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>
49 #include <asm/pgtable.h>
50 #include <asm/tlbflush.h>
51 #include <asm/fixmap.h>
52 #include <asm/mmu_context.h>
53 #include <asm/setup.h>
54 #include <asm/paravirt.h>
56 #include <asm/linkage.h>
59 #include <asm/xen/hypercall.h>
60 #include <asm/xen/hypervisor.h>
64 #include <xen/interface/xen.h>
65 #include <xen/interface/hvm/hvm_op.h>
66 #include <xen/interface/version.h>
67 #include <xen/interface/memory.h>
68 #include <xen/hvc-console.h>
70 #include "multicalls.h"
74 #define MMU_UPDATE_HISTO 30
77 * Protects atomic reservation decrease/increase against concurrent increases.
78 * Also protects non-atomic updates of current_pages and driver_pages, and
81 DEFINE_SPINLOCK(xen_reservation_lock
);
83 #ifdef CONFIG_XEN_DEBUG_FS
87 u32 pgd_update_pinned
;
88 u32 pgd_update_batched
;
91 u32 pud_update_pinned
;
92 u32 pud_update_batched
;
95 u32 pmd_update_pinned
;
96 u32 pmd_update_batched
;
99 u32 pte_update_pinned
;
100 u32 pte_update_batched
;
103 u32 mmu_update_extended
;
104 u32 mmu_update_histo
[MMU_UPDATE_HISTO
];
107 u32 prot_commit_batched
;
110 u32 set_pte_at_batched
;
111 u32 set_pte_at_pinned
;
112 u32 set_pte_at_current
;
113 u32 set_pte_at_kernel
;
116 static u8 zero_stats
;
118 static inline void check_zero(void)
120 if (unlikely(zero_stats
)) {
121 memset(&mmu_stats
, 0, sizeof(mmu_stats
));
126 #define ADD_STATS(elem, val) \
127 do { check_zero(); mmu_stats.elem += (val); } while(0)
129 #else /* !CONFIG_XEN_DEBUG_FS */
131 #define ADD_STATS(elem, val) do { (void)(val); } while(0)
133 #endif /* CONFIG_XEN_DEBUG_FS */
137 * Identity map, in addition to plain kernel map. This needs to be
138 * large enough to allocate page table pages to allocate the rest.
139 * Each page can map 2MB.
141 static pte_t level1_ident_pgt
[PTRS_PER_PTE
* 4] __page_aligned_bss
;
144 /* l3 pud for userspace vsyscall mapping */
145 static pud_t level3_user_vsyscall
[PTRS_PER_PUD
] __page_aligned_bss
;
146 #endif /* CONFIG_X86_64 */
149 * Note about cr3 (pagetable base) values:
151 * xen_cr3 contains the current logical cr3 value; it contains the
152 * last set cr3. This may not be the current effective cr3, because
153 * its update may be being lazily deferred. However, a vcpu looking
154 * at its own cr3 can use this value knowing that it everything will
155 * be self-consistent.
157 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
158 * hypercall to set the vcpu cr3 is complete (so it may be a little
159 * out of date, but it will never be set early). If one vcpu is
160 * looking at another vcpu's cr3 value, it should use this variable.
162 DEFINE_PER_CPU(unsigned long, xen_cr3
); /* cr3 stored as physaddr */
163 DEFINE_PER_CPU(unsigned long, xen_current_cr3
); /* actual vcpu cr3 */
167 * Just beyond the highest usermode address. STACK_TOP_MAX has a
168 * redzone above it, so round it up to a PGD boundary.
170 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
173 #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long))
174 #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE)
176 /* Placeholder for holes in the address space */
177 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing
, P2M_ENTRIES_PER_PAGE
);
179 /* Array of pointers to pages containing p2m entries */
180 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top
, TOP_ENTRIES
);
182 /* Arrays of p2m arrays expressed in mfns used for save/restore */
183 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn
, TOP_ENTRIES
);
185 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn_list
,
186 (TOP_ENTRIES
/ P2M_ENTRIES_PER_PAGE
));
188 static inline unsigned p2m_top_index(unsigned long pfn
)
190 BUG_ON(pfn
>= MAX_DOMAIN_PAGES
);
191 return pfn
/ P2M_ENTRIES_PER_PAGE
;
194 static inline unsigned p2m_index(unsigned long pfn
)
196 return pfn
% P2M_ENTRIES_PER_PAGE
;
199 /* Build the parallel p2m_top_mfn structures */
200 void xen_build_mfn_list_list(void)
204 for (pfn
= 0; pfn
< MAX_DOMAIN_PAGES
; pfn
+= P2M_ENTRIES_PER_PAGE
) {
205 unsigned topidx
= p2m_top_index(pfn
);
207 p2m_top_mfn
[topidx
] = virt_to_mfn(p2m_top
[topidx
]);
210 for (idx
= 0; idx
< TOP_ENTRIES
/P2M_ENTRIES_PER_PAGE
; idx
++) {
211 unsigned topidx
= idx
* P2M_ENTRIES_PER_PAGE
;
212 p2m_top_mfn_list
[idx
] = virt_to_mfn(&p2m_top_mfn
[topidx
]);
216 void xen_setup_mfn_list_list(void)
218 BUG_ON(HYPERVISOR_shared_info
== &xen_dummy_shared_info
);
220 HYPERVISOR_shared_info
->arch
.pfn_to_mfn_frame_list_list
=
221 virt_to_mfn(p2m_top_mfn_list
);
222 HYPERVISOR_shared_info
->arch
.max_pfn
= xen_start_info
->nr_pages
;
225 /* Set up p2m_top to point to the domain-builder provided p2m pages */
226 void __init
xen_build_dynamic_phys_to_machine(void)
228 unsigned long *mfn_list
= (unsigned long *)xen_start_info
->mfn_list
;
229 unsigned long max_pfn
= min(MAX_DOMAIN_PAGES
, xen_start_info
->nr_pages
);
233 p2m_missing
= extend_brk(sizeof(*p2m_missing
) * P2M_ENTRIES_PER_PAGE
,
235 for (i
= 0; i
< P2M_ENTRIES_PER_PAGE
; i
++)
236 p2m_missing
[i
] = ~0UL;
238 p2m_top
= extend_brk(sizeof(*p2m_top
) * TOP_ENTRIES
,
240 for (i
= 0; i
< TOP_ENTRIES
; i
++)
241 p2m_top
[i
] = p2m_missing
;
243 p2m_top_mfn
= extend_brk(sizeof(*p2m_top_mfn
) * TOP_ENTRIES
, PAGE_SIZE
);
244 p2m_top_mfn_list
= extend_brk(sizeof(*p2m_top_mfn_list
) *
245 (TOP_ENTRIES
/ P2M_ENTRIES_PER_PAGE
),
248 for (pfn
= 0; pfn
< max_pfn
; pfn
+= P2M_ENTRIES_PER_PAGE
) {
249 unsigned topidx
= p2m_top_index(pfn
);
251 p2m_top
[topidx
] = &mfn_list
[pfn
];
254 xen_build_mfn_list_list();
257 unsigned long get_phys_to_machine(unsigned long pfn
)
259 unsigned topidx
, idx
;
261 if (unlikely(pfn
>= MAX_DOMAIN_PAGES
))
262 return INVALID_P2M_ENTRY
;
264 topidx
= p2m_top_index(pfn
);
265 idx
= p2m_index(pfn
);
266 return p2m_top
[topidx
][idx
];
268 EXPORT_SYMBOL_GPL(get_phys_to_machine
);
270 /* install a new p2m_top page */
271 bool install_p2mtop_page(unsigned long pfn
, unsigned long *p
)
273 unsigned topidx
= p2m_top_index(pfn
);
274 unsigned long **pfnp
, *mfnp
;
277 pfnp
= &p2m_top
[topidx
];
278 mfnp
= &p2m_top_mfn
[topidx
];
280 for (i
= 0; i
< P2M_ENTRIES_PER_PAGE
; i
++)
281 p
[i
] = INVALID_P2M_ENTRY
;
283 if (cmpxchg(pfnp
, p2m_missing
, p
) == p2m_missing
) {
284 *mfnp
= virt_to_mfn(p
);
291 static void alloc_p2m(unsigned long pfn
)
295 p
= (void *)__get_free_page(GFP_KERNEL
| __GFP_NOFAIL
);
298 if (!install_p2mtop_page(pfn
, p
))
299 free_page((unsigned long)p
);
302 /* Try to install p2m mapping; fail if intermediate bits missing */
303 bool __set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
305 unsigned topidx
, idx
;
307 if (unlikely(pfn
>= MAX_DOMAIN_PAGES
)) {
308 BUG_ON(mfn
!= INVALID_P2M_ENTRY
);
312 topidx
= p2m_top_index(pfn
);
313 if (p2m_top
[topidx
] == p2m_missing
) {
314 if (mfn
== INVALID_P2M_ENTRY
)
319 idx
= p2m_index(pfn
);
320 p2m_top
[topidx
][idx
] = mfn
;
325 void set_phys_to_machine(unsigned long pfn
, unsigned long mfn
)
327 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap
))) {
328 BUG_ON(pfn
!= mfn
&& mfn
!= INVALID_P2M_ENTRY
);
332 if (unlikely(!__set_phys_to_machine(pfn
, mfn
))) {
335 if (!__set_phys_to_machine(pfn
, mfn
))
340 unsigned long arbitrary_virt_to_mfn(void *vaddr
)
342 xmaddr_t maddr
= arbitrary_virt_to_machine(vaddr
);
344 return PFN_DOWN(maddr
.maddr
);
347 xmaddr_t
arbitrary_virt_to_machine(void *vaddr
)
349 unsigned long address
= (unsigned long)vaddr
;
355 * if the PFN is in the linear mapped vaddr range, we can just use
356 * the (quick) virt_to_machine() p2m lookup
358 if (virt_addr_valid(vaddr
))
359 return virt_to_machine(vaddr
);
361 /* otherwise we have to do a (slower) full page-table walk */
363 pte
= lookup_address(address
, &level
);
365 offset
= address
& ~PAGE_MASK
;
366 return XMADDR(((phys_addr_t
)pte_mfn(*pte
) << PAGE_SHIFT
) + offset
);
369 void make_lowmem_page_readonly(void *vaddr
)
372 unsigned long address
= (unsigned long)vaddr
;
375 pte
= lookup_address(address
, &level
);
378 ptev
= pte_wrprotect(*pte
);
380 if (HYPERVISOR_update_va_mapping(address
, ptev
, 0))
384 void make_lowmem_page_readwrite(void *vaddr
)
387 unsigned long address
= (unsigned long)vaddr
;
390 pte
= lookup_address(address
, &level
);
393 ptev
= pte_mkwrite(*pte
);
395 if (HYPERVISOR_update_va_mapping(address
, ptev
, 0))
400 static bool xen_page_pinned(void *ptr
)
402 struct page
*page
= virt_to_page(ptr
);
404 return PagePinned(page
);
407 static bool xen_iomap_pte(pte_t pte
)
409 return pte_flags(pte
) & _PAGE_IOMAP
;
412 static void xen_set_iomap_pte(pte_t
*ptep
, pte_t pteval
)
414 struct multicall_space mcs
;
415 struct mmu_update
*u
;
417 mcs
= xen_mc_entry(sizeof(*u
));
420 /* ptep might be kmapped when using 32-bit HIGHPTE */
421 u
->ptr
= arbitrary_virt_to_machine(ptep
).maddr
;
422 u
->val
= pte_val_ma(pteval
);
424 MULTI_mmu_update(mcs
.mc
, mcs
.args
, 1, NULL
, DOMID_IO
);
426 xen_mc_issue(PARAVIRT_LAZY_MMU
);
429 static void xen_extend_mmu_update(const struct mmu_update
*update
)
431 struct multicall_space mcs
;
432 struct mmu_update
*u
;
434 mcs
= xen_mc_extend_args(__HYPERVISOR_mmu_update
, sizeof(*u
));
436 if (mcs
.mc
!= NULL
) {
437 ADD_STATS(mmu_update_extended
, 1);
438 ADD_STATS(mmu_update_histo
[mcs
.mc
->args
[1]], -1);
442 if (mcs
.mc
->args
[1] < MMU_UPDATE_HISTO
)
443 ADD_STATS(mmu_update_histo
[mcs
.mc
->args
[1]], 1);
445 ADD_STATS(mmu_update_histo
[0], 1);
447 ADD_STATS(mmu_update
, 1);
448 mcs
= __xen_mc_entry(sizeof(*u
));
449 MULTI_mmu_update(mcs
.mc
, mcs
.args
, 1, NULL
, DOMID_SELF
);
450 ADD_STATS(mmu_update_histo
[1], 1);
457 void xen_set_pmd_hyper(pmd_t
*ptr
, pmd_t val
)
465 /* ptr may be ioremapped for 64-bit pagetable setup */
466 u
.ptr
= arbitrary_virt_to_machine(ptr
).maddr
;
467 u
.val
= pmd_val_ma(val
);
468 xen_extend_mmu_update(&u
);
470 ADD_STATS(pmd_update_batched
, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
);
472 xen_mc_issue(PARAVIRT_LAZY_MMU
);
477 void xen_set_pmd(pmd_t
*ptr
, pmd_t val
)
479 ADD_STATS(pmd_update
, 1);
481 /* If page is not pinned, we can just update the entry
483 if (!xen_page_pinned(ptr
)) {
488 ADD_STATS(pmd_update_pinned
, 1);
490 xen_set_pmd_hyper(ptr
, val
);
494 * Associate a virtual page frame with a given physical page frame
495 * and protection flags for that frame.
497 void set_pte_mfn(unsigned long vaddr
, unsigned long mfn
, pgprot_t flags
)
499 set_pte_vaddr(vaddr
, mfn_pte(mfn
, flags
));
502 void xen_set_pte_at(struct mm_struct
*mm
, unsigned long addr
,
503 pte_t
*ptep
, pte_t pteval
)
505 if (xen_iomap_pte(pteval
)) {
506 xen_set_iomap_pte(ptep
, pteval
);
510 ADD_STATS(set_pte_at
, 1);
511 // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep));
512 ADD_STATS(set_pte_at_current
, mm
== current
->mm
);
513 ADD_STATS(set_pte_at_kernel
, mm
== &init_mm
);
515 if (mm
== current
->mm
|| mm
== &init_mm
) {
516 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
) {
517 struct multicall_space mcs
;
518 mcs
= xen_mc_entry(0);
520 MULTI_update_va_mapping(mcs
.mc
, addr
, pteval
, 0);
521 ADD_STATS(set_pte_at_batched
, 1);
522 xen_mc_issue(PARAVIRT_LAZY_MMU
);
525 if (HYPERVISOR_update_va_mapping(addr
, pteval
, 0) == 0)
528 xen_set_pte(ptep
, pteval
);
533 pte_t
xen_ptep_modify_prot_start(struct mm_struct
*mm
,
534 unsigned long addr
, pte_t
*ptep
)
536 /* Just return the pte as-is. We preserve the bits on commit */
540 void xen_ptep_modify_prot_commit(struct mm_struct
*mm
, unsigned long addr
,
541 pte_t
*ptep
, pte_t pte
)
547 u
.ptr
= arbitrary_virt_to_machine(ptep
).maddr
| MMU_PT_UPDATE_PRESERVE_AD
;
548 u
.val
= pte_val_ma(pte
);
549 xen_extend_mmu_update(&u
);
551 ADD_STATS(prot_commit
, 1);
552 ADD_STATS(prot_commit_batched
, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
);
554 xen_mc_issue(PARAVIRT_LAZY_MMU
);
557 /* Assume pteval_t is equivalent to all the other *val_t types. */
558 static pteval_t
pte_mfn_to_pfn(pteval_t val
)
560 if (val
& _PAGE_PRESENT
) {
561 unsigned long mfn
= (val
& PTE_PFN_MASK
) >> PAGE_SHIFT
;
562 pteval_t flags
= val
& PTE_FLAGS_MASK
;
563 val
= ((pteval_t
)mfn_to_pfn(mfn
) << PAGE_SHIFT
) | flags
;
569 static pteval_t
pte_pfn_to_mfn(pteval_t val
)
571 if (val
& _PAGE_PRESENT
) {
572 unsigned long pfn
= (val
& PTE_PFN_MASK
) >> PAGE_SHIFT
;
573 pteval_t flags
= val
& PTE_FLAGS_MASK
;
574 val
= ((pteval_t
)pfn_to_mfn(pfn
) << PAGE_SHIFT
) | flags
;
580 static pteval_t
iomap_pte(pteval_t val
)
582 if (val
& _PAGE_PRESENT
) {
583 unsigned long pfn
= (val
& PTE_PFN_MASK
) >> PAGE_SHIFT
;
584 pteval_t flags
= val
& PTE_FLAGS_MASK
;
586 /* We assume the pte frame number is a MFN, so
587 just use it as-is. */
588 val
= ((pteval_t
)pfn
<< PAGE_SHIFT
) | flags
;
594 pteval_t
xen_pte_val(pte_t pte
)
596 if (xen_initial_domain() && (pte
.pte
& _PAGE_IOMAP
))
599 return pte_mfn_to_pfn(pte
.pte
);
601 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val
);
603 pgdval_t
xen_pgd_val(pgd_t pgd
)
605 return pte_mfn_to_pfn(pgd
.pgd
);
607 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val
);
609 pte_t
xen_make_pte(pteval_t pte
)
611 phys_addr_t addr
= (pte
& PTE_PFN_MASK
);
614 * Unprivileged domains are allowed to do IOMAPpings for
615 * PCI passthrough, but not map ISA space. The ISA
616 * mappings are just dummy local mappings to keep other
617 * parts of the kernel happy.
619 if (unlikely(pte
& _PAGE_IOMAP
) &&
620 (xen_initial_domain() || addr
>= ISA_END_ADDRESS
)) {
621 pte
= iomap_pte(pte
);
624 pte
= pte_pfn_to_mfn(pte
);
627 return native_make_pte(pte
);
629 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte
);
631 pgd_t
xen_make_pgd(pgdval_t pgd
)
633 pgd
= pte_pfn_to_mfn(pgd
);
634 return native_make_pgd(pgd
);
636 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd
);
638 pmdval_t
xen_pmd_val(pmd_t pmd
)
640 return pte_mfn_to_pfn(pmd
.pmd
);
642 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val
);
644 void xen_set_pud_hyper(pud_t
*ptr
, pud_t val
)
652 /* ptr may be ioremapped for 64-bit pagetable setup */
653 u
.ptr
= arbitrary_virt_to_machine(ptr
).maddr
;
654 u
.val
= pud_val_ma(val
);
655 xen_extend_mmu_update(&u
);
657 ADD_STATS(pud_update_batched
, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
);
659 xen_mc_issue(PARAVIRT_LAZY_MMU
);
664 void xen_set_pud(pud_t
*ptr
, pud_t val
)
666 ADD_STATS(pud_update
, 1);
668 /* If page is not pinned, we can just update the entry
670 if (!xen_page_pinned(ptr
)) {
675 ADD_STATS(pud_update_pinned
, 1);
677 xen_set_pud_hyper(ptr
, val
);
680 void xen_set_pte(pte_t
*ptep
, pte_t pte
)
682 if (xen_iomap_pte(pte
)) {
683 xen_set_iomap_pte(ptep
, pte
);
687 ADD_STATS(pte_update
, 1);
688 // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep));
689 ADD_STATS(pte_update_batched
, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
);
691 #ifdef CONFIG_X86_PAE
692 ptep
->pte_high
= pte
.pte_high
;
694 ptep
->pte_low
= pte
.pte_low
;
700 #ifdef CONFIG_X86_PAE
701 void xen_set_pte_atomic(pte_t
*ptep
, pte_t pte
)
703 if (xen_iomap_pte(pte
)) {
704 xen_set_iomap_pte(ptep
, pte
);
708 set_64bit((u64
*)ptep
, native_pte_val(pte
));
711 void xen_pte_clear(struct mm_struct
*mm
, unsigned long addr
, pte_t
*ptep
)
714 smp_wmb(); /* make sure low gets written first */
718 void xen_pmd_clear(pmd_t
*pmdp
)
720 set_pmd(pmdp
, __pmd(0));
722 #endif /* CONFIG_X86_PAE */
724 pmd_t
xen_make_pmd(pmdval_t pmd
)
726 pmd
= pte_pfn_to_mfn(pmd
);
727 return native_make_pmd(pmd
);
729 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd
);
731 #if PAGETABLE_LEVELS == 4
732 pudval_t
xen_pud_val(pud_t pud
)
734 return pte_mfn_to_pfn(pud
.pud
);
736 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val
);
738 pud_t
xen_make_pud(pudval_t pud
)
740 pud
= pte_pfn_to_mfn(pud
);
742 return native_make_pud(pud
);
744 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud
);
746 pgd_t
*xen_get_user_pgd(pgd_t
*pgd
)
748 pgd_t
*pgd_page
= (pgd_t
*)(((unsigned long)pgd
) & PAGE_MASK
);
749 unsigned offset
= pgd
- pgd_page
;
750 pgd_t
*user_ptr
= NULL
;
752 if (offset
< pgd_index(USER_LIMIT
)) {
753 struct page
*page
= virt_to_page(pgd_page
);
754 user_ptr
= (pgd_t
*)page
->private;
762 static void __xen_set_pgd_hyper(pgd_t
*ptr
, pgd_t val
)
766 u
.ptr
= virt_to_machine(ptr
).maddr
;
767 u
.val
= pgd_val_ma(val
);
768 xen_extend_mmu_update(&u
);
772 * Raw hypercall-based set_pgd, intended for in early boot before
773 * there's a page structure. This implies:
774 * 1. The only existing pagetable is the kernel's
775 * 2. It is always pinned
776 * 3. It has no user pagetable attached to it
778 void __init
xen_set_pgd_hyper(pgd_t
*ptr
, pgd_t val
)
784 __xen_set_pgd_hyper(ptr
, val
);
786 xen_mc_issue(PARAVIRT_LAZY_MMU
);
791 void xen_set_pgd(pgd_t
*ptr
, pgd_t val
)
793 pgd_t
*user_ptr
= xen_get_user_pgd(ptr
);
795 ADD_STATS(pgd_update
, 1);
797 /* If page is not pinned, we can just update the entry
799 if (!xen_page_pinned(ptr
)) {
802 WARN_ON(xen_page_pinned(user_ptr
));
808 ADD_STATS(pgd_update_pinned
, 1);
809 ADD_STATS(pgd_update_batched
, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU
);
811 /* If it's pinned, then we can at least batch the kernel and
812 user updates together. */
815 __xen_set_pgd_hyper(ptr
, val
);
817 __xen_set_pgd_hyper(user_ptr
, val
);
819 xen_mc_issue(PARAVIRT_LAZY_MMU
);
821 #endif /* PAGETABLE_LEVELS == 4 */
824 * (Yet another) pagetable walker. This one is intended for pinning a
825 * pagetable. This means that it walks a pagetable and calls the
826 * callback function on each page it finds making up the page table,
827 * at every level. It walks the entire pagetable, but it only bothers
828 * pinning pte pages which are below limit. In the normal case this
829 * will be STACK_TOP_MAX, but at boot we need to pin up to
832 * For 32-bit the important bit is that we don't pin beyond there,
833 * because then we start getting into Xen's ptes.
835 * For 64-bit, we must skip the Xen hole in the middle of the address
836 * space, just after the big x86-64 virtual hole.
838 static int __xen_pgd_walk(struct mm_struct
*mm
, pgd_t
*pgd
,
839 int (*func
)(struct mm_struct
*mm
, struct page
*,
844 unsigned hole_low
, hole_high
;
845 unsigned pgdidx_limit
, pudidx_limit
, pmdidx_limit
;
846 unsigned pgdidx
, pudidx
, pmdidx
;
848 /* The limit is the last byte to be touched */
850 BUG_ON(limit
>= FIXADDR_TOP
);
852 if (xen_feature(XENFEAT_auto_translated_physmap
))
856 * 64-bit has a great big hole in the middle of the address
857 * space, which contains the Xen mappings. On 32-bit these
858 * will end up making a zero-sized hole and so is a no-op.
860 hole_low
= pgd_index(USER_LIMIT
);
861 hole_high
= pgd_index(PAGE_OFFSET
);
863 pgdidx_limit
= pgd_index(limit
);
865 pudidx_limit
= pud_index(limit
);
870 pmdidx_limit
= pmd_index(limit
);
875 for (pgdidx
= 0; pgdidx
<= pgdidx_limit
; pgdidx
++) {
878 if (pgdidx
>= hole_low
&& pgdidx
< hole_high
)
881 if (!pgd_val(pgd
[pgdidx
]))
884 pud
= pud_offset(&pgd
[pgdidx
], 0);
886 if (PTRS_PER_PUD
> 1) /* not folded */
887 flush
|= (*func
)(mm
, virt_to_page(pud
), PT_PUD
);
889 for (pudidx
= 0; pudidx
< PTRS_PER_PUD
; pudidx
++) {
892 if (pgdidx
== pgdidx_limit
&&
893 pudidx
> pudidx_limit
)
896 if (pud_none(pud
[pudidx
]))
899 pmd
= pmd_offset(&pud
[pudidx
], 0);
901 if (PTRS_PER_PMD
> 1) /* not folded */
902 flush
|= (*func
)(mm
, virt_to_page(pmd
), PT_PMD
);
904 for (pmdidx
= 0; pmdidx
< PTRS_PER_PMD
; pmdidx
++) {
907 if (pgdidx
== pgdidx_limit
&&
908 pudidx
== pudidx_limit
&&
909 pmdidx
> pmdidx_limit
)
912 if (pmd_none(pmd
[pmdidx
]))
915 pte
= pmd_page(pmd
[pmdidx
]);
916 flush
|= (*func
)(mm
, pte
, PT_PTE
);
922 /* Do the top level last, so that the callbacks can use it as
923 a cue to do final things like tlb flushes. */
924 flush
|= (*func
)(mm
, virt_to_page(pgd
), PT_PGD
);
929 static int xen_pgd_walk(struct mm_struct
*mm
,
930 int (*func
)(struct mm_struct
*mm
, struct page
*,
934 return __xen_pgd_walk(mm
, mm
->pgd
, func
, limit
);
937 /* If we're using split pte locks, then take the page's lock and
938 return a pointer to it. Otherwise return NULL. */
939 static spinlock_t
*xen_pte_lock(struct page
*page
, struct mm_struct
*mm
)
941 spinlock_t
*ptl
= NULL
;
943 #if USE_SPLIT_PTLOCKS
944 ptl
= __pte_lockptr(page
);
945 spin_lock_nest_lock(ptl
, &mm
->page_table_lock
);
951 static void xen_pte_unlock(void *v
)
957 static void xen_do_pin(unsigned level
, unsigned long pfn
)
959 struct mmuext_op
*op
;
960 struct multicall_space mcs
;
962 mcs
= __xen_mc_entry(sizeof(*op
));
965 op
->arg1
.mfn
= pfn_to_mfn(pfn
);
966 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
969 static int xen_pin_page(struct mm_struct
*mm
, struct page
*page
,
972 unsigned pgfl
= TestSetPagePinned(page
);
976 flush
= 0; /* already pinned */
977 else if (PageHighMem(page
))
978 /* kmaps need flushing if we found an unpinned
982 void *pt
= lowmem_page_address(page
);
983 unsigned long pfn
= page_to_pfn(page
);
984 struct multicall_space mcs
= __xen_mc_entry(0);
990 * We need to hold the pagetable lock between the time
991 * we make the pagetable RO and when we actually pin
992 * it. If we don't, then other users may come in and
993 * attempt to update the pagetable by writing it,
994 * which will fail because the memory is RO but not
995 * pinned, so Xen won't do the trap'n'emulate.
997 * If we're using split pte locks, we can't hold the
998 * entire pagetable's worth of locks during the
999 * traverse, because we may wrap the preempt count (8
1000 * bits). The solution is to mark RO and pin each PTE
1001 * page while holding the lock. This means the number
1002 * of locks we end up holding is never more than a
1003 * batch size (~32 entries, at present).
1005 * If we're not using split pte locks, we needn't pin
1006 * the PTE pages independently, because we're
1007 * protected by the overall pagetable lock.
1010 if (level
== PT_PTE
)
1011 ptl
= xen_pte_lock(page
, mm
);
1013 MULTI_update_va_mapping(mcs
.mc
, (unsigned long)pt
,
1014 pfn_pte(pfn
, PAGE_KERNEL_RO
),
1015 level
== PT_PGD
? UVMF_TLB_FLUSH
: 0);
1018 xen_do_pin(MMUEXT_PIN_L1_TABLE
, pfn
);
1020 /* Queue a deferred unlock for when this batch
1022 xen_mc_callback(xen_pte_unlock
, ptl
);
1029 /* This is called just after a mm has been created, but it has not
1030 been used yet. We need to make sure that its pagetable is all
1031 read-only, and can be pinned. */
1032 static void __xen_pgd_pin(struct mm_struct
*mm
, pgd_t
*pgd
)
1036 if (__xen_pgd_walk(mm
, pgd
, xen_pin_page
, USER_LIMIT
)) {
1037 /* re-enable interrupts for flushing */
1040 kmap_flush_unused();
1045 #ifdef CONFIG_X86_64
1047 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
1049 xen_do_pin(MMUEXT_PIN_L4_TABLE
, PFN_DOWN(__pa(pgd
)));
1052 xen_pin_page(mm
, virt_to_page(user_pgd
), PT_PGD
);
1053 xen_do_pin(MMUEXT_PIN_L4_TABLE
,
1054 PFN_DOWN(__pa(user_pgd
)));
1057 #else /* CONFIG_X86_32 */
1058 #ifdef CONFIG_X86_PAE
1059 /* Need to make sure unshared kernel PMD is pinnable */
1060 xen_pin_page(mm
, pgd_page(pgd
[pgd_index(TASK_SIZE
)]),
1063 xen_do_pin(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(pgd
)));
1064 #endif /* CONFIG_X86_64 */
1068 static void xen_pgd_pin(struct mm_struct
*mm
)
1070 __xen_pgd_pin(mm
, mm
->pgd
);
1074 * On save, we need to pin all pagetables to make sure they get their
1075 * mfns turned into pfns. Search the list for any unpinned pgds and pin
1076 * them (unpinned pgds are not currently in use, probably because the
1077 * process is under construction or destruction).
1079 * Expected to be called in stop_machine() ("equivalent to taking
1080 * every spinlock in the system"), so the locking doesn't really
1081 * matter all that much.
1083 void xen_mm_pin_all(void)
1085 unsigned long flags
;
1088 spin_lock_irqsave(&pgd_lock
, flags
);
1090 list_for_each_entry(page
, &pgd_list
, lru
) {
1091 if (!PagePinned(page
)) {
1092 __xen_pgd_pin(&init_mm
, (pgd_t
*)page_address(page
));
1093 SetPageSavePinned(page
);
1097 spin_unlock_irqrestore(&pgd_lock
, flags
);
1101 * The init_mm pagetable is really pinned as soon as its created, but
1102 * that's before we have page structures to store the bits. So do all
1103 * the book-keeping now.
1105 static __init
int xen_mark_pinned(struct mm_struct
*mm
, struct page
*page
,
1106 enum pt_level level
)
1108 SetPagePinned(page
);
1112 static void __init
xen_mark_init_mm_pinned(void)
1114 xen_pgd_walk(&init_mm
, xen_mark_pinned
, FIXADDR_TOP
);
1117 static int xen_unpin_page(struct mm_struct
*mm
, struct page
*page
,
1118 enum pt_level level
)
1120 unsigned pgfl
= TestClearPagePinned(page
);
1122 if (pgfl
&& !PageHighMem(page
)) {
1123 void *pt
= lowmem_page_address(page
);
1124 unsigned long pfn
= page_to_pfn(page
);
1125 spinlock_t
*ptl
= NULL
;
1126 struct multicall_space mcs
;
1129 * Do the converse to pin_page. If we're using split
1130 * pte locks, we must be holding the lock for while
1131 * the pte page is unpinned but still RO to prevent
1132 * concurrent updates from seeing it in this
1133 * partially-pinned state.
1135 if (level
== PT_PTE
) {
1136 ptl
= xen_pte_lock(page
, mm
);
1139 xen_do_pin(MMUEXT_UNPIN_TABLE
, pfn
);
1142 mcs
= __xen_mc_entry(0);
1144 MULTI_update_va_mapping(mcs
.mc
, (unsigned long)pt
,
1145 pfn_pte(pfn
, PAGE_KERNEL
),
1146 level
== PT_PGD
? UVMF_TLB_FLUSH
: 0);
1149 /* unlock when batch completed */
1150 xen_mc_callback(xen_pte_unlock
, ptl
);
1154 return 0; /* never need to flush on unpin */
1157 /* Release a pagetables pages back as normal RW */
1158 static void __xen_pgd_unpin(struct mm_struct
*mm
, pgd_t
*pgd
)
1162 xen_do_pin(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1164 #ifdef CONFIG_X86_64
1166 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
1169 xen_do_pin(MMUEXT_UNPIN_TABLE
,
1170 PFN_DOWN(__pa(user_pgd
)));
1171 xen_unpin_page(mm
, virt_to_page(user_pgd
), PT_PGD
);
1176 #ifdef CONFIG_X86_PAE
1177 /* Need to make sure unshared kernel PMD is unpinned */
1178 xen_unpin_page(mm
, pgd_page(pgd
[pgd_index(TASK_SIZE
)]),
1182 __xen_pgd_walk(mm
, pgd
, xen_unpin_page
, USER_LIMIT
);
1187 static void xen_pgd_unpin(struct mm_struct
*mm
)
1189 __xen_pgd_unpin(mm
, mm
->pgd
);
1193 * On resume, undo any pinning done at save, so that the rest of the
1194 * kernel doesn't see any unexpected pinned pagetables.
1196 void xen_mm_unpin_all(void)
1198 unsigned long flags
;
1201 spin_lock_irqsave(&pgd_lock
, flags
);
1203 list_for_each_entry(page
, &pgd_list
, lru
) {
1204 if (PageSavePinned(page
)) {
1205 BUG_ON(!PagePinned(page
));
1206 __xen_pgd_unpin(&init_mm
, (pgd_t
*)page_address(page
));
1207 ClearPageSavePinned(page
);
1211 spin_unlock_irqrestore(&pgd_lock
, flags
);
1214 void xen_activate_mm(struct mm_struct
*prev
, struct mm_struct
*next
)
1216 spin_lock(&next
->page_table_lock
);
1218 spin_unlock(&next
->page_table_lock
);
1221 void xen_dup_mmap(struct mm_struct
*oldmm
, struct mm_struct
*mm
)
1223 spin_lock(&mm
->page_table_lock
);
1225 spin_unlock(&mm
->page_table_lock
);
1230 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1231 we need to repoint it somewhere else before we can unpin it. */
1232 static void drop_other_mm_ref(void *info
)
1234 struct mm_struct
*mm
= info
;
1235 struct mm_struct
*active_mm
;
1237 active_mm
= percpu_read(cpu_tlbstate
.active_mm
);
1239 if (active_mm
== mm
)
1240 leave_mm(smp_processor_id());
1242 /* If this cpu still has a stale cr3 reference, then make sure
1243 it has been flushed. */
1244 if (percpu_read(xen_current_cr3
) == __pa(mm
->pgd
))
1245 load_cr3(swapper_pg_dir
);
1248 static void xen_drop_mm_ref(struct mm_struct
*mm
)
1253 if (current
->active_mm
== mm
) {
1254 if (current
->mm
== mm
)
1255 load_cr3(swapper_pg_dir
);
1257 leave_mm(smp_processor_id());
1260 /* Get the "official" set of cpus referring to our pagetable. */
1261 if (!alloc_cpumask_var(&mask
, GFP_ATOMIC
)) {
1262 for_each_online_cpu(cpu
) {
1263 if (!cpumask_test_cpu(cpu
, mm_cpumask(mm
))
1264 && per_cpu(xen_current_cr3
, cpu
) != __pa(mm
->pgd
))
1266 smp_call_function_single(cpu
, drop_other_mm_ref
, mm
, 1);
1270 cpumask_copy(mask
, mm_cpumask(mm
));
1272 /* It's possible that a vcpu may have a stale reference to our
1273 cr3, because its in lazy mode, and it hasn't yet flushed
1274 its set of pending hypercalls yet. In this case, we can
1275 look at its actual current cr3 value, and force it to flush
1277 for_each_online_cpu(cpu
) {
1278 if (per_cpu(xen_current_cr3
, cpu
) == __pa(mm
->pgd
))
1279 cpumask_set_cpu(cpu
, mask
);
1282 if (!cpumask_empty(mask
))
1283 smp_call_function_many(mask
, drop_other_mm_ref
, mm
, 1);
1284 free_cpumask_var(mask
);
1287 static void xen_drop_mm_ref(struct mm_struct
*mm
)
1289 if (current
->active_mm
== mm
)
1290 load_cr3(swapper_pg_dir
);
1295 * While a process runs, Xen pins its pagetables, which means that the
1296 * hypervisor forces it to be read-only, and it controls all updates
1297 * to it. This means that all pagetable updates have to go via the
1298 * hypervisor, which is moderately expensive.
1300 * Since we're pulling the pagetable down, we switch to use init_mm,
1301 * unpin old process pagetable and mark it all read-write, which
1302 * allows further operations on it to be simple memory accesses.
1304 * The only subtle point is that another CPU may be still using the
1305 * pagetable because of lazy tlb flushing. This means we need need to
1306 * switch all CPUs off this pagetable before we can unpin it.
1308 void xen_exit_mmap(struct mm_struct
*mm
)
1310 get_cpu(); /* make sure we don't move around */
1311 xen_drop_mm_ref(mm
);
1314 spin_lock(&mm
->page_table_lock
);
1316 /* pgd may not be pinned in the error exit path of execve */
1317 if (xen_page_pinned(mm
->pgd
))
1320 spin_unlock(&mm
->page_table_lock
);
1323 static __init
void xen_pagetable_setup_start(pgd_t
*base
)
1327 static void xen_post_allocator_init(void);
1329 static __init
void xen_pagetable_setup_done(pgd_t
*base
)
1331 xen_setup_shared_info();
1332 xen_post_allocator_init();
1335 static void xen_write_cr2(unsigned long cr2
)
1337 percpu_read(xen_vcpu
)->arch
.cr2
= cr2
;
1340 static unsigned long xen_read_cr2(void)
1342 return percpu_read(xen_vcpu
)->arch
.cr2
;
1345 unsigned long xen_read_cr2_direct(void)
1347 return percpu_read(xen_vcpu_info
.arch
.cr2
);
1350 static void xen_flush_tlb(void)
1352 struct mmuext_op
*op
;
1353 struct multicall_space mcs
;
1357 mcs
= xen_mc_entry(sizeof(*op
));
1360 op
->cmd
= MMUEXT_TLB_FLUSH_LOCAL
;
1361 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
1363 xen_mc_issue(PARAVIRT_LAZY_MMU
);
1368 static void xen_flush_tlb_single(unsigned long addr
)
1370 struct mmuext_op
*op
;
1371 struct multicall_space mcs
;
1375 mcs
= xen_mc_entry(sizeof(*op
));
1377 op
->cmd
= MMUEXT_INVLPG_LOCAL
;
1378 op
->arg1
.linear_addr
= addr
& PAGE_MASK
;
1379 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
1381 xen_mc_issue(PARAVIRT_LAZY_MMU
);
1386 static void xen_flush_tlb_others(const struct cpumask
*cpus
,
1387 struct mm_struct
*mm
, unsigned long va
)
1390 struct mmuext_op op
;
1391 DECLARE_BITMAP(mask
, NR_CPUS
);
1393 struct multicall_space mcs
;
1395 if (cpumask_empty(cpus
))
1396 return; /* nothing to do */
1398 mcs
= xen_mc_entry(sizeof(*args
));
1400 args
->op
.arg2
.vcpumask
= to_cpumask(args
->mask
);
1402 /* Remove us, and any offline CPUS. */
1403 cpumask_and(to_cpumask(args
->mask
), cpus
, cpu_online_mask
);
1404 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args
->mask
));
1406 if (va
== TLB_FLUSH_ALL
) {
1407 args
->op
.cmd
= MMUEXT_TLB_FLUSH_MULTI
;
1409 args
->op
.cmd
= MMUEXT_INVLPG_MULTI
;
1410 args
->op
.arg1
.linear_addr
= va
;
1413 MULTI_mmuext_op(mcs
.mc
, &args
->op
, 1, NULL
, DOMID_SELF
);
1415 xen_mc_issue(PARAVIRT_LAZY_MMU
);
1418 static unsigned long xen_read_cr3(void)
1420 return percpu_read(xen_cr3
);
1423 static void set_current_cr3(void *v
)
1425 percpu_write(xen_current_cr3
, (unsigned long)v
);
1428 static void __xen_write_cr3(bool kernel
, unsigned long cr3
)
1430 struct mmuext_op
*op
;
1431 struct multicall_space mcs
;
1435 mfn
= pfn_to_mfn(PFN_DOWN(cr3
));
1439 WARN_ON(mfn
== 0 && kernel
);
1441 mcs
= __xen_mc_entry(sizeof(*op
));
1444 op
->cmd
= kernel
? MMUEXT_NEW_BASEPTR
: MMUEXT_NEW_USER_BASEPTR
;
1447 MULTI_mmuext_op(mcs
.mc
, op
, 1, NULL
, DOMID_SELF
);
1450 percpu_write(xen_cr3
, cr3
);
1452 /* Update xen_current_cr3 once the batch has actually
1454 xen_mc_callback(set_current_cr3
, (void *)cr3
);
1458 static void xen_write_cr3(unsigned long cr3
)
1460 BUG_ON(preemptible());
1462 xen_mc_batch(); /* disables interrupts */
1464 /* Update while interrupts are disabled, so its atomic with
1466 percpu_write(xen_cr3
, cr3
);
1468 __xen_write_cr3(true, cr3
);
1470 #ifdef CONFIG_X86_64
1472 pgd_t
*user_pgd
= xen_get_user_pgd(__va(cr3
));
1474 __xen_write_cr3(false, __pa(user_pgd
));
1476 __xen_write_cr3(false, 0);
1480 xen_mc_issue(PARAVIRT_LAZY_CPU
); /* interrupts restored */
1483 static int xen_pgd_alloc(struct mm_struct
*mm
)
1485 pgd_t
*pgd
= mm
->pgd
;
1488 BUG_ON(PagePinned(virt_to_page(pgd
)));
1490 #ifdef CONFIG_X86_64
1492 struct page
*page
= virt_to_page(pgd
);
1495 BUG_ON(page
->private != 0);
1499 user_pgd
= (pgd_t
*)__get_free_page(GFP_KERNEL
| __GFP_ZERO
);
1500 page
->private = (unsigned long)user_pgd
;
1502 if (user_pgd
!= NULL
) {
1503 user_pgd
[pgd_index(VSYSCALL_START
)] =
1504 __pgd(__pa(level3_user_vsyscall
) | _PAGE_TABLE
);
1508 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd
))));
1515 static void xen_pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
1517 #ifdef CONFIG_X86_64
1518 pgd_t
*user_pgd
= xen_get_user_pgd(pgd
);
1521 free_page((unsigned long)user_pgd
);
1525 #ifdef CONFIG_X86_32
1526 static __init pte_t
mask_rw_pte(pte_t
*ptep
, pte_t pte
)
1528 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1529 if (pte_val_ma(*ptep
) & _PAGE_PRESENT
)
1530 pte
= __pte_ma(((pte_val_ma(*ptep
) & _PAGE_RW
) | ~_PAGE_RW
) &
1536 /* Init-time set_pte while constructing initial pagetables, which
1537 doesn't allow RO pagetable pages to be remapped RW */
1538 static __init
void xen_set_pte_init(pte_t
*ptep
, pte_t pte
)
1540 pte
= mask_rw_pte(ptep
, pte
);
1542 xen_set_pte(ptep
, pte
);
1546 static void pin_pagetable_pfn(unsigned cmd
, unsigned long pfn
)
1548 struct mmuext_op op
;
1550 op
.arg1
.mfn
= pfn_to_mfn(pfn
);
1551 if (HYPERVISOR_mmuext_op(&op
, 1, NULL
, DOMID_SELF
))
1555 /* Early in boot, while setting up the initial pagetable, assume
1556 everything is pinned. */
1557 static __init
void xen_alloc_pte_init(struct mm_struct
*mm
, unsigned long pfn
)
1559 #ifdef CONFIG_FLATMEM
1560 BUG_ON(mem_map
); /* should only be used early */
1562 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
1563 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
1566 /* Used for pmd and pud */
1567 static __init
void xen_alloc_pmd_init(struct mm_struct
*mm
, unsigned long pfn
)
1569 #ifdef CONFIG_FLATMEM
1570 BUG_ON(mem_map
); /* should only be used early */
1572 make_lowmem_page_readonly(__va(PFN_PHYS(pfn
)));
1575 /* Early release_pte assumes that all pts are pinned, since there's
1576 only init_mm and anything attached to that is pinned. */
1577 static __init
void xen_release_pte_init(unsigned long pfn
)
1579 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
1580 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
1583 static __init
void xen_release_pmd_init(unsigned long pfn
)
1585 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
1588 /* This needs to make sure the new pte page is pinned iff its being
1589 attached to a pinned pagetable. */
1590 static void xen_alloc_ptpage(struct mm_struct
*mm
, unsigned long pfn
, unsigned level
)
1592 struct page
*page
= pfn_to_page(pfn
);
1594 if (PagePinned(virt_to_page(mm
->pgd
))) {
1595 SetPagePinned(page
);
1597 if (!PageHighMem(page
)) {
1598 make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn
)));
1599 if (level
== PT_PTE
&& USE_SPLIT_PTLOCKS
)
1600 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE
, pfn
);
1602 /* make sure there are no stray mappings of
1604 kmap_flush_unused();
1609 static void xen_alloc_pte(struct mm_struct
*mm
, unsigned long pfn
)
1611 xen_alloc_ptpage(mm
, pfn
, PT_PTE
);
1614 static void xen_alloc_pmd(struct mm_struct
*mm
, unsigned long pfn
)
1616 xen_alloc_ptpage(mm
, pfn
, PT_PMD
);
1619 /* This should never happen until we're OK to use struct page */
1620 static void xen_release_ptpage(unsigned long pfn
, unsigned level
)
1622 struct page
*page
= pfn_to_page(pfn
);
1624 if (PagePinned(page
)) {
1625 if (!PageHighMem(page
)) {
1626 if (level
== PT_PTE
&& USE_SPLIT_PTLOCKS
)
1627 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, pfn
);
1628 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn
)));
1630 ClearPagePinned(page
);
1634 static void xen_release_pte(unsigned long pfn
)
1636 xen_release_ptpage(pfn
, PT_PTE
);
1639 static void xen_release_pmd(unsigned long pfn
)
1641 xen_release_ptpage(pfn
, PT_PMD
);
1644 #if PAGETABLE_LEVELS == 4
1645 static void xen_alloc_pud(struct mm_struct
*mm
, unsigned long pfn
)
1647 xen_alloc_ptpage(mm
, pfn
, PT_PUD
);
1650 static void xen_release_pud(unsigned long pfn
)
1652 xen_release_ptpage(pfn
, PT_PUD
);
1656 void __init
xen_reserve_top(void)
1658 #ifdef CONFIG_X86_32
1659 unsigned long top
= HYPERVISOR_VIRT_START
;
1660 struct xen_platform_parameters pp
;
1662 if (HYPERVISOR_xen_version(XENVER_platform_parameters
, &pp
) == 0)
1663 top
= pp
.virt_start
;
1665 reserve_top_address(-top
);
1666 #endif /* CONFIG_X86_32 */
1670 * Like __va(), but returns address in the kernel mapping (which is
1671 * all we have until the physical memory mapping has been set up.
1673 static void *__ka(phys_addr_t paddr
)
1675 #ifdef CONFIG_X86_64
1676 return (void *)(paddr
+ __START_KERNEL_map
);
1682 /* Convert a machine address to physical address */
1683 static unsigned long m2p(phys_addr_t maddr
)
1687 maddr
&= PTE_PFN_MASK
;
1688 paddr
= mfn_to_pfn(maddr
>> PAGE_SHIFT
) << PAGE_SHIFT
;
1693 /* Convert a machine address to kernel virtual */
1694 static void *m2v(phys_addr_t maddr
)
1696 return __ka(m2p(maddr
));
1699 static void set_page_prot(void *addr
, pgprot_t prot
)
1701 unsigned long pfn
= __pa(addr
) >> PAGE_SHIFT
;
1702 pte_t pte
= pfn_pte(pfn
, prot
);
1704 if (HYPERVISOR_update_va_mapping((unsigned long)addr
, pte
, 0))
1708 static __init
void xen_map_identity_early(pmd_t
*pmd
, unsigned long max_pfn
)
1710 unsigned pmdidx
, pteidx
;
1716 for (pmdidx
= 0; pmdidx
< PTRS_PER_PMD
&& pfn
< max_pfn
; pmdidx
++) {
1719 /* Reuse or allocate a page of ptes */
1720 if (pmd_present(pmd
[pmdidx
]))
1721 pte_page
= m2v(pmd
[pmdidx
].pmd
);
1723 /* Check for free pte pages */
1724 if (ident_pte
== ARRAY_SIZE(level1_ident_pgt
))
1727 pte_page
= &level1_ident_pgt
[ident_pte
];
1728 ident_pte
+= PTRS_PER_PTE
;
1730 pmd
[pmdidx
] = __pmd(__pa(pte_page
) | _PAGE_TABLE
);
1733 /* Install mappings */
1734 for (pteidx
= 0; pteidx
< PTRS_PER_PTE
; pteidx
++, pfn
++) {
1737 if (pfn
> max_pfn_mapped
)
1738 max_pfn_mapped
= pfn
;
1740 if (!pte_none(pte_page
[pteidx
]))
1743 pte
= pfn_pte(pfn
, PAGE_KERNEL_EXEC
);
1744 pte_page
[pteidx
] = pte
;
1748 for (pteidx
= 0; pteidx
< ident_pte
; pteidx
+= PTRS_PER_PTE
)
1749 set_page_prot(&level1_ident_pgt
[pteidx
], PAGE_KERNEL_RO
);
1751 set_page_prot(pmd
, PAGE_KERNEL_RO
);
1754 #ifdef CONFIG_X86_64
1755 static void convert_pfn_mfn(void *v
)
1760 /* All levels are converted the same way, so just treat them
1762 for (i
= 0; i
< PTRS_PER_PTE
; i
++)
1763 pte
[i
] = xen_make_pte(pte
[i
].pte
);
1767 * Set up the inital kernel pagetable.
1769 * We can construct this by grafting the Xen provided pagetable into
1770 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1771 * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1772 * means that only the kernel has a physical mapping to start with -
1773 * but that's enough to get __va working. We need to fill in the rest
1774 * of the physical mapping once some sort of allocator has been set
1777 __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
,
1778 unsigned long max_pfn
)
1783 /* Zap identity mapping */
1784 init_level4_pgt
[0] = __pgd(0);
1786 /* Pre-constructed entries are in pfn, so convert to mfn */
1787 convert_pfn_mfn(init_level4_pgt
);
1788 convert_pfn_mfn(level3_ident_pgt
);
1789 convert_pfn_mfn(level3_kernel_pgt
);
1791 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
)].pgd
);
1792 l2
= m2v(l3
[pud_index(__START_KERNEL_map
)].pud
);
1794 memcpy(level2_ident_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1795 memcpy(level2_kernel_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1797 l3
= m2v(pgd
[pgd_index(__START_KERNEL_map
+ PMD_SIZE
)].pgd
);
1798 l2
= m2v(l3
[pud_index(__START_KERNEL_map
+ PMD_SIZE
)].pud
);
1799 memcpy(level2_fixmap_pgt
, l2
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1801 /* Set up identity map */
1802 xen_map_identity_early(level2_ident_pgt
, max_pfn
);
1804 /* Make pagetable pieces RO */
1805 set_page_prot(init_level4_pgt
, PAGE_KERNEL_RO
);
1806 set_page_prot(level3_ident_pgt
, PAGE_KERNEL_RO
);
1807 set_page_prot(level3_kernel_pgt
, PAGE_KERNEL_RO
);
1808 set_page_prot(level3_user_vsyscall
, PAGE_KERNEL_RO
);
1809 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1810 set_page_prot(level2_fixmap_pgt
, PAGE_KERNEL_RO
);
1812 /* Pin down new L4 */
1813 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE
,
1814 PFN_DOWN(__pa_symbol(init_level4_pgt
)));
1816 /* Unpin Xen-provided one */
1817 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1820 pgd
= init_level4_pgt
;
1823 * At this stage there can be no user pgd, and no page
1824 * structure to attach it to, so make sure we just set kernel
1828 __xen_write_cr3(true, __pa(pgd
));
1829 xen_mc_issue(PARAVIRT_LAZY_CPU
);
1831 reserve_early(__pa(xen_start_info
->pt_base
),
1832 __pa(xen_start_info
->pt_base
+
1833 xen_start_info
->nr_pt_frames
* PAGE_SIZE
),
1838 #else /* !CONFIG_X86_64 */
1839 static pmd_t level2_kernel_pgt
[PTRS_PER_PMD
] __page_aligned_bss
;
1841 __init pgd_t
*xen_setup_kernel_pagetable(pgd_t
*pgd
,
1842 unsigned long max_pfn
)
1846 max_pfn_mapped
= PFN_DOWN(__pa(xen_start_info
->pt_base
) +
1847 xen_start_info
->nr_pt_frames
* PAGE_SIZE
+
1850 kernel_pmd
= m2v(pgd
[KERNEL_PGD_BOUNDARY
].pgd
);
1851 memcpy(level2_kernel_pgt
, kernel_pmd
, sizeof(pmd_t
) * PTRS_PER_PMD
);
1853 xen_map_identity_early(level2_kernel_pgt
, max_pfn
);
1855 memcpy(swapper_pg_dir
, pgd
, sizeof(pgd_t
) * PTRS_PER_PGD
);
1856 set_pgd(&swapper_pg_dir
[KERNEL_PGD_BOUNDARY
],
1857 __pgd(__pa(level2_kernel_pgt
) | _PAGE_PRESENT
));
1859 set_page_prot(level2_kernel_pgt
, PAGE_KERNEL_RO
);
1860 set_page_prot(swapper_pg_dir
, PAGE_KERNEL_RO
);
1861 set_page_prot(empty_zero_page
, PAGE_KERNEL_RO
);
1863 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE
, PFN_DOWN(__pa(pgd
)));
1865 xen_write_cr3(__pa(swapper_pg_dir
));
1867 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE
, PFN_DOWN(__pa(swapper_pg_dir
)));
1869 reserve_early(__pa(xen_start_info
->pt_base
),
1870 __pa(xen_start_info
->pt_base
+
1871 xen_start_info
->nr_pt_frames
* PAGE_SIZE
),
1874 return swapper_pg_dir
;
1876 #endif /* CONFIG_X86_64 */
1878 static void xen_set_fixmap(unsigned idx
, phys_addr_t phys
, pgprot_t prot
)
1882 phys
>>= PAGE_SHIFT
;
1885 case FIX_BTMAP_END
... FIX_BTMAP_BEGIN
:
1886 #ifdef CONFIG_X86_F00F_BUG
1889 #ifdef CONFIG_X86_32
1892 # ifdef CONFIG_HIGHMEM
1893 case FIX_KMAP_BEGIN
... FIX_KMAP_END
:
1896 case VSYSCALL_LAST_PAGE
... VSYSCALL_FIRST_PAGE
:
1898 #ifdef CONFIG_X86_LOCAL_APIC
1899 case FIX_APIC_BASE
: /* maps dummy local APIC */
1901 case FIX_TEXT_POKE0
:
1902 case FIX_TEXT_POKE1
:
1903 /* All local page mappings */
1904 pte
= pfn_pte(phys
, prot
);
1907 case FIX_PARAVIRT_BOOTMAP
:
1908 /* This is an MFN, but it isn't an IO mapping from the
1910 pte
= mfn_pte(phys
, prot
);
1914 /* By default, set_fixmap is used for hardware mappings */
1915 pte
= mfn_pte(phys
, __pgprot(pgprot_val(prot
) | _PAGE_IOMAP
));
1919 __native_set_fixmap(idx
, pte
);
1921 #ifdef CONFIG_X86_64
1922 /* Replicate changes to map the vsyscall page into the user
1923 pagetable vsyscall mapping. */
1924 if (idx
>= VSYSCALL_LAST_PAGE
&& idx
<= VSYSCALL_FIRST_PAGE
) {
1925 unsigned long vaddr
= __fix_to_virt(idx
);
1926 set_pte_vaddr_pud(level3_user_vsyscall
, vaddr
, pte
);
1931 static __init
void xen_post_allocator_init(void)
1933 pv_mmu_ops
.set_pte
= xen_set_pte
;
1934 pv_mmu_ops
.set_pmd
= xen_set_pmd
;
1935 pv_mmu_ops
.set_pud
= xen_set_pud
;
1936 #if PAGETABLE_LEVELS == 4
1937 pv_mmu_ops
.set_pgd
= xen_set_pgd
;
1940 /* This will work as long as patching hasn't happened yet
1941 (which it hasn't) */
1942 pv_mmu_ops
.alloc_pte
= xen_alloc_pte
;
1943 pv_mmu_ops
.alloc_pmd
= xen_alloc_pmd
;
1944 pv_mmu_ops
.release_pte
= xen_release_pte
;
1945 pv_mmu_ops
.release_pmd
= xen_release_pmd
;
1946 #if PAGETABLE_LEVELS == 4
1947 pv_mmu_ops
.alloc_pud
= xen_alloc_pud
;
1948 pv_mmu_ops
.release_pud
= xen_release_pud
;
1951 #ifdef CONFIG_X86_64
1952 SetPagePinned(virt_to_page(level3_user_vsyscall
));
1954 xen_mark_init_mm_pinned();
1957 static void xen_leave_lazy_mmu(void)
1961 paravirt_leave_lazy_mmu();
1965 static const struct pv_mmu_ops xen_mmu_ops __initdata
= {
1966 .read_cr2
= xen_read_cr2
,
1967 .write_cr2
= xen_write_cr2
,
1969 .read_cr3
= xen_read_cr3
,
1970 .write_cr3
= xen_write_cr3
,
1972 .flush_tlb_user
= xen_flush_tlb
,
1973 .flush_tlb_kernel
= xen_flush_tlb
,
1974 .flush_tlb_single
= xen_flush_tlb_single
,
1975 .flush_tlb_others
= xen_flush_tlb_others
,
1977 .pte_update
= paravirt_nop
,
1978 .pte_update_defer
= paravirt_nop
,
1980 .pgd_alloc
= xen_pgd_alloc
,
1981 .pgd_free
= xen_pgd_free
,
1983 .alloc_pte
= xen_alloc_pte_init
,
1984 .release_pte
= xen_release_pte_init
,
1985 .alloc_pmd
= xen_alloc_pmd_init
,
1986 .alloc_pmd_clone
= paravirt_nop
,
1987 .release_pmd
= xen_release_pmd_init
,
1989 #ifdef CONFIG_X86_64
1990 .set_pte
= xen_set_pte
,
1992 .set_pte
= xen_set_pte_init
,
1994 .set_pte_at
= xen_set_pte_at
,
1995 .set_pmd
= xen_set_pmd_hyper
,
1997 .ptep_modify_prot_start
= __ptep_modify_prot_start
,
1998 .ptep_modify_prot_commit
= __ptep_modify_prot_commit
,
2000 .pte_val
= PV_CALLEE_SAVE(xen_pte_val
),
2001 .pgd_val
= PV_CALLEE_SAVE(xen_pgd_val
),
2003 .make_pte
= PV_CALLEE_SAVE(xen_make_pte
),
2004 .make_pgd
= PV_CALLEE_SAVE(xen_make_pgd
),
2006 #ifdef CONFIG_X86_PAE
2007 .set_pte_atomic
= xen_set_pte_atomic
,
2008 .pte_clear
= xen_pte_clear
,
2009 .pmd_clear
= xen_pmd_clear
,
2010 #endif /* CONFIG_X86_PAE */
2011 .set_pud
= xen_set_pud_hyper
,
2013 .make_pmd
= PV_CALLEE_SAVE(xen_make_pmd
),
2014 .pmd_val
= PV_CALLEE_SAVE(xen_pmd_val
),
2016 #if PAGETABLE_LEVELS == 4
2017 .pud_val
= PV_CALLEE_SAVE(xen_pud_val
),
2018 .make_pud
= PV_CALLEE_SAVE(xen_make_pud
),
2019 .set_pgd
= xen_set_pgd_hyper
,
2021 .alloc_pud
= xen_alloc_pmd_init
,
2022 .release_pud
= xen_release_pmd_init
,
2023 #endif /* PAGETABLE_LEVELS == 4 */
2025 .activate_mm
= xen_activate_mm
,
2026 .dup_mmap
= xen_dup_mmap
,
2027 .exit_mmap
= xen_exit_mmap
,
2030 .enter
= paravirt_enter_lazy_mmu
,
2031 .leave
= xen_leave_lazy_mmu
,
2034 .set_fixmap
= xen_set_fixmap
,
2037 void __init
xen_init_mmu_ops(void)
2039 x86_init
.paging
.pagetable_setup_start
= xen_pagetable_setup_start
;
2040 x86_init
.paging
.pagetable_setup_done
= xen_pagetable_setup_done
;
2041 pv_mmu_ops
= xen_mmu_ops
;
2043 vmap_lazy_unmap
= false;
2046 /* Protected by xen_reservation_lock. */
2047 #define MAX_CONTIG_ORDER 9 /* 2MB */
2048 static unsigned long discontig_frames
[1<<MAX_CONTIG_ORDER
];
2050 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2051 static void xen_zap_pfn_range(unsigned long vaddr
, unsigned int order
,
2052 unsigned long *in_frames
,
2053 unsigned long *out_frames
)
2056 struct multicall_space mcs
;
2059 for (i
= 0; i
< (1UL<<order
); i
++, vaddr
+= PAGE_SIZE
) {
2060 mcs
= __xen_mc_entry(0);
2063 in_frames
[i
] = virt_to_mfn(vaddr
);
2065 MULTI_update_va_mapping(mcs
.mc
, vaddr
, VOID_PTE
, 0);
2066 set_phys_to_machine(virt_to_pfn(vaddr
), INVALID_P2M_ENTRY
);
2069 out_frames
[i
] = virt_to_pfn(vaddr
);
2075 * Update the pfn-to-mfn mappings for a virtual address range, either to
2076 * point to an array of mfns, or contiguously from a single starting
2079 static void xen_remap_exchanged_ptes(unsigned long vaddr
, int order
,
2080 unsigned long *mfns
,
2081 unsigned long first_mfn
)
2088 limit
= 1u << order
;
2089 for (i
= 0; i
< limit
; i
++, vaddr
+= PAGE_SIZE
) {
2090 struct multicall_space mcs
;
2093 mcs
= __xen_mc_entry(0);
2097 mfn
= first_mfn
+ i
;
2099 if (i
< (limit
- 1))
2103 flags
= UVMF_INVLPG
| UVMF_ALL
;
2105 flags
= UVMF_TLB_FLUSH
| UVMF_ALL
;
2108 MULTI_update_va_mapping(mcs
.mc
, vaddr
,
2109 mfn_pte(mfn
, PAGE_KERNEL
), flags
);
2111 set_phys_to_machine(virt_to_pfn(vaddr
), mfn
);
2118 * Perform the hypercall to exchange a region of our pfns to point to
2119 * memory with the required contiguous alignment. Takes the pfns as
2120 * input, and populates mfns as output.
2122 * Returns a success code indicating whether the hypervisor was able to
2123 * satisfy the request or not.
2125 static int xen_exchange_memory(unsigned long extents_in
, unsigned int order_in
,
2126 unsigned long *pfns_in
,
2127 unsigned long extents_out
,
2128 unsigned int order_out
,
2129 unsigned long *mfns_out
,
2130 unsigned int address_bits
)
2135 struct xen_memory_exchange exchange
= {
2137 .nr_extents
= extents_in
,
2138 .extent_order
= order_in
,
2139 .extent_start
= pfns_in
,
2143 .nr_extents
= extents_out
,
2144 .extent_order
= order_out
,
2145 .extent_start
= mfns_out
,
2146 .address_bits
= address_bits
,
2151 BUG_ON(extents_in
<< order_in
!= extents_out
<< order_out
);
2153 rc
= HYPERVISOR_memory_op(XENMEM_exchange
, &exchange
);
2154 success
= (exchange
.nr_exchanged
== extents_in
);
2156 BUG_ON(!success
&& ((exchange
.nr_exchanged
!= 0) || (rc
== 0)));
2157 BUG_ON(success
&& (rc
!= 0));
2162 int xen_create_contiguous_region(unsigned long vstart
, unsigned int order
,
2163 unsigned int address_bits
)
2165 unsigned long *in_frames
= discontig_frames
, out_frame
;
2166 unsigned long flags
;
2170 * Currently an auto-translated guest will not perform I/O, nor will
2171 * it require PAE page directories below 4GB. Therefore any calls to
2172 * this function are redundant and can be ignored.
2175 if (xen_feature(XENFEAT_auto_translated_physmap
))
2178 if (unlikely(order
> MAX_CONTIG_ORDER
))
2181 memset((void *) vstart
, 0, PAGE_SIZE
<< order
);
2183 spin_lock_irqsave(&xen_reservation_lock
, flags
);
2185 /* 1. Zap current PTEs, remembering MFNs. */
2186 xen_zap_pfn_range(vstart
, order
, in_frames
, NULL
);
2188 /* 2. Get a new contiguous memory extent. */
2189 out_frame
= virt_to_pfn(vstart
);
2190 success
= xen_exchange_memory(1UL << order
, 0, in_frames
,
2191 1, order
, &out_frame
,
2194 /* 3. Map the new extent in place of old pages. */
2196 xen_remap_exchanged_ptes(vstart
, order
, NULL
, out_frame
);
2198 xen_remap_exchanged_ptes(vstart
, order
, in_frames
, 0);
2200 spin_unlock_irqrestore(&xen_reservation_lock
, flags
);
2202 return success
? 0 : -ENOMEM
;
2204 EXPORT_SYMBOL_GPL(xen_create_contiguous_region
);
2206 void xen_destroy_contiguous_region(unsigned long vstart
, unsigned int order
)
2208 unsigned long *out_frames
= discontig_frames
, in_frame
;
2209 unsigned long flags
;
2212 if (xen_feature(XENFEAT_auto_translated_physmap
))
2215 if (unlikely(order
> MAX_CONTIG_ORDER
))
2218 memset((void *) vstart
, 0, PAGE_SIZE
<< order
);
2220 spin_lock_irqsave(&xen_reservation_lock
, flags
);
2222 /* 1. Find start MFN of contiguous extent. */
2223 in_frame
= virt_to_mfn(vstart
);
2225 /* 2. Zap current PTEs. */
2226 xen_zap_pfn_range(vstart
, order
, NULL
, out_frames
);
2228 /* 3. Do the exchange for non-contiguous MFNs. */
2229 success
= xen_exchange_memory(1, order
, &in_frame
, 1UL << order
,
2232 /* 4. Map new pages in place of old pages. */
2234 xen_remap_exchanged_ptes(vstart
, order
, out_frames
, 0);
2236 xen_remap_exchanged_ptes(vstart
, order
, NULL
, in_frame
);
2238 spin_unlock_irqrestore(&xen_reservation_lock
, flags
);
2240 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region
);
2242 #ifdef CONFIG_XEN_PVHVM
2243 static void xen_hvm_exit_mmap(struct mm_struct
*mm
)
2245 struct xen_hvm_pagetable_dying a
;
2248 a
.domid
= DOMID_SELF
;
2249 a
.gpa
= __pa(mm
->pgd
);
2250 rc
= HYPERVISOR_hvm_op(HVMOP_pagetable_dying
, &a
);
2251 WARN_ON_ONCE(rc
< 0);
2254 static int is_pagetable_dying_supported(void)
2256 struct xen_hvm_pagetable_dying a
;
2259 a
.domid
= DOMID_SELF
;
2261 rc
= HYPERVISOR_hvm_op(HVMOP_pagetable_dying
, &a
);
2263 printk(KERN_DEBUG
"HVMOP_pagetable_dying not supported\n");
2269 void __init
xen_hvm_init_mmu_ops(void)
2271 if (is_pagetable_dying_supported())
2272 pv_mmu_ops
.exit_mmap
= xen_hvm_exit_mmap
;
2276 #ifdef CONFIG_XEN_DEBUG_FS
2278 static struct dentry
*d_mmu_debug
;
2280 static int __init
xen_mmu_debugfs(void)
2282 struct dentry
*d_xen
= xen_init_debugfs();
2287 d_mmu_debug
= debugfs_create_dir("mmu", d_xen
);
2289 debugfs_create_u8("zero_stats", 0644, d_mmu_debug
, &zero_stats
);
2291 debugfs_create_u32("pgd_update", 0444, d_mmu_debug
, &mmu_stats
.pgd_update
);
2292 debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug
,
2293 &mmu_stats
.pgd_update_pinned
);
2294 debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug
,
2295 &mmu_stats
.pgd_update_pinned
);
2297 debugfs_create_u32("pud_update", 0444, d_mmu_debug
, &mmu_stats
.pud_update
);
2298 debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug
,
2299 &mmu_stats
.pud_update_pinned
);
2300 debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug
,
2301 &mmu_stats
.pud_update_pinned
);
2303 debugfs_create_u32("pmd_update", 0444, d_mmu_debug
, &mmu_stats
.pmd_update
);
2304 debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug
,
2305 &mmu_stats
.pmd_update_pinned
);
2306 debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug
,
2307 &mmu_stats
.pmd_update_pinned
);
2309 debugfs_create_u32("pte_update", 0444, d_mmu_debug
, &mmu_stats
.pte_update
);
2310 // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug,
2311 // &mmu_stats.pte_update_pinned);
2312 debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug
,
2313 &mmu_stats
.pte_update_pinned
);
2315 debugfs_create_u32("mmu_update", 0444, d_mmu_debug
, &mmu_stats
.mmu_update
);
2316 debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug
,
2317 &mmu_stats
.mmu_update_extended
);
2318 xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug
,
2319 mmu_stats
.mmu_update_histo
, 20);
2321 debugfs_create_u32("set_pte_at", 0444, d_mmu_debug
, &mmu_stats
.set_pte_at
);
2322 debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug
,
2323 &mmu_stats
.set_pte_at_batched
);
2324 debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug
,
2325 &mmu_stats
.set_pte_at_current
);
2326 debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug
,
2327 &mmu_stats
.set_pte_at_kernel
);
2329 debugfs_create_u32("prot_commit", 0444, d_mmu_debug
, &mmu_stats
.prot_commit
);
2330 debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug
,
2331 &mmu_stats
.prot_commit_batched
);
2335 fs_initcall(xen_mmu_debugfs
);
2337 #endif /* CONFIG_XEN_DEBUG_FS */