~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/xen/mmu.c

Version: ~ [ linux-5.4-rc7 ] ~ [ linux-5.3.11 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.84 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.154 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.201 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.201 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.77 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 <jeremy@xensource.com>, 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>
 73 #include <xen/interface/hvm/hvm_op.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 
133 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
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 
156 void make_lowmem_page_readonly(void *vaddr)
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 
172 void make_lowmem_page_readwrite(void *vaddr)
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 }
214 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
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 
221         mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
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 
239         mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
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 
297         if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
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);
325                 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
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 
342 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
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 
358         u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
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 
475 void xen_set_pat(u64 pat)
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);
801         spin_lock_nest_lock(ptl, &mm->page_table_lock);
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 
896                 kmap_flush_unused();
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  */
1053 void xen_mm_unpin_all(void)
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)
1096                 leave_mm(smp_processor_id());
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 {
1106         cpumask_var_t mask;
1107         unsigned cpu;
1108 
1109         if (current->active_mm == mm) {
1110                 if (current->mm == mm)
1111                         load_cr3(swapper_pg_dir);
1112                 else
1113                         leave_mm(smp_processor_id());
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 #ifdef CONFIG_X86_64
1182 static void __init xen_cleanhighmap(unsigned long vaddr,
1183                                     unsigned long vaddr_end)
1184 {
1185         unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1186         pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1187 
1188         /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1189          * We include the PMD passed in on _both_ boundaries. */
1190         for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PAGE_SIZE));
1191                         pmd++, vaddr += PMD_SIZE) {
1192                 if (pmd_none(*pmd))
1193                         continue;
1194                 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1195                         set_pmd(pmd, __pmd(0));
1196         }
1197         /* In case we did something silly, we should crash in this function
1198          * instead of somewhere later and be confusing. */
1199         xen_mc_flush();
1200 }
1201 #endif
1202 static void __init xen_pagetable_init(void)
1203 {
1204 #ifdef CONFIG_X86_64
1205         unsigned long size;
1206         unsigned long addr;
1207 #endif
1208         paging_init();
1209         xen_setup_shared_info();
1210 #ifdef CONFIG_X86_64
1211         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1212                 unsigned long new_mfn_list;
1213 
1214                 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1215 
1216                 /* On 32-bit, we get zero so this never gets executed. */
1217                 new_mfn_list = xen_revector_p2m_tree();
1218                 if (new_mfn_list && new_mfn_list != xen_start_info->mfn_list) {
1219                         /* using __ka address and sticking INVALID_P2M_ENTRY! */
1220                         memset((void *)xen_start_info->mfn_list, 0xff, size);
1221 
1222                         /* We should be in __ka space. */
1223                         BUG_ON(xen_start_info->mfn_list < __START_KERNEL_map);
1224                         addr = xen_start_info->mfn_list;
1225                         /* We roundup to the PMD, which means that if anybody at this stage is
1226                          * using the __ka address of xen_start_info or xen_start_info->shared_info
1227                          * they are in going to crash. Fortunatly we have already revectored
1228                          * in xen_setup_kernel_pagetable and in xen_setup_shared_info. */
1229                         size = roundup(size, PMD_SIZE);
1230                         xen_cleanhighmap(addr, addr + size);
1231 
1232                         size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1233                         memblock_free(__pa(xen_start_info->mfn_list), size);
1234                         /* And revector! Bye bye old array */
1235                         xen_start_info->mfn_list = new_mfn_list;
1236                 } else
1237                         goto skip;
1238         }
1239         /* At this stage, cleanup_highmap has already cleaned __ka space
1240          * from _brk_limit way up to the max_pfn_mapped (which is the end of
1241          * the ramdisk). We continue on, erasing PMD entries that point to page
1242          * tables - do note that they are accessible at this stage via __va.
1243          * For good measure we also round up to the PMD - which means that if
1244          * anybody is using __ka address to the initial boot-stack - and try
1245          * to use it - they are going to crash. The xen_start_info has been
1246          * taken care of already in xen_setup_kernel_pagetable. */
1247         addr = xen_start_info->pt_base;
1248         size = roundup(xen_start_info->nr_pt_frames * PAGE_SIZE, PMD_SIZE);
1249 
1250         xen_cleanhighmap(addr, addr + size);
1251         xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1252 #ifdef DEBUG
1253         /* This is superflous and is not neccessary, but you know what
1254          * lets do it. The MODULES_VADDR -> MODULES_END should be clear of
1255          * anything at this stage. */
1256         xen_cleanhighmap(MODULES_VADDR, roundup(MODULES_VADDR, PUD_SIZE) - 1);
1257 #endif
1258 skip:
1259 #endif
1260         xen_post_allocator_init();
1261 }
1262 static void xen_write_cr2(unsigned long cr2)
1263 {
1264         this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1265 }
1266 
1267 static unsigned long xen_read_cr2(void)
1268 {
1269         return this_cpu_read(xen_vcpu)->arch.cr2;
1270 }
1271 
1272 unsigned long xen_read_cr2_direct(void)
1273 {
1274         return this_cpu_read(xen_vcpu_info.arch.cr2);
1275 }
1276 
1277 void xen_flush_tlb_all(void)
1278 {
1279         struct mmuext_op *op;
1280         struct multicall_space mcs;
1281 
1282         trace_xen_mmu_flush_tlb_all(0);
1283 
1284         preempt_disable();
1285 
1286         mcs = xen_mc_entry(sizeof(*op));
1287 
1288         op = mcs.args;
1289         op->cmd = MMUEXT_TLB_FLUSH_ALL;
1290         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1291 
1292         xen_mc_issue(PARAVIRT_LAZY_MMU);
1293 
1294         preempt_enable();
1295 }
1296 static void xen_flush_tlb(void)
1297 {
1298         struct mmuext_op *op;
1299         struct multicall_space mcs;
1300 
1301         trace_xen_mmu_flush_tlb(0);
1302 
1303         preempt_disable();
1304 
1305         mcs = xen_mc_entry(sizeof(*op));
1306 
1307         op = mcs.args;
1308         op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1309         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1310 
1311         xen_mc_issue(PARAVIRT_LAZY_MMU);
1312 
1313         preempt_enable();
1314 }
1315 
1316 static void xen_flush_tlb_single(unsigned long addr)
1317 {
1318         struct mmuext_op *op;
1319         struct multicall_space mcs;
1320 
1321         trace_xen_mmu_flush_tlb_single(addr);
1322 
1323         preempt_disable();
1324 
1325         mcs = xen_mc_entry(sizeof(*op));
1326         op = mcs.args;
1327         op->cmd = MMUEXT_INVLPG_LOCAL;
1328         op->arg1.linear_addr = addr & PAGE_MASK;
1329         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1330 
1331         xen_mc_issue(PARAVIRT_LAZY_MMU);
1332 
1333         preempt_enable();
1334 }
1335 
1336 static void xen_flush_tlb_others(const struct cpumask *cpus,
1337                                  struct mm_struct *mm, unsigned long start,
1338                                  unsigned long end)
1339 {
1340         struct {
1341                 struct mmuext_op op;
1342 #ifdef CONFIG_SMP
1343                 DECLARE_BITMAP(mask, num_processors);
1344 #else
1345                 DECLARE_BITMAP(mask, NR_CPUS);
1346 #endif
1347         } *args;
1348         struct multicall_space mcs;
1349 
1350         trace_xen_mmu_flush_tlb_others(cpus, mm, start, end);
1351 
1352         if (cpumask_empty(cpus))
1353                 return;         /* nothing to do */
1354 
1355         mcs = xen_mc_entry(sizeof(*args));
1356         args = mcs.args;
1357         args->op.arg2.vcpumask = to_cpumask(args->mask);
1358 
1359         /* Remove us, and any offline CPUS. */
1360         cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1361         cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1362 
1363         args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1364         if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) {
1365                 args->op.cmd = MMUEXT_INVLPG_MULTI;
1366                 args->op.arg1.linear_addr = start;
1367         }
1368 
1369         MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1370 
1371         xen_mc_issue(PARAVIRT_LAZY_MMU);
1372 }
1373 
1374 static unsigned long xen_read_cr3(void)
1375 {
1376         return this_cpu_read(xen_cr3);
1377 }
1378 
1379 static void set_current_cr3(void *v)
1380 {
1381         this_cpu_write(xen_current_cr3, (unsigned long)v);
1382 }
1383 
1384 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1385 {
1386         struct mmuext_op op;
1387         unsigned long mfn;
1388 
1389         trace_xen_mmu_write_cr3(kernel, cr3);
1390 
1391         if (cr3)
1392                 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1393         else
1394                 mfn = 0;
1395 
1396         WARN_ON(mfn == 0 && kernel);
1397 
1398         op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1399         op.arg1.mfn = mfn;
1400 
1401         xen_extend_mmuext_op(&op);
1402 
1403         if (kernel) {
1404                 this_cpu_write(xen_cr3, cr3);
1405 
1406                 /* Update xen_current_cr3 once the batch has actually
1407                    been submitted. */
1408                 xen_mc_callback(set_current_cr3, (void *)cr3);
1409         }
1410 }
1411 static void xen_write_cr3(unsigned long cr3)
1412 {
1413         BUG_ON(preemptible());
1414 
1415         xen_mc_batch();  /* disables interrupts */
1416 
1417         /* Update while interrupts are disabled, so its atomic with
1418            respect to ipis */
1419         this_cpu_write(xen_cr3, cr3);
1420 
1421         __xen_write_cr3(true, cr3);
1422 
1423 #ifdef CONFIG_X86_64
1424         {
1425                 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1426                 if (user_pgd)
1427                         __xen_write_cr3(false, __pa(user_pgd));
1428                 else
1429                         __xen_write_cr3(false, 0);
1430         }
1431 #endif
1432 
1433         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1434 }
1435 
1436 #ifdef CONFIG_X86_64
1437 /*
1438  * At the start of the day - when Xen launches a guest, it has already
1439  * built pagetables for the guest. We diligently look over them
1440  * in xen_setup_kernel_pagetable and graft as appropiate them in the
1441  * init_level4_pgt and its friends. Then when we are happy we load
1442  * the new init_level4_pgt - and continue on.
1443  *
1444  * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1445  * up the rest of the pagetables. When it has completed it loads the cr3.
1446  * N.B. that baremetal would start at 'start_kernel' (and the early
1447  * #PF handler would create bootstrap pagetables) - so we are running
1448  * with the same assumptions as what to do when write_cr3 is executed
1449  * at this point.
1450  *
1451  * Since there are no user-page tables at all, we have two variants
1452  * of xen_write_cr3 - the early bootup (this one), and the late one
1453  * (xen_write_cr3). The reason we have to do that is that in 64-bit
1454  * the Linux kernel and user-space are both in ring 3 while the
1455  * hypervisor is in ring 0.
1456  */
1457 static void __init xen_write_cr3_init(unsigned long cr3)
1458 {
1459         BUG_ON(preemptible());
1460 
1461         xen_mc_batch();  /* disables interrupts */
1462 
1463         /* Update while interrupts are disabled, so its atomic with
1464            respect to ipis */
1465         this_cpu_write(xen_cr3, cr3);
1466 
1467         __xen_write_cr3(true, cr3);
1468 
1469         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1470 }
1471 #endif
1472 
1473 static int xen_pgd_alloc(struct mm_struct *mm)
1474 {
1475         pgd_t *pgd = mm->pgd;
1476         int ret = 0;
1477 
1478         BUG_ON(PagePinned(virt_to_page(pgd)));
1479 
1480 #ifdef CONFIG_X86_64
1481         {
1482                 struct page *page = virt_to_page(pgd);
1483                 pgd_t *user_pgd;
1484 
1485                 BUG_ON(page->private != 0);
1486 
1487                 ret = -ENOMEM;
1488 
1489                 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1490                 page->private = (unsigned long)user_pgd;
1491 
1492                 if (user_pgd != NULL) {
1493                         user_pgd[pgd_index(VSYSCALL_START)] =
1494                                 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1495                         ret = 0;
1496                 }
1497 
1498                 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1499         }
1500 #endif
1501 
1502         return ret;
1503 }
1504 
1505 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1506 {
1507 #ifdef CONFIG_X86_64
1508         pgd_t *user_pgd = xen_get_user_pgd(pgd);
1509 
1510         if (user_pgd)
1511                 free_page((unsigned long)user_pgd);
1512 #endif
1513 }
1514 
1515 #ifdef CONFIG_X86_32
1516 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1517 {
1518         /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1519         if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1520                 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1521                                pte_val_ma(pte));
1522 
1523         return pte;
1524 }
1525 #else /* CONFIG_X86_64 */
1526 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1527 {
1528         return pte;
1529 }
1530 #endif /* CONFIG_X86_64 */
1531 
1532 /*
1533  * Init-time set_pte while constructing initial pagetables, which
1534  * doesn't allow RO page table pages to be remapped RW.
1535  *
1536  * If there is no MFN for this PFN then this page is initially
1537  * ballooned out so clear the PTE (as in decrease_reservation() in
1538  * drivers/xen/balloon.c).
1539  *
1540  * Many of these PTE updates are done on unpinned and writable pages
1541  * and doing a hypercall for these is unnecessary and expensive.  At
1542  * this point it is not possible to tell if a page is pinned or not,
1543  * so always write the PTE directly and rely on Xen trapping and
1544  * emulating any updates as necessary.
1545  */
1546 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1547 {
1548         if (pte_mfn(pte) != INVALID_P2M_ENTRY)
1549                 pte = mask_rw_pte(ptep, pte);
1550         else
1551                 pte = __pte_ma(0);
1552 
1553         native_set_pte(ptep, pte);
1554 }
1555 
1556 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1557 {
1558         struct mmuext_op op;
1559         op.cmd = cmd;
1560         op.arg1.mfn = pfn_to_mfn(pfn);
1561         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1562                 BUG();
1563 }
1564 
1565 /* Early in boot, while setting up the initial pagetable, assume
1566    everything is pinned. */
1567 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1568 {
1569 #ifdef CONFIG_FLATMEM
1570         BUG_ON(mem_map);        /* should only be used early */
1571 #endif
1572         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1573         pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1574 }
1575 
1576 /* Used for pmd and pud */
1577 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1578 {
1579 #ifdef CONFIG_FLATMEM
1580         BUG_ON(mem_map);        /* should only be used early */
1581 #endif
1582         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1583 }
1584 
1585 /* Early release_pte assumes that all pts are pinned, since there's
1586    only init_mm and anything attached to that is pinned. */
1587 static void __init xen_release_pte_init(unsigned long pfn)
1588 {
1589         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1590         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1591 }
1592 
1593 static void __init xen_release_pmd_init(unsigned long pfn)
1594 {
1595         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1596 }
1597 
1598 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1599 {
1600         struct multicall_space mcs;
1601         struct mmuext_op *op;
1602 
1603         mcs = __xen_mc_entry(sizeof(*op));
1604         op = mcs.args;
1605         op->cmd = cmd;
1606         op->arg1.mfn = pfn_to_mfn(pfn);
1607 
1608         MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1609 }
1610 
1611 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1612 {
1613         struct multicall_space mcs;
1614         unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1615 
1616         mcs = __xen_mc_entry(0);
1617         MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1618                                 pfn_pte(pfn, prot), 0);
1619 }
1620 
1621 /* This needs to make sure the new pte page is pinned iff its being
1622    attached to a pinned pagetable. */
1623 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1624                                     unsigned level)
1625 {
1626         bool pinned = PagePinned(virt_to_page(mm->pgd));
1627 
1628         trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1629 
1630         if (pinned) {
1631                 struct page *page = pfn_to_page(pfn);
1632 
1633                 SetPagePinned(page);
1634 
1635                 if (!PageHighMem(page)) {
1636                         xen_mc_batch();
1637 
1638                         __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1639 
1640                         if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1641                                 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1642 
1643                         xen_mc_issue(PARAVIRT_LAZY_MMU);
1644                 } else {
1645                         /* make sure there are no stray mappings of
1646                            this page */
1647                         kmap_flush_unused();
1648                 }
1649         }
1650 }
1651 
1652 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1653 {
1654         xen_alloc_ptpage(mm, pfn, PT_PTE);
1655 }
1656 
1657 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1658 {
1659         xen_alloc_ptpage(mm, pfn, PT_PMD);
1660 }
1661 
1662 /* This should never happen until we're OK to use struct page */
1663 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1664 {
1665         struct page *page = pfn_to_page(pfn);
1666         bool pinned = PagePinned(page);
1667 
1668         trace_xen_mmu_release_ptpage(pfn, level, pinned);
1669 
1670         if (pinned) {
1671                 if (!PageHighMem(page)) {
1672                         xen_mc_batch();
1673 
1674                         if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1675                                 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1676 
1677                         __set_pfn_prot(pfn, PAGE_KERNEL);
1678 
1679                         xen_mc_issue(PARAVIRT_LAZY_MMU);
1680                 }
1681                 ClearPagePinned(page);
1682         }
1683 }
1684 
1685 static void xen_release_pte(unsigned long pfn)
1686 {
1687         xen_release_ptpage(pfn, PT_PTE);
1688 }
1689 
1690 static void xen_release_pmd(unsigned long pfn)
1691 {
1692         xen_release_ptpage(pfn, PT_PMD);
1693 }
1694 
1695 #if PAGETABLE_LEVELS == 4
1696 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1697 {
1698         xen_alloc_ptpage(mm, pfn, PT_PUD);
1699 }
1700 
1701 static void xen_release_pud(unsigned long pfn)
1702 {
1703         xen_release_ptpage(pfn, PT_PUD);
1704 }
1705 #endif
1706 
1707 void __init xen_reserve_top(void)
1708 {
1709 #ifdef CONFIG_X86_32
1710         unsigned long top = HYPERVISOR_VIRT_START;
1711         struct xen_platform_parameters pp;
1712 
1713         if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1714                 top = pp.virt_start;
1715 
1716         reserve_top_address(-top);
1717 #endif  /* CONFIG_X86_32 */
1718 }
1719 
1720 /*
1721  * Like __va(), but returns address in the kernel mapping (which is
1722  * all we have until the physical memory mapping has been set up.
1723  */
1724 static void *__ka(phys_addr_t paddr)
1725 {
1726 #ifdef CONFIG_X86_64
1727         return (void *)(paddr + __START_KERNEL_map);
1728 #else
1729         return __va(paddr);
1730 #endif
1731 }
1732 
1733 /* Convert a machine address to physical address */
1734 static unsigned long m2p(phys_addr_t maddr)
1735 {
1736         phys_addr_t paddr;
1737 
1738         maddr &= PTE_PFN_MASK;
1739         paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1740 
1741         return paddr;
1742 }
1743 
1744 /* Convert a machine address to kernel virtual */
1745 static void *m2v(phys_addr_t maddr)
1746 {
1747         return __ka(m2p(maddr));
1748 }
1749 
1750 /* Set the page permissions on an identity-mapped pages */
1751 static void set_page_prot_flags(void *addr, pgprot_t prot, unsigned long flags)
1752 {
1753         unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1754         pte_t pte = pfn_pte(pfn, prot);
1755 
1756         if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1757                 BUG();
1758 }
1759 static void set_page_prot(void *addr, pgprot_t prot)
1760 {
1761         return set_page_prot_flags(addr, prot, UVMF_NONE);
1762 }
1763 #ifdef CONFIG_X86_32
1764 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1765 {
1766         unsigned pmdidx, pteidx;
1767         unsigned ident_pte;
1768         unsigned long pfn;
1769 
1770         level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1771                                       PAGE_SIZE);
1772 
1773         ident_pte = 0;
1774         pfn = 0;
1775         for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1776                 pte_t *pte_page;
1777 
1778                 /* Reuse or allocate a page of ptes */
1779                 if (pmd_present(pmd[pmdidx]))
1780                         pte_page = m2v(pmd[pmdidx].pmd);
1781                 else {
1782                         /* Check for free pte pages */
1783                         if (ident_pte == LEVEL1_IDENT_ENTRIES)
1784                                 break;
1785 
1786                         pte_page = &level1_ident_pgt[ident_pte];
1787                         ident_pte += PTRS_PER_PTE;
1788 
1789                         pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1790                 }
1791 
1792                 /* Install mappings */
1793                 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1794                         pte_t pte;
1795 
1796 #ifdef CONFIG_X86_32
1797                         if (pfn > max_pfn_mapped)
1798                                 max_pfn_mapped = pfn;
1799 #endif
1800 
1801                         if (!pte_none(pte_page[pteidx]))
1802                                 continue;
1803 
1804                         pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1805                         pte_page[pteidx] = pte;
1806                 }
1807         }
1808 
1809         for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1810                 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1811 
1812         set_page_prot(pmd, PAGE_KERNEL_RO);
1813 }
1814 #endif
1815 void __init xen_setup_machphys_mapping(void)
1816 {
1817         struct xen_machphys_mapping mapping;
1818 
1819         if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1820                 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1821                 machine_to_phys_nr = mapping.max_mfn + 1;
1822         } else {
1823                 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1824         }
1825 #ifdef CONFIG_X86_32
1826         WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1827                 < machine_to_phys_mapping);
1828 #endif
1829 }
1830 
1831 #ifdef CONFIG_X86_64
1832 static void convert_pfn_mfn(void *v)
1833 {
1834         pte_t *pte = v;
1835         int i;
1836 
1837         /* All levels are converted the same way, so just treat them
1838            as ptes. */
1839         for (i = 0; i < PTRS_PER_PTE; i++)
1840                 pte[i] = xen_make_pte(pte[i].pte);
1841 }
1842 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1843                                  unsigned long addr)
1844 {
1845         if (*pt_base == PFN_DOWN(__pa(addr))) {
1846                 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1847                 clear_page((void *)addr);
1848                 (*pt_base)++;
1849         }
1850         if (*pt_end == PFN_DOWN(__pa(addr))) {
1851                 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1852                 clear_page((void *)addr);
1853                 (*pt_end)--;
1854         }
1855 }
1856 /*
1857  * Set up the initial kernel pagetable.
1858  *
1859  * We can construct this by grafting the Xen provided pagetable into
1860  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1861  * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt.  This
1862  * means that only the kernel has a physical mapping to start with -
1863  * but that's enough to get __va working.  We need to fill in the rest
1864  * of the physical mapping once some sort of allocator has been set
1865  * up.
1866  */
1867 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1868 {
1869         pud_t *l3;
1870         pmd_t *l2;
1871         unsigned long addr[3];
1872         unsigned long pt_base, pt_end;
1873         unsigned i;
1874 
1875         /* max_pfn_mapped is the last pfn mapped in the initial memory
1876          * mappings. Considering that on Xen after the kernel mappings we
1877          * have the mappings of some pages that don't exist in pfn space, we
1878          * set max_pfn_mapped to the last real pfn mapped. */
1879         max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1880 
1881         pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1882         pt_end = pt_base + xen_start_info->nr_pt_frames;
1883 
1884         /* Zap identity mapping */
1885         init_level4_pgt[0] = __pgd(0);
1886 
1887         /* Pre-constructed entries are in pfn, so convert to mfn */
1888         /* L4[272] -> level3_ident_pgt
1889          * L4[511] -> level3_kernel_pgt */
1890         convert_pfn_mfn(init_level4_pgt);
1891 
1892         /* L3_i[0] -> level2_ident_pgt */
1893         convert_pfn_mfn(level3_ident_pgt);
1894         /* L3_k[510] -> level2_kernel_pgt
1895          * L3_i[511] -> level2_fixmap_pgt */
1896         convert_pfn_mfn(level3_kernel_pgt);
1897 
1898         /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1899         l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1900         l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1901 
1902         addr[0] = (unsigned long)pgd;
1903         addr[1] = (unsigned long)l3;
1904         addr[2] = (unsigned long)l2;
1905         /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1906          * Both L4[272][0] and L4[511][511] have entries that point to the same
1907          * L2 (PMD) tables. Meaning that if you modify it in __va space
1908          * it will be also modified in the __ka space! (But if you just
1909          * modify the PMD table to point to other PTE's or none, then you
1910          * are OK - which is what cleanup_highmap does) */
1911         copy_page(level2_ident_pgt, l2);
1912         /* Graft it onto L4[511][511] */
1913         copy_page(level2_kernel_pgt, l2);
1914 
1915         /* Get [511][510] and graft that in level2_fixmap_pgt */
1916         l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1917         l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1918         copy_page(level2_fixmap_pgt, l2);
1919         /* Note that we don't do anything with level1_fixmap_pgt which
1920          * we don't need. */
1921 
1922         /* Make pagetable pieces RO */
1923         set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1924         set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1925         set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1926         set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1927         set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1928         set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1929         set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1930 
1931         /* Pin down new L4 */
1932         pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1933                           PFN_DOWN(__pa_symbol(init_level4_pgt)));
1934 
1935         /* Unpin Xen-provided one */
1936         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1937 
1938         /*
1939          * At this stage there can be no user pgd, and no page
1940          * structure to attach it to, so make sure we just set kernel
1941          * pgd.
1942          */
1943         xen_mc_batch();
1944         __xen_write_cr3(true, __pa(init_level4_pgt));
1945         xen_mc_issue(PARAVIRT_LAZY_CPU);
1946 
1947         /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1948          * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1949          * the initial domain. For guests using the toolstack, they are in:
1950          * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1951          * rip out the [L4] (pgd), but for guests we shave off three pages.
1952          */
1953         for (i = 0; i < ARRAY_SIZE(addr); i++)
1954                 check_pt_base(&pt_base, &pt_end, addr[i]);
1955 
1956         /* Our (by three pages) smaller Xen pagetable that we are using */
1957         memblock_reserve(PFN_PHYS(pt_base), (pt_end - pt_base) * PAGE_SIZE);
1958         /* Revector the xen_start_info */
1959         xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1960 }
1961 #else   /* !CONFIG_X86_64 */
1962 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1963 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1964 
1965 static void __init xen_write_cr3_init(unsigned long cr3)
1966 {
1967         unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1968 
1969         BUG_ON(read_cr3() != __pa(initial_page_table));
1970         BUG_ON(cr3 != __pa(swapper_pg_dir));
1971 
1972         /*
1973          * We are switching to swapper_pg_dir for the first time (from
1974          * initial_page_table) and therefore need to mark that page
1975          * read-only and then pin it.
1976          *
1977          * Xen disallows sharing of kernel PMDs for PAE
1978          * guests. Therefore we must copy the kernel PMD from
1979          * initial_page_table into a new kernel PMD to be used in
1980          * swapper_pg_dir.
1981          */
1982         swapper_kernel_pmd =
1983                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1984         copy_page(swapper_kernel_pmd, initial_kernel_pmd);
1985         swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1986                 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1987         set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1988 
1989         set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1990         xen_write_cr3(cr3);
1991         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1992 
1993         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1994                           PFN_DOWN(__pa(initial_page_table)));
1995         set_page_prot(initial_page_table, PAGE_KERNEL);
1996         set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1997 
1998         pv_mmu_ops.write_cr3 = &xen_write_cr3;
1999 }
2000 
2001 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2002 {
2003         pmd_t *kernel_pmd;
2004 
2005         initial_kernel_pmd =
2006                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2007 
2008         max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
2009                                   xen_start_info->nr_pt_frames * PAGE_SIZE +
2010                                   512*1024);
2011 
2012         kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2013         copy_page(initial_kernel_pmd, kernel_pmd);
2014 
2015         xen_map_identity_early(initial_kernel_pmd, max_pfn);
2016 
2017         copy_page(initial_page_table, pgd);
2018         initial_page_table[KERNEL_PGD_BOUNDARY] =
2019                 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2020 
2021         set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2022         set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2023         set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2024 
2025         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2026 
2027         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2028                           PFN_DOWN(__pa(initial_page_table)));
2029         xen_write_cr3(__pa(initial_page_table));
2030 
2031         memblock_reserve(__pa(xen_start_info->pt_base),
2032                          xen_start_info->nr_pt_frames * PAGE_SIZE);
2033 }
2034 #endif  /* CONFIG_X86_64 */
2035 
2036 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2037 
2038 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2039 {
2040         pte_t pte;
2041 
2042         phys >>= PAGE_SHIFT;
2043 
2044         switch (idx) {
2045         case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2046 #ifdef CONFIG_X86_F00F_BUG
2047         case FIX_F00F_IDT:
2048 #endif
2049 #ifdef CONFIG_X86_32
2050         case FIX_WP_TEST:
2051         case FIX_VDSO:
2052 # ifdef CONFIG_HIGHMEM
2053         case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2054 # endif
2055 #else
2056         case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
2057         case VVAR_PAGE:
2058 #endif
2059         case FIX_TEXT_POKE0:
2060         case FIX_TEXT_POKE1:
2061                 /* All local page mappings */
2062                 pte = pfn_pte(phys, prot);
2063                 break;
2064 
2065 #ifdef CONFIG_X86_LOCAL_APIC
2066         case FIX_APIC_BASE:     /* maps dummy local APIC */
2067                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2068                 break;
2069 #endif
2070 
2071 #ifdef CONFIG_X86_IO_APIC
2072         case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2073                 /*
2074                  * We just don't map the IO APIC - all access is via
2075                  * hypercalls.  Keep the address in the pte for reference.
2076                  */
2077                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2078                 break;
2079 #endif
2080 
2081         case FIX_PARAVIRT_BOOTMAP:
2082                 /* This is an MFN, but it isn't an IO mapping from the
2083                    IO domain */
2084                 pte = mfn_pte(phys, prot);
2085                 break;
2086 
2087         default:
2088                 /* By default, set_fixmap is used for hardware mappings */
2089                 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
2090                 break;
2091         }
2092 
2093         __native_set_fixmap(idx, pte);
2094 
2095 #ifdef CONFIG_X86_64
2096         /* Replicate changes to map the vsyscall page into the user
2097            pagetable vsyscall mapping. */
2098         if ((idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) ||
2099             idx == VVAR_PAGE) {
2100                 unsigned long vaddr = __fix_to_virt(idx);
2101                 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2102         }
2103 #endif
2104 }
2105 
2106 static void __init xen_post_allocator_init(void)
2107 {
2108         pv_mmu_ops.set_pte = xen_set_pte;
2109         pv_mmu_ops.set_pmd = xen_set_pmd;
2110         pv_mmu_ops.set_pud = xen_set_pud;
2111 #if PAGETABLE_LEVELS == 4
2112         pv_mmu_ops.set_pgd = xen_set_pgd;
2113 #endif
2114 
2115         /* This will work as long as patching hasn't happened yet
2116            (which it hasn't) */
2117         pv_mmu_ops.alloc_pte = xen_alloc_pte;
2118         pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2119         pv_mmu_ops.release_pte = xen_release_pte;
2120         pv_mmu_ops.release_pmd = xen_release_pmd;
2121 #if PAGETABLE_LEVELS == 4
2122         pv_mmu_ops.alloc_pud = xen_alloc_pud;
2123         pv_mmu_ops.release_pud = xen_release_pud;
2124 #endif
2125 
2126 #ifdef CONFIG_X86_64
2127         pv_mmu_ops.write_cr3 = &xen_write_cr3;
2128         SetPagePinned(virt_to_page(level3_user_vsyscall));
2129 #endif
2130         xen_mark_init_mm_pinned();
2131 }
2132 
2133 static void xen_leave_lazy_mmu(void)
2134 {
2135         preempt_disable();
2136         xen_mc_flush();
2137         paravirt_leave_lazy_mmu();
2138         preempt_enable();
2139 }
2140 
2141 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2142         .read_cr2 = xen_read_cr2,
2143         .write_cr2 = xen_write_cr2,
2144 
2145         .read_cr3 = xen_read_cr3,
2146         .write_cr3 = xen_write_cr3_init,
2147 
2148         .flush_tlb_user = xen_flush_tlb,
2149         .flush_tlb_kernel = xen_flush_tlb,
2150         .flush_tlb_single = xen_flush_tlb_single,
2151         .flush_tlb_others = xen_flush_tlb_others,
2152 
2153         .pte_update = paravirt_nop,
2154         .pte_update_defer = paravirt_nop,
2155 
2156         .pgd_alloc = xen_pgd_alloc,
2157         .pgd_free = xen_pgd_free,
2158 
2159         .alloc_pte = xen_alloc_pte_init,
2160         .release_pte = xen_release_pte_init,
2161         .alloc_pmd = xen_alloc_pmd_init,
2162         .release_pmd = xen_release_pmd_init,
2163 
2164         .set_pte = xen_set_pte_init,
2165         .set_pte_at = xen_set_pte_at,
2166         .set_pmd = xen_set_pmd_hyper,
2167 
2168         .ptep_modify_prot_start = __ptep_modify_prot_start,
2169         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2170 
2171         .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2172         .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2173 
2174         .make_pte = PV_CALLEE_SAVE(xen_make_pte),
2175         .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2176 
2177 #ifdef CONFIG_X86_PAE
2178         .set_pte_atomic = xen_set_pte_atomic,
2179         .pte_clear = xen_pte_clear,
2180         .pmd_clear = xen_pmd_clear,
2181 #endif  /* CONFIG_X86_PAE */
2182         .set_pud = xen_set_pud_hyper,
2183 
2184         .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2185         .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2186 
2187 #if PAGETABLE_LEVELS == 4
2188         .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2189         .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2190         .set_pgd = xen_set_pgd_hyper,
2191 
2192         .alloc_pud = xen_alloc_pmd_init,
2193         .release_pud = xen_release_pmd_init,
2194 #endif  /* PAGETABLE_LEVELS == 4 */
2195 
2196         .activate_mm = xen_activate_mm,
2197         .dup_mmap = xen_dup_mmap,
2198         .exit_mmap = xen_exit_mmap,
2199 
2200         .lazy_mode = {
2201                 .enter = paravirt_enter_lazy_mmu,
2202                 .leave = xen_leave_lazy_mmu,
2203                 .flush = paravirt_flush_lazy_mmu,
2204         },
2205 
2206         .set_fixmap = xen_set_fixmap,
2207 };
2208 
2209 void __init xen_init_mmu_ops(void)
2210 {
2211         x86_init.paging.pagetable_init = xen_pagetable_init;
2212         pv_mmu_ops = xen_mmu_ops;
2213 
2214         memset(dummy_mapping, 0xff, PAGE_SIZE);
2215 }
2216 
2217 /* Protected by xen_reservation_lock. */
2218 #define MAX_CONTIG_ORDER 9 /* 2MB */
2219 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2220 
2221 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2222 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2223                                 unsigned long *in_frames,
2224                                 unsigned long *out_frames)
2225 {
2226         int i;
2227         struct multicall_space mcs;
2228 
2229         xen_mc_batch();
2230         for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2231                 mcs = __xen_mc_entry(0);
2232 
2233                 if (in_frames)
2234                         in_frames[i] = virt_to_mfn(vaddr);
2235 
2236                 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2237                 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2238 
2239                 if (out_frames)
2240                         out_frames[i] = virt_to_pfn(vaddr);
2241         }
2242         xen_mc_issue(0);
2243 }
2244 
2245 /*
2246  * Update the pfn-to-mfn mappings for a virtual address range, either to
2247  * point to an array of mfns, or contiguously from a single starting
2248  * mfn.
2249  */
2250 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2251                                      unsigned long *mfns,
2252                                      unsigned long first_mfn)
2253 {
2254         unsigned i, limit;
2255         unsigned long mfn;
2256 
2257         xen_mc_batch();
2258 
2259         limit = 1u << order;
2260         for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2261                 struct multicall_space mcs;
2262                 unsigned flags;
2263 
2264                 mcs = __xen_mc_entry(0);
2265                 if (mfns)
2266                         mfn = mfns[i];
2267                 else
2268                         mfn = first_mfn + i;
2269 
2270                 if (i < (limit - 1))
2271                         flags = 0;
2272                 else {
2273                         if (order == 0)
2274                                 flags = UVMF_INVLPG | UVMF_ALL;
2275                         else
2276                                 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2277                 }
2278 
2279                 MULTI_update_va_mapping(mcs.mc, vaddr,
2280                                 mfn_pte(mfn, PAGE_KERNEL), flags);
2281 
2282                 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2283         }
2284 
2285         xen_mc_issue(0);
2286 }
2287 
2288 /*
2289  * Perform the hypercall to exchange a region of our pfns to point to
2290  * memory with the required contiguous alignment.  Takes the pfns as
2291  * input, and populates mfns as output.
2292  *
2293  * Returns a success code indicating whether the hypervisor was able to
2294  * satisfy the request or not.
2295  */
2296 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2297                                unsigned long *pfns_in,
2298                                unsigned long extents_out,
2299                                unsigned int order_out,
2300                                unsigned long *mfns_out,
2301                                unsigned int address_bits)
2302 {
2303         long rc;
2304         int success;
2305 
2306         struct xen_memory_exchange exchange = {
2307                 .in = {
2308                         .nr_extents   = extents_in,
2309                         .extent_order = order_in,
2310                         .extent_start = pfns_in,
2311                         .domid        = DOMID_SELF
2312                 },
2313                 .out = {
2314                         .nr_extents   = extents_out,
2315                         .extent_order = order_out,
2316                         .extent_start = mfns_out,
2317                         .address_bits = address_bits,
2318                         .domid        = DOMID_SELF
2319                 }
2320         };
2321 
2322         BUG_ON(extents_in << order_in != extents_out << order_out);
2323 
2324         rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2325         success = (exchange.nr_exchanged == extents_in);
2326 
2327         BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2328         BUG_ON(success && (rc != 0));
2329 
2330         return success;
2331 }
2332 
2333 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2334                                  unsigned int address_bits)
2335 {
2336         unsigned long *in_frames = discontig_frames, out_frame;
2337         unsigned long  flags;
2338         int            success;
2339 
2340         /*
2341          * Currently an auto-translated guest will not perform I/O, nor will
2342          * it require PAE page directories below 4GB. Therefore any calls to
2343          * this function are redundant and can be ignored.
2344          */
2345 
2346         if (xen_feature(XENFEAT_auto_translated_physmap))
2347                 return 0;
2348 
2349         if (unlikely(order > MAX_CONTIG_ORDER))
2350                 return -ENOMEM;
2351 
2352         memset((void *) vstart, 0, PAGE_SIZE << order);
2353 
2354         spin_lock_irqsave(&xen_reservation_lock, flags);
2355 
2356         /* 1. Zap current PTEs, remembering MFNs. */
2357         xen_zap_pfn_range(vstart, order, in_frames, NULL);
2358 
2359         /* 2. Get a new contiguous memory extent. */
2360         out_frame = virt_to_pfn(vstart);
2361         success = xen_exchange_memory(1UL << order, 0, in_frames,
2362                                       1, order, &out_frame,
2363                                       address_bits);
2364 
2365         /* 3. Map the new extent in place of old pages. */
2366         if (success)
2367                 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2368         else
2369                 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2370 
2371         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2372 
2373         return success ? 0 : -ENOMEM;
2374 }
2375 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2376 
2377 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2378 {
2379         unsigned long *out_frames = discontig_frames, in_frame;
2380         unsigned long  flags;
2381         int success;
2382 
2383         if (xen_feature(XENFEAT_auto_translated_physmap))
2384                 return;
2385 
2386         if (unlikely(order > MAX_CONTIG_ORDER))
2387                 return;
2388 
2389         memset((void *) vstart, 0, PAGE_SIZE << order);
2390 
2391         spin_lock_irqsave(&xen_reservation_lock, flags);
2392 
2393         /* 1. Find start MFN of contiguous extent. */
2394         in_frame = virt_to_mfn(vstart);
2395 
2396         /* 2. Zap current PTEs. */
2397         xen_zap_pfn_range(vstart, order, NULL, out_frames);
2398 
2399         /* 3. Do the exchange for non-contiguous MFNs. */
2400         success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2401                                         0, out_frames, 0);
2402 
2403         /* 4. Map new pages in place of old pages. */
2404         if (success)
2405                 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2406         else
2407                 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2408 
2409         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2410 }
2411 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2412 
2413 #ifdef CONFIG_XEN_PVHVM
2414 #ifdef CONFIG_PROC_VMCORE
2415 /*
2416  * This function is used in two contexts:
2417  * - the kdump kernel has to check whether a pfn of the crashed kernel
2418  *   was a ballooned page. vmcore is using this function to decide
2419  *   whether to access a pfn of the crashed kernel.
2420  * - the kexec kernel has to check whether a pfn was ballooned by the
2421  *   previous kernel. If the pfn is ballooned, handle it properly.
2422  * Returns 0 if the pfn is not backed by a RAM page, the caller may
2423  * handle the pfn special in this case.
2424  */
2425 static int xen_oldmem_pfn_is_ram(unsigned long pfn)
2426 {
2427         struct xen_hvm_get_mem_type a = {
2428                 .domid = DOMID_SELF,
2429                 .pfn = pfn,
2430         };
2431         int ram;
2432 
2433         if (HYPERVISOR_hvm_op(HVMOP_get_mem_type, &a))
2434                 return -ENXIO;
2435 
2436         switch (a.mem_type) {
2437                 case HVMMEM_mmio_dm:
2438                         ram = 0;
2439                         break;
2440                 case HVMMEM_ram_rw:
2441                 case HVMMEM_ram_ro:
2442                 default:
2443                         ram = 1;
2444                         break;
2445         }
2446 
2447         return ram;
2448 }
2449 #endif
2450 
2451 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2452 {
2453         struct xen_hvm_pagetable_dying a;
2454         int rc;
2455 
2456         a.domid = DOMID_SELF;
2457         a.gpa = __pa(mm->pgd);
2458         rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2459         WARN_ON_ONCE(rc < 0);
2460 }
2461 
2462 static int is_pagetable_dying_supported(void)
2463 {
2464         struct xen_hvm_pagetable_dying a;
2465         int rc = 0;
2466 
2467         a.domid = DOMID_SELF;
2468         a.gpa = 0x00;
2469         rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2470         if (rc < 0) {
2471                 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2472                 return 0;
2473         }
2474         return 1;
2475 }
2476 
2477 void __init xen_hvm_init_mmu_ops(void)
2478 {
2479         if (is_pagetable_dying_supported())
2480                 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2481 #ifdef CONFIG_PROC_VMCORE
2482         register_oldmem_pfn_is_ram(&xen_oldmem_pfn_is_ram);
2483 #endif
2484 }
2485 #endif
2486 
2487 #define REMAP_BATCH_SIZE 16
2488 
2489 struct remap_data {
2490         unsigned long mfn;
2491         pgprot_t prot;
2492         struct mmu_update *mmu_update;
2493 };
2494 
2495 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2496                                  unsigned long addr, void *data)
2497 {
2498         struct remap_data *rmd = data;
2499         pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2500 
2501         rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2502         rmd->mmu_update->val = pte_val_ma(pte);
2503         rmd->mmu_update++;
2504 
2505         return 0;
2506 }
2507 
2508 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2509                                unsigned long addr,
2510                                xen_pfn_t mfn, int nr,
2511                                pgprot_t prot, unsigned domid,
2512                                struct page **pages)
2513 
2514 {
2515         struct remap_data rmd;
2516         struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2517         int batch;
2518         unsigned long range;
2519         int err = 0;
2520 
2521         if (xen_feature(XENFEAT_auto_translated_physmap))
2522                 return -EINVAL;
2523 
2524         prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2525 
2526         BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2527 
2528         rmd.mfn = mfn;
2529         rmd.prot = prot;
2530 
2531         while (nr) {
2532                 batch = min(REMAP_BATCH_SIZE, nr);
2533                 range = (unsigned long)batch << PAGE_SHIFT;
2534 
2535                 rmd.mmu_update = mmu_update;
2536                 err = apply_to_page_range(vma->vm_mm, addr, range,
2537                                           remap_area_mfn_pte_fn, &rmd);
2538                 if (err)
2539                         goto out;
2540 
2541                 err = HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid);
2542                 if (err < 0)
2543                         goto out;
2544 
2545                 nr -= batch;
2546                 addr += range;
2547         }
2548 
2549         err = 0;
2550 out:
2551 
2552         xen_flush_tlb_all();
2553 
2554         return err;
2555 }
2556 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2557 
2558 /* Returns: 0 success */
2559 int xen_unmap_domain_mfn_range(struct vm_area_struct *vma,
2560                                int numpgs, struct page **pages)
2561 {
2562         if (!pages || !xen_feature(XENFEAT_auto_translated_physmap))
2563                 return 0;
2564 
2565         return -EINVAL;
2566 }
2567 EXPORT_SYMBOL_GPL(xen_unmap_domain_mfn_range);
2568 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp