xref: /openbmc/linux/arch/x86/xen/mmu_pv.c (revision aa74c44b)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Xen mmu operations
5  *
6  * This file contains the various mmu fetch and update operations.
7  * The most important job they must perform is the mapping between the
8  * domain's pfn and the overall machine mfns.
9  *
10  * Xen allows guests to directly update the pagetable, in a controlled
11  * fashion.  In other words, the guest modifies the same pagetable
12  * that the CPU actually uses, which eliminates the overhead of having
13  * a separate shadow pagetable.
14  *
15  * In order to allow this, it falls on the guest domain to map its
16  * notion of a "physical" pfn - which is just a domain-local linear
17  * address - into a real "machine address" which the CPU's MMU can
18  * use.
19  *
20  * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
21  * inserted directly into the pagetable.  When creating a new
22  * pte/pmd/pgd, it converts the passed pfn into an mfn.  Conversely,
23  * when reading the content back with __(pgd|pmd|pte)_val, it converts
24  * the mfn back into a pfn.
25  *
26  * The other constraint is that all pages which make up a pagetable
27  * must be mapped read-only in the guest.  This prevents uncontrolled
28  * guest updates to the pagetable.  Xen strictly enforces this, and
29  * will disallow any pagetable update which will end up mapping a
30  * pagetable page RW, and will disallow using any writable page as a
31  * pagetable.
32  *
33  * Naively, when loading %cr3 with the base of a new pagetable, Xen
34  * would need to validate the whole pagetable before going on.
35  * Naturally, this is quite slow.  The solution is to "pin" a
36  * pagetable, which enforces all the constraints on the pagetable even
37  * when it is not actively in use.  This menas that Xen can be assured
38  * that it is still valid when you do load it into %cr3, and doesn't
39  * need to revalidate it.
40  *
41  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
42  */
43 #include <linux/sched/mm.h>
44 #include <linux/debugfs.h>
45 #include <linux/bug.h>
46 #include <linux/vmalloc.h>
47 #include <linux/export.h>
48 #include <linux/init.h>
49 #include <linux/gfp.h>
50 #include <linux/memblock.h>
51 #include <linux/seq_file.h>
52 #include <linux/crash_dump.h>
53 #include <linux/pgtable.h>
54 #ifdef CONFIG_KEXEC_CORE
55 #include <linux/kexec.h>
56 #endif
57 
58 #include <trace/events/xen.h>
59 
60 #include <asm/tlbflush.h>
61 #include <asm/fixmap.h>
62 #include <asm/mmu_context.h>
63 #include <asm/setup.h>
64 #include <asm/paravirt.h>
65 #include <asm/e820/api.h>
66 #include <asm/linkage.h>
67 #include <asm/page.h>
68 #include <asm/init.h>
69 #include <asm/memtype.h>
70 #include <asm/smp.h>
71 #include <asm/tlb.h>
72 
73 #include <asm/xen/hypercall.h>
74 #include <asm/xen/hypervisor.h>
75 
76 #include <xen/xen.h>
77 #include <xen/page.h>
78 #include <xen/interface/xen.h>
79 #include <xen/interface/hvm/hvm_op.h>
80 #include <xen/interface/version.h>
81 #include <xen/interface/memory.h>
82 #include <xen/hvc-console.h>
83 
84 #include "multicalls.h"
85 #include "mmu.h"
86 #include "debugfs.h"
87 
88 #ifdef CONFIG_X86_VSYSCALL_EMULATION
89 /* l3 pud for userspace vsyscall mapping */
90 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
91 #endif
92 
93 /*
94  * Protects atomic reservation decrease/increase against concurrent increases.
95  * Also protects non-atomic updates of current_pages and balloon lists.
96  */
97 static DEFINE_SPINLOCK(xen_reservation_lock);
98 
99 /*
100  * Note about cr3 (pagetable base) values:
101  *
102  * xen_cr3 contains the current logical cr3 value; it contains the
103  * last set cr3.  This may not be the current effective cr3, because
104  * its update may be being lazily deferred.  However, a vcpu looking
105  * at its own cr3 can use this value knowing that it everything will
106  * be self-consistent.
107  *
108  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
109  * hypercall to set the vcpu cr3 is complete (so it may be a little
110  * out of date, but it will never be set early).  If one vcpu is
111  * looking at another vcpu's cr3 value, it should use this variable.
112  */
113 DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
114 DEFINE_PER_CPU(unsigned long, xen_current_cr3);	 /* actual vcpu cr3 */
115 
116 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
117 
118 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
119 
120 /*
121  * Just beyond the highest usermode address.  STACK_TOP_MAX has a
122  * redzone above it, so round it up to a PGD boundary.
123  */
124 #define USER_LIMIT	((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
125 
126 void make_lowmem_page_readonly(void *vaddr)
127 {
128 	pte_t *pte, ptev;
129 	unsigned long address = (unsigned long)vaddr;
130 	unsigned int level;
131 
132 	pte = lookup_address(address, &level);
133 	if (pte == NULL)
134 		return;		/* vaddr missing */
135 
136 	ptev = pte_wrprotect(*pte);
137 
138 	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
139 		BUG();
140 }
141 
142 void make_lowmem_page_readwrite(void *vaddr)
143 {
144 	pte_t *pte, ptev;
145 	unsigned long address = (unsigned long)vaddr;
146 	unsigned int level;
147 
148 	pte = lookup_address(address, &level);
149 	if (pte == NULL)
150 		return;		/* vaddr missing */
151 
152 	ptev = pte_mkwrite(*pte);
153 
154 	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
155 		BUG();
156 }
157 
158 
159 /*
160  * During early boot all page table pages are pinned, but we do not have struct
161  * pages, so return true until struct pages are ready.
162  */
163 static bool xen_page_pinned(void *ptr)
164 {
165 	if (static_branch_likely(&xen_struct_pages_ready)) {
166 		struct page *page = virt_to_page(ptr);
167 
168 		return PagePinned(page);
169 	}
170 	return true;
171 }
172 
173 static void xen_extend_mmu_update(const struct mmu_update *update)
174 {
175 	struct multicall_space mcs;
176 	struct mmu_update *u;
177 
178 	mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
179 
180 	if (mcs.mc != NULL) {
181 		mcs.mc->args[1]++;
182 	} else {
183 		mcs = __xen_mc_entry(sizeof(*u));
184 		MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
185 	}
186 
187 	u = mcs.args;
188 	*u = *update;
189 }
190 
191 static void xen_extend_mmuext_op(const struct mmuext_op *op)
192 {
193 	struct multicall_space mcs;
194 	struct mmuext_op *u;
195 
196 	mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
197 
198 	if (mcs.mc != NULL) {
199 		mcs.mc->args[1]++;
200 	} else {
201 		mcs = __xen_mc_entry(sizeof(*u));
202 		MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
203 	}
204 
205 	u = mcs.args;
206 	*u = *op;
207 }
208 
209 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
210 {
211 	struct mmu_update u;
212 
213 	preempt_disable();
214 
215 	xen_mc_batch();
216 
217 	/* ptr may be ioremapped for 64-bit pagetable setup */
218 	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
219 	u.val = pmd_val_ma(val);
220 	xen_extend_mmu_update(&u);
221 
222 	xen_mc_issue(PARAVIRT_LAZY_MMU);
223 
224 	preempt_enable();
225 }
226 
227 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
228 {
229 	trace_xen_mmu_set_pmd(ptr, val);
230 
231 	/* If page is not pinned, we can just update the entry
232 	   directly */
233 	if (!xen_page_pinned(ptr)) {
234 		*ptr = val;
235 		return;
236 	}
237 
238 	xen_set_pmd_hyper(ptr, val);
239 }
240 
241 /*
242  * Associate a virtual page frame with a given physical page frame
243  * and protection flags for that frame.
244  */
245 void __init set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
246 {
247 	if (HYPERVISOR_update_va_mapping(vaddr, mfn_pte(mfn, flags),
248 					 UVMF_INVLPG))
249 		BUG();
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 a hypercall is much cheaper.
276 		 */
277 		struct mmu_update u;
278 
279 		u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
280 		u.val = pte_val_ma(pteval);
281 		HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
282 	}
283 }
284 
285 static void xen_set_pte(pte_t *ptep, pte_t pteval)
286 {
287 	trace_xen_mmu_set_pte(ptep, pteval);
288 	__xen_set_pte(ptep, pteval);
289 }
290 
291 pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
292 				 unsigned long addr, pte_t *ptep)
293 {
294 	/* Just return the pte as-is.  We preserve the bits on commit */
295 	trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
296 	return *ptep;
297 }
298 
299 void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
300 				 pte_t *ptep, pte_t pte)
301 {
302 	struct mmu_update u;
303 
304 	trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
305 	xen_mc_batch();
306 
307 	u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
308 	u.val = pte_val_ma(pte);
309 	xen_extend_mmu_update(&u);
310 
311 	xen_mc_issue(PARAVIRT_LAZY_MMU);
312 }
313 
314 /* Assume pteval_t is equivalent to all the other *val_t types. */
315 static pteval_t pte_mfn_to_pfn(pteval_t val)
316 {
317 	if (val & _PAGE_PRESENT) {
318 		unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
319 		unsigned long pfn = mfn_to_pfn(mfn);
320 
321 		pteval_t flags = val & PTE_FLAGS_MASK;
322 		if (unlikely(pfn == ~0))
323 			val = flags & ~_PAGE_PRESENT;
324 		else
325 			val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
326 	}
327 
328 	return val;
329 }
330 
331 static pteval_t pte_pfn_to_mfn(pteval_t val)
332 {
333 	if (val & _PAGE_PRESENT) {
334 		unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
335 		pteval_t flags = val & PTE_FLAGS_MASK;
336 		unsigned long mfn;
337 
338 		mfn = __pfn_to_mfn(pfn);
339 
340 		/*
341 		 * If there's no mfn for the pfn, then just create an
342 		 * empty non-present pte.  Unfortunately this loses
343 		 * information about the original pfn, so
344 		 * pte_mfn_to_pfn is asymmetric.
345 		 */
346 		if (unlikely(mfn == INVALID_P2M_ENTRY)) {
347 			mfn = 0;
348 			flags = 0;
349 		} else
350 			mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
351 		val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
352 	}
353 
354 	return val;
355 }
356 
357 __visible pteval_t xen_pte_val(pte_t pte)
358 {
359 	pteval_t pteval = pte.pte;
360 
361 	return pte_mfn_to_pfn(pteval);
362 }
363 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
364 
365 __visible pgdval_t xen_pgd_val(pgd_t pgd)
366 {
367 	return pte_mfn_to_pfn(pgd.pgd);
368 }
369 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
370 
371 __visible pte_t xen_make_pte(pteval_t pte)
372 {
373 	pte = pte_pfn_to_mfn(pte);
374 
375 	return native_make_pte(pte);
376 }
377 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
378 
379 __visible pgd_t xen_make_pgd(pgdval_t pgd)
380 {
381 	pgd = pte_pfn_to_mfn(pgd);
382 	return native_make_pgd(pgd);
383 }
384 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
385 
386 __visible pmdval_t xen_pmd_val(pmd_t pmd)
387 {
388 	return pte_mfn_to_pfn(pmd.pmd);
389 }
390 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
391 
392 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
393 {
394 	struct mmu_update u;
395 
396 	preempt_disable();
397 
398 	xen_mc_batch();
399 
400 	/* ptr may be ioremapped for 64-bit pagetable setup */
401 	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
402 	u.val = pud_val_ma(val);
403 	xen_extend_mmu_update(&u);
404 
405 	xen_mc_issue(PARAVIRT_LAZY_MMU);
406 
407 	preempt_enable();
408 }
409 
410 static void xen_set_pud(pud_t *ptr, pud_t val)
411 {
412 	trace_xen_mmu_set_pud(ptr, val);
413 
414 	/* If page is not pinned, we can just update the entry
415 	   directly */
416 	if (!xen_page_pinned(ptr)) {
417 		*ptr = val;
418 		return;
419 	}
420 
421 	xen_set_pud_hyper(ptr, val);
422 }
423 
424 __visible pmd_t xen_make_pmd(pmdval_t pmd)
425 {
426 	pmd = pte_pfn_to_mfn(pmd);
427 	return native_make_pmd(pmd);
428 }
429 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
430 
431 __visible pudval_t xen_pud_val(pud_t pud)
432 {
433 	return pte_mfn_to_pfn(pud.pud);
434 }
435 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
436 
437 __visible pud_t xen_make_pud(pudval_t pud)
438 {
439 	pud = pte_pfn_to_mfn(pud);
440 
441 	return native_make_pud(pud);
442 }
443 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
444 
445 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
446 {
447 	pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
448 	unsigned offset = pgd - pgd_page;
449 	pgd_t *user_ptr = NULL;
450 
451 	if (offset < pgd_index(USER_LIMIT)) {
452 		struct page *page = virt_to_page(pgd_page);
453 		user_ptr = (pgd_t *)page->private;
454 		if (user_ptr)
455 			user_ptr += offset;
456 	}
457 
458 	return user_ptr;
459 }
460 
461 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
462 {
463 	struct mmu_update u;
464 
465 	u.ptr = virt_to_machine(ptr).maddr;
466 	u.val = p4d_val_ma(val);
467 	xen_extend_mmu_update(&u);
468 }
469 
470 /*
471  * Raw hypercall-based set_p4d, intended for in early boot before
472  * there's a page structure.  This implies:
473  *  1. The only existing pagetable is the kernel's
474  *  2. It is always pinned
475  *  3. It has no user pagetable attached to it
476  */
477 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
478 {
479 	preempt_disable();
480 
481 	xen_mc_batch();
482 
483 	__xen_set_p4d_hyper(ptr, val);
484 
485 	xen_mc_issue(PARAVIRT_LAZY_MMU);
486 
487 	preempt_enable();
488 }
489 
490 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
491 {
492 	pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
493 	pgd_t pgd_val;
494 
495 	trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
496 
497 	/* If page is not pinned, we can just update the entry
498 	   directly */
499 	if (!xen_page_pinned(ptr)) {
500 		*ptr = val;
501 		if (user_ptr) {
502 			WARN_ON(xen_page_pinned(user_ptr));
503 			pgd_val.pgd = p4d_val_ma(val);
504 			*user_ptr = pgd_val;
505 		}
506 		return;
507 	}
508 
509 	/* If it's pinned, then we can at least batch the kernel and
510 	   user updates together. */
511 	xen_mc_batch();
512 
513 	__xen_set_p4d_hyper(ptr, val);
514 	if (user_ptr)
515 		__xen_set_p4d_hyper((p4d_t *)user_ptr, val);
516 
517 	xen_mc_issue(PARAVIRT_LAZY_MMU);
518 }
519 
520 #if CONFIG_PGTABLE_LEVELS >= 5
521 __visible p4dval_t xen_p4d_val(p4d_t p4d)
522 {
523 	return pte_mfn_to_pfn(p4d.p4d);
524 }
525 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
526 
527 __visible p4d_t xen_make_p4d(p4dval_t p4d)
528 {
529 	p4d = pte_pfn_to_mfn(p4d);
530 
531 	return native_make_p4d(p4d);
532 }
533 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
534 #endif  /* CONFIG_PGTABLE_LEVELS >= 5 */
535 
536 static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
537 			 void (*func)(struct mm_struct *mm, struct page *,
538 				      enum pt_level),
539 			 bool last, unsigned long limit)
540 {
541 	int i, nr;
542 
543 	nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
544 	for (i = 0; i < nr; i++) {
545 		if (!pmd_none(pmd[i]))
546 			(*func)(mm, pmd_page(pmd[i]), PT_PTE);
547 	}
548 }
549 
550 static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
551 			 void (*func)(struct mm_struct *mm, struct page *,
552 				      enum pt_level),
553 			 bool last, unsigned long limit)
554 {
555 	int i, nr;
556 
557 	nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
558 	for (i = 0; i < nr; i++) {
559 		pmd_t *pmd;
560 
561 		if (pud_none(pud[i]))
562 			continue;
563 
564 		pmd = pmd_offset(&pud[i], 0);
565 		if (PTRS_PER_PMD > 1)
566 			(*func)(mm, virt_to_page(pmd), PT_PMD);
567 		xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
568 	}
569 }
570 
571 static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
572 			 void (*func)(struct mm_struct *mm, struct page *,
573 				      enum pt_level),
574 			 bool last, unsigned long limit)
575 {
576 	pud_t *pud;
577 
578 
579 	if (p4d_none(*p4d))
580 		return;
581 
582 	pud = pud_offset(p4d, 0);
583 	if (PTRS_PER_PUD > 1)
584 		(*func)(mm, virt_to_page(pud), PT_PUD);
585 	xen_pud_walk(mm, pud, func, last, limit);
586 }
587 
588 /*
589  * (Yet another) pagetable walker.  This one is intended for pinning a
590  * pagetable.  This means that it walks a pagetable and calls the
591  * callback function on each page it finds making up the page table,
592  * at every level.  It walks the entire pagetable, but it only bothers
593  * pinning pte pages which are below limit.  In the normal case this
594  * will be STACK_TOP_MAX, but at boot we need to pin up to
595  * FIXADDR_TOP.
596  *
597  * We must skip the Xen hole in the middle of the address space, just after
598  * the big x86-64 virtual hole.
599  */
600 static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
601 			   void (*func)(struct mm_struct *mm, struct page *,
602 					enum pt_level),
603 			   unsigned long limit)
604 {
605 	int i, nr;
606 	unsigned hole_low = 0, hole_high = 0;
607 
608 	/* The limit is the last byte to be touched */
609 	limit--;
610 	BUG_ON(limit >= FIXADDR_TOP);
611 
612 	/*
613 	 * 64-bit has a great big hole in the middle of the address
614 	 * space, which contains the Xen mappings.
615 	 */
616 	hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
617 	hole_high = pgd_index(GUARD_HOLE_END_ADDR);
618 
619 	nr = pgd_index(limit) + 1;
620 	for (i = 0; i < nr; i++) {
621 		p4d_t *p4d;
622 
623 		if (i >= hole_low && i < hole_high)
624 			continue;
625 
626 		if (pgd_none(pgd[i]))
627 			continue;
628 
629 		p4d = p4d_offset(&pgd[i], 0);
630 		xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
631 	}
632 
633 	/* Do the top level last, so that the callbacks can use it as
634 	   a cue to do final things like tlb flushes. */
635 	(*func)(mm, virt_to_page(pgd), PT_PGD);
636 }
637 
638 static void xen_pgd_walk(struct mm_struct *mm,
639 			 void (*func)(struct mm_struct *mm, struct page *,
640 				      enum pt_level),
641 			 unsigned long limit)
642 {
643 	__xen_pgd_walk(mm, mm->pgd, func, limit);
644 }
645 
646 /* If we're using split pte locks, then take the page's lock and
647    return a pointer to it.  Otherwise return NULL. */
648 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
649 {
650 	spinlock_t *ptl = NULL;
651 
652 #if USE_SPLIT_PTE_PTLOCKS
653 	ptl = ptlock_ptr(page);
654 	spin_lock_nest_lock(ptl, &mm->page_table_lock);
655 #endif
656 
657 	return ptl;
658 }
659 
660 static void xen_pte_unlock(void *v)
661 {
662 	spinlock_t *ptl = v;
663 	spin_unlock(ptl);
664 }
665 
666 static void xen_do_pin(unsigned level, unsigned long pfn)
667 {
668 	struct mmuext_op op;
669 
670 	op.cmd = level;
671 	op.arg1.mfn = pfn_to_mfn(pfn);
672 
673 	xen_extend_mmuext_op(&op);
674 }
675 
676 static void xen_pin_page(struct mm_struct *mm, struct page *page,
677 			 enum pt_level level)
678 {
679 	unsigned pgfl = TestSetPagePinned(page);
680 
681 	if (!pgfl) {
682 		void *pt = lowmem_page_address(page);
683 		unsigned long pfn = page_to_pfn(page);
684 		struct multicall_space mcs = __xen_mc_entry(0);
685 		spinlock_t *ptl;
686 
687 		/*
688 		 * We need to hold the pagetable lock between the time
689 		 * we make the pagetable RO and when we actually pin
690 		 * it.  If we don't, then other users may come in and
691 		 * attempt to update the pagetable by writing it,
692 		 * which will fail because the memory is RO but not
693 		 * pinned, so Xen won't do the trap'n'emulate.
694 		 *
695 		 * If we're using split pte locks, we can't hold the
696 		 * entire pagetable's worth of locks during the
697 		 * traverse, because we may wrap the preempt count (8
698 		 * bits).  The solution is to mark RO and pin each PTE
699 		 * page while holding the lock.  This means the number
700 		 * of locks we end up holding is never more than a
701 		 * batch size (~32 entries, at present).
702 		 *
703 		 * If we're not using split pte locks, we needn't pin
704 		 * the PTE pages independently, because we're
705 		 * protected by the overall pagetable lock.
706 		 */
707 		ptl = NULL;
708 		if (level == PT_PTE)
709 			ptl = xen_pte_lock(page, mm);
710 
711 		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
712 					pfn_pte(pfn, PAGE_KERNEL_RO),
713 					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
714 
715 		if (ptl) {
716 			xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
717 
718 			/* Queue a deferred unlock for when this batch
719 			   is completed. */
720 			xen_mc_callback(xen_pte_unlock, ptl);
721 		}
722 	}
723 }
724 
725 /* This is called just after a mm has been created, but it has not
726    been used yet.  We need to make sure that its pagetable is all
727    read-only, and can be pinned. */
728 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
729 {
730 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
731 
732 	trace_xen_mmu_pgd_pin(mm, pgd);
733 
734 	xen_mc_batch();
735 
736 	__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
737 
738 	xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
739 
740 	if (user_pgd) {
741 		xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
742 		xen_do_pin(MMUEXT_PIN_L4_TABLE,
743 			   PFN_DOWN(__pa(user_pgd)));
744 	}
745 
746 	xen_mc_issue(0);
747 }
748 
749 static void xen_pgd_pin(struct mm_struct *mm)
750 {
751 	__xen_pgd_pin(mm, mm->pgd);
752 }
753 
754 /*
755  * On save, we need to pin all pagetables to make sure they get their
756  * mfns turned into pfns.  Search the list for any unpinned pgds and pin
757  * them (unpinned pgds are not currently in use, probably because the
758  * process is under construction or destruction).
759  *
760  * Expected to be called in stop_machine() ("equivalent to taking
761  * every spinlock in the system"), so the locking doesn't really
762  * matter all that much.
763  */
764 void xen_mm_pin_all(void)
765 {
766 	struct page *page;
767 
768 	spin_lock(&pgd_lock);
769 
770 	list_for_each_entry(page, &pgd_list, lru) {
771 		if (!PagePinned(page)) {
772 			__xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
773 			SetPageSavePinned(page);
774 		}
775 	}
776 
777 	spin_unlock(&pgd_lock);
778 }
779 
780 static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
781 				   enum pt_level level)
782 {
783 	SetPagePinned(page);
784 }
785 
786 /*
787  * The init_mm pagetable is really pinned as soon as its created, but
788  * that's before we have page structures to store the bits.  So do all
789  * the book-keeping now once struct pages for allocated pages are
790  * initialized. This happens only after memblock_free_all() is called.
791  */
792 static void __init xen_after_bootmem(void)
793 {
794 	static_branch_enable(&xen_struct_pages_ready);
795 #ifdef CONFIG_X86_VSYSCALL_EMULATION
796 	SetPagePinned(virt_to_page(level3_user_vsyscall));
797 #endif
798 	xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
799 }
800 
801 static void xen_unpin_page(struct mm_struct *mm, struct page *page,
802 			   enum pt_level level)
803 {
804 	unsigned pgfl = TestClearPagePinned(page);
805 
806 	if (pgfl) {
807 		void *pt = lowmem_page_address(page);
808 		unsigned long pfn = page_to_pfn(page);
809 		spinlock_t *ptl = NULL;
810 		struct multicall_space mcs;
811 
812 		/*
813 		 * Do the converse to pin_page.  If we're using split
814 		 * pte locks, we must be holding the lock for while
815 		 * the pte page is unpinned but still RO to prevent
816 		 * concurrent updates from seeing it in this
817 		 * partially-pinned state.
818 		 */
819 		if (level == PT_PTE) {
820 			ptl = xen_pte_lock(page, mm);
821 
822 			if (ptl)
823 				xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
824 		}
825 
826 		mcs = __xen_mc_entry(0);
827 
828 		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
829 					pfn_pte(pfn, PAGE_KERNEL),
830 					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
831 
832 		if (ptl) {
833 			/* unlock when batch completed */
834 			xen_mc_callback(xen_pte_unlock, ptl);
835 		}
836 	}
837 }
838 
839 /* Release a pagetables pages back as normal RW */
840 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
841 {
842 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
843 
844 	trace_xen_mmu_pgd_unpin(mm, pgd);
845 
846 	xen_mc_batch();
847 
848 	xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
849 
850 	if (user_pgd) {
851 		xen_do_pin(MMUEXT_UNPIN_TABLE,
852 			   PFN_DOWN(__pa(user_pgd)));
853 		xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
854 	}
855 
856 	__xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
857 
858 	xen_mc_issue(0);
859 }
860 
861 static void xen_pgd_unpin(struct mm_struct *mm)
862 {
863 	__xen_pgd_unpin(mm, mm->pgd);
864 }
865 
866 /*
867  * On resume, undo any pinning done at save, so that the rest of the
868  * kernel doesn't see any unexpected pinned pagetables.
869  */
870 void xen_mm_unpin_all(void)
871 {
872 	struct page *page;
873 
874 	spin_lock(&pgd_lock);
875 
876 	list_for_each_entry(page, &pgd_list, lru) {
877 		if (PageSavePinned(page)) {
878 			BUG_ON(!PagePinned(page));
879 			__xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
880 			ClearPageSavePinned(page);
881 		}
882 	}
883 
884 	spin_unlock(&pgd_lock);
885 }
886 
887 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
888 {
889 	spin_lock(&next->page_table_lock);
890 	xen_pgd_pin(next);
891 	spin_unlock(&next->page_table_lock);
892 }
893 
894 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
895 {
896 	spin_lock(&mm->page_table_lock);
897 	xen_pgd_pin(mm);
898 	spin_unlock(&mm->page_table_lock);
899 }
900 
901 static void drop_mm_ref_this_cpu(void *info)
902 {
903 	struct mm_struct *mm = info;
904 
905 	if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
906 		leave_mm(smp_processor_id());
907 
908 	/*
909 	 * If this cpu still has a stale cr3 reference, then make sure
910 	 * it has been flushed.
911 	 */
912 	if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
913 		xen_mc_flush();
914 }
915 
916 #ifdef CONFIG_SMP
917 /*
918  * Another cpu may still have their %cr3 pointing at the pagetable, so
919  * we need to repoint it somewhere else before we can unpin it.
920  */
921 static void xen_drop_mm_ref(struct mm_struct *mm)
922 {
923 	cpumask_var_t mask;
924 	unsigned cpu;
925 
926 	drop_mm_ref_this_cpu(mm);
927 
928 	/* Get the "official" set of cpus referring to our pagetable. */
929 	if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
930 		for_each_online_cpu(cpu) {
931 			if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
932 				continue;
933 			smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
934 		}
935 		return;
936 	}
937 
938 	/*
939 	 * It's possible that a vcpu may have a stale reference to our
940 	 * cr3, because its in lazy mode, and it hasn't yet flushed
941 	 * its set of pending hypercalls yet.  In this case, we can
942 	 * look at its actual current cr3 value, and force it to flush
943 	 * if needed.
944 	 */
945 	cpumask_clear(mask);
946 	for_each_online_cpu(cpu) {
947 		if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
948 			cpumask_set_cpu(cpu, mask);
949 	}
950 
951 	smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
952 	free_cpumask_var(mask);
953 }
954 #else
955 static void xen_drop_mm_ref(struct mm_struct *mm)
956 {
957 	drop_mm_ref_this_cpu(mm);
958 }
959 #endif
960 
961 /*
962  * While a process runs, Xen pins its pagetables, which means that the
963  * hypervisor forces it to be read-only, and it controls all updates
964  * to it.  This means that all pagetable updates have to go via the
965  * hypervisor, which is moderately expensive.
966  *
967  * Since we're pulling the pagetable down, we switch to use init_mm,
968  * unpin old process pagetable and mark it all read-write, which
969  * allows further operations on it to be simple memory accesses.
970  *
971  * The only subtle point is that another CPU may be still using the
972  * pagetable because of lazy tlb flushing.  This means we need need to
973  * switch all CPUs off this pagetable before we can unpin it.
974  */
975 static void xen_exit_mmap(struct mm_struct *mm)
976 {
977 	get_cpu();		/* make sure we don't move around */
978 	xen_drop_mm_ref(mm);
979 	put_cpu();
980 
981 	spin_lock(&mm->page_table_lock);
982 
983 	/* pgd may not be pinned in the error exit path of execve */
984 	if (xen_page_pinned(mm->pgd))
985 		xen_pgd_unpin(mm);
986 
987 	spin_unlock(&mm->page_table_lock);
988 }
989 
990 static void xen_post_allocator_init(void);
991 
992 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
993 {
994 	struct mmuext_op op;
995 
996 	op.cmd = cmd;
997 	op.arg1.mfn = pfn_to_mfn(pfn);
998 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
999 		BUG();
1000 }
1001 
1002 static void __init xen_cleanhighmap(unsigned long vaddr,
1003 				    unsigned long vaddr_end)
1004 {
1005 	unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1006 	pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1007 
1008 	/* NOTE: The loop is more greedy than the cleanup_highmap variant.
1009 	 * We include the PMD passed in on _both_ boundaries. */
1010 	for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1011 			pmd++, vaddr += PMD_SIZE) {
1012 		if (pmd_none(*pmd))
1013 			continue;
1014 		if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1015 			set_pmd(pmd, __pmd(0));
1016 	}
1017 	/* In case we did something silly, we should crash in this function
1018 	 * instead of somewhere later and be confusing. */
1019 	xen_mc_flush();
1020 }
1021 
1022 /*
1023  * Make a page range writeable and free it.
1024  */
1025 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1026 {
1027 	void *vaddr = __va(paddr);
1028 	void *vaddr_end = vaddr + size;
1029 
1030 	for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1031 		make_lowmem_page_readwrite(vaddr);
1032 
1033 	memblock_phys_free(paddr, size);
1034 }
1035 
1036 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1037 {
1038 	unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1039 
1040 	if (unpin)
1041 		pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1042 	ClearPagePinned(virt_to_page(__va(pa)));
1043 	xen_free_ro_pages(pa, PAGE_SIZE);
1044 }
1045 
1046 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1047 {
1048 	unsigned long pa;
1049 	pte_t *pte_tbl;
1050 	int i;
1051 
1052 	if (pmd_large(*pmd)) {
1053 		pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1054 		xen_free_ro_pages(pa, PMD_SIZE);
1055 		return;
1056 	}
1057 
1058 	pte_tbl = pte_offset_kernel(pmd, 0);
1059 	for (i = 0; i < PTRS_PER_PTE; i++) {
1060 		if (pte_none(pte_tbl[i]))
1061 			continue;
1062 		pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1063 		xen_free_ro_pages(pa, PAGE_SIZE);
1064 	}
1065 	set_pmd(pmd, __pmd(0));
1066 	xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1067 }
1068 
1069 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1070 {
1071 	unsigned long pa;
1072 	pmd_t *pmd_tbl;
1073 	int i;
1074 
1075 	if (pud_large(*pud)) {
1076 		pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1077 		xen_free_ro_pages(pa, PUD_SIZE);
1078 		return;
1079 	}
1080 
1081 	pmd_tbl = pmd_offset(pud, 0);
1082 	for (i = 0; i < PTRS_PER_PMD; i++) {
1083 		if (pmd_none(pmd_tbl[i]))
1084 			continue;
1085 		xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1086 	}
1087 	set_pud(pud, __pud(0));
1088 	xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1089 }
1090 
1091 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1092 {
1093 	unsigned long pa;
1094 	pud_t *pud_tbl;
1095 	int i;
1096 
1097 	if (p4d_large(*p4d)) {
1098 		pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1099 		xen_free_ro_pages(pa, P4D_SIZE);
1100 		return;
1101 	}
1102 
1103 	pud_tbl = pud_offset(p4d, 0);
1104 	for (i = 0; i < PTRS_PER_PUD; i++) {
1105 		if (pud_none(pud_tbl[i]))
1106 			continue;
1107 		xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1108 	}
1109 	set_p4d(p4d, __p4d(0));
1110 	xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1111 }
1112 
1113 /*
1114  * Since it is well isolated we can (and since it is perhaps large we should)
1115  * also free the page tables mapping the initial P->M table.
1116  */
1117 static void __init xen_cleanmfnmap(unsigned long vaddr)
1118 {
1119 	pgd_t *pgd;
1120 	p4d_t *p4d;
1121 	bool unpin;
1122 
1123 	unpin = (vaddr == 2 * PGDIR_SIZE);
1124 	vaddr &= PMD_MASK;
1125 	pgd = pgd_offset_k(vaddr);
1126 	p4d = p4d_offset(pgd, 0);
1127 	if (!p4d_none(*p4d))
1128 		xen_cleanmfnmap_p4d(p4d, unpin);
1129 }
1130 
1131 static void __init xen_pagetable_p2m_free(void)
1132 {
1133 	unsigned long size;
1134 	unsigned long addr;
1135 
1136 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1137 
1138 	/* No memory or already called. */
1139 	if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1140 		return;
1141 
1142 	/* using __ka address and sticking INVALID_P2M_ENTRY! */
1143 	memset((void *)xen_start_info->mfn_list, 0xff, size);
1144 
1145 	addr = xen_start_info->mfn_list;
1146 	/*
1147 	 * We could be in __ka space.
1148 	 * We roundup to the PMD, which means that if anybody at this stage is
1149 	 * using the __ka address of xen_start_info or
1150 	 * xen_start_info->shared_info they are in going to crash. Fortunately
1151 	 * we have already revectored in xen_setup_kernel_pagetable.
1152 	 */
1153 	size = roundup(size, PMD_SIZE);
1154 
1155 	if (addr >= __START_KERNEL_map) {
1156 		xen_cleanhighmap(addr, addr + size);
1157 		size = PAGE_ALIGN(xen_start_info->nr_pages *
1158 				  sizeof(unsigned long));
1159 		memblock_free((void *)addr, size);
1160 	} else {
1161 		xen_cleanmfnmap(addr);
1162 	}
1163 }
1164 
1165 static void __init xen_pagetable_cleanhighmap(void)
1166 {
1167 	unsigned long size;
1168 	unsigned long addr;
1169 
1170 	/* At this stage, cleanup_highmap has already cleaned __ka space
1171 	 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1172 	 * the ramdisk). We continue on, erasing PMD entries that point to page
1173 	 * tables - do note that they are accessible at this stage via __va.
1174 	 * As Xen is aligning the memory end to a 4MB boundary, for good
1175 	 * measure we also round up to PMD_SIZE * 2 - which means that if
1176 	 * anybody is using __ka address to the initial boot-stack - and try
1177 	 * to use it - they are going to crash. The xen_start_info has been
1178 	 * taken care of already in xen_setup_kernel_pagetable. */
1179 	addr = xen_start_info->pt_base;
1180 	size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1181 
1182 	xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1183 	xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1184 }
1185 
1186 static void __init xen_pagetable_p2m_setup(void)
1187 {
1188 	xen_vmalloc_p2m_tree();
1189 
1190 	xen_pagetable_p2m_free();
1191 
1192 	xen_pagetable_cleanhighmap();
1193 
1194 	/* And revector! Bye bye old array */
1195 	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1196 }
1197 
1198 static void __init xen_pagetable_init(void)
1199 {
1200 	/*
1201 	 * The majority of further PTE writes is to pagetables already
1202 	 * announced as such to Xen. Hence it is more efficient to use
1203 	 * hypercalls for these updates.
1204 	 */
1205 	pv_ops.mmu.set_pte = __xen_set_pte;
1206 
1207 	paging_init();
1208 	xen_post_allocator_init();
1209 
1210 	xen_pagetable_p2m_setup();
1211 
1212 	/* Allocate and initialize top and mid mfn levels for p2m structure */
1213 	xen_build_mfn_list_list();
1214 
1215 	/* Remap memory freed due to conflicts with E820 map */
1216 	xen_remap_memory();
1217 	xen_setup_mfn_list_list();
1218 }
1219 
1220 static noinstr void xen_write_cr2(unsigned long cr2)
1221 {
1222 	this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1223 }
1224 
1225 static noinline void xen_flush_tlb(void)
1226 {
1227 	struct mmuext_op *op;
1228 	struct multicall_space mcs;
1229 
1230 	preempt_disable();
1231 
1232 	mcs = xen_mc_entry(sizeof(*op));
1233 
1234 	op = mcs.args;
1235 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1236 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1237 
1238 	xen_mc_issue(PARAVIRT_LAZY_MMU);
1239 
1240 	preempt_enable();
1241 }
1242 
1243 static void xen_flush_tlb_one_user(unsigned long addr)
1244 {
1245 	struct mmuext_op *op;
1246 	struct multicall_space mcs;
1247 
1248 	trace_xen_mmu_flush_tlb_one_user(addr);
1249 
1250 	preempt_disable();
1251 
1252 	mcs = xen_mc_entry(sizeof(*op));
1253 	op = mcs.args;
1254 	op->cmd = MMUEXT_INVLPG_LOCAL;
1255 	op->arg1.linear_addr = addr & PAGE_MASK;
1256 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1257 
1258 	xen_mc_issue(PARAVIRT_LAZY_MMU);
1259 
1260 	preempt_enable();
1261 }
1262 
1263 static void xen_flush_tlb_multi(const struct cpumask *cpus,
1264 				const struct flush_tlb_info *info)
1265 {
1266 	struct {
1267 		struct mmuext_op op;
1268 		DECLARE_BITMAP(mask, NR_CPUS);
1269 	} *args;
1270 	struct multicall_space mcs;
1271 	const size_t mc_entry_size = sizeof(args->op) +
1272 		sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1273 
1274 	trace_xen_mmu_flush_tlb_multi(cpus, info->mm, info->start, info->end);
1275 
1276 	if (cpumask_empty(cpus))
1277 		return;		/* nothing to do */
1278 
1279 	mcs = xen_mc_entry(mc_entry_size);
1280 	args = mcs.args;
1281 	args->op.arg2.vcpumask = to_cpumask(args->mask);
1282 
1283 	/* Remove any offline CPUs */
1284 	cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1285 
1286 	args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1287 	if (info->end != TLB_FLUSH_ALL &&
1288 	    (info->end - info->start) <= PAGE_SIZE) {
1289 		args->op.cmd = MMUEXT_INVLPG_MULTI;
1290 		args->op.arg1.linear_addr = info->start;
1291 	}
1292 
1293 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1294 
1295 	xen_mc_issue(PARAVIRT_LAZY_MMU);
1296 }
1297 
1298 static unsigned long xen_read_cr3(void)
1299 {
1300 	return this_cpu_read(xen_cr3);
1301 }
1302 
1303 static void set_current_cr3(void *v)
1304 {
1305 	this_cpu_write(xen_current_cr3, (unsigned long)v);
1306 }
1307 
1308 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1309 {
1310 	struct mmuext_op op;
1311 	unsigned long mfn;
1312 
1313 	trace_xen_mmu_write_cr3(kernel, cr3);
1314 
1315 	if (cr3)
1316 		mfn = pfn_to_mfn(PFN_DOWN(cr3));
1317 	else
1318 		mfn = 0;
1319 
1320 	WARN_ON(mfn == 0 && kernel);
1321 
1322 	op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1323 	op.arg1.mfn = mfn;
1324 
1325 	xen_extend_mmuext_op(&op);
1326 
1327 	if (kernel) {
1328 		this_cpu_write(xen_cr3, cr3);
1329 
1330 		/* Update xen_current_cr3 once the batch has actually
1331 		   been submitted. */
1332 		xen_mc_callback(set_current_cr3, (void *)cr3);
1333 	}
1334 }
1335 static void xen_write_cr3(unsigned long cr3)
1336 {
1337 	pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1338 
1339 	BUG_ON(preemptible());
1340 
1341 	xen_mc_batch();  /* disables interrupts */
1342 
1343 	/* Update while interrupts are disabled, so its atomic with
1344 	   respect to ipis */
1345 	this_cpu_write(xen_cr3, cr3);
1346 
1347 	__xen_write_cr3(true, cr3);
1348 
1349 	if (user_pgd)
1350 		__xen_write_cr3(false, __pa(user_pgd));
1351 	else
1352 		__xen_write_cr3(false, 0);
1353 
1354 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1355 }
1356 
1357 /*
1358  * At the start of the day - when Xen launches a guest, it has already
1359  * built pagetables for the guest. We diligently look over them
1360  * in xen_setup_kernel_pagetable and graft as appropriate them in the
1361  * init_top_pgt and its friends. Then when we are happy we load
1362  * the new init_top_pgt - and continue on.
1363  *
1364  * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1365  * up the rest of the pagetables. When it has completed it loads the cr3.
1366  * N.B. that baremetal would start at 'start_kernel' (and the early
1367  * #PF handler would create bootstrap pagetables) - so we are running
1368  * with the same assumptions as what to do when write_cr3 is executed
1369  * at this point.
1370  *
1371  * Since there are no user-page tables at all, we have two variants
1372  * of xen_write_cr3 - the early bootup (this one), and the late one
1373  * (xen_write_cr3). The reason we have to do that is that in 64-bit
1374  * the Linux kernel and user-space are both in ring 3 while the
1375  * hypervisor is in ring 0.
1376  */
1377 static void __init xen_write_cr3_init(unsigned long cr3)
1378 {
1379 	BUG_ON(preemptible());
1380 
1381 	xen_mc_batch();  /* disables interrupts */
1382 
1383 	/* Update while interrupts are disabled, so its atomic with
1384 	   respect to ipis */
1385 	this_cpu_write(xen_cr3, cr3);
1386 
1387 	__xen_write_cr3(true, cr3);
1388 
1389 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1390 }
1391 
1392 static int xen_pgd_alloc(struct mm_struct *mm)
1393 {
1394 	pgd_t *pgd = mm->pgd;
1395 	struct page *page = virt_to_page(pgd);
1396 	pgd_t *user_pgd;
1397 	int ret = -ENOMEM;
1398 
1399 	BUG_ON(PagePinned(virt_to_page(pgd)));
1400 	BUG_ON(page->private != 0);
1401 
1402 	user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1403 	page->private = (unsigned long)user_pgd;
1404 
1405 	if (user_pgd != NULL) {
1406 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1407 		user_pgd[pgd_index(VSYSCALL_ADDR)] =
1408 			__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1409 #endif
1410 		ret = 0;
1411 	}
1412 
1413 	BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1414 
1415 	return ret;
1416 }
1417 
1418 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1419 {
1420 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
1421 
1422 	if (user_pgd)
1423 		free_page((unsigned long)user_pgd);
1424 }
1425 
1426 /*
1427  * Init-time set_pte while constructing initial pagetables, which
1428  * doesn't allow RO page table pages to be remapped RW.
1429  *
1430  * If there is no MFN for this PFN then this page is initially
1431  * ballooned out so clear the PTE (as in decrease_reservation() in
1432  * drivers/xen/balloon.c).
1433  *
1434  * Many of these PTE updates are done on unpinned and writable pages
1435  * and doing a hypercall for these is unnecessary and expensive.  At
1436  * this point it is rarely possible to tell if a page is pinned, so
1437  * mostly write the PTE directly and rely on Xen trapping and
1438  * emulating any updates as necessary.
1439  */
1440 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1441 {
1442 	if (unlikely(is_early_ioremap_ptep(ptep)))
1443 		__xen_set_pte(ptep, pte);
1444 	else
1445 		native_set_pte(ptep, pte);
1446 }
1447 
1448 __visible pte_t xen_make_pte_init(pteval_t pte)
1449 {
1450 	unsigned long pfn;
1451 
1452 	/*
1453 	 * Pages belonging to the initial p2m list mapped outside the default
1454 	 * address range must be mapped read-only. This region contains the
1455 	 * page tables for mapping the p2m list, too, and page tables MUST be
1456 	 * mapped read-only.
1457 	 */
1458 	pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1459 	if (xen_start_info->mfn_list < __START_KERNEL_map &&
1460 	    pfn >= xen_start_info->first_p2m_pfn &&
1461 	    pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1462 		pte &= ~_PAGE_RW;
1463 
1464 	pte = pte_pfn_to_mfn(pte);
1465 	return native_make_pte(pte);
1466 }
1467 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1468 
1469 /* Early in boot, while setting up the initial pagetable, assume
1470    everything is pinned. */
1471 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1472 {
1473 #ifdef CONFIG_FLATMEM
1474 	BUG_ON(mem_map);	/* should only be used early */
1475 #endif
1476 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1477 	pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1478 }
1479 
1480 /* Used for pmd and pud */
1481 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1482 {
1483 #ifdef CONFIG_FLATMEM
1484 	BUG_ON(mem_map);	/* should only be used early */
1485 #endif
1486 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1487 }
1488 
1489 /* Early release_pte assumes that all pts are pinned, since there's
1490    only init_mm and anything attached to that is pinned. */
1491 static void __init xen_release_pte_init(unsigned long pfn)
1492 {
1493 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1494 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1495 }
1496 
1497 static void __init xen_release_pmd_init(unsigned long pfn)
1498 {
1499 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1500 }
1501 
1502 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1503 {
1504 	struct multicall_space mcs;
1505 	struct mmuext_op *op;
1506 
1507 	mcs = __xen_mc_entry(sizeof(*op));
1508 	op = mcs.args;
1509 	op->cmd = cmd;
1510 	op->arg1.mfn = pfn_to_mfn(pfn);
1511 
1512 	MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1513 }
1514 
1515 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1516 {
1517 	struct multicall_space mcs;
1518 	unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1519 
1520 	mcs = __xen_mc_entry(0);
1521 	MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1522 				pfn_pte(pfn, prot), 0);
1523 }
1524 
1525 /* This needs to make sure the new pte page is pinned iff its being
1526    attached to a pinned pagetable. */
1527 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1528 				    unsigned level)
1529 {
1530 	bool pinned = xen_page_pinned(mm->pgd);
1531 
1532 	trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1533 
1534 	if (pinned) {
1535 		struct page *page = pfn_to_page(pfn);
1536 
1537 		pinned = false;
1538 		if (static_branch_likely(&xen_struct_pages_ready)) {
1539 			pinned = PagePinned(page);
1540 			SetPagePinned(page);
1541 		}
1542 
1543 		xen_mc_batch();
1544 
1545 		__set_pfn_prot(pfn, PAGE_KERNEL_RO);
1546 
1547 		if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS && !pinned)
1548 			__pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1549 
1550 		xen_mc_issue(PARAVIRT_LAZY_MMU);
1551 	}
1552 }
1553 
1554 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1555 {
1556 	xen_alloc_ptpage(mm, pfn, PT_PTE);
1557 }
1558 
1559 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1560 {
1561 	xen_alloc_ptpage(mm, pfn, PT_PMD);
1562 }
1563 
1564 /* This should never happen until we're OK to use struct page */
1565 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1566 {
1567 	struct page *page = pfn_to_page(pfn);
1568 	bool pinned = PagePinned(page);
1569 
1570 	trace_xen_mmu_release_ptpage(pfn, level, pinned);
1571 
1572 	if (pinned) {
1573 		xen_mc_batch();
1574 
1575 		if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1576 			__pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1577 
1578 		__set_pfn_prot(pfn, PAGE_KERNEL);
1579 
1580 		xen_mc_issue(PARAVIRT_LAZY_MMU);
1581 
1582 		ClearPagePinned(page);
1583 	}
1584 }
1585 
1586 static void xen_release_pte(unsigned long pfn)
1587 {
1588 	xen_release_ptpage(pfn, PT_PTE);
1589 }
1590 
1591 static void xen_release_pmd(unsigned long pfn)
1592 {
1593 	xen_release_ptpage(pfn, PT_PMD);
1594 }
1595 
1596 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1597 {
1598 	xen_alloc_ptpage(mm, pfn, PT_PUD);
1599 }
1600 
1601 static void xen_release_pud(unsigned long pfn)
1602 {
1603 	xen_release_ptpage(pfn, PT_PUD);
1604 }
1605 
1606 /*
1607  * Like __va(), but returns address in the kernel mapping (which is
1608  * all we have until the physical memory mapping has been set up.
1609  */
1610 static void * __init __ka(phys_addr_t paddr)
1611 {
1612 	return (void *)(paddr + __START_KERNEL_map);
1613 }
1614 
1615 /* Convert a machine address to physical address */
1616 static unsigned long __init m2p(phys_addr_t maddr)
1617 {
1618 	phys_addr_t paddr;
1619 
1620 	maddr &= XEN_PTE_MFN_MASK;
1621 	paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1622 
1623 	return paddr;
1624 }
1625 
1626 /* Convert a machine address to kernel virtual */
1627 static void * __init m2v(phys_addr_t maddr)
1628 {
1629 	return __ka(m2p(maddr));
1630 }
1631 
1632 /* Set the page permissions on an identity-mapped pages */
1633 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1634 				       unsigned long flags)
1635 {
1636 	unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1637 	pte_t pte = pfn_pte(pfn, prot);
1638 
1639 	if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1640 		BUG();
1641 }
1642 static void __init set_page_prot(void *addr, pgprot_t prot)
1643 {
1644 	return set_page_prot_flags(addr, prot, UVMF_NONE);
1645 }
1646 
1647 void __init xen_setup_machphys_mapping(void)
1648 {
1649 	struct xen_machphys_mapping mapping;
1650 
1651 	if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1652 		machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1653 		machine_to_phys_nr = mapping.max_mfn + 1;
1654 	} else {
1655 		machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1656 	}
1657 }
1658 
1659 static void __init convert_pfn_mfn(void *v)
1660 {
1661 	pte_t *pte = v;
1662 	int i;
1663 
1664 	/* All levels are converted the same way, so just treat them
1665 	   as ptes. */
1666 	for (i = 0; i < PTRS_PER_PTE; i++)
1667 		pte[i] = xen_make_pte(pte[i].pte);
1668 }
1669 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1670 				 unsigned long addr)
1671 {
1672 	if (*pt_base == PFN_DOWN(__pa(addr))) {
1673 		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1674 		clear_page((void *)addr);
1675 		(*pt_base)++;
1676 	}
1677 	if (*pt_end == PFN_DOWN(__pa(addr))) {
1678 		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1679 		clear_page((void *)addr);
1680 		(*pt_end)--;
1681 	}
1682 }
1683 /*
1684  * Set up the initial kernel pagetable.
1685  *
1686  * We can construct this by grafting the Xen provided pagetable into
1687  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1688  * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1689  * kernel has a physical mapping to start with - but that's enough to
1690  * get __va working.  We need to fill in the rest of the physical
1691  * mapping once some sort of allocator has been set up.
1692  */
1693 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1694 {
1695 	pud_t *l3;
1696 	pmd_t *l2;
1697 	unsigned long addr[3];
1698 	unsigned long pt_base, pt_end;
1699 	unsigned i;
1700 
1701 	/* max_pfn_mapped is the last pfn mapped in the initial memory
1702 	 * mappings. Considering that on Xen after the kernel mappings we
1703 	 * have the mappings of some pages that don't exist in pfn space, we
1704 	 * set max_pfn_mapped to the last real pfn mapped. */
1705 	if (xen_start_info->mfn_list < __START_KERNEL_map)
1706 		max_pfn_mapped = xen_start_info->first_p2m_pfn;
1707 	else
1708 		max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1709 
1710 	pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1711 	pt_end = pt_base + xen_start_info->nr_pt_frames;
1712 
1713 	/* Zap identity mapping */
1714 	init_top_pgt[0] = __pgd(0);
1715 
1716 	/* Pre-constructed entries are in pfn, so convert to mfn */
1717 	/* L4[273] -> level3_ident_pgt  */
1718 	/* L4[511] -> level3_kernel_pgt */
1719 	convert_pfn_mfn(init_top_pgt);
1720 
1721 	/* L3_i[0] -> level2_ident_pgt */
1722 	convert_pfn_mfn(level3_ident_pgt);
1723 	/* L3_k[510] -> level2_kernel_pgt */
1724 	/* L3_k[511] -> level2_fixmap_pgt */
1725 	convert_pfn_mfn(level3_kernel_pgt);
1726 
1727 	/* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1728 	convert_pfn_mfn(level2_fixmap_pgt);
1729 
1730 	/* We get [511][511] and have Xen's version of level2_kernel_pgt */
1731 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1732 	l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1733 
1734 	addr[0] = (unsigned long)pgd;
1735 	addr[1] = (unsigned long)l3;
1736 	addr[2] = (unsigned long)l2;
1737 	/* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1738 	 * Both L4[273][0] and L4[511][510] have entries that point to the same
1739 	 * L2 (PMD) tables. Meaning that if you modify it in __va space
1740 	 * it will be also modified in the __ka space! (But if you just
1741 	 * modify the PMD table to point to other PTE's or none, then you
1742 	 * are OK - which is what cleanup_highmap does) */
1743 	copy_page(level2_ident_pgt, l2);
1744 	/* Graft it onto L4[511][510] */
1745 	copy_page(level2_kernel_pgt, l2);
1746 
1747 	/*
1748 	 * Zap execute permission from the ident map. Due to the sharing of
1749 	 * L1 entries we need to do this in the L2.
1750 	 */
1751 	if (__supported_pte_mask & _PAGE_NX) {
1752 		for (i = 0; i < PTRS_PER_PMD; ++i) {
1753 			if (pmd_none(level2_ident_pgt[i]))
1754 				continue;
1755 			level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1756 		}
1757 	}
1758 
1759 	/* Copy the initial P->M table mappings if necessary. */
1760 	i = pgd_index(xen_start_info->mfn_list);
1761 	if (i && i < pgd_index(__START_KERNEL_map))
1762 		init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1763 
1764 	/* Make pagetable pieces RO */
1765 	set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1766 	set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1767 	set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1768 	set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1769 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1770 	set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1771 
1772 	for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1773 		set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1774 			      PAGE_KERNEL_RO);
1775 	}
1776 
1777 	/* Pin down new L4 */
1778 	pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1779 			  PFN_DOWN(__pa_symbol(init_top_pgt)));
1780 
1781 	/* Unpin Xen-provided one */
1782 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1783 
1784 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1785 	/* Pin user vsyscall L3 */
1786 	set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1787 	pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1788 			  PFN_DOWN(__pa_symbol(level3_user_vsyscall)));
1789 #endif
1790 
1791 	/*
1792 	 * At this stage there can be no user pgd, and no page structure to
1793 	 * attach it to, so make sure we just set kernel pgd.
1794 	 */
1795 	xen_mc_batch();
1796 	__xen_write_cr3(true, __pa(init_top_pgt));
1797 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1798 
1799 	/* We can't that easily rip out L3 and L2, as the Xen pagetables are
1800 	 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1801 	 * the initial domain. For guests using the toolstack, they are in:
1802 	 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1803 	 * rip out the [L4] (pgd), but for guests we shave off three pages.
1804 	 */
1805 	for (i = 0; i < ARRAY_SIZE(addr); i++)
1806 		check_pt_base(&pt_base, &pt_end, addr[i]);
1807 
1808 	/* Our (by three pages) smaller Xen pagetable that we are using */
1809 	xen_pt_base = PFN_PHYS(pt_base);
1810 	xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1811 	memblock_reserve(xen_pt_base, xen_pt_size);
1812 
1813 	/* Revector the xen_start_info */
1814 	xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1815 }
1816 
1817 /*
1818  * Read a value from a physical address.
1819  */
1820 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1821 {
1822 	unsigned long *vaddr;
1823 	unsigned long val;
1824 
1825 	vaddr = early_memremap_ro(addr, sizeof(val));
1826 	val = *vaddr;
1827 	early_memunmap(vaddr, sizeof(val));
1828 	return val;
1829 }
1830 
1831 /*
1832  * Translate a virtual address to a physical one without relying on mapped
1833  * page tables. Don't rely on big pages being aligned in (guest) physical
1834  * space!
1835  */
1836 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1837 {
1838 	phys_addr_t pa;
1839 	pgd_t pgd;
1840 	pud_t pud;
1841 	pmd_t pmd;
1842 	pte_t pte;
1843 
1844 	pa = read_cr3_pa();
1845 	pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1846 						       sizeof(pgd)));
1847 	if (!pgd_present(pgd))
1848 		return 0;
1849 
1850 	pa = pgd_val(pgd) & PTE_PFN_MASK;
1851 	pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1852 						       sizeof(pud)));
1853 	if (!pud_present(pud))
1854 		return 0;
1855 	pa = pud_val(pud) & PTE_PFN_MASK;
1856 	if (pud_large(pud))
1857 		return pa + (vaddr & ~PUD_MASK);
1858 
1859 	pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1860 						       sizeof(pmd)));
1861 	if (!pmd_present(pmd))
1862 		return 0;
1863 	pa = pmd_val(pmd) & PTE_PFN_MASK;
1864 	if (pmd_large(pmd))
1865 		return pa + (vaddr & ~PMD_MASK);
1866 
1867 	pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1868 						       sizeof(pte)));
1869 	if (!pte_present(pte))
1870 		return 0;
1871 	pa = pte_pfn(pte) << PAGE_SHIFT;
1872 
1873 	return pa | (vaddr & ~PAGE_MASK);
1874 }
1875 
1876 /*
1877  * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1878  * this area.
1879  */
1880 void __init xen_relocate_p2m(void)
1881 {
1882 	phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1883 	unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1884 	int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1885 	pte_t *pt;
1886 	pmd_t *pmd;
1887 	pud_t *pud;
1888 	pgd_t *pgd;
1889 	unsigned long *new_p2m;
1890 
1891 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1892 	n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1893 	n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1894 	n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1895 	n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1896 	n_frames = n_pte + n_pt + n_pmd + n_pud;
1897 
1898 	new_area = xen_find_free_area(PFN_PHYS(n_frames));
1899 	if (!new_area) {
1900 		xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1901 		BUG();
1902 	}
1903 
1904 	/*
1905 	 * Setup the page tables for addressing the new p2m list.
1906 	 * We have asked the hypervisor to map the p2m list at the user address
1907 	 * PUD_SIZE. It may have done so, or it may have used a kernel space
1908 	 * address depending on the Xen version.
1909 	 * To avoid any possible virtual address collision, just use
1910 	 * 2 * PUD_SIZE for the new area.
1911 	 */
1912 	pud_phys = new_area;
1913 	pmd_phys = pud_phys + PFN_PHYS(n_pud);
1914 	pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1915 	p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1916 
1917 	pgd = __va(read_cr3_pa());
1918 	new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1919 	for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1920 		pud = early_memremap(pud_phys, PAGE_SIZE);
1921 		clear_page(pud);
1922 		for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1923 				idx_pmd++) {
1924 			pmd = early_memremap(pmd_phys, PAGE_SIZE);
1925 			clear_page(pmd);
1926 			for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1927 					idx_pt++) {
1928 				pt = early_memremap(pt_phys, PAGE_SIZE);
1929 				clear_page(pt);
1930 				for (idx_pte = 0;
1931 				     idx_pte < min(n_pte, PTRS_PER_PTE);
1932 				     idx_pte++) {
1933 					pt[idx_pte] = pfn_pte(p2m_pfn,
1934 							      PAGE_KERNEL);
1935 					p2m_pfn++;
1936 				}
1937 				n_pte -= PTRS_PER_PTE;
1938 				early_memunmap(pt, PAGE_SIZE);
1939 				make_lowmem_page_readonly(__va(pt_phys));
1940 				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
1941 						PFN_DOWN(pt_phys));
1942 				pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
1943 				pt_phys += PAGE_SIZE;
1944 			}
1945 			n_pt -= PTRS_PER_PMD;
1946 			early_memunmap(pmd, PAGE_SIZE);
1947 			make_lowmem_page_readonly(__va(pmd_phys));
1948 			pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
1949 					PFN_DOWN(pmd_phys));
1950 			pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
1951 			pmd_phys += PAGE_SIZE;
1952 		}
1953 		n_pmd -= PTRS_PER_PUD;
1954 		early_memunmap(pud, PAGE_SIZE);
1955 		make_lowmem_page_readonly(__va(pud_phys));
1956 		pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
1957 		set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
1958 		pud_phys += PAGE_SIZE;
1959 	}
1960 
1961 	/* Now copy the old p2m info to the new area. */
1962 	memcpy(new_p2m, xen_p2m_addr, size);
1963 	xen_p2m_addr = new_p2m;
1964 
1965 	/* Release the old p2m list and set new list info. */
1966 	p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
1967 	BUG_ON(!p2m_pfn);
1968 	p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
1969 
1970 	if (xen_start_info->mfn_list < __START_KERNEL_map) {
1971 		pfn = xen_start_info->first_p2m_pfn;
1972 		pfn_end = xen_start_info->first_p2m_pfn +
1973 			  xen_start_info->nr_p2m_frames;
1974 		set_pgd(pgd + 1, __pgd(0));
1975 	} else {
1976 		pfn = p2m_pfn;
1977 		pfn_end = p2m_pfn_end;
1978 	}
1979 
1980 	memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
1981 	while (pfn < pfn_end) {
1982 		if (pfn == p2m_pfn) {
1983 			pfn = p2m_pfn_end;
1984 			continue;
1985 		}
1986 		make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1987 		pfn++;
1988 	}
1989 
1990 	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1991 	xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
1992 	xen_start_info->nr_p2m_frames = n_frames;
1993 }
1994 
1995 void __init xen_reserve_special_pages(void)
1996 {
1997 	phys_addr_t paddr;
1998 
1999 	memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2000 	if (xen_start_info->store_mfn) {
2001 		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2002 		memblock_reserve(paddr, PAGE_SIZE);
2003 	}
2004 	if (!xen_initial_domain()) {
2005 		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2006 		memblock_reserve(paddr, PAGE_SIZE);
2007 	}
2008 }
2009 
2010 void __init xen_pt_check_e820(void)
2011 {
2012 	if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2013 		xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2014 		BUG();
2015 	}
2016 }
2017 
2018 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2019 
2020 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2021 {
2022 	pte_t pte;
2023 	unsigned long vaddr;
2024 
2025 	phys >>= PAGE_SHIFT;
2026 
2027 	switch (idx) {
2028 	case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2029 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2030 	case VSYSCALL_PAGE:
2031 #endif
2032 		/* All local page mappings */
2033 		pte = pfn_pte(phys, prot);
2034 		break;
2035 
2036 #ifdef CONFIG_X86_LOCAL_APIC
2037 	case FIX_APIC_BASE:	/* maps dummy local APIC */
2038 		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2039 		break;
2040 #endif
2041 
2042 #ifdef CONFIG_X86_IO_APIC
2043 	case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2044 		/*
2045 		 * We just don't map the IO APIC - all access is via
2046 		 * hypercalls.  Keep the address in the pte for reference.
2047 		 */
2048 		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2049 		break;
2050 #endif
2051 
2052 	case FIX_PARAVIRT_BOOTMAP:
2053 		/* This is an MFN, but it isn't an IO mapping from the
2054 		   IO domain */
2055 		pte = mfn_pte(phys, prot);
2056 		break;
2057 
2058 	default:
2059 		/* By default, set_fixmap is used for hardware mappings */
2060 		pte = mfn_pte(phys, prot);
2061 		break;
2062 	}
2063 
2064 	vaddr = __fix_to_virt(idx);
2065 	if (HYPERVISOR_update_va_mapping(vaddr, pte, UVMF_INVLPG))
2066 		BUG();
2067 
2068 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2069 	/* Replicate changes to map the vsyscall page into the user
2070 	   pagetable vsyscall mapping. */
2071 	if (idx == VSYSCALL_PAGE)
2072 		set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2073 #endif
2074 }
2075 
2076 static void __init xen_post_allocator_init(void)
2077 {
2078 	pv_ops.mmu.set_pte = xen_set_pte;
2079 	pv_ops.mmu.set_pmd = xen_set_pmd;
2080 	pv_ops.mmu.set_pud = xen_set_pud;
2081 	pv_ops.mmu.set_p4d = xen_set_p4d;
2082 
2083 	/* This will work as long as patching hasn't happened yet
2084 	   (which it hasn't) */
2085 	pv_ops.mmu.alloc_pte = xen_alloc_pte;
2086 	pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2087 	pv_ops.mmu.release_pte = xen_release_pte;
2088 	pv_ops.mmu.release_pmd = xen_release_pmd;
2089 	pv_ops.mmu.alloc_pud = xen_alloc_pud;
2090 	pv_ops.mmu.release_pud = xen_release_pud;
2091 	pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2092 
2093 	pv_ops.mmu.write_cr3 = &xen_write_cr3;
2094 }
2095 
2096 static void xen_leave_lazy_mmu(void)
2097 {
2098 	preempt_disable();
2099 	xen_mc_flush();
2100 	paravirt_leave_lazy_mmu();
2101 	preempt_enable();
2102 }
2103 
2104 static const typeof(pv_ops) xen_mmu_ops __initconst = {
2105 	.mmu = {
2106 		.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2107 		.write_cr2 = xen_write_cr2,
2108 
2109 		.read_cr3 = xen_read_cr3,
2110 		.write_cr3 = xen_write_cr3_init,
2111 
2112 		.flush_tlb_user = xen_flush_tlb,
2113 		.flush_tlb_kernel = xen_flush_tlb,
2114 		.flush_tlb_one_user = xen_flush_tlb_one_user,
2115 		.flush_tlb_multi = xen_flush_tlb_multi,
2116 		.tlb_remove_table = tlb_remove_table,
2117 
2118 		.pgd_alloc = xen_pgd_alloc,
2119 		.pgd_free = xen_pgd_free,
2120 
2121 		.alloc_pte = xen_alloc_pte_init,
2122 		.release_pte = xen_release_pte_init,
2123 		.alloc_pmd = xen_alloc_pmd_init,
2124 		.release_pmd = xen_release_pmd_init,
2125 
2126 		.set_pte = xen_set_pte_init,
2127 		.set_pmd = xen_set_pmd_hyper,
2128 
2129 		.ptep_modify_prot_start = xen_ptep_modify_prot_start,
2130 		.ptep_modify_prot_commit = xen_ptep_modify_prot_commit,
2131 
2132 		.pte_val = PV_CALLEE_SAVE(xen_pte_val),
2133 		.pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2134 
2135 		.make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2136 		.make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2137 
2138 		.set_pud = xen_set_pud_hyper,
2139 
2140 		.make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2141 		.pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2142 
2143 		.pud_val = PV_CALLEE_SAVE(xen_pud_val),
2144 		.make_pud = PV_CALLEE_SAVE(xen_make_pud),
2145 		.set_p4d = xen_set_p4d_hyper,
2146 
2147 		.alloc_pud = xen_alloc_pmd_init,
2148 		.release_pud = xen_release_pmd_init,
2149 
2150 #if CONFIG_PGTABLE_LEVELS >= 5
2151 		.p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2152 		.make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2153 #endif
2154 
2155 		.activate_mm = xen_activate_mm,
2156 		.dup_mmap = xen_dup_mmap,
2157 		.exit_mmap = xen_exit_mmap,
2158 
2159 		.lazy_mode = {
2160 			.enter = paravirt_enter_lazy_mmu,
2161 			.leave = xen_leave_lazy_mmu,
2162 			.flush = paravirt_flush_lazy_mmu,
2163 		},
2164 
2165 		.set_fixmap = xen_set_fixmap,
2166 	},
2167 };
2168 
2169 void __init xen_init_mmu_ops(void)
2170 {
2171 	x86_init.paging.pagetable_init = xen_pagetable_init;
2172 	x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2173 
2174 	pv_ops.mmu = xen_mmu_ops.mmu;
2175 
2176 	memset(dummy_mapping, 0xff, PAGE_SIZE);
2177 }
2178 
2179 /* Protected by xen_reservation_lock. */
2180 #define MAX_CONTIG_ORDER 9 /* 2MB */
2181 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2182 
2183 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2184 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2185 				unsigned long *in_frames,
2186 				unsigned long *out_frames)
2187 {
2188 	int i;
2189 	struct multicall_space mcs;
2190 
2191 	xen_mc_batch();
2192 	for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2193 		mcs = __xen_mc_entry(0);
2194 
2195 		if (in_frames)
2196 			in_frames[i] = virt_to_mfn(vaddr);
2197 
2198 		MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2199 		__set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2200 
2201 		if (out_frames)
2202 			out_frames[i] = virt_to_pfn(vaddr);
2203 	}
2204 	xen_mc_issue(0);
2205 }
2206 
2207 /*
2208  * Update the pfn-to-mfn mappings for a virtual address range, either to
2209  * point to an array of mfns, or contiguously from a single starting
2210  * mfn.
2211  */
2212 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2213 				     unsigned long *mfns,
2214 				     unsigned long first_mfn)
2215 {
2216 	unsigned i, limit;
2217 	unsigned long mfn;
2218 
2219 	xen_mc_batch();
2220 
2221 	limit = 1u << order;
2222 	for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2223 		struct multicall_space mcs;
2224 		unsigned flags;
2225 
2226 		mcs = __xen_mc_entry(0);
2227 		if (mfns)
2228 			mfn = mfns[i];
2229 		else
2230 			mfn = first_mfn + i;
2231 
2232 		if (i < (limit - 1))
2233 			flags = 0;
2234 		else {
2235 			if (order == 0)
2236 				flags = UVMF_INVLPG | UVMF_ALL;
2237 			else
2238 				flags = UVMF_TLB_FLUSH | UVMF_ALL;
2239 		}
2240 
2241 		MULTI_update_va_mapping(mcs.mc, vaddr,
2242 				mfn_pte(mfn, PAGE_KERNEL), flags);
2243 
2244 		set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2245 	}
2246 
2247 	xen_mc_issue(0);
2248 }
2249 
2250 /*
2251  * Perform the hypercall to exchange a region of our pfns to point to
2252  * memory with the required contiguous alignment.  Takes the pfns as
2253  * input, and populates mfns as output.
2254  *
2255  * Returns a success code indicating whether the hypervisor was able to
2256  * satisfy the request or not.
2257  */
2258 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2259 			       unsigned long *pfns_in,
2260 			       unsigned long extents_out,
2261 			       unsigned int order_out,
2262 			       unsigned long *mfns_out,
2263 			       unsigned int address_bits)
2264 {
2265 	long rc;
2266 	int success;
2267 
2268 	struct xen_memory_exchange exchange = {
2269 		.in = {
2270 			.nr_extents   = extents_in,
2271 			.extent_order = order_in,
2272 			.extent_start = pfns_in,
2273 			.domid        = DOMID_SELF
2274 		},
2275 		.out = {
2276 			.nr_extents   = extents_out,
2277 			.extent_order = order_out,
2278 			.extent_start = mfns_out,
2279 			.address_bits = address_bits,
2280 			.domid        = DOMID_SELF
2281 		}
2282 	};
2283 
2284 	BUG_ON(extents_in << order_in != extents_out << order_out);
2285 
2286 	rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2287 	success = (exchange.nr_exchanged == extents_in);
2288 
2289 	BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2290 	BUG_ON(success && (rc != 0));
2291 
2292 	return success;
2293 }
2294 
2295 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2296 				 unsigned int address_bits,
2297 				 dma_addr_t *dma_handle)
2298 {
2299 	unsigned long *in_frames = discontig_frames, out_frame;
2300 	unsigned long  flags;
2301 	int            success;
2302 	unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2303 
2304 	/*
2305 	 * Currently an auto-translated guest will not perform I/O, nor will
2306 	 * it require PAE page directories below 4GB. Therefore any calls to
2307 	 * this function are redundant and can be ignored.
2308 	 */
2309 
2310 	if (unlikely(order > MAX_CONTIG_ORDER))
2311 		return -ENOMEM;
2312 
2313 	memset((void *) vstart, 0, PAGE_SIZE << order);
2314 
2315 	spin_lock_irqsave(&xen_reservation_lock, flags);
2316 
2317 	/* 1. Zap current PTEs, remembering MFNs. */
2318 	xen_zap_pfn_range(vstart, order, in_frames, NULL);
2319 
2320 	/* 2. Get a new contiguous memory extent. */
2321 	out_frame = virt_to_pfn(vstart);
2322 	success = xen_exchange_memory(1UL << order, 0, in_frames,
2323 				      1, order, &out_frame,
2324 				      address_bits);
2325 
2326 	/* 3. Map the new extent in place of old pages. */
2327 	if (success)
2328 		xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2329 	else
2330 		xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2331 
2332 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2333 
2334 	*dma_handle = virt_to_machine(vstart).maddr;
2335 	return success ? 0 : -ENOMEM;
2336 }
2337 
2338 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2339 {
2340 	unsigned long *out_frames = discontig_frames, in_frame;
2341 	unsigned long  flags;
2342 	int success;
2343 	unsigned long vstart;
2344 
2345 	if (unlikely(order > MAX_CONTIG_ORDER))
2346 		return;
2347 
2348 	vstart = (unsigned long)phys_to_virt(pstart);
2349 	memset((void *) vstart, 0, PAGE_SIZE << order);
2350 
2351 	spin_lock_irqsave(&xen_reservation_lock, flags);
2352 
2353 	/* 1. Find start MFN of contiguous extent. */
2354 	in_frame = virt_to_mfn(vstart);
2355 
2356 	/* 2. Zap current PTEs. */
2357 	xen_zap_pfn_range(vstart, order, NULL, out_frames);
2358 
2359 	/* 3. Do the exchange for non-contiguous MFNs. */
2360 	success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2361 					0, out_frames, 0);
2362 
2363 	/* 4. Map new pages in place of old pages. */
2364 	if (success)
2365 		xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2366 	else
2367 		xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2368 
2369 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2370 }
2371 
2372 static noinline void xen_flush_tlb_all(void)
2373 {
2374 	struct mmuext_op *op;
2375 	struct multicall_space mcs;
2376 
2377 	preempt_disable();
2378 
2379 	mcs = xen_mc_entry(sizeof(*op));
2380 
2381 	op = mcs.args;
2382 	op->cmd = MMUEXT_TLB_FLUSH_ALL;
2383 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2384 
2385 	xen_mc_issue(PARAVIRT_LAZY_MMU);
2386 
2387 	preempt_enable();
2388 }
2389 
2390 #define REMAP_BATCH_SIZE 16
2391 
2392 struct remap_data {
2393 	xen_pfn_t *pfn;
2394 	bool contiguous;
2395 	bool no_translate;
2396 	pgprot_t prot;
2397 	struct mmu_update *mmu_update;
2398 };
2399 
2400 static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2401 {
2402 	struct remap_data *rmd = data;
2403 	pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2404 
2405 	/*
2406 	 * If we have a contiguous range, just update the pfn itself,
2407 	 * else update pointer to be "next pfn".
2408 	 */
2409 	if (rmd->contiguous)
2410 		(*rmd->pfn)++;
2411 	else
2412 		rmd->pfn++;
2413 
2414 	rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2415 	rmd->mmu_update->ptr |= rmd->no_translate ?
2416 		MMU_PT_UPDATE_NO_TRANSLATE :
2417 		MMU_NORMAL_PT_UPDATE;
2418 	rmd->mmu_update->val = pte_val_ma(pte);
2419 	rmd->mmu_update++;
2420 
2421 	return 0;
2422 }
2423 
2424 int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2425 		  xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2426 		  unsigned int domid, bool no_translate)
2427 {
2428 	int err = 0;
2429 	struct remap_data rmd;
2430 	struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2431 	unsigned long range;
2432 	int mapped = 0;
2433 
2434 	BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2435 
2436 	rmd.pfn = pfn;
2437 	rmd.prot = prot;
2438 	/*
2439 	 * We use the err_ptr to indicate if there we are doing a contiguous
2440 	 * mapping or a discontiguous mapping.
2441 	 */
2442 	rmd.contiguous = !err_ptr;
2443 	rmd.no_translate = no_translate;
2444 
2445 	while (nr) {
2446 		int index = 0;
2447 		int done = 0;
2448 		int batch = min(REMAP_BATCH_SIZE, nr);
2449 		int batch_left = batch;
2450 
2451 		range = (unsigned long)batch << PAGE_SHIFT;
2452 
2453 		rmd.mmu_update = mmu_update;
2454 		err = apply_to_page_range(vma->vm_mm, addr, range,
2455 					  remap_area_pfn_pte_fn, &rmd);
2456 		if (err)
2457 			goto out;
2458 
2459 		/*
2460 		 * We record the error for each page that gives an error, but
2461 		 * continue mapping until the whole set is done
2462 		 */
2463 		do {
2464 			int i;
2465 
2466 			err = HYPERVISOR_mmu_update(&mmu_update[index],
2467 						    batch_left, &done, domid);
2468 
2469 			/*
2470 			 * @err_ptr may be the same buffer as @gfn, so
2471 			 * only clear it after each chunk of @gfn is
2472 			 * used.
2473 			 */
2474 			if (err_ptr) {
2475 				for (i = index; i < index + done; i++)
2476 					err_ptr[i] = 0;
2477 			}
2478 			if (err < 0) {
2479 				if (!err_ptr)
2480 					goto out;
2481 				err_ptr[i] = err;
2482 				done++; /* Skip failed frame. */
2483 			} else
2484 				mapped += done;
2485 			batch_left -= done;
2486 			index += done;
2487 		} while (batch_left);
2488 
2489 		nr -= batch;
2490 		addr += range;
2491 		if (err_ptr)
2492 			err_ptr += batch;
2493 		cond_resched();
2494 	}
2495 out:
2496 
2497 	xen_flush_tlb_all();
2498 
2499 	return err < 0 ? err : mapped;
2500 }
2501 EXPORT_SYMBOL_GPL(xen_remap_pfn);
2502 
2503 #ifdef CONFIG_KEXEC_CORE
2504 phys_addr_t paddr_vmcoreinfo_note(void)
2505 {
2506 	if (xen_pv_domain())
2507 		return virt_to_machine(vmcoreinfo_note).maddr;
2508 	else
2509 		return __pa(vmcoreinfo_note);
2510 }
2511 #endif /* CONFIG_KEXEC_CORE */
2512