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

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

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.3 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.30 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.107 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.164 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ 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-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ 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 
 51 #include <asm/pgtable.h>
 52 #include <asm/tlbflush.h>
 53 #include <asm/fixmap.h>
 54 #include <asm/mmu_context.h>
 55 #include <asm/setup.h>
 56 #include <asm/paravirt.h>
 57 #include <asm/e820.h>
 58 #include <asm/linkage.h>
 59 #include <asm/page.h>
 60 #include <asm/init.h>
 61 #include <asm/pat.h>
 62 #include <asm/smp.h>
 63 
 64 #include <asm/xen/hypercall.h>
 65 #include <asm/xen/hypervisor.h>
 66 
 67 #include <xen/xen.h>
 68 #include <xen/page.h>
 69 #include <xen/interface/xen.h>
 70 #include <xen/interface/hvm/hvm_op.h>
 71 #include <xen/interface/version.h>
 72 #include <xen/interface/memory.h>
 73 #include <xen/hvc-console.h>
 74 
 75 #include "multicalls.h"
 76 #include "mmu.h"
 77 #include "debugfs.h"
 78 
 79 /*
 80  * Protects atomic reservation decrease/increase against concurrent increases.
 81  * Also protects non-atomic updates of current_pages and balloon lists.
 82  */
 83 DEFINE_SPINLOCK(xen_reservation_lock);
 84 
 85 /*
 86  * Identity map, in addition to plain kernel map.  This needs to be
 87  * large enough to allocate page table pages to allocate the rest.
 88  * Each page can map 2MB.
 89  */
 90 #define LEVEL1_IDENT_ENTRIES    (PTRS_PER_PTE * 4)
 91 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
 92 
 93 #ifdef CONFIG_X86_64
 94 /* l3 pud for userspace vsyscall mapping */
 95 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
 96 #endif /* CONFIG_X86_64 */
 97 
 98 /*
 99  * Note about cr3 (pagetable base) values:
100  *
101  * xen_cr3 contains the current logical cr3 value; it contains the
102  * last set cr3.  This may not be the current effective cr3, because
103  * its update may be being lazily deferred.  However, a vcpu looking
104  * at its own cr3 can use this value knowing that it everything will
105  * be self-consistent.
106  *
107  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
108  * hypercall to set the vcpu cr3 is complete (so it may be a little
109  * out of date, but it will never be set early).  If one vcpu is
110  * looking at another vcpu's cr3 value, it should use this variable.
111  */
112 DEFINE_PER_CPU(unsigned long, xen_cr3);  /* cr3 stored as physaddr */
113 DEFINE_PER_CPU(unsigned long, xen_current_cr3);  /* actual vcpu cr3 */
114 
115 
116 /*
117  * Just beyond the highest usermode address.  STACK_TOP_MAX has a
118  * redzone above it, so round it up to a PGD boundary.
119  */
120 #define USER_LIMIT      ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
121 
122 unsigned long arbitrary_virt_to_mfn(void *vaddr)
123 {
124         xmaddr_t maddr = arbitrary_virt_to_machine(vaddr);
125 
126         return PFN_DOWN(maddr.maddr);
127 }
128 
129 xmaddr_t arbitrary_virt_to_machine(void *vaddr)
130 {
131         unsigned long address = (unsigned long)vaddr;
132         unsigned int level;
133         pte_t *pte;
134         unsigned offset;
135 
136         /*
137          * if the PFN is in the linear mapped vaddr range, we can just use
138          * the (quick) virt_to_machine() p2m lookup
139          */
140         if (virt_addr_valid(vaddr))
141                 return virt_to_machine(vaddr);
142 
143         /* otherwise we have to do a (slower) full page-table walk */
144 
145         pte = lookup_address(address, &level);
146         BUG_ON(pte == NULL);
147         offset = address & ~PAGE_MASK;
148         return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset);
149 }
150 EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine);
151 
152 void make_lowmem_page_readonly(void *vaddr)
153 {
154         pte_t *pte, ptev;
155         unsigned long address = (unsigned long)vaddr;
156         unsigned int level;
157 
158         pte = lookup_address(address, &level);
159         if (pte == NULL)
160                 return;         /* vaddr missing */
161 
162         ptev = pte_wrprotect(*pte);
163 
164         if (HYPERVISOR_update_va_mapping(address, ptev, 0))
165                 BUG();
166 }
167 
168 void make_lowmem_page_readwrite(void *vaddr)
169 {
170         pte_t *pte, ptev;
171         unsigned long address = (unsigned long)vaddr;
172         unsigned int level;
173 
174         pte = lookup_address(address, &level);
175         if (pte == NULL)
176                 return;         /* vaddr missing */
177 
178         ptev = pte_mkwrite(*pte);
179 
180         if (HYPERVISOR_update_va_mapping(address, ptev, 0))
181                 BUG();
182 }
183 
184 
185 static bool xen_page_pinned(void *ptr)
186 {
187         struct page *page = virt_to_page(ptr);
188 
189         return PagePinned(page);
190 }
191 
192 void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid)
193 {
194         struct multicall_space mcs;
195         struct mmu_update *u;
196 
197         mcs = xen_mc_entry(sizeof(*u));
198         u = mcs.args;
199 
200         /* ptep might be kmapped when using 32-bit HIGHPTE */
201         u->ptr = virt_to_machine(ptep).maddr;
202         u->val = pte_val_ma(pteval);
203 
204         MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid);
205 
206         xen_mc_issue(PARAVIRT_LAZY_MMU);
207 }
208 EXPORT_SYMBOL_GPL(xen_set_domain_pte);
209 
210 static void xen_extend_mmu_update(const struct mmu_update *update)
211 {
212         struct multicall_space mcs;
213         struct mmu_update *u;
214 
215         mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
216 
217         if (mcs.mc != NULL) {
218                 mcs.mc->args[1]++;
219         } else {
220                 mcs = __xen_mc_entry(sizeof(*u));
221                 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
222         }
223 
224         u = mcs.args;
225         *u = *update;
226 }
227 
228 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
229 {
230         struct mmu_update u;
231 
232         preempt_disable();
233 
234         xen_mc_batch();
235 
236         /* ptr may be ioremapped for 64-bit pagetable setup */
237         u.ptr = arbitrary_virt_to_machine(ptr).maddr;
238         u.val = pmd_val_ma(val);
239         xen_extend_mmu_update(&u);
240 
241         xen_mc_issue(PARAVIRT_LAZY_MMU);
242 
243         preempt_enable();
244 }
245 
246 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
247 {
248         /* If page is not pinned, we can just update the entry
249            directly */
250         if (!xen_page_pinned(ptr)) {
251                 *ptr = val;
252                 return;
253         }
254 
255         xen_set_pmd_hyper(ptr, val);
256 }
257 
258 /*
259  * Associate a virtual page frame with a given physical page frame
260  * and protection flags for that frame.
261  */
262 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
263 {
264         set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
265 }
266 
267 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
268 {
269         struct mmu_update u;
270 
271         if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
272                 return false;
273 
274         xen_mc_batch();
275 
276         u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
277         u.val = pte_val_ma(pteval);
278         xen_extend_mmu_update(&u);
279 
280         xen_mc_issue(PARAVIRT_LAZY_MMU);
281 
282         return true;
283 }
284 
285 static void xen_set_pte(pte_t *ptep, pte_t pteval)
286 {
287         if (!xen_batched_set_pte(ptep, pteval))
288                 native_set_pte(ptep, pteval);
289 }
290 
291 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
292                     pte_t *ptep, pte_t pteval)
293 {
294         xen_set_pte(ptep, pteval);
295 }
296 
297 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
298                                  unsigned long addr, pte_t *ptep)
299 {
300         /* Just return the pte as-is.  We preserve the bits on commit */
301         return *ptep;
302 }
303 
304 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
305                                  pte_t *ptep, pte_t pte)
306 {
307         struct mmu_update u;
308 
309         xen_mc_batch();
310 
311         u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
312         u.val = pte_val_ma(pte);
313         xen_extend_mmu_update(&u);
314 
315         xen_mc_issue(PARAVIRT_LAZY_MMU);
316 }
317 
318 /* Assume pteval_t is equivalent to all the other *val_t types. */
319 static pteval_t pte_mfn_to_pfn(pteval_t val)
320 {
321         if (val & _PAGE_PRESENT) {
322                 unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
323                 unsigned long pfn = mfn_to_pfn(mfn);
324 
325                 pteval_t flags = val & PTE_FLAGS_MASK;
326                 if (unlikely(pfn == ~0))
327                         val = flags & ~_PAGE_PRESENT;
328                 else
329                         val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
330         }
331 
332         return val;
333 }
334 
335 static pteval_t pte_pfn_to_mfn(pteval_t val)
336 {
337         if (val & _PAGE_PRESENT) {
338                 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
339                 pteval_t flags = val & PTE_FLAGS_MASK;
340                 unsigned long mfn;
341 
342                 if (!xen_feature(XENFEAT_auto_translated_physmap))
343                         mfn = get_phys_to_machine(pfn);
344                 else
345                         mfn = pfn;
346                 /*
347                  * If there's no mfn for the pfn, then just create an
348                  * empty non-present pte.  Unfortunately this loses
349                  * information about the original pfn, so
350                  * pte_mfn_to_pfn is asymmetric.
351                  */
352                 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
353                         mfn = 0;
354                         flags = 0;
355                 } else {
356                         /*
357                          * Paramount to do this test _after_ the
358                          * INVALID_P2M_ENTRY as INVALID_P2M_ENTRY &
359                          * IDENTITY_FRAME_BIT resolves to true.
360                          */
361                         mfn &= ~FOREIGN_FRAME_BIT;
362                         if (mfn & IDENTITY_FRAME_BIT) {
363                                 mfn &= ~IDENTITY_FRAME_BIT;
364                                 flags |= _PAGE_IOMAP;
365                         }
366                 }
367                 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
368         }
369 
370         return val;
371 }
372 
373 static pteval_t iomap_pte(pteval_t val)
374 {
375         if (val & _PAGE_PRESENT) {
376                 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
377                 pteval_t flags = val & PTE_FLAGS_MASK;
378 
379                 /* We assume the pte frame number is a MFN, so
380                    just use it as-is. */
381                 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
382         }
383 
384         return val;
385 }
386 
387 static pteval_t xen_pte_val(pte_t pte)
388 {
389         pteval_t pteval = pte.pte;
390 
391         /* If this is a WC pte, convert back from Xen WC to Linux WC */
392         if ((pteval & (_PAGE_PAT | _PAGE_PCD | _PAGE_PWT)) == _PAGE_PAT) {
393                 WARN_ON(!pat_enabled);
394                 pteval = (pteval & ~_PAGE_PAT) | _PAGE_PWT;
395         }
396 
397         if (xen_initial_domain() && (pteval & _PAGE_IOMAP))
398                 return pteval;
399 
400         return pte_mfn_to_pfn(pteval);
401 }
402 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
403 
404 static pgdval_t xen_pgd_val(pgd_t pgd)
405 {
406         return pte_mfn_to_pfn(pgd.pgd);
407 }
408 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
409 
410 /*
411  * Xen's PAT setup is part of its ABI, though I assume entries 6 & 7
412  * are reserved for now, to correspond to the Intel-reserved PAT
413  * types.
414  *
415  * We expect Linux's PAT set as follows:
416  *
417  * Idx  PTE flags        Linux    Xen    Default
418  * 0                     WB       WB     WB
419  * 1            PWT      WC       WT     WT
420  * 2        PCD          UC-      UC-    UC-
421  * 3        PCD PWT      UC       UC     UC
422  * 4    PAT              WB       WC     WB
423  * 5    PAT     PWT      WC       WP     WT
424  * 6    PAT PCD          UC-      UC     UC-
425  * 7    PAT PCD PWT      UC       UC     UC
426  */
427 
428 void xen_set_pat(u64 pat)
429 {
430         /* We expect Linux to use a PAT setting of
431          * UC UC- WC WB (ignoring the PAT flag) */
432         WARN_ON(pat != 0x0007010600070106ull);
433 }
434 
435 static pte_t xen_make_pte(pteval_t pte)
436 {
437         phys_addr_t addr = (pte & PTE_PFN_MASK);
438 
439         /* If Linux is trying to set a WC pte, then map to the Xen WC.
440          * If _PAGE_PAT is set, then it probably means it is really
441          * _PAGE_PSE, so avoid fiddling with the PAT mapping and hope
442          * things work out OK...
443          *
444          * (We should never see kernel mappings with _PAGE_PSE set,
445          * but we could see hugetlbfs mappings, I think.).
446          */
447         if (pat_enabled && !WARN_ON(pte & _PAGE_PAT)) {
448                 if ((pte & (_PAGE_PCD | _PAGE_PWT)) == _PAGE_PWT)
449                         pte = (pte & ~(_PAGE_PCD | _PAGE_PWT)) | _PAGE_PAT;
450         }
451 
452         /*
453          * Unprivileged domains are allowed to do IOMAPpings for
454          * PCI passthrough, but not map ISA space.  The ISA
455          * mappings are just dummy local mappings to keep other
456          * parts of the kernel happy.
457          */
458         if (unlikely(pte & _PAGE_IOMAP) &&
459             (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
460                 pte = iomap_pte(pte);
461         } else {
462                 pte &= ~_PAGE_IOMAP;
463                 pte = pte_pfn_to_mfn(pte);
464         }
465 
466         return native_make_pte(pte);
467 }
468 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
469 
470 #ifdef CONFIG_XEN_DEBUG
471 pte_t xen_make_pte_debug(pteval_t pte)
472 {
473         phys_addr_t addr = (pte & PTE_PFN_MASK);
474         phys_addr_t other_addr;
475         bool io_page = false;
476         pte_t _pte;
477 
478         if (pte & _PAGE_IOMAP)
479                 io_page = true;
480 
481         _pte = xen_make_pte(pte);
482 
483         if (!addr)
484                 return _pte;
485 
486         if (io_page &&
487             (xen_initial_domain() || addr >= ISA_END_ADDRESS)) {
488                 other_addr = pfn_to_mfn(addr >> PAGE_SHIFT) << PAGE_SHIFT;
489                 WARN_ONCE(addr != other_addr,
490                         "0x%lx is using VM_IO, but it is 0x%lx!\n",
491                         (unsigned long)addr, (unsigned long)other_addr);
492         } else {
493                 pteval_t iomap_set = (_pte.pte & PTE_FLAGS_MASK) & _PAGE_IOMAP;
494                 other_addr = (_pte.pte & PTE_PFN_MASK);
495                 WARN_ONCE((addr == other_addr) && (!io_page) && (!iomap_set),
496                         "0x%lx is missing VM_IO (and wasn't fixed)!\n",
497                         (unsigned long)addr);
498         }
499 
500         return _pte;
501 }
502 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_debug);
503 #endif
504 
505 static pgd_t xen_make_pgd(pgdval_t pgd)
506 {
507         pgd = pte_pfn_to_mfn(pgd);
508         return native_make_pgd(pgd);
509 }
510 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
511 
512 static pmdval_t xen_pmd_val(pmd_t pmd)
513 {
514         return pte_mfn_to_pfn(pmd.pmd);
515 }
516 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
517 
518 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
519 {
520         struct mmu_update u;
521 
522         preempt_disable();
523 
524         xen_mc_batch();
525 
526         /* ptr may be ioremapped for 64-bit pagetable setup */
527         u.ptr = arbitrary_virt_to_machine(ptr).maddr;
528         u.val = pud_val_ma(val);
529         xen_extend_mmu_update(&u);
530 
531         xen_mc_issue(PARAVIRT_LAZY_MMU);
532 
533         preempt_enable();
534 }
535 
536 static void xen_set_pud(pud_t *ptr, pud_t val)
537 {
538         /* If page is not pinned, we can just update the entry
539            directly */
540         if (!xen_page_pinned(ptr)) {
541                 *ptr = val;
542                 return;
543         }
544 
545         xen_set_pud_hyper(ptr, val);
546 }
547 
548 #ifdef CONFIG_X86_PAE
549 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
550 {
551         set_64bit((u64 *)ptep, native_pte_val(pte));
552 }
553 
554 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
555 {
556         if (!xen_batched_set_pte(ptep, native_make_pte(0)))
557                 native_pte_clear(mm, addr, ptep);
558 }
559 
560 static void xen_pmd_clear(pmd_t *pmdp)
561 {
562         set_pmd(pmdp, __pmd(0));
563 }
564 #endif  /* CONFIG_X86_PAE */
565 
566 static pmd_t xen_make_pmd(pmdval_t pmd)
567 {
568         pmd = pte_pfn_to_mfn(pmd);
569         return native_make_pmd(pmd);
570 }
571 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
572 
573 #if PAGETABLE_LEVELS == 4
574 static pudval_t xen_pud_val(pud_t pud)
575 {
576         return pte_mfn_to_pfn(pud.pud);
577 }
578 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
579 
580 static pud_t xen_make_pud(pudval_t pud)
581 {
582         pud = pte_pfn_to_mfn(pud);
583 
584         return native_make_pud(pud);
585 }
586 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
587 
588 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
589 {
590         pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
591         unsigned offset = pgd - pgd_page;
592         pgd_t *user_ptr = NULL;
593 
594         if (offset < pgd_index(USER_LIMIT)) {
595                 struct page *page = virt_to_page(pgd_page);
596                 user_ptr = (pgd_t *)page->private;
597                 if (user_ptr)
598                         user_ptr += offset;
599         }
600 
601         return user_ptr;
602 }
603 
604 static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
605 {
606         struct mmu_update u;
607 
608         u.ptr = virt_to_machine(ptr).maddr;
609         u.val = pgd_val_ma(val);
610         xen_extend_mmu_update(&u);
611 }
612 
613 /*
614  * Raw hypercall-based set_pgd, intended for in early boot before
615  * there's a page structure.  This implies:
616  *  1. The only existing pagetable is the kernel's
617  *  2. It is always pinned
618  *  3. It has no user pagetable attached to it
619  */
620 static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val)
621 {
622         preempt_disable();
623 
624         xen_mc_batch();
625 
626         __xen_set_pgd_hyper(ptr, val);
627 
628         xen_mc_issue(PARAVIRT_LAZY_MMU);
629 
630         preempt_enable();
631 }
632 
633 static void xen_set_pgd(pgd_t *ptr, pgd_t val)
634 {
635         pgd_t *user_ptr = xen_get_user_pgd(ptr);
636 
637         /* If page is not pinned, we can just update the entry
638            directly */
639         if (!xen_page_pinned(ptr)) {
640                 *ptr = val;
641                 if (user_ptr) {
642                         WARN_ON(xen_page_pinned(user_ptr));
643                         *user_ptr = val;
644                 }
645                 return;
646         }
647 
648         /* If it's pinned, then we can at least batch the kernel and
649            user updates together. */
650         xen_mc_batch();
651 
652         __xen_set_pgd_hyper(ptr, val);
653         if (user_ptr)
654                 __xen_set_pgd_hyper(user_ptr, val);
655 
656         xen_mc_issue(PARAVIRT_LAZY_MMU);
657 }
658 #endif  /* PAGETABLE_LEVELS == 4 */
659 
660 /*
661  * (Yet another) pagetable walker.  This one is intended for pinning a
662  * pagetable.  This means that it walks a pagetable and calls the
663  * callback function on each page it finds making up the page table,
664  * at every level.  It walks the entire pagetable, but it only bothers
665  * pinning pte pages which are below limit.  In the normal case this
666  * will be STACK_TOP_MAX, but at boot we need to pin up to
667  * FIXADDR_TOP.
668  *
669  * For 32-bit the important bit is that we don't pin beyond there,
670  * because then we start getting into Xen's ptes.
671  *
672  * For 64-bit, we must skip the Xen hole in the middle of the address
673  * space, just after the big x86-64 virtual hole.
674  */
675 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
676                           int (*func)(struct mm_struct *mm, struct page *,
677                                       enum pt_level),
678                           unsigned long limit)
679 {
680         int flush = 0;
681         unsigned hole_low, hole_high;
682         unsigned pgdidx_limit, pudidx_limit, pmdidx_limit;
683         unsigned pgdidx, pudidx, pmdidx;
684 
685         /* The limit is the last byte to be touched */
686         limit--;
687         BUG_ON(limit >= FIXADDR_TOP);
688 
689         if (xen_feature(XENFEAT_auto_translated_physmap))
690                 return 0;
691 
692         /*
693          * 64-bit has a great big hole in the middle of the address
694          * space, which contains the Xen mappings.  On 32-bit these
695          * will end up making a zero-sized hole and so is a no-op.
696          */
697         hole_low = pgd_index(USER_LIMIT);
698         hole_high = pgd_index(PAGE_OFFSET);
699 
700         pgdidx_limit = pgd_index(limit);
701 #if PTRS_PER_PUD > 1
702         pudidx_limit = pud_index(limit);
703 #else
704         pudidx_limit = 0;
705 #endif
706 #if PTRS_PER_PMD > 1
707         pmdidx_limit = pmd_index(limit);
708 #else
709         pmdidx_limit = 0;
710 #endif
711 
712         for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) {
713                 pud_t *pud;
714 
715                 if (pgdidx >= hole_low && pgdidx < hole_high)
716                         continue;
717 
718                 if (!pgd_val(pgd[pgdidx]))
719                         continue;
720 
721                 pud = pud_offset(&pgd[pgdidx], 0);
722 
723                 if (PTRS_PER_PUD > 1) /* not folded */
724                         flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
725 
726                 for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) {
727                         pmd_t *pmd;
728 
729                         if (pgdidx == pgdidx_limit &&
730                             pudidx > pudidx_limit)
731                                 goto out;
732 
733                         if (pud_none(pud[pudidx]))
734                                 continue;
735 
736                         pmd = pmd_offset(&pud[pudidx], 0);
737 
738                         if (PTRS_PER_PMD > 1) /* not folded */
739                                 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
740 
741                         for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) {
742                                 struct page *pte;
743 
744                                 if (pgdidx == pgdidx_limit &&
745                                     pudidx == pudidx_limit &&
746                                     pmdidx > pmdidx_limit)
747                                         goto out;
748 
749                                 if (pmd_none(pmd[pmdidx]))
750                                         continue;
751 
752                                 pte = pmd_page(pmd[pmdidx]);
753                                 flush |= (*func)(mm, pte, PT_PTE);
754                         }
755                 }
756         }
757 
758 out:
759         /* Do the top level last, so that the callbacks can use it as
760            a cue to do final things like tlb flushes. */
761         flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
762 
763         return flush;
764 }
765 
766 static int xen_pgd_walk(struct mm_struct *mm,
767                         int (*func)(struct mm_struct *mm, struct page *,
768                                     enum pt_level),
769                         unsigned long limit)
770 {
771         return __xen_pgd_walk(mm, mm->pgd, func, limit);
772 }
773 
774 /* If we're using split pte locks, then take the page's lock and
775    return a pointer to it.  Otherwise return NULL. */
776 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
777 {
778         spinlock_t *ptl = NULL;
779 
780 #if USE_SPLIT_PTLOCKS
781         ptl = __pte_lockptr(page);
782         spin_lock_nest_lock(ptl, &mm->page_table_lock);
783 #endif
784 
785         return ptl;
786 }
787 
788 static void xen_pte_unlock(void *v)
789 {
790         spinlock_t *ptl = v;
791         spin_unlock(ptl);
792 }
793 
794 static void xen_do_pin(unsigned level, unsigned long pfn)
795 {
796         struct mmuext_op *op;
797         struct multicall_space mcs;
798 
799         mcs = __xen_mc_entry(sizeof(*op));
800         op = mcs.args;
801         op->cmd = level;
802         op->arg1.mfn = pfn_to_mfn(pfn);
803         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
804 }
805 
806 static int xen_pin_page(struct mm_struct *mm, struct page *page,
807                         enum pt_level level)
808 {
809         unsigned pgfl = TestSetPagePinned(page);
810         int flush;
811 
812         if (pgfl)
813                 flush = 0;              /* already pinned */
814         else if (PageHighMem(page))
815                 /* kmaps need flushing if we found an unpinned
816                    highpage */
817                 flush = 1;
818         else {
819                 void *pt = lowmem_page_address(page);
820                 unsigned long pfn = page_to_pfn(page);
821                 struct multicall_space mcs = __xen_mc_entry(0);
822                 spinlock_t *ptl;
823 
824                 flush = 0;
825 
826                 /*
827                  * We need to hold the pagetable lock between the time
828                  * we make the pagetable RO and when we actually pin
829                  * it.  If we don't, then other users may come in and
830                  * attempt to update the pagetable by writing it,
831                  * which will fail because the memory is RO but not
832                  * pinned, so Xen won't do the trap'n'emulate.
833                  *
834                  * If we're using split pte locks, we can't hold the
835                  * entire pagetable's worth of locks during the
836                  * traverse, because we may wrap the preempt count (8
837                  * bits).  The solution is to mark RO and pin each PTE
838                  * page while holding the lock.  This means the number
839                  * of locks we end up holding is never more than a
840                  * batch size (~32 entries, at present).
841                  *
842                  * If we're not using split pte locks, we needn't pin
843                  * the PTE pages independently, because we're
844                  * protected by the overall pagetable lock.
845                  */
846                 ptl = NULL;
847                 if (level == PT_PTE)
848                         ptl = xen_pte_lock(page, mm);
849 
850                 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
851                                         pfn_pte(pfn, PAGE_KERNEL_RO),
852                                         level == PT_PGD ? UVMF_TLB_FLUSH : 0);
853 
854                 if (ptl) {
855                         xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
856 
857                         /* Queue a deferred unlock for when this batch
858                            is completed. */
859                         xen_mc_callback(xen_pte_unlock, ptl);
860                 }
861         }
862 
863         return flush;
864 }
865 
866 /* This is called just after a mm has been created, but it has not
867    been used yet.  We need to make sure that its pagetable is all
868    read-only, and can be pinned. */
869 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
870 {
871         xen_mc_batch();
872 
873         if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
874                 /* re-enable interrupts for flushing */
875                 xen_mc_issue(0);
876 
877                 kmap_flush_unused();
878 
879                 xen_mc_batch();
880         }
881 
882 #ifdef CONFIG_X86_64
883         {
884                 pgd_t *user_pgd = xen_get_user_pgd(pgd);
885 
886                 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
887 
888                 if (user_pgd) {
889                         xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
890                         xen_do_pin(MMUEXT_PIN_L4_TABLE,
891                                    PFN_DOWN(__pa(user_pgd)));
892                 }
893         }
894 #else /* CONFIG_X86_32 */
895 #ifdef CONFIG_X86_PAE
896         /* Need to make sure unshared kernel PMD is pinnable */
897         xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
898                      PT_PMD);
899 #endif
900         xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
901 #endif /* CONFIG_X86_64 */
902         xen_mc_issue(0);
903 }
904 
905 static void xen_pgd_pin(struct mm_struct *mm)
906 {
907         __xen_pgd_pin(mm, mm->pgd);
908 }
909 
910 /*
911  * On save, we need to pin all pagetables to make sure they get their
912  * mfns turned into pfns.  Search the list for any unpinned pgds and pin
913  * them (unpinned pgds are not currently in use, probably because the
914  * process is under construction or destruction).
915  *
916  * Expected to be called in stop_machine() ("equivalent to taking
917  * every spinlock in the system"), so the locking doesn't really
918  * matter all that much.
919  */
920 void xen_mm_pin_all(void)
921 {
922         struct page *page;
923 
924         spin_lock(&pgd_lock);
925 
926         list_for_each_entry(page, &pgd_list, lru) {
927                 if (!PagePinned(page)) {
928                         __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
929                         SetPageSavePinned(page);
930                 }
931         }
932 
933         spin_unlock(&pgd_lock);
934 }
935 
936 /*
937  * The init_mm pagetable is really pinned as soon as its created, but
938  * that's before we have page structures to store the bits.  So do all
939  * the book-keeping now.
940  */
941 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
942                                   enum pt_level level)
943 {
944         SetPagePinned(page);
945         return 0;
946 }
947 
948 static void __init xen_mark_init_mm_pinned(void)
949 {
950         xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
951 }
952 
953 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
954                           enum pt_level level)
955 {
956         unsigned pgfl = TestClearPagePinned(page);
957 
958         if (pgfl && !PageHighMem(page)) {
959                 void *pt = lowmem_page_address(page);
960                 unsigned long pfn = page_to_pfn(page);
961                 spinlock_t *ptl = NULL;
962                 struct multicall_space mcs;
963 
964                 /*
965                  * Do the converse to pin_page.  If we're using split
966                  * pte locks, we must be holding the lock for while
967                  * the pte page is unpinned but still RO to prevent
968                  * concurrent updates from seeing it in this
969                  * partially-pinned state.
970                  */
971                 if (level == PT_PTE) {
972                         ptl = xen_pte_lock(page, mm);
973 
974                         if (ptl)
975                                 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
976                 }
977 
978                 mcs = __xen_mc_entry(0);
979 
980                 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
981                                         pfn_pte(pfn, PAGE_KERNEL),
982                                         level == PT_PGD ? UVMF_TLB_FLUSH : 0);
983 
984                 if (ptl) {
985                         /* unlock when batch completed */
986                         xen_mc_callback(xen_pte_unlock, ptl);
987                 }
988         }
989 
990         return 0;               /* never need to flush on unpin */
991 }
992 
993 /* Release a pagetables pages back as normal RW */
994 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
995 {
996         xen_mc_batch();
997 
998         xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
999 
1000 #ifdef CONFIG_X86_64
1001         {
1002                 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1003 
1004                 if (user_pgd) {
1005                         xen_do_pin(MMUEXT_UNPIN_TABLE,
1006                                    PFN_DOWN(__pa(user_pgd)));
1007                         xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
1008                 }
1009         }
1010 #endif
1011 
1012 #ifdef CONFIG_X86_PAE
1013         /* Need to make sure unshared kernel PMD is unpinned */
1014         xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
1015                        PT_PMD);
1016 #endif
1017 
1018         __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
1019 
1020         xen_mc_issue(0);
1021 }
1022 
1023 static void xen_pgd_unpin(struct mm_struct *mm)
1024 {
1025         __xen_pgd_unpin(mm, mm->pgd);
1026 }
1027 
1028 /*
1029  * On resume, undo any pinning done at save, so that the rest of the
1030  * kernel doesn't see any unexpected pinned pagetables.
1031  */
1032 void xen_mm_unpin_all(void)
1033 {
1034         struct page *page;
1035 
1036         spin_lock(&pgd_lock);
1037 
1038         list_for_each_entry(page, &pgd_list, lru) {
1039                 if (PageSavePinned(page)) {
1040                         BUG_ON(!PagePinned(page));
1041                         __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
1042                         ClearPageSavePinned(page);
1043                 }
1044         }
1045 
1046         spin_unlock(&pgd_lock);
1047 }
1048 
1049 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
1050 {
1051         spin_lock(&next->page_table_lock);
1052         xen_pgd_pin(next);
1053         spin_unlock(&next->page_table_lock);
1054 }
1055 
1056 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
1057 {
1058         spin_lock(&mm->page_table_lock);
1059         xen_pgd_pin(mm);
1060         spin_unlock(&mm->page_table_lock);
1061 }
1062 
1063 
1064 #ifdef CONFIG_SMP
1065 /* Another cpu may still have their %cr3 pointing at the pagetable, so
1066    we need to repoint it somewhere else before we can unpin it. */
1067 static void drop_other_mm_ref(void *info)
1068 {
1069         struct mm_struct *mm = info;
1070         struct mm_struct *active_mm;
1071 
1072         active_mm = percpu_read(cpu_tlbstate.active_mm);
1073 
1074         if (active_mm == mm && percpu_read(cpu_tlbstate.state) != TLBSTATE_OK)
1075                 leave_mm(smp_processor_id());
1076 
1077         /* If this cpu still has a stale cr3 reference, then make sure
1078            it has been flushed. */
1079         if (percpu_read(xen_current_cr3) == __pa(mm->pgd))
1080                 load_cr3(swapper_pg_dir);
1081 }
1082 
1083 static void xen_drop_mm_ref(struct mm_struct *mm)
1084 {
1085         cpumask_var_t mask;
1086         unsigned cpu;
1087 
1088         if (current->active_mm == mm) {
1089                 if (current->mm == mm)
1090                         load_cr3(swapper_pg_dir);
1091                 else
1092                         leave_mm(smp_processor_id());
1093         }
1094 
1095         /* Get the "official" set of cpus referring to our pagetable. */
1096         if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1097                 for_each_online_cpu(cpu) {
1098                         if (!cpumask_test_cpu(cpu, mm_cpumask(mm))
1099                             && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1100                                 continue;
1101                         smp_call_function_single(cpu, drop_other_mm_ref, mm, 1);
1102                 }
1103                 return;
1104         }
1105         cpumask_copy(mask, mm_cpumask(mm));
1106 
1107         /* It's possible that a vcpu may have a stale reference to our
1108            cr3, because its in lazy mode, and it hasn't yet flushed
1109            its set of pending hypercalls yet.  In this case, we can
1110            look at its actual current cr3 value, and force it to flush
1111            if needed. */
1112         for_each_online_cpu(cpu) {
1113                 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1114                         cpumask_set_cpu(cpu, mask);
1115         }
1116 
1117         if (!cpumask_empty(mask))
1118                 smp_call_function_many(mask, drop_other_mm_ref, mm, 1);
1119         free_cpumask_var(mask);
1120 }
1121 #else
1122 static void xen_drop_mm_ref(struct mm_struct *mm)
1123 {
1124         if (current->active_mm == mm)
1125                 load_cr3(swapper_pg_dir);
1126 }
1127 #endif
1128 
1129 /*
1130  * While a process runs, Xen pins its pagetables, which means that the
1131  * hypervisor forces it to be read-only, and it controls all updates
1132  * to it.  This means that all pagetable updates have to go via the
1133  * hypervisor, which is moderately expensive.
1134  *
1135  * Since we're pulling the pagetable down, we switch to use init_mm,
1136  * unpin old process pagetable and mark it all read-write, which
1137  * allows further operations on it to be simple memory accesses.
1138  *
1139  * The only subtle point is that another CPU may be still using the
1140  * pagetable because of lazy tlb flushing.  This means we need need to
1141  * switch all CPUs off this pagetable before we can unpin it.
1142  */
1143 static void xen_exit_mmap(struct mm_struct *mm)
1144 {
1145         get_cpu();              /* make sure we don't move around */
1146         xen_drop_mm_ref(mm);
1147         put_cpu();
1148 
1149         spin_lock(&mm->page_table_lock);
1150 
1151         /* pgd may not be pinned in the error exit path of execve */
1152         if (xen_page_pinned(mm->pgd))
1153                 xen_pgd_unpin(mm);
1154 
1155         spin_unlock(&mm->page_table_lock);
1156 }
1157 
1158 static void __init xen_pagetable_setup_start(pgd_t *base)
1159 {
1160 }
1161 
1162 static __init void xen_mapping_pagetable_reserve(u64 start, u64 end)
1163 {
1164         /* reserve the range used */
1165         native_pagetable_reserve(start, end);
1166 
1167         /* set as RW the rest */
1168         printk(KERN_DEBUG "xen: setting RW the range %llx - %llx\n", end,
1169                         PFN_PHYS(pgt_buf_top));
1170         while (end < PFN_PHYS(pgt_buf_top)) {
1171                 make_lowmem_page_readwrite(__va(end));
1172                 end += PAGE_SIZE;
1173         }
1174 }
1175 
1176 static void xen_post_allocator_init(void);
1177 
1178 static void __init xen_pagetable_setup_done(pgd_t *base)
1179 {
1180         xen_setup_shared_info();
1181         xen_post_allocator_init();
1182 }
1183 
1184 static void xen_write_cr2(unsigned long cr2)
1185 {
1186         percpu_read(xen_vcpu)->arch.cr2 = cr2;
1187 }
1188 
1189 static unsigned long xen_read_cr2(void)
1190 {
1191         return percpu_read(xen_vcpu)->arch.cr2;
1192 }
1193 
1194 unsigned long xen_read_cr2_direct(void)
1195 {
1196         return percpu_read(xen_vcpu_info.arch.cr2);
1197 }
1198 
1199 static void xen_flush_tlb(void)
1200 {
1201         struct mmuext_op *op;
1202         struct multicall_space mcs;
1203 
1204         preempt_disable();
1205 
1206         mcs = xen_mc_entry(sizeof(*op));
1207 
1208         op = mcs.args;
1209         op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1210         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1211 
1212         xen_mc_issue(PARAVIRT_LAZY_MMU);
1213 
1214         preempt_enable();
1215 }
1216 
1217 static void xen_flush_tlb_single(unsigned long addr)
1218 {
1219         struct mmuext_op *op;
1220         struct multicall_space mcs;
1221 
1222         preempt_disable();
1223 
1224         mcs = xen_mc_entry(sizeof(*op));
1225         op = mcs.args;
1226         op->cmd = MMUEXT_INVLPG_LOCAL;
1227         op->arg1.linear_addr = addr & PAGE_MASK;
1228         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1229 
1230         xen_mc_issue(PARAVIRT_LAZY_MMU);
1231 
1232         preempt_enable();
1233 }
1234 
1235 static void xen_flush_tlb_others(const struct cpumask *cpus,
1236                                  struct mm_struct *mm, unsigned long va)
1237 {
1238         struct {
1239                 struct mmuext_op op;
1240 #ifdef CONFIG_SMP
1241                 DECLARE_BITMAP(mask, num_processors);
1242 #else
1243                 DECLARE_BITMAP(mask, NR_CPUS);
1244 #endif
1245         } *args;
1246         struct multicall_space mcs;
1247 
1248         if (cpumask_empty(cpus))
1249                 return;         /* nothing to do */
1250 
1251         mcs = xen_mc_entry(sizeof(*args));
1252         args = mcs.args;
1253         args->op.arg2.vcpumask = to_cpumask(args->mask);
1254 
1255         /* Remove us, and any offline CPUS. */
1256         cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1257         cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1258 
1259         if (va == TLB_FLUSH_ALL) {
1260                 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1261         } else {
1262                 args->op.cmd = MMUEXT_INVLPG_MULTI;
1263                 args->op.arg1.linear_addr = va;
1264         }
1265 
1266         MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1267 
1268         xen_mc_issue(PARAVIRT_LAZY_MMU);
1269 }
1270 
1271 static unsigned long xen_read_cr3(void)
1272 {
1273         return percpu_read(xen_cr3);
1274 }
1275 
1276 static void set_current_cr3(void *v)
1277 {
1278         percpu_write(xen_current_cr3, (unsigned long)v);
1279 }
1280 
1281 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1282 {
1283         struct mmuext_op *op;
1284         struct multicall_space mcs;
1285         unsigned long mfn;
1286 
1287         if (cr3)
1288                 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1289         else
1290                 mfn = 0;
1291 
1292         WARN_ON(mfn == 0 && kernel);
1293 
1294         mcs = __xen_mc_entry(sizeof(*op));
1295 
1296         op = mcs.args;
1297         op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1298         op->arg1.mfn = mfn;
1299 
1300         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1301 
1302         if (kernel) {
1303                 percpu_write(xen_cr3, cr3);
1304 
1305                 /* Update xen_current_cr3 once the batch has actually
1306                    been submitted. */
1307                 xen_mc_callback(set_current_cr3, (void *)cr3);
1308         }
1309 }
1310 
1311 static void xen_write_cr3(unsigned long cr3)
1312 {
1313         BUG_ON(preemptible());
1314 
1315         xen_mc_batch();  /* disables interrupts */
1316 
1317         /* Update while interrupts are disabled, so its atomic with
1318            respect to ipis */
1319         percpu_write(xen_cr3, cr3);
1320 
1321         __xen_write_cr3(true, cr3);
1322 
1323 #ifdef CONFIG_X86_64
1324         {
1325                 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1326                 if (user_pgd)
1327                         __xen_write_cr3(false, __pa(user_pgd));
1328                 else
1329                         __xen_write_cr3(false, 0);
1330         }
1331 #endif
1332 
1333         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1334 }
1335 
1336 static int xen_pgd_alloc(struct mm_struct *mm)
1337 {
1338         pgd_t *pgd = mm->pgd;
1339         int ret = 0;
1340 
1341         BUG_ON(PagePinned(virt_to_page(pgd)));
1342 
1343 #ifdef CONFIG_X86_64
1344         {
1345                 struct page *page = virt_to_page(pgd);
1346                 pgd_t *user_pgd;
1347 
1348                 BUG_ON(page->private != 0);
1349 
1350                 ret = -ENOMEM;
1351 
1352                 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1353                 page->private = (unsigned long)user_pgd;
1354 
1355                 if (user_pgd != NULL) {
1356                         user_pgd[pgd_index(VSYSCALL_START)] =
1357                                 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1358                         ret = 0;
1359                 }
1360 
1361                 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1362         }
1363 #endif
1364 
1365         return ret;
1366 }
1367 
1368 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1369 {
1370 #ifdef CONFIG_X86_64
1371         pgd_t *user_pgd = xen_get_user_pgd(pgd);
1372 
1373         if (user_pgd)
1374                 free_page((unsigned long)user_pgd);
1375 #endif
1376 }
1377 
1378 #ifdef CONFIG_X86_32
1379 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1380 {
1381         /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1382         if (pte_val_ma(*ptep) & _PAGE_PRESENT)
1383                 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1384                                pte_val_ma(pte));
1385 
1386         return pte;
1387 }
1388 #else /* CONFIG_X86_64 */
1389 static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte)
1390 {
1391         unsigned long pfn = pte_pfn(pte);
1392 
1393         /*
1394          * If the new pfn is within the range of the newly allocated
1395          * kernel pagetable, and it isn't being mapped into an
1396          * early_ioremap fixmap slot as a freshly allocated page, make sure
1397          * it is RO.
1398          */
1399         if (((!is_early_ioremap_ptep(ptep) &&
1400                         pfn >= pgt_buf_start && pfn < pgt_buf_top)) ||
1401                         (is_early_ioremap_ptep(ptep) && pfn != (pgt_buf_end - 1)))
1402                 pte = pte_wrprotect(pte);
1403 
1404         return pte;
1405 }
1406 #endif /* CONFIG_X86_64 */
1407 
1408 /* Init-time set_pte while constructing initial pagetables, which
1409    doesn't allow RO pagetable pages to be remapped RW */
1410 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1411 {
1412         pte = mask_rw_pte(ptep, pte);
1413 
1414         xen_set_pte(ptep, pte);
1415 }
1416 
1417 static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1418 {
1419         struct mmuext_op op;
1420         op.cmd = cmd;
1421         op.arg1.mfn = pfn_to_mfn(pfn);
1422         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1423                 BUG();
1424 }
1425 
1426 /* Early in boot, while setting up the initial pagetable, assume
1427    everything is pinned. */
1428 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1429 {
1430 #ifdef CONFIG_FLATMEM
1431         BUG_ON(mem_map);        /* should only be used early */
1432 #endif
1433         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1434         pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1435 }
1436 
1437 /* Used for pmd and pud */
1438 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1439 {
1440 #ifdef CONFIG_FLATMEM
1441         BUG_ON(mem_map);        /* should only be used early */
1442 #endif
1443         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1444 }
1445 
1446 /* Early release_pte assumes that all pts are pinned, since there's
1447    only init_mm and anything attached to that is pinned. */
1448 static void __init xen_release_pte_init(unsigned long pfn)
1449 {
1450         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1451         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1452 }
1453 
1454 static void __init xen_release_pmd_init(unsigned long pfn)
1455 {
1456         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1457 }
1458 
1459 /* This needs to make sure the new pte page is pinned iff its being
1460    attached to a pinned pagetable. */
1461 static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level)
1462 {
1463         struct page *page = pfn_to_page(pfn);
1464 
1465         if (PagePinned(virt_to_page(mm->pgd))) {
1466                 SetPagePinned(page);
1467 
1468                 if (!PageHighMem(page)) {
1469                         make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn)));
1470                         if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1471                                 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1472                 } else {
1473                         /* make sure there are no stray mappings of
1474                            this page */
1475                         kmap_flush_unused();
1476                 }
1477         }
1478 }
1479 
1480 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1481 {
1482         xen_alloc_ptpage(mm, pfn, PT_PTE);
1483 }
1484 
1485 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1486 {
1487         xen_alloc_ptpage(mm, pfn, PT_PMD);
1488 }
1489 
1490 /* This should never happen until we're OK to use struct page */
1491 static void xen_release_ptpage(unsigned long pfn, unsigned level)
1492 {
1493         struct page *page = pfn_to_page(pfn);
1494 
1495         if (PagePinned(page)) {
1496                 if (!PageHighMem(page)) {
1497                         if (level == PT_PTE && USE_SPLIT_PTLOCKS)
1498                                 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1499                         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1500                 }
1501                 ClearPagePinned(page);
1502         }
1503 }
1504 
1505 static void xen_release_pte(unsigned long pfn)
1506 {
1507         xen_release_ptpage(pfn, PT_PTE);
1508 }
1509 
1510 static void xen_release_pmd(unsigned long pfn)
1511 {
1512         xen_release_ptpage(pfn, PT_PMD);
1513 }
1514 
1515 #if PAGETABLE_LEVELS == 4
1516 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1517 {
1518         xen_alloc_ptpage(mm, pfn, PT_PUD);
1519 }
1520 
1521 static void xen_release_pud(unsigned long pfn)
1522 {
1523         xen_release_ptpage(pfn, PT_PUD);
1524 }
1525 #endif
1526 
1527 void __init xen_reserve_top(void)
1528 {
1529 #ifdef CONFIG_X86_32
1530         unsigned long top = HYPERVISOR_VIRT_START;
1531         struct xen_platform_parameters pp;
1532 
1533         if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1534                 top = pp.virt_start;
1535 
1536         reserve_top_address(-top);
1537 #endif  /* CONFIG_X86_32 */
1538 }
1539 
1540 /*
1541  * Like __va(), but returns address in the kernel mapping (which is
1542  * all we have until the physical memory mapping has been set up.
1543  */
1544 static void *__ka(phys_addr_t paddr)
1545 {
1546 #ifdef CONFIG_X86_64
1547         return (void *)(paddr + __START_KERNEL_map);
1548 #else
1549         return __va(paddr);
1550 #endif
1551 }
1552 
1553 /* Convert a machine address to physical address */
1554 static unsigned long m2p(phys_addr_t maddr)
1555 {
1556         phys_addr_t paddr;
1557 
1558         maddr &= PTE_PFN_MASK;
1559         paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1560 
1561         return paddr;
1562 }
1563 
1564 /* Convert a machine address to kernel virtual */
1565 static void *m2v(phys_addr_t maddr)
1566 {
1567         return __ka(m2p(maddr));
1568 }
1569 
1570 /* Set the page permissions on an identity-mapped pages */
1571 static void set_page_prot(void *addr, pgprot_t prot)
1572 {
1573         unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1574         pte_t pte = pfn_pte(pfn, prot);
1575 
1576         if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0))
1577                 BUG();
1578 }
1579 
1580 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1581 {
1582         unsigned pmdidx, pteidx;
1583         unsigned ident_pte;
1584         unsigned long pfn;
1585 
1586         level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1587                                       PAGE_SIZE);
1588 
1589         ident_pte = 0;
1590         pfn = 0;
1591         for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1592                 pte_t *pte_page;
1593 
1594                 /* Reuse or allocate a page of ptes */
1595                 if (pmd_present(pmd[pmdidx]))
1596                         pte_page = m2v(pmd[pmdidx].pmd);
1597                 else {
1598                         /* Check for free pte pages */
1599                         if (ident_pte == LEVEL1_IDENT_ENTRIES)
1600                                 break;
1601 
1602                         pte_page = &level1_ident_pgt[ident_pte];
1603                         ident_pte += PTRS_PER_PTE;
1604 
1605                         pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1606                 }
1607 
1608                 /* Install mappings */
1609                 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1610                         pte_t pte;
1611 
1612 #ifdef CONFIG_X86_32
1613                         if (pfn > max_pfn_mapped)
1614                                 max_pfn_mapped = pfn;
1615 #endif
1616 
1617                         if (!pte_none(pte_page[pteidx]))
1618                                 continue;
1619 
1620                         pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1621                         pte_page[pteidx] = pte;
1622                 }
1623         }
1624 
1625         for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1626                 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1627 
1628         set_page_prot(pmd, PAGE_KERNEL_RO);
1629 }
1630 
1631 void __init xen_setup_machphys_mapping(void)
1632 {
1633         struct xen_machphys_mapping mapping;
1634 
1635         if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1636                 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1637                 machine_to_phys_nr = mapping.max_mfn + 1;
1638         } else {
1639                 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1640         }
1641 #ifdef CONFIG_X86_32
1642         if ((machine_to_phys_mapping + machine_to_phys_nr)
1643             < machine_to_phys_mapping)
1644                 machine_to_phys_nr = (unsigned long *)NULL
1645                                      - machine_to_phys_mapping;
1646 #endif
1647 }
1648 
1649 #ifdef CONFIG_X86_64
1650 static void convert_pfn_mfn(void *v)
1651 {
1652         pte_t *pte = v;
1653         int i;
1654 
1655         /* All levels are converted the same way, so just treat them
1656            as ptes. */
1657         for (i = 0; i < PTRS_PER_PTE; i++)
1658                 pte[i] = xen_make_pte(pte[i].pte);
1659 }
1660 
1661 /*
1662  * Set up the initial kernel pagetable.
1663  *
1664  * We can construct this by grafting the Xen provided pagetable into
1665  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1666  * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt.  This
1667  * means that only the kernel has a physical mapping to start with -
1668  * but that's enough to get __va working.  We need to fill in the rest
1669  * of the physical mapping once some sort of allocator has been set
1670  * up.
1671  */
1672 pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1673                                          unsigned long max_pfn)
1674 {
1675         pud_t *l3;
1676         pmd_t *l2;
1677 
1678         /* max_pfn_mapped is the last pfn mapped in the initial memory
1679          * mappings. Considering that on Xen after the kernel mappings we
1680          * have the mappings of some pages that don't exist in pfn space, we
1681          * set max_pfn_mapped to the last real pfn mapped. */
1682         max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1683 
1684         /* Zap identity mapping */
1685         init_level4_pgt[0] = __pgd(0);
1686 
1687         /* Pre-constructed entries are in pfn, so convert to mfn */
1688         convert_pfn_mfn(init_level4_pgt);
1689         convert_pfn_mfn(level3_ident_pgt);
1690         convert_pfn_mfn(level3_kernel_pgt);
1691 
1692         l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1693         l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1694 
1695         memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1696         memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1697 
1698         l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd);
1699         l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud);
1700         memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD);
1701 
1702         /* Set up identity map */
1703         xen_map_identity_early(level2_ident_pgt, max_pfn);
1704 
1705         /* Make pagetable pieces RO */
1706         set_page_prot(init_level4_pgt, PAGE_KERNEL_RO);
1707         set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1708         set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1709         set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1710         set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1711         set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1712 
1713         /* Pin down new L4 */
1714         pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1715                           PFN_DOWN(__pa_symbol(init_level4_pgt)));
1716 
1717         /* Unpin Xen-provided one */
1718         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1719 
1720         /* Switch over */
1721         pgd = init_level4_pgt;
1722 
1723         /*
1724          * At this stage there can be no user pgd, and no page
1725          * structure to attach it to, so make sure we just set kernel
1726          * pgd.
1727          */
1728         xen_mc_batch();
1729         __xen_write_cr3(true, __pa(pgd));
1730         xen_mc_issue(PARAVIRT_LAZY_CPU);
1731 
1732         memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1733                       __pa(xen_start_info->pt_base +
1734                            xen_start_info->nr_pt_frames * PAGE_SIZE),
1735                       "XEN PAGETABLES");
1736 
1737         return pgd;
1738 }
1739 #else   /* !CONFIG_X86_64 */
1740 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
1741 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
1742 
1743 static void __init xen_write_cr3_init(unsigned long cr3)
1744 {
1745         unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
1746 
1747         BUG_ON(read_cr3() != __pa(initial_page_table));
1748         BUG_ON(cr3 != __pa(swapper_pg_dir));
1749 
1750         /*
1751          * We are switching to swapper_pg_dir for the first time (from
1752          * initial_page_table) and therefore need to mark that page
1753          * read-only and then pin it.
1754          *
1755          * Xen disallows sharing of kernel PMDs for PAE
1756          * guests. Therefore we must copy the kernel PMD from
1757          * initial_page_table into a new kernel PMD to be used in
1758          * swapper_pg_dir.
1759          */
1760         swapper_kernel_pmd =
1761                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1762         memcpy(swapper_kernel_pmd, initial_kernel_pmd,
1763                sizeof(pmd_t) * PTRS_PER_PMD);
1764         swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
1765                 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
1766         set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
1767 
1768         set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
1769         xen_write_cr3(cr3);
1770         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
1771 
1772         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
1773                           PFN_DOWN(__pa(initial_page_table)));
1774         set_page_prot(initial_page_table, PAGE_KERNEL);
1775         set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
1776 
1777         pv_mmu_ops.write_cr3 = &xen_write_cr3;
1778 }
1779 
1780 pgd_t * __init xen_setup_kernel_pagetable(pgd_t *pgd,
1781                                          unsigned long max_pfn)
1782 {
1783         pmd_t *kernel_pmd;
1784 
1785         initial_kernel_pmd =
1786                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
1787 
1788         max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) +
1789                                   xen_start_info->nr_pt_frames * PAGE_SIZE +
1790                                   512*1024);
1791 
1792         kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
1793         memcpy(initial_kernel_pmd, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD);
1794 
1795         xen_map_identity_early(initial_kernel_pmd, max_pfn);
1796 
1797         memcpy(initial_page_table, pgd, sizeof(pgd_t) * PTRS_PER_PGD);
1798         initial_page_table[KERNEL_PGD_BOUNDARY] =
1799                 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
1800 
1801         set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
1802         set_page_prot(initial_page_table, PAGE_KERNEL_RO);
1803         set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
1804 
1805         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1806 
1807         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1808                           PFN_DOWN(__pa(initial_page_table)));
1809         xen_write_cr3(__pa(initial_page_table));
1810 
1811         memblock_x86_reserve_range(__pa(xen_start_info->pt_base),
1812                       __pa(xen_start_info->pt_base +
1813                            xen_start_info->nr_pt_frames * PAGE_SIZE),
1814                       "XEN PAGETABLES");
1815 
1816         return initial_page_table;
1817 }
1818 #endif  /* CONFIG_X86_64 */
1819 
1820 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
1821 
1822 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
1823 {
1824         pte_t pte;
1825 
1826         phys >>= PAGE_SHIFT;
1827 
1828         switch (idx) {
1829         case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
1830 #ifdef CONFIG_X86_F00F_BUG
1831         case FIX_F00F_IDT:
1832 #endif
1833 #ifdef CONFIG_X86_32
1834         case FIX_WP_TEST:
1835         case FIX_VDSO:
1836 # ifdef CONFIG_HIGHMEM
1837         case FIX_KMAP_BEGIN ... FIX_KMAP_END:
1838 # endif
1839 #else
1840         case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE:
1841 #endif
1842         case FIX_TEXT_POKE0:
1843         case FIX_TEXT_POKE1:
1844                 /* All local page mappings */
1845                 pte = pfn_pte(phys, prot);
1846                 break;
1847 
1848 #ifdef CONFIG_X86_LOCAL_APIC
1849         case FIX_APIC_BASE:     /* maps dummy local APIC */
1850                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1851                 break;
1852 #endif
1853 
1854 #ifdef CONFIG_X86_IO_APIC
1855         case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
1856                 /*
1857                  * We just don't map the IO APIC - all access is via
1858                  * hypercalls.  Keep the address in the pte for reference.
1859                  */
1860                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
1861                 break;
1862 #endif
1863 
1864         case FIX_PARAVIRT_BOOTMAP:
1865                 /* This is an MFN, but it isn't an IO mapping from the
1866                    IO domain */
1867                 pte = mfn_pte(phys, prot);
1868                 break;
1869 
1870         default:
1871                 /* By default, set_fixmap is used for hardware mappings */
1872                 pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP));
1873                 break;
1874         }
1875 
1876         __native_set_fixmap(idx, pte);
1877 
1878 #ifdef CONFIG_X86_64
1879         /* Replicate changes to map the vsyscall page into the user
1880            pagetable vsyscall mapping. */
1881         if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) {
1882                 unsigned long vaddr = __fix_to_virt(idx);
1883                 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
1884         }
1885 #endif
1886 }
1887 
1888 void __init xen_ident_map_ISA(void)
1889 {
1890         unsigned long pa;
1891 
1892         /*
1893          * If we're dom0, then linear map the ISA machine addresses into
1894          * the kernel's address space.
1895          */
1896         if (!xen_initial_domain())
1897                 return;
1898 
1899         xen_raw_printk("Xen: setup ISA identity maps\n");
1900 
1901         for (pa = ISA_START_ADDRESS; pa < ISA_END_ADDRESS; pa += PAGE_SIZE) {
1902                 pte_t pte = mfn_pte(PFN_DOWN(pa), PAGE_KERNEL_IO);
1903 
1904                 if (HYPERVISOR_update_va_mapping(PAGE_OFFSET + pa, pte, 0))
1905                         BUG();
1906         }
1907 
1908         xen_flush_tlb();
1909 }
1910 
1911 static void __init xen_post_allocator_init(void)
1912 {
1913 #ifdef CONFIG_XEN_DEBUG
1914         pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte_debug);
1915 #endif
1916         pv_mmu_ops.set_pte = xen_set_pte;
1917         pv_mmu_ops.set_pmd = xen_set_pmd;
1918         pv_mmu_ops.set_pud = xen_set_pud;
1919 #if PAGETABLE_LEVELS == 4
1920         pv_mmu_ops.set_pgd = xen_set_pgd;
1921 #endif
1922 
1923         /* This will work as long as patching hasn't happened yet
1924            (which it hasn't) */
1925         pv_mmu_ops.alloc_pte = xen_alloc_pte;
1926         pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
1927         pv_mmu_ops.release_pte = xen_release_pte;
1928         pv_mmu_ops.release_pmd = xen_release_pmd;
1929 #if PAGETABLE_LEVELS == 4
1930         pv_mmu_ops.alloc_pud = xen_alloc_pud;
1931         pv_mmu_ops.release_pud = xen_release_pud;
1932 #endif
1933 
1934 #ifdef CONFIG_X86_64
1935         SetPagePinned(virt_to_page(level3_user_vsyscall));
1936 #endif
1937         xen_mark_init_mm_pinned();
1938 }
1939 
1940 static void xen_leave_lazy_mmu(void)
1941 {
1942         preempt_disable();
1943         xen_mc_flush();
1944         paravirt_leave_lazy_mmu();
1945         preempt_enable();
1946 }
1947 
1948 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
1949         .read_cr2 = xen_read_cr2,
1950         .write_cr2 = xen_write_cr2,
1951 
1952         .read_cr3 = xen_read_cr3,
1953 #ifdef CONFIG_X86_32
1954         .write_cr3 = xen_write_cr3_init,
1955 #else
1956         .write_cr3 = xen_write_cr3,
1957 #endif
1958 
1959         .flush_tlb_user = xen_flush_tlb,
1960         .flush_tlb_kernel = xen_flush_tlb,
1961         .flush_tlb_single = xen_flush_tlb_single,
1962         .flush_tlb_others = xen_flush_tlb_others,
1963 
1964         .pte_update = paravirt_nop,
1965         .pte_update_defer = paravirt_nop,
1966 
1967         .pgd_alloc = xen_pgd_alloc,
1968         .pgd_free = xen_pgd_free,
1969 
1970         .alloc_pte = xen_alloc_pte_init,
1971         .release_pte = xen_release_pte_init,
1972         .alloc_pmd = xen_alloc_pmd_init,
1973         .release_pmd = xen_release_pmd_init,
1974 
1975         .set_pte = xen_set_pte_init,
1976         .set_pte_at = xen_set_pte_at,
1977         .set_pmd = xen_set_pmd_hyper,
1978 
1979         .ptep_modify_prot_start = __ptep_modify_prot_start,
1980         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
1981 
1982         .pte_val = PV_CALLEE_SAVE(xen_pte_val),
1983         .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
1984 
1985         .make_pte = PV_CALLEE_SAVE(xen_make_pte),
1986         .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
1987 
1988 #ifdef CONFIG_X86_PAE
1989         .set_pte_atomic = xen_set_pte_atomic,
1990         .pte_clear = xen_pte_clear,
1991         .pmd_clear = xen_pmd_clear,
1992 #endif  /* CONFIG_X86_PAE */
1993         .set_pud = xen_set_pud_hyper,
1994 
1995         .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
1996         .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
1997 
1998 #if PAGETABLE_LEVELS == 4
1999         .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2000         .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2001         .set_pgd = xen_set_pgd_hyper,
2002 
2003         .alloc_pud = xen_alloc_pmd_init,
2004         .release_pud = xen_release_pmd_init,
2005 #endif  /* PAGETABLE_LEVELS == 4 */
2006 
2007         .activate_mm = xen_activate_mm,
2008         .dup_mmap = xen_dup_mmap,
2009         .exit_mmap = xen_exit_mmap,
2010 
2011         .lazy_mode = {
2012                 .enter = paravirt_enter_lazy_mmu,
2013                 .leave = xen_leave_lazy_mmu,
2014                 .flush = paravirt_flush_lazy_mmu,
2015         },
2016 
2017         .set_fixmap = xen_set_fixmap,
2018 };
2019 
2020 void __init xen_init_mmu_ops(void)
2021 {
2022         x86_init.mapping.pagetable_reserve = xen_mapping_pagetable_reserve;
2023         x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start;
2024         x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done;
2025         pv_mmu_ops = xen_mmu_ops;
2026 
2027         memset(dummy_mapping, 0xff, PAGE_SIZE);
2028 }
2029 
2030 /* Protected by xen_reservation_lock. */
2031 #define MAX_CONTIG_ORDER 9 /* 2MB */
2032 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2033 
2034 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2035 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2036                                 unsigned long *in_frames,
2037                                 unsigned long *out_frames)
2038 {
2039         int i;
2040         struct multicall_space mcs;
2041 
2042         xen_mc_batch();
2043         for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2044                 mcs = __xen_mc_entry(0);
2045 
2046                 if (in_frames)
2047                         in_frames[i] = virt_to_mfn(vaddr);
2048 
2049                 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2050                 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2051 
2052                 if (out_frames)
2053                         out_frames[i] = virt_to_pfn(vaddr);
2054         }
2055         xen_mc_issue(0);
2056 }
2057 
2058 /*
2059  * Update the pfn-to-mfn mappings for a virtual address range, either to
2060  * point to an array of mfns, or contiguously from a single starting
2061  * mfn.
2062  */
2063 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2064                                      unsigned long *mfns,
2065                                      unsigned long first_mfn)
2066 {
2067         unsigned i, limit;
2068         unsigned long mfn;
2069 
2070         xen_mc_batch();
2071 
2072         limit = 1u << order;
2073         for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2074                 struct multicall_space mcs;
2075                 unsigned flags;
2076 
2077                 mcs = __xen_mc_entry(0);
2078                 if (mfns)
2079                         mfn = mfns[i];
2080                 else
2081                         mfn = first_mfn + i;
2082 
2083                 if (i < (limit - 1))
2084                         flags = 0;
2085                 else {
2086                         if (order == 0)
2087                                 flags = UVMF_INVLPG | UVMF_ALL;
2088                         else
2089                                 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2090                 }
2091 
2092                 MULTI_update_va_mapping(mcs.mc, vaddr,
2093                                 mfn_pte(mfn, PAGE_KERNEL), flags);
2094 
2095                 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2096         }
2097 
2098         xen_mc_issue(0);
2099 }
2100 
2101 /*
2102  * Perform the hypercall to exchange a region of our pfns to point to
2103  * memory with the required contiguous alignment.  Takes the pfns as
2104  * input, and populates mfns as output.
2105  *
2106  * Returns a success code indicating whether the hypervisor was able to
2107  * satisfy the request or not.
2108  */
2109 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2110                                unsigned long *pfns_in,
2111                                unsigned long extents_out,
2112                                unsigned int order_out,
2113                                unsigned long *mfns_out,
2114                                unsigned int address_bits)
2115 {
2116         long rc;
2117         int success;
2118 
2119         struct xen_memory_exchange exchange = {
2120                 .in = {
2121                         .nr_extents   = extents_in,
2122                         .extent_order = order_in,
2123                         .extent_start = pfns_in,
2124                         .domid        = DOMID_SELF
2125                 },
2126                 .out = {
2127                         .nr_extents   = extents_out,
2128                         .extent_order = order_out,
2129                         .extent_start = mfns_out,
2130                         .address_bits = address_bits,
2131                         .domid        = DOMID_SELF
2132                 }
2133         };
2134 
2135         BUG_ON(extents_in << order_in != extents_out << order_out);
2136 
2137         rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2138         success = (exchange.nr_exchanged == extents_in);
2139 
2140         BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2141         BUG_ON(success && (rc != 0));
2142 
2143         return success;
2144 }
2145 
2146 int xen_create_contiguous_region(unsigned long vstart, unsigned int order,
2147                                  unsigned int address_bits)
2148 {
2149         unsigned long *in_frames = discontig_frames, out_frame;
2150         unsigned long  flags;
2151         int            success;
2152 
2153         /*
2154          * Currently an auto-translated guest will not perform I/O, nor will
2155          * it require PAE page directories below 4GB. Therefore any calls to
2156          * this function are redundant and can be ignored.
2157          */
2158 
2159         if (xen_feature(XENFEAT_auto_translated_physmap))
2160                 return 0;
2161 
2162         if (unlikely(order > MAX_CONTIG_ORDER))
2163                 return -ENOMEM;
2164 
2165         memset((void *) vstart, 0, PAGE_SIZE << order);
2166 
2167         spin_lock_irqsave(&xen_reservation_lock, flags);
2168 
2169         /* 1. Zap current PTEs, remembering MFNs. */
2170         xen_zap_pfn_range(vstart, order, in_frames, NULL);
2171 
2172         /* 2. Get a new contiguous memory extent. */
2173         out_frame = virt_to_pfn(vstart);
2174         success = xen_exchange_memory(1UL << order, 0, in_frames,
2175                                       1, order, &out_frame,
2176                                       address_bits);
2177 
2178         /* 3. Map the new extent in place of old pages. */
2179         if (success)
2180                 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2181         else
2182                 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2183 
2184         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2185 
2186         return success ? 0 : -ENOMEM;
2187 }
2188 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2189 
2190 void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order)
2191 {
2192         unsigned long *out_frames = discontig_frames, in_frame;
2193         unsigned long  flags;
2194         int success;
2195 
2196         if (xen_feature(XENFEAT_auto_translated_physmap))
2197                 return;
2198 
2199         if (unlikely(order > MAX_CONTIG_ORDER))
2200                 return;
2201 
2202         memset((void *) vstart, 0, PAGE_SIZE << order);
2203 
2204         spin_lock_irqsave(&xen_reservation_lock, flags);
2205 
2206         /* 1. Find start MFN of contiguous extent. */
2207         in_frame = virt_to_mfn(vstart);
2208 
2209         /* 2. Zap current PTEs. */
2210         xen_zap_pfn_range(vstart, order, NULL, out_frames);
2211 
2212         /* 3. Do the exchange for non-contiguous MFNs. */
2213         success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2214                                         0, out_frames, 0);
2215 
2216         /* 4. Map new pages in place of old pages. */
2217         if (success)
2218                 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2219         else
2220                 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2221 
2222         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2223 }
2224 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2225 
2226 #ifdef CONFIG_XEN_PVHVM
2227 static void xen_hvm_exit_mmap(struct mm_struct *mm)
2228 {
2229         struct xen_hvm_pagetable_dying a;
2230         int rc;
2231 
2232         a.domid = DOMID_SELF;
2233         a.gpa = __pa(mm->pgd);
2234         rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2235         WARN_ON_ONCE(rc < 0);
2236 }
2237 
2238 static int is_pagetable_dying_supported(void)
2239 {
2240         struct xen_hvm_pagetable_dying a;
2241         int rc = 0;
2242 
2243         a.domid = DOMID_SELF;
2244         a.gpa = 0x00;
2245         rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a);
2246         if (rc < 0) {
2247                 printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n");
2248                 return 0;
2249         }
2250         return 1;
2251 }
2252 
2253 void __init xen_hvm_init_mmu_ops(void)
2254 {
2255         if (is_pagetable_dying_supported())
2256                 pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap;
2257 }
2258 #endif
2259 
2260 #define REMAP_BATCH_SIZE 16
2261 
2262 struct remap_data {
2263         unsigned long mfn;
2264         pgprot_t prot;
2265         struct mmu_update *mmu_update;
2266 };
2267 
2268 static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token,
2269                                  unsigned long addr, void *data)
2270 {
2271         struct remap_data *rmd = data;
2272         pte_t pte = pte_mkspecial(pfn_pte(rmd->mfn++, rmd->prot));
2273 
2274         rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2275         rmd->mmu_update->val = pte_val_ma(pte);
2276         rmd->mmu_update++;
2277 
2278         return 0;
2279 }
2280 
2281 int xen_remap_domain_mfn_range(struct vm_area_struct *vma,
2282                                unsigned long addr,
2283                                unsigned long mfn, int nr,
2284                                pgprot_t prot, unsigned domid)
2285 {
2286         struct remap_data rmd;
2287         struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2288         int batch;
2289         unsigned long range;
2290         int err = 0;
2291 
2292         prot = __pgprot(pgprot_val(prot) | _PAGE_IOMAP);
2293 
2294         BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_RESERVED | VM_IO)) ==
2295                                 (VM_PFNMAP | VM_RESERVED | VM_IO)));
2296 
2297         rmd.mfn = mfn;
2298         rmd.prot = prot;
2299 
2300         while (nr) {
2301                 batch = min(REMAP_BATCH_SIZE, nr);
2302                 range = (unsigned long)batch << PAGE_SHIFT;
2303 
2304                 rmd.mmu_update = mmu_update;
2305                 err = apply_to_page_range(vma->vm_mm, addr, range,
2306                                           remap_area_mfn_pte_fn, &rmd);
2307                 if (err)
2308                         goto out;
2309 
2310                 err = -EFAULT;
2311                 if (HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid) < 0)
2312                         goto out;
2313 
2314                 nr -= batch;
2315                 addr += range;
2316         }
2317 
2318         err = 0;
2319 out:
2320 
2321         flush_tlb_all();
2322 
2323         return err;
2324 }
2325 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range);
2326 
2327 #ifdef CONFIG_XEN_DEBUG_FS
2328 static int p2m_dump_open(struct inode *inode, struct file *filp)
2329 {
2330         return single_open(filp, p2m_dump_show, NULL);
2331 }
2332 
2333 static const struct file_operations p2m_dump_fops = {
2334         .open           = p2m_dump_open,
2335         .read           = seq_read,
2336         .llseek         = seq_lseek,
2337         .release        = single_release,
2338 };
2339 #endif /* CONFIG_XEN_DEBUG_FS */
2340 

~ [ 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