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

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

Version: ~ [ linux-4.18 ] ~ [ linux-4.17.14 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.62 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.119 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.147 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.118 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.57 ] ~ [ 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-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~ [ linux-next-20180810 ] ~ [ linux-next-20180813 ] ~
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/mm.h>
 42 #include <linux/highmem.h>
 43 #include <linux/debugfs.h>
 44 #include <linux/bug.h>
 45 #include <linux/vmalloc.h>
 46 #include <linux/export.h>
 47 #include <linux/init.h>
 48 #include <linux/gfp.h>
 49 #include <linux/memblock.h>
 50 #include <linux/seq_file.h>
 51 #include <linux/crash_dump.h>
 52 #ifdef CONFIG_KEXEC_CORE
 53 #include <linux/kexec.h>
 54 #endif
 55 
 56 #include <trace/events/xen.h>
 57 
 58 #include <asm/pgtable.h>
 59 #include <asm/tlbflush.h>
 60 #include <asm/fixmap.h>
 61 #include <asm/mmu_context.h>
 62 #include <asm/setup.h>
 63 #include <asm/paravirt.h>
 64 #include <asm/e820/api.h>
 65 #include <asm/linkage.h>
 66 #include <asm/page.h>
 67 #include <asm/init.h>
 68 #include <asm/pat.h>
 69 #include <asm/smp.h>
 70 
 71 #include <asm/xen/hypercall.h>
 72 #include <asm/xen/hypervisor.h>
 73 
 74 #include <xen/xen.h>
 75 #include <xen/page.h>
 76 #include <xen/interface/xen.h>
 77 #include <xen/interface/hvm/hvm_op.h>
 78 #include <xen/interface/version.h>
 79 #include <xen/interface/memory.h>
 80 #include <xen/hvc-console.h>
 81 
 82 #include "multicalls.h"
 83 #include "mmu.h"
 84 #include "debugfs.h"
 85 
 86 #ifdef CONFIG_X86_32
 87 /*
 88  * Identity map, in addition to plain kernel map.  This needs to be
 89  * large enough to allocate page table pages to allocate the rest.
 90  * Each page can map 2MB.
 91  */
 92 #define LEVEL1_IDENT_ENTRIES    (PTRS_PER_PTE * 4)
 93 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
 94 #endif
 95 #ifdef CONFIG_X86_64
 96 /* l3 pud for userspace vsyscall mapping */
 97 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
 98 #endif /* CONFIG_X86_64 */
 99 
100 /*
101  * Note about cr3 (pagetable base) values:
102  *
103  * xen_cr3 contains the current logical cr3 value; it contains the
104  * last set cr3.  This may not be the current effective cr3, because
105  * its update may be being lazily deferred.  However, a vcpu looking
106  * at its own cr3 can use this value knowing that it everything will
107  * be self-consistent.
108  *
109  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
110  * hypercall to set the vcpu cr3 is complete (so it may be a little
111  * out of date, but it will never be set early).  If one vcpu is
112  * looking at another vcpu's cr3 value, it should use this variable.
113  */
114 DEFINE_PER_CPU(unsigned long, xen_cr3);  /* cr3 stored as physaddr */
115 DEFINE_PER_CPU(unsigned long, xen_current_cr3);  /* actual vcpu cr3 */
116 
117 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
118 
119 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
120 
121 /*
122  * Just beyond the highest usermode address.  STACK_TOP_MAX has a
123  * redzone above it, so round it up to a PGD boundary.
124  */
125 #define USER_LIMIT      ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
126 
127 void make_lowmem_page_readonly(void *vaddr)
128 {
129         pte_t *pte, ptev;
130         unsigned long address = (unsigned long)vaddr;
131         unsigned int level;
132 
133         pte = lookup_address(address, &level);
134         if (pte == NULL)
135                 return;         /* vaddr missing */
136 
137         ptev = pte_wrprotect(*pte);
138 
139         if (HYPERVISOR_update_va_mapping(address, ptev, 0))
140                 BUG();
141 }
142 
143 void make_lowmem_page_readwrite(void *vaddr)
144 {
145         pte_t *pte, ptev;
146         unsigned long address = (unsigned long)vaddr;
147         unsigned int level;
148 
149         pte = lookup_address(address, &level);
150         if (pte == NULL)
151                 return;         /* vaddr missing */
152 
153         ptev = pte_mkwrite(*pte);
154 
155         if (HYPERVISOR_update_va_mapping(address, ptev, 0))
156                 BUG();
157 }
158 
159 
160 /*
161  * During early boot all page table pages are pinned, but we do not have struct
162  * pages, so return true until struct pages are ready.
163  */
164 static bool xen_page_pinned(void *ptr)
165 {
166         if (static_branch_likely(&xen_struct_pages_ready)) {
167                 struct page *page = virt_to_page(ptr);
168 
169                 return PagePinned(page);
170         }
171         return true;
172 }
173 
174 static void xen_extend_mmu_update(const struct mmu_update *update)
175 {
176         struct multicall_space mcs;
177         struct mmu_update *u;
178 
179         mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
180 
181         if (mcs.mc != NULL) {
182                 mcs.mc->args[1]++;
183         } else {
184                 mcs = __xen_mc_entry(sizeof(*u));
185                 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
186         }
187 
188         u = mcs.args;
189         *u = *update;
190 }
191 
192 static void xen_extend_mmuext_op(const struct mmuext_op *op)
193 {
194         struct multicall_space mcs;
195         struct mmuext_op *u;
196 
197         mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
198 
199         if (mcs.mc != NULL) {
200                 mcs.mc->args[1]++;
201         } else {
202                 mcs = __xen_mc_entry(sizeof(*u));
203                 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
204         }
205 
206         u = mcs.args;
207         *u = *op;
208 }
209 
210 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
211 {
212         struct mmu_update u;
213 
214         preempt_disable();
215 
216         xen_mc_batch();
217 
218         /* ptr may be ioremapped for 64-bit pagetable setup */
219         u.ptr = arbitrary_virt_to_machine(ptr).maddr;
220         u.val = pmd_val_ma(val);
221         xen_extend_mmu_update(&u);
222 
223         xen_mc_issue(PARAVIRT_LAZY_MMU);
224 
225         preempt_enable();
226 }
227 
228 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
229 {
230         trace_xen_mmu_set_pmd(ptr, val);
231 
232         /* If page is not pinned, we can just update the entry
233            directly */
234         if (!xen_page_pinned(ptr)) {
235                 *ptr = val;
236                 return;
237         }
238 
239         xen_set_pmd_hyper(ptr, val);
240 }
241 
242 /*
243  * Associate a virtual page frame with a given physical page frame
244  * and protection flags for that frame.
245  */
246 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
247 {
248         set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
249 }
250 
251 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
252 {
253         struct mmu_update u;
254 
255         if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
256                 return false;
257 
258         xen_mc_batch();
259 
260         u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
261         u.val = pte_val_ma(pteval);
262         xen_extend_mmu_update(&u);
263 
264         xen_mc_issue(PARAVIRT_LAZY_MMU);
265 
266         return true;
267 }
268 
269 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
270 {
271         if (!xen_batched_set_pte(ptep, pteval)) {
272                 /*
273                  * Could call native_set_pte() here and trap and
274                  * emulate the PTE write but with 32-bit guests this
275                  * needs two traps (one for each of the two 32-bit
276                  * words in the PTE) so do one hypercall directly
277                  * instead.
278                  */
279                 struct mmu_update u;
280 
281                 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
282                 u.val = pte_val_ma(pteval);
283                 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
284         }
285 }
286 
287 static void xen_set_pte(pte_t *ptep, pte_t pteval)
288 {
289         trace_xen_mmu_set_pte(ptep, pteval);
290         __xen_set_pte(ptep, pteval);
291 }
292 
293 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
294                     pte_t *ptep, pte_t pteval)
295 {
296         trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
297         __xen_set_pte(ptep, pteval);
298 }
299 
300 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
301                                  unsigned long addr, pte_t *ptep)
302 {
303         /* Just return the pte as-is.  We preserve the bits on commit */
304         trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
305         return *ptep;
306 }
307 
308 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
309                                  pte_t *ptep, pte_t pte)
310 {
311         struct mmu_update u;
312 
313         trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
314         xen_mc_batch();
315 
316         u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
317         u.val = pte_val_ma(pte);
318         xen_extend_mmu_update(&u);
319 
320         xen_mc_issue(PARAVIRT_LAZY_MMU);
321 }
322 
323 /* Assume pteval_t is equivalent to all the other *val_t types. */
324 static pteval_t pte_mfn_to_pfn(pteval_t val)
325 {
326         if (val & _PAGE_PRESENT) {
327                 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
328                 unsigned long pfn = mfn_to_pfn(mfn);
329 
330                 pteval_t flags = val & PTE_FLAGS_MASK;
331                 if (unlikely(pfn == ~0))
332                         val = flags & ~_PAGE_PRESENT;
333                 else
334                         val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
335         }
336 
337         return val;
338 }
339 
340 static pteval_t pte_pfn_to_mfn(pteval_t val)
341 {
342         if (val & _PAGE_PRESENT) {
343                 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
344                 pteval_t flags = val & PTE_FLAGS_MASK;
345                 unsigned long mfn;
346 
347                 mfn = __pfn_to_mfn(pfn);
348 
349                 /*
350                  * If there's no mfn for the pfn, then just create an
351                  * empty non-present pte.  Unfortunately this loses
352                  * information about the original pfn, so
353                  * pte_mfn_to_pfn is asymmetric.
354                  */
355                 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
356                         mfn = 0;
357                         flags = 0;
358                 } else
359                         mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
360                 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
361         }
362 
363         return val;
364 }
365 
366 __visible pteval_t xen_pte_val(pte_t pte)
367 {
368         pteval_t pteval = pte.pte;
369 
370         return pte_mfn_to_pfn(pteval);
371 }
372 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
373 
374 __visible pgdval_t xen_pgd_val(pgd_t pgd)
375 {
376         return pte_mfn_to_pfn(pgd.pgd);
377 }
378 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
379 
380 __visible pte_t xen_make_pte(pteval_t pte)
381 {
382         pte = pte_pfn_to_mfn(pte);
383 
384         return native_make_pte(pte);
385 }
386 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
387 
388 __visible pgd_t xen_make_pgd(pgdval_t pgd)
389 {
390         pgd = pte_pfn_to_mfn(pgd);
391         return native_make_pgd(pgd);
392 }
393 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
394 
395 __visible pmdval_t xen_pmd_val(pmd_t pmd)
396 {
397         return pte_mfn_to_pfn(pmd.pmd);
398 }
399 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
400 
401 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
402 {
403         struct mmu_update u;
404 
405         preempt_disable();
406 
407         xen_mc_batch();
408 
409         /* ptr may be ioremapped for 64-bit pagetable setup */
410         u.ptr = arbitrary_virt_to_machine(ptr).maddr;
411         u.val = pud_val_ma(val);
412         xen_extend_mmu_update(&u);
413 
414         xen_mc_issue(PARAVIRT_LAZY_MMU);
415 
416         preempt_enable();
417 }
418 
419 static void xen_set_pud(pud_t *ptr, pud_t val)
420 {
421         trace_xen_mmu_set_pud(ptr, val);
422 
423         /* If page is not pinned, we can just update the entry
424            directly */
425         if (!xen_page_pinned(ptr)) {
426                 *ptr = val;
427                 return;
428         }
429 
430         xen_set_pud_hyper(ptr, val);
431 }
432 
433 #ifdef CONFIG_X86_PAE
434 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
435 {
436         trace_xen_mmu_set_pte_atomic(ptep, pte);
437         set_64bit((u64 *)ptep, native_pte_val(pte));
438 }
439 
440 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
441 {
442         trace_xen_mmu_pte_clear(mm, addr, ptep);
443         if (!xen_batched_set_pte(ptep, native_make_pte(0)))
444                 native_pte_clear(mm, addr, ptep);
445 }
446 
447 static void xen_pmd_clear(pmd_t *pmdp)
448 {
449         trace_xen_mmu_pmd_clear(pmdp);
450         set_pmd(pmdp, __pmd(0));
451 }
452 #endif  /* CONFIG_X86_PAE */
453 
454 __visible pmd_t xen_make_pmd(pmdval_t pmd)
455 {
456         pmd = pte_pfn_to_mfn(pmd);
457         return native_make_pmd(pmd);
458 }
459 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
460 
461 #ifdef CONFIG_X86_64
462 __visible pudval_t xen_pud_val(pud_t pud)
463 {
464         return pte_mfn_to_pfn(pud.pud);
465 }
466 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
467 
468 __visible pud_t xen_make_pud(pudval_t pud)
469 {
470         pud = pte_pfn_to_mfn(pud);
471 
472         return native_make_pud(pud);
473 }
474 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
475 
476 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
477 {
478         pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
479         unsigned offset = pgd - pgd_page;
480         pgd_t *user_ptr = NULL;
481 
482         if (offset < pgd_index(USER_LIMIT)) {
483                 struct page *page = virt_to_page(pgd_page);
484                 user_ptr = (pgd_t *)page->private;
485                 if (user_ptr)
486                         user_ptr += offset;
487         }
488 
489         return user_ptr;
490 }
491 
492 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
493 {
494         struct mmu_update u;
495 
496         u.ptr = virt_to_machine(ptr).maddr;
497         u.val = p4d_val_ma(val);
498         xen_extend_mmu_update(&u);
499 }
500 
501 /*
502  * Raw hypercall-based set_p4d, intended for in early boot before
503  * there's a page structure.  This implies:
504  *  1. The only existing pagetable is the kernel's
505  *  2. It is always pinned
506  *  3. It has no user pagetable attached to it
507  */
508 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
509 {
510         preempt_disable();
511 
512         xen_mc_batch();
513 
514         __xen_set_p4d_hyper(ptr, val);
515 
516         xen_mc_issue(PARAVIRT_LAZY_MMU);
517 
518         preempt_enable();
519 }
520 
521 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
522 {
523         pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
524         pgd_t pgd_val;
525 
526         trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
527 
528         /* If page is not pinned, we can just update the entry
529            directly */
530         if (!xen_page_pinned(ptr)) {
531                 *ptr = val;
532                 if (user_ptr) {
533                         WARN_ON(xen_page_pinned(user_ptr));
534                         pgd_val.pgd = p4d_val_ma(val);
535                         *user_ptr = pgd_val;
536                 }
537                 return;
538         }
539 
540         /* If it's pinned, then we can at least batch the kernel and
541            user updates together. */
542         xen_mc_batch();
543 
544         __xen_set_p4d_hyper(ptr, val);
545         if (user_ptr)
546                 __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
547 
548         xen_mc_issue(PARAVIRT_LAZY_MMU);
549 }
550 
551 #if CONFIG_PGTABLE_LEVELS >= 5
552 __visible p4dval_t xen_p4d_val(p4d_t p4d)
553 {
554         return pte_mfn_to_pfn(p4d.p4d);
555 }
556 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
557 
558 __visible p4d_t xen_make_p4d(p4dval_t p4d)
559 {
560         p4d = pte_pfn_to_mfn(p4d);
561 
562         return native_make_p4d(p4d);
563 }
564 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
565 #endif  /* CONFIG_PGTABLE_LEVELS >= 5 */
566 #endif  /* CONFIG_X86_64 */
567 
568 static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
569                 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
570                 bool last, unsigned long limit)
571 {
572         int i, nr, flush = 0;
573 
574         nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
575         for (i = 0; i < nr; i++) {
576                 if (!pmd_none(pmd[i]))
577                         flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE);
578         }
579         return flush;
580 }
581 
582 static int xen_pud_walk(struct mm_struct *mm, pud_t *pud,
583                 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
584                 bool last, unsigned long limit)
585 {
586         int i, nr, flush = 0;
587 
588         nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
589         for (i = 0; i < nr; i++) {
590                 pmd_t *pmd;
591 
592                 if (pud_none(pud[i]))
593                         continue;
594 
595                 pmd = pmd_offset(&pud[i], 0);
596                 if (PTRS_PER_PMD > 1)
597                         flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
598                 flush |= xen_pmd_walk(mm, pmd, func,
599                                 last && i == nr - 1, limit);
600         }
601         return flush;
602 }
603 
604 static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
605                 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
606                 bool last, unsigned long limit)
607 {
608         int flush = 0;
609         pud_t *pud;
610 
611 
612         if (p4d_none(*p4d))
613                 return flush;
614 
615         pud = pud_offset(p4d, 0);
616         if (PTRS_PER_PUD > 1)
617                 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
618         flush |= xen_pud_walk(mm, pud, func, last, limit);
619         return flush;
620 }
621 
622 /*
623  * (Yet another) pagetable walker.  This one is intended for pinning a
624  * pagetable.  This means that it walks a pagetable and calls the
625  * callback function on each page it finds making up the page table,
626  * at every level.  It walks the entire pagetable, but it only bothers
627  * pinning pte pages which are below limit.  In the normal case this
628  * will be STACK_TOP_MAX, but at boot we need to pin up to
629  * FIXADDR_TOP.
630  *
631  * For 32-bit the important bit is that we don't pin beyond there,
632  * because then we start getting into Xen's ptes.
633  *
634  * For 64-bit, we must skip the Xen hole in the middle of the address
635  * space, just after the big x86-64 virtual hole.
636  */
637 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
638                           int (*func)(struct mm_struct *mm, struct page *,
639                                       enum pt_level),
640                           unsigned long limit)
641 {
642         int i, nr, flush = 0;
643         unsigned hole_low, hole_high;
644 
645         /* The limit is the last byte to be touched */
646         limit--;
647         BUG_ON(limit >= FIXADDR_TOP);
648 
649         /*
650          * 64-bit has a great big hole in the middle of the address
651          * space, which contains the Xen mappings.  On 32-bit these
652          * will end up making a zero-sized hole and so is a no-op.
653          */
654         hole_low = pgd_index(USER_LIMIT);
655         hole_high = pgd_index(PAGE_OFFSET);
656 
657         nr = pgd_index(limit) + 1;
658         for (i = 0; i < nr; i++) {
659                 p4d_t *p4d;
660 
661                 if (i >= hole_low && i < hole_high)
662                         continue;
663 
664                 if (pgd_none(pgd[i]))
665                         continue;
666 
667                 p4d = p4d_offset(&pgd[i], 0);
668                 flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
669         }
670 
671         /* Do the top level last, so that the callbacks can use it as
672            a cue to do final things like tlb flushes. */
673         flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
674 
675         return flush;
676 }
677 
678 static int xen_pgd_walk(struct mm_struct *mm,
679                         int (*func)(struct mm_struct *mm, struct page *,
680                                     enum pt_level),
681                         unsigned long limit)
682 {
683         return __xen_pgd_walk(mm, mm->pgd, func, limit);
684 }
685 
686 /* If we're using split pte locks, then take the page's lock and
687    return a pointer to it.  Otherwise return NULL. */
688 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
689 {
690         spinlock_t *ptl = NULL;
691 
692 #if USE_SPLIT_PTE_PTLOCKS
693         ptl = ptlock_ptr(page);
694         spin_lock_nest_lock(ptl, &mm->page_table_lock);
695 #endif
696 
697         return ptl;
698 }
699 
700 static void xen_pte_unlock(void *v)
701 {
702         spinlock_t *ptl = v;
703         spin_unlock(ptl);
704 }
705 
706 static void xen_do_pin(unsigned level, unsigned long pfn)
707 {
708         struct mmuext_op op;
709 
710         op.cmd = level;
711         op.arg1.mfn = pfn_to_mfn(pfn);
712 
713         xen_extend_mmuext_op(&op);
714 }
715 
716 static int xen_pin_page(struct mm_struct *mm, struct page *page,
717                         enum pt_level level)
718 {
719         unsigned pgfl = TestSetPagePinned(page);
720         int flush;
721 
722         if (pgfl)
723                 flush = 0;              /* already pinned */
724         else if (PageHighMem(page))
725                 /* kmaps need flushing if we found an unpinned
726                    highpage */
727                 flush = 1;
728         else {
729                 void *pt = lowmem_page_address(page);
730                 unsigned long pfn = page_to_pfn(page);
731                 struct multicall_space mcs = __xen_mc_entry(0);
732                 spinlock_t *ptl;
733 
734                 flush = 0;
735 
736                 /*
737                  * We need to hold the pagetable lock between the time
738                  * we make the pagetable RO and when we actually pin
739                  * it.  If we don't, then other users may come in and
740                  * attempt to update the pagetable by writing it,
741                  * which will fail because the memory is RO but not
742                  * pinned, so Xen won't do the trap'n'emulate.
743                  *
744                  * If we're using split pte locks, we can't hold the
745                  * entire pagetable's worth of locks during the
746                  * traverse, because we may wrap the preempt count (8
747                  * bits).  The solution is to mark RO and pin each PTE
748                  * page while holding the lock.  This means the number
749                  * of locks we end up holding is never more than a
750                  * batch size (~32 entries, at present).
751                  *
752                  * If we're not using split pte locks, we needn't pin
753                  * the PTE pages independently, because we're
754                  * protected by the overall pagetable lock.
755                  */
756                 ptl = NULL;
757                 if (level == PT_PTE)
758                         ptl = xen_pte_lock(page, mm);
759 
760                 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
761                                         pfn_pte(pfn, PAGE_KERNEL_RO),
762                                         level == PT_PGD ? UVMF_TLB_FLUSH : 0);
763 
764                 if (ptl) {
765                         xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
766 
767                         /* Queue a deferred unlock for when this batch
768                            is completed. */
769                         xen_mc_callback(xen_pte_unlock, ptl);
770                 }
771         }
772 
773         return flush;
774 }
775 
776 /* This is called just after a mm has been created, but it has not
777    been used yet.  We need to make sure that its pagetable is all
778    read-only, and can be pinned. */
779 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
780 {
781         trace_xen_mmu_pgd_pin(mm, pgd);
782 
783         xen_mc_batch();
784 
785         if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
786                 /* re-enable interrupts for flushing */
787                 xen_mc_issue(0);
788 
789                 kmap_flush_unused();
790 
791                 xen_mc_batch();
792         }
793 
794 #ifdef CONFIG_X86_64
795         {
796                 pgd_t *user_pgd = xen_get_user_pgd(pgd);
797 
798                 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
799 
800                 if (user_pgd) {
801                         xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
802                         xen_do_pin(MMUEXT_PIN_L4_TABLE,
803                                    PFN_DOWN(__pa(user_pgd)));
804                 }
805         }
806 #else /* CONFIG_X86_32 */
807 #ifdef CONFIG_X86_PAE
808         /* Need to make sure unshared kernel PMD is pinnable */
809         xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
810                      PT_PMD);
811 #endif
812         xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
813 #endif /* CONFIG_X86_64 */
814         xen_mc_issue(0);
815 }
816 
817 static void xen_pgd_pin(struct mm_struct *mm)
818 {
819         __xen_pgd_pin(mm, mm->pgd);
820 }
821 
822 /*
823  * On save, we need to pin all pagetables to make sure they get their
824  * mfns turned into pfns.  Search the list for any unpinned pgds and pin
825  * them (unpinned pgds are not currently in use, probably because the
826  * process is under construction or destruction).
827  *
828  * Expected to be called in stop_machine() ("equivalent to taking
829  * every spinlock in the system"), so the locking doesn't really
830  * matter all that much.
831  */
832 void xen_mm_pin_all(void)
833 {
834         struct page *page;
835 
836         spin_lock(&pgd_lock);
837 
838         list_for_each_entry(page, &pgd_list, lru) {
839                 if (!PagePinned(page)) {
840                         __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
841                         SetPageSavePinned(page);
842                 }
843         }
844 
845         spin_unlock(&pgd_lock);
846 }
847 
848 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
849                                   enum pt_level level)
850 {
851         SetPagePinned(page);
852         return 0;
853 }
854 
855 /*
856  * The init_mm pagetable is really pinned as soon as its created, but
857  * that's before we have page structures to store the bits.  So do all
858  * the book-keeping now once struct pages for allocated pages are
859  * initialized. This happens only after free_all_bootmem() is called.
860  */
861 static void __init xen_after_bootmem(void)
862 {
863         static_branch_enable(&xen_struct_pages_ready);
864 #ifdef CONFIG_X86_64
865         SetPagePinned(virt_to_page(level3_user_vsyscall));
866 #endif
867         xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
868 }
869 
870 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
871                           enum pt_level level)
872 {
873         unsigned pgfl = TestClearPagePinned(page);
874 
875         if (pgfl && !PageHighMem(page)) {
876                 void *pt = lowmem_page_address(page);
877                 unsigned long pfn = page_to_pfn(page);
878                 spinlock_t *ptl = NULL;
879                 struct multicall_space mcs;
880 
881                 /*
882                  * Do the converse to pin_page.  If we're using split
883                  * pte locks, we must be holding the lock for while
884                  * the pte page is unpinned but still RO to prevent
885                  * concurrent updates from seeing it in this
886                  * partially-pinned state.
887                  */
888                 if (level == PT_PTE) {
889                         ptl = xen_pte_lock(page, mm);
890 
891                         if (ptl)
892                                 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
893                 }
894 
895                 mcs = __xen_mc_entry(0);
896 
897                 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
898                                         pfn_pte(pfn, PAGE_KERNEL),
899                                         level == PT_PGD ? UVMF_TLB_FLUSH : 0);
900 
901                 if (ptl) {
902                         /* unlock when batch completed */
903                         xen_mc_callback(xen_pte_unlock, ptl);
904                 }
905         }
906 
907         return 0;               /* never need to flush on unpin */
908 }
909 
910 /* Release a pagetables pages back as normal RW */
911 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
912 {
913         trace_xen_mmu_pgd_unpin(mm, pgd);
914 
915         xen_mc_batch();
916 
917         xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
918 
919 #ifdef CONFIG_X86_64
920         {
921                 pgd_t *user_pgd = xen_get_user_pgd(pgd);
922 
923                 if (user_pgd) {
924                         xen_do_pin(MMUEXT_UNPIN_TABLE,
925                                    PFN_DOWN(__pa(user_pgd)));
926                         xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
927                 }
928         }
929 #endif
930 
931 #ifdef CONFIG_X86_PAE
932         /* Need to make sure unshared kernel PMD is unpinned */
933         xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
934                        PT_PMD);
935 #endif
936 
937         __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
938 
939         xen_mc_issue(0);
940 }
941 
942 static void xen_pgd_unpin(struct mm_struct *mm)
943 {
944         __xen_pgd_unpin(mm, mm->pgd);
945 }
946 
947 /*
948  * On resume, undo any pinning done at save, so that the rest of the
949  * kernel doesn't see any unexpected pinned pagetables.
950  */
951 void xen_mm_unpin_all(void)
952 {
953         struct page *page;
954 
955         spin_lock(&pgd_lock);
956 
957         list_for_each_entry(page, &pgd_list, lru) {
958                 if (PageSavePinned(page)) {
959                         BUG_ON(!PagePinned(page));
960                         __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
961                         ClearPageSavePinned(page);
962                 }
963         }
964 
965         spin_unlock(&pgd_lock);
966 }
967 
968 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
969 {
970         spin_lock(&next->page_table_lock);
971         xen_pgd_pin(next);
972         spin_unlock(&next->page_table_lock);
973 }
974 
975 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
976 {
977         spin_lock(&mm->page_table_lock);
978         xen_pgd_pin(mm);
979         spin_unlock(&mm->page_table_lock);
980 }
981 
982 static void drop_mm_ref_this_cpu(void *info)
983 {
984         struct mm_struct *mm = info;
985 
986         if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
987                 leave_mm(smp_processor_id());
988 
989         /*
990          * If this cpu still has a stale cr3 reference, then make sure
991          * it has been flushed.
992          */
993         if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
994                 xen_mc_flush();
995 }
996 
997 #ifdef CONFIG_SMP
998 /*
999  * Another cpu may still have their %cr3 pointing at the pagetable, so
1000  * we need to repoint it somewhere else before we can unpin it.
1001  */
1002 static void xen_drop_mm_ref(struct mm_struct *mm)
1003 {
1004         cpumask_var_t mask;
1005         unsigned cpu;
1006 
1007         drop_mm_ref_this_cpu(mm);
1008 
1009         /* Get the "official" set of cpus referring to our pagetable. */
1010         if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1011                 for_each_online_cpu(cpu) {
1012                         if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1013                                 continue;
1014                         smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
1015                 }
1016                 return;
1017         }
1018 
1019         /*
1020          * It's possible that a vcpu may have a stale reference to our
1021          * cr3, because its in lazy mode, and it hasn't yet flushed
1022          * its set of pending hypercalls yet.  In this case, we can
1023          * look at its actual current cr3 value, and force it to flush
1024          * if needed.
1025          */
1026         cpumask_clear(mask);
1027         for_each_online_cpu(cpu) {
1028                 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1029                         cpumask_set_cpu(cpu, mask);
1030         }
1031 
1032         smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
1033         free_cpumask_var(mask);
1034 }
1035 #else
1036 static void xen_drop_mm_ref(struct mm_struct *mm)
1037 {
1038         drop_mm_ref_this_cpu(mm);
1039 }
1040 #endif
1041 
1042 /*
1043  * While a process runs, Xen pins its pagetables, which means that the
1044  * hypervisor forces it to be read-only, and it controls all updates
1045  * to it.  This means that all pagetable updates have to go via the
1046  * hypervisor, which is moderately expensive.
1047  *
1048  * Since we're pulling the pagetable down, we switch to use init_mm,
1049  * unpin old process pagetable and mark it all read-write, which
1050  * allows further operations on it to be simple memory accesses.
1051  *
1052  * The only subtle point is that another CPU may be still using the
1053  * pagetable because of lazy tlb flushing.  This means we need need to
1054  * switch all CPUs off this pagetable before we can unpin it.
1055  */
1056 static void xen_exit_mmap(struct mm_struct *mm)
1057 {
1058         get_cpu();              /* make sure we don't move around */
1059         xen_drop_mm_ref(mm);
1060         put_cpu();
1061 
1062         spin_lock(&mm->page_table_lock);
1063 
1064         /* pgd may not be pinned in the error exit path of execve */
1065         if (xen_page_pinned(mm->pgd))
1066                 xen_pgd_unpin(mm);
1067 
1068         spin_unlock(&mm->page_table_lock);
1069 }
1070 
1071 static void xen_post_allocator_init(void);
1072 
1073 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1074 {
1075         struct mmuext_op op;
1076 
1077         op.cmd = cmd;
1078         op.arg1.mfn = pfn_to_mfn(pfn);
1079         if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1080                 BUG();
1081 }
1082 
1083 #ifdef CONFIG_X86_64
1084 static void __init xen_cleanhighmap(unsigned long vaddr,
1085                                     unsigned long vaddr_end)
1086 {
1087         unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1088         pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1089 
1090         /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1091          * We include the PMD passed in on _both_ boundaries. */
1092         for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1093                         pmd++, vaddr += PMD_SIZE) {
1094                 if (pmd_none(*pmd))
1095                         continue;
1096                 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1097                         set_pmd(pmd, __pmd(0));
1098         }
1099         /* In case we did something silly, we should crash in this function
1100          * instead of somewhere later and be confusing. */
1101         xen_mc_flush();
1102 }
1103 
1104 /*
1105  * Make a page range writeable and free it.
1106  */
1107 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1108 {
1109         void *vaddr = __va(paddr);
1110         void *vaddr_end = vaddr + size;
1111 
1112         for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1113                 make_lowmem_page_readwrite(vaddr);
1114 
1115         memblock_free(paddr, size);
1116 }
1117 
1118 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1119 {
1120         unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1121 
1122         if (unpin)
1123                 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1124         ClearPagePinned(virt_to_page(__va(pa)));
1125         xen_free_ro_pages(pa, PAGE_SIZE);
1126 }
1127 
1128 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1129 {
1130         unsigned long pa;
1131         pte_t *pte_tbl;
1132         int i;
1133 
1134         if (pmd_large(*pmd)) {
1135                 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1136                 xen_free_ro_pages(pa, PMD_SIZE);
1137                 return;
1138         }
1139 
1140         pte_tbl = pte_offset_kernel(pmd, 0);
1141         for (i = 0; i < PTRS_PER_PTE; i++) {
1142                 if (pte_none(pte_tbl[i]))
1143                         continue;
1144                 pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1145                 xen_free_ro_pages(pa, PAGE_SIZE);
1146         }
1147         set_pmd(pmd, __pmd(0));
1148         xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1149 }
1150 
1151 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1152 {
1153         unsigned long pa;
1154         pmd_t *pmd_tbl;
1155         int i;
1156 
1157         if (pud_large(*pud)) {
1158                 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1159                 xen_free_ro_pages(pa, PUD_SIZE);
1160                 return;
1161         }
1162 
1163         pmd_tbl = pmd_offset(pud, 0);
1164         for (i = 0; i < PTRS_PER_PMD; i++) {
1165                 if (pmd_none(pmd_tbl[i]))
1166                         continue;
1167                 xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1168         }
1169         set_pud(pud, __pud(0));
1170         xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1171 }
1172 
1173 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1174 {
1175         unsigned long pa;
1176         pud_t *pud_tbl;
1177         int i;
1178 
1179         if (p4d_large(*p4d)) {
1180                 pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1181                 xen_free_ro_pages(pa, P4D_SIZE);
1182                 return;
1183         }
1184 
1185         pud_tbl = pud_offset(p4d, 0);
1186         for (i = 0; i < PTRS_PER_PUD; i++) {
1187                 if (pud_none(pud_tbl[i]))
1188                         continue;
1189                 xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1190         }
1191         set_p4d(p4d, __p4d(0));
1192         xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1193 }
1194 
1195 /*
1196  * Since it is well isolated we can (and since it is perhaps large we should)
1197  * also free the page tables mapping the initial P->M table.
1198  */
1199 static void __init xen_cleanmfnmap(unsigned long vaddr)
1200 {
1201         pgd_t *pgd;
1202         p4d_t *p4d;
1203         bool unpin;
1204 
1205         unpin = (vaddr == 2 * PGDIR_SIZE);
1206         vaddr &= PMD_MASK;
1207         pgd = pgd_offset_k(vaddr);
1208         p4d = p4d_offset(pgd, 0);
1209         if (!p4d_none(*p4d))
1210                 xen_cleanmfnmap_p4d(p4d, unpin);
1211 }
1212 
1213 static void __init xen_pagetable_p2m_free(void)
1214 {
1215         unsigned long size;
1216         unsigned long addr;
1217 
1218         size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1219 
1220         /* No memory or already called. */
1221         if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1222                 return;
1223 
1224         /* using __ka address and sticking INVALID_P2M_ENTRY! */
1225         memset((void *)xen_start_info->mfn_list, 0xff, size);
1226 
1227         addr = xen_start_info->mfn_list;
1228         /*
1229          * We could be in __ka space.
1230          * We roundup to the PMD, which means that if anybody at this stage is
1231          * using the __ka address of xen_start_info or
1232          * xen_start_info->shared_info they are in going to crash. Fortunatly
1233          * we have already revectored in xen_setup_kernel_pagetable and in
1234          * xen_setup_shared_info.
1235          */
1236         size = roundup(size, PMD_SIZE);
1237 
1238         if (addr >= __START_KERNEL_map) {
1239                 xen_cleanhighmap(addr, addr + size);
1240                 size = PAGE_ALIGN(xen_start_info->nr_pages *
1241                                   sizeof(unsigned long));
1242                 memblock_free(__pa(addr), size);
1243         } else {
1244                 xen_cleanmfnmap(addr);
1245         }
1246 }
1247 
1248 static void __init xen_pagetable_cleanhighmap(void)
1249 {
1250         unsigned long size;
1251         unsigned long addr;
1252 
1253         /* At this stage, cleanup_highmap has already cleaned __ka space
1254          * from _brk_limit way up to the max_pfn_mapped (which is the end of
1255          * the ramdisk). We continue on, erasing PMD entries that point to page
1256          * tables - do note that they are accessible at this stage via __va.
1257          * As Xen is aligning the memory end to a 4MB boundary, for good
1258          * measure we also round up to PMD_SIZE * 2 - which means that if
1259          * anybody is using __ka address to the initial boot-stack - and try
1260          * to use it - they are going to crash. The xen_start_info has been
1261          * taken care of already in xen_setup_kernel_pagetable. */
1262         addr = xen_start_info->pt_base;
1263         size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1264 
1265         xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1266         xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1267 }
1268 #endif
1269 
1270 static void __init xen_pagetable_p2m_setup(void)
1271 {
1272         xen_vmalloc_p2m_tree();
1273 
1274 #ifdef CONFIG_X86_64
1275         xen_pagetable_p2m_free();
1276 
1277         xen_pagetable_cleanhighmap();
1278 #endif
1279         /* And revector! Bye bye old array */
1280         xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1281 }
1282 
1283 static void __init xen_pagetable_init(void)
1284 {
1285         paging_init();
1286         xen_post_allocator_init();
1287 
1288         xen_pagetable_p2m_setup();
1289 
1290         /* Allocate and initialize top and mid mfn levels for p2m structure */
1291         xen_build_mfn_list_list();
1292 
1293         /* Remap memory freed due to conflicts with E820 map */
1294         xen_remap_memory();
1295 
1296         xen_setup_shared_info();
1297 }
1298 static void xen_write_cr2(unsigned long cr2)
1299 {
1300         this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1301 }
1302 
1303 static unsigned long xen_read_cr2(void)
1304 {
1305         return this_cpu_read(xen_vcpu)->arch.cr2;
1306 }
1307 
1308 unsigned long xen_read_cr2_direct(void)
1309 {
1310         return this_cpu_read(xen_vcpu_info.arch.cr2);
1311 }
1312 
1313 static noinline void xen_flush_tlb(void)
1314 {
1315         struct mmuext_op *op;
1316         struct multicall_space mcs;
1317 
1318         preempt_disable();
1319 
1320         mcs = xen_mc_entry(sizeof(*op));
1321 
1322         op = mcs.args;
1323         op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1324         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1325 
1326         xen_mc_issue(PARAVIRT_LAZY_MMU);
1327 
1328         preempt_enable();
1329 }
1330 
1331 static void xen_flush_tlb_one_user(unsigned long addr)
1332 {
1333         struct mmuext_op *op;
1334         struct multicall_space mcs;
1335 
1336         trace_xen_mmu_flush_tlb_one_user(addr);
1337 
1338         preempt_disable();
1339 
1340         mcs = xen_mc_entry(sizeof(*op));
1341         op = mcs.args;
1342         op->cmd = MMUEXT_INVLPG_LOCAL;
1343         op->arg1.linear_addr = addr & PAGE_MASK;
1344         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1345 
1346         xen_mc_issue(PARAVIRT_LAZY_MMU);
1347 
1348         preempt_enable();
1349 }
1350 
1351 static void xen_flush_tlb_others(const struct cpumask *cpus,
1352                                  const struct flush_tlb_info *info)
1353 {
1354         struct {
1355                 struct mmuext_op op;
1356                 DECLARE_BITMAP(mask, NR_CPUS);
1357         } *args;
1358         struct multicall_space mcs;
1359         const size_t mc_entry_size = sizeof(args->op) +
1360                 sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1361 
1362         trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end);
1363 
1364         if (cpumask_empty(cpus))
1365                 return;         /* nothing to do */
1366 
1367         mcs = xen_mc_entry(mc_entry_size);
1368         args = mcs.args;
1369         args->op.arg2.vcpumask = to_cpumask(args->mask);
1370 
1371         /* Remove us, and any offline CPUS. */
1372         cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1373         cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1374 
1375         args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1376         if (info->end != TLB_FLUSH_ALL &&
1377             (info->end - info->start) <= PAGE_SIZE) {
1378                 args->op.cmd = MMUEXT_INVLPG_MULTI;
1379                 args->op.arg1.linear_addr = info->start;
1380         }
1381 
1382         MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1383 
1384         xen_mc_issue(PARAVIRT_LAZY_MMU);
1385 }
1386 
1387 static unsigned long xen_read_cr3(void)
1388 {
1389         return this_cpu_read(xen_cr3);
1390 }
1391 
1392 static void set_current_cr3(void *v)
1393 {
1394         this_cpu_write(xen_current_cr3, (unsigned long)v);
1395 }
1396 
1397 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1398 {
1399         struct mmuext_op op;
1400         unsigned long mfn;
1401 
1402         trace_xen_mmu_write_cr3(kernel, cr3);
1403 
1404         if (cr3)
1405                 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1406         else
1407                 mfn = 0;
1408 
1409         WARN_ON(mfn == 0 && kernel);
1410 
1411         op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1412         op.arg1.mfn = mfn;
1413 
1414         xen_extend_mmuext_op(&op);
1415 
1416         if (kernel) {
1417                 this_cpu_write(xen_cr3, cr3);
1418 
1419                 /* Update xen_current_cr3 once the batch has actually
1420                    been submitted. */
1421                 xen_mc_callback(set_current_cr3, (void *)cr3);
1422         }
1423 }
1424 static void xen_write_cr3(unsigned long cr3)
1425 {
1426         BUG_ON(preemptible());
1427 
1428         xen_mc_batch();  /* disables interrupts */
1429 
1430         /* Update while interrupts are disabled, so its atomic with
1431            respect to ipis */
1432         this_cpu_write(xen_cr3, cr3);
1433 
1434         __xen_write_cr3(true, cr3);
1435 
1436 #ifdef CONFIG_X86_64
1437         {
1438                 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1439                 if (user_pgd)
1440                         __xen_write_cr3(false, __pa(user_pgd));
1441                 else
1442                         __xen_write_cr3(false, 0);
1443         }
1444 #endif
1445 
1446         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1447 }
1448 
1449 #ifdef CONFIG_X86_64
1450 /*
1451  * At the start of the day - when Xen launches a guest, it has already
1452  * built pagetables for the guest. We diligently look over them
1453  * in xen_setup_kernel_pagetable and graft as appropriate them in the
1454  * init_top_pgt and its friends. Then when we are happy we load
1455  * the new init_top_pgt - and continue on.
1456  *
1457  * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1458  * up the rest of the pagetables. When it has completed it loads the cr3.
1459  * N.B. that baremetal would start at 'start_kernel' (and the early
1460  * #PF handler would create bootstrap pagetables) - so we are running
1461  * with the same assumptions as what to do when write_cr3 is executed
1462  * at this point.
1463  *
1464  * Since there are no user-page tables at all, we have two variants
1465  * of xen_write_cr3 - the early bootup (this one), and the late one
1466  * (xen_write_cr3). The reason we have to do that is that in 64-bit
1467  * the Linux kernel and user-space are both in ring 3 while the
1468  * hypervisor is in ring 0.
1469  */
1470 static void __init xen_write_cr3_init(unsigned long cr3)
1471 {
1472         BUG_ON(preemptible());
1473 
1474         xen_mc_batch();  /* disables interrupts */
1475 
1476         /* Update while interrupts are disabled, so its atomic with
1477            respect to ipis */
1478         this_cpu_write(xen_cr3, cr3);
1479 
1480         __xen_write_cr3(true, cr3);
1481 
1482         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1483 }
1484 #endif
1485 
1486 static int xen_pgd_alloc(struct mm_struct *mm)
1487 {
1488         pgd_t *pgd = mm->pgd;
1489         int ret = 0;
1490 
1491         BUG_ON(PagePinned(virt_to_page(pgd)));
1492 
1493 #ifdef CONFIG_X86_64
1494         {
1495                 struct page *page = virt_to_page(pgd);
1496                 pgd_t *user_pgd;
1497 
1498                 BUG_ON(page->private != 0);
1499 
1500                 ret = -ENOMEM;
1501 
1502                 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1503                 page->private = (unsigned long)user_pgd;
1504 
1505                 if (user_pgd != NULL) {
1506 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1507                         user_pgd[pgd_index(VSYSCALL_ADDR)] =
1508                                 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1509 #endif
1510                         ret = 0;
1511                 }
1512 
1513                 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1514         }
1515 #endif
1516         return ret;
1517 }
1518 
1519 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1520 {
1521 #ifdef CONFIG_X86_64
1522         pgd_t *user_pgd = xen_get_user_pgd(pgd);
1523 
1524         if (user_pgd)
1525                 free_page((unsigned long)user_pgd);
1526 #endif
1527 }
1528 
1529 /*
1530  * Init-time set_pte while constructing initial pagetables, which
1531  * doesn't allow RO page table pages to be remapped RW.
1532  *
1533  * If there is no MFN for this PFN then this page is initially
1534  * ballooned out so clear the PTE (as in decrease_reservation() in
1535  * drivers/xen/balloon.c).
1536  *
1537  * Many of these PTE updates are done on unpinned and writable pages
1538  * and doing a hypercall for these is unnecessary and expensive.  At
1539  * this point it is not possible to tell if a page is pinned or not,
1540  * so always write the PTE directly and rely on Xen trapping and
1541  * emulating any updates as necessary.
1542  */
1543 __visible pte_t xen_make_pte_init(pteval_t pte)
1544 {
1545 #ifdef CONFIG_X86_64
1546         unsigned long pfn;
1547 
1548         /*
1549          * Pages belonging to the initial p2m list mapped outside the default
1550          * address range must be mapped read-only. This region contains the
1551          * page tables for mapping the p2m list, too, and page tables MUST be
1552          * mapped read-only.
1553          */
1554         pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1555         if (xen_start_info->mfn_list < __START_KERNEL_map &&
1556             pfn >= xen_start_info->first_p2m_pfn &&
1557             pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1558                 pte &= ~_PAGE_RW;
1559 #endif
1560         pte = pte_pfn_to_mfn(pte);
1561         return native_make_pte(pte);
1562 }
1563 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1564 
1565 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1566 {
1567 #ifdef CONFIG_X86_32
1568         /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1569         if (pte_mfn(pte) != INVALID_P2M_ENTRY
1570             && pte_val_ma(*ptep) & _PAGE_PRESENT)
1571                 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1572                                pte_val_ma(pte));
1573 #endif
1574         native_set_pte(ptep, pte);
1575 }
1576 
1577 /* Early in boot, while setting up the initial pagetable, assume
1578    everything is pinned. */
1579 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1580 {
1581 #ifdef CONFIG_FLATMEM
1582         BUG_ON(mem_map);        /* should only be used early */
1583 #endif
1584         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1585         pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1586 }
1587 
1588 /* Used for pmd and pud */
1589 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1590 {
1591 #ifdef CONFIG_FLATMEM
1592         BUG_ON(mem_map);        /* should only be used early */
1593 #endif
1594         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1595 }
1596 
1597 /* Early release_pte assumes that all pts are pinned, since there's
1598    only init_mm and anything attached to that is pinned. */
1599 static void __init xen_release_pte_init(unsigned long pfn)
1600 {
1601         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1602         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1603 }
1604 
1605 static void __init xen_release_pmd_init(unsigned long pfn)
1606 {
1607         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1608 }
1609 
1610 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1611 {
1612         struct multicall_space mcs;
1613         struct mmuext_op *op;
1614 
1615         mcs = __xen_mc_entry(sizeof(*op));
1616         op = mcs.args;
1617         op->cmd = cmd;
1618         op->arg1.mfn = pfn_to_mfn(pfn);
1619 
1620         MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1621 }
1622 
1623 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1624 {
1625         struct multicall_space mcs;
1626         unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1627 
1628         mcs = __xen_mc_entry(0);
1629         MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1630                                 pfn_pte(pfn, prot), 0);
1631 }
1632 
1633 /* This needs to make sure the new pte page is pinned iff its being
1634    attached to a pinned pagetable. */
1635 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1636                                     unsigned level)
1637 {
1638         bool pinned = xen_page_pinned(mm->pgd);
1639 
1640         trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1641 
1642         if (pinned) {
1643                 struct page *page = pfn_to_page(pfn);
1644 
1645                 if (static_branch_likely(&xen_struct_pages_ready))
1646                         SetPagePinned(page);
1647 
1648                 if (!PageHighMem(page)) {
1649                         xen_mc_batch();
1650 
1651                         __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1652 
1653                         if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1654                                 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1655 
1656                         xen_mc_issue(PARAVIRT_LAZY_MMU);
1657                 } else {
1658                         /* make sure there are no stray mappings of
1659                            this page */
1660                         kmap_flush_unused();
1661                 }
1662         }
1663 }
1664 
1665 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1666 {
1667         xen_alloc_ptpage(mm, pfn, PT_PTE);
1668 }
1669 
1670 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1671 {
1672         xen_alloc_ptpage(mm, pfn, PT_PMD);
1673 }
1674 
1675 /* This should never happen until we're OK to use struct page */
1676 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1677 {
1678         struct page *page = pfn_to_page(pfn);
1679         bool pinned = PagePinned(page);
1680 
1681         trace_xen_mmu_release_ptpage(pfn, level, pinned);
1682 
1683         if (pinned) {
1684                 if (!PageHighMem(page)) {
1685                         xen_mc_batch();
1686 
1687                         if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1688                                 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1689 
1690                         __set_pfn_prot(pfn, PAGE_KERNEL);
1691 
1692                         xen_mc_issue(PARAVIRT_LAZY_MMU);
1693                 }
1694                 ClearPagePinned(page);
1695         }
1696 }
1697 
1698 static void xen_release_pte(unsigned long pfn)
1699 {
1700         xen_release_ptpage(pfn, PT_PTE);
1701 }
1702 
1703 static void xen_release_pmd(unsigned long pfn)
1704 {
1705         xen_release_ptpage(pfn, PT_PMD);
1706 }
1707 
1708 #ifdef CONFIG_X86_64
1709 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1710 {
1711         xen_alloc_ptpage(mm, pfn, PT_PUD);
1712 }
1713 
1714 static void xen_release_pud(unsigned long pfn)
1715 {
1716         xen_release_ptpage(pfn, PT_PUD);
1717 }
1718 #endif
1719 
1720 void __init xen_reserve_top(void)
1721 {
1722 #ifdef CONFIG_X86_32
1723         unsigned long top = HYPERVISOR_VIRT_START;
1724         struct xen_platform_parameters pp;
1725 
1726         if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1727                 top = pp.virt_start;
1728 
1729         reserve_top_address(-top);
1730 #endif  /* CONFIG_X86_32 */
1731 }
1732 
1733 /*
1734  * Like __va(), but returns address in the kernel mapping (which is
1735  * all we have until the physical memory mapping has been set up.
1736  */
1737 static void * __init __ka(phys_addr_t paddr)
1738 {
1739 #ifdef CONFIG_X86_64
1740         return (void *)(paddr + __START_KERNEL_map);
1741 #else
1742         return __va(paddr);
1743 #endif
1744 }
1745 
1746 /* Convert a machine address to physical address */
1747 static unsigned long __init m2p(phys_addr_t maddr)
1748 {
1749         phys_addr_t paddr;
1750 
1751         maddr &= XEN_PTE_MFN_MASK;
1752         paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1753 
1754         return paddr;
1755 }
1756 
1757 /* Convert a machine address to kernel virtual */
1758 static void * __init m2v(phys_addr_t maddr)
1759 {
1760         return __ka(m2p(maddr));
1761 }
1762 
1763 /* Set the page permissions on an identity-mapped pages */
1764 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1765                                        unsigned long flags)
1766 {
1767         unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1768         pte_t pte = pfn_pte(pfn, prot);
1769 
1770         if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1771                 BUG();
1772 }
1773 static void __init set_page_prot(void *addr, pgprot_t prot)
1774 {
1775         return set_page_prot_flags(addr, prot, UVMF_NONE);
1776 }
1777 #ifdef CONFIG_X86_32
1778 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1779 {
1780         unsigned pmdidx, pteidx;
1781         unsigned ident_pte;
1782         unsigned long pfn;
1783 
1784         level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1785                                       PAGE_SIZE);
1786 
1787         ident_pte = 0;
1788         pfn = 0;
1789         for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1790                 pte_t *pte_page;
1791 
1792                 /* Reuse or allocate a page of ptes */
1793                 if (pmd_present(pmd[pmdidx]))
1794                         pte_page = m2v(pmd[pmdidx].pmd);
1795                 else {
1796                         /* Check for free pte pages */
1797                         if (ident_pte == LEVEL1_IDENT_ENTRIES)
1798                                 break;
1799 
1800                         pte_page = &level1_ident_pgt[ident_pte];
1801                         ident_pte += PTRS_PER_PTE;
1802 
1803                         pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1804                 }
1805 
1806                 /* Install mappings */
1807                 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1808                         pte_t pte;
1809 
1810                         if (pfn > max_pfn_mapped)
1811                                 max_pfn_mapped = pfn;
1812 
1813                         if (!pte_none(pte_page[pteidx]))
1814                                 continue;
1815 
1816                         pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1817                         pte_page[pteidx] = pte;
1818                 }
1819         }
1820 
1821         for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1822                 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1823 
1824         set_page_prot(pmd, PAGE_KERNEL_RO);
1825 }
1826 #endif
1827 void __init xen_setup_machphys_mapping(void)
1828 {
1829         struct xen_machphys_mapping mapping;
1830 
1831         if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1832                 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1833                 machine_to_phys_nr = mapping.max_mfn + 1;
1834         } else {
1835                 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1836         }
1837 #ifdef CONFIG_X86_32
1838         WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1839                 < machine_to_phys_mapping);
1840 #endif
1841 }
1842 
1843 #ifdef CONFIG_X86_64
1844 static void __init convert_pfn_mfn(void *v)
1845 {
1846         pte_t *pte = v;
1847         int i;
1848 
1849         /* All levels are converted the same way, so just treat them
1850            as ptes. */
1851         for (i = 0; i < PTRS_PER_PTE; i++)
1852                 pte[i] = xen_make_pte(pte[i].pte);
1853 }
1854 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1855                                  unsigned long addr)
1856 {
1857         if (*pt_base == PFN_DOWN(__pa(addr))) {
1858                 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1859                 clear_page((void *)addr);
1860                 (*pt_base)++;
1861         }
1862         if (*pt_end == PFN_DOWN(__pa(addr))) {
1863                 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1864                 clear_page((void *)addr);
1865                 (*pt_end)--;
1866         }
1867 }
1868 /*
1869  * Set up the initial kernel pagetable.
1870  *
1871  * We can construct this by grafting the Xen provided pagetable into
1872  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1873  * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1874  * kernel has a physical mapping to start with - but that's enough to
1875  * get __va working.  We need to fill in the rest of the physical
1876  * mapping once some sort of allocator has been set up.
1877  */
1878 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1879 {
1880         pud_t *l3;
1881         pmd_t *l2;
1882         unsigned long addr[3];
1883         unsigned long pt_base, pt_end;
1884         unsigned i;
1885 
1886         /* max_pfn_mapped is the last pfn mapped in the initial memory
1887          * mappings. Considering that on Xen after the kernel mappings we
1888          * have the mappings of some pages that don't exist in pfn space, we
1889          * set max_pfn_mapped to the last real pfn mapped. */
1890         if (xen_start_info->mfn_list < __START_KERNEL_map)
1891                 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1892         else
1893                 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1894 
1895         pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1896         pt_end = pt_base + xen_start_info->nr_pt_frames;
1897 
1898         /* Zap identity mapping */
1899         init_top_pgt[0] = __pgd(0);
1900 
1901         /* Pre-constructed entries are in pfn, so convert to mfn */
1902         /* L4[272] -> level3_ident_pgt  */
1903         /* L4[511] -> level3_kernel_pgt */
1904         convert_pfn_mfn(init_top_pgt);
1905 
1906         /* L3_i[0] -> level2_ident_pgt */
1907         convert_pfn_mfn(level3_ident_pgt);
1908         /* L3_k[510] -> level2_kernel_pgt */
1909         /* L3_k[511] -> level2_fixmap_pgt */
1910         convert_pfn_mfn(level3_kernel_pgt);
1911 
1912         /* L3_k[511][506] -> level1_fixmap_pgt */
1913         convert_pfn_mfn(level2_fixmap_pgt);
1914 
1915         /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1916         l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1917         l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1918 
1919         addr[0] = (unsigned long)pgd;
1920         addr[1] = (unsigned long)l3;
1921         addr[2] = (unsigned long)l2;
1922         /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1923          * Both L4[272][0] and L4[511][510] have entries that point to the same
1924          * L2 (PMD) tables. Meaning that if you modify it in __va space
1925          * it will be also modified in the __ka space! (But if you just
1926          * modify the PMD table to point to other PTE's or none, then you
1927          * are OK - which is what cleanup_highmap does) */
1928         copy_page(level2_ident_pgt, l2);
1929         /* Graft it onto L4[511][510] */
1930         copy_page(level2_kernel_pgt, l2);
1931 
1932         /*
1933          * Zap execute permission from the ident map. Due to the sharing of
1934          * L1 entries we need to do this in the L2.
1935          */
1936         if (__supported_pte_mask & _PAGE_NX) {
1937                 for (i = 0; i < PTRS_PER_PMD; ++i) {
1938                         if (pmd_none(level2_ident_pgt[i]))
1939                                 continue;
1940                         level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1941                 }
1942         }
1943 
1944         /* Copy the initial P->M table mappings if necessary. */
1945         i = pgd_index(xen_start_info->mfn_list);
1946         if (i && i < pgd_index(__START_KERNEL_map))
1947                 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1948 
1949         /* Make pagetable pieces RO */
1950         set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1951         set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1952         set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1953         set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1954         set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1955         set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1956         set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1957         set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
1958 
1959         /* Pin down new L4 */
1960         pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1961                           PFN_DOWN(__pa_symbol(init_top_pgt)));
1962 
1963         /* Unpin Xen-provided one */
1964         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1965 
1966         /*
1967          * At this stage there can be no user pgd, and no page structure to
1968          * attach it to, so make sure we just set kernel pgd.
1969          */
1970         xen_mc_batch();
1971         __xen_write_cr3(true, __pa(init_top_pgt));
1972         xen_mc_issue(PARAVIRT_LAZY_CPU);
1973 
1974         /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1975          * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1976          * the initial domain. For guests using the toolstack, they are in:
1977          * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1978          * rip out the [L4] (pgd), but for guests we shave off three pages.
1979          */
1980         for (i = 0; i < ARRAY_SIZE(addr); i++)
1981                 check_pt_base(&pt_base, &pt_end, addr[i]);
1982 
1983         /* Our (by three pages) smaller Xen pagetable that we are using */
1984         xen_pt_base = PFN_PHYS(pt_base);
1985         xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1986         memblock_reserve(xen_pt_base, xen_pt_size);
1987 
1988         /* Revector the xen_start_info */
1989         xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1990 }
1991 
1992 /*
1993  * Read a value from a physical address.
1994  */
1995 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1996 {
1997         unsigned long *vaddr;
1998         unsigned long val;
1999 
2000         vaddr = early_memremap_ro(addr, sizeof(val));
2001         val = *vaddr;
2002         early_memunmap(vaddr, sizeof(val));
2003         return val;
2004 }
2005 
2006 /*
2007  * Translate a virtual address to a physical one without relying on mapped
2008  * page tables. Don't rely on big pages being aligned in (guest) physical
2009  * space!
2010  */
2011 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
2012 {
2013         phys_addr_t pa;
2014         pgd_t pgd;
2015         pud_t pud;
2016         pmd_t pmd;
2017         pte_t pte;
2018 
2019         pa = read_cr3_pa();
2020         pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
2021                                                        sizeof(pgd)));
2022         if (!pgd_present(pgd))
2023                 return 0;
2024 
2025         pa = pgd_val(pgd) & PTE_PFN_MASK;
2026         pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
2027                                                        sizeof(pud)));
2028         if (!pud_present(pud))
2029                 return 0;
2030         pa = pud_val(pud) & PTE_PFN_MASK;
2031         if (pud_large(pud))
2032                 return pa + (vaddr & ~PUD_MASK);
2033 
2034         pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
2035                                                        sizeof(pmd)));
2036         if (!pmd_present(pmd))
2037                 return 0;
2038         pa = pmd_val(pmd) & PTE_PFN_MASK;
2039         if (pmd_large(pmd))
2040                 return pa + (vaddr & ~PMD_MASK);
2041 
2042         pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
2043                                                        sizeof(pte)));
2044         if (!pte_present(pte))
2045                 return 0;
2046         pa = pte_pfn(pte) << PAGE_SHIFT;
2047 
2048         return pa | (vaddr & ~PAGE_MASK);
2049 }
2050 
2051 /*
2052  * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
2053  * this area.
2054  */
2055 void __init xen_relocate_p2m(void)
2056 {
2057         phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
2058         unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
2059         int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
2060         pte_t *pt;
2061         pmd_t *pmd;
2062         pud_t *pud;
2063         pgd_t *pgd;
2064         unsigned long *new_p2m;
2065         int save_pud;
2066 
2067         size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
2068         n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
2069         n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
2070         n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
2071         n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
2072         n_frames = n_pte + n_pt + n_pmd + n_pud;
2073 
2074         new_area = xen_find_free_area(PFN_PHYS(n_frames));
2075         if (!new_area) {
2076                 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
2077                 BUG();
2078         }
2079 
2080         /*
2081          * Setup the page tables for addressing the new p2m list.
2082          * We have asked the hypervisor to map the p2m list at the user address
2083          * PUD_SIZE. It may have done so, or it may have used a kernel space
2084          * address depending on the Xen version.
2085          * To avoid any possible virtual address collision, just use
2086          * 2 * PUD_SIZE for the new area.
2087          */
2088         pud_phys = new_area;
2089         pmd_phys = pud_phys + PFN_PHYS(n_pud);
2090         pt_phys = pmd_phys + PFN_PHYS(n_pmd);
2091         p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
2092 
2093         pgd = __va(read_cr3_pa());
2094         new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
2095         save_pud = n_pud;
2096         for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
2097                 pud = early_memremap(pud_phys, PAGE_SIZE);
2098                 clear_page(pud);
2099                 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
2100                                 idx_pmd++) {
2101                         pmd = early_memremap(pmd_phys, PAGE_SIZE);
2102                         clear_page(pmd);
2103                         for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
2104                                         idx_pt++) {
2105                                 pt = early_memremap(pt_phys, PAGE_SIZE);
2106                                 clear_page(pt);
2107                                 for (idx_pte = 0;
2108                                                 idx_pte < min(n_pte, PTRS_PER_PTE);
2109                                                 idx_pte++) {
2110                                         set_pte(pt + idx_pte,
2111                                                         pfn_pte(p2m_pfn, PAGE_KERNEL));
2112                                         p2m_pfn++;
2113                                 }
2114                                 n_pte -= PTRS_PER_PTE;
2115                                 early_memunmap(pt, PAGE_SIZE);
2116                                 make_lowmem_page_readonly(__va(pt_phys));
2117                                 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2118                                                 PFN_DOWN(pt_phys));
2119                                 set_pmd(pmd + idx_pt,
2120                                                 __pmd(_PAGE_TABLE | pt_phys));
2121                                 pt_phys += PAGE_SIZE;
2122                         }
2123                         n_pt -= PTRS_PER_PMD;
2124                         early_memunmap(pmd, PAGE_SIZE);
2125                         make_lowmem_page_readonly(__va(pmd_phys));
2126                         pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2127                                         PFN_DOWN(pmd_phys));
2128                         set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
2129                         pmd_phys += PAGE_SIZE;
2130                 }
2131                 n_pmd -= PTRS_PER_PUD;
2132                 early_memunmap(pud, PAGE_SIZE);
2133                 make_lowmem_page_readonly(__va(pud_phys));
2134                 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2135                 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2136                 pud_phys += PAGE_SIZE;
2137         }
2138 
2139         /* Now copy the old p2m info to the new area. */
2140         memcpy(new_p2m, xen_p2m_addr, size);
2141         xen_p2m_addr = new_p2m;
2142 
2143         /* Release the old p2m list and set new list info. */
2144         p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2145         BUG_ON(!p2m_pfn);
2146         p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2147 
2148         if (xen_start_info->mfn_list < __START_KERNEL_map) {
2149                 pfn = xen_start_info->first_p2m_pfn;
2150                 pfn_end = xen_start_info->first_p2m_pfn +
2151                           xen_start_info->nr_p2m_frames;
2152                 set_pgd(pgd + 1, __pgd(0));
2153         } else {
2154                 pfn = p2m_pfn;
2155                 pfn_end = p2m_pfn_end;
2156         }
2157 
2158         memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2159         while (pfn < pfn_end) {
2160                 if (pfn == p2m_pfn) {
2161                         pfn = p2m_pfn_end;
2162                         continue;
2163                 }
2164                 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2165                 pfn++;
2166         }
2167 
2168         xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2169         xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
2170         xen_start_info->nr_p2m_frames = n_frames;
2171 }
2172 
2173 #else   /* !CONFIG_X86_64 */
2174 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
2175 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
2176 
2177 static void __init xen_write_cr3_init(unsigned long cr3)
2178 {
2179         unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
2180 
2181         BUG_ON(read_cr3_pa() != __pa(initial_page_table));
2182         BUG_ON(cr3 != __pa(swapper_pg_dir));
2183 
2184         /*
2185          * We are switching to swapper_pg_dir for the first time (from
2186          * initial_page_table) and therefore need to mark that page
2187          * read-only and then pin it.
2188          *
2189          * Xen disallows sharing of kernel PMDs for PAE
2190          * guests. Therefore we must copy the kernel PMD from
2191          * initial_page_table into a new kernel PMD to be used in
2192          * swapper_pg_dir.
2193          */
2194         swapper_kernel_pmd =
2195                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2196         copy_page(swapper_kernel_pmd, initial_kernel_pmd);
2197         swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
2198                 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
2199         set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
2200 
2201         set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
2202         xen_write_cr3(cr3);
2203         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
2204 
2205         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
2206                           PFN_DOWN(__pa(initial_page_table)));
2207         set_page_prot(initial_page_table, PAGE_KERNEL);
2208         set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
2209 
2210         pv_mmu_ops.write_cr3 = &xen_write_cr3;
2211 }
2212 
2213 /*
2214  * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
2215  * not the first page table in the page table pool.
2216  * Iterate through the initial page tables to find the real page table base.
2217  */
2218 static phys_addr_t __init xen_find_pt_base(pmd_t *pmd)
2219 {
2220         phys_addr_t pt_base, paddr;
2221         unsigned pmdidx;
2222 
2223         pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
2224 
2225         for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
2226                 if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
2227                         paddr = m2p(pmd[pmdidx].pmd);
2228                         pt_base = min(pt_base, paddr);
2229                 }
2230 
2231         return pt_base;
2232 }
2233 
2234 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2235 {
2236         pmd_t *kernel_pmd;
2237 
2238         kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2239 
2240         xen_pt_base = xen_find_pt_base(kernel_pmd);
2241         xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
2242 
2243         initial_kernel_pmd =
2244                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2245 
2246         max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
2247 
2248         copy_page(initial_kernel_pmd, kernel_pmd);
2249 
2250         xen_map_identity_early(initial_kernel_pmd, max_pfn);
2251 
2252         copy_page(initial_page_table, pgd);
2253         initial_page_table[KERNEL_PGD_BOUNDARY] =
2254                 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2255 
2256         set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2257         set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2258         set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2259 
2260         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2261 
2262         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2263                           PFN_DOWN(__pa(initial_page_table)));
2264         xen_write_cr3(__pa(initial_page_table));
2265 
2266         memblock_reserve(xen_pt_base, xen_pt_size);
2267 }
2268 #endif  /* CONFIG_X86_64 */
2269 
2270 void __init xen_reserve_special_pages(void)
2271 {
2272         phys_addr_t paddr;
2273 
2274         memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2275         if (xen_start_info->store_mfn) {
2276                 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2277                 memblock_reserve(paddr, PAGE_SIZE);
2278         }
2279         if (!xen_initial_domain()) {
2280                 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2281                 memblock_reserve(paddr, PAGE_SIZE);
2282         }
2283 }
2284 
2285 void __init xen_pt_check_e820(void)
2286 {
2287         if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2288                 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2289                 BUG();
2290         }
2291 }
2292 
2293 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2294 
2295 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2296 {
2297         pte_t pte;
2298 
2299         phys >>= PAGE_SHIFT;
2300 
2301         switch (idx) {
2302         case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2303 #ifdef CONFIG_X86_32
2304         case FIX_WP_TEST:
2305 # ifdef CONFIG_HIGHMEM
2306         case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2307 # endif
2308 #elif defined(CONFIG_X86_VSYSCALL_EMULATION)
2309         case VSYSCALL_PAGE:
2310 #endif
2311         case FIX_TEXT_POKE0:
2312         case FIX_TEXT_POKE1:
2313                 /* All local page mappings */
2314                 pte = pfn_pte(phys, prot);
2315                 break;
2316 
2317 #ifdef CONFIG_X86_LOCAL_APIC
2318         case FIX_APIC_BASE:     /* maps dummy local APIC */
2319                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2320                 break;
2321 #endif
2322 
2323 #ifdef CONFIG_X86_IO_APIC
2324         case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2325                 /*
2326                  * We just don't map the IO APIC - all access is via
2327                  * hypercalls.  Keep the address in the pte for reference.
2328                  */
2329                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2330                 break;
2331 #endif
2332 
2333         case FIX_PARAVIRT_BOOTMAP:
2334                 /* This is an MFN, but it isn't an IO mapping from the
2335                    IO domain */
2336                 pte = mfn_pte(phys, prot);
2337                 break;
2338 
2339         default:
2340                 /* By default, set_fixmap is used for hardware mappings */
2341                 pte = mfn_pte(phys, prot);
2342                 break;
2343         }
2344 
2345         __native_set_fixmap(idx, pte);
2346 
2347 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2348         /* Replicate changes to map the vsyscall page into the user
2349            pagetable vsyscall mapping. */
2350         if (idx == VSYSCALL_PAGE) {
2351                 unsigned long vaddr = __fix_to_virt(idx);
2352                 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2353         }
2354 #endif
2355 }
2356 
2357 static void __init xen_post_allocator_init(void)
2358 {
2359         pv_mmu_ops.set_pte = xen_set_pte;
2360         pv_mmu_ops.set_pmd = xen_set_pmd;
2361         pv_mmu_ops.set_pud = xen_set_pud;
2362 #ifdef CONFIG_X86_64
2363         pv_mmu_ops.set_p4d = xen_set_p4d;
2364 #endif
2365 
2366         /* This will work as long as patching hasn't happened yet
2367            (which it hasn't) */
2368         pv_mmu_ops.alloc_pte = xen_alloc_pte;
2369         pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2370         pv_mmu_ops.release_pte = xen_release_pte;
2371         pv_mmu_ops.release_pmd = xen_release_pmd;
2372 #ifdef CONFIG_X86_64
2373         pv_mmu_ops.alloc_pud = xen_alloc_pud;
2374         pv_mmu_ops.release_pud = xen_release_pud;
2375 #endif
2376         pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2377 
2378 #ifdef CONFIG_X86_64
2379         pv_mmu_ops.write_cr3 = &xen_write_cr3;
2380 #endif
2381 }
2382 
2383 static void xen_leave_lazy_mmu(void)
2384 {
2385         preempt_disable();
2386         xen_mc_flush();
2387         paravirt_leave_lazy_mmu();
2388         preempt_enable();
2389 }
2390 
2391 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2392         .read_cr2 = xen_read_cr2,
2393         .write_cr2 = xen_write_cr2,
2394 
2395         .read_cr3 = xen_read_cr3,
2396         .write_cr3 = xen_write_cr3_init,
2397 
2398         .flush_tlb_user = xen_flush_tlb,
2399         .flush_tlb_kernel = xen_flush_tlb,
2400         .flush_tlb_one_user = xen_flush_tlb_one_user,
2401         .flush_tlb_others = xen_flush_tlb_others,
2402 
2403         .pgd_alloc = xen_pgd_alloc,
2404         .pgd_free = xen_pgd_free,
2405 
2406         .alloc_pte = xen_alloc_pte_init,
2407         .release_pte = xen_release_pte_init,
2408         .alloc_pmd = xen_alloc_pmd_init,
2409         .release_pmd = xen_release_pmd_init,
2410 
2411         .set_pte = xen_set_pte_init,
2412         .set_pte_at = xen_set_pte_at,
2413         .set_pmd = xen_set_pmd_hyper,
2414 
2415         .ptep_modify_prot_start = __ptep_modify_prot_start,
2416         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2417 
2418         .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2419         .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2420 
2421         .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2422         .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2423 
2424 #ifdef CONFIG_X86_PAE
2425         .set_pte_atomic = xen_set_pte_atomic,
2426         .pte_clear = xen_pte_clear,
2427         .pmd_clear = xen_pmd_clear,
2428 #endif  /* CONFIG_X86_PAE */
2429         .set_pud = xen_set_pud_hyper,
2430 
2431         .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2432         .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2433 
2434 #ifdef CONFIG_X86_64
2435         .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2436         .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2437         .set_p4d = xen_set_p4d_hyper,
2438 
2439         .alloc_pud = xen_alloc_pmd_init,
2440         .release_pud = xen_release_pmd_init,
2441 
2442 #if CONFIG_PGTABLE_LEVELS >= 5
2443         .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2444         .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2445 #endif
2446 #endif  /* CONFIG_X86_64 */
2447 
2448         .activate_mm = xen_activate_mm,
2449         .dup_mmap = xen_dup_mmap,
2450         .exit_mmap = xen_exit_mmap,
2451 
2452         .lazy_mode = {
2453                 .enter = paravirt_enter_lazy_mmu,
2454                 .leave = xen_leave_lazy_mmu,
2455                 .flush = paravirt_flush_lazy_mmu,
2456         },
2457 
2458         .set_fixmap = xen_set_fixmap,
2459 };
2460 
2461 void __init xen_init_mmu_ops(void)
2462 {
2463         x86_init.paging.pagetable_init = xen_pagetable_init;
2464         x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2465 
2466         pv_mmu_ops = xen_mmu_ops;
2467 
2468         memset(dummy_mapping, 0xff, PAGE_SIZE);
2469 }
2470 
2471 /* Protected by xen_reservation_lock. */
2472 #define MAX_CONTIG_ORDER 9 /* 2MB */
2473 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2474 
2475 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2476 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2477                                 unsigned long *in_frames,
2478                                 unsigned long *out_frames)
2479 {
2480         int i;
2481         struct multicall_space mcs;
2482 
2483         xen_mc_batch();
2484         for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2485                 mcs = __xen_mc_entry(0);
2486 
2487                 if (in_frames)
2488                         in_frames[i] = virt_to_mfn(vaddr);
2489 
2490                 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2491                 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2492 
2493                 if (out_frames)
2494                         out_frames[i] = virt_to_pfn(vaddr);
2495         }
2496         xen_mc_issue(0);
2497 }
2498 
2499 /*
2500  * Update the pfn-to-mfn mappings for a virtual address range, either to
2501  * point to an array of mfns, or contiguously from a single starting
2502  * mfn.
2503  */
2504 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2505                                      unsigned long *mfns,
2506                                      unsigned long first_mfn)
2507 {
2508         unsigned i, limit;
2509         unsigned long mfn;
2510 
2511         xen_mc_batch();
2512 
2513         limit = 1u << order;
2514         for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2515                 struct multicall_space mcs;
2516                 unsigned flags;
2517 
2518                 mcs = __xen_mc_entry(0);
2519                 if (mfns)
2520                         mfn = mfns[i];
2521                 else
2522                         mfn = first_mfn + i;
2523 
2524                 if (i < (limit - 1))
2525                         flags = 0;
2526                 else {
2527                         if (order == 0)
2528                                 flags = UVMF_INVLPG | UVMF_ALL;
2529                         else
2530                                 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2531                 }
2532 
2533                 MULTI_update_va_mapping(mcs.mc, vaddr,
2534                                 mfn_pte(mfn, PAGE_KERNEL), flags);
2535 
2536                 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2537         }
2538 
2539         xen_mc_issue(0);
2540 }
2541 
2542 /*
2543  * Perform the hypercall to exchange a region of our pfns to point to
2544  * memory with the required contiguous alignment.  Takes the pfns as
2545  * input, and populates mfns as output.
2546  *
2547  * Returns a success code indicating whether the hypervisor was able to
2548  * satisfy the request or not.
2549  */
2550 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2551                                unsigned long *pfns_in,
2552                                unsigned long extents_out,
2553                                unsigned int order_out,
2554                                unsigned long *mfns_out,
2555                                unsigned int address_bits)
2556 {
2557         long rc;
2558         int success;
2559 
2560         struct xen_memory_exchange exchange = {
2561                 .in = {
2562                         .nr_extents   = extents_in,
2563                         .extent_order = order_in,
2564                         .extent_start = pfns_in,
2565                         .domid        = DOMID_SELF
2566                 },
2567                 .out = {
2568                         .nr_extents   = extents_out,
2569                         .extent_order = order_out,
2570                         .extent_start = mfns_out,
2571                         .address_bits = address_bits,
2572                         .domid        = DOMID_SELF
2573                 }
2574         };
2575 
2576         BUG_ON(extents_in << order_in != extents_out << order_out);
2577 
2578         rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2579         success = (exchange.nr_exchanged == extents_in);
2580 
2581         BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2582         BUG_ON(success && (rc != 0));
2583 
2584         return success;
2585 }
2586 
2587 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2588                                  unsigned int address_bits,
2589                                  dma_addr_t *dma_handle)
2590 {
2591         unsigned long *in_frames = discontig_frames, out_frame;
2592         unsigned long  flags;
2593         int            success;
2594         unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2595 
2596         /*
2597          * Currently an auto-translated guest will not perform I/O, nor will
2598          * it require PAE page directories below 4GB. Therefore any calls to
2599          * this function are redundant and can be ignored.
2600          */
2601 
2602         if (unlikely(order > MAX_CONTIG_ORDER))
2603                 return -ENOMEM;
2604 
2605         memset((void *) vstart, 0, PAGE_SIZE << order);
2606 
2607         spin_lock_irqsave(&xen_reservation_lock, flags);
2608 
2609         /* 1. Zap current PTEs, remembering MFNs. */
2610         xen_zap_pfn_range(vstart, order, in_frames, NULL);
2611 
2612         /* 2. Get a new contiguous memory extent. */
2613         out_frame = virt_to_pfn(vstart);
2614         success = xen_exchange_memory(1UL << order, 0, in_frames,
2615                                       1, order, &out_frame,
2616                                       address_bits);
2617 
2618         /* 3. Map the new extent in place of old pages. */
2619         if (success)
2620                 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2621         else
2622                 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2623 
2624         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2625 
2626         *dma_handle = virt_to_machine(vstart).maddr;
2627         return success ? 0 : -ENOMEM;
2628 }
2629 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2630 
2631 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2632 {
2633         unsigned long *out_frames = discontig_frames, in_frame;
2634         unsigned long  flags;
2635         int success;
2636         unsigned long vstart;
2637 
2638         if (unlikely(order > MAX_CONTIG_ORDER))
2639                 return;
2640 
2641         vstart = (unsigned long)phys_to_virt(pstart);
2642         memset((void *) vstart, 0, PAGE_SIZE << order);
2643 
2644         spin_lock_irqsave(&xen_reservation_lock, flags);
2645 
2646         /* 1. Find start MFN of contiguous extent. */
2647         in_frame = virt_to_mfn(vstart);
2648 
2649         /* 2. Zap current PTEs. */
2650         xen_zap_pfn_range(vstart, order, NULL, out_frames);
2651 
2652         /* 3. Do the exchange for non-contiguous MFNs. */
2653         success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2654                                         0, out_frames, 0);
2655 
2656         /* 4. Map new pages in place of old pages. */
2657         if (success)
2658                 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2659         else
2660                 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2661 
2662         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2663 }
2664 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2665 
2666 #ifdef CONFIG_KEXEC_CORE
2667 phys_addr_t paddr_vmcoreinfo_note(void)
2668 {
2669         if (xen_pv_domain())
2670                 return virt_to_machine(vmcoreinfo_note).maddr;
2671         else
2672                 return __pa(vmcoreinfo_note);
2673 }
2674 #endif /* CONFIG_KEXEC_CORE */
2675 

~ [ 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