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 #include <xen/swiotlb-xen.h>
84 
85 #include "multicalls.h"
86 #include "mmu.h"
87 #include "debugfs.h"
88 
89 #ifdef CONFIG_X86_VSYSCALL_EMULATION
90 /* l3 pud for userspace vsyscall mapping */
91 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
92 #endif
93 
94 /*
95  * Protects atomic reservation decrease/increase against concurrent increases.
96  * Also protects non-atomic updates of current_pages and balloon lists.
97  */
98 static DEFINE_SPINLOCK(xen_reservation_lock);
99 
100 /*
101  * Note about cr3 (pagetable base) values:
102  *
103  * xen_cr3 contains the current logical cr3 value; it contains the
104  * last set cr3.  This may not be the current effective cr3, because
105  * its update may be being lazily deferred.  However, a vcpu looking
106  * at its own cr3 can use this value knowing that it everything will
107  * be self-consistent.
108  *
109  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
110  * hypercall to set the vcpu cr3 is complete (so it may be a little
111  * out of date, but it will never be set early).  If one vcpu is
112  * looking at another vcpu's cr3 value, it should use this variable.
113  */
114 DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
115 DEFINE_PER_CPU(unsigned long, xen_current_cr3);	 /* actual vcpu cr3 */
116 
117 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
118 
119 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
120 
121 /*
122  * Just beyond the highest usermode address.  STACK_TOP_MAX has a
123  * redzone above it, so round it up to a PGD boundary.
124  */
125 #define USER_LIMIT	((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
126 
make_lowmem_page_readonly(void * vaddr)127 void make_lowmem_page_readonly(void *vaddr)
128 {
129 	pte_t *pte, ptev;
130 	unsigned long address = (unsigned long)vaddr;
131 	unsigned int level;
132 
133 	pte = lookup_address(address, &level);
134 	if (pte == NULL)
135 		return;		/* vaddr missing */
136 
137 	ptev = pte_wrprotect(*pte);
138 
139 	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
140 		BUG();
141 }
142 
make_lowmem_page_readwrite(void * vaddr)143 void make_lowmem_page_readwrite(void *vaddr)
144 {
145 	pte_t *pte, ptev;
146 	unsigned long address = (unsigned long)vaddr;
147 	unsigned int level;
148 
149 	pte = lookup_address(address, &level);
150 	if (pte == NULL)
151 		return;		/* vaddr missing */
152 
153 	ptev = pte_mkwrite(*pte);
154 
155 	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
156 		BUG();
157 }
158 
159 
160 /*
161  * During early boot all page table pages are pinned, but we do not have struct
162  * pages, so return true until struct pages are ready.
163  */
xen_page_pinned(void * ptr)164 static bool xen_page_pinned(void *ptr)
165 {
166 	if (static_branch_likely(&xen_struct_pages_ready)) {
167 		struct page *page = virt_to_page(ptr);
168 
169 		return PagePinned(page);
170 	}
171 	return true;
172 }
173 
xen_extend_mmu_update(const struct mmu_update * update)174 static void xen_extend_mmu_update(const struct mmu_update *update)
175 {
176 	struct multicall_space mcs;
177 	struct mmu_update *u;
178 
179 	mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
180 
181 	if (mcs.mc != NULL) {
182 		mcs.mc->args[1]++;
183 	} else {
184 		mcs = __xen_mc_entry(sizeof(*u));
185 		MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
186 	}
187 
188 	u = mcs.args;
189 	*u = *update;
190 }
191 
xen_extend_mmuext_op(const struct mmuext_op * op)192 static void xen_extend_mmuext_op(const struct mmuext_op *op)
193 {
194 	struct multicall_space mcs;
195 	struct mmuext_op *u;
196 
197 	mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
198 
199 	if (mcs.mc != NULL) {
200 		mcs.mc->args[1]++;
201 	} else {
202 		mcs = __xen_mc_entry(sizeof(*u));
203 		MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
204 	}
205 
206 	u = mcs.args;
207 	*u = *op;
208 }
209 
xen_set_pmd_hyper(pmd_t * ptr,pmd_t val)210 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
211 {
212 	struct mmu_update u;
213 
214 	preempt_disable();
215 
216 	xen_mc_batch();
217 
218 	/* ptr may be ioremapped for 64-bit pagetable setup */
219 	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
220 	u.val = pmd_val_ma(val);
221 	xen_extend_mmu_update(&u);
222 
223 	xen_mc_issue(PARAVIRT_LAZY_MMU);
224 
225 	preempt_enable();
226 }
227 
xen_set_pmd(pmd_t * ptr,pmd_t val)228 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
229 {
230 	trace_xen_mmu_set_pmd(ptr, val);
231 
232 	/* If page is not pinned, we can just update the entry
233 	   directly */
234 	if (!xen_page_pinned(ptr)) {
235 		*ptr = val;
236 		return;
237 	}
238 
239 	xen_set_pmd_hyper(ptr, val);
240 }
241 
242 /*
243  * Associate a virtual page frame with a given physical page frame
244  * and protection flags for that frame.
245  */
set_pte_mfn(unsigned long vaddr,unsigned long mfn,pgprot_t flags)246 void __init set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
247 {
248 	if (HYPERVISOR_update_va_mapping(vaddr, mfn_pte(mfn, flags),
249 					 UVMF_INVLPG))
250 		BUG();
251 }
252 
xen_batched_set_pte(pte_t * ptep,pte_t pteval)253 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
254 {
255 	struct mmu_update u;
256 
257 	if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
258 		return false;
259 
260 	xen_mc_batch();
261 
262 	u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
263 	u.val = pte_val_ma(pteval);
264 	xen_extend_mmu_update(&u);
265 
266 	xen_mc_issue(PARAVIRT_LAZY_MMU);
267 
268 	return true;
269 }
270 
__xen_set_pte(pte_t * ptep,pte_t pteval)271 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
272 {
273 	if (!xen_batched_set_pte(ptep, pteval)) {
274 		/*
275 		 * Could call native_set_pte() here and trap and
276 		 * emulate the PTE write, but a hypercall is much cheaper.
277 		 */
278 		struct mmu_update u;
279 
280 		u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
281 		u.val = pte_val_ma(pteval);
282 		HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
283 	}
284 }
285 
xen_set_pte(pte_t * ptep,pte_t pteval)286 static void xen_set_pte(pte_t *ptep, pte_t pteval)
287 {
288 	trace_xen_mmu_set_pte(ptep, pteval);
289 	__xen_set_pte(ptep, pteval);
290 }
291 
xen_ptep_modify_prot_start(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep)292 pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
293 				 unsigned long addr, pte_t *ptep)
294 {
295 	/* Just return the pte as-is.  We preserve the bits on commit */
296 	trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
297 	return *ptep;
298 }
299 
xen_ptep_modify_prot_commit(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep,pte_t pte)300 void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
301 				 pte_t *ptep, pte_t pte)
302 {
303 	struct mmu_update u;
304 
305 	trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
306 	xen_mc_batch();
307 
308 	u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
309 	u.val = pte_val_ma(pte);
310 	xen_extend_mmu_update(&u);
311 
312 	xen_mc_issue(PARAVIRT_LAZY_MMU);
313 }
314 
315 /* Assume pteval_t is equivalent to all the other *val_t types. */
pte_mfn_to_pfn(pteval_t val)316 static pteval_t pte_mfn_to_pfn(pteval_t val)
317 {
318 	if (val & _PAGE_PRESENT) {
319 		unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
320 		unsigned long pfn = mfn_to_pfn(mfn);
321 
322 		pteval_t flags = val & PTE_FLAGS_MASK;
323 		if (unlikely(pfn == ~0))
324 			val = flags & ~_PAGE_PRESENT;
325 		else
326 			val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
327 	}
328 
329 	return val;
330 }
331 
pte_pfn_to_mfn(pteval_t val)332 static pteval_t pte_pfn_to_mfn(pteval_t val)
333 {
334 	if (val & _PAGE_PRESENT) {
335 		unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
336 		pteval_t flags = val & PTE_FLAGS_MASK;
337 		unsigned long mfn;
338 
339 		mfn = __pfn_to_mfn(pfn);
340 
341 		/*
342 		 * If there's no mfn for the pfn, then just create an
343 		 * empty non-present pte.  Unfortunately this loses
344 		 * information about the original pfn, so
345 		 * pte_mfn_to_pfn is asymmetric.
346 		 */
347 		if (unlikely(mfn == INVALID_P2M_ENTRY)) {
348 			mfn = 0;
349 			flags = 0;
350 		} else
351 			mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
352 		val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
353 	}
354 
355 	return val;
356 }
357 
xen_pte_val(pte_t pte)358 __visible pteval_t xen_pte_val(pte_t pte)
359 {
360 	pteval_t pteval = pte.pte;
361 
362 	return pte_mfn_to_pfn(pteval);
363 }
364 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
365 
xen_pgd_val(pgd_t pgd)366 __visible pgdval_t xen_pgd_val(pgd_t pgd)
367 {
368 	return pte_mfn_to_pfn(pgd.pgd);
369 }
370 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
371 
xen_make_pte(pteval_t pte)372 __visible pte_t xen_make_pte(pteval_t pte)
373 {
374 	pte = pte_pfn_to_mfn(pte);
375 
376 	return native_make_pte(pte);
377 }
378 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
379 
xen_make_pgd(pgdval_t pgd)380 __visible pgd_t xen_make_pgd(pgdval_t pgd)
381 {
382 	pgd = pte_pfn_to_mfn(pgd);
383 	return native_make_pgd(pgd);
384 }
385 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
386 
xen_pmd_val(pmd_t pmd)387 __visible pmdval_t xen_pmd_val(pmd_t pmd)
388 {
389 	return pte_mfn_to_pfn(pmd.pmd);
390 }
391 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
392 
xen_set_pud_hyper(pud_t * ptr,pud_t val)393 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
394 {
395 	struct mmu_update u;
396 
397 	preempt_disable();
398 
399 	xen_mc_batch();
400 
401 	/* ptr may be ioremapped for 64-bit pagetable setup */
402 	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
403 	u.val = pud_val_ma(val);
404 	xen_extend_mmu_update(&u);
405 
406 	xen_mc_issue(PARAVIRT_LAZY_MMU);
407 
408 	preempt_enable();
409 }
410 
xen_set_pud(pud_t * ptr,pud_t val)411 static void xen_set_pud(pud_t *ptr, pud_t val)
412 {
413 	trace_xen_mmu_set_pud(ptr, val);
414 
415 	/* If page is not pinned, we can just update the entry
416 	   directly */
417 	if (!xen_page_pinned(ptr)) {
418 		*ptr = val;
419 		return;
420 	}
421 
422 	xen_set_pud_hyper(ptr, val);
423 }
424 
xen_make_pmd(pmdval_t pmd)425 __visible pmd_t xen_make_pmd(pmdval_t pmd)
426 {
427 	pmd = pte_pfn_to_mfn(pmd);
428 	return native_make_pmd(pmd);
429 }
430 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
431 
xen_pud_val(pud_t pud)432 __visible pudval_t xen_pud_val(pud_t pud)
433 {
434 	return pte_mfn_to_pfn(pud.pud);
435 }
436 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
437 
xen_make_pud(pudval_t pud)438 __visible pud_t xen_make_pud(pudval_t pud)
439 {
440 	pud = pte_pfn_to_mfn(pud);
441 
442 	return native_make_pud(pud);
443 }
444 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
445 
xen_get_user_pgd(pgd_t * pgd)446 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
447 {
448 	pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
449 	unsigned offset = pgd - pgd_page;
450 	pgd_t *user_ptr = NULL;
451 
452 	if (offset < pgd_index(USER_LIMIT)) {
453 		struct page *page = virt_to_page(pgd_page);
454 		user_ptr = (pgd_t *)page->private;
455 		if (user_ptr)
456 			user_ptr += offset;
457 	}
458 
459 	return user_ptr;
460 }
461 
__xen_set_p4d_hyper(p4d_t * ptr,p4d_t val)462 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
463 {
464 	struct mmu_update u;
465 
466 	u.ptr = virt_to_machine(ptr).maddr;
467 	u.val = p4d_val_ma(val);
468 	xen_extend_mmu_update(&u);
469 }
470 
471 /*
472  * Raw hypercall-based set_p4d, intended for in early boot before
473  * there's a page structure.  This implies:
474  *  1. The only existing pagetable is the kernel's
475  *  2. It is always pinned
476  *  3. It has no user pagetable attached to it
477  */
xen_set_p4d_hyper(p4d_t * ptr,p4d_t val)478 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
479 {
480 	preempt_disable();
481 
482 	xen_mc_batch();
483 
484 	__xen_set_p4d_hyper(ptr, val);
485 
486 	xen_mc_issue(PARAVIRT_LAZY_MMU);
487 
488 	preempt_enable();
489 }
490 
xen_set_p4d(p4d_t * ptr,p4d_t val)491 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
492 {
493 	pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
494 	pgd_t pgd_val;
495 
496 	trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
497 
498 	/* If page is not pinned, we can just update the entry
499 	   directly */
500 	if (!xen_page_pinned(ptr)) {
501 		*ptr = val;
502 		if (user_ptr) {
503 			WARN_ON(xen_page_pinned(user_ptr));
504 			pgd_val.pgd = p4d_val_ma(val);
505 			*user_ptr = pgd_val;
506 		}
507 		return;
508 	}
509 
510 	/* If it's pinned, then we can at least batch the kernel and
511 	   user updates together. */
512 	xen_mc_batch();
513 
514 	__xen_set_p4d_hyper(ptr, val);
515 	if (user_ptr)
516 		__xen_set_p4d_hyper((p4d_t *)user_ptr, val);
517 
518 	xen_mc_issue(PARAVIRT_LAZY_MMU);
519 }
520 
521 #if CONFIG_PGTABLE_LEVELS >= 5
xen_p4d_val(p4d_t p4d)522 __visible p4dval_t xen_p4d_val(p4d_t p4d)
523 {
524 	return pte_mfn_to_pfn(p4d.p4d);
525 }
526 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
527 
xen_make_p4d(p4dval_t p4d)528 __visible p4d_t xen_make_p4d(p4dval_t p4d)
529 {
530 	p4d = pte_pfn_to_mfn(p4d);
531 
532 	return native_make_p4d(p4d);
533 }
534 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
535 #endif  /* CONFIG_PGTABLE_LEVELS >= 5 */
536 
xen_pmd_walk(struct mm_struct * mm,pmd_t * pmd,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),bool last,unsigned long limit)537 static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
538 			 void (*func)(struct mm_struct *mm, struct page *,
539 				      enum pt_level),
540 			 bool last, unsigned long limit)
541 {
542 	int i, nr;
543 
544 	nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
545 	for (i = 0; i < nr; i++) {
546 		if (!pmd_none(pmd[i]))
547 			(*func)(mm, pmd_page(pmd[i]), PT_PTE);
548 	}
549 }
550 
xen_pud_walk(struct mm_struct * mm,pud_t * pud,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),bool last,unsigned long limit)551 static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
552 			 void (*func)(struct mm_struct *mm, struct page *,
553 				      enum pt_level),
554 			 bool last, unsigned long limit)
555 {
556 	int i, nr;
557 
558 	nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
559 	for (i = 0; i < nr; i++) {
560 		pmd_t *pmd;
561 
562 		if (pud_none(pud[i]))
563 			continue;
564 
565 		pmd = pmd_offset(&pud[i], 0);
566 		if (PTRS_PER_PMD > 1)
567 			(*func)(mm, virt_to_page(pmd), PT_PMD);
568 		xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
569 	}
570 }
571 
xen_p4d_walk(struct mm_struct * mm,p4d_t * p4d,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),bool last,unsigned long limit)572 static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
573 			 void (*func)(struct mm_struct *mm, struct page *,
574 				      enum pt_level),
575 			 bool last, unsigned long limit)
576 {
577 	pud_t *pud;
578 
579 
580 	if (p4d_none(*p4d))
581 		return;
582 
583 	pud = pud_offset(p4d, 0);
584 	if (PTRS_PER_PUD > 1)
585 		(*func)(mm, virt_to_page(pud), PT_PUD);
586 	xen_pud_walk(mm, pud, func, last, limit);
587 }
588 
589 /*
590  * (Yet another) pagetable walker.  This one is intended for pinning a
591  * pagetable.  This means that it walks a pagetable and calls the
592  * callback function on each page it finds making up the page table,
593  * at every level.  It walks the entire pagetable, but it only bothers
594  * pinning pte pages which are below limit.  In the normal case this
595  * will be STACK_TOP_MAX, but at boot we need to pin up to
596  * FIXADDR_TOP.
597  *
598  * We must skip the Xen hole in the middle of the address space, just after
599  * the big x86-64 virtual hole.
600  */
__xen_pgd_walk(struct mm_struct * mm,pgd_t * pgd,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),unsigned long limit)601 static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
602 			   void (*func)(struct mm_struct *mm, struct page *,
603 					enum pt_level),
604 			   unsigned long limit)
605 {
606 	int i, nr;
607 	unsigned hole_low = 0, hole_high = 0;
608 
609 	/* The limit is the last byte to be touched */
610 	limit--;
611 	BUG_ON(limit >= FIXADDR_TOP);
612 
613 	/*
614 	 * 64-bit has a great big hole in the middle of the address
615 	 * space, which contains the Xen mappings.
616 	 */
617 	hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
618 	hole_high = pgd_index(GUARD_HOLE_END_ADDR);
619 
620 	nr = pgd_index(limit) + 1;
621 	for (i = 0; i < nr; i++) {
622 		p4d_t *p4d;
623 
624 		if (i >= hole_low && i < hole_high)
625 			continue;
626 
627 		if (pgd_none(pgd[i]))
628 			continue;
629 
630 		p4d = p4d_offset(&pgd[i], 0);
631 		xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
632 	}
633 
634 	/* Do the top level last, so that the callbacks can use it as
635 	   a cue to do final things like tlb flushes. */
636 	(*func)(mm, virt_to_page(pgd), PT_PGD);
637 }
638 
xen_pgd_walk(struct mm_struct * mm,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),unsigned long limit)639 static void xen_pgd_walk(struct mm_struct *mm,
640 			 void (*func)(struct mm_struct *mm, struct page *,
641 				      enum pt_level),
642 			 unsigned long limit)
643 {
644 	__xen_pgd_walk(mm, mm->pgd, func, limit);
645 }
646 
647 /* If we're using split pte locks, then take the page's lock and
648    return a pointer to it.  Otherwise return NULL. */
xen_pte_lock(struct page * page,struct mm_struct * mm)649 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
650 {
651 	spinlock_t *ptl = NULL;
652 
653 #if USE_SPLIT_PTE_PTLOCKS
654 	ptl = ptlock_ptr(page);
655 	spin_lock_nest_lock(ptl, &mm->page_table_lock);
656 #endif
657 
658 	return ptl;
659 }
660 
xen_pte_unlock(void * v)661 static void xen_pte_unlock(void *v)
662 {
663 	spinlock_t *ptl = v;
664 	spin_unlock(ptl);
665 }
666 
xen_do_pin(unsigned level,unsigned long pfn)667 static void xen_do_pin(unsigned level, unsigned long pfn)
668 {
669 	struct mmuext_op op;
670 
671 	op.cmd = level;
672 	op.arg1.mfn = pfn_to_mfn(pfn);
673 
674 	xen_extend_mmuext_op(&op);
675 }
676 
xen_pin_page(struct mm_struct * mm,struct page * page,enum pt_level level)677 static void xen_pin_page(struct mm_struct *mm, struct page *page,
678 			 enum pt_level level)
679 {
680 	unsigned pgfl = TestSetPagePinned(page);
681 
682 	if (!pgfl) {
683 		void *pt = lowmem_page_address(page);
684 		unsigned long pfn = page_to_pfn(page);
685 		struct multicall_space mcs = __xen_mc_entry(0);
686 		spinlock_t *ptl;
687 
688 		/*
689 		 * We need to hold the pagetable lock between the time
690 		 * we make the pagetable RO and when we actually pin
691 		 * it.  If we don't, then other users may come in and
692 		 * attempt to update the pagetable by writing it,
693 		 * which will fail because the memory is RO but not
694 		 * pinned, so Xen won't do the trap'n'emulate.
695 		 *
696 		 * If we're using split pte locks, we can't hold the
697 		 * entire pagetable's worth of locks during the
698 		 * traverse, because we may wrap the preempt count (8
699 		 * bits).  The solution is to mark RO and pin each PTE
700 		 * page while holding the lock.  This means the number
701 		 * of locks we end up holding is never more than a
702 		 * batch size (~32 entries, at present).
703 		 *
704 		 * If we're not using split pte locks, we needn't pin
705 		 * the PTE pages independently, because we're
706 		 * protected by the overall pagetable lock.
707 		 */
708 		ptl = NULL;
709 		if (level == PT_PTE)
710 			ptl = xen_pte_lock(page, mm);
711 
712 		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
713 					pfn_pte(pfn, PAGE_KERNEL_RO),
714 					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
715 
716 		if (ptl) {
717 			xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
718 
719 			/* Queue a deferred unlock for when this batch
720 			   is completed. */
721 			xen_mc_callback(xen_pte_unlock, ptl);
722 		}
723 	}
724 }
725 
726 /* This is called just after a mm has been created, but it has not
727    been used yet.  We need to make sure that its pagetable is all
728    read-only, and can be pinned. */
__xen_pgd_pin(struct mm_struct * mm,pgd_t * pgd)729 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
730 {
731 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
732 
733 	trace_xen_mmu_pgd_pin(mm, pgd);
734 
735 	xen_mc_batch();
736 
737 	__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
738 
739 	xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
740 
741 	if (user_pgd) {
742 		xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
743 		xen_do_pin(MMUEXT_PIN_L4_TABLE,
744 			   PFN_DOWN(__pa(user_pgd)));
745 	}
746 
747 	xen_mc_issue(0);
748 }
749 
xen_pgd_pin(struct mm_struct * mm)750 static void xen_pgd_pin(struct mm_struct *mm)
751 {
752 	__xen_pgd_pin(mm, mm->pgd);
753 }
754 
755 /*
756  * On save, we need to pin all pagetables to make sure they get their
757  * mfns turned into pfns.  Search the list for any unpinned pgds and pin
758  * them (unpinned pgds are not currently in use, probably because the
759  * process is under construction or destruction).
760  *
761  * Expected to be called in stop_machine() ("equivalent to taking
762  * every spinlock in the system"), so the locking doesn't really
763  * matter all that much.
764  */
xen_mm_pin_all(void)765 void xen_mm_pin_all(void)
766 {
767 	struct page *page;
768 
769 	spin_lock(&pgd_lock);
770 
771 	list_for_each_entry(page, &pgd_list, lru) {
772 		if (!PagePinned(page)) {
773 			__xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
774 			SetPageSavePinned(page);
775 		}
776 	}
777 
778 	spin_unlock(&pgd_lock);
779 }
780 
xen_mark_pinned(struct mm_struct * mm,struct page * page,enum pt_level level)781 static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
782 				   enum pt_level level)
783 {
784 	SetPagePinned(page);
785 }
786 
787 /*
788  * The init_mm pagetable is really pinned as soon as its created, but
789  * that's before we have page structures to store the bits.  So do all
790  * the book-keeping now once struct pages for allocated pages are
791  * initialized. This happens only after memblock_free_all() is called.
792  */
xen_after_bootmem(void)793 static void __init xen_after_bootmem(void)
794 {
795 	static_branch_enable(&xen_struct_pages_ready);
796 #ifdef CONFIG_X86_VSYSCALL_EMULATION
797 	SetPagePinned(virt_to_page(level3_user_vsyscall));
798 #endif
799 	xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
800 }
801 
xen_unpin_page(struct mm_struct * mm,struct page * page,enum pt_level level)802 static void xen_unpin_page(struct mm_struct *mm, struct page *page,
803 			   enum pt_level level)
804 {
805 	unsigned pgfl = TestClearPagePinned(page);
806 
807 	if (pgfl) {
808 		void *pt = lowmem_page_address(page);
809 		unsigned long pfn = page_to_pfn(page);
810 		spinlock_t *ptl = NULL;
811 		struct multicall_space mcs;
812 
813 		/*
814 		 * Do the converse to pin_page.  If we're using split
815 		 * pte locks, we must be holding the lock for while
816 		 * the pte page is unpinned but still RO to prevent
817 		 * concurrent updates from seeing it in this
818 		 * partially-pinned state.
819 		 */
820 		if (level == PT_PTE) {
821 			ptl = xen_pte_lock(page, mm);
822 
823 			if (ptl)
824 				xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
825 		}
826 
827 		mcs = __xen_mc_entry(0);
828 
829 		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
830 					pfn_pte(pfn, PAGE_KERNEL),
831 					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
832 
833 		if (ptl) {
834 			/* unlock when batch completed */
835 			xen_mc_callback(xen_pte_unlock, ptl);
836 		}
837 	}
838 }
839 
840 /* Release a pagetables pages back as normal RW */
__xen_pgd_unpin(struct mm_struct * mm,pgd_t * pgd)841 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
842 {
843 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
844 
845 	trace_xen_mmu_pgd_unpin(mm, pgd);
846 
847 	xen_mc_batch();
848 
849 	xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
850 
851 	if (user_pgd) {
852 		xen_do_pin(MMUEXT_UNPIN_TABLE,
853 			   PFN_DOWN(__pa(user_pgd)));
854 		xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
855 	}
856 
857 	__xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
858 
859 	xen_mc_issue(0);
860 }
861 
xen_pgd_unpin(struct mm_struct * mm)862 static void xen_pgd_unpin(struct mm_struct *mm)
863 {
864 	__xen_pgd_unpin(mm, mm->pgd);
865 }
866 
867 /*
868  * On resume, undo any pinning done at save, so that the rest of the
869  * kernel doesn't see any unexpected pinned pagetables.
870  */
xen_mm_unpin_all(void)871 void xen_mm_unpin_all(void)
872 {
873 	struct page *page;
874 
875 	spin_lock(&pgd_lock);
876 
877 	list_for_each_entry(page, &pgd_list, lru) {
878 		if (PageSavePinned(page)) {
879 			BUG_ON(!PagePinned(page));
880 			__xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
881 			ClearPageSavePinned(page);
882 		}
883 	}
884 
885 	spin_unlock(&pgd_lock);
886 }
887 
xen_activate_mm(struct mm_struct * prev,struct mm_struct * next)888 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
889 {
890 	spin_lock(&next->page_table_lock);
891 	xen_pgd_pin(next);
892 	spin_unlock(&next->page_table_lock);
893 }
894 
xen_dup_mmap(struct mm_struct * oldmm,struct mm_struct * mm)895 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
896 {
897 	spin_lock(&mm->page_table_lock);
898 	xen_pgd_pin(mm);
899 	spin_unlock(&mm->page_table_lock);
900 }
901 
drop_mm_ref_this_cpu(void * info)902 static void drop_mm_ref_this_cpu(void *info)
903 {
904 	struct mm_struct *mm = info;
905 
906 	if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
907 		leave_mm(smp_processor_id());
908 
909 	/*
910 	 * If this cpu still has a stale cr3 reference, then make sure
911 	 * it has been flushed.
912 	 */
913 	if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
914 		xen_mc_flush();
915 }
916 
917 #ifdef CONFIG_SMP
918 /*
919  * Another cpu may still have their %cr3 pointing at the pagetable, so
920  * we need to repoint it somewhere else before we can unpin it.
921  */
xen_drop_mm_ref(struct mm_struct * mm)922 static void xen_drop_mm_ref(struct mm_struct *mm)
923 {
924 	cpumask_var_t mask;
925 	unsigned cpu;
926 
927 	drop_mm_ref_this_cpu(mm);
928 
929 	/* Get the "official" set of cpus referring to our pagetable. */
930 	if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
931 		for_each_online_cpu(cpu) {
932 			if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
933 				continue;
934 			smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
935 		}
936 		return;
937 	}
938 
939 	/*
940 	 * It's possible that a vcpu may have a stale reference to our
941 	 * cr3, because its in lazy mode, and it hasn't yet flushed
942 	 * its set of pending hypercalls yet.  In this case, we can
943 	 * look at its actual current cr3 value, and force it to flush
944 	 * if needed.
945 	 */
946 	cpumask_clear(mask);
947 	for_each_online_cpu(cpu) {
948 		if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
949 			cpumask_set_cpu(cpu, mask);
950 	}
951 
952 	smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
953 	free_cpumask_var(mask);
954 }
955 #else
xen_drop_mm_ref(struct mm_struct * mm)956 static void xen_drop_mm_ref(struct mm_struct *mm)
957 {
958 	drop_mm_ref_this_cpu(mm);
959 }
960 #endif
961 
962 /*
963  * While a process runs, Xen pins its pagetables, which means that the
964  * hypervisor forces it to be read-only, and it controls all updates
965  * to it.  This means that all pagetable updates have to go via the
966  * hypervisor, which is moderately expensive.
967  *
968  * Since we're pulling the pagetable down, we switch to use init_mm,
969  * unpin old process pagetable and mark it all read-write, which
970  * allows further operations on it to be simple memory accesses.
971  *
972  * The only subtle point is that another CPU may be still using the
973  * pagetable because of lazy tlb flushing.  This means we need need to
974  * switch all CPUs off this pagetable before we can unpin it.
975  */
xen_exit_mmap(struct mm_struct * mm)976 static void xen_exit_mmap(struct mm_struct *mm)
977 {
978 	get_cpu();		/* make sure we don't move around */
979 	xen_drop_mm_ref(mm);
980 	put_cpu();
981 
982 	spin_lock(&mm->page_table_lock);
983 
984 	/* pgd may not be pinned in the error exit path of execve */
985 	if (xen_page_pinned(mm->pgd))
986 		xen_pgd_unpin(mm);
987 
988 	spin_unlock(&mm->page_table_lock);
989 }
990 
991 static void xen_post_allocator_init(void);
992 
pin_pagetable_pfn(unsigned cmd,unsigned long pfn)993 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
994 {
995 	struct mmuext_op op;
996 
997 	op.cmd = cmd;
998 	op.arg1.mfn = pfn_to_mfn(pfn);
999 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1000 		BUG();
1001 }
1002 
xen_cleanhighmap(unsigned long vaddr,unsigned long vaddr_end)1003 static void __init xen_cleanhighmap(unsigned long vaddr,
1004 				    unsigned long vaddr_end)
1005 {
1006 	unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1007 	pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1008 
1009 	/* NOTE: The loop is more greedy than the cleanup_highmap variant.
1010 	 * We include the PMD passed in on _both_ boundaries. */
1011 	for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1012 			pmd++, vaddr += PMD_SIZE) {
1013 		if (pmd_none(*pmd))
1014 			continue;
1015 		if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1016 			set_pmd(pmd, __pmd(0));
1017 	}
1018 	/* In case we did something silly, we should crash in this function
1019 	 * instead of somewhere later and be confusing. */
1020 	xen_mc_flush();
1021 }
1022 
1023 /*
1024  * Make a page range writeable and free it.
1025  */
xen_free_ro_pages(unsigned long paddr,unsigned long size)1026 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1027 {
1028 	void *vaddr = __va(paddr);
1029 	void *vaddr_end = vaddr + size;
1030 
1031 	for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1032 		make_lowmem_page_readwrite(vaddr);
1033 
1034 	memblock_phys_free(paddr, size);
1035 }
1036 
xen_cleanmfnmap_free_pgtbl(void * pgtbl,bool unpin)1037 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1038 {
1039 	unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1040 
1041 	if (unpin)
1042 		pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1043 	ClearPagePinned(virt_to_page(__va(pa)));
1044 	xen_free_ro_pages(pa, PAGE_SIZE);
1045 }
1046 
xen_cleanmfnmap_pmd(pmd_t * pmd,bool unpin)1047 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1048 {
1049 	unsigned long pa;
1050 	pte_t *pte_tbl;
1051 	int i;
1052 
1053 	if (pmd_large(*pmd)) {
1054 		pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1055 		xen_free_ro_pages(pa, PMD_SIZE);
1056 		return;
1057 	}
1058 
1059 	pte_tbl = pte_offset_kernel(pmd, 0);
1060 	for (i = 0; i < PTRS_PER_PTE; i++) {
1061 		if (pte_none(pte_tbl[i]))
1062 			continue;
1063 		pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1064 		xen_free_ro_pages(pa, PAGE_SIZE);
1065 	}
1066 	set_pmd(pmd, __pmd(0));
1067 	xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1068 }
1069 
xen_cleanmfnmap_pud(pud_t * pud,bool unpin)1070 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1071 {
1072 	unsigned long pa;
1073 	pmd_t *pmd_tbl;
1074 	int i;
1075 
1076 	if (pud_large(*pud)) {
1077 		pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1078 		xen_free_ro_pages(pa, PUD_SIZE);
1079 		return;
1080 	}
1081 
1082 	pmd_tbl = pmd_offset(pud, 0);
1083 	for (i = 0; i < PTRS_PER_PMD; i++) {
1084 		if (pmd_none(pmd_tbl[i]))
1085 			continue;
1086 		xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1087 	}
1088 	set_pud(pud, __pud(0));
1089 	xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1090 }
1091 
xen_cleanmfnmap_p4d(p4d_t * p4d,bool unpin)1092 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1093 {
1094 	unsigned long pa;
1095 	pud_t *pud_tbl;
1096 	int i;
1097 
1098 	if (p4d_large(*p4d)) {
1099 		pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1100 		xen_free_ro_pages(pa, P4D_SIZE);
1101 		return;
1102 	}
1103 
1104 	pud_tbl = pud_offset(p4d, 0);
1105 	for (i = 0; i < PTRS_PER_PUD; i++) {
1106 		if (pud_none(pud_tbl[i]))
1107 			continue;
1108 		xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1109 	}
1110 	set_p4d(p4d, __p4d(0));
1111 	xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1112 }
1113 
1114 /*
1115  * Since it is well isolated we can (and since it is perhaps large we should)
1116  * also free the page tables mapping the initial P->M table.
1117  */
xen_cleanmfnmap(unsigned long vaddr)1118 static void __init xen_cleanmfnmap(unsigned long vaddr)
1119 {
1120 	pgd_t *pgd;
1121 	p4d_t *p4d;
1122 	bool unpin;
1123 
1124 	unpin = (vaddr == 2 * PGDIR_SIZE);
1125 	vaddr &= PMD_MASK;
1126 	pgd = pgd_offset_k(vaddr);
1127 	p4d = p4d_offset(pgd, 0);
1128 	if (!p4d_none(*p4d))
1129 		xen_cleanmfnmap_p4d(p4d, unpin);
1130 }
1131 
xen_pagetable_p2m_free(void)1132 static void __init xen_pagetable_p2m_free(void)
1133 {
1134 	unsigned long size;
1135 	unsigned long addr;
1136 
1137 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1138 
1139 	/* No memory or already called. */
1140 	if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1141 		return;
1142 
1143 	/* using __ka address and sticking INVALID_P2M_ENTRY! */
1144 	memset((void *)xen_start_info->mfn_list, 0xff, size);
1145 
1146 	addr = xen_start_info->mfn_list;
1147 	/*
1148 	 * We could be in __ka space.
1149 	 * We roundup to the PMD, which means that if anybody at this stage is
1150 	 * using the __ka address of xen_start_info or
1151 	 * xen_start_info->shared_info they are in going to crash. Fortunately
1152 	 * we have already revectored in xen_setup_kernel_pagetable.
1153 	 */
1154 	size = roundup(size, PMD_SIZE);
1155 
1156 	if (addr >= __START_KERNEL_map) {
1157 		xen_cleanhighmap(addr, addr + size);
1158 		size = PAGE_ALIGN(xen_start_info->nr_pages *
1159 				  sizeof(unsigned long));
1160 		memblock_free((void *)addr, size);
1161 	} else {
1162 		xen_cleanmfnmap(addr);
1163 	}
1164 }
1165 
xen_pagetable_cleanhighmap(void)1166 static void __init xen_pagetable_cleanhighmap(void)
1167 {
1168 	unsigned long size;
1169 	unsigned long addr;
1170 
1171 	/* At this stage, cleanup_highmap has already cleaned __ka space
1172 	 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1173 	 * the ramdisk). We continue on, erasing PMD entries that point to page
1174 	 * tables - do note that they are accessible at this stage via __va.
1175 	 * As Xen is aligning the memory end to a 4MB boundary, for good
1176 	 * measure we also round up to PMD_SIZE * 2 - which means that if
1177 	 * anybody is using __ka address to the initial boot-stack - and try
1178 	 * to use it - they are going to crash. The xen_start_info has been
1179 	 * taken care of already in xen_setup_kernel_pagetable. */
1180 	addr = xen_start_info->pt_base;
1181 	size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1182 
1183 	xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1184 	xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1185 }
1186 
xen_pagetable_p2m_setup(void)1187 static void __init xen_pagetable_p2m_setup(void)
1188 {
1189 	xen_vmalloc_p2m_tree();
1190 
1191 	xen_pagetable_p2m_free();
1192 
1193 	xen_pagetable_cleanhighmap();
1194 
1195 	/* And revector! Bye bye old array */
1196 	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1197 }
1198 
xen_pagetable_init(void)1199 static void __init xen_pagetable_init(void)
1200 {
1201 	/*
1202 	 * The majority of further PTE writes is to pagetables already
1203 	 * announced as such to Xen. Hence it is more efficient to use
1204 	 * hypercalls for these updates.
1205 	 */
1206 	pv_ops.mmu.set_pte = __xen_set_pte;
1207 
1208 	paging_init();
1209 	xen_post_allocator_init();
1210 
1211 	xen_pagetable_p2m_setup();
1212 
1213 	/* Allocate and initialize top and mid mfn levels for p2m structure */
1214 	xen_build_mfn_list_list();
1215 
1216 	/* Remap memory freed due to conflicts with E820 map */
1217 	xen_remap_memory();
1218 	xen_setup_mfn_list_list();
1219 }
1220 
xen_write_cr2(unsigned long cr2)1221 static noinstr void xen_write_cr2(unsigned long cr2)
1222 {
1223 	this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1224 }
1225 
xen_flush_tlb(void)1226 static noinline void xen_flush_tlb(void)
1227 {
1228 	struct mmuext_op *op;
1229 	struct multicall_space mcs;
1230 
1231 	preempt_disable();
1232 
1233 	mcs = xen_mc_entry(sizeof(*op));
1234 
1235 	op = mcs.args;
1236 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1237 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1238 
1239 	xen_mc_issue(PARAVIRT_LAZY_MMU);
1240 
1241 	preempt_enable();
1242 }
1243 
xen_flush_tlb_one_user(unsigned long addr)1244 static void xen_flush_tlb_one_user(unsigned long addr)
1245 {
1246 	struct mmuext_op *op;
1247 	struct multicall_space mcs;
1248 
1249 	trace_xen_mmu_flush_tlb_one_user(addr);
1250 
1251 	preempt_disable();
1252 
1253 	mcs = xen_mc_entry(sizeof(*op));
1254 	op = mcs.args;
1255 	op->cmd = MMUEXT_INVLPG_LOCAL;
1256 	op->arg1.linear_addr = addr & PAGE_MASK;
1257 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1258 
1259 	xen_mc_issue(PARAVIRT_LAZY_MMU);
1260 
1261 	preempt_enable();
1262 }
1263 
xen_flush_tlb_multi(const struct cpumask * cpus,const struct flush_tlb_info * info)1264 static void xen_flush_tlb_multi(const struct cpumask *cpus,
1265 				const struct flush_tlb_info *info)
1266 {
1267 	struct {
1268 		struct mmuext_op op;
1269 		DECLARE_BITMAP(mask, NR_CPUS);
1270 	} *args;
1271 	struct multicall_space mcs;
1272 	const size_t mc_entry_size = sizeof(args->op) +
1273 		sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1274 
1275 	trace_xen_mmu_flush_tlb_multi(cpus, info->mm, info->start, info->end);
1276 
1277 	if (cpumask_empty(cpus))
1278 		return;		/* nothing to do */
1279 
1280 	mcs = xen_mc_entry(mc_entry_size);
1281 	args = mcs.args;
1282 	args->op.arg2.vcpumask = to_cpumask(args->mask);
1283 
1284 	/* Remove any offline CPUs */
1285 	cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1286 
1287 	args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1288 	if (info->end != TLB_FLUSH_ALL &&
1289 	    (info->end - info->start) <= PAGE_SIZE) {
1290 		args->op.cmd = MMUEXT_INVLPG_MULTI;
1291 		args->op.arg1.linear_addr = info->start;
1292 	}
1293 
1294 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1295 
1296 	xen_mc_issue(PARAVIRT_LAZY_MMU);
1297 }
1298 
xen_read_cr3(void)1299 static unsigned long xen_read_cr3(void)
1300 {
1301 	return this_cpu_read(xen_cr3);
1302 }
1303 
set_current_cr3(void * v)1304 static void set_current_cr3(void *v)
1305 {
1306 	this_cpu_write(xen_current_cr3, (unsigned long)v);
1307 }
1308 
__xen_write_cr3(bool kernel,unsigned long cr3)1309 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1310 {
1311 	struct mmuext_op op;
1312 	unsigned long mfn;
1313 
1314 	trace_xen_mmu_write_cr3(kernel, cr3);
1315 
1316 	if (cr3)
1317 		mfn = pfn_to_mfn(PFN_DOWN(cr3));
1318 	else
1319 		mfn = 0;
1320 
1321 	WARN_ON(mfn == 0 && kernel);
1322 
1323 	op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1324 	op.arg1.mfn = mfn;
1325 
1326 	xen_extend_mmuext_op(&op);
1327 
1328 	if (kernel) {
1329 		this_cpu_write(xen_cr3, cr3);
1330 
1331 		/* Update xen_current_cr3 once the batch has actually
1332 		   been submitted. */
1333 		xen_mc_callback(set_current_cr3, (void *)cr3);
1334 	}
1335 }
xen_write_cr3(unsigned long cr3)1336 static void xen_write_cr3(unsigned long cr3)
1337 {
1338 	pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1339 
1340 	BUG_ON(preemptible());
1341 
1342 	xen_mc_batch();  /* disables interrupts */
1343 
1344 	/* Update while interrupts are disabled, so its atomic with
1345 	   respect to ipis */
1346 	this_cpu_write(xen_cr3, cr3);
1347 
1348 	__xen_write_cr3(true, cr3);
1349 
1350 	if (user_pgd)
1351 		__xen_write_cr3(false, __pa(user_pgd));
1352 	else
1353 		__xen_write_cr3(false, 0);
1354 
1355 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1356 }
1357 
1358 /*
1359  * At the start of the day - when Xen launches a guest, it has already
1360  * built pagetables for the guest. We diligently look over them
1361  * in xen_setup_kernel_pagetable and graft as appropriate them in the
1362  * init_top_pgt and its friends. Then when we are happy we load
1363  * the new init_top_pgt - and continue on.
1364  *
1365  * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1366  * up the rest of the pagetables. When it has completed it loads the cr3.
1367  * N.B. that baremetal would start at 'start_kernel' (and the early
1368  * #PF handler would create bootstrap pagetables) - so we are running
1369  * with the same assumptions as what to do when write_cr3 is executed
1370  * at this point.
1371  *
1372  * Since there are no user-page tables at all, we have two variants
1373  * of xen_write_cr3 - the early bootup (this one), and the late one
1374  * (xen_write_cr3). The reason we have to do that is that in 64-bit
1375  * the Linux kernel and user-space are both in ring 3 while the
1376  * hypervisor is in ring 0.
1377  */
xen_write_cr3_init(unsigned long cr3)1378 static void __init xen_write_cr3_init(unsigned long cr3)
1379 {
1380 	BUG_ON(preemptible());
1381 
1382 	xen_mc_batch();  /* disables interrupts */
1383 
1384 	/* Update while interrupts are disabled, so its atomic with
1385 	   respect to ipis */
1386 	this_cpu_write(xen_cr3, cr3);
1387 
1388 	__xen_write_cr3(true, cr3);
1389 
1390 	xen_mc_issue(PARAVIRT_LAZY_CPU);  /* interrupts restored */
1391 }
1392 
xen_pgd_alloc(struct mm_struct * mm)1393 static int xen_pgd_alloc(struct mm_struct *mm)
1394 {
1395 	pgd_t *pgd = mm->pgd;
1396 	struct page *page = virt_to_page(pgd);
1397 	pgd_t *user_pgd;
1398 	int ret = -ENOMEM;
1399 
1400 	BUG_ON(PagePinned(virt_to_page(pgd)));
1401 	BUG_ON(page->private != 0);
1402 
1403 	user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1404 	page->private = (unsigned long)user_pgd;
1405 
1406 	if (user_pgd != NULL) {
1407 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1408 		user_pgd[pgd_index(VSYSCALL_ADDR)] =
1409 			__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1410 #endif
1411 		ret = 0;
1412 	}
1413 
1414 	BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1415 
1416 	return ret;
1417 }
1418 
xen_pgd_free(struct mm_struct * mm,pgd_t * pgd)1419 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1420 {
1421 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
1422 
1423 	if (user_pgd)
1424 		free_page((unsigned long)user_pgd);
1425 }
1426 
1427 /*
1428  * Init-time set_pte while constructing initial pagetables, which
1429  * doesn't allow RO page table pages to be remapped RW.
1430  *
1431  * If there is no MFN for this PFN then this page is initially
1432  * ballooned out so clear the PTE (as in decrease_reservation() in
1433  * drivers/xen/balloon.c).
1434  *
1435  * Many of these PTE updates are done on unpinned and writable pages
1436  * and doing a hypercall for these is unnecessary and expensive.  At
1437  * this point it is rarely possible to tell if a page is pinned, so
1438  * mostly write the PTE directly and rely on Xen trapping and
1439  * emulating any updates as necessary.
1440  */
xen_set_pte_init(pte_t * ptep,pte_t pte)1441 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1442 {
1443 	if (unlikely(is_early_ioremap_ptep(ptep)))
1444 		__xen_set_pte(ptep, pte);
1445 	else
1446 		native_set_pte(ptep, pte);
1447 }
1448 
xen_make_pte_init(pteval_t pte)1449 __visible pte_t xen_make_pte_init(pteval_t pte)
1450 {
1451 	unsigned long pfn;
1452 
1453 	/*
1454 	 * Pages belonging to the initial p2m list mapped outside the default
1455 	 * address range must be mapped read-only. This region contains the
1456 	 * page tables for mapping the p2m list, too, and page tables MUST be
1457 	 * mapped read-only.
1458 	 */
1459 	pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1460 	if (xen_start_info->mfn_list < __START_KERNEL_map &&
1461 	    pfn >= xen_start_info->first_p2m_pfn &&
1462 	    pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1463 		pte &= ~_PAGE_RW;
1464 
1465 	pte = pte_pfn_to_mfn(pte);
1466 	return native_make_pte(pte);
1467 }
1468 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1469 
1470 /* Early in boot, while setting up the initial pagetable, assume
1471    everything is pinned. */
xen_alloc_pte_init(struct mm_struct * mm,unsigned long pfn)1472 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1473 {
1474 #ifdef CONFIG_FLATMEM
1475 	BUG_ON(mem_map);	/* should only be used early */
1476 #endif
1477 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1478 	pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1479 }
1480 
1481 /* Used for pmd and pud */
xen_alloc_pmd_init(struct mm_struct * mm,unsigned long pfn)1482 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1483 {
1484 #ifdef CONFIG_FLATMEM
1485 	BUG_ON(mem_map);	/* should only be used early */
1486 #endif
1487 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1488 }
1489 
1490 /* Early release_pte assumes that all pts are pinned, since there's
1491    only init_mm and anything attached to that is pinned. */
xen_release_pte_init(unsigned long pfn)1492 static void __init xen_release_pte_init(unsigned long pfn)
1493 {
1494 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1495 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1496 }
1497 
xen_release_pmd_init(unsigned long pfn)1498 static void __init xen_release_pmd_init(unsigned long pfn)
1499 {
1500 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1501 }
1502 
__pin_pagetable_pfn(unsigned cmd,unsigned long pfn)1503 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1504 {
1505 	struct multicall_space mcs;
1506 	struct mmuext_op *op;
1507 
1508 	mcs = __xen_mc_entry(sizeof(*op));
1509 	op = mcs.args;
1510 	op->cmd = cmd;
1511 	op->arg1.mfn = pfn_to_mfn(pfn);
1512 
1513 	MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1514 }
1515 
__set_pfn_prot(unsigned long pfn,pgprot_t prot)1516 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1517 {
1518 	struct multicall_space mcs;
1519 	unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1520 
1521 	mcs = __xen_mc_entry(0);
1522 	MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1523 				pfn_pte(pfn, prot), 0);
1524 }
1525 
1526 /* This needs to make sure the new pte page is pinned iff its being
1527    attached to a pinned pagetable. */
xen_alloc_ptpage(struct mm_struct * mm,unsigned long pfn,unsigned level)1528 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1529 				    unsigned level)
1530 {
1531 	bool pinned = xen_page_pinned(mm->pgd);
1532 
1533 	trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1534 
1535 	if (pinned) {
1536 		struct page *page = pfn_to_page(pfn);
1537 
1538 		pinned = false;
1539 		if (static_branch_likely(&xen_struct_pages_ready)) {
1540 			pinned = PagePinned(page);
1541 			SetPagePinned(page);
1542 		}
1543 
1544 		xen_mc_batch();
1545 
1546 		__set_pfn_prot(pfn, PAGE_KERNEL_RO);
1547 
1548 		if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS && !pinned)
1549 			__pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1550 
1551 		xen_mc_issue(PARAVIRT_LAZY_MMU);
1552 	}
1553 }
1554 
xen_alloc_pte(struct mm_struct * mm,unsigned long pfn)1555 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1556 {
1557 	xen_alloc_ptpage(mm, pfn, PT_PTE);
1558 }
1559 
xen_alloc_pmd(struct mm_struct * mm,unsigned long pfn)1560 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1561 {
1562 	xen_alloc_ptpage(mm, pfn, PT_PMD);
1563 }
1564 
1565 /* This should never happen until we're OK to use struct page */
xen_release_ptpage(unsigned long pfn,unsigned level)1566 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1567 {
1568 	struct page *page = pfn_to_page(pfn);
1569 	bool pinned = PagePinned(page);
1570 
1571 	trace_xen_mmu_release_ptpage(pfn, level, pinned);
1572 
1573 	if (pinned) {
1574 		xen_mc_batch();
1575 
1576 		if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1577 			__pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1578 
1579 		__set_pfn_prot(pfn, PAGE_KERNEL);
1580 
1581 		xen_mc_issue(PARAVIRT_LAZY_MMU);
1582 
1583 		ClearPagePinned(page);
1584 	}
1585 }
1586 
xen_release_pte(unsigned long pfn)1587 static void xen_release_pte(unsigned long pfn)
1588 {
1589 	xen_release_ptpage(pfn, PT_PTE);
1590 }
1591 
xen_release_pmd(unsigned long pfn)1592 static void xen_release_pmd(unsigned long pfn)
1593 {
1594 	xen_release_ptpage(pfn, PT_PMD);
1595 }
1596 
xen_alloc_pud(struct mm_struct * mm,unsigned long pfn)1597 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1598 {
1599 	xen_alloc_ptpage(mm, pfn, PT_PUD);
1600 }
1601 
xen_release_pud(unsigned long pfn)1602 static void xen_release_pud(unsigned long pfn)
1603 {
1604 	xen_release_ptpage(pfn, PT_PUD);
1605 }
1606 
1607 /*
1608  * Like __va(), but returns address in the kernel mapping (which is
1609  * all we have until the physical memory mapping has been set up.
1610  */
__ka(phys_addr_t paddr)1611 static void * __init __ka(phys_addr_t paddr)
1612 {
1613 	return (void *)(paddr + __START_KERNEL_map);
1614 }
1615 
1616 /* Convert a machine address to physical address */
m2p(phys_addr_t maddr)1617 static unsigned long __init m2p(phys_addr_t maddr)
1618 {
1619 	phys_addr_t paddr;
1620 
1621 	maddr &= XEN_PTE_MFN_MASK;
1622 	paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1623 
1624 	return paddr;
1625 }
1626 
1627 /* Convert a machine address to kernel virtual */
m2v(phys_addr_t maddr)1628 static void * __init m2v(phys_addr_t maddr)
1629 {
1630 	return __ka(m2p(maddr));
1631 }
1632 
1633 /* Set the page permissions on an identity-mapped pages */
set_page_prot_flags(void * addr,pgprot_t prot,unsigned long flags)1634 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1635 				       unsigned long flags)
1636 {
1637 	unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1638 	pte_t pte = pfn_pte(pfn, prot);
1639 
1640 	if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1641 		BUG();
1642 }
set_page_prot(void * addr,pgprot_t prot)1643 static void __init set_page_prot(void *addr, pgprot_t prot)
1644 {
1645 	return set_page_prot_flags(addr, prot, UVMF_NONE);
1646 }
1647 
xen_setup_machphys_mapping(void)1648 void __init xen_setup_machphys_mapping(void)
1649 {
1650 	struct xen_machphys_mapping mapping;
1651 
1652 	if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1653 		machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1654 		machine_to_phys_nr = mapping.max_mfn + 1;
1655 	} else {
1656 		machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1657 	}
1658 }
1659 
convert_pfn_mfn(void * v)1660 static void __init convert_pfn_mfn(void *v)
1661 {
1662 	pte_t *pte = v;
1663 	int i;
1664 
1665 	/* All levels are converted the same way, so just treat them
1666 	   as ptes. */
1667 	for (i = 0; i < PTRS_PER_PTE; i++)
1668 		pte[i] = xen_make_pte(pte[i].pte);
1669 }
check_pt_base(unsigned long * pt_base,unsigned long * pt_end,unsigned long addr)1670 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1671 				 unsigned long addr)
1672 {
1673 	if (*pt_base == PFN_DOWN(__pa(addr))) {
1674 		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1675 		clear_page((void *)addr);
1676 		(*pt_base)++;
1677 	}
1678 	if (*pt_end == PFN_DOWN(__pa(addr))) {
1679 		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1680 		clear_page((void *)addr);
1681 		(*pt_end)--;
1682 	}
1683 }
1684 /*
1685  * Set up the initial kernel pagetable.
1686  *
1687  * We can construct this by grafting the Xen provided pagetable into
1688  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1689  * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1690  * kernel has a physical mapping to start with - but that's enough to
1691  * get __va working.  We need to fill in the rest of the physical
1692  * mapping once some sort of allocator has been set up.
1693  */
xen_setup_kernel_pagetable(pgd_t * pgd,unsigned long max_pfn)1694 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1695 {
1696 	pud_t *l3;
1697 	pmd_t *l2;
1698 	unsigned long addr[3];
1699 	unsigned long pt_base, pt_end;
1700 	unsigned i;
1701 
1702 	/* max_pfn_mapped is the last pfn mapped in the initial memory
1703 	 * mappings. Considering that on Xen after the kernel mappings we
1704 	 * have the mappings of some pages that don't exist in pfn space, we
1705 	 * set max_pfn_mapped to the last real pfn mapped. */
1706 	if (xen_start_info->mfn_list < __START_KERNEL_map)
1707 		max_pfn_mapped = xen_start_info->first_p2m_pfn;
1708 	else
1709 		max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1710 
1711 	pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1712 	pt_end = pt_base + xen_start_info->nr_pt_frames;
1713 
1714 	/* Zap identity mapping */
1715 	init_top_pgt[0] = __pgd(0);
1716 
1717 	/* Pre-constructed entries are in pfn, so convert to mfn */
1718 	/* L4[273] -> level3_ident_pgt  */
1719 	/* L4[511] -> level3_kernel_pgt */
1720 	convert_pfn_mfn(init_top_pgt);
1721 
1722 	/* L3_i[0] -> level2_ident_pgt */
1723 	convert_pfn_mfn(level3_ident_pgt);
1724 	/* L3_k[510] -> level2_kernel_pgt */
1725 	/* L3_k[511] -> level2_fixmap_pgt */
1726 	convert_pfn_mfn(level3_kernel_pgt);
1727 
1728 	/* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1729 	convert_pfn_mfn(level2_fixmap_pgt);
1730 
1731 	/* We get [511][511] and have Xen's version of level2_kernel_pgt */
1732 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1733 	l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1734 
1735 	addr[0] = (unsigned long)pgd;
1736 	addr[1] = (unsigned long)l3;
1737 	addr[2] = (unsigned long)l2;
1738 	/* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1739 	 * Both L4[273][0] and L4[511][510] have entries that point to the same
1740 	 * L2 (PMD) tables. Meaning that if you modify it in __va space
1741 	 * it will be also modified in the __ka space! (But if you just
1742 	 * modify the PMD table to point to other PTE's or none, then you
1743 	 * are OK - which is what cleanup_highmap does) */
1744 	copy_page(level2_ident_pgt, l2);
1745 	/* Graft it onto L4[511][510] */
1746 	copy_page(level2_kernel_pgt, l2);
1747 
1748 	/*
1749 	 * Zap execute permission from the ident map. Due to the sharing of
1750 	 * L1 entries we need to do this in the L2.
1751 	 */
1752 	if (__supported_pte_mask & _PAGE_NX) {
1753 		for (i = 0; i < PTRS_PER_PMD; ++i) {
1754 			if (pmd_none(level2_ident_pgt[i]))
1755 				continue;
1756 			level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1757 		}
1758 	}
1759 
1760 	/* Copy the initial P->M table mappings if necessary. */
1761 	i = pgd_index(xen_start_info->mfn_list);
1762 	if (i && i < pgd_index(__START_KERNEL_map))
1763 		init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1764 
1765 	/* Make pagetable pieces RO */
1766 	set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1767 	set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1768 	set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1769 	set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1770 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1771 	set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1772 
1773 	for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1774 		set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1775 			      PAGE_KERNEL_RO);
1776 	}
1777 
1778 	/* Pin down new L4 */
1779 	pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1780 			  PFN_DOWN(__pa_symbol(init_top_pgt)));
1781 
1782 	/* Unpin Xen-provided one */
1783 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1784 
1785 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1786 	/* Pin user vsyscall L3 */
1787 	set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1788 	pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1789 			  PFN_DOWN(__pa_symbol(level3_user_vsyscall)));
1790 #endif
1791 
1792 	/*
1793 	 * At this stage there can be no user pgd, and no page structure to
1794 	 * attach it to, so make sure we just set kernel pgd.
1795 	 */
1796 	xen_mc_batch();
1797 	__xen_write_cr3(true, __pa(init_top_pgt));
1798 	xen_mc_issue(PARAVIRT_LAZY_CPU);
1799 
1800 	/* We can't that easily rip out L3 and L2, as the Xen pagetables are
1801 	 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1802 	 * the initial domain. For guests using the toolstack, they are in:
1803 	 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1804 	 * rip out the [L4] (pgd), but for guests we shave off three pages.
1805 	 */
1806 	for (i = 0; i < ARRAY_SIZE(addr); i++)
1807 		check_pt_base(&pt_base, &pt_end, addr[i]);
1808 
1809 	/* Our (by three pages) smaller Xen pagetable that we are using */
1810 	xen_pt_base = PFN_PHYS(pt_base);
1811 	xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1812 	memblock_reserve(xen_pt_base, xen_pt_size);
1813 
1814 	/* Revector the xen_start_info */
1815 	xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1816 }
1817 
1818 /*
1819  * Read a value from a physical address.
1820  */
xen_read_phys_ulong(phys_addr_t addr)1821 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1822 {
1823 	unsigned long *vaddr;
1824 	unsigned long val;
1825 
1826 	vaddr = early_memremap_ro(addr, sizeof(val));
1827 	val = *vaddr;
1828 	early_memunmap(vaddr, sizeof(val));
1829 	return val;
1830 }
1831 
1832 /*
1833  * Translate a virtual address to a physical one without relying on mapped
1834  * page tables. Don't rely on big pages being aligned in (guest) physical
1835  * space!
1836  */
xen_early_virt_to_phys(unsigned long vaddr)1837 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1838 {
1839 	phys_addr_t pa;
1840 	pgd_t pgd;
1841 	pud_t pud;
1842 	pmd_t pmd;
1843 	pte_t pte;
1844 
1845 	pa = read_cr3_pa();
1846 	pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1847 						       sizeof(pgd)));
1848 	if (!pgd_present(pgd))
1849 		return 0;
1850 
1851 	pa = pgd_val(pgd) & PTE_PFN_MASK;
1852 	pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1853 						       sizeof(pud)));
1854 	if (!pud_present(pud))
1855 		return 0;
1856 	pa = pud_val(pud) & PTE_PFN_MASK;
1857 	if (pud_large(pud))
1858 		return pa + (vaddr & ~PUD_MASK);
1859 
1860 	pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1861 						       sizeof(pmd)));
1862 	if (!pmd_present(pmd))
1863 		return 0;
1864 	pa = pmd_val(pmd) & PTE_PFN_MASK;
1865 	if (pmd_large(pmd))
1866 		return pa + (vaddr & ~PMD_MASK);
1867 
1868 	pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1869 						       sizeof(pte)));
1870 	if (!pte_present(pte))
1871 		return 0;
1872 	pa = pte_pfn(pte) << PAGE_SHIFT;
1873 
1874 	return pa | (vaddr & ~PAGE_MASK);
1875 }
1876 
1877 /*
1878  * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1879  * this area.
1880  */
xen_relocate_p2m(void)1881 void __init xen_relocate_p2m(void)
1882 {
1883 	phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1884 	unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1885 	int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1886 	pte_t *pt;
1887 	pmd_t *pmd;
1888 	pud_t *pud;
1889 	pgd_t *pgd;
1890 	unsigned long *new_p2m;
1891 
1892 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1893 	n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1894 	n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1895 	n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1896 	n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1897 	n_frames = n_pte + n_pt + n_pmd + n_pud;
1898 
1899 	new_area = xen_find_free_area(PFN_PHYS(n_frames));
1900 	if (!new_area) {
1901 		xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1902 		BUG();
1903 	}
1904 
1905 	/*
1906 	 * Setup the page tables for addressing the new p2m list.
1907 	 * We have asked the hypervisor to map the p2m list at the user address
1908 	 * PUD_SIZE. It may have done so, or it may have used a kernel space
1909 	 * address depending on the Xen version.
1910 	 * To avoid any possible virtual address collision, just use
1911 	 * 2 * PUD_SIZE for the new area.
1912 	 */
1913 	pud_phys = new_area;
1914 	pmd_phys = pud_phys + PFN_PHYS(n_pud);
1915 	pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1916 	p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1917 
1918 	pgd = __va(read_cr3_pa());
1919 	new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1920 	for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1921 		pud = early_memremap(pud_phys, PAGE_SIZE);
1922 		clear_page(pud);
1923 		for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1924 				idx_pmd++) {
1925 			pmd = early_memremap(pmd_phys, PAGE_SIZE);
1926 			clear_page(pmd);
1927 			for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1928 					idx_pt++) {
1929 				pt = early_memremap(pt_phys, PAGE_SIZE);
1930 				clear_page(pt);
1931 				for (idx_pte = 0;
1932 				     idx_pte < min(n_pte, PTRS_PER_PTE);
1933 				     idx_pte++) {
1934 					pt[idx_pte] = pfn_pte(p2m_pfn,
1935 							      PAGE_KERNEL);
1936 					p2m_pfn++;
1937 				}
1938 				n_pte -= PTRS_PER_PTE;
1939 				early_memunmap(pt, PAGE_SIZE);
1940 				make_lowmem_page_readonly(__va(pt_phys));
1941 				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
1942 						PFN_DOWN(pt_phys));
1943 				pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
1944 				pt_phys += PAGE_SIZE;
1945 			}
1946 			n_pt -= PTRS_PER_PMD;
1947 			early_memunmap(pmd, PAGE_SIZE);
1948 			make_lowmem_page_readonly(__va(pmd_phys));
1949 			pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
1950 					PFN_DOWN(pmd_phys));
1951 			pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
1952 			pmd_phys += PAGE_SIZE;
1953 		}
1954 		n_pmd -= PTRS_PER_PUD;
1955 		early_memunmap(pud, PAGE_SIZE);
1956 		make_lowmem_page_readonly(__va(pud_phys));
1957 		pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
1958 		set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
1959 		pud_phys += PAGE_SIZE;
1960 	}
1961 
1962 	/* Now copy the old p2m info to the new area. */
1963 	memcpy(new_p2m, xen_p2m_addr, size);
1964 	xen_p2m_addr = new_p2m;
1965 
1966 	/* Release the old p2m list and set new list info. */
1967 	p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
1968 	BUG_ON(!p2m_pfn);
1969 	p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
1970 
1971 	if (xen_start_info->mfn_list < __START_KERNEL_map) {
1972 		pfn = xen_start_info->first_p2m_pfn;
1973 		pfn_end = xen_start_info->first_p2m_pfn +
1974 			  xen_start_info->nr_p2m_frames;
1975 		set_pgd(pgd + 1, __pgd(0));
1976 	} else {
1977 		pfn = p2m_pfn;
1978 		pfn_end = p2m_pfn_end;
1979 	}
1980 
1981 	memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
1982 	while (pfn < pfn_end) {
1983 		if (pfn == p2m_pfn) {
1984 			pfn = p2m_pfn_end;
1985 			continue;
1986 		}
1987 		make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1988 		pfn++;
1989 	}
1990 
1991 	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1992 	xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
1993 	xen_start_info->nr_p2m_frames = n_frames;
1994 }
1995 
xen_reserve_special_pages(void)1996 void __init xen_reserve_special_pages(void)
1997 {
1998 	phys_addr_t paddr;
1999 
2000 	memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2001 	if (xen_start_info->store_mfn) {
2002 		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2003 		memblock_reserve(paddr, PAGE_SIZE);
2004 	}
2005 	if (!xen_initial_domain()) {
2006 		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2007 		memblock_reserve(paddr, PAGE_SIZE);
2008 	}
2009 }
2010 
xen_pt_check_e820(void)2011 void __init xen_pt_check_e820(void)
2012 {
2013 	if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2014 		xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2015 		BUG();
2016 	}
2017 }
2018 
2019 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2020 
xen_set_fixmap(unsigned idx,phys_addr_t phys,pgprot_t prot)2021 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2022 {
2023 	pte_t pte;
2024 	unsigned long vaddr;
2025 
2026 	phys >>= PAGE_SHIFT;
2027 
2028 	switch (idx) {
2029 	case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2030 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2031 	case VSYSCALL_PAGE:
2032 #endif
2033 		/* All local page mappings */
2034 		pte = pfn_pte(phys, prot);
2035 		break;
2036 
2037 #ifdef CONFIG_X86_LOCAL_APIC
2038 	case FIX_APIC_BASE:	/* maps dummy local APIC */
2039 		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2040 		break;
2041 #endif
2042 
2043 #ifdef CONFIG_X86_IO_APIC
2044 	case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2045 		/*
2046 		 * We just don't map the IO APIC - all access is via
2047 		 * hypercalls.  Keep the address in the pte for reference.
2048 		 */
2049 		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2050 		break;
2051 #endif
2052 
2053 	case FIX_PARAVIRT_BOOTMAP:
2054 		/* This is an MFN, but it isn't an IO mapping from the
2055 		   IO domain */
2056 		pte = mfn_pte(phys, prot);
2057 		break;
2058 
2059 	default:
2060 		/* By default, set_fixmap is used for hardware mappings */
2061 		pte = mfn_pte(phys, prot);
2062 		break;
2063 	}
2064 
2065 	vaddr = __fix_to_virt(idx);
2066 	if (HYPERVISOR_update_va_mapping(vaddr, pte, UVMF_INVLPG))
2067 		BUG();
2068 
2069 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2070 	/* Replicate changes to map the vsyscall page into the user
2071 	   pagetable vsyscall mapping. */
2072 	if (idx == VSYSCALL_PAGE)
2073 		set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2074 #endif
2075 }
2076 
xen_post_allocator_init(void)2077 static void __init xen_post_allocator_init(void)
2078 {
2079 	pv_ops.mmu.set_pte = xen_set_pte;
2080 	pv_ops.mmu.set_pmd = xen_set_pmd;
2081 	pv_ops.mmu.set_pud = xen_set_pud;
2082 	pv_ops.mmu.set_p4d = xen_set_p4d;
2083 
2084 	/* This will work as long as patching hasn't happened yet
2085 	   (which it hasn't) */
2086 	pv_ops.mmu.alloc_pte = xen_alloc_pte;
2087 	pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2088 	pv_ops.mmu.release_pte = xen_release_pte;
2089 	pv_ops.mmu.release_pmd = xen_release_pmd;
2090 	pv_ops.mmu.alloc_pud = xen_alloc_pud;
2091 	pv_ops.mmu.release_pud = xen_release_pud;
2092 	pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2093 
2094 	pv_ops.mmu.write_cr3 = &xen_write_cr3;
2095 }
2096 
xen_leave_lazy_mmu(void)2097 static void xen_leave_lazy_mmu(void)
2098 {
2099 	preempt_disable();
2100 	xen_mc_flush();
2101 	paravirt_leave_lazy_mmu();
2102 	preempt_enable();
2103 }
2104 
2105 static const typeof(pv_ops) xen_mmu_ops __initconst = {
2106 	.mmu = {
2107 		.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2108 		.write_cr2 = xen_write_cr2,
2109 
2110 		.read_cr3 = xen_read_cr3,
2111 		.write_cr3 = xen_write_cr3_init,
2112 
2113 		.flush_tlb_user = xen_flush_tlb,
2114 		.flush_tlb_kernel = xen_flush_tlb,
2115 		.flush_tlb_one_user = xen_flush_tlb_one_user,
2116 		.flush_tlb_multi = xen_flush_tlb_multi,
2117 		.tlb_remove_table = tlb_remove_table,
2118 
2119 		.pgd_alloc = xen_pgd_alloc,
2120 		.pgd_free = xen_pgd_free,
2121 
2122 		.alloc_pte = xen_alloc_pte_init,
2123 		.release_pte = xen_release_pte_init,
2124 		.alloc_pmd = xen_alloc_pmd_init,
2125 		.release_pmd = xen_release_pmd_init,
2126 
2127 		.set_pte = xen_set_pte_init,
2128 		.set_pmd = xen_set_pmd_hyper,
2129 
2130 		.ptep_modify_prot_start = xen_ptep_modify_prot_start,
2131 		.ptep_modify_prot_commit = xen_ptep_modify_prot_commit,
2132 
2133 		.pte_val = PV_CALLEE_SAVE(xen_pte_val),
2134 		.pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2135 
2136 		.make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2137 		.make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2138 
2139 		.set_pud = xen_set_pud_hyper,
2140 
2141 		.make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2142 		.pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2143 
2144 		.pud_val = PV_CALLEE_SAVE(xen_pud_val),
2145 		.make_pud = PV_CALLEE_SAVE(xen_make_pud),
2146 		.set_p4d = xen_set_p4d_hyper,
2147 
2148 		.alloc_pud = xen_alloc_pmd_init,
2149 		.release_pud = xen_release_pmd_init,
2150 
2151 #if CONFIG_PGTABLE_LEVELS >= 5
2152 		.p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2153 		.make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2154 #endif
2155 
2156 		.activate_mm = xen_activate_mm,
2157 		.dup_mmap = xen_dup_mmap,
2158 		.exit_mmap = xen_exit_mmap,
2159 
2160 		.lazy_mode = {
2161 			.enter = paravirt_enter_lazy_mmu,
2162 			.leave = xen_leave_lazy_mmu,
2163 			.flush = paravirt_flush_lazy_mmu,
2164 		},
2165 
2166 		.set_fixmap = xen_set_fixmap,
2167 	},
2168 };
2169 
xen_init_mmu_ops(void)2170 void __init xen_init_mmu_ops(void)
2171 {
2172 	x86_init.paging.pagetable_init = xen_pagetable_init;
2173 	x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2174 
2175 	pv_ops.mmu = xen_mmu_ops.mmu;
2176 
2177 	memset(dummy_mapping, 0xff, PAGE_SIZE);
2178 }
2179 
2180 /* Protected by xen_reservation_lock. */
2181 #define MAX_CONTIG_ORDER 9 /* 2MB */
2182 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2183 
2184 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
xen_zap_pfn_range(unsigned long vaddr,unsigned int order,unsigned long * in_frames,unsigned long * out_frames)2185 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2186 				unsigned long *in_frames,
2187 				unsigned long *out_frames)
2188 {
2189 	int i;
2190 	struct multicall_space mcs;
2191 
2192 	xen_mc_batch();
2193 	for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2194 		mcs = __xen_mc_entry(0);
2195 
2196 		if (in_frames)
2197 			in_frames[i] = virt_to_mfn(vaddr);
2198 
2199 		MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2200 		__set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2201 
2202 		if (out_frames)
2203 			out_frames[i] = virt_to_pfn(vaddr);
2204 	}
2205 	xen_mc_issue(0);
2206 }
2207 
2208 /*
2209  * Update the pfn-to-mfn mappings for a virtual address range, either to
2210  * point to an array of mfns, or contiguously from a single starting
2211  * mfn.
2212  */
xen_remap_exchanged_ptes(unsigned long vaddr,int order,unsigned long * mfns,unsigned long first_mfn)2213 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2214 				     unsigned long *mfns,
2215 				     unsigned long first_mfn)
2216 {
2217 	unsigned i, limit;
2218 	unsigned long mfn;
2219 
2220 	xen_mc_batch();
2221 
2222 	limit = 1u << order;
2223 	for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2224 		struct multicall_space mcs;
2225 		unsigned flags;
2226 
2227 		mcs = __xen_mc_entry(0);
2228 		if (mfns)
2229 			mfn = mfns[i];
2230 		else
2231 			mfn = first_mfn + i;
2232 
2233 		if (i < (limit - 1))
2234 			flags = 0;
2235 		else {
2236 			if (order == 0)
2237 				flags = UVMF_INVLPG | UVMF_ALL;
2238 			else
2239 				flags = UVMF_TLB_FLUSH | UVMF_ALL;
2240 		}
2241 
2242 		MULTI_update_va_mapping(mcs.mc, vaddr,
2243 				mfn_pte(mfn, PAGE_KERNEL), flags);
2244 
2245 		set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2246 	}
2247 
2248 	xen_mc_issue(0);
2249 }
2250 
2251 /*
2252  * Perform the hypercall to exchange a region of our pfns to point to
2253  * memory with the required contiguous alignment.  Takes the pfns as
2254  * input, and populates mfns as output.
2255  *
2256  * Returns a success code indicating whether the hypervisor was able to
2257  * satisfy the request or not.
2258  */
xen_exchange_memory(unsigned long extents_in,unsigned int order_in,unsigned long * pfns_in,unsigned long extents_out,unsigned int order_out,unsigned long * mfns_out,unsigned int address_bits)2259 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2260 			       unsigned long *pfns_in,
2261 			       unsigned long extents_out,
2262 			       unsigned int order_out,
2263 			       unsigned long *mfns_out,
2264 			       unsigned int address_bits)
2265 {
2266 	long rc;
2267 	int success;
2268 
2269 	struct xen_memory_exchange exchange = {
2270 		.in = {
2271 			.nr_extents   = extents_in,
2272 			.extent_order = order_in,
2273 			.extent_start = pfns_in,
2274 			.domid        = DOMID_SELF
2275 		},
2276 		.out = {
2277 			.nr_extents   = extents_out,
2278 			.extent_order = order_out,
2279 			.extent_start = mfns_out,
2280 			.address_bits = address_bits,
2281 			.domid        = DOMID_SELF
2282 		}
2283 	};
2284 
2285 	BUG_ON(extents_in << order_in != extents_out << order_out);
2286 
2287 	rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2288 	success = (exchange.nr_exchanged == extents_in);
2289 
2290 	BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2291 	BUG_ON(success && (rc != 0));
2292 
2293 	return success;
2294 }
2295 
xen_create_contiguous_region(phys_addr_t pstart,unsigned int order,unsigned int address_bits,dma_addr_t * dma_handle)2296 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2297 				 unsigned int address_bits,
2298 				 dma_addr_t *dma_handle)
2299 {
2300 	unsigned long *in_frames = discontig_frames, out_frame;
2301 	unsigned long  flags;
2302 	int            success;
2303 	unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2304 
2305 	/*
2306 	 * Currently an auto-translated guest will not perform I/O, nor will
2307 	 * it require PAE page directories below 4GB. Therefore any calls to
2308 	 * this function are redundant and can be ignored.
2309 	 */
2310 
2311 	if (unlikely(order > MAX_CONTIG_ORDER))
2312 		return -ENOMEM;
2313 
2314 	memset((void *) vstart, 0, PAGE_SIZE << order);
2315 
2316 	spin_lock_irqsave(&xen_reservation_lock, flags);
2317 
2318 	/* 1. Zap current PTEs, remembering MFNs. */
2319 	xen_zap_pfn_range(vstart, order, in_frames, NULL);
2320 
2321 	/* 2. Get a new contiguous memory extent. */
2322 	out_frame = virt_to_pfn(vstart);
2323 	success = xen_exchange_memory(1UL << order, 0, in_frames,
2324 				      1, order, &out_frame,
2325 				      address_bits);
2326 
2327 	/* 3. Map the new extent in place of old pages. */
2328 	if (success)
2329 		xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2330 	else
2331 		xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2332 
2333 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2334 
2335 	*dma_handle = virt_to_machine(vstart).maddr;
2336 	return success ? 0 : -ENOMEM;
2337 }
2338 
xen_destroy_contiguous_region(phys_addr_t pstart,unsigned int order)2339 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2340 {
2341 	unsigned long *out_frames = discontig_frames, in_frame;
2342 	unsigned long  flags;
2343 	int success;
2344 	unsigned long vstart;
2345 
2346 	if (unlikely(order > MAX_CONTIG_ORDER))
2347 		return;
2348 
2349 	vstart = (unsigned long)phys_to_virt(pstart);
2350 	memset((void *) vstart, 0, PAGE_SIZE << order);
2351 
2352 	spin_lock_irqsave(&xen_reservation_lock, flags);
2353 
2354 	/* 1. Find start MFN of contiguous extent. */
2355 	in_frame = virt_to_mfn(vstart);
2356 
2357 	/* 2. Zap current PTEs. */
2358 	xen_zap_pfn_range(vstart, order, NULL, out_frames);
2359 
2360 	/* 3. Do the exchange for non-contiguous MFNs. */
2361 	success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2362 					0, out_frames, 0);
2363 
2364 	/* 4. Map new pages in place of old pages. */
2365 	if (success)
2366 		xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2367 	else
2368 		xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2369 
2370 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2371 }
2372 
xen_flush_tlb_all(void)2373 static noinline void xen_flush_tlb_all(void)
2374 {
2375 	struct mmuext_op *op;
2376 	struct multicall_space mcs;
2377 
2378 	preempt_disable();
2379 
2380 	mcs = xen_mc_entry(sizeof(*op));
2381 
2382 	op = mcs.args;
2383 	op->cmd = MMUEXT_TLB_FLUSH_ALL;
2384 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2385 
2386 	xen_mc_issue(PARAVIRT_LAZY_MMU);
2387 
2388 	preempt_enable();
2389 }
2390 
2391 #define REMAP_BATCH_SIZE 16
2392 
2393 struct remap_data {
2394 	xen_pfn_t *pfn;
2395 	bool contiguous;
2396 	bool no_translate;
2397 	pgprot_t prot;
2398 	struct mmu_update *mmu_update;
2399 };
2400 
remap_area_pfn_pte_fn(pte_t * ptep,unsigned long addr,void * data)2401 static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2402 {
2403 	struct remap_data *rmd = data;
2404 	pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2405 
2406 	/*
2407 	 * If we have a contiguous range, just update the pfn itself,
2408 	 * else update pointer to be "next pfn".
2409 	 */
2410 	if (rmd->contiguous)
2411 		(*rmd->pfn)++;
2412 	else
2413 		rmd->pfn++;
2414 
2415 	rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2416 	rmd->mmu_update->ptr |= rmd->no_translate ?
2417 		MMU_PT_UPDATE_NO_TRANSLATE :
2418 		MMU_NORMAL_PT_UPDATE;
2419 	rmd->mmu_update->val = pte_val_ma(pte);
2420 	rmd->mmu_update++;
2421 
2422 	return 0;
2423 }
2424 
xen_remap_pfn(struct vm_area_struct * vma,unsigned long addr,xen_pfn_t * pfn,int nr,int * err_ptr,pgprot_t prot,unsigned int domid,bool no_translate)2425 int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2426 		  xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2427 		  unsigned int domid, bool no_translate)
2428 {
2429 	int err = 0;
2430 	struct remap_data rmd;
2431 	struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2432 	unsigned long range;
2433 	int mapped = 0;
2434 
2435 	BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2436 
2437 	rmd.pfn = pfn;
2438 	rmd.prot = prot;
2439 	/*
2440 	 * We use the err_ptr to indicate if there we are doing a contiguous
2441 	 * mapping or a discontiguous mapping.
2442 	 */
2443 	rmd.contiguous = !err_ptr;
2444 	rmd.no_translate = no_translate;
2445 
2446 	while (nr) {
2447 		int index = 0;
2448 		int done = 0;
2449 		int batch = min(REMAP_BATCH_SIZE, nr);
2450 		int batch_left = batch;
2451 
2452 		range = (unsigned long)batch << PAGE_SHIFT;
2453 
2454 		rmd.mmu_update = mmu_update;
2455 		err = apply_to_page_range(vma->vm_mm, addr, range,
2456 					  remap_area_pfn_pte_fn, &rmd);
2457 		if (err)
2458 			goto out;
2459 
2460 		/*
2461 		 * We record the error for each page that gives an error, but
2462 		 * continue mapping until the whole set is done
2463 		 */
2464 		do {
2465 			int i;
2466 
2467 			err = HYPERVISOR_mmu_update(&mmu_update[index],
2468 						    batch_left, &done, domid);
2469 
2470 			/*
2471 			 * @err_ptr may be the same buffer as @gfn, so
2472 			 * only clear it after each chunk of @gfn is
2473 			 * used.
2474 			 */
2475 			if (err_ptr) {
2476 				for (i = index; i < index + done; i++)
2477 					err_ptr[i] = 0;
2478 			}
2479 			if (err < 0) {
2480 				if (!err_ptr)
2481 					goto out;
2482 				err_ptr[i] = err;
2483 				done++; /* Skip failed frame. */
2484 			} else
2485 				mapped += done;
2486 			batch_left -= done;
2487 			index += done;
2488 		} while (batch_left);
2489 
2490 		nr -= batch;
2491 		addr += range;
2492 		if (err_ptr)
2493 			err_ptr += batch;
2494 		cond_resched();
2495 	}
2496 out:
2497 
2498 	xen_flush_tlb_all();
2499 
2500 	return err < 0 ? err : mapped;
2501 }
2502 EXPORT_SYMBOL_GPL(xen_remap_pfn);
2503 
2504 #ifdef CONFIG_KEXEC_CORE
paddr_vmcoreinfo_note(void)2505 phys_addr_t paddr_vmcoreinfo_note(void)
2506 {
2507 	if (xen_pv_domain())
2508 		return virt_to_machine(vmcoreinfo_note).maddr;
2509 	else
2510 		return __pa(vmcoreinfo_note);
2511 }
2512 #endif /* CONFIG_KEXEC_CORE */
2513