4
* This file contains the various mmu fetch and update operations.
5
* The most important job they must perform is the mapping between the
6
* domain's pfn and the overall machine mfns.
8
* Xen allows guests to directly update the pagetable, in a controlled
9
* fashion. In other words, the guest modifies the same pagetable
10
* that the CPU actually uses, which eliminates the overhead of having
11
* a separate shadow pagetable.
13
* In order to allow this, it falls on the guest domain to map its
14
* notion of a "physical" pfn - which is just a domain-local linear
15
* address - into a real "machine address" which the CPU's MMU can
18
* A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19
* inserted directly into the pagetable. When creating a new
20
* pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21
* when reading the content back with __(pgd|pmd|pte)_val, it converts
22
* the mfn back into a pfn.
24
* The other constraint is that all pages which make up a pagetable
25
* must be mapped read-only in the guest. This prevents uncontrolled
26
* guest updates to the pagetable. Xen strictly enforces this, and
27
* will disallow any pagetable update which will end up mapping a
28
* pagetable page RW, and will disallow using any writable page as a
31
* Naively, when loading %cr3 with the base of a new pagetable, Xen
32
* would need to validate the whole pagetable before going on.
33
* Naturally, this is quite slow. The solution is to "pin" a
34
* pagetable, which enforces all the constraints on the pagetable even
35
* when it is not actively in use. This menas that Xen can be assured
36
* that it is still valid when you do load it into %cr3, and doesn't
37
* need to revalidate it.
39
* Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41
#include <linux/sched.h>
42
#include <linux/highmem.h>
43
#include <linux/debugfs.h>
44
#include <linux/bug.h>
45
#include <linux/vmalloc.h>
46
#include <linux/module.h>
47
#include <linux/gfp.h>
48
#include <linux/memblock.h>
49
#include <linux/seq_file.h>
51
#include <trace/events/xen.h>
53
#include <asm/pgtable.h>
54
#include <asm/tlbflush.h>
55
#include <asm/fixmap.h>
56
#include <asm/mmu_context.h>
57
#include <asm/setup.h>
58
#include <asm/paravirt.h>
60
#include <asm/linkage.h>
66
#include <asm/xen/hypercall.h>
67
#include <asm/xen/hypervisor.h>
71
#include <xen/interface/xen.h>
72
#include <xen/interface/hvm/hvm_op.h>
73
#include <xen/interface/version.h>
74
#include <xen/interface/memory.h>
75
#include <xen/hvc-console.h>
77
#include "multicalls.h"
82
* Protects atomic reservation decrease/increase against concurrent increases.
83
* Also protects non-atomic updates of current_pages and balloon lists.
85
DEFINE_SPINLOCK(xen_reservation_lock);
88
* Identity map, in addition to plain kernel map. This needs to be
89
* large enough to allocate page table pages to allocate the rest.
90
* Each page can map 2MB.
92
#define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
93
static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
96
/* l3 pud for userspace vsyscall mapping */
97
static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
98
#endif /* CONFIG_X86_64 */
101
* Note about cr3 (pagetable base) values:
103
* xen_cr3 contains the current logical cr3 value; it contains the
104
* last set cr3. This may not be the current effective cr3, because
105
* its update may be being lazily deferred. However, a vcpu looking
106
* at its own cr3 can use this value knowing that it everything will
107
* be self-consistent.
109
* xen_current_cr3 contains the actual vcpu cr3; it is set once the
110
* hypercall to set the vcpu cr3 is complete (so it may be a little
111
* out of date, but it will never be set early). If one vcpu is
112
* looking at another vcpu's cr3 value, it should use this variable.
114
DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
115
DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
119
* Just beyond the highest usermode address. STACK_TOP_MAX has a
120
* redzone above it, so round it up to a PGD boundary.
122
#define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
124
unsigned long arbitrary_virt_to_mfn(void *vaddr)
126
xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
128
return PFN_DOWN(maddr.maddr);
131
xmaddr_t arbitrary_virt_to_machine(void *vaddr)
133
unsigned long address = (unsigned long)vaddr;
139
* if the PFN is in the linear mapped vaddr range, we can just use
140
* the (quick) virt_to_machine() p2m lookup
142
if (virt_addr_valid(vaddr))
143
return virt_to_machine(vaddr);
145
/* otherwise we have to do a (slower) full page-table walk */
147
pte = lookup_address(address, &level);
149
offset = address & ~PAGE_MASK;
150
return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
152
EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
154
void make_lowmem_page_readonly(void *vaddr)
157
unsigned long address = (unsigned long)vaddr;
160
pte = lookup_address(address, &level);
162
return; /* vaddr missing */
164
ptev = pte_wrprotect(*pte);
166
if (HYPERVISOR_update_va_mapping(address, ptev, 0))
170
void make_lowmem_page_readwrite(void *vaddr)
173
unsigned long address = (unsigned long)vaddr;
176
pte = lookup_address(address, &level);
178
return; /* vaddr missing */
180
ptev = pte_mkwrite(*pte);
182
if (HYPERVISOR_update_va_mapping(address, ptev, 0))
187
static bool xen_page_pinned(void *ptr)
189
struct page *page = virt_to_page(ptr);
191
return PagePinned(page);
194
void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
196
struct multicall_space mcs;
197
struct mmu_update *u;
199
trace_xen_mmu_set_domain_pte(ptep, pteval, domid);
201
mcs = xen_mc_entry(sizeof(*u));
204
/* ptep might be kmapped when using 32-bit HIGHPTE */
205
u->ptr = virt_to_machine(ptep).maddr;
206
u->val = pte_val_ma(pteval);
208
MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
210
xen_mc_issue(PARAVIRT_LAZY_MMU);
212
EXPORT_SYMBOL_GPL(xen_set_domain_pte);
214
static void xen_extend_mmu_update(const struct mmu_update *update)
216
struct multicall_space mcs;
217
struct mmu_update *u;
219
mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
221
if (mcs.mc != NULL) {
224
mcs = __xen_mc_entry(sizeof(*u));
225
MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
232
static void xen_extend_mmuext_op(const struct mmuext_op *op)
234
struct multicall_space mcs;
237
mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
239
if (mcs.mc != NULL) {
242
mcs = __xen_mc_entry(sizeof(*u));
243
MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
250
static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
258
/* ptr may be ioremapped for 64-bit pagetable setup */
259
u.ptr = arbitrary_virt_to_machine(ptr).maddr;
260
u.val = pmd_val_ma(val);
261
xen_extend_mmu_update(&u);
263
xen_mc_issue(PARAVIRT_LAZY_MMU);
268
static void xen_set_pmd(pmd_t *ptr, pmd_t val)
270
trace_xen_mmu_set_pmd(ptr, val);
272
/* If page is not pinned, we can just update the entry
274
if (!xen_page_pinned(ptr)) {
279
xen_set_pmd_hyper(ptr, val);
283
* Associate a virtual page frame with a given physical page frame
284
* and protection flags for that frame.
286
void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
288
set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
291
static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
295
if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
300
u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
301
u.val = pte_val_ma(pteval);
302
xen_extend_mmu_update(&u);
304
xen_mc_issue(PARAVIRT_LAZY_MMU);
309
static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
311
if (!xen_batched_set_pte(ptep, pteval))
312
native_set_pte(ptep, pteval);
315
static void xen_set_pte(pte_t *ptep, pte_t pteval)
317
trace_xen_mmu_set_pte(ptep, pteval);
318
__xen_set_pte(ptep, pteval);
321
static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
322
pte_t *ptep, pte_t pteval)
324
trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
325
__xen_set_pte(ptep, pteval);
328
pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
329
unsigned long addr, pte_t *ptep)
331
/* Just return the pte as-is. We preserve the bits on commit */
332
trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
336
void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
337
pte_t *ptep, pte_t pte)
341
trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
344
u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
345
u.val = pte_val_ma(pte);
346
xen_extend_mmu_update(&u);
348
xen_mc_issue(PARAVIRT_LAZY_MMU);
351
/* Assume pteval_t is equivalent to all the other *val_t types. */
352
static pteval_t pte_mfn_to_pfn(pteval_t val)
354
if (val & _PAGE_PRESENT) {
355
unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
356
pteval_t flags = val & PTE_FLAGS_MASK;
357
val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags;
363
static pteval_t pte_pfn_to_mfn(pteval_t val)
365
if (val & _PAGE_PRESENT) {
366
unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
367
pteval_t flags = val & PTE_FLAGS_MASK;
370
if (!xen_feature(XENFEAT_auto_translated_physmap))
371
mfn = get_phys_to_machine(pfn);
375
* If there's no mfn for the pfn, then just create an
376
* empty non-present pte. Unfortunately this loses
377
* information about the original pfn, so
378
* pte_mfn_to_pfn is asymmetric.
380
if (unlikely(mfn == INVALID_P2M_ENTRY)) {
385
* Paramount to do this test _after_ the
386
* INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
387
* IDENTITY_FRAME_BIT resolves to true.
389
mfn &= ~FOREIGN_FRAME_BIT;
390
if (mfn & IDENTITY_FRAME_BIT) {
391
mfn &= ~IDENTITY_FRAME_BIT;
392
flags |= _PAGE_IOMAP;
395
val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
401
static pteval_t iomap_pte(pteval_t val)
403
if (val & _PAGE_PRESENT) {
404
unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
405
pteval_t flags = val & PTE_FLAGS_MASK;
407
/* We assume the pte frame number is a MFN, so
408
just use it as-is. */
409
val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
415
static pteval_t xen_pte_val(pte_t pte)
417
pteval_t pteval = pte.pte;
419
/* If this is a WC pte, convert back from Xen WC to Linux WC */
420
if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
421
WARN_ON(!pat_enabled);
422
pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
425
if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
428
return pte_mfn_to_pfn(pteval);
430
PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
432
static pgdval_t xen_pgd_val(pgd_t pgd)
434
return pte_mfn_to_pfn(pgd.pgd);
436
PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
439
* Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
440
* are reserved for now, to correspond to the Intel-reserved PAT
443
* We expect Linux's PAT set as follows:
445
* Idx PTE flags Linux Xen Default
452
* 6 PAT PCD UC- UC UC-
453
* 7 PAT PCD PWT UC UC UC
456
void xen_set_pat(u64 pat)
458
/* We expect Linux to use a PAT setting of
459
* UC UC- WC WB (ignoring the PAT flag) */
460
WARN_ON(pat != 0x0007010600070106ull);
463
static pte_t xen_make_pte(pteval_t pte)
465
phys_addr_t addr = (pte & PTE_PFN_MASK);
467
/* If Linux is trying to set a WC pte, then map to the Xen WC.
468
* If _PAGE_PAT is set, then it probably means it is really
469
* _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
470
* things work out OK...
472
* (We should never see kernel mappings with _PAGE_PSE set,
473
* but we could see hugetlbfs mappings, I think.).
475
if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
476
if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
477
pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
481
* Unprivileged domains are allowed to do IOMAPpings for
482
* PCI passthrough, but not map ISA space. The ISA
483
* mappings are just dummy local mappings to keep other
484
* parts of the kernel happy.
486
if (unlikely(pte & _PAGE_IOMAP) &&
487
(xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
488
pte = iomap_pte(pte);
491
pte = pte_pfn_to_mfn(pte);
494
return native_make_pte(pte);
496
PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
498
static pgd_t xen_make_pgd(pgdval_t pgd)
500
pgd = pte_pfn_to_mfn(pgd);
501
return native_make_pgd(pgd);
503
PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
505
static pmdval_t xen_pmd_val(pmd_t pmd)
507
return pte_mfn_to_pfn(pmd.pmd);
509
PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
511
static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
519
/* ptr may be ioremapped for 64-bit pagetable setup */
520
u.ptr = arbitrary_virt_to_machine(ptr).maddr;
521
u.val = pud_val_ma(val);
522
xen_extend_mmu_update(&u);
524
xen_mc_issue(PARAVIRT_LAZY_MMU);
529
static void xen_set_pud(pud_t *ptr, pud_t val)
531
trace_xen_mmu_set_pud(ptr, val);
533
/* If page is not pinned, we can just update the entry
535
if (!xen_page_pinned(ptr)) {
540
xen_set_pud_hyper(ptr, val);
543
#ifdef CONFIG_X86_PAE
544
static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
546
trace_xen_mmu_set_pte_atomic(ptep, pte);
547
set_64bit((u64 *)ptep, native_pte_val(pte));
550
static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
552
trace_xen_mmu_pte_clear(mm, addr, ptep);
553
if (!xen_batched_set_pte(ptep, native_make_pte(0)))
554
native_pte_clear(mm, addr, ptep);
557
static void xen_pmd_clear(pmd_t *pmdp)
559
trace_xen_mmu_pmd_clear(pmdp);
560
set_pmd(pmdp, __pmd(0));
562
#endif /* CONFIG_X86_PAE */
564
static pmd_t xen_make_pmd(pmdval_t pmd)
566
pmd = pte_pfn_to_mfn(pmd);
567
return native_make_pmd(pmd);
569
PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
571
#if PAGETABLE_LEVELS == 4
572
static pudval_t xen_pud_val(pud_t pud)
574
return pte_mfn_to_pfn(pud.pud);
576
PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
578
static pud_t xen_make_pud(pudval_t pud)
580
pud = pte_pfn_to_mfn(pud);
582
return native_make_pud(pud);
584
PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
586
static pgd_t *xen_get_user_pgd(pgd_t *pgd)
588
pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
589
unsigned offset = pgd - pgd_page;
590
pgd_t *user_ptr = NULL;
592
if (offset < pgd_index(USER_LIMIT)) {
593
struct page *page = virt_to_page(pgd_page);
594
user_ptr = (pgd_t *)page->private;
602
static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
606
u.ptr = virt_to_machine(ptr).maddr;
607
u.val = pgd_val_ma(val);
608
xen_extend_mmu_update(&u);
612
* Raw hypercall-based set_pgd, intended for in early boot before
613
* there's a page structure. This implies:
614
* 1. The only existing pagetable is the kernel's
615
* 2. It is always pinned
616
* 3. It has no user pagetable attached to it
618
static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
624
__xen_set_pgd_hyper(ptr, val);
626
xen_mc_issue(PARAVIRT_LAZY_MMU);
631
static void xen_set_pgd(pgd_t *ptr, pgd_t val)
633
pgd_t *user_ptr = xen_get_user_pgd(ptr);
635
trace_xen_mmu_set_pgd(ptr, user_ptr, val);
637
/* If page is not pinned, we can just update the entry
639
if (!xen_page_pinned(ptr)) {
642
WARN_ON(xen_page_pinned(user_ptr));
648
/* If it's pinned, then we can at least batch the kernel and
649
user updates together. */
652
__xen_set_pgd_hyper(ptr, val);
654
__xen_set_pgd_hyper(user_ptr, val);
656
xen_mc_issue(PARAVIRT_LAZY_MMU);
658
#endif /* PAGETABLE_LEVELS == 4 */
661
* (Yet another) pagetable walker. This one is intended for pinning a
662
* pagetable. This means that it walks a pagetable and calls the
663
* callback function on each page it finds making up the page table,
664
* at every level. It walks the entire pagetable, but it only bothers
665
* pinning pte pages which are below limit. In the normal case this
666
* will be STACK_TOP_MAX, but at boot we need to pin up to
669
* For 32-bit the important bit is that we don't pin beyond there,
670
* because then we start getting into Xen's ptes.
672
* For 64-bit, we must skip the Xen hole in the middle of the address
673
* space, just after the big x86-64 virtual hole.
675
static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
676
int (*func)(struct mm_struct *mm, struct page *,
681
unsigned hole_low, hole_high;
682
unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
683
unsigned pgdidx, pudidx, pmdidx;
685
/* The limit is the last byte to be touched */
687
BUG_ON(limit >= FIXADDR_TOP);
689
if (xen_feature(XENFEAT_auto_translated_physmap))
693
* 64-bit has a great big hole in the middle of the address
694
* space, which contains the Xen mappings. On 32-bit these
695
* will end up making a zero-sized hole and so is a no-op.
697
hole_low = pgd_index(USER_LIMIT);
698
hole_high = pgd_index(PAGE_OFFSET);
700
pgdidx_limit = pgd_index(limit);
702
pudidx_limit = pud_index(limit);
707
pmdidx_limit = pmd_index(limit);
712
for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
715
if (pgdidx >= hole_low && pgdidx < hole_high)
718
if (!pgd_val(pgd[pgdidx]))
721
pud = pud_offset(&pgd[pgdidx], 0);
723
if (PTRS_PER_PUD > 1) /* not folded */
724
flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
726
for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
729
if (pgdidx == pgdidx_limit &&
730
pudidx > pudidx_limit)
733
if (pud_none(pud[pudidx]))
736
pmd = pmd_offset(&pud[pudidx], 0);
738
if (PTRS_PER_PMD > 1) /* not folded */
739
flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
741
for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
744
if (pgdidx == pgdidx_limit &&
745
pudidx == pudidx_limit &&
746
pmdidx > pmdidx_limit)
749
if (pmd_none(pmd[pmdidx]))
752
pte = pmd_page(pmd[pmdidx]);
753
flush |= (*func)(mm, pte, PT_PTE);
759
/* Do the top level last, so that the callbacks can use it as
760
a cue to do final things like tlb flushes. */
761
flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
766
static int xen_pgd_walk(struct mm_struct *mm,
767
int (*func)(struct mm_struct *mm, struct page *,
771
return __xen_pgd_walk(mm, mm->pgd, func, limit);
774
/* If we're using split pte locks, then take the page's lock and
775
return a pointer to it. Otherwise return NULL. */
776
static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
778
spinlock_t *ptl = NULL;
780
#if USE_SPLIT_PTLOCKS
781
ptl = __pte_lockptr(page);
782
spin_lock_nest_lock(ptl, &mm->page_table_lock);
788
static void xen_pte_unlock(void *v)
794
static void xen_do_pin(unsigned level, unsigned long pfn)
799
op.arg1.mfn = pfn_to_mfn(pfn);
801
xen_extend_mmuext_op(&op);
804
static int xen_pin_page(struct mm_struct *mm, struct page *page,
807
unsigned pgfl = TestSetPagePinned(page);
811
flush = 0; /* already pinned */
812
else if (PageHighMem(page))
813
/* kmaps need flushing if we found an unpinned
817
void *pt = lowmem_page_address(page);
818
unsigned long pfn = page_to_pfn(page);
819
struct multicall_space mcs = __xen_mc_entry(0);
825
* We need to hold the pagetable lock between the time
826
* we make the pagetable RO and when we actually pin
827
* it. If we don't, then other users may come in and
828
* attempt to update the pagetable by writing it,
829
* which will fail because the memory is RO but not
830
* pinned, so Xen won't do the trap'n'emulate.
832
* If we're using split pte locks, we can't hold the
833
* entire pagetable's worth of locks during the
834
* traverse, because we may wrap the preempt count (8
835
* bits). The solution is to mark RO and pin each PTE
836
* page while holding the lock. This means the number
837
* of locks we end up holding is never more than a
838
* batch size (~32 entries, at present).
840
* If we're not using split pte locks, we needn't pin
841
* the PTE pages independently, because we're
842
* protected by the overall pagetable lock.
846
ptl = xen_pte_lock(page, mm);
848
MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
849
pfn_pte(pfn, PAGE_KERNEL_RO),
850
level == PT_PGD ? UVMF_TLB_FLUSH : 0);
853
xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
855
/* Queue a deferred unlock for when this batch
857
xen_mc_callback(xen_pte_unlock, ptl);
864
/* This is called just after a mm has been created, but it has not
865
been used yet. We need to make sure that its pagetable is all
866
read-only, and can be pinned. */
867
static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
869
trace_xen_mmu_pgd_pin(mm, pgd);
873
if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
874
/* re-enable interrupts for flushing */
884
pgd_t *user_pgd = xen_get_user_pgd(pgd);
886
xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
889
xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
890
xen_do_pin(MMUEXT_PIN_L4_TABLE,
891
PFN_DOWN(__pa(user_pgd)));
894
#else /* CONFIG_X86_32 */
895
#ifdef CONFIG_X86_PAE
896
/* Need to make sure unshared kernel PMD is pinnable */
897
xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
900
xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
901
#endif /* CONFIG_X86_64 */
905
static void xen_pgd_pin(struct mm_struct *mm)
907
__xen_pgd_pin(mm, mm->pgd);
911
* On save, we need to pin all pagetables to make sure they get their
912
* mfns turned into pfns. Search the list for any unpinned pgds and pin
913
* them (unpinned pgds are not currently in use, probably because the
914
* process is under construction or destruction).
916
* Expected to be called in stop_machine() ("equivalent to taking
917
* every spinlock in the system"), so the locking doesn't really
918
* matter all that much.
920
void xen_mm_pin_all(void)
924
spin_lock(&pgd_lock);
926
list_for_each_entry(page, &pgd_list, lru) {
927
if (!PagePinned(page)) {
928
__xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
929
SetPageSavePinned(page);
933
spin_unlock(&pgd_lock);
937
* The init_mm pagetable is really pinned as soon as its created, but
938
* that's before we have page structures to store the bits. So do all
939
* the book-keeping now.
941
static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
948
static void __init xen_mark_init_mm_pinned(void)
950
xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
953
static int xen_unpin_page(struct mm_struct *mm, struct page *page,
956
unsigned pgfl = TestClearPagePinned(page);
958
if (pgfl && !PageHighMem(page)) {
959
void *pt = lowmem_page_address(page);
960
unsigned long pfn = page_to_pfn(page);
961
spinlock_t *ptl = NULL;
962
struct multicall_space mcs;
965
* Do the converse to pin_page. If we're using split
966
* pte locks, we must be holding the lock for while
967
* the pte page is unpinned but still RO to prevent
968
* concurrent updates from seeing it in this
969
* partially-pinned state.
971
if (level == PT_PTE) {
972
ptl = xen_pte_lock(page, mm);
975
xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
978
mcs = __xen_mc_entry(0);
980
MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
981
pfn_pte(pfn, PAGE_KERNEL),
982
level == PT_PGD ? UVMF_TLB_FLUSH : 0);
985
/* unlock when batch completed */
986
xen_mc_callback(xen_pte_unlock, ptl);
990
return 0; /* never need to flush on unpin */
993
/* Release a pagetables pages back as normal RW */
994
static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
996
trace_xen_mmu_pgd_unpin(mm, pgd);
1000
xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1002
#ifdef CONFIG_X86_64
1004
pgd_t *user_pgd = xen_get_user_pgd(pgd);
1007
xen_do_pin(MMUEXT_UNPIN_TABLE,
1008
PFN_DOWN(__pa(user_pgd)));
1009
xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1014
#ifdef CONFIG_X86_PAE
1015
/* Need to make sure unshared kernel PMD is unpinned */
1016
xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1020
__xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1025
static void xen_pgd_unpin(struct mm_struct *mm)
1027
__xen_pgd_unpin(mm, mm->pgd);
1031
* On resume, undo any pinning done at save, so that the rest of the
1032
* kernel doesn't see any unexpected pinned pagetables.
1034
void xen_mm_unpin_all(void)
1038
spin_lock(&pgd_lock);
1040
list_for_each_entry(page, &pgd_list, lru) {
1041
if (PageSavePinned(page)) {
1042
BUG_ON(!PagePinned(page));
1043
__xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1044
ClearPageSavePinned(page);
1048
spin_unlock(&pgd_lock);
1051
static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1053
spin_lock(&next->page_table_lock);
1055
spin_unlock(&next->page_table_lock);
1058
static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1060
spin_lock(&mm->page_table_lock);
1062
spin_unlock(&mm->page_table_lock);
1067
/* Another cpu may still have their %cr3 pointing at the pagetable, so
1068
we need to repoint it somewhere else before we can unpin it. */
1069
static void drop_other_mm_ref(void *info)
1071
struct mm_struct *mm = info;
1072
struct mm_struct *active_mm;
1074
active_mm = percpu_read(cpu_tlbstate.active_mm);
1076
if (active_mm == mm && percpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
1077
leave_mm(smp_processor_id());
1079
/* If this cpu still has a stale cr3 reference, then make sure
1080
it has been flushed. */
1081
if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1082
load_cr3(swapper_pg_dir);
1085
static void xen_drop_mm_ref(struct mm_struct *mm)
1090
if (current->active_mm == mm) {
1091
if (current->mm == mm)
1092
load_cr3(swapper_pg_dir);
1094
leave_mm(smp_processor_id());
1097
/* Get the "official" set of cpus referring to our pagetable. */
1098
if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1099
for_each_online_cpu(cpu) {
1100
if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1101
&& per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1103
smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1107
cpumask_copy(mask, mm_cpumask(mm));
1109
/* It's possible that a vcpu may have a stale reference to our
1110
cr3, because its in lazy mode, and it hasn't yet flushed
1111
its set of pending hypercalls yet. In this case, we can
1112
look at its actual current cr3 value, and force it to flush
1114
for_each_online_cpu(cpu) {
1115
if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1116
cpumask_set_cpu(cpu, mask);
1119
if (!cpumask_empty(mask))
1120
smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1121
free_cpumask_var(mask);
1124
static void xen_drop_mm_ref(struct mm_struct *mm)
1126
if (current->active_mm == mm)
1127
load_cr3(swapper_pg_dir);
1132
* While a process runs, Xen pins its pagetables, which means that the
1133
* hypervisor forces it to be read-only, and it controls all updates
1134
* to it. This means that all pagetable updates have to go via the
1135
* hypervisor, which is moderately expensive.
1137
* Since we're pulling the pagetable down, we switch to use init_mm,
1138
* unpin old process pagetable and mark it all read-write, which
1139
* allows further operations on it to be simple memory accesses.
1141
* The only subtle point is that another CPU may be still using the
1142
* pagetable because of lazy tlb flushing. This means we need need to
1143
* switch all CPUs off this pagetable before we can unpin it.
1145
static void xen_exit_mmap(struct mm_struct *mm)
1147
get_cpu(); /* make sure we don't move around */
1148
xen_drop_mm_ref(mm);
1151
spin_lock(&mm->page_table_lock);
1153
/* pgd may not be pinned in the error exit path of execve */
1154
if (xen_page_pinned(mm->pgd))
1157
spin_unlock(&mm->page_table_lock);
1160
static void __init xen_pagetable_setup_start(pgd_t *base)
1164
static __init void xen_mapping_pagetable_reserve(u64 start, u64 end)
1166
/* reserve the range used */
1167
native_pagetable_reserve(start, end);
1169
/* set as RW the rest */
1170
printk(KERN_DEBUG "xen: setting RW the range %llx - %llx\n", end,
1171
PFN_PHYS(pgt_buf_top));
1172
while (end < PFN_PHYS(pgt_buf_top)) {
1173
make_lowmem_page_readwrite(__va(end));
1178
static void xen_post_allocator_init(void);
1180
static void __init xen_pagetable_setup_done(pgd_t *base)
1182
xen_setup_shared_info();
1183
xen_post_allocator_init();
1186
static void xen_write_cr2(unsigned long cr2)
1188
percpu_read(xen_vcpu)->arch.cr2 = cr2;
1191
static unsigned long xen_read_cr2(void)
1193
return percpu_read(xen_vcpu)->arch.cr2;
1196
unsigned long xen_read_cr2_direct(void)
1198
return percpu_read(xen_vcpu_info.arch.cr2);
1201
static void xen_flush_tlb(void)
1203
struct mmuext_op *op;
1204
struct multicall_space mcs;
1206
trace_xen_mmu_flush_tlb(0);
1210
mcs = xen_mc_entry(sizeof(*op));
1213
op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1214
MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1216
xen_mc_issue(PARAVIRT_LAZY_MMU);
1221
static void xen_flush_tlb_single(unsigned long addr)
1223
struct mmuext_op *op;
1224
struct multicall_space mcs;
1226
trace_xen_mmu_flush_tlb_single(addr);
1230
mcs = xen_mc_entry(sizeof(*op));
1232
op->cmd = MMUEXT_INVLPG_LOCAL;
1233
op->arg1.linear_addr = addr & PAGE_MASK;
1234
MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1236
xen_mc_issue(PARAVIRT_LAZY_MMU);
1241
static void xen_flush_tlb_others(const struct cpumask *cpus,
1242
struct mm_struct *mm, unsigned long va)
1245
struct mmuext_op op;
1247
DECLARE_BITMAP(mask, num_processors);
1249
DECLARE_BITMAP(mask, NR_CPUS);
1252
struct multicall_space mcs;
1254
trace_xen_mmu_flush_tlb_others(cpus, mm, va);
1256
if (cpumask_empty(cpus))
1257
return; /* nothing to do */
1259
mcs = xen_mc_entry(sizeof(*args));
1261
args->op.arg2.vcpumask = to_cpumask(args->mask);
1263
/* Remove us, and any offline CPUS. */
1264
cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1265
cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1267
if (va == TLB_FLUSH_ALL) {
1268
args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1270
args->op.cmd = MMUEXT_INVLPG_MULTI;
1271
args->op.arg1.linear_addr = va;
1274
MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1276
xen_mc_issue(PARAVIRT_LAZY_MMU);
1279
static unsigned long xen_read_cr3(void)
1281
return percpu_read(xen_cr3);
1284
static void set_current_cr3(void *v)
1286
percpu_write(xen_current_cr3, (unsigned long)v);
1289
static void __xen_write_cr3(bool kernel, unsigned long cr3)
1291
struct mmuext_op op;
1294
trace_xen_mmu_write_cr3(kernel, cr3);
1297
mfn = pfn_to_mfn(PFN_DOWN(cr3));
1301
WARN_ON(mfn == 0 && kernel);
1303
op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1306
xen_extend_mmuext_op(&op);
1309
percpu_write(xen_cr3, cr3);
1311
/* Update xen_current_cr3 once the batch has actually
1313
xen_mc_callback(set_current_cr3, (void *)cr3);
1317
static void xen_write_cr3(unsigned long cr3)
1319
BUG_ON(preemptible());
1321
xen_mc_batch(); /* disables interrupts */
1323
/* Update while interrupts are disabled, so its atomic with
1325
percpu_write(xen_cr3, cr3);
1327
__xen_write_cr3(true, cr3);
1329
#ifdef CONFIG_X86_64
1331
pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1333
__xen_write_cr3(false, __pa(user_pgd));
1335
__xen_write_cr3(false, 0);
1339
xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1342
static int xen_pgd_alloc(struct mm_struct *mm)
1344
pgd_t *pgd = mm->pgd;
1347
BUG_ON(PagePinned(virt_to_page(pgd)));
1349
#ifdef CONFIG_X86_64
1351
struct page *page = virt_to_page(pgd);
1354
BUG_ON(page->private != 0);
1358
user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1359
page->private = (unsigned long)user_pgd;
1361
if (user_pgd != NULL) {
1362
user_pgd[pgd_index(VSYSCALL_START)] =
1363
__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1367
BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1374
static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1376
#ifdef CONFIG_X86_64
1377
pgd_t *user_pgd = xen_get_user_pgd(pgd);
1380
free_page((unsigned long)user_pgd);
1384
#ifdef CONFIG_X86_32
1385
static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1387
/* If there's an existing pte, then don't allow _PAGE_RW to be set */
1388
if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1389
pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1394
#else /* CONFIG_X86_64 */
1395
static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1397
unsigned long pfn = pte_pfn(pte);
1400
* If the new pfn is within the range of the newly allocated
1401
* kernel pagetable, and it isn't being mapped into an
1402
* early_ioremap fixmap slot as a freshly allocated page, make sure
1405
if (((!is_early_ioremap_ptep(ptep) &&
1406
pfn >= pgt_buf_start && pfn < pgt_buf_top)) ||
1407
(is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1408
pte = pte_wrprotect(pte);
1412
#endif /* CONFIG_X86_64 */
1414
/* Init-time set_pte while constructing initial pagetables, which
1415
doesn't allow RO pagetable pages to be remapped RW */
1416
static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1418
pte = mask_rw_pte(ptep, pte);
1420
xen_set_pte(ptep, pte);
1423
static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1425
struct mmuext_op op;
1427
op.arg1.mfn = pfn_to_mfn(pfn);
1428
if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1432
/* Early in boot, while setting up the initial pagetable, assume
1433
everything is pinned. */
1434
static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1436
#ifdef CONFIG_FLATMEM
1437
BUG_ON(mem_map); /* should only be used early */
1439
make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1440
pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1443
/* Used for pmd and pud */
1444
static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1446
#ifdef CONFIG_FLATMEM
1447
BUG_ON(mem_map); /* should only be used early */
1449
make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1452
/* Early release_pte assumes that all pts are pinned, since there's
1453
only init_mm and anything attached to that is pinned. */
1454
static void __init xen_release_pte_init(unsigned long pfn)
1456
pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1457
make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1460
static void __init xen_release_pmd_init(unsigned long pfn)
1462
make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1465
static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1467
struct multicall_space mcs;
1468
struct mmuext_op *op;
1470
mcs = __xen_mc_entry(sizeof(*op));
1473
op->arg1.mfn = pfn_to_mfn(pfn);
1475
MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1478
static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1480
struct multicall_space mcs;
1481
unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1483
mcs = __xen_mc_entry(0);
1484
MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1485
pfn_pte(pfn, prot), 0);
1488
/* This needs to make sure the new pte page is pinned iff its being
1489
attached to a pinned pagetable. */
1490
static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1493
bool pinned = PagePinned(virt_to_page(mm->pgd));
1495
trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1498
struct page *page = pfn_to_page(pfn);
1500
SetPagePinned(page);
1502
if (!PageHighMem(page)) {
1505
__set_pfn_prot(pfn, PAGE_KERNEL_RO);
1507
if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1508
__pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1510
xen_mc_issue(PARAVIRT_LAZY_MMU);
1512
/* make sure there are no stray mappings of
1514
kmap_flush_unused();
1519
static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1521
xen_alloc_ptpage(mm, pfn, PT_PTE);
1524
static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1526
xen_alloc_ptpage(mm, pfn, PT_PMD);
1529
/* This should never happen until we're OK to use struct page */
1530
static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1532
struct page *page = pfn_to_page(pfn);
1533
bool pinned = PagePinned(page);
1535
trace_xen_mmu_release_ptpage(pfn, level, pinned);
1538
if (!PageHighMem(page)) {
1541
if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1542
__pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1544
__set_pfn_prot(pfn, PAGE_KERNEL);
1546
xen_mc_issue(PARAVIRT_LAZY_MMU);
1548
ClearPagePinned(page);
1552
static void xen_release_pte(unsigned long pfn)
1554
xen_release_ptpage(pfn, PT_PTE);
1557
static void xen_release_pmd(unsigned long pfn)
1559
xen_release_ptpage(pfn, PT_PMD);
1562
#if PAGETABLE_LEVELS == 4
1563
static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1565
xen_alloc_ptpage(mm, pfn, PT_PUD);
1568
static void xen_release_pud(unsigned long pfn)
1570
xen_release_ptpage(pfn, PT_PUD);
1574
void __init xen_reserve_top(void)
1576
#ifdef CONFIG_X86_32
1577
unsigned long top = HYPERVISOR_VIRT_START;
1578
struct xen_platform_parameters pp;
1580
if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1581
top = pp.virt_start;
1583
reserve_top_address(-top);
1584
#endif /* CONFIG_X86_32 */
1588
* Like __va(), but returns address in the kernel mapping (which is
1589
* all we have until the physical memory mapping has been set up.
1591
static void *__ka(phys_addr_t paddr)
1593
#ifdef CONFIG_X86_64
1594
return (void *)(paddr + __START_KERNEL_map);
1600
/* Convert a machine address to physical address */
1601
static unsigned long m2p(phys_addr_t maddr)
1605
maddr &= PTE_PFN_MASK;
1606
paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1611
/* Convert a machine address to kernel virtual */
1612
static void *m2v(phys_addr_t maddr)
1614
return __ka(m2p(maddr));
1617
/* Set the page permissions on an identity-mapped pages */
1618
static void set_page_prot(void *addr, pgprot_t prot)
1620
unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1621
pte_t pte = pfn_pte(pfn, prot);
1623
if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1627
static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1629
unsigned pmdidx, pteidx;
1633
level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1638
for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1641
/* Reuse or allocate a page of ptes */
1642
if (pmd_present(pmd[pmdidx]))
1643
pte_page = m2v(pmd[pmdidx].pmd);
1645
/* Check for free pte pages */
1646
if (ident_pte == LEVEL1_IDENT_ENTRIES)
1649
pte_page = &level1_ident_pgt[ident_pte];
1650
ident_pte += PTRS_PER_PTE;
1652
pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1655
/* Install mappings */
1656
for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1659
#ifdef CONFIG_X86_32
1660
if (pfn > max_pfn_mapped)
1661
max_pfn_mapped = pfn;
1664
if (!pte_none(pte_page[pteidx]))
1667
pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1668
pte_page[pteidx] = pte;
1672
for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1673
set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1675
set_page_prot(pmd, PAGE_KERNEL_RO);
1678
void __init xen_setup_machphys_mapping(void)
1680
struct xen_machphys_mapping mapping;
1682
if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1683
machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1684
machine_to_phys_nr = mapping.max_mfn + 1;
1686
machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1688
#ifdef CONFIG_X86_32
1689
WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1690
< machine_to_phys_mapping);
1694
#ifdef CONFIG_X86_64
1695
static void convert_pfn_mfn(void *v)
1700
/* All levels are converted the same way, so just treat them
1702
for (i = 0; i < PTRS_PER_PTE; i++)
1703
pte[i] = xen_make_pte(pte[i].pte);
1707
* Set up the initial kernel pagetable.
1709
* We can construct this by grafting the Xen provided pagetable into
1710
* head_64.S's preconstructed pagetables. We copy the Xen L2's into
1711
* level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1712
* means that only the kernel has a physical mapping to start with -
1713
* but that's enough to get __va working. We need to fill in the rest
1714
* of the physical mapping once some sort of allocator has been set
1717
pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1718
unsigned long max_pfn)
1723
/* max_pfn_mapped is the last pfn mapped in the initial memory
1724
* mappings. Considering that on Xen after the kernel mappings we
1725
* have the mappings of some pages that don't exist in pfn space, we
1726
* set max_pfn_mapped to the last real pfn mapped. */
1727
max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1729
/* Zap identity mapping */
1730
init_level4_pgt[0] = __pgd(0);
1732
/* Pre-constructed entries are in pfn, so convert to mfn */
1733
convert_pfn_mfn(init_level4_pgt);
1734
convert_pfn_mfn(level3_ident_pgt);
1735
convert_pfn_mfn(level3_kernel_pgt);
1737
l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1738
l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1740
memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1741
memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1743
l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1744
l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1745
memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1747
/* Set up identity map */
1748
xen_map_identity_early(level2_ident_pgt, max_pfn);
1750
/* Make pagetable pieces RO */
1751
set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1752
set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1753
set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1754
set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1755
set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1756
set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1758
/* Pin down new L4 */
1759
pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1760
PFN_DOWN(__pa_symbol(init_level4_pgt)));
1762
/* Unpin Xen-provided one */
1763
pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1766
pgd = init_level4_pgt;
1769
* At this stage there can be no user pgd, and no page
1770
* structure to attach it to, so make sure we just set kernel
1774
__xen_write_cr3(true, __pa(pgd));
1775
xen_mc_issue(PARAVIRT_LAZY_CPU);
1777
memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1778
__pa(xen_start_info->pt_base +
1779
xen_start_info->nr_pt_frames * PAGE_SIZE),
1784
#else /* !CONFIG_X86_64 */
1785
static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1786
static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1788
static void __init xen_write_cr3_init(unsigned long cr3)
1790
unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1792
BUG_ON(read_cr3() != __pa(initial_page_table));
1793
BUG_ON(cr3 != __pa(swapper_pg_dir));
1796
* We are switching to swapper_pg_dir for the first time (from
1797
* initial_page_table) and therefore need to mark that page
1798
* read-only and then pin it.
1800
* Xen disallows sharing of kernel PMDs for PAE
1801
* guests. Therefore we must copy the kernel PMD from
1802
* initial_page_table into a new kernel PMD to be used in
1805
swapper_kernel_pmd =
1806
extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1807
memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1808
sizeof(pmd_t) * PTRS_PER_PMD);
1809
swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1810
__pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1811
set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1813
set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1815
pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1817
pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1818
PFN_DOWN(__pa(initial_page_table)));
1819
set_page_prot(initial_page_table, PAGE_KERNEL);
1820
set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1822
pv_mmu_ops.write_cr3 = &xen_write_cr3;
1825
pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1826
unsigned long max_pfn)
1830
initial_kernel_pmd =
1831
extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1833
max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1834
xen_start_info->nr_pt_frames * PAGE_SIZE +
1837
kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1838
memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1840
xen_map_identity_early(initial_kernel_pmd, max_pfn);
1842
memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1843
initial_page_table[KERNEL_PGD_BOUNDARY] =
1844
__pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1846
set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1847
set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1848
set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1850
pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1852
pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1853
PFN_DOWN(__pa(initial_page_table)));
1854
xen_write_cr3(__pa(initial_page_table));
1856
memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1857
__pa(xen_start_info->pt_base +
1858
xen_start_info->nr_pt_frames * PAGE_SIZE),
1861
return initial_page_table;
1863
#endif /* CONFIG_X86_64 */
1865
static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1867
static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1871
phys >>= PAGE_SHIFT;
1874
case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1875
#ifdef CONFIG_X86_F00F_BUG
1878
#ifdef CONFIG_X86_32
1881
# ifdef CONFIG_HIGHMEM
1882
case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1885
case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1888
case FIX_TEXT_POKE0:
1889
case FIX_TEXT_POKE1:
1890
/* All local page mappings */
1891
pte = pfn_pte(phys, prot);
1894
#ifdef CONFIG_X86_LOCAL_APIC
1895
case FIX_APIC_BASE: /* maps dummy local APIC */
1896
pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1900
#ifdef CONFIG_X86_IO_APIC
1901
case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1903
* We just don't map the IO APIC - all access is via
1904
* hypercalls. Keep the address in the pte for reference.
1906
pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1910
case FIX_PARAVIRT_BOOTMAP:
1911
/* This is an MFN, but it isn't an IO mapping from the
1913
pte = mfn_pte(phys, prot);
1917
/* By default, set_fixmap is used for hardware mappings */
1918
pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1922
__native_set_fixmap(idx, pte);
1924
#ifdef CONFIG_X86_64
1925
/* Replicate changes to map the vsyscall page into the user
1926
pagetable vsyscall mapping. */
1927
if ((idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) ||
1929
unsigned long vaddr = __fix_to_virt(idx);
1930
set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1935
void __init xen_ident_map_ISA(void)
1940
* If we're dom0, then linear map the ISA machine addresses into
1941
* the kernel's address space.
1943
if (!xen_initial_domain())
1946
xen_raw_printk("Xen: setup ISA identity maps\n");
1948
for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1949
pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1951
if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1958
static void __init xen_post_allocator_init(void)
1960
pv_mmu_ops.set_pte = xen_set_pte;
1961
pv_mmu_ops.set_pmd = xen_set_pmd;
1962
pv_mmu_ops.set_pud = xen_set_pud;
1963
#if PAGETABLE_LEVELS == 4
1964
pv_mmu_ops.set_pgd = xen_set_pgd;
1967
/* This will work as long as patching hasn't happened yet
1968
(which it hasn't) */
1969
pv_mmu_ops.alloc_pte = xen_alloc_pte;
1970
pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1971
pv_mmu_ops.release_pte = xen_release_pte;
1972
pv_mmu_ops.release_pmd = xen_release_pmd;
1973
#if PAGETABLE_LEVELS == 4
1974
pv_mmu_ops.alloc_pud = xen_alloc_pud;
1975
pv_mmu_ops.release_pud = xen_release_pud;
1978
#ifdef CONFIG_X86_64
1979
SetPagePinned(virt_to_page(level3_user_vsyscall));
1981
xen_mark_init_mm_pinned();
1984
static void xen_leave_lazy_mmu(void)
1988
paravirt_leave_lazy_mmu();
1992
static const struct pv_mmu_ops xen_mmu_ops __initconst = {
1993
.read_cr2 = xen_read_cr2,
1994
.write_cr2 = xen_write_cr2,
1996
.read_cr3 = xen_read_cr3,
1997
#ifdef CONFIG_X86_32
1998
.write_cr3 = xen_write_cr3_init,
2000
.write_cr3 = xen_write_cr3,
2003
.flush_tlb_user = xen_flush_tlb,
2004
.flush_tlb_kernel = xen_flush_tlb,
2005
.flush_tlb_single = xen_flush_tlb_single,
2006
.flush_tlb_others = xen_flush_tlb_others,
2008
.pte_update = paravirt_nop,
2009
.pte_update_defer = paravirt_nop,
2011
.pgd_alloc = xen_pgd_alloc,
2012
.pgd_free = xen_pgd_free,
2014
.alloc_pte = xen_alloc_pte_init,
2015
.release_pte = xen_release_pte_init,
2016
.alloc_pmd = xen_alloc_pmd_init,
2017
.release_pmd = xen_release_pmd_init,
2019
.set_pte = xen_set_pte_init,
2020
.set_pte_at = xen_set_pte_at,
2021
.set_pmd = xen_set_pmd_hyper,
2023
.ptep_modify_prot_start = __ptep_modify_prot_start,
2024
.ptep_modify_prot_commit = __ptep_modify_prot_commit,
2026
.pte_val = PV_CALLEE_SAVE(xen_pte_val),
2027
.pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2029
.make_pte = PV_CALLEE_SAVE(xen_make_pte),
2030
.make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2032
#ifdef CONFIG_X86_PAE
2033
.set_pte_atomic = xen_set_pte_atomic,
2034
.pte_clear = xen_pte_clear,
2035
.pmd_clear = xen_pmd_clear,
2036
#endif /* CONFIG_X86_PAE */
2037
.set_pud = xen_set_pud_hyper,
2039
.make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2040
.pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2042
#if PAGETABLE_LEVELS == 4
2043
.pud_val = PV_CALLEE_SAVE(xen_pud_val),
2044
.make_pud = PV_CALLEE_SAVE(xen_make_pud),
2045
.set_pgd = xen_set_pgd_hyper,
2047
.alloc_pud = xen_alloc_pmd_init,
2048
.release_pud = xen_release_pmd_init,
2049
#endif /* PAGETABLE_LEVELS == 4 */
2051
.activate_mm = xen_activate_mm,
2052
.dup_mmap = xen_dup_mmap,
2053
.exit_mmap = xen_exit_mmap,
2056
.enter = paravirt_enter_lazy_mmu,
2057
.leave = xen_leave_lazy_mmu,
2060
.set_fixmap = xen_set_fixmap,
2063
void __init xen_init_mmu_ops(void)
2065
x86_init.mapping.pagetable_reserve = xen_mapping_pagetable_reserve;
2066
x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2067
x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2068
pv_mmu_ops = xen_mmu_ops;
2070
memset(dummy_mapping, 0xff, PAGE_SIZE);
2073
/* Protected by xen_reservation_lock. */
2074
#define MAX_CONTIG_ORDER 9 /* 2MB */
2075
static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2077
#define VOID_PTE (mfn_pte(0, __pgprot(0)))
2078
static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2079
unsigned long *in_frames,
2080
unsigned long *out_frames)
2083
struct multicall_space mcs;
2086
for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2087
mcs = __xen_mc_entry(0);
2090
in_frames[i] = virt_to_mfn(vaddr);
2092
MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2093
__set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2096
out_frames[i] = virt_to_pfn(vaddr);
2102
* Update the pfn-to-mfn mappings for a virtual address range, either to
2103
* point to an array of mfns, or contiguously from a single starting
2106
static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2107
unsigned long *mfns,
2108
unsigned long first_mfn)
2115
limit = 1u << order;
2116
for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2117
struct multicall_space mcs;
2120
mcs = __xen_mc_entry(0);
2124
mfn = first_mfn + i;
2126
if (i < (limit - 1))
2130
flags = UVMF_INVLPG | UVMF_ALL;
2132
flags = UVMF_TLB_FLUSH | UVMF_ALL;
2135
MULTI_update_va_mapping(mcs.mc, vaddr,
2136
mfn_pte(mfn, PAGE_KERNEL), flags);
2138
set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2145
* Perform the hypercall to exchange a region of our pfns to point to
2146
* memory with the required contiguous alignment. Takes the pfns as
2147
* input, and populates mfns as output.
2149
* Returns a success code indicating whether the hypervisor was able to
2150
* satisfy the request or not.
2152
static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2153
unsigned long *pfns_in,
2154
unsigned long extents_out,
2155
unsigned int order_out,
2156
unsigned long *mfns_out,
2157
unsigned int address_bits)
2162
struct xen_memory_exchange exchange = {
2164
.nr_extents = extents_in,
2165
.extent_order = order_in,
2166
.extent_start = pfns_in,
2170
.nr_extents = extents_out,
2171
.extent_order = order_out,
2172
.extent_start = mfns_out,
2173
.address_bits = address_bits,
2178
BUG_ON(extents_in << order_in != extents_out << order_out);
2180
rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2181
success = (exchange.nr_exchanged == extents_in);
2183
BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2184
BUG_ON(success && (rc != 0));
2189
int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2190
unsigned int address_bits)
2192
unsigned long *in_frames = discontig_frames, out_frame;
2193
unsigned long flags;
2197
* Currently an auto-translated guest will not perform I/O, nor will
2198
* it require PAE page directories below 4GB. Therefore any calls to
2199
* this function are redundant and can be ignored.
2202
if (xen_feature(XENFEAT_auto_translated_physmap))
2205
if (unlikely(order > MAX_CONTIG_ORDER))
2208
memset((void *) vstart, 0, PAGE_SIZE << order);
2210
spin_lock_irqsave(&xen_reservation_lock, flags);
2212
/* 1. Zap current PTEs, remembering MFNs. */
2213
xen_zap_pfn_range(vstart, order, in_frames, NULL);
2215
/* 2. Get a new contiguous memory extent. */
2216
out_frame = virt_to_pfn(vstart);
2217
success = xen_exchange_memory(1UL << order, 0, in_frames,
2218
1, order, &out_frame,
2221
/* 3. Map the new extent in place of old pages. */
2223
xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2225
xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2227
spin_unlock_irqrestore(&xen_reservation_lock, flags);
2229
return success ? 0 : -ENOMEM;
2231
EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2233
void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2235
unsigned long *out_frames = discontig_frames, in_frame;
2236
unsigned long flags;
2239
if (xen_feature(XENFEAT_auto_translated_physmap))
2242
if (unlikely(order > MAX_CONTIG_ORDER))
2245
memset((void *) vstart, 0, PAGE_SIZE << order);
2247
spin_lock_irqsave(&xen_reservation_lock, flags);
2249
/* 1. Find start MFN of contiguous extent. */
2250
in_frame = virt_to_mfn(vstart);
2252
/* 2. Zap current PTEs. */
2253
xen_zap_pfn_range(vstart, order, NULL, out_frames);
2255
/* 3. Do the exchange for non-contiguous MFNs. */
2256
success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2259
/* 4. Map new pages in place of old pages. */
2261
xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2263
xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2265
spin_unlock_irqrestore(&xen_reservation_lock, flags);
2267
EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2269
#ifdef CONFIG_XEN_PVHVM
2270
static void xen_hvm_exit_mmap(struct mm_struct *mm)
2272
struct xen_hvm_pagetable_dying a;
2275
a.domid = DOMID_SELF;
2276
a.gpa = __pa(mm->pgd);
2277
rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2278
WARN_ON_ONCE(rc < 0);
2281
static int is_pagetable_dying_supported(void)
2283
struct xen_hvm_pagetable_dying a;
2286
a.domid = DOMID_SELF;
2288
rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2290
printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2296
void __init xen_hvm_init_mmu_ops(void)
2298
if (is_pagetable_dying_supported())
2299
pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2303
#define REMAP_BATCH_SIZE 16
2308
struct mmu_update *mmu_update;
2311
static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2312
unsigned long addr, void *data)
2314
struct remap_data *rmd = data;
2315
pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2317
rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2318
rmd->mmu_update->val = pte_val_ma(pte);
2324
int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2326
unsigned long mfn, int nr,
2327
pgprot_t prot, unsigned domid)
2329
struct remap_data rmd;
2330
struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2332
unsigned long range;
2335
prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2337
BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2338
(VM_PFNMAP | VM_RESERVED | VM_IO)));
2344
batch = min(REMAP_BATCH_SIZE, nr);
2345
range = (unsigned long)batch << PAGE_SHIFT;
2347
rmd.mmu_update = mmu_update;
2348
err = apply_to_page_range(vma->vm_mm, addr, range,
2349
remap_area_mfn_pte_fn, &rmd);
2354
if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2368
EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);