101 * kernel doesn't see any unexpected pinned pagetables.
1103 void xen_mm_unpin_all(void)
1105 unsigned long flags;
1108 spin_lock_irqsave(&pgd_lock, flags);
1110 list_for_each_entry(page, &pgd_list, lru) {
1111 if (PageSavePinned(page)) {
1112 BUG_ON(!PagePinned(page));
1113 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1114 ClearPageSavePinned(page);
1118 spin_unlock_irqrestore(&pgd_lock, flags);
1121 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1123 spin_lock(&next->page_table_lock);
1125 spin_unlock(&next->page_table_lock);
1128 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1130 spin_lock(&mm->page_table_lock);
1132 spin_unlock(&mm->page_table_lock);
1137 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1138 we need to repoint it somewhere else before we can unpin it. */
1139 static void drop_other_mm_ref(void *info)
1141 struct mm_struct *mm = info;
1142 struct mm_struct *active_mm;
1144 active_mm = percpu_read(cpu_tlbstate.active_mm);
1146 if (active_mm == mm)
1147 leave_mm(smp_processor_id());
1149 /* If this cpu still has a stale cr3 reference, then make sure
1150 it has been flushed. */
1151 if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1152 load_cr3(swapper_pg_dir);
1155 static void xen_drop_mm_ref(struct mm_struct *mm)
1160 if (current->active_mm == mm) {
1161 if (current->mm == mm)
1162 load_cr3(swapper_pg_dir);
1164 leave_mm(smp_processor_id());
1167 /* Get the "official" set of cpus referring to our pagetable. */
1168 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1169 for_each_online_cpu(cpu) {
1170 if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1171 && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1173 smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1177 cpumask_copy(mask, mm_cpumask(mm));
1179 /* It's possible that a vcpu may have a stale reference to our
1180 cr3, because its in lazy mode, and it hasn't yet flushed
1181 its set of pending hypercalls yet. In this case, we can
1182 look at its actual current cr3 value, and force it to flush
1184 for_each_online_cpu(cpu) {
1185 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1186 cpumask_set_cpu(cpu, mask);
1189 if (!cpumask_empty(mask))
1190 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1191 free_cpumask_var(mask);
1194 static void xen_drop_mm_ref(struct mm_struct *mm)
1196 if (current->active_mm == mm)
1197 load_cr3(swapper_pg_dir);
1202 * While a process runs, Xen pins its pagetables, which means that the
1203 * hypervisor forces it to be read-only, and it controls all updates
1204 * to it. This means that all pagetable updates have to go via the
1205 * hypervisor, which is moderately expensive.
1207 * Since we're pulling the pagetable down, we switch to use init_mm,
1208 * unpin old process pagetable and mark it all read-write, which
1209 * allows further operations on it to be simple memory accesses.
1211 * The only subtle point is that another CPU may be still using the
1212 * pagetable because of lazy tlb flushing. This means we need need to
1213 * switch all CPUs off this pagetable before we can unpin it.
1215 void xen_exit_mmap(struct mm_struct *mm)
1217 get_cpu(); /* make sure we don't move around */
1218 xen_drop_mm_ref(mm);
1221 spin_lock(&mm->page_table_lock);
1223 /* pgd may not be pinned in the error exit path of execve */
1224 if (xen_page_pinned(mm->pgd))
1227 spin_unlock(&mm->page_table_lock);
1230 static __init void xen_pagetable_setup_start(pgd_t *base)
1234 static void xen_post_allocator_init(void);
1236 static __init void xen_pagetable_setup_done(pgd_t *base)
1238 xen_setup_shared_info();
1239 xen_post_allocator_init();
1242 static void xen_write_cr2(unsigned long cr2)
1244 percpu_read(xen_vcpu)->arch.cr2 = cr2;
1247 static unsigned long xen_read_cr2(void)
1249 return percpu_read(xen_vcpu)->arch.cr2;
1252 unsigned long xen_read_cr2_direct(void)
1254 return percpu_read(xen_vcpu_info.arch.cr2);
1257 static void xen_flush_tlb(void)
1259 struct mmuext_op *op;
1260 struct multicall_space mcs;
1264 mcs = xen_mc_entry(sizeof(*op));
1267 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1268 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1270 xen_mc_issue(PARAVIRT_LAZY_MMU);
1275 static void xen_flush_tlb_single(unsigned long addr)
1277 struct mmuext_op *op;
1278 struct multicall_space mcs;
1282 mcs = xen_mc_entry(sizeof(*op));
1284 op->cmd = MMUEXT_INVLPG_LOCAL;
1285 op->arg1.linear_addr = addr & PAGE_MASK;
1286 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1288 xen_mc_issue(PARAVIRT_LAZY_MMU);
1293 static void xen_flush_tlb_others(const struct cpumask *cpus,
1294 struct mm_struct *mm, unsigned long va)
1297 struct mmuext_op op;
1298 DECLARE_BITMAP(mask, NR_CPUS);
1300 struct multicall_space mcs;
1302 if (cpumask_empty(cpus))
1303 return; /* nothing to do */
1305 mcs = xen_mc_entry(sizeof(*args));
1307 args->op.arg2.vcpumask = to_cpumask(args->mask);
1309 /* Remove us, and any offline CPUS. */
1310 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1311 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1313 if (va == TLB_FLUSH_ALL) {
1314 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1316 args->op.cmd = MMUEXT_INVLPG_MULTI;
1317 args->op.arg1.linear_addr = va;
1320 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1322 xen_mc_issue(PARAVIRT_LAZY_MMU);
1325 static unsigned long xen_read_cr3(void)
1327 return percpu_read(xen_cr3);
1330 static void set_current_cr3(void *v)
1332 percpu_write(xen_current_cr3, (unsigned long)v);
1335 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1337 struct mmuext_op *op;
1338 struct multicall_space mcs;
1342 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1346 WARN_ON(mfn == 0 && kernel);
1348 mcs = __xen_mc_entry(sizeof(*op));
1351 op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1354 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1357 percpu_write(xen_cr3, cr3);
1359 /* Update xen_current_cr3 once the batch has actually
1361 xen_mc_callback(set_current_cr3, (void *)cr3);
1365 static void xen_write_cr3(unsigned long cr3)
1367 BUG_ON(preemptible());
1369 xen_mc_batch(); /* disables interrupts */
1371 /* Update while interrupts are disabled, so its atomic with
1373 percpu_write(xen_cr3, cr3);
1375 __xen_write_cr3(true, cr3);
1377 #ifdef CONFIG_X86_64
1379 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1381 __xen_write_cr3(false, __pa(user_pgd));
1383 __xen_write_cr3(false, 0);
1387 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1390 static int xen_pgd_alloc(struct mm_struct *mm)
1392 pgd_t *pgd = mm->pgd;
1395 BUG_ON(PagePinned(virt_to_page(pgd)));
1397 #ifdef CONFIG_X86_64
1399 struct page *page = virt_to_page(pgd);
1402 BUG_ON(page->private != 0);
1406 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1407 page->private = (unsigned long)user_pgd;
1409 if (user_pgd != NULL) {
1410 user_pgd[pgd_index(VSYSCALL_START)] =
1411 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1415 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1422 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1424 #ifdef CONFIG_X86_64
1425 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1428 free_page((unsigned long)user_pgd);
1432 #ifdef CONFIG_X86_32
1433 static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte)
1435 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1436 if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1437 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1443 /* Init-time set_pte while constructing initial pagetables, which
1444 doesn't allow RO pagetable pages to be remapped RW */
1445 static __init void xen_set_pte_init(pte_t *ptep, pte_t pte)
1447 pte = mask_rw_pte(ptep, pte);
1449 xen_set_pte(ptep, pte);
1453 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1455 struct mmuext_op op;
1457 op.arg1.mfn = pfn_to_mfn(pfn);
1458 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1462 /* Early in boot, while setting up the initial pagetable, assume
1463 everything is pinned. */
1464 static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1466 #ifdef CONFIG_FLATMEM
1467 BUG_ON(mem_map); /* should only be used early */
1469 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1470 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1473 /* Used for pmd and pud */
1474 static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1476 #ifdef CONFIG_FLATMEM
1477 BUG_ON(mem_map); /* should only be used early */
1479 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1482 /* Early release_pte assumes that all pts are pinned, since there's
1483 only init_mm and anything attached to that is pinned. */
1484 static __init void xen_release_pte_init(unsigned long pfn)
1486 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1487 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1490 static __init void xen_release_pmd_init(unsigned long pfn)
1492 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1495 /* This needs to make sure the new pte page is pinned iff its being
1496 attached to a pinned pagetable. */
1497 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1499 struct page *page = pfn_to_page(pfn);
1501 if (PagePinned(virt_to_page(mm->pgd))) {
1502 SetPagePinned(page);
1505 if (!PageHighMem(page)) {
1506 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1507 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1508 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1510 /* make sure there are no stray mappings of
1512 kmap_flush_unused();
1517 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1519 xen_alloc_ptpage(mm, pfn, PT_PTE);
1522 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1524 xen_alloc_ptpage(mm, pfn, PT_PMD);
1527 /* This should never happen until we're OK to use struct page */
1528 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1530 struct page *page = pfn_to_page(pfn);
1532 if (PagePinned(page)) {
1533 if (!PageHighMem(page)) {
1534 if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1535 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1536 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1538 ClearPagePinned(page);
1542 static void xen_release_pte(unsigned long pfn)
1544 xen_release_ptpage(pfn, PT_PTE);
1547 static void xen_release_pmd(unsigned long pfn)
1549 xen_release_ptpage(pfn, PT_PMD);
1552 #if PAGETABLE_LEVELS == 4
1553 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1555 xen_alloc_ptpage(mm, pfn, PT_PUD);
1558 static void xen_release_pud(unsigned long pfn)
1560 xen_release_ptpage(pfn, PT_PUD);
1564 void __init xen_reserve_top(void)
1566 #ifdef CONFIG_X86_32
1567 unsigned long top = HYPERVISOR_VIRT_START;
1568 struct xen_platform_parameters pp;
1570 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1571 top = pp.virt_start;
1573 reserve_top_address(-top);
1574 #endif /* CONFIG_X86_32 */
1578 * Like __va(), but returns address in the kernel mapping (which is
1579 * all we have until the physical memory mapping has been set up.
1581 static void *__ka(phys_addr_t paddr)
1583 #ifdef CONFIG_X86_64
1584 return (void *)(paddr + __START_KERNEL_map);
1590 /* Convert a machine address to physical address */
1591 static unsigned long m2p(phys_addr_t maddr)
1595 maddr &= PTE_PFN_MASK;
1596 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1601 /* Convert a machine address to kernel virtual */
1602 static void *m2v(phys_addr_t maddr)
1604 return __ka(m2p(maddr));
1607 static void set_page_prot(void *addr, pgprot_t prot)
1609 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1610 pte_t pte = pfn_pte(pfn, prot);
1612 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1616 static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1618 unsigned pmdidx, pteidx;
1624 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1627 /* Reuse or allocate a page of ptes */
1628 if (pmd_present(pmd[pmdidx]))
1629 pte_page = m2v(pmd[pmdidx].pmd);
1631 /* Check for free pte pages */
1632 if (ident_pte == ARRAY_SIZE(level1_ident_pgt))
1635 pte_page = &level1_ident_pgt[ident_pte];
1636 ident_pte += PTRS_PER_PTE;
1638 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1641 /* Install mappings */
1642 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1645 if (pfn > max_pfn_mapped)
1646 max_pfn_mapped = pfn;
1648 if (!pte_none(pte_page[pteidx]))
1651 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1652 pte_page[pteidx] = pte;
1656 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1657 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1659 set_page_prot(pmd, PAGE_KERNEL_RO);
1662 #ifdef CONFIG_X86_64
1663 static void convert_pfn_mfn(void *v)
1668 /* All levels are converted the same way, so just treat them
1670 for (i = 0; i < PTRS_PER_PTE; i++)
1671 pte[i] = xen_make_pte(pte[i].pte);
1675 * Set up the inital kernel pagetable.
1677 * We can construct this by grafting the Xen provided pagetable into
1678 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1679 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1680 * means that only the kernel has a physical mapping to start with -
1681 * but that's enough to get __va working. We need to fill in the rest
1682 * of the physical mapping once some sort of allocator has been set
1685 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1686 unsigned long max_pfn)
1691 /* Zap identity mapping */
1692 init_level4_pgt[0] = __pgd(0);
1694 /* Pre-constructed entries are in pfn, so convert to mfn */
1695 convert_pfn_mfn(init_level4_pgt);
1696 convert_pfn_mfn(level3_ident_pgt);
1697 convert_pfn_mfn(level3_kernel_pgt);
1699 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1700 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1702 memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1703 memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1705 l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1706 l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1707 memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1709 /* Set up identity map */
1710 xen_map_identity_early(level2_ident_pgt, max_pfn);
1712 /* Make pagetable pieces RO */
1713 set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1714 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1715 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1716 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1717 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1718 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1720 /* Pin down new L4 */
1721 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1722 PFN_DOWN(__pa_symbol(init_level4_pgt)));
1724 /* Unpin Xen-provided one */
1725 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1728 pgd = init_level4_pgt;
1731 * At this stage there can be no user pgd, and no page
1732 * structure to attach it to, so make sure we just set kernel
1736 __xen_write_cr3(true, __pa(pgd));
1737 xen_mc_issue(PARAVIRT_LAZY_CPU);
1739 reserve_early(__pa(xen_start_info->pt_base),
1740 __pa(xen_start_info->pt_base +
1741 xen_start_info->nr_pt_frames * PAGE_SIZE),
1746 #else /* !CONFIG_X86_64 */
1747 static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss;
1749 __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd,
1750 unsigned long max_pfn)
1754 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1755 xen_start_info->nr_pt_frames * PAGE_SIZE +
1758 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1759 memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1761 xen_map_identity_early(level2_kernel_pgt, max_pfn);
1763 memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1764 set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY],
1765 __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT));
1767 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1768 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1769 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1771 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1773 xen_write_cr3(__pa(swapper_pg_dir));
1775 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir)));
1777 reserve_early(__pa(xen_start_info->pt_base),
1778 __pa(xen_start_info->pt_base +
1779 xen_start_info->nr_pt_frames * PAGE_SIZE),
1782 return swapper_pg_dir;
1784 #endif /* CONFIG_X86_64 */
1786 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1790 phys >>= PAGE_SHIFT;
1793 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1794 #ifdef CONFIG_X86_F00F_BUG
1797 #ifdef CONFIG_X86_32
1800 # ifdef CONFIG_HIGHMEM
1801 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1804 case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1806 #ifdef CONFIG_X86_LOCAL_APIC
1807 case FIX_APIC_BASE: /* maps dummy local APIC */
1809 case FIX_TEXT_POKE0:
1810 case FIX_TEXT_POKE1:
1811 /* All local page mappings */
1812 pte = pfn_pte(phys, prot);
1816 pte = mfn_pte(phys, prot);
1820 __native_set_fixmap(idx, pte);
1822 #ifdef CONFIG_X86_64
1823 /* Replicate changes to map the vsyscall page into the user
1824 pagetable vsyscall mapping. */
1825 if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1826 unsigned long vaddr = __fix_to_virt(idx);
1827 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1832 static __init void xen_post_allocator_init(void)
1834 pv_mmu_ops.set_pte = xen_set_pte;
1835 pv_mmu_ops.set_pmd = xen_set_pmd;
1836 pv_mmu_ops.set_pud = xen_set_pud;
1837 #if PAGETABLE_LEVELS == 4
1838 pv_mmu_ops.set_pgd = xen_set_pgd;
1841 /* This will work as long as patching hasn't happened yet
1842 (which it hasn't) */
1843 pv_mmu_ops.alloc_pte = xen_alloc_pte;
1844 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1845 pv_mmu_ops.release_pte = xen_release_pte;
1846 pv_mmu_ops.release_pmd = xen_release_pmd;
1847 #if PAGETABLE_LEVELS == 4
1848 pv_mmu_ops.alloc_pud = xen_alloc_pud;
1849 pv_mmu_ops.release_pud = xen_release_pud;
1852 #ifdef CONFIG_X86_64
1853 SetPagePinned(virt_to_page(level3_user_vsyscall));
1855 xen_mark_init_mm_pinned();
1858 static void xen_leave_lazy_mmu(void)
1862 paravirt_leave_lazy_mmu();
1866 static const struct pv_mmu_ops xen_mmu_ops __initdata = {
1867 .read_cr2 = xen_read_cr2,
1868 .write_cr2 = xen_write_cr2,
1870 .read_cr3 = xen_read_cr3,
1871 .write_cr3 = xen_write_cr3,
1873 .flush_tlb_user = xen_flush_tlb,
1874 .flush_tlb_kernel = xen_flush_tlb,
1875 .flush_tlb_single = xen_flush_tlb_single,
1876 .flush_tlb_others = xen_flush_tlb_others,
1878 .pte_update = paravirt_nop,
1879 .pte_update_defer = paravirt_nop,
1881 .pgd_alloc = xen_pgd_alloc,
1882 .pgd_free = xen_pgd_free,
1884 .alloc_pte = xen_alloc_pte_init,
1885 .release_pte = xen_release_pte_init,
1886 .alloc_pmd = xen_alloc_pmd_init,
1887 .alloc_pmd_clone = paravirt_nop,
1888 .release_pmd = xen_release_pmd_init,
1890 #ifdef CONFIG_X86_64
1891 .set_pte = xen_set_pte,
1893 .set_pte = xen_set_pte_init,
1895 .set_pte_at = xen_set_pte_at,
1896 .set_pmd = xen_set_pmd_hyper,
1898 .ptep_modify_prot_start = __ptep_modify_prot_start,
1899 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1901 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1902 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1904 .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1905 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1907 #ifdef CONFIG_X86_PAE
1908 .set_pte_atomic = xen_set_pte_atomic,
1909 .pte_clear = xen_pte_clear,
1910 .pmd_clear = xen_pmd_clear,
1911 #endif /* CONFIG_X86_PAE */
1912 .set_pud = xen_set_pud_hyper,
1914 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
1915 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
1917 #if PAGETABLE_LEVELS == 4
1918 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
1919 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
1920 .set_pgd = xen_set_pgd_hyper,
1922 .alloc_pud = xen_alloc_pmd_init,
1923 .release_pud = xen_release_pmd_init,
1924 #endif /* PAGETABLE_LEVELS == 4 */
1926 .activate_mm = xen_activate_mm,
1927 .dup_mmap = xen_dup_mmap,
1928 .exit_mmap = xen_exit_mmap,
1931 .enter = paravirt_enter_lazy_mmu,
1932 .leave = xen_leave_lazy_mmu,
1935 .set_fixmap = xen_set_fixmap,
1938 void __init xen_init_mmu_ops(void)
1940 x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
1941 x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
1942 pv_mmu_ops = xen_mmu_ops;
1945 #ifdef CONFIG_XEN_PVHVM
1946 static void xen_hvm_exit_mmap(struct mm_struct *mm)
1948 struct xen_hvm_pagetable_dying a;
1951 a.domid = DOMID_SELF;
1952 a.gpa = __pa(mm->pgd);
1953 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
1954 WARN_ON_ONCE(rc < 0);
1957 static int is_pagetable_dying_supported(void)
1959 struct xen_hvm_pagetable_dying a;
1962 a.domid = DOMID_SELF;
1964 rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
1966 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
1972 void __init xen_hvm_init_mmu_ops(void)
1974 if (is_pagetable_dying_supported())
1975 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
1979 #ifdef CONFIG_XEN_DEBUG_FS
1981 static struct dentry *d_mmu_debug;
1983 static int __init xen_mmu_debugfs(void)
1985 struct dentry *d_xen = xen_init_debugfs();
1990 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1992 debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats);
1994 debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update);
1995 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug,
1996 &mmu_stats.pgd_update_pinned);
1997 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug,
1998 &mmu_stats.pgd_update_pinned);
2000 debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update);
2001 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug,
2002 &mmu_stats.pud_update_pinned);
2003 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug,
2004 &mmu_stats.pud_update_pinned);
2006 debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update);
2007 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug,
2008 &mmu_stats.pmd_update_pinned);
2009 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug,
2010 &mmu_stats.pmd_update_pinned);
2012 debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update);
2013 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2014 // &mmu_stats.pte_update_pinned);
2015 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug,
2016 &mmu_stats.pte_update_pinned);
2018 debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update);
2019 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug,
2020 &mmu_stats.mmu_update_extended);
2021 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug,
2022 mmu_stats.mmu_update_histo, 20);
2024 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at);
2025 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug,
2026 &mmu_stats.set_pte_at_batched);
2027 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug,
2028 &mmu_stats.set_pte_at_current);
2029 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug,
2030 &mmu_stats.set_pte_at_kernel);
2032 debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit);
2033 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug,
2034 &mmu_stats.prot_commit_batched);
2038 fs_initcall(xen_mmu_debugfs);
2040 #endif /* CONFIG_XEN_DEBUG_FS */