Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
mmu.c
Go to the documentation of this file.
1 /*
2  * Xen mmu operations
3  *
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.
7  *
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.
12  *
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
16  * use.
17  *
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.
23  *
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
29  * pagetable.
30  *
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.
38  *
39  * Jeremy Fitzhardinge <[email protected]>, XenSource Inc, 2007
40  */
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>
50 #include <linux/crash_dump.h>
51 
52 #include <trace/events/xen.h>
53 
54 #include <asm/pgtable.h>
55 #include <asm/tlbflush.h>
56 #include <asm/fixmap.h>
57 #include <asm/mmu_context.h>
58 #include <asm/setup.h>
59 #include <asm/paravirt.h>
60 #include <asm/e820.h>
61 #include <asm/linkage.h>
62 #include <asm/page.h>
63 #include <asm/init.h>
64 #include <asm/pat.h>
65 #include <asm/smp.h>
66 
67 #include <asm/xen/hypercall.h>
68 #include <asm/xen/hypervisor.h>
69 
70 #include <xen/xen.h>
71 #include <xen/page.h>
72 #include <xen/interface/xen.h>
74 #include <xen/interface/version.h>
75 #include <xen/interface/memory.h>
76 #include <xen/hvc-console.h>
77 
78 #include "multicalls.h"
79 #include "mmu.h"
80 #include "debugfs.h"
81 
82 /*
83  * Protects atomic reservation decrease/increase against concurrent increases.
84  * Also protects non-atomic updates of current_pages and balloon lists.
85  */
86 DEFINE_SPINLOCK(xen_reservation_lock);
87 
88 #ifdef CONFIG_X86_32
89 /*
90  * Identity map, in addition to plain kernel map. This needs to be
91  * large enough to allocate page table pages to allocate the rest.
92  * Each page can map 2MB.
93  */
94 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
95 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
96 #endif
97 #ifdef CONFIG_X86_64
98 /* l3 pud for userspace vsyscall mapping */
99 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
100 #endif /* CONFIG_X86_64 */
101 
102 /*
103  * Note about cr3 (pagetable base) values:
104  *
105  * xen_cr3 contains the current logical cr3 value; it contains the
106  * last set cr3. This may not be the current effective cr3, because
107  * its update may be being lazily deferred. However, a vcpu looking
108  * at its own cr3 can use this value knowing that it everything will
109  * be self-consistent.
110  *
111  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
112  * hypercall to set the vcpu cr3 is complete (so it may be a little
113  * out of date, but it will never be set early). If one vcpu is
114  * looking at another vcpu's cr3 value, it should use this variable.
115  */
116 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
117 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
118 
119 
120 /*
121  * Just beyond the highest usermode address. STACK_TOP_MAX has a
122  * redzone above it, so round it up to a PGD boundary.
123  */
124 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
125 
126 unsigned long arbitrary_virt_to_mfn(void *vaddr)
127 {
128  xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
129 
130  return PFN_DOWN(maddr.maddr);
131 }
132 
134 {
135  unsigned long address = (unsigned long)vaddr;
136  unsigned int level;
137  pte_t *pte;
138  unsigned offset;
139 
140  /*
141  * if the PFN is in the linear mapped vaddr range, we can just use
142  * the (quick) virt_to_machine() p2m lookup
143  */
144  if (virt_addr_valid(vaddr))
145  return virt_to_machine(vaddr);
146 
147  /* otherwise we have to do a (slower) full page-table walk */
148 
149  pte = lookup_address(address, &level);
150  BUG_ON(pte == NULL);
151  offset = address & ~PAGE_MASK;
152  return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
153 }
154 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
155 
157 {
158  pte_t *pte, ptev;
159  unsigned long address = (unsigned long)vaddr;
160  unsigned int level;
161 
162  pte = lookup_address(address, &level);
163  if (pte == NULL)
164  return; /* vaddr missing */
165 
166  ptev = pte_wrprotect(*pte);
167 
168  if (HYPERVISOR_update_va_mapping(address, ptev, 0))
169  BUG();
170 }
171 
173 {
174  pte_t *pte, ptev;
175  unsigned long address = (unsigned long)vaddr;
176  unsigned int level;
177 
178  pte = lookup_address(address, &level);
179  if (pte == NULL)
180  return; /* vaddr missing */
181 
182  ptev = pte_mkwrite(*pte);
183 
184  if (HYPERVISOR_update_va_mapping(address, ptev, 0))
185  BUG();
186 }
187 
188 
189 static bool xen_page_pinned(void *ptr)
190 {
191  struct page *page = virt_to_page(ptr);
192 
193  return PagePinned(page);
194 }
195 
196 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
197 {
198  struct multicall_space mcs;
199  struct mmu_update *u;
200 
201  trace_xen_mmu_set_domain_pte(ptep, pteval, domid);
202 
203  mcs = xen_mc_entry(sizeof(*u));
204  u = mcs.args;
205 
206  /* ptep might be kmapped when using 32-bit HIGHPTE */
207  u->ptr = virt_to_machine(ptep).maddr;
208  u->val = pte_val_ma(pteval);
209 
210  MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
211 
212  xen_mc_issue(PARAVIRT_LAZY_MMU);
213 }
215 
216 static void xen_extend_mmu_update(const struct mmu_update *update)
217 {
218  struct multicall_space mcs;
219  struct mmu_update *u;
220 
222 
223  if (mcs.mc != NULL) {
224  mcs.mc->args[1]++;
225  } else {
226  mcs = __xen_mc_entry(sizeof(*u));
227  MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
228  }
229 
230  u = mcs.args;
231  *u = *update;
232 }
233 
234 static void xen_extend_mmuext_op(const struct mmuext_op *op)
235 {
236  struct multicall_space mcs;
237  struct mmuext_op *u;
238 
240 
241  if (mcs.mc != NULL) {
242  mcs.mc->args[1]++;
243  } else {
244  mcs = __xen_mc_entry(sizeof(*u));
245  MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
246  }
247 
248  u = mcs.args;
249  *u = *op;
250 }
251 
252 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
253 {
254  struct mmu_update u;
255 
256  preempt_disable();
257 
258  xen_mc_batch();
259 
260  /* ptr may be ioremapped for 64-bit pagetable setup */
261  u.ptr = arbitrary_virt_to_machine(ptr).maddr;
262  u.val = pmd_val_ma(val);
263  xen_extend_mmu_update(&u);
264 
265  xen_mc_issue(PARAVIRT_LAZY_MMU);
266 
267  preempt_enable();
268 }
269 
270 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
271 {
272  trace_xen_mmu_set_pmd(ptr, val);
273 
274  /* If page is not pinned, we can just update the entry
275  directly */
276  if (!xen_page_pinned(ptr)) {
277  *ptr = val;
278  return;
279  }
280 
281  xen_set_pmd_hyper(ptr, val);
282 }
283 
284 /*
285  * Associate a virtual page frame with a given physical page frame
286  * and protection flags for that frame.
287  */
288 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
289 {
290  set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
291 }
292 
293 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
294 {
295  struct mmu_update u;
296 
298  return false;
299 
300  xen_mc_batch();
301 
302  u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
303  u.val = pte_val_ma(pteval);
304  xen_extend_mmu_update(&u);
305 
306  xen_mc_issue(PARAVIRT_LAZY_MMU);
307 
308  return true;
309 }
310 
311 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
312 {
313  if (!xen_batched_set_pte(ptep, pteval)) {
314  /*
315  * Could call native_set_pte() here and trap and
316  * emulate the PTE write but with 32-bit guests this
317  * needs two traps (one for each of the two 32-bit
318  * words in the PTE) so do one hypercall directly
319  * instead.
320  */
321  struct mmu_update u;
322 
323  u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
324  u.val = pte_val_ma(pteval);
326  }
327 }
328 
329 static void xen_set_pte(pte_t *ptep, pte_t pteval)
330 {
331  trace_xen_mmu_set_pte(ptep, pteval);
332  __xen_set_pte(ptep, pteval);
333 }
334 
335 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
336  pte_t *ptep, pte_t pteval)
337 {
338  trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
339  __xen_set_pte(ptep, pteval);
340 }
341 
343  unsigned long addr, pte_t *ptep)
344 {
345  /* Just return the pte as-is. We preserve the bits on commit */
346  trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
347  return *ptep;
348 }
349 
350 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
351  pte_t *ptep, pte_t pte)
352 {
353  struct mmu_update u;
354 
355  trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
356  xen_mc_batch();
357 
359  u.val = pte_val_ma(pte);
360  xen_extend_mmu_update(&u);
361 
362  xen_mc_issue(PARAVIRT_LAZY_MMU);
363 }
364 
365 /* Assume pteval_t is equivalent to all the other *val_t types. */
366 static pteval_t pte_mfn_to_pfn(pteval_t val)
367 {
368  if (val & _PAGE_PRESENT) {
369  unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
370  unsigned long pfn = mfn_to_pfn(mfn);
371 
372  pteval_t flags = val & PTE_FLAGS_MASK;
373  if (unlikely(pfn == ~0))
374  val = flags & ~_PAGE_PRESENT;
375  else
376  val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
377  }
378 
379  return val;
380 }
381 
382 static pteval_t pte_pfn_to_mfn(pteval_t val)
383 {
384  if (val & _PAGE_PRESENT) {
385  unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
386  pteval_t flags = val & PTE_FLAGS_MASK;
387  unsigned long mfn;
388 
389  if (!xen_feature(XENFEAT_auto_translated_physmap))
390  mfn = get_phys_to_machine(pfn);
391  else
392  mfn = pfn;
393  /*
394  * If there's no mfn for the pfn, then just create an
395  * empty non-present pte. Unfortunately this loses
396  * information about the original pfn, so
397  * pte_mfn_to_pfn is asymmetric.
398  */
399  if (unlikely(mfn == INVALID_P2M_ENTRY)) {
400  mfn = 0;
401  flags = 0;
402  } else {
403  /*
404  * Paramount to do this test _after_ the
405  * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
406  * IDENTITY_FRAME_BIT resolves to true.
407  */
408  mfn &= ~FOREIGN_FRAME_BIT;
409  if (mfn & IDENTITY_FRAME_BIT) {
410  mfn &= ~IDENTITY_FRAME_BIT;
411  flags |= _PAGE_IOMAP;
412  }
413  }
414  val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
415  }
416 
417  return val;
418 }
419 
420 static pteval_t iomap_pte(pteval_t val)
421 {
422  if (val & _PAGE_PRESENT) {
423  unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
424  pteval_t flags = val & PTE_FLAGS_MASK;
425 
426  /* We assume the pte frame number is a MFN, so
427  just use it as-is. */
428  val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
429  }
430 
431  return val;
432 }
433 
434 static pteval_t xen_pte_val(pte_t pte)
435 {
436  pteval_t pteval = pte.pte;
437 #if 0
438  /* If this is a WC pte, convert back from Xen WC to Linux WC */
439  if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
440  WARN_ON(!pat_enabled);
441  pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
442  }
443 #endif
444  if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
445  return pteval;
446 
447  return pte_mfn_to_pfn(pteval);
448 }
449 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
450 
451 static pgdval_t xen_pgd_val(pgd_t pgd)
452 {
453  return pte_mfn_to_pfn(pgd.pgd);
454 }
455 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
456 
457 /*
458  * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
459  * are reserved for now, to correspond to the Intel-reserved PAT
460  * types.
461  *
462  * We expect Linux's PAT set as follows:
463  *
464  * Idx PTE flags Linux Xen Default
465  * 0 WB WB WB
466  * 1 PWT WC WT WT
467  * 2 PCD UC- UC- UC-
468  * 3 PCD PWT UC UC UC
469  * 4 PAT WB WC WB
470  * 5 PAT PWT WC WP WT
471  * 6 PAT PCD UC- UC UC-
472  * 7 PAT PCD PWT UC UC UC
473  */
474 
476 {
477  /* We expect Linux to use a PAT setting of
478  * UC UC- WC WB (ignoring the PAT flag) */
479  WARN_ON(pat != 0x0007010600070106ull);
480 }
481 
482 static pte_t xen_make_pte(pteval_t pte)
483 {
484  phys_addr_t addr = (pte & PTE_PFN_MASK);
485 #if 0
486  /* If Linux is trying to set a WC pte, then map to the Xen WC.
487  * If _PAGE_PAT is set, then it probably means it is really
488  * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
489  * things work out OK...
490  *
491  * (We should never see kernel mappings with _PAGE_PSE set,
492  * but we could see hugetlbfs mappings, I think.).
493  */
494  if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
495  if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
496  pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
497  }
498 #endif
499  /*
500  * Unprivileged domains are allowed to do IOMAPpings for
501  * PCI passthrough, but not map ISA space. The ISA
502  * mappings are just dummy local mappings to keep other
503  * parts of the kernel happy.
504  */
505  if (unlikely(pte & _PAGE_IOMAP) &&
506  (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
507  pte = iomap_pte(pte);
508  } else {
509  pte &= ~_PAGE_IOMAP;
510  pte = pte_pfn_to_mfn(pte);
511  }
512 
513  return native_make_pte(pte);
514 }
515 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
516 
517 static pgd_t xen_make_pgd(pgdval_t pgd)
518 {
519  pgd = pte_pfn_to_mfn(pgd);
520  return native_make_pgd(pgd);
521 }
522 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
523 
524 static pmdval_t xen_pmd_val(pmd_t pmd)
525 {
526  return pte_mfn_to_pfn(pmd.pmd);
527 }
528 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
529 
530 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
531 {
532  struct mmu_update u;
533 
534  preempt_disable();
535 
536  xen_mc_batch();
537 
538  /* ptr may be ioremapped for 64-bit pagetable setup */
539  u.ptr = arbitrary_virt_to_machine(ptr).maddr;
540  u.val = pud_val_ma(val);
541  xen_extend_mmu_update(&u);
542 
543  xen_mc_issue(PARAVIRT_LAZY_MMU);
544 
545  preempt_enable();
546 }
547 
548 static void xen_set_pud(pud_t *ptr, pud_t val)
549 {
550  trace_xen_mmu_set_pud(ptr, val);
551 
552  /* If page is not pinned, we can just update the entry
553  directly */
554  if (!xen_page_pinned(ptr)) {
555  *ptr = val;
556  return;
557  }
558 
559  xen_set_pud_hyper(ptr, val);
560 }
561 
562 #ifdef CONFIG_X86_PAE
563 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
564 {
565  trace_xen_mmu_set_pte_atomic(ptep, pte);
566  set_64bit((u64 *)ptep, native_pte_val(pte));
567 }
568 
569 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
570 {
571  trace_xen_mmu_pte_clear(mm, addr, ptep);
572  if (!xen_batched_set_pte(ptep, native_make_pte(0)))
573  native_pte_clear(mm, addr, ptep);
574 }
575 
576 static void xen_pmd_clear(pmd_t *pmdp)
577 {
578  trace_xen_mmu_pmd_clear(pmdp);
579  set_pmd(pmdp, __pmd(0));
580 }
581 #endif /* CONFIG_X86_PAE */
582 
583 static pmd_t xen_make_pmd(pmdval_t pmd)
584 {
585  pmd = pte_pfn_to_mfn(pmd);
586  return native_make_pmd(pmd);
587 }
588 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
589 
590 #if PAGETABLE_LEVELS == 4
591 static pudval_t xen_pud_val(pud_t pud)
592 {
593  return pte_mfn_to_pfn(pud.pud);
594 }
595 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
596 
597 static pud_t xen_make_pud(pudval_t pud)
598 {
599  pud = pte_pfn_to_mfn(pud);
600 
601  return native_make_pud(pud);
602 }
603 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
604 
605 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
606 {
607  pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
608  unsigned offset = pgd - pgd_page;
609  pgd_t *user_ptr = NULL;
610 
611  if (offset < pgd_index(USER_LIMIT)) {
612  struct page *page = virt_to_page(pgd_page);
613  user_ptr = (pgd_t *)page->private;
614  if (user_ptr)
615  user_ptr += offset;
616  }
617 
618  return user_ptr;
619 }
620 
621 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
622 {
623  struct mmu_update u;
624 
625  u.ptr = virt_to_machine(ptr).maddr;
626  u.val = pgd_val_ma(val);
627  xen_extend_mmu_update(&u);
628 }
629 
630 /*
631  * Raw hypercall-based set_pgd, intended for in early boot before
632  * there's a page structure. This implies:
633  * 1. The only existing pagetable is the kernel's
634  * 2. It is always pinned
635  * 3. It has no user pagetable attached to it
636  */
637 static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
638 {
639  preempt_disable();
640 
641  xen_mc_batch();
642 
643  __xen_set_pgd_hyper(ptr, val);
644 
645  xen_mc_issue(PARAVIRT_LAZY_MMU);
646 
647  preempt_enable();
648 }
649 
650 static void xen_set_pgd(pgd_t *ptr, pgd_t val)
651 {
652  pgd_t *user_ptr = xen_get_user_pgd(ptr);
653 
654  trace_xen_mmu_set_pgd(ptr, user_ptr, val);
655 
656  /* If page is not pinned, we can just update the entry
657  directly */
658  if (!xen_page_pinned(ptr)) {
659  *ptr = val;
660  if (user_ptr) {
661  WARN_ON(xen_page_pinned(user_ptr));
662  *user_ptr = val;
663  }
664  return;
665  }
666 
667  /* If it's pinned, then we can at least batch the kernel and
668  user updates together. */
669  xen_mc_batch();
670 
671  __xen_set_pgd_hyper(ptr, val);
672  if (user_ptr)
673  __xen_set_pgd_hyper(user_ptr, val);
674 
675  xen_mc_issue(PARAVIRT_LAZY_MMU);
676 }
677 #endif /* PAGETABLE_LEVELS == 4 */
678 
679 /*
680  * (Yet another) pagetable walker. This one is intended for pinning a
681  * pagetable. This means that it walks a pagetable and calls the
682  * callback function on each page it finds making up the page table,
683  * at every level. It walks the entire pagetable, but it only bothers
684  * pinning pte pages which are below limit. In the normal case this
685  * will be STACK_TOP_MAX, but at boot we need to pin up to
686  * FIXADDR_TOP.
687  *
688  * For 32-bit the important bit is that we don't pin beyond there,
689  * because then we start getting into Xen's ptes.
690  *
691  * For 64-bit, we must skip the Xen hole in the middle of the address
692  * space, just after the big x86-64 virtual hole.
693  */
694 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
695  int (*func)(struct mm_struct *mm, struct page *,
696  enum pt_level),
697  unsigned long limit)
698 {
699  int flush = 0;
700  unsigned hole_low, hole_high;
701  unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
702  unsigned pgdidx, pudidx, pmdidx;
703 
704  /* The limit is the last byte to be touched */
705  limit--;
706  BUG_ON(limit >= FIXADDR_TOP);
707 
708  if (xen_feature(XENFEAT_auto_translated_physmap))
709  return 0;
710 
711  /*
712  * 64-bit has a great big hole in the middle of the address
713  * space, which contains the Xen mappings. On 32-bit these
714  * will end up making a zero-sized hole and so is a no-op.
715  */
716  hole_low = pgd_index(USER_LIMIT);
717  hole_high = pgd_index(PAGE_OFFSET);
718 
719  pgdidx_limit = pgd_index(limit);
720 #if PTRS_PER_PUD > 1
721  pudidx_limit = pud_index(limit);
722 #else
723  pudidx_limit = 0;
724 #endif
725 #if PTRS_PER_PMD > 1
726  pmdidx_limit = pmd_index(limit);
727 #else
728  pmdidx_limit = 0;
729 #endif
730 
731  for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
732  pud_t *pud;
733 
734  if (pgdidx >= hole_low && pgdidx < hole_high)
735  continue;
736 
737  if (!pgd_val(pgd[pgdidx]))
738  continue;
739 
740  pud = pud_offset(&pgd[pgdidx], 0);
741 
742  if (PTRS_PER_PUD > 1) /* not folded */
743  flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
744 
745  for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
746  pmd_t *pmd;
747 
748  if (pgdidx == pgdidx_limit &&
749  pudidx > pudidx_limit)
750  goto out;
751 
752  if (pud_none(pud[pudidx]))
753  continue;
754 
755  pmd = pmd_offset(&pud[pudidx], 0);
756 
757  if (PTRS_PER_PMD > 1) /* not folded */
758  flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
759 
760  for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
761  struct page *pte;
762 
763  if (pgdidx == pgdidx_limit &&
764  pudidx == pudidx_limit &&
765  pmdidx > pmdidx_limit)
766  goto out;
767 
768  if (pmd_none(pmd[pmdidx]))
769  continue;
770 
771  pte = pmd_page(pmd[pmdidx]);
772  flush |= (*func)(mm, pte, PT_PTE);
773  }
774  }
775  }
776 
777 out:
778  /* Do the top level last, so that the callbacks can use it as
779  a cue to do final things like tlb flushes. */
780  flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
781 
782  return flush;
783 }
784 
785 static int xen_pgd_walk(struct mm_struct *mm,
786  int (*func)(struct mm_struct *mm, struct page *,
787  enum pt_level),
788  unsigned long limit)
789 {
790  return __xen_pgd_walk(mm, mm->pgd, func, limit);
791 }
792 
793 /* If we're using split pte locks, then take the page's lock and
794  return a pointer to it. Otherwise return NULL. */
795 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
796 {
797  spinlock_t *ptl = NULL;
798 
799 #if USE_SPLIT_PTLOCKS
800  ptl = __pte_lockptr(page);
802 #endif
803 
804  return ptl;
805 }
806 
807 static void xen_pte_unlock(void *v)
808 {
809  spinlock_t *ptl = v;
810  spin_unlock(ptl);
811 }
812 
813 static void xen_do_pin(unsigned level, unsigned long pfn)
814 {
815  struct mmuext_op op;
816 
817  op.cmd = level;
818  op.arg1.mfn = pfn_to_mfn(pfn);
819 
820  xen_extend_mmuext_op(&op);
821 }
822 
823 static int xen_pin_page(struct mm_struct *mm, struct page *page,
824  enum pt_level level)
825 {
826  unsigned pgfl = TestSetPagePinned(page);
827  int flush;
828 
829  if (pgfl)
830  flush = 0; /* already pinned */
831  else if (PageHighMem(page))
832  /* kmaps need flushing if we found an unpinned
833  highpage */
834  flush = 1;
835  else {
836  void *pt = lowmem_page_address(page);
837  unsigned long pfn = page_to_pfn(page);
838  struct multicall_space mcs = __xen_mc_entry(0);
839  spinlock_t *ptl;
840 
841  flush = 0;
842 
843  /*
844  * We need to hold the pagetable lock between the time
845  * we make the pagetable RO and when we actually pin
846  * it. If we don't, then other users may come in and
847  * attempt to update the pagetable by writing it,
848  * which will fail because the memory is RO but not
849  * pinned, so Xen won't do the trap'n'emulate.
850  *
851  * If we're using split pte locks, we can't hold the
852  * entire pagetable's worth of locks during the
853  * traverse, because we may wrap the preempt count (8
854  * bits). The solution is to mark RO and pin each PTE
855  * page while holding the lock. This means the number
856  * of locks we end up holding is never more than a
857  * batch size (~32 entries, at present).
858  *
859  * If we're not using split pte locks, we needn't pin
860  * the PTE pages independently, because we're
861  * protected by the overall pagetable lock.
862  */
863  ptl = NULL;
864  if (level == PT_PTE)
865  ptl = xen_pte_lock(page, mm);
866 
867  MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
868  pfn_pte(pfn, PAGE_KERNEL_RO),
869  level == PT_PGD ? UVMF_TLB_FLUSH : 0);
870 
871  if (ptl) {
872  xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
873 
874  /* Queue a deferred unlock for when this batch
875  is completed. */
876  xen_mc_callback(xen_pte_unlock, ptl);
877  }
878  }
879 
880  return flush;
881 }
882 
883 /* This is called just after a mm has been created, but it has not
884  been used yet. We need to make sure that its pagetable is all
885  read-only, and can be pinned. */
886 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
887 {
888  trace_xen_mmu_pgd_pin(mm, pgd);
889 
890  xen_mc_batch();
891 
892  if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
893  /* re-enable interrupts for flushing */
894  xen_mc_issue(0);
895 
897 
898  xen_mc_batch();
899  }
900 
901 #ifdef CONFIG_X86_64
902  {
903  pgd_t *user_pgd = xen_get_user_pgd(pgd);
904 
905  xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
906 
907  if (user_pgd) {
908  xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
909  xen_do_pin(MMUEXT_PIN_L4_TABLE,
910  PFN_DOWN(__pa(user_pgd)));
911  }
912  }
913 #else /* CONFIG_X86_32 */
914 #ifdef CONFIG_X86_PAE
915  /* Need to make sure unshared kernel PMD is pinnable */
916  xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
917  PT_PMD);
918 #endif
919  xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
920 #endif /* CONFIG_X86_64 */
921  xen_mc_issue(0);
922 }
923 
924 static void xen_pgd_pin(struct mm_struct *mm)
925 {
926  __xen_pgd_pin(mm, mm->pgd);
927 }
928 
929 /*
930  * On save, we need to pin all pagetables to make sure they get their
931  * mfns turned into pfns. Search the list for any unpinned pgds and pin
932  * them (unpinned pgds are not currently in use, probably because the
933  * process is under construction or destruction).
934  *
935  * Expected to be called in stop_machine() ("equivalent to taking
936  * every spinlock in the system"), so the locking doesn't really
937  * matter all that much.
938  */
939 void xen_mm_pin_all(void)
940 {
941  struct page *page;
942 
943  spin_lock(&pgd_lock);
944 
945  list_for_each_entry(page, &pgd_list, lru) {
946  if (!PagePinned(page)) {
947  __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
948  SetPageSavePinned(page);
949  }
950  }
951 
952  spin_unlock(&pgd_lock);
953 }
954 
955 /*
956  * The init_mm pagetable is really pinned as soon as its created, but
957  * that's before we have page structures to store the bits. So do all
958  * the book-keeping now.
959  */
960 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
961  enum pt_level level)
962 {
963  SetPagePinned(page);
964  return 0;
965 }
966 
967 static void __init xen_mark_init_mm_pinned(void)
968 {
969  xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
970 }
971 
972 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
973  enum pt_level level)
974 {
975  unsigned pgfl = TestClearPagePinned(page);
976 
977  if (pgfl && !PageHighMem(page)) {
978  void *pt = lowmem_page_address(page);
979  unsigned long pfn = page_to_pfn(page);
980  spinlock_t *ptl = NULL;
981  struct multicall_space mcs;
982 
983  /*
984  * Do the converse to pin_page. If we're using split
985  * pte locks, we must be holding the lock for while
986  * the pte page is unpinned but still RO to prevent
987  * concurrent updates from seeing it in this
988  * partially-pinned state.
989  */
990  if (level == PT_PTE) {
991  ptl = xen_pte_lock(page, mm);
992 
993  if (ptl)
994  xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
995  }
996 
997  mcs = __xen_mc_entry(0);
998 
999  MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
1000  pfn_pte(pfn, PAGE_KERNEL),
1001  level == PT_PGD ? UVMF_TLB_FLUSH : 0);
1002 
1003  if (ptl) {
1004  /* unlock when batch completed */
1005  xen_mc_callback(xen_pte_unlock, ptl);
1006  }
1007  }
1008 
1009  return 0; /* never need to flush on unpin */
1010 }
1011 
1012 /* Release a pagetables pages back as normal RW */
1013 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
1014 {
1015  trace_xen_mmu_pgd_unpin(mm, pgd);
1016 
1017  xen_mc_batch();
1018 
1019  xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1020 
1021 #ifdef CONFIG_X86_64
1022  {
1023  pgd_t *user_pgd = xen_get_user_pgd(pgd);
1024 
1025  if (user_pgd) {
1026  xen_do_pin(MMUEXT_UNPIN_TABLE,
1027  PFN_DOWN(__pa(user_pgd)));
1028  xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1029  }
1030  }
1031 #endif
1032 
1033 #ifdef CONFIG_X86_PAE
1034  /* Need to make sure unshared kernel PMD is unpinned */
1035  xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1036  PT_PMD);
1037 #endif
1038 
1039  __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1040 
1041  xen_mc_issue(0);
1042 }
1043 
1044 static void xen_pgd_unpin(struct mm_struct *mm)
1045 {
1046  __xen_pgd_unpin(mm, mm->pgd);
1047 }
1048 
1049 /*
1050  * On resume, undo any pinning done at save, so that the rest of the
1051  * kernel doesn't see any unexpected pinned pagetables.
1052  */
1054 {
1055  struct page *page;
1056 
1057  spin_lock(&pgd_lock);
1058 
1059  list_for_each_entry(page, &pgd_list, lru) {
1060  if (PageSavePinned(page)) {
1061  BUG_ON(!PagePinned(page));
1062  __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1063  ClearPageSavePinned(page);
1064  }
1065  }
1066 
1067  spin_unlock(&pgd_lock);
1068 }
1069 
1070 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1071 {
1072  spin_lock(&next->page_table_lock);
1073  xen_pgd_pin(next);
1074  spin_unlock(&next->page_table_lock);
1075 }
1076 
1077 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1078 {
1079  spin_lock(&mm->page_table_lock);
1080  xen_pgd_pin(mm);
1081  spin_unlock(&mm->page_table_lock);
1082 }
1083 
1084 
1085 #ifdef CONFIG_SMP
1086 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1087  we need to repoint it somewhere else before we can unpin it. */
1088 static void drop_other_mm_ref(void *info)
1089 {
1090  struct mm_struct *mm = info;
1091  struct mm_struct *active_mm;
1092 
1093  active_mm = this_cpu_read(cpu_tlbstate.active_mm);
1094 
1095  if (active_mm == mm && this_cpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
1097 
1098  /* If this cpu still has a stale cr3 reference, then make sure
1099  it has been flushed. */
1100  if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
1101  load_cr3(swapper_pg_dir);
1102 }
1103 
1104 static void xen_drop_mm_ref(struct mm_struct *mm)
1105 {
1107  unsigned cpu;
1108 
1109  if (current->active_mm == mm) {
1110  if (current->mm == mm)
1111  load_cr3(swapper_pg_dir);
1112  else
1114  }
1115 
1116  /* Get the "official" set of cpus referring to our pagetable. */
1117  if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1118  for_each_online_cpu(cpu) {
1119  if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1120  && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1121  continue;
1122  smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1123  }
1124  return;
1125  }
1126  cpumask_copy(mask, mm_cpumask(mm));
1127 
1128  /* It's possible that a vcpu may have a stale reference to our
1129  cr3, because its in lazy mode, and it hasn't yet flushed
1130  its set of pending hypercalls yet. In this case, we can
1131  look at its actual current cr3 value, and force it to flush
1132  if needed. */
1133  for_each_online_cpu(cpu) {
1134  if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1135  cpumask_set_cpu(cpu, mask);
1136  }
1137 
1138  if (!cpumask_empty(mask))
1139  smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1140  free_cpumask_var(mask);
1141 }
1142 #else
1143 static void xen_drop_mm_ref(struct mm_struct *mm)
1144 {
1145  if (current->active_mm == mm)
1146  load_cr3(swapper_pg_dir);
1147 }
1148 #endif
1149 
1150 /*
1151  * While a process runs, Xen pins its pagetables, which means that the
1152  * hypervisor forces it to be read-only, and it controls all updates
1153  * to it. This means that all pagetable updates have to go via the
1154  * hypervisor, which is moderately expensive.
1155  *
1156  * Since we're pulling the pagetable down, we switch to use init_mm,
1157  * unpin old process pagetable and mark it all read-write, which
1158  * allows further operations on it to be simple memory accesses.
1159  *
1160  * The only subtle point is that another CPU may be still using the
1161  * pagetable because of lazy tlb flushing. This means we need need to
1162  * switch all CPUs off this pagetable before we can unpin it.
1163  */
1164 static void xen_exit_mmap(struct mm_struct *mm)
1165 {
1166  get_cpu(); /* make sure we don't move around */
1167  xen_drop_mm_ref(mm);
1168  put_cpu();
1169 
1170  spin_lock(&mm->page_table_lock);
1171 
1172  /* pgd may not be pinned in the error exit path of execve */
1173  if (xen_page_pinned(mm->pgd))
1174  xen_pgd_unpin(mm);
1175 
1176  spin_unlock(&mm->page_table_lock);
1177 }
1178 
1179 static void xen_post_allocator_init(void);
1180 
1181 static __init void xen_mapping_pagetable_reserve(u64 start, u64 end)
1182 {
1183  /* reserve the range used */
1184  native_pagetable_reserve(start, end);
1185 
1186  /* set as RW the rest */
1187  printk(KERN_DEBUG "xen: setting RW the range %llx - %llx\n", end,
1189  while (end < PFN_PHYS(pgt_buf_top)) {
1191  end += PAGE_SIZE;
1192  }
1193 }
1194 
1195 #ifdef CONFIG_X86_64
1196 static void __init xen_cleanhighmap(unsigned long vaddr,
1197  unsigned long vaddr_end)
1198 {
1199  unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1200  pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1201 
1202  /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1203  * We include the PMD passed in on _both_ boundaries. */
1204  for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PAGE_SIZE));
1205  pmd++, vaddr += PMD_SIZE) {
1206  if (pmd_none(*pmd))
1207  continue;
1208  if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1209  set_pmd(pmd, __pmd(0));
1210  }
1211  /* In case we did something silly, we should crash in this function
1212  * instead of somewhere later and be confusing. */
1213  xen_mc_flush();
1214 }
1215 #endif
1216 static void __init xen_pagetable_init(void)
1217 {
1218 #ifdef CONFIG_X86_64
1219  unsigned long size;
1220  unsigned long addr;
1221 #endif
1222  paging_init();
1224 #ifdef CONFIG_X86_64
1225  if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1226  unsigned long new_mfn_list;
1227 
1228  size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1229 
1230  /* On 32-bit, we get zero so this never gets executed. */
1231  new_mfn_list = xen_revector_p2m_tree();
1232  if (new_mfn_list && new_mfn_list != xen_start_info->mfn_list) {
1233  /* using __ka address and sticking INVALID_P2M_ENTRY! */
1234  memset((void *)xen_start_info->mfn_list, 0xff, size);
1235 
1236  /* We should be in __ka space. */
1238  addr = xen_start_info->mfn_list;
1239  /* We roundup to the PMD, which means that if anybody at this stage is
1240  * using the __ka address of xen_start_info or xen_start_info->shared_info
1241  * they are in going to crash. Fortunatly we have already revectored
1242  * in xen_setup_kernel_pagetable and in xen_setup_shared_info. */
1243  size = roundup(size, PMD_SIZE);
1244  xen_cleanhighmap(addr, addr + size);
1245 
1246  size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1247  memblock_free(__pa(xen_start_info->mfn_list), size);
1248  /* And revector! Bye bye old array */
1249  xen_start_info->mfn_list = new_mfn_list;
1250  } else
1251  goto skip;
1252  }
1253  /* At this stage, cleanup_highmap has already cleaned __ka space
1254  * from _brk_limit way up to the max_pfn_mapped (which is the end of
1255  * the ramdisk). We continue on, erasing PMD entries that point to page
1256  * tables - do note that they are accessible at this stage via __va.
1257  * For good measure we also round up to the PMD - which means that if
1258  * anybody is using __ka address to the initial boot-stack - and try
1259  * to use it - they are going to crash. The xen_start_info has been
1260  * taken care of already in xen_setup_kernel_pagetable. */
1261  addr = xen_start_info->pt_base;
1262  size = roundup(xen_start_info->nr_pt_frames * PAGE_SIZE, PMD_SIZE);
1263 
1264  xen_cleanhighmap(addr, addr + size);
1265  xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1266 #ifdef DEBUG
1267  /* This is superflous and is not neccessary, but you know what
1268  * lets do it. The MODULES_VADDR -> MODULES_END should be clear of
1269  * anything at this stage. */
1270  xen_cleanhighmap(MODULES_VADDR, roundup(MODULES_VADDR, PUD_SIZE) - 1);
1271 #endif
1272 skip:
1273 #endif
1274  xen_post_allocator_init();
1275 }
1276 static void xen_write_cr2(unsigned long cr2)
1277 {
1278  this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1279 }
1280 
1281 static unsigned long xen_read_cr2(void)
1282 {
1283  return this_cpu_read(xen_vcpu)->arch.cr2;
1284 }
1285 
1286 unsigned long xen_read_cr2_direct(void)
1287 {
1288  return this_cpu_read(xen_vcpu_info.arch.cr2);
1289 }
1290 
1292 {
1293  struct mmuext_op *op;
1294  struct multicall_space mcs;
1295 
1296  trace_xen_mmu_flush_tlb_all(0);
1297 
1298  preempt_disable();
1299 
1300  mcs = xen_mc_entry(sizeof(*op));
1301 
1302  op = mcs.args;
1303  op->cmd = MMUEXT_TLB_FLUSH_ALL;
1304  MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1305 
1306  xen_mc_issue(PARAVIRT_LAZY_MMU);
1307 
1308  preempt_enable();
1309 }
1310 static void xen_flush_tlb(void)
1311 {
1312  struct mmuext_op *op;
1313  struct multicall_space mcs;
1314 
1315  trace_xen_mmu_flush_tlb(0);
1316 
1317  preempt_disable();
1318 
1319  mcs = xen_mc_entry(sizeof(*op));
1320 
1321  op = mcs.args;
1323  MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1324 
1325  xen_mc_issue(PARAVIRT_LAZY_MMU);
1326 
1327  preempt_enable();
1328 }
1329 
1330 static void xen_flush_tlb_single(unsigned long addr)
1331 {
1332  struct mmuext_op *op;
1333  struct multicall_space mcs;
1334 
1335  trace_xen_mmu_flush_tlb_single(addr);
1336 
1337  preempt_disable();
1338 
1339  mcs = xen_mc_entry(sizeof(*op));
1340  op = mcs.args;
1341  op->cmd = MMUEXT_INVLPG_LOCAL;
1342  op->arg1.linear_addr = addr & PAGE_MASK;
1343  MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1344 
1345  xen_mc_issue(PARAVIRT_LAZY_MMU);
1346 
1347  preempt_enable();
1348 }
1349 
1350 static void xen_flush_tlb_others(const struct cpumask *cpus,
1351  struct mm_struct *mm, unsigned long start,
1352  unsigned long end)
1353 {
1354  struct {
1355  struct mmuext_op op;
1356 #ifdef CONFIG_SMP
1358 #else
1359  DECLARE_BITMAP(mask, NR_CPUS);
1360 #endif
1361  } *args;
1362  struct multicall_space mcs;
1363 
1364  trace_xen_mmu_flush_tlb_others(cpus, mm, start, end);
1365 
1366  if (cpumask_empty(cpus))
1367  return; /* nothing to do */
1368 
1369  mcs = xen_mc_entry(sizeof(*args));
1370  args = mcs.args;
1371  args->op.arg2.vcpumask = to_cpumask(args->mask);
1372 
1373  /* Remove us, and any offline CPUS. */
1374  cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1375  cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1376 
1377  args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1378  if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) {
1379  args->op.cmd = MMUEXT_INVLPG_MULTI;
1380  args->op.arg1.linear_addr = start;
1381  }
1382 
1383  MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1384 
1385  xen_mc_issue(PARAVIRT_LAZY_MMU);
1386 }
1387 
1388 static unsigned long xen_read_cr3(void)
1389 {
1390  return this_cpu_read(xen_cr3);
1391 }
1392 
1393 static void set_current_cr3(void *v)
1394 {
1395  this_cpu_write(xen_current_cr3, (unsigned long)v);
1396 }
1397 
1398 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1399 {
1400  struct mmuext_op op;
1401  unsigned long mfn;
1402 
1403  trace_xen_mmu_write_cr3(kernel, cr3);
1404 
1405  if (cr3)
1406  mfn = pfn_to_mfn(PFN_DOWN(cr3));
1407  else
1408  mfn = 0;
1409 
1410  WARN_ON(mfn == 0 && kernel);
1411 
1413  op.arg1.mfn = mfn;
1414 
1415  xen_extend_mmuext_op(&op);
1416 
1417  if (kernel) {
1418  this_cpu_write(xen_cr3, cr3);
1419 
1420  /* Update xen_current_cr3 once the batch has actually
1421  been submitted. */
1422  xen_mc_callback(set_current_cr3, (void *)cr3);
1423  }
1424 }
1425 
1426 static void xen_write_cr3(unsigned long cr3)
1427 {
1428  BUG_ON(preemptible());
1429 
1430  xen_mc_batch(); /* disables interrupts */
1431 
1432  /* Update while interrupts are disabled, so its atomic with
1433  respect to ipis */
1434  this_cpu_write(xen_cr3, cr3);
1435 
1436  __xen_write_cr3(true, cr3);
1437 
1438 #ifdef CONFIG_X86_64
1439  {
1440  pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1441  if (user_pgd)
1442  __xen_write_cr3(false, __pa(user_pgd));
1443  else
1444  __xen_write_cr3(false, 0);
1445  }
1446 #endif
1447 
1448  xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1449 }
1450 
1451 static int xen_pgd_alloc(struct mm_struct *mm)
1452 {
1453  pgd_t *pgd = mm->pgd;
1454  int ret = 0;
1455 
1456  BUG_ON(PagePinned(virt_to_page(pgd)));
1457 
1458 #ifdef CONFIG_X86_64
1459  {
1460  struct page *page = virt_to_page(pgd);
1461  pgd_t *user_pgd;
1462 
1463  BUG_ON(page->private != 0);
1464 
1465  ret = -ENOMEM;
1466 
1467  user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1468  page->private = (unsigned long)user_pgd;
1469 
1470  if (user_pgd != NULL) {
1471  user_pgd[pgd_index(VSYSCALL_START)] =
1472  __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1473  ret = 0;
1474  }
1475 
1476  BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1477  }
1478 #endif
1479 
1480  return ret;
1481 }
1482 
1483 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1484 {
1485 #ifdef CONFIG_X86_64
1486  pgd_t *user_pgd = xen_get_user_pgd(pgd);
1487 
1488  if (user_pgd)
1489  free_page((unsigned long)user_pgd);
1490 #endif
1491 }
1492 
1493 #ifdef CONFIG_X86_32
1494 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1495 {
1496  /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1497  if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1498  pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1499  pte_val_ma(pte));
1500 
1501  return pte;
1502 }
1503 #else /* CONFIG_X86_64 */
1504 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1505 {
1506  unsigned long pfn = pte_pfn(pte);
1507 
1508  /*
1509  * If the new pfn is within the range of the newly allocated
1510  * kernel pagetable, and it isn't being mapped into an
1511  * early_ioremap fixmap slot as a freshly allocated page, make sure
1512  * it is RO.
1513  */
1514  if (((!is_early_ioremap_ptep(ptep) &&
1515  pfn >= pgt_buf_start && pfn < pgt_buf_top)) ||
1516  (is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1517  pte = pte_wrprotect(pte);
1518 
1519  return pte;
1520 }
1521 #endif /* CONFIG_X86_64 */
1522 
1523 /*
1524  * Init-time set_pte while constructing initial pagetables, which
1525  * doesn't allow RO page table pages to be remapped RW.
1526  *
1527  * If there is no MFN for this PFN then this page is initially
1528  * ballooned out so clear the PTE (as in decrease_reservation() in
1529  * drivers/xen/balloon.c).
1530  *
1531  * Many of these PTE updates are done on unpinned and writable pages
1532  * and doing a hypercall for these is unnecessary and expensive. At
1533  * this point it is not possible to tell if a page is pinned or not,
1534  * so always write the PTE directly and rely on Xen trapping and
1535  * emulating any updates as necessary.
1536  */
1537 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1538 {
1539  if (pte_mfn(pte) != INVALID_P2M_ENTRY)
1540  pte = mask_rw_pte(ptep, pte);
1541  else
1542  pte = __pte_ma(0);
1543 
1544  native_set_pte(ptep, pte);
1545 }
1546 
1547 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1548 {
1549  struct mmuext_op op;
1550  op.cmd = cmd;
1551  op.arg1.mfn = pfn_to_mfn(pfn);
1552  if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1553  BUG();
1554 }
1555 
1556 /* Early in boot, while setting up the initial pagetable, assume
1557  everything is pinned. */
1558 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1559 {
1560 #ifdef CONFIG_FLATMEM
1561  BUG_ON(mem_map); /* should only be used early */
1562 #endif
1564  pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1565 }
1566 
1567 /* Used for pmd and pud */
1568 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1569 {
1570 #ifdef CONFIG_FLATMEM
1571  BUG_ON(mem_map); /* should only be used early */
1572 #endif
1574 }
1575 
1576 /* Early release_pte assumes that all pts are pinned, since there's
1577  only init_mm and anything attached to that is pinned. */
1578 static void __init xen_release_pte_init(unsigned long pfn)
1579 {
1580  pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1582 }
1583 
1584 static void __init xen_release_pmd_init(unsigned long pfn)
1585 {
1587 }
1588 
1589 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1590 {
1591  struct multicall_space mcs;
1592  struct mmuext_op *op;
1593 
1594  mcs = __xen_mc_entry(sizeof(*op));
1595  op = mcs.args;
1596  op->cmd = cmd;
1597  op->arg1.mfn = pfn_to_mfn(pfn);
1598 
1599  MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1600 }
1601 
1602 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1603 {
1604  struct multicall_space mcs;
1605  unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1606 
1607  mcs = __xen_mc_entry(0);
1608  MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1609  pfn_pte(pfn, prot), 0);
1610 }
1611 
1612 /* This needs to make sure the new pte page is pinned iff its being
1613  attached to a pinned pagetable. */
1614 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1615  unsigned level)
1616 {
1617  bool pinned = PagePinned(virt_to_page(mm->pgd));
1618 
1619  trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1620 
1621  if (pinned) {
1622  struct page *page = pfn_to_page(pfn);
1623 
1624  SetPagePinned(page);
1625 
1626  if (!PageHighMem(page)) {
1627  xen_mc_batch();
1628 
1629  __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1630 
1631  if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1632  __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1633 
1634  xen_mc_issue(PARAVIRT_LAZY_MMU);
1635  } else {
1636  /* make sure there are no stray mappings of
1637  this page */
1639  }
1640  }
1641 }
1642 
1643 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1644 {
1645  xen_alloc_ptpage(mm, pfn, PT_PTE);
1646 }
1647 
1648 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1649 {
1650  xen_alloc_ptpage(mm, pfn, PT_PMD);
1651 }
1652 
1653 /* This should never happen until we're OK to use struct page */
1654 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1655 {
1656  struct page *page = pfn_to_page(pfn);
1657  bool pinned = PagePinned(page);
1658 
1659  trace_xen_mmu_release_ptpage(pfn, level, pinned);
1660 
1661  if (pinned) {
1662  if (!PageHighMem(page)) {
1663  xen_mc_batch();
1664 
1665  if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1666  __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1667 
1668  __set_pfn_prot(pfn, PAGE_KERNEL);
1669 
1670  xen_mc_issue(PARAVIRT_LAZY_MMU);
1671  }
1672  ClearPagePinned(page);
1673  }
1674 }
1675 
1676 static void xen_release_pte(unsigned long pfn)
1677 {
1678  xen_release_ptpage(pfn, PT_PTE);
1679 }
1680 
1681 static void xen_release_pmd(unsigned long pfn)
1682 {
1683  xen_release_ptpage(pfn, PT_PMD);
1684 }
1685 
1686 #if PAGETABLE_LEVELS == 4
1687 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1688 {
1689  xen_alloc_ptpage(mm, pfn, PT_PUD);
1690 }
1691 
1692 static void xen_release_pud(unsigned long pfn)
1693 {
1694  xen_release_ptpage(pfn, PT_PUD);
1695 }
1696 #endif
1697 
1699 {
1700 #ifdef CONFIG_X86_32
1701  unsigned long top = HYPERVISOR_VIRT_START;
1702  struct xen_platform_parameters pp;
1703 
1705  top = pp.virt_start;
1706 
1707  reserve_top_address(-top);
1708 #endif /* CONFIG_X86_32 */
1709 }
1710 
1711 /*
1712  * Like __va(), but returns address in the kernel mapping (which is
1713  * all we have until the physical memory mapping has been set up.
1714  */
1715 static void *__ka(phys_addr_t paddr)
1716 {
1717 #ifdef CONFIG_X86_64
1718  return (void *)(paddr + __START_KERNEL_map);
1719 #else
1720  return __va(paddr);
1721 #endif
1722 }
1723 
1724 /* Convert a machine address to physical address */
1725 static unsigned long m2p(phys_addr_t maddr)
1726 {
1728 
1729  maddr &= PTE_PFN_MASK;
1730  paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1731 
1732  return paddr;
1733 }
1734 
1735 /* Convert a machine address to kernel virtual */
1736 static void *m2v(phys_addr_t maddr)
1737 {
1738  return __ka(m2p(maddr));
1739 }
1740 
1741 /* Set the page permissions on an identity-mapped pages */
1742 static void set_page_prot(void *addr, pgprot_t prot)
1743 {
1744  unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1745  pte_t pte = pfn_pte(pfn, prot);
1746 
1747  if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1748  BUG();
1749 }
1750 #ifdef CONFIG_X86_32
1751 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1752 {
1753  unsigned pmdidx, pteidx;
1754  unsigned ident_pte;
1755  unsigned long pfn;
1756 
1757  level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1758  PAGE_SIZE);
1759 
1760  ident_pte = 0;
1761  pfn = 0;
1762  for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1763  pte_t *pte_page;
1764 
1765  /* Reuse or allocate a page of ptes */
1766  if (pmd_present(pmd[pmdidx]))
1767  pte_page = m2v(pmd[pmdidx].pmd);
1768  else {
1769  /* Check for free pte pages */
1770  if (ident_pte == LEVEL1_IDENT_ENTRIES)
1771  break;
1772 
1773  pte_page = &level1_ident_pgt[ident_pte];
1774  ident_pte += PTRS_PER_PTE;
1775 
1776  pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1777  }
1778 
1779  /* Install mappings */
1780  for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1781  pte_t pte;
1782 
1783 #ifdef CONFIG_X86_32
1784  if (pfn > max_pfn_mapped)
1785  max_pfn_mapped = pfn;
1786 #endif
1787 
1788  if (!pte_none(pte_page[pteidx]))
1789  continue;
1790 
1791  pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1792  pte_page[pteidx] = pte;
1793  }
1794  }
1795 
1796  for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1797  set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1798 
1799  set_page_prot(pmd, PAGE_KERNEL_RO);
1800 }
1801 #endif
1803 {
1804  struct xen_machphys_mapping mapping;
1805 
1806  if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1807  machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1808  machine_to_phys_nr = mapping.max_mfn + 1;
1809  } else {
1811  }
1812 #ifdef CONFIG_X86_32
1815 #endif
1816 }
1817 
1818 #ifdef CONFIG_X86_64
1819 static void convert_pfn_mfn(void *v)
1820 {
1821  pte_t *pte = v;
1822  int i;
1823 
1824  /* All levels are converted the same way, so just treat them
1825  as ptes. */
1826  for (i = 0; i < PTRS_PER_PTE; i++)
1827  pte[i] = xen_make_pte(pte[i].pte);
1828 }
1829 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1830  unsigned long addr)
1831 {
1832  if (*pt_base == PFN_DOWN(__pa(addr))) {
1833  set_page_prot((void *)addr, PAGE_KERNEL);
1834  clear_page((void *)addr);
1835  (*pt_base)++;
1836  }
1837  if (*pt_end == PFN_DOWN(__pa(addr))) {
1838  set_page_prot((void *)addr, PAGE_KERNEL);
1839  clear_page((void *)addr);
1840  (*pt_end)--;
1841  }
1842 }
1843 /*
1844  * Set up the initial kernel pagetable.
1845  *
1846  * We can construct this by grafting the Xen provided pagetable into
1847  * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1848  * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This
1849  * means that only the kernel has a physical mapping to start with -
1850  * but that's enough to get __va working. We need to fill in the rest
1851  * of the physical mapping once some sort of allocator has been set
1852  * up.
1853  */
1854 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1855 {
1856  pud_t *l3;
1857  pmd_t *l2;
1858  unsigned long addr[3];
1859  unsigned long pt_base, pt_end;
1860  unsigned i;
1861 
1862  /* max_pfn_mapped is the last pfn mapped in the initial memory
1863  * mappings. Considering that on Xen after the kernel mappings we
1864  * have the mappings of some pages that don't exist in pfn space, we
1865  * set max_pfn_mapped to the last real pfn mapped. */
1867 
1868  pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1869  pt_end = pt_base + xen_start_info->nr_pt_frames;
1870 
1871  /* Zap identity mapping */
1872  init_level4_pgt[0] = __pgd(0);
1873 
1874  /* Pre-constructed entries are in pfn, so convert to mfn */
1875  /* L4[272] -> level3_ident_pgt
1876  * L4[511] -> level3_kernel_pgt */
1877  convert_pfn_mfn(init_level4_pgt);
1878 
1879  /* L3_i[0] -> level2_ident_pgt */
1880  convert_pfn_mfn(level3_ident_pgt);
1881  /* L3_k[510] -> level2_kernel_pgt
1882  * L3_i[511] -> level2_fixmap_pgt */
1883  convert_pfn_mfn(level3_kernel_pgt);
1884 
1885  /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1886  l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1887  l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1888 
1889  addr[0] = (unsigned long)pgd;
1890  addr[1] = (unsigned long)l3;
1891  addr[2] = (unsigned long)l2;
1892  /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1893  * Both L4[272][0] and L4[511][511] have entries that point to the same
1894  * L2 (PMD) tables. Meaning that if you modify it in __va space
1895  * it will be also modified in the __ka space! (But if you just
1896  * modify the PMD table to point to other PTE's or none, then you
1897  * are OK - which is what cleanup_highmap does) */
1898  copy_page(level2_ident_pgt, l2);
1899  /* Graft it onto L4[511][511] */
1900  copy_page(level2_kernel_pgt, l2);
1901 
1902  /* Get [511][510] and graft that in level2_fixmap_pgt */
1903  l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1904  l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1905  copy_page(level2_fixmap_pgt, l2);
1906  /* Note that we don't do anything with level1_fixmap_pgt which
1907  * we don't need. */
1908 
1909  /* Make pagetable pieces RO */
1910  set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1911  set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1912  set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1913  set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1914  set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1915  set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1916  set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1917 
1918  /* Pin down new L4 */
1919  pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1920  PFN_DOWN(__pa_symbol(init_level4_pgt)));
1921 
1922  /* Unpin Xen-provided one */
1923  pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1924 
1925  /*
1926  * At this stage there can be no user pgd, and no page
1927  * structure to attach it to, so make sure we just set kernel
1928  * pgd.
1929  */
1930  xen_mc_batch();
1931  __xen_write_cr3(true, __pa(init_level4_pgt));
1932  xen_mc_issue(PARAVIRT_LAZY_CPU);
1933 
1934  /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1935  * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
1936  * the initial domain. For guests using the toolstack, they are in:
1937  * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1938  * rip out the [L4] (pgd), but for guests we shave off three pages.
1939  */
1940  for (i = 0; i < ARRAY_SIZE(addr); i++)
1941  check_pt_base(&pt_base, &pt_end, addr[i]);
1942 
1943  /* Our (by three pages) smaller Xen pagetable that we are using */
1944  memblock_reserve(PFN_PHYS(pt_base), (pt_end - pt_base) * PAGE_SIZE);
1945  /* Revector the xen_start_info */
1947 }
1948 #else /* !CONFIG_X86_64 */
1949 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1950 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1951 
1952 static void __init xen_write_cr3_init(unsigned long cr3)
1953 {
1954  unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1955 
1956  BUG_ON(read_cr3() != __pa(initial_page_table));
1957  BUG_ON(cr3 != __pa(swapper_pg_dir));
1958 
1959  /*
1960  * We are switching to swapper_pg_dir for the first time (from
1961  * initial_page_table) and therefore need to mark that page
1962  * read-only and then pin it.
1963  *
1964  * Xen disallows sharing of kernel PMDs for PAE
1965  * guests. Therefore we must copy the kernel PMD from
1966  * initial_page_table into a new kernel PMD to be used in
1967  * swapper_pg_dir.
1968  */
1969  swapper_kernel_pmd =
1970  extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1971  copy_page(swapper_kernel_pmd, initial_kernel_pmd);
1973  __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1974  set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1975 
1976  set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1977  xen_write_cr3(cr3);
1978  pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1979 
1980  pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1981  PFN_DOWN(__pa(initial_page_table)));
1982  set_page_prot(initial_page_table, PAGE_KERNEL);
1983  set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1984 
1985  pv_mmu_ops.write_cr3 = &xen_write_cr3;
1986 }
1987 
1988 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1989 {
1990  pmd_t *kernel_pmd;
1991 
1992  initial_kernel_pmd =
1993  extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1994 
1996  xen_start_info->nr_pt_frames * PAGE_SIZE +
1997  512*1024);
1998 
1999  kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2000  copy_page(initial_kernel_pmd, kernel_pmd);
2001 
2002  xen_map_identity_early(initial_kernel_pmd, max_pfn);
2003 
2006  __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2007 
2008  set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2009  set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2010  set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2011 
2012  pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2013 
2014  pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2016  xen_write_cr3(__pa(initial_page_table));
2017 
2019  xen_start_info->nr_pt_frames * PAGE_SIZE);
2020 }
2021 #endif /* CONFIG_X86_64 */
2022 
2023 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2024 
2025 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2026 {
2027  pte_t pte;
2028 
2029  phys >>= PAGE_SHIFT;
2030 
2031  switch (idx) {
2032  case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2033 #ifdef CONFIG_X86_F00F_BUG
2034  case FIX_F00F_IDT:
2035 #endif
2036 #ifdef CONFIG_X86_32
2037  case FIX_WP_TEST:
2038  case FIX_VDSO:
2039 # ifdef CONFIG_HIGHMEM
2040  case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2041 # endif
2042 #else
2044  case VVAR_PAGE:
2045 #endif
2046  case FIX_TEXT_POKE0:
2047  case FIX_TEXT_POKE1:
2048  /* All local page mappings */
2049  pte = pfn_pte(phys, prot);
2050  break;
2051 
2052 #ifdef CONFIG_X86_LOCAL_APIC
2053  case FIX_APIC_BASE: /* maps dummy local APIC */
2054  pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2055  break;
2056 #endif
2057 
2058 #ifdef CONFIG_X86_IO_APIC
2059  case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2060  /*
2061  * We just don't map the IO APIC - all access is via
2062  * hypercalls. Keep the address in the pte for reference.
2063  */
2064  pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2065  break;
2066 #endif
2067 
2068  case FIX_PARAVIRT_BOOTMAP:
2069  /* This is an MFN, but it isn't an IO mapping from the
2070  IO domain */
2071  pte = mfn_pte(phys, prot);
2072  break;
2073 
2074  default:
2075  /* By default, set_fixmap is used for hardware mappings */
2076  pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
2077  break;
2078  }
2079 
2080  __native_set_fixmap(idx, pte);
2081 
2082 #ifdef CONFIG_X86_64
2083  /* Replicate changes to map the vsyscall page into the user
2084  pagetable vsyscall mapping. */
2085  if ((idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) ||
2086  idx == VVAR_PAGE) {
2087  unsigned long vaddr = __fix_to_virt(idx);
2088  set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2089  }
2090 #endif
2091 }
2092 
2093 static void __init xen_post_allocator_init(void)
2094 {
2095  pv_mmu_ops.set_pte = xen_set_pte;
2096  pv_mmu_ops.set_pmd = xen_set_pmd;
2097  pv_mmu_ops.set_pud = xen_set_pud;
2098 #if PAGETABLE_LEVELS == 4
2099  pv_mmu_ops.set_pgd = xen_set_pgd;
2100 #endif
2101 
2102  /* This will work as long as patching hasn't happened yet
2103  (which it hasn't) */
2104  pv_mmu_ops.alloc_pte = xen_alloc_pte;
2105  pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2106  pv_mmu_ops.release_pte = xen_release_pte;
2107  pv_mmu_ops.release_pmd = xen_release_pmd;
2108 #if PAGETABLE_LEVELS == 4
2109  pv_mmu_ops.alloc_pud = xen_alloc_pud;
2110  pv_mmu_ops.release_pud = xen_release_pud;
2111 #endif
2112 
2113 #ifdef CONFIG_X86_64
2114  SetPagePinned(virt_to_page(level3_user_vsyscall));
2115 #endif
2116  xen_mark_init_mm_pinned();
2117 }
2118 
2119 static void xen_leave_lazy_mmu(void)
2120 {
2121  preempt_disable();
2122  xen_mc_flush();
2124  preempt_enable();
2125 }
2126 
2127 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2128  .read_cr2 = xen_read_cr2,
2129  .write_cr2 = xen_write_cr2,
2130 
2131  .read_cr3 = xen_read_cr3,
2132 #ifdef CONFIG_X86_32
2133  .write_cr3 = xen_write_cr3_init,
2134 #else
2135  .write_cr3 = xen_write_cr3,
2136 #endif
2137 
2138  .flush_tlb_user = xen_flush_tlb,
2139  .flush_tlb_kernel = xen_flush_tlb,
2140  .flush_tlb_single = xen_flush_tlb_single,
2141  .flush_tlb_others = xen_flush_tlb_others,
2142 
2143  .pte_update = paravirt_nop,
2144  .pte_update_defer = paravirt_nop,
2145 
2146  .pgd_alloc = xen_pgd_alloc,
2147  .pgd_free = xen_pgd_free,
2148 
2149  .alloc_pte = xen_alloc_pte_init,
2150  .release_pte = xen_release_pte_init,
2151  .alloc_pmd = xen_alloc_pmd_init,
2152  .release_pmd = xen_release_pmd_init,
2153 
2154  .set_pte = xen_set_pte_init,
2155  .set_pte_at = xen_set_pte_at,
2156  .set_pmd = xen_set_pmd_hyper,
2157 
2158  .ptep_modify_prot_start = __ptep_modify_prot_start,
2159  .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2160 
2161  .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2162  .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2163 
2164  .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2165  .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2166 
2167 #ifdef CONFIG_X86_PAE
2168  .set_pte_atomic = xen_set_pte_atomic,
2169  .pte_clear = xen_pte_clear,
2170  .pmd_clear = xen_pmd_clear,
2171 #endif /* CONFIG_X86_PAE */
2172  .set_pud = xen_set_pud_hyper,
2173 
2174  .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2175  .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2176 
2177 #if PAGETABLE_LEVELS == 4
2178  .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2179  .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2180  .set_pgd = xen_set_pgd_hyper,
2181 
2182  .alloc_pud = xen_alloc_pmd_init,
2183  .release_pud = xen_release_pmd_init,
2184 #endif /* PAGETABLE_LEVELS == 4 */
2185 
2186  .activate_mm = xen_activate_mm,
2187  .dup_mmap = xen_dup_mmap,
2188  .exit_mmap = xen_exit_mmap,
2189 
2190  .lazy_mode = {
2191  .enter = paravirt_enter_lazy_mmu,
2192  .leave = xen_leave_lazy_mmu,
2193  },
2194 
2195  .set_fixmap = xen_set_fixmap,
2196 };
2197 
2199 {
2200  x86_init.mapping.pagetable_reserve = xen_mapping_pagetable_reserve;
2201  x86_init.paging.pagetable_init = xen_pagetable_init;
2202  pv_mmu_ops = xen_mmu_ops;
2203 
2204  memset(dummy_mapping, 0xff, PAGE_SIZE);
2205 }
2206 
2207 /* Protected by xen_reservation_lock. */
2208 #define MAX_CONTIG_ORDER 9 /* 2MB */
2209 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2210 
2211 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2212 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2213  unsigned long *in_frames,
2214  unsigned long *out_frames)
2215 {
2216  int i;
2217  struct multicall_space mcs;
2218 
2219  xen_mc_batch();
2220  for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2221  mcs = __xen_mc_entry(0);
2222 
2223  if (in_frames)
2224  in_frames[i] = virt_to_mfn(vaddr);
2225 
2226  MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2228 
2229  if (out_frames)
2230  out_frames[i] = virt_to_pfn(vaddr);
2231  }
2232  xen_mc_issue(0);
2233 }
2234 
2235 /*
2236  * Update the pfn-to-mfn mappings for a virtual address range, either to
2237  * point to an array of mfns, or contiguously from a single starting
2238  * mfn.
2239  */
2240 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2241  unsigned long *mfns,
2242  unsigned long first_mfn)
2243 {
2244  unsigned i, limit;
2245  unsigned long mfn;
2246 
2247  xen_mc_batch();
2248 
2249  limit = 1u << order;
2250  for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2251  struct multicall_space mcs;
2252  unsigned flags;
2253 
2254  mcs = __xen_mc_entry(0);
2255  if (mfns)
2256  mfn = mfns[i];
2257  else
2258  mfn = first_mfn + i;
2259 
2260  if (i < (limit - 1))
2261  flags = 0;
2262  else {
2263  if (order == 0)
2264  flags = UVMF_INVLPG | UVMF_ALL;
2265  else
2266  flags = UVMF_TLB_FLUSH | UVMF_ALL;
2267  }
2268 
2269  MULTI_update_va_mapping(mcs.mc, vaddr,
2270  mfn_pte(mfn, PAGE_KERNEL), flags);
2271 
2272  set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2273  }
2274 
2275  xen_mc_issue(0);
2276 }
2277 
2278 /*
2279  * Perform the hypercall to exchange a region of our pfns to point to
2280  * memory with the required contiguous alignment. Takes the pfns as
2281  * input, and populates mfns as output.
2282  *
2283  * Returns a success code indicating whether the hypervisor was able to
2284  * satisfy the request or not.
2285  */
2286 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2287  unsigned long *pfns_in,
2288  unsigned long extents_out,
2289  unsigned int order_out,
2290  unsigned long *mfns_out,
2291  unsigned int address_bits)
2292 {
2293  long rc;
2294  int success;
2295 
2296  struct xen_memory_exchange exchange = {
2297  .in = {
2298  .nr_extents = extents_in,
2299  .extent_order = order_in,
2300  .extent_start = pfns_in,
2301  .domid = DOMID_SELF
2302  },
2303  .out = {
2304  .nr_extents = extents_out,
2305  .extent_order = order_out,
2306  .extent_start = mfns_out,
2307  .address_bits = address_bits,
2308  .domid = DOMID_SELF
2309  }
2310  };
2311 
2312  BUG_ON(extents_in << order_in != extents_out << order_out);
2313 
2314  rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2315  success = (exchange.nr_exchanged == extents_in);
2316 
2317  BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2318  BUG_ON(success && (rc != 0));
2319 
2320  return success;
2321 }
2322 
2323 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2324  unsigned int address_bits)
2325 {
2326  unsigned long *in_frames = discontig_frames, out_frame;
2327  unsigned long flags;
2328  int success;
2329 
2330  /*
2331  * Currently an auto-translated guest will not perform I/O, nor will
2332  * it require PAE page directories below 4GB. Therefore any calls to
2333  * this function are redundant and can be ignored.
2334  */
2335 
2336  if (xen_feature(XENFEAT_auto_translated_physmap))
2337  return 0;
2338 
2339  if (unlikely(order > MAX_CONTIG_ORDER))
2340  return -ENOMEM;
2341 
2342  memset((void *) vstart, 0, PAGE_SIZE << order);
2343 
2345 
2346  /* 1. Zap current PTEs, remembering MFNs. */
2347  xen_zap_pfn_range(vstart, order, in_frames, NULL);
2348 
2349  /* 2. Get a new contiguous memory extent. */
2350  out_frame = virt_to_pfn(vstart);
2351  success = xen_exchange_memory(1UL << order, 0, in_frames,
2352  1, order, &out_frame,
2353  address_bits);
2354 
2355  /* 3. Map the new extent in place of old pages. */
2356  if (success)
2357  xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2358  else
2359  xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2360 
2361  spin_unlock_irqrestore(&xen_reservation_lock, flags);
2362 
2363  return success ? 0 : -ENOMEM;
2364 }
2366 
2367 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2368 {
2369  unsigned long *out_frames = discontig_frames, in_frame;
2370  unsigned long flags;
2371  int success;
2372 
2373  if (xen_feature(XENFEAT_auto_translated_physmap))
2374  return;
2375 
2376  if (unlikely(order > MAX_CONTIG_ORDER))
2377  return;
2378 
2379  memset((void *) vstart, 0, PAGE_SIZE << order);
2380 
2382 
2383  /* 1. Find start MFN of contiguous extent. */
2384  in_frame = virt_to_mfn(vstart);
2385 
2386  /* 2. Zap current PTEs. */
2387  xen_zap_pfn_range(vstart, order, NULL, out_frames);
2388 
2389  /* 3. Do the exchange for non-contiguous MFNs. */
2390  success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2391  0, out_frames, 0);
2392 
2393  /* 4. Map new pages in place of old pages. */
2394  if (success)
2395  xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2396  else
2397  xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2398 
2399  spin_unlock_irqrestore(&xen_reservation_lock, flags);
2400 }
2402 
2403 #ifdef CONFIG_XEN_PVHVM
2404 #ifdef CONFIG_PROC_VMCORE
2405 /*
2406  * This function is used in two contexts:
2407  * - the kdump kernel has to check whether a pfn of the crashed kernel
2408  * was a ballooned page. vmcore is using this function to decide
2409  * whether to access a pfn of the crashed kernel.
2410  * - the kexec kernel has to check whether a pfn was ballooned by the
2411  * previous kernel. If the pfn is ballooned, handle it properly.
2412  * Returns 0 if the pfn is not backed by a RAM page, the caller may
2413  * handle the pfn special in this case.
2414  */
2415 static int xen_oldmem_pfn_is_ram(unsigned long pfn)
2416 {
2417  struct xen_hvm_get_mem_type a = {
2418  .domid = DOMID_SELF,
2419  .pfn = pfn,
2420  };
2421  int ram;
2422 
2424  return -ENXIO;
2425 
2426  switch (a.mem_type) {
2427  case HVMMEM_mmio_dm:
2428  ram = 0;
2429  break;
2430  case HVMMEM_ram_rw:
2431  case HVMMEM_ram_ro:
2432  default:
2433  ram = 1;
2434  break;
2435  }
2436 
2437  return ram;
2438 }
2439 #endif
2440 
2441 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2442 {
2443  struct xen_hvm_pagetable_dying a;
2444  int rc;
2445 
2446  a.domid = DOMID_SELF;
2447  a.gpa = __pa(mm->pgd);
2449  WARN_ON_ONCE(rc < 0);
2450 }
2451 
2452 static int is_pagetable_dying_supported(void)
2453 {
2454  struct xen_hvm_pagetable_dying a;
2455  int rc = 0;
2456 
2457  a.domid = DOMID_SELF;
2458  a.gpa = 0x00;
2460  if (rc < 0) {
2461  printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2462  return 0;
2463  }
2464  return 1;
2465 }
2466 
2467 void __init xen_hvm_init_mmu_ops(void)
2468 {
2469  if (is_pagetable_dying_supported())
2470  pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2471 #ifdef CONFIG_PROC_VMCORE
2472  register_oldmem_pfn_is_ram(&xen_oldmem_pfn_is_ram);
2473 #endif
2474 }
2475 #endif
2476 
2477 #define REMAP_BATCH_SIZE 16
2478 
2479 struct remap_data {
2480  unsigned long mfn;
2483 };
2484 
2485 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2486  unsigned long addr, void *data)
2487 {
2488  struct remap_data *rmd = data;
2489  pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2490 
2491  rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2492  rmd->mmu_update->val = pte_val_ma(pte);
2493  rmd->mmu_update++;
2494 
2495  return 0;
2496 }
2497 
2499  unsigned long addr,
2500  unsigned long mfn, int nr,
2501  pgprot_t prot, unsigned domid)
2502 {
2503  struct remap_data rmd;
2504  struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2505  int batch;
2506  unsigned long range;
2507  int err = 0;
2508 
2509  if (xen_feature(XENFEAT_auto_translated_physmap))
2510  return -EINVAL;
2511 
2512  prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2513 
2514  BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2515 
2516  rmd.mfn = mfn;
2517  rmd.prot = prot;
2518 
2519  while (nr) {
2520  batch = min(REMAP_BATCH_SIZE, nr);
2521  range = (unsigned long)batch << PAGE_SHIFT;
2522 
2523  rmd.mmu_update = mmu_update;
2524  err = apply_to_page_range(vma->vm_mm, addr, range,
2525  remap_area_mfn_pte_fn, &rmd);
2526  if (err)
2527  goto out;
2528 
2529  err = HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid);
2530  if (err < 0)
2531  goto out;
2532 
2533  nr -= batch;
2534  addr += range;
2535  }
2536 
2537  err = 0;
2538 out:
2539 
2541 
2542  return err;
2543 }