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

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

Version: ~ [ linux-4.19-rc7 ] ~ [ linux-4.18.12 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.74 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.131 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.159 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.123 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.59 ] ~ [ 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-2.6.31.14 ] ~ [ linux-2.6.30.10 ] ~ [ linux-2.6.29.6 ] ~ [ linux-2.6.28.10 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Xen mmu operations
  3  *
  4  * This file contains the various mmu fetch and update operations.
  5  * The most important job they must perform is the mapping between the
  6  * domain's pfn and the overall machine mfns.
  7  *
  8  * Xen allows guests to directly update the pagetable, in a controlled
  9  * fashion.  In other words, the guest modifies the same pagetable
 10  * that the CPU actually uses, which eliminates the overhead of having
 11  * a separate shadow pagetable.
 12  *
 13  * In order to allow this, it falls on the guest domain to map its
 14  * notion of a "physical" pfn - which is just a domain-local linear
 15  * address - into a real "machine address" which the CPU's MMU can
 16  * use.
 17  *
 18  * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
 19  * inserted directly into the pagetable.  When creating a new
 20  * pte/pmd/pgd, it converts the passed pfn into an mfn.  Conversely,
 21  * when reading the content back with __(pgd|pmd|pte)_val, it converts
 22  * the mfn back into a pfn.
 23  *
 24  * The other constraint is that all pages which make up a pagetable
 25  * must be mapped read-only in the guest.  This prevents uncontrolled
 26  * guest updates to the pagetable.  Xen strictly enforces this, and
 27  * will disallow any pagetable update which will end up mapping a
 28  * pagetable page RW, and will disallow using any writable page as a
 29  * pagetable.
 30  *
 31  * Naively, when loading %cr3 with the base of a new pagetable, Xen
 32  * would need to validate the whole pagetable before going on.
 33  * Naturally, this is quite slow.  The solution is to "pin" a
 34  * pagetable, which enforces all the constraints on the pagetable even
 35  * when it is not actively in use.  This menas that Xen can be assured
 36  * that it is still valid when you do load it into %cr3, and doesn't
 37  * need to revalidate it.
 38  *
 39  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
 40  */
 41 #include <linux/sched/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 #include <asm/tlb.h>
 71 
 72 #include <asm/xen/hypercall.h>
 73 #include <asm/xen/hypervisor.h>
 74 
 75 #include <xen/xen.h>
 76 #include <xen/page.h>
 77 #include <xen/interface/xen.h>
 78 #include <xen/interface/hvm/hvm_op.h>
 79 #include <xen/interface/version.h>
 80 #include <xen/interface/memory.h>
 81 #include <xen/hvc-console.h>
 82 
 83 #include "multicalls.h"
 84 #include "mmu.h"
 85 #include "debugfs.h"
 86 
 87 #ifdef CONFIG_X86_32
 88 /*
 89  * Identity map, in addition to plain kernel map.  This needs to be
 90  * large enough to allocate page table pages to allocate the rest.
 91  * Each page can map 2MB.
 92  */
 93 #define LEVEL1_IDENT_ENTRIES    (PTRS_PER_PTE * 4)
 94 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
 95 #endif
 96 #ifdef CONFIG_X86_64
 97 /* l3 pud for userspace vsyscall mapping */
 98 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
 99 #endif /* CONFIG_X86_64 */
100 
101 /*
102  * Note about cr3 (pagetable base) values:
103  *
104  * xen_cr3 contains the current logical cr3 value; it contains the
105  * last set cr3.  This may not be the current effective cr3, because
106  * its update may be being lazily deferred.  However, a vcpu looking
107  * at its own cr3 can use this value knowing that it everything will
108  * be self-consistent.
109  *
110  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
111  * hypercall to set the vcpu cr3 is complete (so it may be a little
112  * out of date, but it will never be set early).  If one vcpu is
113  * looking at another vcpu's cr3 value, it should use this variable.
114  */
115 DEFINE_PER_CPU(unsigned long, xen_cr3);  /* cr3 stored as physaddr */
116 DEFINE_PER_CPU(unsigned long, xen_current_cr3);  /* actual vcpu cr3 */
117 
118 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
119 
120 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
121 
122 /*
123  * Just beyond the highest usermode address.  STACK_TOP_MAX has a
124  * redzone above it, so round it up to a PGD boundary.
125  */
126 #define USER_LIMIT      ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
127 
128 void make_lowmem_page_readonly(void *vaddr)
129 {
130         pte_t *pte, ptev;
131         unsigned long address = (unsigned long)vaddr;
132         unsigned int level;
133 
134         pte = lookup_address(address, &level);
135         if (pte == NULL)
136                 return;         /* vaddr missing */
137 
138         ptev = pte_wrprotect(*pte);
139 
140         if (HYPERVISOR_update_va_mapping(address, ptev, 0))
141                 BUG();
142 }
143 
144 void make_lowmem_page_readwrite(void *vaddr)
145 {
146         pte_t *pte, ptev;
147         unsigned long address = (unsigned long)vaddr;
148         unsigned int level;
149 
150         pte = lookup_address(address, &level);
151         if (pte == NULL)
152                 return;         /* vaddr missing */
153 
154         ptev = pte_mkwrite(*pte);
155 
156         if (HYPERVISOR_update_va_mapping(address, ptev, 0))
157                 BUG();
158 }
159 
160 
161 /*
162  * During early boot all page table pages are pinned, but we do not have struct
163  * pages, so return true until struct pages are ready.
164  */
165 static bool xen_page_pinned(void *ptr)
166 {
167         if (static_branch_likely(&xen_struct_pages_ready)) {
168                 struct page *page = virt_to_page(ptr);
169 
170                 return PagePinned(page);
171         }
172         return true;
173 }
174 
175 static void xen_extend_mmu_update(const struct mmu_update *update)
176 {
177         struct multicall_space mcs;
178         struct mmu_update *u;
179 
180         mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
181 
182         if (mcs.mc != NULL) {
183                 mcs.mc->args[1]++;
184         } else {
185                 mcs = __xen_mc_entry(sizeof(*u));
186                 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
187         }
188 
189         u = mcs.args;
190         *u = *update;
191 }
192 
193 static void xen_extend_mmuext_op(const struct mmuext_op *op)
194 {
195         struct multicall_space mcs;
196         struct mmuext_op *u;
197 
198         mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
199 
200         if (mcs.mc != NULL) {
201                 mcs.mc->args[1]++;
202         } else {
203                 mcs = __xen_mc_entry(sizeof(*u));
204                 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
205         }
206 
207         u = mcs.args;
208         *u = *op;
209 }
210 
211 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
212 {
213         struct mmu_update u;
214 
215         preempt_disable();
216 
217         xen_mc_batch();
218 
219         /* ptr may be ioremapped for 64-bit pagetable setup */
220         u.ptr = arbitrary_virt_to_machine(ptr).maddr;
221         u.val = pmd_val_ma(val);
222         xen_extend_mmu_update(&u);
223 
224         xen_mc_issue(PARAVIRT_LAZY_MMU);
225 
226         preempt_enable();
227 }
228 
229 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
230 {
231         trace_xen_mmu_set_pmd(ptr, val);
232 
233         /* If page is not pinned, we can just update the entry
234            directly */
235         if (!xen_page_pinned(ptr)) {
236                 *ptr = val;
237                 return;
238         }
239 
240         xen_set_pmd_hyper(ptr, val);
241 }
242 
243 /*
244  * Associate a virtual page frame with a given physical page frame
245  * and protection flags for that frame.
246  */
247 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
248 {
249         set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
250 }
251 
252 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
253 {
254         struct mmu_update u;
255 
256         if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
257                 return false;
258 
259         xen_mc_batch();
260 
261         u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
262         u.val = pte_val_ma(pteval);
263         xen_extend_mmu_update(&u);
264 
265         xen_mc_issue(PARAVIRT_LAZY_MMU);
266 
267         return true;
268 }
269 
270 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
271 {
272         if (!xen_batched_set_pte(ptep, pteval)) {
273                 /*
274                  * Could call native_set_pte() here and trap and
275                  * emulate the PTE write but with 32-bit guests this
276                  * needs two traps (one for each of the two 32-bit
277                  * words in the PTE) so do one hypercall directly
278                  * instead.
279                  */
280                 struct mmu_update u;
281 
282                 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
283                 u.val = pte_val_ma(pteval);
284                 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
285         }
286 }
287 
288 static void xen_set_pte(pte_t *ptep, pte_t pteval)
289 {
290         trace_xen_mmu_set_pte(ptep, pteval);
291         __xen_set_pte(ptep, pteval);
292 }
293 
294 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
295                     pte_t *ptep, pte_t pteval)
296 {
297         trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
298         __xen_set_pte(ptep, pteval);
299 }
300 
301 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
302                                  unsigned long addr, pte_t *ptep)
303 {
304         /* Just return the pte as-is.  We preserve the bits on commit */
305         trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
306         return *ptep;
307 }
308 
309 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
310                                  pte_t *ptep, pte_t pte)
311 {
312         struct mmu_update u;
313 
314         trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
315         xen_mc_batch();
316 
317         u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
318         u.val = pte_val_ma(pte);
319         xen_extend_mmu_update(&u);
320 
321         xen_mc_issue(PARAVIRT_LAZY_MMU);
322 }
323 
324 /* Assume pteval_t is equivalent to all the other *val_t types. */
325 static pteval_t pte_mfn_to_pfn(pteval_t val)
326 {
327         if (val & _PAGE_PRESENT) {
328                 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
329                 unsigned long pfn = mfn_to_pfn(mfn);
330 
331                 pteval_t flags = val & PTE_FLAGS_MASK;
332                 if (unlikely(pfn == ~0))
333                         val = flags & ~_PAGE_PRESENT;
334                 else
335                         val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
336         }
337 
338         return val;
339 }
340 
341 static pteval_t pte_pfn_to_mfn(pteval_t val)
342 {
343         if (val & _PAGE_PRESENT) {
344                 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
345                 pteval_t flags = val & PTE_FLAGS_MASK;
346                 unsigned long mfn;
347 
348                 mfn = __pfn_to_mfn(pfn);
349 
350                 /*
351                  * If there's no mfn for the pfn, then just create an
352                  * empty non-present pte.  Unfortunately this loses
353                  * information about the original pfn, so
354                  * pte_mfn_to_pfn is asymmetric.
355                  */
356                 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
357                         mfn = 0;
358                         flags = 0;
359                 } else
360                         mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
361                 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
362         }
363 
364         return val;
365 }
366 
367 __visible pteval_t xen_pte_val(pte_t pte)
368 {
369         pteval_t pteval = pte.pte;
370 
371         return pte_mfn_to_pfn(pteval);
372 }
373 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
374 
375 __visible pgdval_t xen_pgd_val(pgd_t pgd)
376 {
377         return pte_mfn_to_pfn(pgd.pgd);
378 }
379 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
380 
381 __visible pte_t xen_make_pte(pteval_t pte)
382 {
383         pte = pte_pfn_to_mfn(pte);
384 
385         return native_make_pte(pte);
386 }
387 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
388 
389 __visible pgd_t xen_make_pgd(pgdval_t pgd)
390 {
391         pgd = pte_pfn_to_mfn(pgd);
392         return native_make_pgd(pgd);
393 }
394 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
395 
396 __visible pmdval_t xen_pmd_val(pmd_t pmd)
397 {
398         return pte_mfn_to_pfn(pmd.pmd);
399 }
400 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
401 
402 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
403 {
404         struct mmu_update u;
405 
406         preempt_disable();
407 
408         xen_mc_batch();
409 
410         /* ptr may be ioremapped for 64-bit pagetable setup */
411         u.ptr = arbitrary_virt_to_machine(ptr).maddr;
412         u.val = pud_val_ma(val);
413         xen_extend_mmu_update(&u);
414 
415         xen_mc_issue(PARAVIRT_LAZY_MMU);
416 
417         preempt_enable();
418 }
419 
420 static void xen_set_pud(pud_t *ptr, pud_t val)
421 {
422         trace_xen_mmu_set_pud(ptr, val);
423 
424         /* If page is not pinned, we can just update the entry
425            directly */
426         if (!xen_page_pinned(ptr)) {
427                 *ptr = val;
428                 return;
429         }
430 
431         xen_set_pud_hyper(ptr, val);
432 }
433 
434 #ifdef CONFIG_X86_PAE
435 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
436 {
437         trace_xen_mmu_set_pte_atomic(ptep, pte);
438         __xen_set_pte(ptep, pte);
439 }
440 
441 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
442 {
443         trace_xen_mmu_pte_clear(mm, addr, ptep);
444         __xen_set_pte(ptep, native_make_pte(0));
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.
1234          */
1235         size = roundup(size, PMD_SIZE);
1236 
1237         if (addr >= __START_KERNEL_map) {
1238                 xen_cleanhighmap(addr, addr + size);
1239                 size = PAGE_ALIGN(xen_start_info->nr_pages *
1240                                   sizeof(unsigned long));
1241                 memblock_free(__pa(addr), size);
1242         } else {
1243                 xen_cleanmfnmap(addr);
1244         }
1245 }
1246 
1247 static void __init xen_pagetable_cleanhighmap(void)
1248 {
1249         unsigned long size;
1250         unsigned long addr;
1251 
1252         /* At this stage, cleanup_highmap has already cleaned __ka space
1253          * from _brk_limit way up to the max_pfn_mapped (which is the end of
1254          * the ramdisk). We continue on, erasing PMD entries that point to page
1255          * tables - do note that they are accessible at this stage via __va.
1256          * As Xen is aligning the memory end to a 4MB boundary, for good
1257          * measure we also round up to PMD_SIZE * 2 - which means that if
1258          * anybody is using __ka address to the initial boot-stack - and try
1259          * to use it - they are going to crash. The xen_start_info has been
1260          * taken care of already in xen_setup_kernel_pagetable. */
1261         addr = xen_start_info->pt_base;
1262         size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1263 
1264         xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1265         xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1266 }
1267 #endif
1268 
1269 static void __init xen_pagetable_p2m_setup(void)
1270 {
1271         xen_vmalloc_p2m_tree();
1272 
1273 #ifdef CONFIG_X86_64
1274         xen_pagetable_p2m_free();
1275 
1276         xen_pagetable_cleanhighmap();
1277 #endif
1278         /* And revector! Bye bye old array */
1279         xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1280 }
1281 
1282 static void __init xen_pagetable_init(void)
1283 {
1284         paging_init();
1285         xen_post_allocator_init();
1286 
1287         xen_pagetable_p2m_setup();
1288 
1289         /* Allocate and initialize top and mid mfn levels for p2m structure */
1290         xen_build_mfn_list_list();
1291 
1292         /* Remap memory freed due to conflicts with E820 map */
1293         xen_remap_memory();
1294         xen_setup_mfn_list_list();
1295 }
1296 static void xen_write_cr2(unsigned long cr2)
1297 {
1298         this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1299 }
1300 
1301 static unsigned long xen_read_cr2(void)
1302 {
1303         return this_cpu_read(xen_vcpu)->arch.cr2;
1304 }
1305 
1306 unsigned long xen_read_cr2_direct(void)
1307 {
1308         return this_cpu_read(xen_vcpu_info.arch.cr2);
1309 }
1310 
1311 static noinline void xen_flush_tlb(void)
1312 {
1313         struct mmuext_op *op;
1314         struct multicall_space mcs;
1315 
1316         preempt_disable();
1317 
1318         mcs = xen_mc_entry(sizeof(*op));
1319 
1320         op = mcs.args;
1321         op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1322         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1323 
1324         xen_mc_issue(PARAVIRT_LAZY_MMU);
1325 
1326         preempt_enable();
1327 }
1328 
1329 static void xen_flush_tlb_one_user(unsigned long addr)
1330 {
1331         struct mmuext_op *op;
1332         struct multicall_space mcs;
1333 
1334         trace_xen_mmu_flush_tlb_one_user(addr);
1335 
1336         preempt_disable();
1337 
1338         mcs = xen_mc_entry(sizeof(*op));
1339         op = mcs.args;
1340         op->cmd = MMUEXT_INVLPG_LOCAL;
1341         op->arg1.linear_addr = addr & PAGE_MASK;
1342         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1343 
1344         xen_mc_issue(PARAVIRT_LAZY_MMU);
1345 
1346         preempt_enable();
1347 }
1348 
1349 static void xen_flush_tlb_others(const struct cpumask *cpus,
1350                                  const struct flush_tlb_info *info)
1351 {
1352         struct {
1353                 struct mmuext_op op;
1354                 DECLARE_BITMAP(mask, NR_CPUS);
1355         } *args;
1356         struct multicall_space mcs;
1357         const size_t mc_entry_size = sizeof(args->op) +
1358                 sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1359 
1360         trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end);
1361 
1362         if (cpumask_empty(cpus))
1363                 return;         /* nothing to do */
1364 
1365         mcs = xen_mc_entry(mc_entry_size);
1366         args = mcs.args;
1367         args->op.arg2.vcpumask = to_cpumask(args->mask);
1368 
1369         /* Remove us, and any offline CPUS. */
1370         cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1371         cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1372 
1373         args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1374         if (info->end != TLB_FLUSH_ALL &&
1375             (info->end - info->start) <= PAGE_SIZE) {
1376                 args->op.cmd = MMUEXT_INVLPG_MULTI;
1377                 args->op.arg1.linear_addr = info->start;
1378         }
1379 
1380         MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1381 
1382         xen_mc_issue(PARAVIRT_LAZY_MMU);
1383 }
1384 
1385 static unsigned long xen_read_cr3(void)
1386 {
1387         return this_cpu_read(xen_cr3);
1388 }
1389 
1390 static void set_current_cr3(void *v)
1391 {
1392         this_cpu_write(xen_current_cr3, (unsigned long)v);
1393 }
1394 
1395 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1396 {
1397         struct mmuext_op op;
1398         unsigned long mfn;
1399 
1400         trace_xen_mmu_write_cr3(kernel, cr3);
1401 
1402         if (cr3)
1403                 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1404         else
1405                 mfn = 0;
1406 
1407         WARN_ON(mfn == 0 && kernel);
1408 
1409         op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1410         op.arg1.mfn = mfn;
1411 
1412         xen_extend_mmuext_op(&op);
1413 
1414         if (kernel) {
1415                 this_cpu_write(xen_cr3, cr3);
1416 
1417                 /* Update xen_current_cr3 once the batch has actually
1418                    been submitted. */
1419                 xen_mc_callback(set_current_cr3, (void *)cr3);
1420         }
1421 }
1422 static void xen_write_cr3(unsigned long cr3)
1423 {
1424         BUG_ON(preemptible());
1425 
1426         xen_mc_batch();  /* disables interrupts */
1427 
1428         /* Update while interrupts are disabled, so its atomic with
1429            respect to ipis */
1430         this_cpu_write(xen_cr3, cr3);
1431 
1432         __xen_write_cr3(true, cr3);
1433 
1434 #ifdef CONFIG_X86_64
1435         {
1436                 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1437                 if (user_pgd)
1438                         __xen_write_cr3(false, __pa(user_pgd));
1439                 else
1440                         __xen_write_cr3(false, 0);
1441         }
1442 #endif
1443 
1444         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1445 }
1446 
1447 #ifdef CONFIG_X86_64
1448 /*
1449  * At the start of the day - when Xen launches a guest, it has already
1450  * built pagetables for the guest. We diligently look over them
1451  * in xen_setup_kernel_pagetable and graft as appropriate them in the
1452  * init_top_pgt and its friends. Then when we are happy we load
1453  * the new init_top_pgt - and continue on.
1454  *
1455  * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1456  * up the rest of the pagetables. When it has completed it loads the cr3.
1457  * N.B. that baremetal would start at 'start_kernel' (and the early
1458  * #PF handler would create bootstrap pagetables) - so we are running
1459  * with the same assumptions as what to do when write_cr3 is executed
1460  * at this point.
1461  *
1462  * Since there are no user-page tables at all, we have two variants
1463  * of xen_write_cr3 - the early bootup (this one), and the late one
1464  * (xen_write_cr3). The reason we have to do that is that in 64-bit
1465  * the Linux kernel and user-space are both in ring 3 while the
1466  * hypervisor is in ring 0.
1467  */
1468 static void __init xen_write_cr3_init(unsigned long cr3)
1469 {
1470         BUG_ON(preemptible());
1471 
1472         xen_mc_batch();  /* disables interrupts */
1473 
1474         /* Update while interrupts are disabled, so its atomic with
1475            respect to ipis */
1476         this_cpu_write(xen_cr3, cr3);
1477 
1478         __xen_write_cr3(true, cr3);
1479 
1480         xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1481 }
1482 #endif
1483 
1484 static int xen_pgd_alloc(struct mm_struct *mm)
1485 {
1486         pgd_t *pgd = mm->pgd;
1487         int ret = 0;
1488 
1489         BUG_ON(PagePinned(virt_to_page(pgd)));
1490 
1491 #ifdef CONFIG_X86_64
1492         {
1493                 struct page *page = virt_to_page(pgd);
1494                 pgd_t *user_pgd;
1495 
1496                 BUG_ON(page->private != 0);
1497 
1498                 ret = -ENOMEM;
1499 
1500                 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1501                 page->private = (unsigned long)user_pgd;
1502 
1503                 if (user_pgd != NULL) {
1504 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1505                         user_pgd[pgd_index(VSYSCALL_ADDR)] =
1506                                 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1507 #endif
1508                         ret = 0;
1509                 }
1510 
1511                 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1512         }
1513 #endif
1514         return ret;
1515 }
1516 
1517 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1518 {
1519 #ifdef CONFIG_X86_64
1520         pgd_t *user_pgd = xen_get_user_pgd(pgd);
1521 
1522         if (user_pgd)
1523                 free_page((unsigned long)user_pgd);
1524 #endif
1525 }
1526 
1527 /*
1528  * Init-time set_pte while constructing initial pagetables, which
1529  * doesn't allow RO page table pages to be remapped RW.
1530  *
1531  * If there is no MFN for this PFN then this page is initially
1532  * ballooned out so clear the PTE (as in decrease_reservation() in
1533  * drivers/xen/balloon.c).
1534  *
1535  * Many of these PTE updates are done on unpinned and writable pages
1536  * and doing a hypercall for these is unnecessary and expensive.  At
1537  * this point it is not possible to tell if a page is pinned or not,
1538  * so always write the PTE directly and rely on Xen trapping and
1539  * emulating any updates as necessary.
1540  */
1541 __visible pte_t xen_make_pte_init(pteval_t pte)
1542 {
1543 #ifdef CONFIG_X86_64
1544         unsigned long pfn;
1545 
1546         /*
1547          * Pages belonging to the initial p2m list mapped outside the default
1548          * address range must be mapped read-only. This region contains the
1549          * page tables for mapping the p2m list, too, and page tables MUST be
1550          * mapped read-only.
1551          */
1552         pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1553         if (xen_start_info->mfn_list < __START_KERNEL_map &&
1554             pfn >= xen_start_info->first_p2m_pfn &&
1555             pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1556                 pte &= ~_PAGE_RW;
1557 #endif
1558         pte = pte_pfn_to_mfn(pte);
1559         return native_make_pte(pte);
1560 }
1561 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1562 
1563 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1564 {
1565 #ifdef CONFIG_X86_32
1566         /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1567         if (pte_mfn(pte) != INVALID_P2M_ENTRY
1568             && pte_val_ma(*ptep) & _PAGE_PRESENT)
1569                 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1570                                pte_val_ma(pte));
1571 #endif
1572         __xen_set_pte(ptep, pte);
1573 }
1574 
1575 /* Early in boot, while setting up the initial pagetable, assume
1576    everything is pinned. */
1577 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1578 {
1579 #ifdef CONFIG_FLATMEM
1580         BUG_ON(mem_map);        /* should only be used early */
1581 #endif
1582         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1583         pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1584 }
1585 
1586 /* Used for pmd and pud */
1587 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1588 {
1589 #ifdef CONFIG_FLATMEM
1590         BUG_ON(mem_map);        /* should only be used early */
1591 #endif
1592         make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1593 }
1594 
1595 /* Early release_pte assumes that all pts are pinned, since there's
1596    only init_mm and anything attached to that is pinned. */
1597 static void __init xen_release_pte_init(unsigned long pfn)
1598 {
1599         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1600         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1601 }
1602 
1603 static void __init xen_release_pmd_init(unsigned long pfn)
1604 {
1605         make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1606 }
1607 
1608 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1609 {
1610         struct multicall_space mcs;
1611         struct mmuext_op *op;
1612 
1613         mcs = __xen_mc_entry(sizeof(*op));
1614         op = mcs.args;
1615         op->cmd = cmd;
1616         op->arg1.mfn = pfn_to_mfn(pfn);
1617 
1618         MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1619 }
1620 
1621 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1622 {
1623         struct multicall_space mcs;
1624         unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1625 
1626         mcs = __xen_mc_entry(0);
1627         MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1628                                 pfn_pte(pfn, prot), 0);
1629 }
1630 
1631 /* This needs to make sure the new pte page is pinned iff its being
1632    attached to a pinned pagetable. */
1633 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1634                                     unsigned level)
1635 {
1636         bool pinned = xen_page_pinned(mm->pgd);
1637 
1638         trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1639 
1640         if (pinned) {
1641                 struct page *page = pfn_to_page(pfn);
1642 
1643                 if (static_branch_likely(&xen_struct_pages_ready))
1644                         SetPagePinned(page);
1645 
1646                 if (!PageHighMem(page)) {
1647                         xen_mc_batch();
1648 
1649                         __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1650 
1651                         if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1652                                 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1653 
1654                         xen_mc_issue(PARAVIRT_LAZY_MMU);
1655                 } else {
1656                         /* make sure there are no stray mappings of
1657                            this page */
1658                         kmap_flush_unused();
1659                 }
1660         }
1661 }
1662 
1663 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1664 {
1665         xen_alloc_ptpage(mm, pfn, PT_PTE);
1666 }
1667 
1668 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1669 {
1670         xen_alloc_ptpage(mm, pfn, PT_PMD);
1671 }
1672 
1673 /* This should never happen until we're OK to use struct page */
1674 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1675 {
1676         struct page *page = pfn_to_page(pfn);
1677         bool pinned = PagePinned(page);
1678 
1679         trace_xen_mmu_release_ptpage(pfn, level, pinned);
1680 
1681         if (pinned) {
1682                 if (!PageHighMem(page)) {
1683                         xen_mc_batch();
1684 
1685                         if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1686                                 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1687 
1688                         __set_pfn_prot(pfn, PAGE_KERNEL);
1689 
1690                         xen_mc_issue(PARAVIRT_LAZY_MMU);
1691                 }
1692                 ClearPagePinned(page);
1693         }
1694 }
1695 
1696 static void xen_release_pte(unsigned long pfn)
1697 {
1698         xen_release_ptpage(pfn, PT_PTE);
1699 }
1700 
1701 static void xen_release_pmd(unsigned long pfn)
1702 {
1703         xen_release_ptpage(pfn, PT_PMD);
1704 }
1705 
1706 #ifdef CONFIG_X86_64
1707 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1708 {
1709         xen_alloc_ptpage(mm, pfn, PT_PUD);
1710 }
1711 
1712 static void xen_release_pud(unsigned long pfn)
1713 {
1714         xen_release_ptpage(pfn, PT_PUD);
1715 }
1716 #endif
1717 
1718 void __init xen_reserve_top(void)
1719 {
1720 #ifdef CONFIG_X86_32
1721         unsigned long top = HYPERVISOR_VIRT_START;
1722         struct xen_platform_parameters pp;
1723 
1724         if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1725                 top = pp.virt_start;
1726 
1727         reserve_top_address(-top);
1728 #endif  /* CONFIG_X86_32 */
1729 }
1730 
1731 /*
1732  * Like __va(), but returns address in the kernel mapping (which is
1733  * all we have until the physical memory mapping has been set up.
1734  */
1735 static void * __init __ka(phys_addr_t paddr)
1736 {
1737 #ifdef CONFIG_X86_64
1738         return (void *)(paddr + __START_KERNEL_map);
1739 #else
1740         return __va(paddr);
1741 #endif
1742 }
1743 
1744 /* Convert a machine address to physical address */
1745 static unsigned long __init m2p(phys_addr_t maddr)
1746 {
1747         phys_addr_t paddr;
1748 
1749         maddr &= XEN_PTE_MFN_MASK;
1750         paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1751 
1752         return paddr;
1753 }
1754 
1755 /* Convert a machine address to kernel virtual */
1756 static void * __init m2v(phys_addr_t maddr)
1757 {
1758         return __ka(m2p(maddr));
1759 }
1760 
1761 /* Set the page permissions on an identity-mapped pages */
1762 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1763                                        unsigned long flags)
1764 {
1765         unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1766         pte_t pte = pfn_pte(pfn, prot);
1767 
1768         if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1769                 BUG();
1770 }
1771 static void __init set_page_prot(void *addr, pgprot_t prot)
1772 {
1773         return set_page_prot_flags(addr, prot, UVMF_NONE);
1774 }
1775 #ifdef CONFIG_X86_32
1776 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1777 {
1778         unsigned pmdidx, pteidx;
1779         unsigned ident_pte;
1780         unsigned long pfn;
1781 
1782         level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1783                                       PAGE_SIZE);
1784 
1785         ident_pte = 0;
1786         pfn = 0;
1787         for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1788                 pte_t *pte_page;
1789 
1790                 /* Reuse or allocate a page of ptes */
1791                 if (pmd_present(pmd[pmdidx]))
1792                         pte_page = m2v(pmd[pmdidx].pmd);
1793                 else {
1794                         /* Check for free pte pages */
1795                         if (ident_pte == LEVEL1_IDENT_ENTRIES)
1796                                 break;
1797 
1798                         pte_page = &level1_ident_pgt[ident_pte];
1799                         ident_pte += PTRS_PER_PTE;
1800 
1801                         pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1802                 }
1803 
1804                 /* Install mappings */
1805                 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1806                         pte_t pte;
1807 
1808                         if (pfn > max_pfn_mapped)
1809                                 max_pfn_mapped = pfn;
1810 
1811                         if (!pte_none(pte_page[pteidx]))
1812                                 continue;
1813 
1814                         pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1815                         pte_page[pteidx] = pte;
1816                 }
1817         }
1818 
1819         for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1820                 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1821 
1822         set_page_prot(pmd, PAGE_KERNEL_RO);
1823 }
1824 #endif
1825 void __init xen_setup_machphys_mapping(void)
1826 {
1827         struct xen_machphys_mapping mapping;
1828 
1829         if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1830                 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1831                 machine_to_phys_nr = mapping.max_mfn + 1;
1832         } else {
1833                 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1834         }
1835 #ifdef CONFIG_X86_32
1836         WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1837                 < machine_to_phys_mapping);
1838 #endif
1839 }
1840 
1841 #ifdef CONFIG_X86_64
1842 static void __init convert_pfn_mfn(void *v)
1843 {
1844         pte_t *pte = v;
1845         int i;
1846 
1847         /* All levels are converted the same way, so just treat them
1848            as ptes. */
1849         for (i = 0; i < PTRS_PER_PTE; i++)
1850                 pte[i] = xen_make_pte(pte[i].pte);
1851 }
1852 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1853                                  unsigned long addr)
1854 {
1855         if (*pt_base == PFN_DOWN(__pa(addr))) {
1856                 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1857                 clear_page((void *)addr);
1858                 (*pt_base)++;
1859         }
1860         if (*pt_end == PFN_DOWN(__pa(addr))) {
1861                 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1862                 clear_page((void *)addr);
1863                 (*pt_end)--;
1864         }
1865 }
1866 /*
1867  * Set up the initial kernel pagetable.
1868  *
1869  * We can construct this by grafting the Xen provided pagetable into
1870  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1871  * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1872  * kernel has a physical mapping to start with - but that's enough to
1873  * get __va working.  We need to fill in the rest of the physical
1874  * mapping once some sort of allocator has been set up.
1875  */
1876 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1877 {
1878         pud_t *l3;
1879         pmd_t *l2;
1880         unsigned long addr[3];
1881         unsigned long pt_base, pt_end;
1882         unsigned i;
1883 
1884         /* max_pfn_mapped is the last pfn mapped in the initial memory
1885          * mappings. Considering that on Xen after the kernel mappings we
1886          * have the mappings of some pages that don't exist in pfn space, we
1887          * set max_pfn_mapped to the last real pfn mapped. */
1888         if (xen_start_info->mfn_list < __START_KERNEL_map)
1889                 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1890         else
1891                 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1892 
1893         pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1894         pt_end = pt_base + xen_start_info->nr_pt_frames;
1895 
1896         /* Zap identity mapping */
1897         init_top_pgt[0] = __pgd(0);
1898 
1899         /* Pre-constructed entries are in pfn, so convert to mfn */
1900         /* L4[272] -> level3_ident_pgt  */
1901         /* L4[511] -> level3_kernel_pgt */
1902         convert_pfn_mfn(init_top_pgt);
1903 
1904         /* L3_i[0] -> level2_ident_pgt */
1905         convert_pfn_mfn(level3_ident_pgt);
1906         /* L3_k[510] -> level2_kernel_pgt */
1907         /* L3_k[511] -> level2_fixmap_pgt */
1908         convert_pfn_mfn(level3_kernel_pgt);
1909 
1910         /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1911         convert_pfn_mfn(level2_fixmap_pgt);
1912 
1913         /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1914         l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1915         l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1916 
1917         addr[0] = (unsigned long)pgd;
1918         addr[1] = (unsigned long)l3;
1919         addr[2] = (unsigned long)l2;
1920         /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1921          * Both L4[272][0] and L4[511][510] have entries that point to the same
1922          * L2 (PMD) tables. Meaning that if you modify it in __va space
1923          * it will be also modified in the __ka space! (But if you just
1924          * modify the PMD table to point to other PTE's or none, then you
1925          * are OK - which is what cleanup_highmap does) */
1926         copy_page(level2_ident_pgt, l2);
1927         /* Graft it onto L4[511][510] */
1928         copy_page(level2_kernel_pgt, l2);
1929 
1930         /*
1931          * Zap execute permission from the ident map. Due to the sharing of
1932          * L1 entries we need to do this in the L2.
1933          */
1934         if (__supported_pte_mask & _PAGE_NX) {
1935                 for (i = 0; i < PTRS_PER_PMD; ++i) {
1936                         if (pmd_none(level2_ident_pgt[i]))
1937                                 continue;
1938                         level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1939                 }
1940         }
1941 
1942         /* Copy the initial P->M table mappings if necessary. */
1943         i = pgd_index(xen_start_info->mfn_list);
1944         if (i && i < pgd_index(__START_KERNEL_map))
1945                 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1946 
1947         /* Make pagetable pieces RO */
1948         set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1949         set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1950         set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1951         set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1952         set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1953         set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1954         set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1955 
1956         for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1957                 set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1958                               PAGE_KERNEL_RO);
1959         }
1960 
1961         /* Pin down new L4 */
1962         pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1963                           PFN_DOWN(__pa_symbol(init_top_pgt)));
1964 
1965         /* Unpin Xen-provided one */
1966         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1967 
1968         /*
1969          * At this stage there can be no user pgd, and no page structure to
1970          * attach it to, so make sure we just set kernel pgd.
1971          */
1972         xen_mc_batch();
1973         __xen_write_cr3(true, __pa(init_top_pgt));
1974         xen_mc_issue(PARAVIRT_LAZY_CPU);
1975 
1976         /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1977          * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1978          * the initial domain. For guests using the toolstack, they are in:
1979          * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1980          * rip out the [L4] (pgd), but for guests we shave off three pages.
1981          */
1982         for (i = 0; i < ARRAY_SIZE(addr); i++)
1983                 check_pt_base(&pt_base, &pt_end, addr[i]);
1984 
1985         /* Our (by three pages) smaller Xen pagetable that we are using */
1986         xen_pt_base = PFN_PHYS(pt_base);
1987         xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1988         memblock_reserve(xen_pt_base, xen_pt_size);
1989 
1990         /* Revector the xen_start_info */
1991         xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1992 }
1993 
1994 /*
1995  * Read a value from a physical address.
1996  */
1997 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1998 {
1999         unsigned long *vaddr;
2000         unsigned long val;
2001 
2002         vaddr = early_memremap_ro(addr, sizeof(val));
2003         val = *vaddr;
2004         early_memunmap(vaddr, sizeof(val));
2005         return val;
2006 }
2007 
2008 /*
2009  * Translate a virtual address to a physical one without relying on mapped
2010  * page tables. Don't rely on big pages being aligned in (guest) physical
2011  * space!
2012  */
2013 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
2014 {
2015         phys_addr_t pa;
2016         pgd_t pgd;
2017         pud_t pud;
2018         pmd_t pmd;
2019         pte_t pte;
2020 
2021         pa = read_cr3_pa();
2022         pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
2023                                                        sizeof(pgd)));
2024         if (!pgd_present(pgd))
2025                 return 0;
2026 
2027         pa = pgd_val(pgd) & PTE_PFN_MASK;
2028         pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
2029                                                        sizeof(pud)));
2030         if (!pud_present(pud))
2031                 return 0;
2032         pa = pud_val(pud) & PTE_PFN_MASK;
2033         if (pud_large(pud))
2034                 return pa + (vaddr & ~PUD_MASK);
2035 
2036         pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
2037                                                        sizeof(pmd)));
2038         if (!pmd_present(pmd))
2039                 return 0;
2040         pa = pmd_val(pmd) & PTE_PFN_MASK;
2041         if (pmd_large(pmd))
2042                 return pa + (vaddr & ~PMD_MASK);
2043 
2044         pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
2045                                                        sizeof(pte)));
2046         if (!pte_present(pte))
2047                 return 0;
2048         pa = pte_pfn(pte) << PAGE_SHIFT;
2049 
2050         return pa | (vaddr & ~PAGE_MASK);
2051 }
2052 
2053 /*
2054  * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
2055  * this area.
2056  */
2057 void __init xen_relocate_p2m(void)
2058 {
2059         phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
2060         unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
2061         int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
2062         pte_t *pt;
2063         pmd_t *pmd;
2064         pud_t *pud;
2065         pgd_t *pgd;
2066         unsigned long *new_p2m;
2067 
2068         size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
2069         n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
2070         n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
2071         n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
2072         n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
2073         n_frames = n_pte + n_pt + n_pmd + n_pud;
2074 
2075         new_area = xen_find_free_area(PFN_PHYS(n_frames));
2076         if (!new_area) {
2077                 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
2078                 BUG();
2079         }
2080 
2081         /*
2082          * Setup the page tables for addressing the new p2m list.
2083          * We have asked the hypervisor to map the p2m list at the user address
2084          * PUD_SIZE. It may have done so, or it may have used a kernel space
2085          * address depending on the Xen version.
2086          * To avoid any possible virtual address collision, just use
2087          * 2 * PUD_SIZE for the new area.
2088          */
2089         pud_phys = new_area;
2090         pmd_phys = pud_phys + PFN_PHYS(n_pud);
2091         pt_phys = pmd_phys + PFN_PHYS(n_pmd);
2092         p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
2093 
2094         pgd = __va(read_cr3_pa());
2095         new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
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 RESERVE_BRK(fixup_kernel_pmd, PAGE_SIZE);
2177 RESERVE_BRK(fixup_kernel_pte, PAGE_SIZE);
2178 
2179 static void __init xen_write_cr3_init(unsigned long cr3)
2180 {
2181         unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
2182 
2183         BUG_ON(read_cr3_pa() != __pa(initial_page_table));
2184         BUG_ON(cr3 != __pa(swapper_pg_dir));
2185 
2186         /*
2187          * We are switching to swapper_pg_dir for the first time (from
2188          * initial_page_table) and therefore need to mark that page
2189          * read-only and then pin it.
2190          *
2191          * Xen disallows sharing of kernel PMDs for PAE
2192          * guests. Therefore we must copy the kernel PMD from
2193          * initial_page_table into a new kernel PMD to be used in
2194          * swapper_pg_dir.
2195          */
2196         swapper_kernel_pmd =
2197                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2198         copy_page(swapper_kernel_pmd, initial_kernel_pmd);
2199         swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
2200                 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
2201         set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
2202 
2203         set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
2204         xen_write_cr3(cr3);
2205         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
2206 
2207         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
2208                           PFN_DOWN(__pa(initial_page_table)));
2209         set_page_prot(initial_page_table, PAGE_KERNEL);
2210         set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
2211 
2212         pv_mmu_ops.write_cr3 = &xen_write_cr3;
2213 }
2214 
2215 /*
2216  * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
2217  * not the first page table in the page table pool.
2218  * Iterate through the initial page tables to find the real page table base.
2219  */
2220 static phys_addr_t __init xen_find_pt_base(pmd_t *pmd)
2221 {
2222         phys_addr_t pt_base, paddr;
2223         unsigned pmdidx;
2224 
2225         pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
2226 
2227         for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
2228                 if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
2229                         paddr = m2p(pmd[pmdidx].pmd);
2230                         pt_base = min(pt_base, paddr);
2231                 }
2232 
2233         return pt_base;
2234 }
2235 
2236 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2237 {
2238         pmd_t *kernel_pmd;
2239 
2240         kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2241 
2242         xen_pt_base = xen_find_pt_base(kernel_pmd);
2243         xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
2244 
2245         initial_kernel_pmd =
2246                 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2247 
2248         max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
2249 
2250         copy_page(initial_kernel_pmd, kernel_pmd);
2251 
2252         xen_map_identity_early(initial_kernel_pmd, max_pfn);
2253 
2254         copy_page(initial_page_table, pgd);
2255         initial_page_table[KERNEL_PGD_BOUNDARY] =
2256                 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2257 
2258         set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2259         set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2260         set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2261 
2262         pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2263 
2264         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2265                           PFN_DOWN(__pa(initial_page_table)));
2266         xen_write_cr3(__pa(initial_page_table));
2267 
2268         memblock_reserve(xen_pt_base, xen_pt_size);
2269 }
2270 #endif  /* CONFIG_X86_64 */
2271 
2272 void __init xen_reserve_special_pages(void)
2273 {
2274         phys_addr_t paddr;
2275 
2276         memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2277         if (xen_start_info->store_mfn) {
2278                 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2279                 memblock_reserve(paddr, PAGE_SIZE);
2280         }
2281         if (!xen_initial_domain()) {
2282                 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2283                 memblock_reserve(paddr, PAGE_SIZE);
2284         }
2285 }
2286 
2287 void __init xen_pt_check_e820(void)
2288 {
2289         if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2290                 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2291                 BUG();
2292         }
2293 }
2294 
2295 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2296 
2297 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2298 {
2299         pte_t pte;
2300 
2301         phys >>= PAGE_SHIFT;
2302 
2303         switch (idx) {
2304         case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2305 #ifdef CONFIG_X86_32
2306         case FIX_WP_TEST:
2307 # ifdef CONFIG_HIGHMEM
2308         case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2309 # endif
2310 #elif defined(CONFIG_X86_VSYSCALL_EMULATION)
2311         case VSYSCALL_PAGE:
2312 #endif
2313         case FIX_TEXT_POKE0:
2314         case FIX_TEXT_POKE1:
2315                 /* All local page mappings */
2316                 pte = pfn_pte(phys, prot);
2317                 break;
2318 
2319 #ifdef CONFIG_X86_LOCAL_APIC
2320         case FIX_APIC_BASE:     /* maps dummy local APIC */
2321                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2322                 break;
2323 #endif
2324 
2325 #ifdef CONFIG_X86_IO_APIC
2326         case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2327                 /*
2328                  * We just don't map the IO APIC - all access is via
2329                  * hypercalls.  Keep the address in the pte for reference.
2330                  */
2331                 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2332                 break;
2333 #endif
2334 
2335         case FIX_PARAVIRT_BOOTMAP:
2336                 /* This is an MFN, but it isn't an IO mapping from the
2337                    IO domain */
2338                 pte = mfn_pte(phys, prot);
2339                 break;
2340 
2341         default:
2342                 /* By default, set_fixmap is used for hardware mappings */
2343                 pte = mfn_pte(phys, prot);
2344                 break;
2345         }
2346 
2347         __native_set_fixmap(idx, pte);
2348 
2349 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2350         /* Replicate changes to map the vsyscall page into the user
2351            pagetable vsyscall mapping. */
2352         if (idx == VSYSCALL_PAGE) {
2353                 unsigned long vaddr = __fix_to_virt(idx);
2354                 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2355         }
2356 #endif
2357 }
2358 
2359 static void __init xen_post_allocator_init(void)
2360 {
2361         pv_mmu_ops.set_pte = xen_set_pte;
2362         pv_mmu_ops.set_pmd = xen_set_pmd;
2363         pv_mmu_ops.set_pud = xen_set_pud;
2364 #ifdef CONFIG_X86_64
2365         pv_mmu_ops.set_p4d = xen_set_p4d;
2366 #endif
2367 
2368         /* This will work as long as patching hasn't happened yet
2369            (which it hasn't) */
2370         pv_mmu_ops.alloc_pte = xen_alloc_pte;
2371         pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2372         pv_mmu_ops.release_pte = xen_release_pte;
2373         pv_mmu_ops.release_pmd = xen_release_pmd;
2374 #ifdef CONFIG_X86_64
2375         pv_mmu_ops.alloc_pud = xen_alloc_pud;
2376         pv_mmu_ops.release_pud = xen_release_pud;
2377 #endif
2378         pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2379 
2380 #ifdef CONFIG_X86_64
2381         pv_mmu_ops.write_cr3 = &xen_write_cr3;
2382 #endif
2383 }
2384 
2385 static void xen_leave_lazy_mmu(void)
2386 {
2387         preempt_disable();
2388         xen_mc_flush();
2389         paravirt_leave_lazy_mmu();
2390         preempt_enable();
2391 }
2392 
2393 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2394         .read_cr2 = xen_read_cr2,
2395         .write_cr2 = xen_write_cr2,
2396 
2397         .read_cr3 = xen_read_cr3,
2398         .write_cr3 = xen_write_cr3_init,
2399 
2400         .flush_tlb_user = xen_flush_tlb,
2401         .flush_tlb_kernel = xen_flush_tlb,
2402         .flush_tlb_one_user = xen_flush_tlb_one_user,
2403         .flush_tlb_others = xen_flush_tlb_others,
2404         .tlb_remove_table = tlb_remove_table,
2405 
2406         .pgd_alloc = xen_pgd_alloc,
2407         .pgd_free = xen_pgd_free,
2408 
2409         .alloc_pte = xen_alloc_pte_init,
2410         .release_pte = xen_release_pte_init,
2411         .alloc_pmd = xen_alloc_pmd_init,
2412         .release_pmd = xen_release_pmd_init,
2413 
2414         .set_pte = xen_set_pte_init,
2415         .set_pte_at = xen_set_pte_at,
2416         .set_pmd = xen_set_pmd_hyper,
2417 
2418         .ptep_modify_prot_start = __ptep_modify_prot_start,
2419         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2420 
2421         .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2422         .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2423 
2424         .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2425         .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2426 
2427 #ifdef CONFIG_X86_PAE
2428         .set_pte_atomic = xen_set_pte_atomic,
2429         .pte_clear = xen_pte_clear,
2430         .pmd_clear = xen_pmd_clear,
2431 #endif  /* CONFIG_X86_PAE */
2432         .set_pud = xen_set_pud_hyper,
2433 
2434         .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2435         .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2436 
2437 #ifdef CONFIG_X86_64
2438         .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2439         .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2440         .set_p4d = xen_set_p4d_hyper,
2441 
2442         .alloc_pud = xen_alloc_pmd_init,
2443         .release_pud = xen_release_pmd_init,
2444 
2445 #if CONFIG_PGTABLE_LEVELS >= 5
2446         .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2447         .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2448 #endif
2449 #endif  /* CONFIG_X86_64 */
2450 
2451         .activate_mm = xen_activate_mm,
2452         .dup_mmap = xen_dup_mmap,
2453         .exit_mmap = xen_exit_mmap,
2454 
2455         .lazy_mode = {
2456                 .enter = paravirt_enter_lazy_mmu,
2457                 .leave = xen_leave_lazy_mmu,
2458                 .flush = paravirt_flush_lazy_mmu,
2459         },
2460 
2461         .set_fixmap = xen_set_fixmap,
2462 };
2463 
2464 void __init xen_init_mmu_ops(void)
2465 {
2466         x86_init.paging.pagetable_init = xen_pagetable_init;
2467         x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2468 
2469         pv_mmu_ops = xen_mmu_ops;
2470 
2471         memset(dummy_mapping, 0xff, PAGE_SIZE);
2472 }
2473 
2474 /* Protected by xen_reservation_lock. */
2475 #define MAX_CONTIG_ORDER 9 /* 2MB */
2476 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2477 
2478 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2479 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2480                                 unsigned long *in_frames,
2481                                 unsigned long *out_frames)
2482 {
2483         int i;
2484         struct multicall_space mcs;
2485 
2486         xen_mc_batch();
2487         for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2488                 mcs = __xen_mc_entry(0);
2489 
2490                 if (in_frames)
2491                         in_frames[i] = virt_to_mfn(vaddr);
2492 
2493                 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2494                 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2495 
2496                 if (out_frames)
2497                         out_frames[i] = virt_to_pfn(vaddr);
2498         }
2499         xen_mc_issue(0);
2500 }
2501 
2502 /*
2503  * Update the pfn-to-mfn mappings for a virtual address range, either to
2504  * point to an array of mfns, or contiguously from a single starting
2505  * mfn.
2506  */
2507 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2508                                      unsigned long *mfns,
2509                                      unsigned long first_mfn)
2510 {
2511         unsigned i, limit;
2512         unsigned long mfn;
2513 
2514         xen_mc_batch();
2515 
2516         limit = 1u << order;
2517         for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2518                 struct multicall_space mcs;
2519                 unsigned flags;
2520 
2521                 mcs = __xen_mc_entry(0);
2522                 if (mfns)
2523                         mfn = mfns[i];
2524                 else
2525                         mfn = first_mfn + i;
2526 
2527                 if (i < (limit - 1))
2528                         flags = 0;
2529                 else {
2530                         if (order == 0)
2531                                 flags = UVMF_INVLPG | UVMF_ALL;
2532                         else
2533                                 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2534                 }
2535 
2536                 MULTI_update_va_mapping(mcs.mc, vaddr,
2537                                 mfn_pte(mfn, PAGE_KERNEL), flags);
2538 
2539                 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2540         }
2541 
2542         xen_mc_issue(0);
2543 }
2544 
2545 /*
2546  * Perform the hypercall to exchange a region of our pfns to point to
2547  * memory with the required contiguous alignment.  Takes the pfns as
2548  * input, and populates mfns as output.
2549  *
2550  * Returns a success code indicating whether the hypervisor was able to
2551  * satisfy the request or not.
2552  */
2553 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2554                                unsigned long *pfns_in,
2555                                unsigned long extents_out,
2556                                unsigned int order_out,
2557                                unsigned long *mfns_out,
2558                                unsigned int address_bits)
2559 {
2560         long rc;
2561         int success;
2562 
2563         struct xen_memory_exchange exchange = {
2564                 .in = {
2565                         .nr_extents   = extents_in,
2566                         .extent_order = order_in,
2567                         .extent_start = pfns_in,
2568                         .domid        = DOMID_SELF
2569                 },
2570                 .out = {
2571                         .nr_extents   = extents_out,
2572                         .extent_order = order_out,
2573                         .extent_start = mfns_out,
2574                         .address_bits = address_bits,
2575                         .domid        = DOMID_SELF
2576                 }
2577         };
2578 
2579         BUG_ON(extents_in << order_in != extents_out << order_out);
2580 
2581         rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2582         success = (exchange.nr_exchanged == extents_in);
2583 
2584         BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2585         BUG_ON(success && (rc != 0));
2586 
2587         return success;
2588 }
2589 
2590 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2591                                  unsigned int address_bits,
2592                                  dma_addr_t *dma_handle)
2593 {
2594         unsigned long *in_frames = discontig_frames, out_frame;
2595         unsigned long  flags;
2596         int            success;
2597         unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2598 
2599         /*
2600          * Currently an auto-translated guest will not perform I/O, nor will
2601          * it require PAE page directories below 4GB. Therefore any calls to
2602          * this function are redundant and can be ignored.
2603          */
2604 
2605         if (unlikely(order > MAX_CONTIG_ORDER))
2606                 return -ENOMEM;
2607 
2608         memset((void *) vstart, 0, PAGE_SIZE << order);
2609 
2610         spin_lock_irqsave(&xen_reservation_lock, flags);
2611 
2612         /* 1. Zap current PTEs, remembering MFNs. */
2613         xen_zap_pfn_range(vstart, order, in_frames, NULL);
2614 
2615         /* 2. Get a new contiguous memory extent. */
2616         out_frame = virt_to_pfn(vstart);
2617         success = xen_exchange_memory(1UL << order, 0, in_frames,
2618                                       1, order, &out_frame,
2619                                       address_bits);
2620 
2621         /* 3. Map the new extent in place of old pages. */
2622         if (success)
2623                 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2624         else
2625                 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2626 
2627         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2628 
2629         *dma_handle = virt_to_machine(vstart).maddr;
2630         return success ? 0 : -ENOMEM;
2631 }
2632 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2633 
2634 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2635 {
2636         unsigned long *out_frames = discontig_frames, in_frame;
2637         unsigned long  flags;
2638         int success;
2639         unsigned long vstart;
2640 
2641         if (unlikely(order > MAX_CONTIG_ORDER))
2642                 return;
2643 
2644         vstart = (unsigned long)phys_to_virt(pstart);
2645         memset((void *) vstart, 0, PAGE_SIZE << order);
2646 
2647         spin_lock_irqsave(&xen_reservation_lock, flags);
2648 
2649         /* 1. Find start MFN of contiguous extent. */
2650         in_frame = virt_to_mfn(vstart);
2651 
2652         /* 2. Zap current PTEs. */
2653         xen_zap_pfn_range(vstart, order, NULL, out_frames);
2654 
2655         /* 3. Do the exchange for non-contiguous MFNs. */
2656         success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2657                                         0, out_frames, 0);
2658 
2659         /* 4. Map new pages in place of old pages. */
2660         if (success)
2661                 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2662         else
2663                 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2664 
2665         spin_unlock_irqrestore(&xen_reservation_lock, flags);
2666 }
2667 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2668 
2669 #ifdef CONFIG_KEXEC_CORE
2670 phys_addr_t paddr_vmcoreinfo_note(void)
2671 {
2672         if (xen_pv_domain())
2673                 return virt_to_machine(vmcoreinfo_note).maddr;
2674         else
2675                 return __pa(vmcoreinfo_note);
2676 }
2677 #endif /* CONFIG_KEXEC_CORE */
2678 

~ [ 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