xref: /openbmc/linux/arch/x86/xen/mmu_pv.c (revision fd5e9fccbd504c5179ab57ff695c610bca8809d6)
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 /*
90  * Prototypes for functions called via PV_CALLEE_SAVE_REGS_THUNK() in order
91  * to avoid warnings with "-Wmissing-prototypes".
92  */
93 pteval_t xen_pte_val(pte_t pte);
94 pgdval_t xen_pgd_val(pgd_t pgd);
95 pmdval_t xen_pmd_val(pmd_t pmd);
96 pudval_t xen_pud_val(pud_t pud);
97 p4dval_t xen_p4d_val(p4d_t p4d);
98 pte_t xen_make_pte(pteval_t pte);
99 pgd_t xen_make_pgd(pgdval_t pgd);
100 pmd_t xen_make_pmd(pmdval_t pmd);
101 pud_t xen_make_pud(pudval_t pud);
102 p4d_t xen_make_p4d(p4dval_t p4d);
103 pte_t xen_make_pte_init(pteval_t pte);
104 
105 #ifdef CONFIG_X86_VSYSCALL_EMULATION
106 /* l3 pud for userspace vsyscall mapping */
107 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
108 #endif
109 
110 /*
111  * Protects atomic reservation decrease/increase against concurrent increases.
112  * Also protects non-atomic updates of current_pages and balloon lists.
113  */
114 static DEFINE_SPINLOCK(xen_reservation_lock);
115 
116 /* Protected by xen_reservation_lock. */
117 #define MIN_CONTIG_ORDER 9 /* 2MB */
118 static unsigned int discontig_frames_order = MIN_CONTIG_ORDER;
119 static unsigned long discontig_frames_early[1UL << MIN_CONTIG_ORDER] __initdata;
120 static unsigned long *discontig_frames __refdata = discontig_frames_early;
121 static bool discontig_frames_dyn;
122 
alloc_discontig_frames(unsigned int order)123 static int alloc_discontig_frames(unsigned int order)
124 {
125 	unsigned long *new_array, *old_array;
126 	unsigned int old_order;
127 	unsigned long flags;
128 
129 	BUG_ON(order < MIN_CONTIG_ORDER);
130 	BUILD_BUG_ON(sizeof(discontig_frames_early) != PAGE_SIZE);
131 
132 	new_array = (unsigned long *)__get_free_pages(GFP_KERNEL,
133 						      order - MIN_CONTIG_ORDER);
134 	if (!new_array)
135 		return -ENOMEM;
136 
137 	spin_lock_irqsave(&xen_reservation_lock, flags);
138 
139 	old_order = discontig_frames_order;
140 
141 	if (order > discontig_frames_order || !discontig_frames_dyn) {
142 		if (!discontig_frames_dyn)
143 			old_array = NULL;
144 		else
145 			old_array = discontig_frames;
146 
147 		discontig_frames = new_array;
148 		discontig_frames_order = order;
149 		discontig_frames_dyn = true;
150 	} else {
151 		old_array = new_array;
152 	}
153 
154 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
155 
156 	free_pages((unsigned long)old_array, old_order - MIN_CONTIG_ORDER);
157 
158 	return 0;
159 }
160 
161 /*
162  * Note about cr3 (pagetable base) values:
163  *
164  * xen_cr3 contains the current logical cr3 value; it contains the
165  * last set cr3.  This may not be the current effective cr3, because
166  * its update may be being lazily deferred.  However, a vcpu looking
167  * at its own cr3 can use this value knowing that it everything will
168  * be self-consistent.
169  *
170  * xen_current_cr3 contains the actual vcpu cr3; it is set once the
171  * hypercall to set the vcpu cr3 is complete (so it may be a little
172  * out of date, but it will never be set early).  If one vcpu is
173  * looking at another vcpu's cr3 value, it should use this variable.
174  */
175 DEFINE_PER_CPU(unsigned long, xen_cr3);	 /* cr3 stored as physaddr */
176 DEFINE_PER_CPU(unsigned long, xen_current_cr3);	 /* actual vcpu cr3 */
177 
178 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
179 
180 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
181 
182 /*
183  * Just beyond the highest usermode address.  STACK_TOP_MAX has a
184  * redzone above it, so round it up to a PGD boundary.
185  */
186 #define USER_LIMIT	((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
187 
make_lowmem_page_readonly(void * vaddr)188 void make_lowmem_page_readonly(void *vaddr)
189 {
190 	pte_t *pte, ptev;
191 	unsigned long address = (unsigned long)vaddr;
192 	unsigned int level;
193 
194 	pte = lookup_address(address, &level);
195 	if (pte == NULL)
196 		return;		/* vaddr missing */
197 
198 	ptev = pte_wrprotect(*pte);
199 
200 	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
201 		BUG();
202 }
203 
make_lowmem_page_readwrite(void * vaddr)204 void make_lowmem_page_readwrite(void *vaddr)
205 {
206 	pte_t *pte, ptev;
207 	unsigned long address = (unsigned long)vaddr;
208 	unsigned int level;
209 
210 	pte = lookup_address(address, &level);
211 	if (pte == NULL)
212 		return;		/* vaddr missing */
213 
214 	ptev = pte_mkwrite_novma(*pte);
215 
216 	if (HYPERVISOR_update_va_mapping(address, ptev, 0))
217 		BUG();
218 }
219 
220 
221 /*
222  * During early boot all page table pages are pinned, but we do not have struct
223  * pages, so return true until struct pages are ready.
224  */
xen_page_pinned(void * ptr)225 static bool xen_page_pinned(void *ptr)
226 {
227 	if (static_branch_likely(&xen_struct_pages_ready)) {
228 		struct page *page = virt_to_page(ptr);
229 
230 		return PagePinned(page);
231 	}
232 	return true;
233 }
234 
xen_extend_mmu_update(const struct mmu_update * update)235 static void xen_extend_mmu_update(const struct mmu_update *update)
236 {
237 	struct multicall_space mcs;
238 	struct mmu_update *u;
239 
240 	mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
241 
242 	if (mcs.mc != NULL) {
243 		mcs.mc->args[1]++;
244 	} else {
245 		mcs = __xen_mc_entry(sizeof(*u));
246 		MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
247 	}
248 
249 	u = mcs.args;
250 	*u = *update;
251 }
252 
xen_extend_mmuext_op(const struct mmuext_op * op)253 static void xen_extend_mmuext_op(const struct mmuext_op *op)
254 {
255 	struct multicall_space mcs;
256 	struct mmuext_op *u;
257 
258 	mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
259 
260 	if (mcs.mc != NULL) {
261 		mcs.mc->args[1]++;
262 	} else {
263 		mcs = __xen_mc_entry(sizeof(*u));
264 		MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
265 	}
266 
267 	u = mcs.args;
268 	*u = *op;
269 }
270 
xen_set_pmd_hyper(pmd_t * ptr,pmd_t val)271 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
272 {
273 	struct mmu_update u;
274 
275 	preempt_disable();
276 
277 	xen_mc_batch();
278 
279 	/* ptr may be ioremapped for 64-bit pagetable setup */
280 	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
281 	u.val = pmd_val_ma(val);
282 	xen_extend_mmu_update(&u);
283 
284 	xen_mc_issue(XEN_LAZY_MMU);
285 
286 	preempt_enable();
287 }
288 
xen_set_pmd(pmd_t * ptr,pmd_t val)289 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
290 {
291 	trace_xen_mmu_set_pmd(ptr, val);
292 
293 	/* If page is not pinned, we can just update the entry
294 	   directly */
295 	if (!xen_page_pinned(ptr)) {
296 		*ptr = val;
297 		return;
298 	}
299 
300 	xen_set_pmd_hyper(ptr, val);
301 }
302 
303 /*
304  * Associate a virtual page frame with a given physical page frame
305  * and protection flags for that frame.
306  */
set_pte_mfn(unsigned long vaddr,unsigned long mfn,pgprot_t flags)307 void __init set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
308 {
309 	if (HYPERVISOR_update_va_mapping(vaddr, mfn_pte(mfn, flags),
310 					 UVMF_INVLPG))
311 		BUG();
312 }
313 
xen_batched_set_pte(pte_t * ptep,pte_t pteval)314 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
315 {
316 	struct mmu_update u;
317 
318 	if (xen_get_lazy_mode() != XEN_LAZY_MMU)
319 		return false;
320 
321 	xen_mc_batch();
322 
323 	u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
324 	u.val = pte_val_ma(pteval);
325 	xen_extend_mmu_update(&u);
326 
327 	xen_mc_issue(XEN_LAZY_MMU);
328 
329 	return true;
330 }
331 
__xen_set_pte(pte_t * ptep,pte_t pteval)332 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
333 {
334 	if (!xen_batched_set_pte(ptep, pteval)) {
335 		/*
336 		 * Could call native_set_pte() here and trap and
337 		 * emulate the PTE write, but a hypercall is much cheaper.
338 		 */
339 		struct mmu_update u;
340 
341 		u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
342 		u.val = pte_val_ma(pteval);
343 		HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
344 	}
345 }
346 
xen_set_pte(pte_t * ptep,pte_t pteval)347 static void xen_set_pte(pte_t *ptep, pte_t pteval)
348 {
349 	trace_xen_mmu_set_pte(ptep, pteval);
350 	__xen_set_pte(ptep, pteval);
351 }
352 
xen_ptep_modify_prot_start(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep)353 pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma,
354 				 unsigned long addr, pte_t *ptep)
355 {
356 	/* Just return the pte as-is.  We preserve the bits on commit */
357 	trace_xen_mmu_ptep_modify_prot_start(vma->vm_mm, addr, ptep, *ptep);
358 	return *ptep;
359 }
360 
xen_ptep_modify_prot_commit(struct vm_area_struct * vma,unsigned long addr,pte_t * ptep,pte_t pte)361 void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
362 				 pte_t *ptep, pte_t pte)
363 {
364 	struct mmu_update u;
365 
366 	trace_xen_mmu_ptep_modify_prot_commit(vma->vm_mm, addr, ptep, pte);
367 	xen_mc_batch();
368 
369 	u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
370 	u.val = pte_val_ma(pte);
371 	xen_extend_mmu_update(&u);
372 
373 	xen_mc_issue(XEN_LAZY_MMU);
374 }
375 
376 /* Assume pteval_t is equivalent to all the other *val_t types. */
pte_mfn_to_pfn(pteval_t val)377 static pteval_t pte_mfn_to_pfn(pteval_t val)
378 {
379 	if (val & _PAGE_PRESENT) {
380 		unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
381 		unsigned long pfn = mfn_to_pfn(mfn);
382 
383 		pteval_t flags = val & PTE_FLAGS_MASK;
384 		if (unlikely(pfn == ~0))
385 			val = flags & ~_PAGE_PRESENT;
386 		else
387 			val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
388 	}
389 
390 	return val;
391 }
392 
pte_pfn_to_mfn(pteval_t val)393 static pteval_t pte_pfn_to_mfn(pteval_t val)
394 {
395 	if (val & _PAGE_PRESENT) {
396 		unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
397 		pteval_t flags = val & PTE_FLAGS_MASK;
398 		unsigned long mfn;
399 
400 		mfn = __pfn_to_mfn(pfn);
401 
402 		/*
403 		 * If there's no mfn for the pfn, then just create an
404 		 * empty non-present pte.  Unfortunately this loses
405 		 * information about the original pfn, so
406 		 * pte_mfn_to_pfn is asymmetric.
407 		 */
408 		if (unlikely(mfn == INVALID_P2M_ENTRY)) {
409 			mfn = 0;
410 			flags = 0;
411 		} else
412 			mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
413 		val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
414 	}
415 
416 	return val;
417 }
418 
xen_pte_val(pte_t pte)419 __visible pteval_t xen_pte_val(pte_t pte)
420 {
421 	pteval_t pteval = pte.pte;
422 
423 	return pte_mfn_to_pfn(pteval);
424 }
425 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
426 
xen_pgd_val(pgd_t pgd)427 __visible pgdval_t xen_pgd_val(pgd_t pgd)
428 {
429 	return pte_mfn_to_pfn(pgd.pgd);
430 }
431 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
432 
xen_make_pte(pteval_t pte)433 __visible pte_t xen_make_pte(pteval_t pte)
434 {
435 	pte = pte_pfn_to_mfn(pte);
436 
437 	return native_make_pte(pte);
438 }
439 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
440 
xen_make_pgd(pgdval_t pgd)441 __visible pgd_t xen_make_pgd(pgdval_t pgd)
442 {
443 	pgd = pte_pfn_to_mfn(pgd);
444 	return native_make_pgd(pgd);
445 }
446 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
447 
xen_pmd_val(pmd_t pmd)448 __visible pmdval_t xen_pmd_val(pmd_t pmd)
449 {
450 	return pte_mfn_to_pfn(pmd.pmd);
451 }
452 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
453 
xen_set_pud_hyper(pud_t * ptr,pud_t val)454 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
455 {
456 	struct mmu_update u;
457 
458 	preempt_disable();
459 
460 	xen_mc_batch();
461 
462 	/* ptr may be ioremapped for 64-bit pagetable setup */
463 	u.ptr = arbitrary_virt_to_machine(ptr).maddr;
464 	u.val = pud_val_ma(val);
465 	xen_extend_mmu_update(&u);
466 
467 	xen_mc_issue(XEN_LAZY_MMU);
468 
469 	preempt_enable();
470 }
471 
xen_set_pud(pud_t * ptr,pud_t val)472 static void xen_set_pud(pud_t *ptr, pud_t val)
473 {
474 	trace_xen_mmu_set_pud(ptr, val);
475 
476 	/* If page is not pinned, we can just update the entry
477 	   directly */
478 	if (!xen_page_pinned(ptr)) {
479 		*ptr = val;
480 		return;
481 	}
482 
483 	xen_set_pud_hyper(ptr, val);
484 }
485 
xen_make_pmd(pmdval_t pmd)486 __visible pmd_t xen_make_pmd(pmdval_t pmd)
487 {
488 	pmd = pte_pfn_to_mfn(pmd);
489 	return native_make_pmd(pmd);
490 }
491 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
492 
xen_pud_val(pud_t pud)493 __visible pudval_t xen_pud_val(pud_t pud)
494 {
495 	return pte_mfn_to_pfn(pud.pud);
496 }
497 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
498 
xen_make_pud(pudval_t pud)499 __visible pud_t xen_make_pud(pudval_t pud)
500 {
501 	pud = pte_pfn_to_mfn(pud);
502 
503 	return native_make_pud(pud);
504 }
505 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
506 
xen_get_user_pgd(pgd_t * pgd)507 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
508 {
509 	pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
510 	unsigned offset = pgd - pgd_page;
511 	pgd_t *user_ptr = NULL;
512 
513 	if (offset < pgd_index(USER_LIMIT)) {
514 		struct page *page = virt_to_page(pgd_page);
515 		user_ptr = (pgd_t *)page->private;
516 		if (user_ptr)
517 			user_ptr += offset;
518 	}
519 
520 	return user_ptr;
521 }
522 
__xen_set_p4d_hyper(p4d_t * ptr,p4d_t val)523 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
524 {
525 	struct mmu_update u;
526 
527 	u.ptr = virt_to_machine(ptr).maddr;
528 	u.val = p4d_val_ma(val);
529 	xen_extend_mmu_update(&u);
530 }
531 
532 /*
533  * Raw hypercall-based set_p4d, intended for in early boot before
534  * there's a page structure.  This implies:
535  *  1. The only existing pagetable is the kernel's
536  *  2. It is always pinned
537  *  3. It has no user pagetable attached to it
538  */
xen_set_p4d_hyper(p4d_t * ptr,p4d_t val)539 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
540 {
541 	preempt_disable();
542 
543 	xen_mc_batch();
544 
545 	__xen_set_p4d_hyper(ptr, val);
546 
547 	xen_mc_issue(XEN_LAZY_MMU);
548 
549 	preempt_enable();
550 }
551 
xen_set_p4d(p4d_t * ptr,p4d_t val)552 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
553 {
554 	pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
555 	pgd_t pgd_val;
556 
557 	trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
558 
559 	/* If page is not pinned, we can just update the entry
560 	   directly */
561 	if (!xen_page_pinned(ptr)) {
562 		*ptr = val;
563 		if (user_ptr) {
564 			WARN_ON(xen_page_pinned(user_ptr));
565 			pgd_val.pgd = p4d_val_ma(val);
566 			*user_ptr = pgd_val;
567 		}
568 		return;
569 	}
570 
571 	/* If it's pinned, then we can at least batch the kernel and
572 	   user updates together. */
573 	xen_mc_batch();
574 
575 	__xen_set_p4d_hyper(ptr, val);
576 	if (user_ptr)
577 		__xen_set_p4d_hyper((p4d_t *)user_ptr, val);
578 
579 	xen_mc_issue(XEN_LAZY_MMU);
580 }
581 
582 #if CONFIG_PGTABLE_LEVELS >= 5
xen_p4d_val(p4d_t p4d)583 __visible p4dval_t xen_p4d_val(p4d_t p4d)
584 {
585 	return pte_mfn_to_pfn(p4d.p4d);
586 }
587 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
588 
xen_make_p4d(p4dval_t p4d)589 __visible p4d_t xen_make_p4d(p4dval_t p4d)
590 {
591 	p4d = pte_pfn_to_mfn(p4d);
592 
593 	return native_make_p4d(p4d);
594 }
595 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
596 #endif  /* CONFIG_PGTABLE_LEVELS >= 5 */
597 
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)598 static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
599 			 void (*func)(struct mm_struct *mm, struct page *,
600 				      enum pt_level),
601 			 bool last, unsigned long limit)
602 {
603 	int i, nr;
604 
605 	nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
606 	for (i = 0; i < nr; i++) {
607 		if (!pmd_none(pmd[i]))
608 			(*func)(mm, pmd_page(pmd[i]), PT_PTE);
609 	}
610 }
611 
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)612 static void xen_pud_walk(struct mm_struct *mm, pud_t *pud,
613 			 void (*func)(struct mm_struct *mm, struct page *,
614 				      enum pt_level),
615 			 bool last, unsigned long limit)
616 {
617 	int i, nr;
618 
619 	nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
620 	for (i = 0; i < nr; i++) {
621 		pmd_t *pmd;
622 
623 		if (pud_none(pud[i]))
624 			continue;
625 
626 		pmd = pmd_offset(&pud[i], 0);
627 		if (PTRS_PER_PMD > 1)
628 			(*func)(mm, virt_to_page(pmd), PT_PMD);
629 		xen_pmd_walk(mm, pmd, func, last && i == nr - 1, limit);
630 	}
631 }
632 
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)633 static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
634 			 void (*func)(struct mm_struct *mm, struct page *,
635 				      enum pt_level),
636 			 bool last, unsigned long limit)
637 {
638 	pud_t *pud;
639 
640 
641 	if (p4d_none(*p4d))
642 		return;
643 
644 	pud = pud_offset(p4d, 0);
645 	if (PTRS_PER_PUD > 1)
646 		(*func)(mm, virt_to_page(pud), PT_PUD);
647 	xen_pud_walk(mm, pud, func, last, limit);
648 }
649 
650 /*
651  * (Yet another) pagetable walker.  This one is intended for pinning a
652  * pagetable.  This means that it walks a pagetable and calls the
653  * callback function on each page it finds making up the page table,
654  * at every level.  It walks the entire pagetable, but it only bothers
655  * pinning pte pages which are below limit.  In the normal case this
656  * will be STACK_TOP_MAX, but at boot we need to pin up to
657  * FIXADDR_TOP.
658  *
659  * We must skip the Xen hole in the middle of the address space, just after
660  * the big x86-64 virtual hole.
661  */
__xen_pgd_walk(struct mm_struct * mm,pgd_t * pgd,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),unsigned long limit)662 static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
663 			   void (*func)(struct mm_struct *mm, struct page *,
664 					enum pt_level),
665 			   unsigned long limit)
666 {
667 	int i, nr;
668 	unsigned hole_low = 0, hole_high = 0;
669 
670 	/* The limit is the last byte to be touched */
671 	limit--;
672 	BUG_ON(limit >= FIXADDR_TOP);
673 
674 	/*
675 	 * 64-bit has a great big hole in the middle of the address
676 	 * space, which contains the Xen mappings.
677 	 */
678 	hole_low = pgd_index(GUARD_HOLE_BASE_ADDR);
679 	hole_high = pgd_index(GUARD_HOLE_END_ADDR);
680 
681 	nr = pgd_index(limit) + 1;
682 	for (i = 0; i < nr; i++) {
683 		p4d_t *p4d;
684 
685 		if (i >= hole_low && i < hole_high)
686 			continue;
687 
688 		if (pgd_none(pgd[i]))
689 			continue;
690 
691 		p4d = p4d_offset(&pgd[i], 0);
692 		xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
693 	}
694 
695 	/* Do the top level last, so that the callbacks can use it as
696 	   a cue to do final things like tlb flushes. */
697 	(*func)(mm, virt_to_page(pgd), PT_PGD);
698 }
699 
xen_pgd_walk(struct mm_struct * mm,void (* func)(struct mm_struct * mm,struct page *,enum pt_level),unsigned long limit)700 static void xen_pgd_walk(struct mm_struct *mm,
701 			 void (*func)(struct mm_struct *mm, struct page *,
702 				      enum pt_level),
703 			 unsigned long limit)
704 {
705 	__xen_pgd_walk(mm, mm->pgd, func, limit);
706 }
707 
708 /* If we're using split pte locks, then take the page's lock and
709    return a pointer to it.  Otherwise return NULL. */
xen_pte_lock(struct page * page,struct mm_struct * mm)710 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
711 {
712 	spinlock_t *ptl = NULL;
713 
714 #if USE_SPLIT_PTE_PTLOCKS
715 	ptl = ptlock_ptr(page_ptdesc(page));
716 	spin_lock_nest_lock(ptl, &mm->page_table_lock);
717 #endif
718 
719 	return ptl;
720 }
721 
xen_pte_unlock(void * v)722 static void xen_pte_unlock(void *v)
723 {
724 	spinlock_t *ptl = v;
725 	spin_unlock(ptl);
726 }
727 
xen_do_pin(unsigned level,unsigned long pfn)728 static void xen_do_pin(unsigned level, unsigned long pfn)
729 {
730 	struct mmuext_op op;
731 
732 	op.cmd = level;
733 	op.arg1.mfn = pfn_to_mfn(pfn);
734 
735 	xen_extend_mmuext_op(&op);
736 }
737 
xen_pin_page(struct mm_struct * mm,struct page * page,enum pt_level level)738 static void xen_pin_page(struct mm_struct *mm, struct page *page,
739 			 enum pt_level level)
740 {
741 	unsigned pgfl = TestSetPagePinned(page);
742 
743 	if (!pgfl) {
744 		void *pt = lowmem_page_address(page);
745 		unsigned long pfn = page_to_pfn(page);
746 		struct multicall_space mcs = __xen_mc_entry(0);
747 		spinlock_t *ptl;
748 
749 		/*
750 		 * We need to hold the pagetable lock between the time
751 		 * we make the pagetable RO and when we actually pin
752 		 * it.  If we don't, then other users may come in and
753 		 * attempt to update the pagetable by writing it,
754 		 * which will fail because the memory is RO but not
755 		 * pinned, so Xen won't do the trap'n'emulate.
756 		 *
757 		 * If we're using split pte locks, we can't hold the
758 		 * entire pagetable's worth of locks during the
759 		 * traverse, because we may wrap the preempt count (8
760 		 * bits).  The solution is to mark RO and pin each PTE
761 		 * page while holding the lock.  This means the number
762 		 * of locks we end up holding is never more than a
763 		 * batch size (~32 entries, at present).
764 		 *
765 		 * If we're not using split pte locks, we needn't pin
766 		 * the PTE pages independently, because we're
767 		 * protected by the overall pagetable lock.
768 		 */
769 		ptl = NULL;
770 		if (level == PT_PTE)
771 			ptl = xen_pte_lock(page, mm);
772 
773 		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
774 					pfn_pte(pfn, PAGE_KERNEL_RO),
775 					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
776 
777 		if (ptl) {
778 			xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
779 
780 			/* Queue a deferred unlock for when this batch
781 			   is completed. */
782 			xen_mc_callback(xen_pte_unlock, ptl);
783 		}
784 	}
785 }
786 
787 /* This is called just after a mm has been created, but it has not
788    been used yet.  We need to make sure that its pagetable is all
789    read-only, and can be pinned. */
__xen_pgd_pin(struct mm_struct * mm,pgd_t * pgd)790 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
791 {
792 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
793 
794 	trace_xen_mmu_pgd_pin(mm, pgd);
795 
796 	xen_mc_batch();
797 
798 	__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT);
799 
800 	xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
801 
802 	if (user_pgd) {
803 		xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
804 		xen_do_pin(MMUEXT_PIN_L4_TABLE,
805 			   PFN_DOWN(__pa(user_pgd)));
806 	}
807 
808 	xen_mc_issue(0);
809 }
810 
xen_pgd_pin(struct mm_struct * mm)811 static void xen_pgd_pin(struct mm_struct *mm)
812 {
813 	__xen_pgd_pin(mm, mm->pgd);
814 }
815 
816 /*
817  * On save, we need to pin all pagetables to make sure they get their
818  * mfns turned into pfns.  Search the list for any unpinned pgds and pin
819  * them (unpinned pgds are not currently in use, probably because the
820  * process is under construction or destruction).
821  *
822  * Expected to be called in stop_machine() ("equivalent to taking
823  * every spinlock in the system"), so the locking doesn't really
824  * matter all that much.
825  */
xen_mm_pin_all(void)826 void xen_mm_pin_all(void)
827 {
828 	struct page *page;
829 
830 	spin_lock(&init_mm.page_table_lock);
831 	spin_lock(&pgd_lock);
832 
833 	list_for_each_entry(page, &pgd_list, lru) {
834 		if (!PagePinned(page)) {
835 			__xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
836 			SetPageSavePinned(page);
837 		}
838 	}
839 
840 	spin_unlock(&pgd_lock);
841 	spin_unlock(&init_mm.page_table_lock);
842 }
843 
xen_mark_pinned(struct mm_struct * mm,struct page * page,enum pt_level level)844 static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
845 				   enum pt_level level)
846 {
847 	SetPagePinned(page);
848 }
849 
850 /*
851  * The init_mm pagetable is really pinned as soon as its created, but
852  * that's before we have page structures to store the bits.  So do all
853  * the book-keeping now once struct pages for allocated pages are
854  * initialized. This happens only after memblock_free_all() is called.
855  */
xen_after_bootmem(void)856 static void __init xen_after_bootmem(void)
857 {
858 	static_branch_enable(&xen_struct_pages_ready);
859 #ifdef CONFIG_X86_VSYSCALL_EMULATION
860 	SetPagePinned(virt_to_page(level3_user_vsyscall));
861 #endif
862 	xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
863 
864 	if (alloc_discontig_frames(MIN_CONTIG_ORDER))
865 		BUG();
866 }
867 
xen_unpin_page(struct mm_struct * mm,struct page * page,enum pt_level level)868 static void xen_unpin_page(struct mm_struct *mm, struct page *page,
869 			   enum pt_level level)
870 {
871 	unsigned pgfl = TestClearPagePinned(page);
872 
873 	if (pgfl) {
874 		void *pt = lowmem_page_address(page);
875 		unsigned long pfn = page_to_pfn(page);
876 		spinlock_t *ptl = NULL;
877 		struct multicall_space mcs;
878 
879 		/*
880 		 * Do the converse to pin_page.  If we're using split
881 		 * pte locks, we must be holding the lock for while
882 		 * the pte page is unpinned but still RO to prevent
883 		 * concurrent updates from seeing it in this
884 		 * partially-pinned state.
885 		 */
886 		if (level == PT_PTE) {
887 			ptl = xen_pte_lock(page, mm);
888 
889 			if (ptl)
890 				xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
891 		}
892 
893 		mcs = __xen_mc_entry(0);
894 
895 		MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
896 					pfn_pte(pfn, PAGE_KERNEL),
897 					level == PT_PGD ? UVMF_TLB_FLUSH : 0);
898 
899 		if (ptl) {
900 			/* unlock when batch completed */
901 			xen_mc_callback(xen_pte_unlock, ptl);
902 		}
903 	}
904 }
905 
906 /* Release a pagetables pages back as normal RW */
__xen_pgd_unpin(struct mm_struct * mm,pgd_t * pgd)907 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
908 {
909 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
910 
911 	trace_xen_mmu_pgd_unpin(mm, pgd);
912 
913 	xen_mc_batch();
914 
915 	xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
916 
917 	if (user_pgd) {
918 		xen_do_pin(MMUEXT_UNPIN_TABLE,
919 			   PFN_DOWN(__pa(user_pgd)));
920 		xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
921 	}
922 
923 	__xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
924 
925 	xen_mc_issue(0);
926 }
927 
xen_pgd_unpin(struct mm_struct * mm)928 static void xen_pgd_unpin(struct mm_struct *mm)
929 {
930 	__xen_pgd_unpin(mm, mm->pgd);
931 }
932 
933 /*
934  * On resume, undo any pinning done at save, so that the rest of the
935  * kernel doesn't see any unexpected pinned pagetables.
936  */
xen_mm_unpin_all(void)937 void xen_mm_unpin_all(void)
938 {
939 	struct page *page;
940 
941 	spin_lock(&init_mm.page_table_lock);
942 	spin_lock(&pgd_lock);
943 
944 	list_for_each_entry(page, &pgd_list, lru) {
945 		if (PageSavePinned(page)) {
946 			BUG_ON(!PagePinned(page));
947 			__xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
948 			ClearPageSavePinned(page);
949 		}
950 	}
951 
952 	spin_unlock(&pgd_lock);
953 	spin_unlock(&init_mm.page_table_lock);
954 }
955 
xen_enter_mmap(struct mm_struct * mm)956 static void xen_enter_mmap(struct mm_struct *mm)
957 {
958 	spin_lock(&mm->page_table_lock);
959 	xen_pgd_pin(mm);
960 	spin_unlock(&mm->page_table_lock);
961 }
962 
drop_mm_ref_this_cpu(void * info)963 static void drop_mm_ref_this_cpu(void *info)
964 {
965 	struct mm_struct *mm = info;
966 
967 	if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
968 		leave_mm(smp_processor_id());
969 
970 	/*
971 	 * If this cpu still has a stale cr3 reference, then make sure
972 	 * it has been flushed.
973 	 */
974 	if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
975 		xen_mc_flush();
976 }
977 
978 #ifdef CONFIG_SMP
979 /*
980  * Another cpu may still have their %cr3 pointing at the pagetable, so
981  * we need to repoint it somewhere else before we can unpin it.
982  */
xen_drop_mm_ref(struct mm_struct * mm)983 static void xen_drop_mm_ref(struct mm_struct *mm)
984 {
985 	cpumask_var_t mask;
986 	unsigned cpu;
987 
988 	drop_mm_ref_this_cpu(mm);
989 
990 	/* Get the "official" set of cpus referring to our pagetable. */
991 	if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
992 		for_each_online_cpu(cpu) {
993 			if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
994 				continue;
995 			smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
996 		}
997 		return;
998 	}
999 
1000 	/*
1001 	 * It's possible that a vcpu may have a stale reference to our
1002 	 * cr3, because its in lazy mode, and it hasn't yet flushed
1003 	 * its set of pending hypercalls yet.  In this case, we can
1004 	 * look at its actual current cr3 value, and force it to flush
1005 	 * if needed.
1006 	 */
1007 	cpumask_clear(mask);
1008 	for_each_online_cpu(cpu) {
1009 		if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1010 			cpumask_set_cpu(cpu, mask);
1011 	}
1012 
1013 	smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
1014 	free_cpumask_var(mask);
1015 }
1016 #else
xen_drop_mm_ref(struct mm_struct * mm)1017 static void xen_drop_mm_ref(struct mm_struct *mm)
1018 {
1019 	drop_mm_ref_this_cpu(mm);
1020 }
1021 #endif
1022 
1023 /*
1024  * While a process runs, Xen pins its pagetables, which means that the
1025  * hypervisor forces it to be read-only, and it controls all updates
1026  * to it.  This means that all pagetable updates have to go via the
1027  * hypervisor, which is moderately expensive.
1028  *
1029  * Since we're pulling the pagetable down, we switch to use init_mm,
1030  * unpin old process pagetable and mark it all read-write, which
1031  * allows further operations on it to be simple memory accesses.
1032  *
1033  * The only subtle point is that another CPU may be still using the
1034  * pagetable because of lazy tlb flushing.  This means we need need to
1035  * switch all CPUs off this pagetable before we can unpin it.
1036  */
xen_exit_mmap(struct mm_struct * mm)1037 static void xen_exit_mmap(struct mm_struct *mm)
1038 {
1039 	get_cpu();		/* make sure we don't move around */
1040 	xen_drop_mm_ref(mm);
1041 	put_cpu();
1042 
1043 	spin_lock(&mm->page_table_lock);
1044 
1045 	/* pgd may not be pinned in the error exit path of execve */
1046 	if (xen_page_pinned(mm->pgd))
1047 		xen_pgd_unpin(mm);
1048 
1049 	spin_unlock(&mm->page_table_lock);
1050 }
1051 
1052 static void xen_post_allocator_init(void);
1053 
pin_pagetable_pfn(unsigned cmd,unsigned long pfn)1054 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1055 {
1056 	struct mmuext_op op;
1057 
1058 	op.cmd = cmd;
1059 	op.arg1.mfn = pfn_to_mfn(pfn);
1060 	if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1061 		BUG();
1062 }
1063 
xen_cleanhighmap(unsigned long vaddr,unsigned long vaddr_end)1064 static void __init xen_cleanhighmap(unsigned long vaddr,
1065 				    unsigned long vaddr_end)
1066 {
1067 	unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1068 	pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1069 
1070 	/* NOTE: The loop is more greedy than the cleanup_highmap variant.
1071 	 * We include the PMD passed in on _both_ boundaries. */
1072 	for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1073 			pmd++, vaddr += PMD_SIZE) {
1074 		if (pmd_none(*pmd))
1075 			continue;
1076 		if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1077 			set_pmd(pmd, __pmd(0));
1078 	}
1079 	/* In case we did something silly, we should crash in this function
1080 	 * instead of somewhere later and be confusing. */
1081 	xen_mc_flush();
1082 }
1083 
1084 /*
1085  * Make a page range writeable and free it.
1086  */
xen_free_ro_pages(unsigned long paddr,unsigned long size)1087 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1088 {
1089 	void *vaddr = __va(paddr);
1090 	void *vaddr_end = vaddr + size;
1091 
1092 	for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1093 		make_lowmem_page_readwrite(vaddr);
1094 
1095 	memblock_phys_free(paddr, size);
1096 }
1097 
xen_cleanmfnmap_free_pgtbl(void * pgtbl,bool unpin)1098 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1099 {
1100 	unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1101 
1102 	if (unpin)
1103 		pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1104 	ClearPagePinned(virt_to_page(__va(pa)));
1105 	xen_free_ro_pages(pa, PAGE_SIZE);
1106 }
1107 
xen_cleanmfnmap_pmd(pmd_t * pmd,bool unpin)1108 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1109 {
1110 	unsigned long pa;
1111 	pte_t *pte_tbl;
1112 	int i;
1113 
1114 	if (pmd_large(*pmd)) {
1115 		pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1116 		xen_free_ro_pages(pa, PMD_SIZE);
1117 		return;
1118 	}
1119 
1120 	pte_tbl = pte_offset_kernel(pmd, 0);
1121 	for (i = 0; i < PTRS_PER_PTE; i++) {
1122 		if (pte_none(pte_tbl[i]))
1123 			continue;
1124 		pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1125 		xen_free_ro_pages(pa, PAGE_SIZE);
1126 	}
1127 	set_pmd(pmd, __pmd(0));
1128 	xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1129 }
1130 
xen_cleanmfnmap_pud(pud_t * pud,bool unpin)1131 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1132 {
1133 	unsigned long pa;
1134 	pmd_t *pmd_tbl;
1135 	int i;
1136 
1137 	if (pud_leaf(*pud)) {
1138 		pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1139 		xen_free_ro_pages(pa, PUD_SIZE);
1140 		return;
1141 	}
1142 
1143 	pmd_tbl = pmd_offset(pud, 0);
1144 	for (i = 0; i < PTRS_PER_PMD; i++) {
1145 		if (pmd_none(pmd_tbl[i]))
1146 			continue;
1147 		xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1148 	}
1149 	set_pud(pud, __pud(0));
1150 	xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1151 }
1152 
xen_cleanmfnmap_p4d(p4d_t * p4d,bool unpin)1153 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1154 {
1155 	unsigned long pa;
1156 	pud_t *pud_tbl;
1157 	int i;
1158 
1159 	if (p4d_large(*p4d)) {
1160 		pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1161 		xen_free_ro_pages(pa, P4D_SIZE);
1162 		return;
1163 	}
1164 
1165 	pud_tbl = pud_offset(p4d, 0);
1166 	for (i = 0; i < PTRS_PER_PUD; i++) {
1167 		if (pud_none(pud_tbl[i]))
1168 			continue;
1169 		xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1170 	}
1171 	set_p4d(p4d, __p4d(0));
1172 	xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1173 }
1174 
1175 /*
1176  * Since it is well isolated we can (and since it is perhaps large we should)
1177  * also free the page tables mapping the initial P->M table.
1178  */
xen_cleanmfnmap(unsigned long vaddr)1179 static void __init xen_cleanmfnmap(unsigned long vaddr)
1180 {
1181 	pgd_t *pgd;
1182 	p4d_t *p4d;
1183 	bool unpin;
1184 
1185 	unpin = (vaddr == 2 * PGDIR_SIZE);
1186 	vaddr &= PMD_MASK;
1187 	pgd = pgd_offset_k(vaddr);
1188 	p4d = p4d_offset(pgd, 0);
1189 	if (!p4d_none(*p4d))
1190 		xen_cleanmfnmap_p4d(p4d, unpin);
1191 }
1192 
xen_pagetable_p2m_free(void)1193 static void __init xen_pagetable_p2m_free(void)
1194 {
1195 	unsigned long size;
1196 	unsigned long addr;
1197 
1198 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1199 
1200 	/* No memory or already called. */
1201 	if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1202 		return;
1203 
1204 	/* using __ka address and sticking INVALID_P2M_ENTRY! */
1205 	memset((void *)xen_start_info->mfn_list, 0xff, size);
1206 
1207 	addr = xen_start_info->mfn_list;
1208 	/*
1209 	 * We could be in __ka space.
1210 	 * We roundup to the PMD, which means that if anybody at this stage is
1211 	 * using the __ka address of xen_start_info or
1212 	 * xen_start_info->shared_info they are in going to crash. Fortunately
1213 	 * we have already revectored in xen_setup_kernel_pagetable.
1214 	 */
1215 	size = roundup(size, PMD_SIZE);
1216 
1217 	if (addr >= __START_KERNEL_map) {
1218 		xen_cleanhighmap(addr, addr + size);
1219 		size = PAGE_ALIGN(xen_start_info->nr_pages *
1220 				  sizeof(unsigned long));
1221 		memblock_free((void *)addr, size);
1222 	} else {
1223 		xen_cleanmfnmap(addr);
1224 	}
1225 }
1226 
xen_pagetable_cleanhighmap(void)1227 static void __init xen_pagetable_cleanhighmap(void)
1228 {
1229 	unsigned long size;
1230 	unsigned long addr;
1231 
1232 	/* At this stage, cleanup_highmap has already cleaned __ka space
1233 	 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1234 	 * the ramdisk). We continue on, erasing PMD entries that point to page
1235 	 * tables - do note that they are accessible at this stage via __va.
1236 	 * As Xen is aligning the memory end to a 4MB boundary, for good
1237 	 * measure we also round up to PMD_SIZE * 2 - which means that if
1238 	 * anybody is using __ka address to the initial boot-stack - and try
1239 	 * to use it - they are going to crash. The xen_start_info has been
1240 	 * taken care of already in xen_setup_kernel_pagetable. */
1241 	addr = xen_start_info->pt_base;
1242 	size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1243 
1244 	xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1245 	xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1246 }
1247 
xen_pagetable_p2m_setup(void)1248 static void __init xen_pagetable_p2m_setup(void)
1249 {
1250 	xen_vmalloc_p2m_tree();
1251 
1252 	xen_pagetable_p2m_free();
1253 
1254 	xen_pagetable_cleanhighmap();
1255 
1256 	/* And revector! Bye bye old array */
1257 	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1258 }
1259 
xen_pagetable_init(void)1260 static void __init xen_pagetable_init(void)
1261 {
1262 	/*
1263 	 * The majority of further PTE writes is to pagetables already
1264 	 * announced as such to Xen. Hence it is more efficient to use
1265 	 * hypercalls for these updates.
1266 	 */
1267 	pv_ops.mmu.set_pte = __xen_set_pte;
1268 
1269 	paging_init();
1270 	xen_post_allocator_init();
1271 
1272 	xen_pagetable_p2m_setup();
1273 
1274 	/* Allocate and initialize top and mid mfn levels for p2m structure */
1275 	xen_build_mfn_list_list();
1276 
1277 	/* Remap memory freed due to conflicts with E820 map */
1278 	xen_remap_memory();
1279 	xen_setup_mfn_list_list();
1280 }
1281 
xen_write_cr2(unsigned long cr2)1282 static noinstr void xen_write_cr2(unsigned long cr2)
1283 {
1284 	this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1285 }
1286 
xen_flush_tlb(void)1287 static noinline void xen_flush_tlb(void)
1288 {
1289 	struct mmuext_op *op;
1290 	struct multicall_space mcs;
1291 
1292 	preempt_disable();
1293 
1294 	mcs = xen_mc_entry(sizeof(*op));
1295 
1296 	op = mcs.args;
1297 	op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1298 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1299 
1300 	xen_mc_issue(XEN_LAZY_MMU);
1301 
1302 	preempt_enable();
1303 }
1304 
xen_flush_tlb_one_user(unsigned long addr)1305 static void xen_flush_tlb_one_user(unsigned long addr)
1306 {
1307 	struct mmuext_op *op;
1308 	struct multicall_space mcs;
1309 
1310 	trace_xen_mmu_flush_tlb_one_user(addr);
1311 
1312 	preempt_disable();
1313 
1314 	mcs = xen_mc_entry(sizeof(*op));
1315 	op = mcs.args;
1316 	op->cmd = MMUEXT_INVLPG_LOCAL;
1317 	op->arg1.linear_addr = addr & PAGE_MASK;
1318 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1319 
1320 	xen_mc_issue(XEN_LAZY_MMU);
1321 
1322 	preempt_enable();
1323 }
1324 
xen_flush_tlb_multi(const struct cpumask * cpus,const struct flush_tlb_info * info)1325 static void xen_flush_tlb_multi(const struct cpumask *cpus,
1326 				const struct flush_tlb_info *info)
1327 {
1328 	struct {
1329 		struct mmuext_op op;
1330 		DECLARE_BITMAP(mask, NR_CPUS);
1331 	} *args;
1332 	struct multicall_space mcs;
1333 	const size_t mc_entry_size = sizeof(args->op) +
1334 		sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1335 
1336 	trace_xen_mmu_flush_tlb_multi(cpus, info->mm, info->start, info->end);
1337 
1338 	if (cpumask_empty(cpus))
1339 		return;		/* nothing to do */
1340 
1341 	mcs = xen_mc_entry(mc_entry_size);
1342 	args = mcs.args;
1343 	args->op.arg2.vcpumask = to_cpumask(args->mask);
1344 
1345 	/* Remove any offline CPUs */
1346 	cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1347 
1348 	args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1349 	if (info->end != TLB_FLUSH_ALL &&
1350 	    (info->end - info->start) <= PAGE_SIZE) {
1351 		args->op.cmd = MMUEXT_INVLPG_MULTI;
1352 		args->op.arg1.linear_addr = info->start;
1353 	}
1354 
1355 	MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1356 
1357 	xen_mc_issue(XEN_LAZY_MMU);
1358 }
1359 
xen_read_cr3(void)1360 static unsigned long xen_read_cr3(void)
1361 {
1362 	return this_cpu_read(xen_cr3);
1363 }
1364 
set_current_cr3(void * v)1365 static void set_current_cr3(void *v)
1366 {
1367 	this_cpu_write(xen_current_cr3, (unsigned long)v);
1368 }
1369 
__xen_write_cr3(bool kernel,unsigned long cr3)1370 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1371 {
1372 	struct mmuext_op op;
1373 	unsigned long mfn;
1374 
1375 	trace_xen_mmu_write_cr3(kernel, cr3);
1376 
1377 	if (cr3)
1378 		mfn = pfn_to_mfn(PFN_DOWN(cr3));
1379 	else
1380 		mfn = 0;
1381 
1382 	WARN_ON(mfn == 0 && kernel);
1383 
1384 	op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1385 	op.arg1.mfn = mfn;
1386 
1387 	xen_extend_mmuext_op(&op);
1388 
1389 	if (kernel) {
1390 		this_cpu_write(xen_cr3, cr3);
1391 
1392 		/* Update xen_current_cr3 once the batch has actually
1393 		   been submitted. */
1394 		xen_mc_callback(set_current_cr3, (void *)cr3);
1395 	}
1396 }
xen_write_cr3(unsigned long cr3)1397 static void xen_write_cr3(unsigned long cr3)
1398 {
1399 	pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1400 
1401 	BUG_ON(preemptible());
1402 
1403 	xen_mc_batch();  /* disables interrupts */
1404 
1405 	/* Update while interrupts are disabled, so its atomic with
1406 	   respect to ipis */
1407 	this_cpu_write(xen_cr3, cr3);
1408 
1409 	__xen_write_cr3(true, cr3);
1410 
1411 	if (user_pgd)
1412 		__xen_write_cr3(false, __pa(user_pgd));
1413 	else
1414 		__xen_write_cr3(false, 0);
1415 
1416 	xen_mc_issue(XEN_LAZY_CPU);  /* interrupts restored */
1417 }
1418 
1419 /*
1420  * At the start of the day - when Xen launches a guest, it has already
1421  * built pagetables for the guest. We diligently look over them
1422  * in xen_setup_kernel_pagetable and graft as appropriate them in the
1423  * init_top_pgt and its friends. Then when we are happy we load
1424  * the new init_top_pgt - and continue on.
1425  *
1426  * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1427  * up the rest of the pagetables. When it has completed it loads the cr3.
1428  * N.B. that baremetal would start at 'start_kernel' (and the early
1429  * #PF handler would create bootstrap pagetables) - so we are running
1430  * with the same assumptions as what to do when write_cr3 is executed
1431  * at this point.
1432  *
1433  * Since there are no user-page tables at all, we have two variants
1434  * of xen_write_cr3 - the early bootup (this one), and the late one
1435  * (xen_write_cr3). The reason we have to do that is that in 64-bit
1436  * the Linux kernel and user-space are both in ring 3 while the
1437  * hypervisor is in ring 0.
1438  */
xen_write_cr3_init(unsigned long cr3)1439 static void __init xen_write_cr3_init(unsigned long cr3)
1440 {
1441 	BUG_ON(preemptible());
1442 
1443 	xen_mc_batch();  /* disables interrupts */
1444 
1445 	/* Update while interrupts are disabled, so its atomic with
1446 	   respect to ipis */
1447 	this_cpu_write(xen_cr3, cr3);
1448 
1449 	__xen_write_cr3(true, cr3);
1450 
1451 	xen_mc_issue(XEN_LAZY_CPU);  /* interrupts restored */
1452 }
1453 
xen_pgd_alloc(struct mm_struct * mm)1454 static int xen_pgd_alloc(struct mm_struct *mm)
1455 {
1456 	pgd_t *pgd = mm->pgd;
1457 	struct page *page = virt_to_page(pgd);
1458 	pgd_t *user_pgd;
1459 	int ret = -ENOMEM;
1460 
1461 	BUG_ON(PagePinned(virt_to_page(pgd)));
1462 	BUG_ON(page->private != 0);
1463 
1464 	user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1465 	page->private = (unsigned long)user_pgd;
1466 
1467 	if (user_pgd != NULL) {
1468 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1469 		user_pgd[pgd_index(VSYSCALL_ADDR)] =
1470 			__pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1471 #endif
1472 		ret = 0;
1473 	}
1474 
1475 	BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1476 
1477 	return ret;
1478 }
1479 
xen_pgd_free(struct mm_struct * mm,pgd_t * pgd)1480 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1481 {
1482 	pgd_t *user_pgd = xen_get_user_pgd(pgd);
1483 
1484 	if (user_pgd)
1485 		free_page((unsigned long)user_pgd);
1486 }
1487 
1488 /*
1489  * Init-time set_pte while constructing initial pagetables, which
1490  * doesn't allow RO page table pages to be remapped RW.
1491  *
1492  * If there is no MFN for this PFN then this page is initially
1493  * ballooned out so clear the PTE (as in decrease_reservation() in
1494  * drivers/xen/balloon.c).
1495  *
1496  * Many of these PTE updates are done on unpinned and writable pages
1497  * and doing a hypercall for these is unnecessary and expensive.  At
1498  * this point it is rarely possible to tell if a page is pinned, so
1499  * mostly write the PTE directly and rely on Xen trapping and
1500  * emulating any updates as necessary.
1501  */
xen_set_pte_init(pte_t * ptep,pte_t pte)1502 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1503 {
1504 	if (unlikely(is_early_ioremap_ptep(ptep)))
1505 		__xen_set_pte(ptep, pte);
1506 	else
1507 		native_set_pte(ptep, pte);
1508 }
1509 
xen_make_pte_init(pteval_t pte)1510 __visible pte_t xen_make_pte_init(pteval_t pte)
1511 {
1512 	unsigned long pfn;
1513 
1514 	/*
1515 	 * Pages belonging to the initial p2m list mapped outside the default
1516 	 * address range must be mapped read-only. This region contains the
1517 	 * page tables for mapping the p2m list, too, and page tables MUST be
1518 	 * mapped read-only.
1519 	 */
1520 	pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1521 	if (xen_start_info->mfn_list < __START_KERNEL_map &&
1522 	    pfn >= xen_start_info->first_p2m_pfn &&
1523 	    pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1524 		pte &= ~_PAGE_RW;
1525 
1526 	pte = pte_pfn_to_mfn(pte);
1527 	return native_make_pte(pte);
1528 }
1529 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1530 
1531 /* Early in boot, while setting up the initial pagetable, assume
1532    everything is pinned. */
xen_alloc_pte_init(struct mm_struct * mm,unsigned long pfn)1533 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1534 {
1535 #ifdef CONFIG_FLATMEM
1536 	BUG_ON(mem_map);	/* should only be used early */
1537 #endif
1538 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1539 	pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1540 }
1541 
1542 /* Used for pmd and pud */
xen_alloc_pmd_init(struct mm_struct * mm,unsigned long pfn)1543 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1544 {
1545 #ifdef CONFIG_FLATMEM
1546 	BUG_ON(mem_map);	/* should only be used early */
1547 #endif
1548 	make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1549 }
1550 
1551 /* Early release_pte assumes that all pts are pinned, since there's
1552    only init_mm and anything attached to that is pinned. */
xen_release_pte_init(unsigned long pfn)1553 static void __init xen_release_pte_init(unsigned long pfn)
1554 {
1555 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1556 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1557 }
1558 
xen_release_pmd_init(unsigned long pfn)1559 static void __init xen_release_pmd_init(unsigned long pfn)
1560 {
1561 	make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1562 }
1563 
__pin_pagetable_pfn(unsigned cmd,unsigned long pfn)1564 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1565 {
1566 	struct multicall_space mcs;
1567 	struct mmuext_op *op;
1568 
1569 	mcs = __xen_mc_entry(sizeof(*op));
1570 	op = mcs.args;
1571 	op->cmd = cmd;
1572 	op->arg1.mfn = pfn_to_mfn(pfn);
1573 
1574 	MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1575 }
1576 
__set_pfn_prot(unsigned long pfn,pgprot_t prot)1577 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1578 {
1579 	struct multicall_space mcs;
1580 	unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1581 
1582 	mcs = __xen_mc_entry(0);
1583 	MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1584 				pfn_pte(pfn, prot), 0);
1585 }
1586 
1587 /* This needs to make sure the new pte page is pinned iff its being
1588    attached to a pinned pagetable. */
xen_alloc_ptpage(struct mm_struct * mm,unsigned long pfn,unsigned level)1589 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1590 				    unsigned level)
1591 {
1592 	bool pinned = xen_page_pinned(mm->pgd);
1593 
1594 	trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1595 
1596 	if (pinned) {
1597 		struct page *page = pfn_to_page(pfn);
1598 
1599 		pinned = false;
1600 		if (static_branch_likely(&xen_struct_pages_ready)) {
1601 			pinned = PagePinned(page);
1602 			SetPagePinned(page);
1603 		}
1604 
1605 		xen_mc_batch();
1606 
1607 		__set_pfn_prot(pfn, PAGE_KERNEL_RO);
1608 
1609 		if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS && !pinned)
1610 			__pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1611 
1612 		xen_mc_issue(XEN_LAZY_MMU);
1613 	}
1614 }
1615 
xen_alloc_pte(struct mm_struct * mm,unsigned long pfn)1616 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1617 {
1618 	xen_alloc_ptpage(mm, pfn, PT_PTE);
1619 }
1620 
xen_alloc_pmd(struct mm_struct * mm,unsigned long pfn)1621 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1622 {
1623 	xen_alloc_ptpage(mm, pfn, PT_PMD);
1624 }
1625 
1626 /* This should never happen until we're OK to use struct page */
xen_release_ptpage(unsigned long pfn,unsigned level)1627 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1628 {
1629 	struct page *page = pfn_to_page(pfn);
1630 	bool pinned = PagePinned(page);
1631 
1632 	trace_xen_mmu_release_ptpage(pfn, level, pinned);
1633 
1634 	if (pinned) {
1635 		xen_mc_batch();
1636 
1637 		if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1638 			__pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1639 
1640 		__set_pfn_prot(pfn, PAGE_KERNEL);
1641 
1642 		xen_mc_issue(XEN_LAZY_MMU);
1643 
1644 		ClearPagePinned(page);
1645 	}
1646 }
1647 
xen_release_pte(unsigned long pfn)1648 static void xen_release_pte(unsigned long pfn)
1649 {
1650 	xen_release_ptpage(pfn, PT_PTE);
1651 }
1652 
xen_release_pmd(unsigned long pfn)1653 static void xen_release_pmd(unsigned long pfn)
1654 {
1655 	xen_release_ptpage(pfn, PT_PMD);
1656 }
1657 
xen_alloc_pud(struct mm_struct * mm,unsigned long pfn)1658 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1659 {
1660 	xen_alloc_ptpage(mm, pfn, PT_PUD);
1661 }
1662 
xen_release_pud(unsigned long pfn)1663 static void xen_release_pud(unsigned long pfn)
1664 {
1665 	xen_release_ptpage(pfn, PT_PUD);
1666 }
1667 
1668 /*
1669  * Like __va(), but returns address in the kernel mapping (which is
1670  * all we have until the physical memory mapping has been set up.
1671  */
__ka(phys_addr_t paddr)1672 static void * __init __ka(phys_addr_t paddr)
1673 {
1674 	return (void *)(paddr + __START_KERNEL_map);
1675 }
1676 
1677 /* Convert a machine address to physical address */
m2p(phys_addr_t maddr)1678 static unsigned long __init m2p(phys_addr_t maddr)
1679 {
1680 	phys_addr_t paddr;
1681 
1682 	maddr &= XEN_PTE_MFN_MASK;
1683 	paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1684 
1685 	return paddr;
1686 }
1687 
1688 /* Convert a machine address to kernel virtual */
m2v(phys_addr_t maddr)1689 static void * __init m2v(phys_addr_t maddr)
1690 {
1691 	return __ka(m2p(maddr));
1692 }
1693 
1694 /* Set the page permissions on an identity-mapped pages */
set_page_prot_flags(void * addr,pgprot_t prot,unsigned long flags)1695 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1696 				       unsigned long flags)
1697 {
1698 	unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1699 	pte_t pte = pfn_pte(pfn, prot);
1700 
1701 	if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1702 		BUG();
1703 }
set_page_prot(void * addr,pgprot_t prot)1704 static void __init set_page_prot(void *addr, pgprot_t prot)
1705 {
1706 	return set_page_prot_flags(addr, prot, UVMF_NONE);
1707 }
1708 
xen_setup_machphys_mapping(void)1709 void __init xen_setup_machphys_mapping(void)
1710 {
1711 	struct xen_machphys_mapping mapping;
1712 
1713 	if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1714 		machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1715 		machine_to_phys_nr = mapping.max_mfn + 1;
1716 	} else {
1717 		machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1718 	}
1719 }
1720 
convert_pfn_mfn(void * v)1721 static void __init convert_pfn_mfn(void *v)
1722 {
1723 	pte_t *pte = v;
1724 	int i;
1725 
1726 	/* All levels are converted the same way, so just treat them
1727 	   as ptes. */
1728 	for (i = 0; i < PTRS_PER_PTE; i++)
1729 		pte[i] = xen_make_pte(pte[i].pte);
1730 }
check_pt_base(unsigned long * pt_base,unsigned long * pt_end,unsigned long addr)1731 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1732 				 unsigned long addr)
1733 {
1734 	if (*pt_base == PFN_DOWN(__pa(addr))) {
1735 		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1736 		clear_page((void *)addr);
1737 		(*pt_base)++;
1738 	}
1739 	if (*pt_end == PFN_DOWN(__pa(addr))) {
1740 		set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1741 		clear_page((void *)addr);
1742 		(*pt_end)--;
1743 	}
1744 }
1745 /*
1746  * Set up the initial kernel pagetable.
1747  *
1748  * We can construct this by grafting the Xen provided pagetable into
1749  * head_64.S's preconstructed pagetables.  We copy the Xen L2's into
1750  * level2_ident_pgt, and level2_kernel_pgt.  This means that only the
1751  * kernel has a physical mapping to start with - but that's enough to
1752  * get __va working.  We need to fill in the rest of the physical
1753  * mapping once some sort of allocator has been set up.
1754  */
xen_setup_kernel_pagetable(pgd_t * pgd,unsigned long max_pfn)1755 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1756 {
1757 	pud_t *l3;
1758 	pmd_t *l2;
1759 	unsigned long addr[3];
1760 	unsigned long pt_base, pt_end;
1761 	unsigned i;
1762 
1763 	/* max_pfn_mapped is the last pfn mapped in the initial memory
1764 	 * mappings. Considering that on Xen after the kernel mappings we
1765 	 * have the mappings of some pages that don't exist in pfn space, we
1766 	 * set max_pfn_mapped to the last real pfn mapped. */
1767 	if (xen_start_info->mfn_list < __START_KERNEL_map)
1768 		max_pfn_mapped = xen_start_info->first_p2m_pfn;
1769 	else
1770 		max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1771 
1772 	pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1773 	pt_end = pt_base + xen_start_info->nr_pt_frames;
1774 
1775 	/* Zap identity mapping */
1776 	init_top_pgt[0] = __pgd(0);
1777 
1778 	/* Pre-constructed entries are in pfn, so convert to mfn */
1779 	/* L4[273] -> level3_ident_pgt  */
1780 	/* L4[511] -> level3_kernel_pgt */
1781 	convert_pfn_mfn(init_top_pgt);
1782 
1783 	/* L3_i[0] -> level2_ident_pgt */
1784 	convert_pfn_mfn(level3_ident_pgt);
1785 	/* L3_k[510] -> level2_kernel_pgt */
1786 	/* L3_k[511] -> level2_fixmap_pgt */
1787 	convert_pfn_mfn(level3_kernel_pgt);
1788 
1789 	/* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */
1790 	convert_pfn_mfn(level2_fixmap_pgt);
1791 
1792 	/* We get [511][511] and have Xen's version of level2_kernel_pgt */
1793 	l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1794 	l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1795 
1796 	addr[0] = (unsigned long)pgd;
1797 	addr[1] = (unsigned long)l3;
1798 	addr[2] = (unsigned long)l2;
1799 	/* Graft it onto L4[273][0]. Note that we creating an aliasing problem:
1800 	 * Both L4[273][0] and L4[511][510] have entries that point to the same
1801 	 * L2 (PMD) tables. Meaning that if you modify it in __va space
1802 	 * it will be also modified in the __ka space! (But if you just
1803 	 * modify the PMD table to point to other PTE's or none, then you
1804 	 * are OK - which is what cleanup_highmap does) */
1805 	copy_page(level2_ident_pgt, l2);
1806 	/* Graft it onto L4[511][510] */
1807 	copy_page(level2_kernel_pgt, l2);
1808 
1809 	/*
1810 	 * Zap execute permission from the ident map. Due to the sharing of
1811 	 * L1 entries we need to do this in the L2.
1812 	 */
1813 	if (__supported_pte_mask & _PAGE_NX) {
1814 		for (i = 0; i < PTRS_PER_PMD; ++i) {
1815 			if (pmd_none(level2_ident_pgt[i]))
1816 				continue;
1817 			level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1818 		}
1819 	}
1820 
1821 	/* Copy the initial P->M table mappings if necessary. */
1822 	i = pgd_index(xen_start_info->mfn_list);
1823 	if (i && i < pgd_index(__START_KERNEL_map))
1824 		init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1825 
1826 	/* Make pagetable pieces RO */
1827 	set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1828 	set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1829 	set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1830 	set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1831 	set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1832 	set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1833 
1834 	for (i = 0; i < FIXMAP_PMD_NUM; i++) {
1835 		set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE,
1836 			      PAGE_KERNEL_RO);
1837 	}
1838 
1839 	/* Pin down new L4 */
1840 	pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1841 			  PFN_DOWN(__pa_symbol(init_top_pgt)));
1842 
1843 	/* Unpin Xen-provided one */
1844 	pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1845 
1846 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1847 	/* Pin user vsyscall L3 */
1848 	set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1849 	pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
1850 			  PFN_DOWN(__pa_symbol(level3_user_vsyscall)));
1851 #endif
1852 
1853 	/*
1854 	 * At this stage there can be no user pgd, and no page structure to
1855 	 * attach it to, so make sure we just set kernel pgd.
1856 	 */
1857 	xen_mc_batch();
1858 	__xen_write_cr3(true, __pa(init_top_pgt));
1859 	xen_mc_issue(XEN_LAZY_CPU);
1860 
1861 	/* We can't that easily rip out L3 and L2, as the Xen pagetables are
1862 	 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ...  for
1863 	 * the initial domain. For guests using the toolstack, they are in:
1864 	 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1865 	 * rip out the [L4] (pgd), but for guests we shave off three pages.
1866 	 */
1867 	for (i = 0; i < ARRAY_SIZE(addr); i++)
1868 		check_pt_base(&pt_base, &pt_end, addr[i]);
1869 
1870 	/* Our (by three pages) smaller Xen pagetable that we are using */
1871 	xen_pt_base = PFN_PHYS(pt_base);
1872 	xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1873 	memblock_reserve(xen_pt_base, xen_pt_size);
1874 
1875 	/* Revector the xen_start_info */
1876 	xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1877 }
1878 
1879 /*
1880  * Read a value from a physical address.
1881  */
xen_read_phys_ulong(phys_addr_t addr)1882 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1883 {
1884 	unsigned long *vaddr;
1885 	unsigned long val;
1886 
1887 	vaddr = early_memremap_ro(addr, sizeof(val));
1888 	val = *vaddr;
1889 	early_memunmap(vaddr, sizeof(val));
1890 	return val;
1891 }
1892 
1893 /*
1894  * Translate a virtual address to a physical one without relying on mapped
1895  * page tables. Don't rely on big pages being aligned in (guest) physical
1896  * space!
1897  */
xen_early_virt_to_phys(unsigned long vaddr)1898 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
1899 {
1900 	phys_addr_t pa;
1901 	pgd_t pgd;
1902 	pud_t pud;
1903 	pmd_t pmd;
1904 	pte_t pte;
1905 
1906 	pa = read_cr3_pa();
1907 	pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
1908 						       sizeof(pgd)));
1909 	if (!pgd_present(pgd))
1910 		return 0;
1911 
1912 	pa = pgd_val(pgd) & PTE_PFN_MASK;
1913 	pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
1914 						       sizeof(pud)));
1915 	if (!pud_present(pud))
1916 		return 0;
1917 	pa = pud_val(pud) & PTE_PFN_MASK;
1918 	if (pud_leaf(pud))
1919 		return pa + (vaddr & ~PUD_MASK);
1920 
1921 	pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
1922 						       sizeof(pmd)));
1923 	if (!pmd_present(pmd))
1924 		return 0;
1925 	pa = pmd_val(pmd) & PTE_PFN_MASK;
1926 	if (pmd_large(pmd))
1927 		return pa + (vaddr & ~PMD_MASK);
1928 
1929 	pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
1930 						       sizeof(pte)));
1931 	if (!pte_present(pte))
1932 		return 0;
1933 	pa = pte_pfn(pte) << PAGE_SHIFT;
1934 
1935 	return pa | (vaddr & ~PAGE_MASK);
1936 }
1937 
1938 /*
1939  * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
1940  * this area.
1941  */
xen_relocate_p2m(void)1942 void __init xen_relocate_p2m(void)
1943 {
1944 	phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
1945 	unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
1946 	int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
1947 	pte_t *pt;
1948 	pmd_t *pmd;
1949 	pud_t *pud;
1950 	pgd_t *pgd;
1951 	unsigned long *new_p2m;
1952 
1953 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1954 	n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
1955 	n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
1956 	n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
1957 	n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
1958 	n_frames = n_pte + n_pt + n_pmd + n_pud;
1959 
1960 	new_area = xen_find_free_area(PFN_PHYS(n_frames));
1961 	if (!new_area) {
1962 		xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
1963 		BUG();
1964 	}
1965 
1966 	/*
1967 	 * Setup the page tables for addressing the new p2m list.
1968 	 * We have asked the hypervisor to map the p2m list at the user address
1969 	 * PUD_SIZE. It may have done so, or it may have used a kernel space
1970 	 * address depending on the Xen version.
1971 	 * To avoid any possible virtual address collision, just use
1972 	 * 2 * PUD_SIZE for the new area.
1973 	 */
1974 	pud_phys = new_area;
1975 	pmd_phys = pud_phys + PFN_PHYS(n_pud);
1976 	pt_phys = pmd_phys + PFN_PHYS(n_pmd);
1977 	p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
1978 
1979 	pgd = __va(read_cr3_pa());
1980 	new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
1981 	for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
1982 		pud = early_memremap(pud_phys, PAGE_SIZE);
1983 		clear_page(pud);
1984 		for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
1985 				idx_pmd++) {
1986 			pmd = early_memremap(pmd_phys, PAGE_SIZE);
1987 			clear_page(pmd);
1988 			for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
1989 					idx_pt++) {
1990 				pt = early_memremap(pt_phys, PAGE_SIZE);
1991 				clear_page(pt);
1992 				for (idx_pte = 0;
1993 				     idx_pte < min(n_pte, PTRS_PER_PTE);
1994 				     idx_pte++) {
1995 					pt[idx_pte] = pfn_pte(p2m_pfn,
1996 							      PAGE_KERNEL);
1997 					p2m_pfn++;
1998 				}
1999 				n_pte -= PTRS_PER_PTE;
2000 				early_memunmap(pt, PAGE_SIZE);
2001 				make_lowmem_page_readonly(__va(pt_phys));
2002 				pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2003 						PFN_DOWN(pt_phys));
2004 				pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys);
2005 				pt_phys += PAGE_SIZE;
2006 			}
2007 			n_pt -= PTRS_PER_PMD;
2008 			early_memunmap(pmd, PAGE_SIZE);
2009 			make_lowmem_page_readonly(__va(pmd_phys));
2010 			pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2011 					PFN_DOWN(pmd_phys));
2012 			pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys);
2013 			pmd_phys += PAGE_SIZE;
2014 		}
2015 		n_pmd -= PTRS_PER_PUD;
2016 		early_memunmap(pud, PAGE_SIZE);
2017 		make_lowmem_page_readonly(__va(pud_phys));
2018 		pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2019 		set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2020 		pud_phys += PAGE_SIZE;
2021 	}
2022 
2023 	/* Now copy the old p2m info to the new area. */
2024 	memcpy(new_p2m, xen_p2m_addr, size);
2025 	xen_p2m_addr = new_p2m;
2026 
2027 	/* Release the old p2m list and set new list info. */
2028 	p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2029 	BUG_ON(!p2m_pfn);
2030 	p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2031 
2032 	if (xen_start_info->mfn_list < __START_KERNEL_map) {
2033 		pfn = xen_start_info->first_p2m_pfn;
2034 		pfn_end = xen_start_info->first_p2m_pfn +
2035 			  xen_start_info->nr_p2m_frames;
2036 		set_pgd(pgd + 1, __pgd(0));
2037 	} else {
2038 		pfn = p2m_pfn;
2039 		pfn_end = p2m_pfn_end;
2040 	}
2041 
2042 	memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2043 	while (pfn < pfn_end) {
2044 		if (pfn == p2m_pfn) {
2045 			pfn = p2m_pfn_end;
2046 			continue;
2047 		}
2048 		make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2049 		pfn++;
2050 	}
2051 
2052 	xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2053 	xen_start_info->first_p2m_pfn =  PFN_DOWN(new_area);
2054 	xen_start_info->nr_p2m_frames = n_frames;
2055 }
2056 
xen_reserve_special_pages(void)2057 void __init xen_reserve_special_pages(void)
2058 {
2059 	phys_addr_t paddr;
2060 
2061 	memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2062 	if (xen_start_info->store_mfn) {
2063 		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2064 		memblock_reserve(paddr, PAGE_SIZE);
2065 	}
2066 	if (!xen_initial_domain()) {
2067 		paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2068 		memblock_reserve(paddr, PAGE_SIZE);
2069 	}
2070 }
2071 
xen_pt_check_e820(void)2072 void __init xen_pt_check_e820(void)
2073 {
2074 	xen_chk_is_e820_usable(xen_pt_base, xen_pt_size, "page table");
2075 }
2076 
2077 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2078 
xen_set_fixmap(unsigned idx,phys_addr_t phys,pgprot_t prot)2079 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2080 {
2081 	pte_t pte;
2082 	unsigned long vaddr;
2083 
2084 	phys >>= PAGE_SHIFT;
2085 
2086 	switch (idx) {
2087 	case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2088 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2089 	case VSYSCALL_PAGE:
2090 #endif
2091 		/* All local page mappings */
2092 		pte = pfn_pte(phys, prot);
2093 		break;
2094 
2095 #ifdef CONFIG_X86_LOCAL_APIC
2096 	case FIX_APIC_BASE:	/* maps dummy local APIC */
2097 		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2098 		break;
2099 #endif
2100 
2101 #ifdef CONFIG_X86_IO_APIC
2102 	case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2103 		/*
2104 		 * We just don't map the IO APIC - all access is via
2105 		 * hypercalls.  Keep the address in the pte for reference.
2106 		 */
2107 		pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2108 		break;
2109 #endif
2110 
2111 	case FIX_PARAVIRT_BOOTMAP:
2112 		/* This is an MFN, but it isn't an IO mapping from the
2113 		   IO domain */
2114 		pte = mfn_pte(phys, prot);
2115 		break;
2116 
2117 	default:
2118 		/* By default, set_fixmap is used for hardware mappings */
2119 		pte = mfn_pte(phys, prot);
2120 		break;
2121 	}
2122 
2123 	vaddr = __fix_to_virt(idx);
2124 	if (HYPERVISOR_update_va_mapping(vaddr, pte, UVMF_INVLPG))
2125 		BUG();
2126 
2127 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2128 	/* Replicate changes to map the vsyscall page into the user
2129 	   pagetable vsyscall mapping. */
2130 	if (idx == VSYSCALL_PAGE)
2131 		set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2132 #endif
2133 }
2134 
xen_enter_lazy_mmu(void)2135 static void xen_enter_lazy_mmu(void)
2136 {
2137 	enter_lazy(XEN_LAZY_MMU);
2138 }
2139 
xen_flush_lazy_mmu(void)2140 static void xen_flush_lazy_mmu(void)
2141 {
2142 	preempt_disable();
2143 
2144 	if (xen_get_lazy_mode() == XEN_LAZY_MMU) {
2145 		arch_leave_lazy_mmu_mode();
2146 		arch_enter_lazy_mmu_mode();
2147 	}
2148 
2149 	preempt_enable();
2150 }
2151 
xen_post_allocator_init(void)2152 static void __init xen_post_allocator_init(void)
2153 {
2154 	pv_ops.mmu.set_pte = xen_set_pte;
2155 	pv_ops.mmu.set_pmd = xen_set_pmd;
2156 	pv_ops.mmu.set_pud = xen_set_pud;
2157 	pv_ops.mmu.set_p4d = xen_set_p4d;
2158 
2159 	/* This will work as long as patching hasn't happened yet
2160 	   (which it hasn't) */
2161 	pv_ops.mmu.alloc_pte = xen_alloc_pte;
2162 	pv_ops.mmu.alloc_pmd = xen_alloc_pmd;
2163 	pv_ops.mmu.release_pte = xen_release_pte;
2164 	pv_ops.mmu.release_pmd = xen_release_pmd;
2165 	pv_ops.mmu.alloc_pud = xen_alloc_pud;
2166 	pv_ops.mmu.release_pud = xen_release_pud;
2167 	pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2168 
2169 	pv_ops.mmu.write_cr3 = &xen_write_cr3;
2170 }
2171 
xen_leave_lazy_mmu(void)2172 static void xen_leave_lazy_mmu(void)
2173 {
2174 	preempt_disable();
2175 	xen_mc_flush();
2176 	leave_lazy(XEN_LAZY_MMU);
2177 	preempt_enable();
2178 }
2179 
2180 static const typeof(pv_ops) xen_mmu_ops __initconst = {
2181 	.mmu = {
2182 		.read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2),
2183 		.write_cr2 = xen_write_cr2,
2184 
2185 		.read_cr3 = xen_read_cr3,
2186 		.write_cr3 = xen_write_cr3_init,
2187 
2188 		.flush_tlb_user = xen_flush_tlb,
2189 		.flush_tlb_kernel = xen_flush_tlb,
2190 		.flush_tlb_one_user = xen_flush_tlb_one_user,
2191 		.flush_tlb_multi = xen_flush_tlb_multi,
2192 		.tlb_remove_table = tlb_remove_table,
2193 
2194 		.pgd_alloc = xen_pgd_alloc,
2195 		.pgd_free = xen_pgd_free,
2196 
2197 		.alloc_pte = xen_alloc_pte_init,
2198 		.release_pte = xen_release_pte_init,
2199 		.alloc_pmd = xen_alloc_pmd_init,
2200 		.release_pmd = xen_release_pmd_init,
2201 
2202 		.set_pte = xen_set_pte_init,
2203 		.set_pmd = xen_set_pmd_hyper,
2204 
2205 		.ptep_modify_prot_start = xen_ptep_modify_prot_start,
2206 		.ptep_modify_prot_commit = xen_ptep_modify_prot_commit,
2207 
2208 		.pte_val = PV_CALLEE_SAVE(xen_pte_val),
2209 		.pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2210 
2211 		.make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2212 		.make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2213 
2214 		.set_pud = xen_set_pud_hyper,
2215 
2216 		.make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2217 		.pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2218 
2219 		.pud_val = PV_CALLEE_SAVE(xen_pud_val),
2220 		.make_pud = PV_CALLEE_SAVE(xen_make_pud),
2221 		.set_p4d = xen_set_p4d_hyper,
2222 
2223 		.alloc_pud = xen_alloc_pmd_init,
2224 		.release_pud = xen_release_pmd_init,
2225 
2226 #if CONFIG_PGTABLE_LEVELS >= 5
2227 		.p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2228 		.make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2229 #endif
2230 
2231 		.enter_mmap = xen_enter_mmap,
2232 		.exit_mmap = xen_exit_mmap,
2233 
2234 		.lazy_mode = {
2235 			.enter = xen_enter_lazy_mmu,
2236 			.leave = xen_leave_lazy_mmu,
2237 			.flush = xen_flush_lazy_mmu,
2238 		},
2239 
2240 		.set_fixmap = xen_set_fixmap,
2241 	},
2242 };
2243 
xen_init_mmu_ops(void)2244 void __init xen_init_mmu_ops(void)
2245 {
2246 	x86_init.paging.pagetable_init = xen_pagetable_init;
2247 	x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2248 
2249 	pv_ops.mmu = xen_mmu_ops.mmu;
2250 
2251 	memset(dummy_mapping, 0xff, PAGE_SIZE);
2252 }
2253 
2254 #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)2255 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2256 				unsigned long *in_frames,
2257 				unsigned long *out_frames)
2258 {
2259 	int i;
2260 	struct multicall_space mcs;
2261 
2262 	xen_mc_batch();
2263 	for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2264 		mcs = __xen_mc_entry(0);
2265 
2266 		if (in_frames)
2267 			in_frames[i] = virt_to_mfn((void *)vaddr);
2268 
2269 		MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2270 		__set_phys_to_machine(virt_to_pfn((void *)vaddr), INVALID_P2M_ENTRY);
2271 
2272 		if (out_frames)
2273 			out_frames[i] = virt_to_pfn((void *)vaddr);
2274 	}
2275 	xen_mc_issue(0);
2276 }
2277 
2278 /*
2279  * Update the pfn-to-mfn mappings for a virtual address range, either to
2280  * point to an array of mfns, or contiguously from a single starting
2281  * mfn.
2282  */
xen_remap_exchanged_ptes(unsigned long vaddr,int order,unsigned long * mfns,unsigned long first_mfn)2283 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2284 				     unsigned long *mfns,
2285 				     unsigned long first_mfn)
2286 {
2287 	unsigned i, limit;
2288 	unsigned long mfn;
2289 
2290 	xen_mc_batch();
2291 
2292 	limit = 1u << order;
2293 	for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2294 		struct multicall_space mcs;
2295 		unsigned flags;
2296 
2297 		mcs = __xen_mc_entry(0);
2298 		if (mfns)
2299 			mfn = mfns[i];
2300 		else
2301 			mfn = first_mfn + i;
2302 
2303 		if (i < (limit - 1))
2304 			flags = 0;
2305 		else {
2306 			if (order == 0)
2307 				flags = UVMF_INVLPG | UVMF_ALL;
2308 			else
2309 				flags = UVMF_TLB_FLUSH | UVMF_ALL;
2310 		}
2311 
2312 		MULTI_update_va_mapping(mcs.mc, vaddr,
2313 				mfn_pte(mfn, PAGE_KERNEL), flags);
2314 
2315 		set_phys_to_machine(virt_to_pfn((void *)vaddr), mfn);
2316 	}
2317 
2318 	xen_mc_issue(0);
2319 }
2320 
2321 /*
2322  * Perform the hypercall to exchange a region of our pfns to point to
2323  * memory with the required contiguous alignment.  Takes the pfns as
2324  * input, and populates mfns as output.
2325  *
2326  * Returns a success code indicating whether the hypervisor was able to
2327  * satisfy the request or not.
2328  */
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)2329 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2330 			       unsigned long *pfns_in,
2331 			       unsigned long extents_out,
2332 			       unsigned int order_out,
2333 			       unsigned long *mfns_out,
2334 			       unsigned int address_bits)
2335 {
2336 	long rc;
2337 	int success;
2338 
2339 	struct xen_memory_exchange exchange = {
2340 		.in = {
2341 			.nr_extents   = extents_in,
2342 			.extent_order = order_in,
2343 			.extent_start = pfns_in,
2344 			.domid        = DOMID_SELF
2345 		},
2346 		.out = {
2347 			.nr_extents   = extents_out,
2348 			.extent_order = order_out,
2349 			.extent_start = mfns_out,
2350 			.address_bits = address_bits,
2351 			.domid        = DOMID_SELF
2352 		}
2353 	};
2354 
2355 	BUG_ON(extents_in << order_in != extents_out << order_out);
2356 
2357 	rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2358 	success = (exchange.nr_exchanged == extents_in);
2359 
2360 	BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2361 	BUG_ON(success && (rc != 0));
2362 
2363 	return success;
2364 }
2365 
xen_create_contiguous_region(phys_addr_t pstart,unsigned int order,unsigned int address_bits,dma_addr_t * dma_handle)2366 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2367 				 unsigned int address_bits,
2368 				 dma_addr_t *dma_handle)
2369 {
2370 	unsigned long *in_frames, out_frame;
2371 	unsigned long  flags;
2372 	int            success;
2373 	unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2374 
2375 	if (unlikely(order > discontig_frames_order)) {
2376 		if (!discontig_frames_dyn)
2377 			return -ENOMEM;
2378 
2379 		if (alloc_discontig_frames(order))
2380 			return -ENOMEM;
2381 	}
2382 
2383 	memset((void *) vstart, 0, PAGE_SIZE << order);
2384 
2385 	spin_lock_irqsave(&xen_reservation_lock, flags);
2386 
2387 	in_frames = discontig_frames;
2388 
2389 	/* 1. Zap current PTEs, remembering MFNs. */
2390 	xen_zap_pfn_range(vstart, order, in_frames, NULL);
2391 
2392 	/* 2. Get a new contiguous memory extent. */
2393 	out_frame = virt_to_pfn((void *)vstart);
2394 	success = xen_exchange_memory(1UL << order, 0, in_frames,
2395 				      1, order, &out_frame,
2396 				      address_bits);
2397 
2398 	/* 3. Map the new extent in place of old pages. */
2399 	if (success)
2400 		xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2401 	else
2402 		xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2403 
2404 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2405 
2406 	*dma_handle = virt_to_machine(vstart).maddr;
2407 	return success ? 0 : -ENOMEM;
2408 }
2409 
xen_destroy_contiguous_region(phys_addr_t pstart,unsigned int order)2410 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2411 {
2412 	unsigned long *out_frames, in_frame;
2413 	unsigned long  flags;
2414 	int success;
2415 	unsigned long vstart;
2416 
2417 	if (unlikely(order > discontig_frames_order))
2418 		return;
2419 
2420 	vstart = (unsigned long)phys_to_virt(pstart);
2421 	memset((void *) vstart, 0, PAGE_SIZE << order);
2422 
2423 	spin_lock_irqsave(&xen_reservation_lock, flags);
2424 
2425 	out_frames = discontig_frames;
2426 
2427 	/* 1. Find start MFN of contiguous extent. */
2428 	in_frame = virt_to_mfn((void *)vstart);
2429 
2430 	/* 2. Zap current PTEs. */
2431 	xen_zap_pfn_range(vstart, order, NULL, out_frames);
2432 
2433 	/* 3. Do the exchange for non-contiguous MFNs. */
2434 	success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2435 					0, out_frames, 0);
2436 
2437 	/* 4. Map new pages in place of old pages. */
2438 	if (success)
2439 		xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2440 	else
2441 		xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2442 
2443 	spin_unlock_irqrestore(&xen_reservation_lock, flags);
2444 }
2445 
xen_flush_tlb_all(void)2446 static noinline void xen_flush_tlb_all(void)
2447 {
2448 	struct mmuext_op *op;
2449 	struct multicall_space mcs;
2450 
2451 	preempt_disable();
2452 
2453 	mcs = xen_mc_entry(sizeof(*op));
2454 
2455 	op = mcs.args;
2456 	op->cmd = MMUEXT_TLB_FLUSH_ALL;
2457 	MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
2458 
2459 	xen_mc_issue(XEN_LAZY_MMU);
2460 
2461 	preempt_enable();
2462 }
2463 
2464 #define REMAP_BATCH_SIZE 16
2465 
2466 struct remap_data {
2467 	xen_pfn_t *pfn;
2468 	bool contiguous;
2469 	bool no_translate;
2470 	pgprot_t prot;
2471 	struct mmu_update *mmu_update;
2472 };
2473 
remap_area_pfn_pte_fn(pte_t * ptep,unsigned long addr,void * data)2474 static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data)
2475 {
2476 	struct remap_data *rmd = data;
2477 	pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot));
2478 
2479 	/*
2480 	 * If we have a contiguous range, just update the pfn itself,
2481 	 * else update pointer to be "next pfn".
2482 	 */
2483 	if (rmd->contiguous)
2484 		(*rmd->pfn)++;
2485 	else
2486 		rmd->pfn++;
2487 
2488 	rmd->mmu_update->ptr = virt_to_machine(ptep).maddr;
2489 	rmd->mmu_update->ptr |= rmd->no_translate ?
2490 		MMU_PT_UPDATE_NO_TRANSLATE :
2491 		MMU_NORMAL_PT_UPDATE;
2492 	rmd->mmu_update->val = pte_val_ma(pte);
2493 	rmd->mmu_update++;
2494 
2495 	return 0;
2496 }
2497 
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)2498 int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr,
2499 		  xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot,
2500 		  unsigned int domid, bool no_translate)
2501 {
2502 	int err = 0;
2503 	struct remap_data rmd;
2504 	struct mmu_update mmu_update[REMAP_BATCH_SIZE];
2505 	unsigned long range;
2506 	int mapped = 0;
2507 
2508 	BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO)));
2509 
2510 	rmd.pfn = pfn;
2511 	rmd.prot = prot;
2512 	/*
2513 	 * We use the err_ptr to indicate if there we are doing a contiguous
2514 	 * mapping or a discontiguous mapping.
2515 	 */
2516 	rmd.contiguous = !err_ptr;
2517 	rmd.no_translate = no_translate;
2518 
2519 	while (nr) {
2520 		int index = 0;
2521 		int done = 0;
2522 		int batch = min(REMAP_BATCH_SIZE, nr);
2523 		int batch_left = batch;
2524 
2525 		range = (unsigned long)batch << PAGE_SHIFT;
2526 
2527 		rmd.mmu_update = mmu_update;
2528 		err = apply_to_page_range(vma->vm_mm, addr, range,
2529 					  remap_area_pfn_pte_fn, &rmd);
2530 		if (err)
2531 			goto out;
2532 
2533 		/*
2534 		 * We record the error for each page that gives an error, but
2535 		 * continue mapping until the whole set is done
2536 		 */
2537 		do {
2538 			int i;
2539 
2540 			err = HYPERVISOR_mmu_update(&mmu_update[index],
2541 						    batch_left, &done, domid);
2542 
2543 			/*
2544 			 * @err_ptr may be the same buffer as @gfn, so
2545 			 * only clear it after each chunk of @gfn is
2546 			 * used.
2547 			 */
2548 			if (err_ptr) {
2549 				for (i = index; i < index + done; i++)
2550 					err_ptr[i] = 0;
2551 			}
2552 			if (err < 0) {
2553 				if (!err_ptr)
2554 					goto out;
2555 				err_ptr[i] = err;
2556 				done++; /* Skip failed frame. */
2557 			} else
2558 				mapped += done;
2559 			batch_left -= done;
2560 			index += done;
2561 		} while (batch_left);
2562 
2563 		nr -= batch;
2564 		addr += range;
2565 		if (err_ptr)
2566 			err_ptr += batch;
2567 		cond_resched();
2568 	}
2569 out:
2570 
2571 	xen_flush_tlb_all();
2572 
2573 	return err < 0 ? err : mapped;
2574 }
2575 EXPORT_SYMBOL_GPL(xen_remap_pfn);
2576 
2577 #ifdef CONFIG_KEXEC_CORE
paddr_vmcoreinfo_note(void)2578 phys_addr_t paddr_vmcoreinfo_note(void)
2579 {
2580 	if (xen_pv_domain())
2581 		return virt_to_machine(vmcoreinfo_note).maddr;
2582 	else
2583 		return __pa(vmcoreinfo_note);
2584 }
2585 #endif /* CONFIG_KEXEC_CORE */
2586