xref: /linux/arch/x86/xen/p2m.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
1 /*
2  * Xen leaves the responsibility for maintaining p2m mappings to the
3  * guests themselves, but it must also access and update the p2m array
4  * during suspend/resume when all the pages are reallocated.
5  *
6  * The logical flat p2m table is mapped to a linear kernel memory area.
7  * For accesses by Xen a three-level tree linked via mfns only is set up to
8  * allow the address space to be sparse.
9  *
10  *               Xen
11  *                |
12  *          p2m_top_mfn
13  *              /   \
14  * p2m_mid_mfn p2m_mid_mfn
15  *         /           /
16  *  p2m p2m p2m ...
17  *
18  * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
19  *
20  * The p2m_top_mfn level is limited to 1 page, so the maximum representable
21  * pseudo-physical address space is:
22  *  P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
23  *
24  * P2M_PER_PAGE depends on the architecture, as a mfn is always
25  * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
26  * 512 and 1024 entries respectively.
27  *
28  * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
29  *
30  * However not all entries are filled with MFNs. Specifically for all other
31  * leaf entries, or for the top  root, or middle one, for which there is a void
32  * entry, we assume it is  "missing". So (for example)
33  *  pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
34  * We have a dedicated page p2m_missing with all entries being
35  * INVALID_P2M_ENTRY. This page may be referenced multiple times in the p2m
36  * list/tree in case there are multiple areas with P2M_PER_PAGE invalid pfns.
37  *
38  * We also have the possibility of setting 1-1 mappings on certain regions, so
39  * that:
40  *  pfn_to_mfn(0xc0000)=0xc0000
41  *
42  * The benefit of this is, that we can assume for non-RAM regions (think
43  * PCI BARs, or ACPI spaces), we can create mappings easily because we
44  * get the PFN value to match the MFN.
45  *
46  * For this to work efficiently we have one new page p2m_identity. All entries
47  * in p2m_identity are set to INVALID_P2M_ENTRY type (Xen toolstack only
48  * recognizes that and MFNs, no other fancy value).
49  *
50  * On lookup we spot that the entry points to p2m_identity and return the
51  * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
52  * If the entry points to an allocated page, we just proceed as before and
53  * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
54  * appropriate functions (pfn_to_mfn).
55  *
56  * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
57  * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
58  * non-identity pfn. To protect ourselves against we elect to set (and get) the
59  * IDENTITY_FRAME_BIT on all identity mapped PFNs.
60  */
61 
62 #include <linux/init.h>
63 #include <linux/export.h>
64 #include <linux/list.h>
65 #include <linux/hash.h>
66 #include <linux/sched.h>
67 #include <linux/seq_file.h>
68 #include <linux/bootmem.h>
69 #include <linux/slab.h>
70 #include <linux/vmalloc.h>
71 
72 #include <asm/cache.h>
73 #include <asm/setup.h>
74 #include <asm/uaccess.h>
75 
76 #include <asm/xen/page.h>
77 #include <asm/xen/hypercall.h>
78 #include <asm/xen/hypervisor.h>
79 #include <xen/balloon.h>
80 #include <xen/grant_table.h>
81 
82 #include "multicalls.h"
83 #include "xen-ops.h"
84 
85 #define P2M_MID_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long *))
86 #define P2M_TOP_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long **))
87 
88 #define MAX_P2M_PFN	(P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
89 
90 #define PMDS_PER_MID_PAGE	(P2M_MID_PER_PAGE / PTRS_PER_PTE)
91 
92 unsigned long *xen_p2m_addr __read_mostly;
93 EXPORT_SYMBOL_GPL(xen_p2m_addr);
94 unsigned long xen_p2m_size __read_mostly;
95 EXPORT_SYMBOL_GPL(xen_p2m_size);
96 unsigned long xen_max_p2m_pfn __read_mostly;
97 EXPORT_SYMBOL_GPL(xen_max_p2m_pfn);
98 
99 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
100 #define P2M_LIMIT CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
101 #else
102 #define P2M_LIMIT 0
103 #endif
104 
105 static DEFINE_SPINLOCK(p2m_update_lock);
106 
107 static unsigned long *p2m_mid_missing_mfn;
108 static unsigned long *p2m_top_mfn;
109 static unsigned long **p2m_top_mfn_p;
110 static unsigned long *p2m_missing;
111 static unsigned long *p2m_identity;
112 static pte_t *p2m_missing_pte;
113 static pte_t *p2m_identity_pte;
114 
115 /*
116  * Hint at last populated PFN.
117  *
118  * Used to set HYPERVISOR_shared_info->arch.max_pfn so the toolstack
119  * can avoid scanning the whole P2M (which may be sized to account for
120  * hotplugged memory).
121  */
122 static unsigned long xen_p2m_last_pfn;
123 
124 static inline unsigned p2m_top_index(unsigned long pfn)
125 {
126 	BUG_ON(pfn >= MAX_P2M_PFN);
127 	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
128 }
129 
130 static inline unsigned p2m_mid_index(unsigned long pfn)
131 {
132 	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
133 }
134 
135 static inline unsigned p2m_index(unsigned long pfn)
136 {
137 	return pfn % P2M_PER_PAGE;
138 }
139 
140 static void p2m_top_mfn_init(unsigned long *top)
141 {
142 	unsigned i;
143 
144 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
145 		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
146 }
147 
148 static void p2m_top_mfn_p_init(unsigned long **top)
149 {
150 	unsigned i;
151 
152 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
153 		top[i] = p2m_mid_missing_mfn;
154 }
155 
156 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
157 {
158 	unsigned i;
159 
160 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
161 		mid[i] = virt_to_mfn(leaf);
162 }
163 
164 static void p2m_init(unsigned long *p2m)
165 {
166 	unsigned i;
167 
168 	for (i = 0; i < P2M_PER_PAGE; i++)
169 		p2m[i] = INVALID_P2M_ENTRY;
170 }
171 
172 static void p2m_init_identity(unsigned long *p2m, unsigned long pfn)
173 {
174 	unsigned i;
175 
176 	for (i = 0; i < P2M_PER_PAGE; i++)
177 		p2m[i] = IDENTITY_FRAME(pfn + i);
178 }
179 
180 static void * __ref alloc_p2m_page(void)
181 {
182 	if (unlikely(!slab_is_available()))
183 		return alloc_bootmem_align(PAGE_SIZE, PAGE_SIZE);
184 
185 	return (void *)__get_free_page(GFP_KERNEL);
186 }
187 
188 static void __ref free_p2m_page(void *p)
189 {
190 	if (unlikely(!slab_is_available())) {
191 		free_bootmem((unsigned long)p, PAGE_SIZE);
192 		return;
193 	}
194 
195 	free_page((unsigned long)p);
196 }
197 
198 /*
199  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
200  *
201  * This is called both at boot time, and after resuming from suspend:
202  * - At boot time we're called rather early, and must use alloc_bootmem*()
203  *   to allocate memory.
204  *
205  * - After resume we're called from within stop_machine, but the mfn
206  *   tree should already be completely allocated.
207  */
208 void __ref xen_build_mfn_list_list(void)
209 {
210 	unsigned long pfn, mfn;
211 	pte_t *ptep;
212 	unsigned int level, topidx, mididx;
213 	unsigned long *mid_mfn_p;
214 
215 	if (xen_feature(XENFEAT_auto_translated_physmap) ||
216 	    xen_start_info->flags & SIF_VIRT_P2M_4TOOLS)
217 		return;
218 
219 	/* Pre-initialize p2m_top_mfn to be completely missing */
220 	if (p2m_top_mfn == NULL) {
221 		p2m_mid_missing_mfn = alloc_p2m_page();
222 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
223 
224 		p2m_top_mfn_p = alloc_p2m_page();
225 		p2m_top_mfn_p_init(p2m_top_mfn_p);
226 
227 		p2m_top_mfn = alloc_p2m_page();
228 		p2m_top_mfn_init(p2m_top_mfn);
229 	} else {
230 		/* Reinitialise, mfn's all change after migration */
231 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
232 	}
233 
234 	for (pfn = 0; pfn < xen_max_p2m_pfn && pfn < MAX_P2M_PFN;
235 	     pfn += P2M_PER_PAGE) {
236 		topidx = p2m_top_index(pfn);
237 		mididx = p2m_mid_index(pfn);
238 
239 		mid_mfn_p = p2m_top_mfn_p[topidx];
240 		ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn),
241 				      &level);
242 		BUG_ON(!ptep || level != PG_LEVEL_4K);
243 		mfn = pte_mfn(*ptep);
244 		ptep = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
245 
246 		/* Don't bother allocating any mfn mid levels if
247 		 * they're just missing, just update the stored mfn,
248 		 * since all could have changed over a migrate.
249 		 */
250 		if (ptep == p2m_missing_pte || ptep == p2m_identity_pte) {
251 			BUG_ON(mididx);
252 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
253 			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
254 			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
255 			continue;
256 		}
257 
258 		if (mid_mfn_p == p2m_mid_missing_mfn) {
259 			mid_mfn_p = alloc_p2m_page();
260 			p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
261 
262 			p2m_top_mfn_p[topidx] = mid_mfn_p;
263 		}
264 
265 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
266 		mid_mfn_p[mididx] = mfn;
267 	}
268 }
269 
270 void xen_setup_mfn_list_list(void)
271 {
272 	if (xen_feature(XENFEAT_auto_translated_physmap))
273 		return;
274 
275 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
276 
277 	if (xen_start_info->flags & SIF_VIRT_P2M_4TOOLS)
278 		HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = ~0UL;
279 	else
280 		HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
281 			virt_to_mfn(p2m_top_mfn);
282 	HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
283 	HYPERVISOR_shared_info->arch.p2m_generation = 0;
284 	HYPERVISOR_shared_info->arch.p2m_vaddr = (unsigned long)xen_p2m_addr;
285 	HYPERVISOR_shared_info->arch.p2m_cr3 =
286 		xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
287 }
288 
289 /* Set up p2m_top to point to the domain-builder provided p2m pages */
290 void __init xen_build_dynamic_phys_to_machine(void)
291 {
292 	unsigned long pfn;
293 
294 	 if (xen_feature(XENFEAT_auto_translated_physmap))
295 		return;
296 
297 	xen_p2m_addr = (unsigned long *)xen_start_info->mfn_list;
298 	xen_p2m_size = ALIGN(xen_start_info->nr_pages, P2M_PER_PAGE);
299 
300 	for (pfn = xen_start_info->nr_pages; pfn < xen_p2m_size; pfn++)
301 		xen_p2m_addr[pfn] = INVALID_P2M_ENTRY;
302 
303 	xen_max_p2m_pfn = xen_p2m_size;
304 }
305 
306 #define P2M_TYPE_IDENTITY	0
307 #define P2M_TYPE_MISSING	1
308 #define P2M_TYPE_PFN		2
309 #define P2M_TYPE_UNKNOWN	3
310 
311 static int xen_p2m_elem_type(unsigned long pfn)
312 {
313 	unsigned long mfn;
314 
315 	if (pfn >= xen_p2m_size)
316 		return P2M_TYPE_IDENTITY;
317 
318 	mfn = xen_p2m_addr[pfn];
319 
320 	if (mfn == INVALID_P2M_ENTRY)
321 		return P2M_TYPE_MISSING;
322 
323 	if (mfn & IDENTITY_FRAME_BIT)
324 		return P2M_TYPE_IDENTITY;
325 
326 	return P2M_TYPE_PFN;
327 }
328 
329 static void __init xen_rebuild_p2m_list(unsigned long *p2m)
330 {
331 	unsigned int i, chunk;
332 	unsigned long pfn;
333 	unsigned long *mfns;
334 	pte_t *ptep;
335 	pmd_t *pmdp;
336 	int type;
337 
338 	p2m_missing = alloc_p2m_page();
339 	p2m_init(p2m_missing);
340 	p2m_identity = alloc_p2m_page();
341 	p2m_init(p2m_identity);
342 
343 	p2m_missing_pte = alloc_p2m_page();
344 	paravirt_alloc_pte(&init_mm, __pa(p2m_missing_pte) >> PAGE_SHIFT);
345 	p2m_identity_pte = alloc_p2m_page();
346 	paravirt_alloc_pte(&init_mm, __pa(p2m_identity_pte) >> PAGE_SHIFT);
347 	for (i = 0; i < PTRS_PER_PTE; i++) {
348 		set_pte(p2m_missing_pte + i,
349 			pfn_pte(PFN_DOWN(__pa(p2m_missing)), PAGE_KERNEL_RO));
350 		set_pte(p2m_identity_pte + i,
351 			pfn_pte(PFN_DOWN(__pa(p2m_identity)), PAGE_KERNEL_RO));
352 	}
353 
354 	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += chunk) {
355 		/*
356 		 * Try to map missing/identity PMDs or p2m-pages if possible.
357 		 * We have to respect the structure of the mfn_list_list
358 		 * which will be built just afterwards.
359 		 * Chunk size to test is one p2m page if we are in the middle
360 		 * of a mfn_list_list mid page and the complete mid page area
361 		 * if we are at index 0 of the mid page. Please note that a
362 		 * mid page might cover more than one PMD, e.g. on 32 bit PAE
363 		 * kernels.
364 		 */
365 		chunk = (pfn & (P2M_PER_PAGE * P2M_MID_PER_PAGE - 1)) ?
366 			P2M_PER_PAGE : P2M_PER_PAGE * P2M_MID_PER_PAGE;
367 
368 		type = xen_p2m_elem_type(pfn);
369 		i = 0;
370 		if (type != P2M_TYPE_PFN)
371 			for (i = 1; i < chunk; i++)
372 				if (xen_p2m_elem_type(pfn + i) != type)
373 					break;
374 		if (i < chunk)
375 			/* Reset to minimal chunk size. */
376 			chunk = P2M_PER_PAGE;
377 
378 		if (type == P2M_TYPE_PFN || i < chunk) {
379 			/* Use initial p2m page contents. */
380 #ifdef CONFIG_X86_64
381 			mfns = alloc_p2m_page();
382 			copy_page(mfns, xen_p2m_addr + pfn);
383 #else
384 			mfns = xen_p2m_addr + pfn;
385 #endif
386 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
387 			set_pte(ptep,
388 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
389 			continue;
390 		}
391 
392 		if (chunk == P2M_PER_PAGE) {
393 			/* Map complete missing or identity p2m-page. */
394 			mfns = (type == P2M_TYPE_MISSING) ?
395 				p2m_missing : p2m_identity;
396 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
397 			set_pte(ptep,
398 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
399 			continue;
400 		}
401 
402 		/* Complete missing or identity PMD(s) can be mapped. */
403 		ptep = (type == P2M_TYPE_MISSING) ?
404 			p2m_missing_pte : p2m_identity_pte;
405 		for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
406 			pmdp = populate_extra_pmd(
407 				(unsigned long)(p2m + pfn) + i * PMD_SIZE);
408 			set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
409 		}
410 	}
411 }
412 
413 void __init xen_vmalloc_p2m_tree(void)
414 {
415 	static struct vm_struct vm;
416 	unsigned long p2m_limit;
417 
418 	xen_p2m_last_pfn = xen_max_p2m_pfn;
419 
420 	p2m_limit = (phys_addr_t)P2M_LIMIT * 1024 * 1024 * 1024 / PAGE_SIZE;
421 	vm.flags = VM_ALLOC;
422 	vm.size = ALIGN(sizeof(unsigned long) * max(xen_max_p2m_pfn, p2m_limit),
423 			PMD_SIZE * PMDS_PER_MID_PAGE);
424 	vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
425 	pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
426 
427 	xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
428 
429 	xen_rebuild_p2m_list(vm.addr);
430 
431 	xen_p2m_addr = vm.addr;
432 	xen_p2m_size = xen_max_p2m_pfn;
433 
434 	xen_inv_extra_mem();
435 }
436 
437 unsigned long get_phys_to_machine(unsigned long pfn)
438 {
439 	pte_t *ptep;
440 	unsigned int level;
441 
442 	if (unlikely(pfn >= xen_p2m_size)) {
443 		if (pfn < xen_max_p2m_pfn)
444 			return xen_chk_extra_mem(pfn);
445 
446 		return IDENTITY_FRAME(pfn);
447 	}
448 
449 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
450 	BUG_ON(!ptep || level != PG_LEVEL_4K);
451 
452 	/*
453 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
454 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
455 	 * would be wrong.
456 	 */
457 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
458 		return IDENTITY_FRAME(pfn);
459 
460 	return xen_p2m_addr[pfn];
461 }
462 EXPORT_SYMBOL_GPL(get_phys_to_machine);
463 
464 /*
465  * Allocate new pmd(s). It is checked whether the old pmd is still in place.
466  * If not, nothing is changed. This is okay as the only reason for allocating
467  * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
468  * pmd. In case of PAE/x86-32 there are multiple pmds to allocate!
469  */
470 static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
471 {
472 	pte_t *ptechk;
473 	pte_t *pte_newpg[PMDS_PER_MID_PAGE];
474 	pmd_t *pmdp;
475 	unsigned int level;
476 	unsigned long flags;
477 	unsigned long vaddr;
478 	int i;
479 
480 	/* Do all allocations first to bail out in error case. */
481 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
482 		pte_newpg[i] = alloc_p2m_page();
483 		if (!pte_newpg[i]) {
484 			for (i--; i >= 0; i--)
485 				free_p2m_page(pte_newpg[i]);
486 
487 			return NULL;
488 		}
489 	}
490 
491 	vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
492 
493 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
494 		copy_page(pte_newpg[i], pte_pg);
495 		paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
496 
497 		pmdp = lookup_pmd_address(vaddr);
498 		BUG_ON(!pmdp);
499 
500 		spin_lock_irqsave(&p2m_update_lock, flags);
501 
502 		ptechk = lookup_address(vaddr, &level);
503 		if (ptechk == pte_pg) {
504 			HYPERVISOR_shared_info->arch.p2m_generation++;
505 			wmb(); /* Tools are synchronizing via p2m_generation. */
506 			set_pmd(pmdp,
507 				__pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
508 			wmb(); /* Tools are synchronizing via p2m_generation. */
509 			HYPERVISOR_shared_info->arch.p2m_generation++;
510 			pte_newpg[i] = NULL;
511 		}
512 
513 		spin_unlock_irqrestore(&p2m_update_lock, flags);
514 
515 		if (pte_newpg[i]) {
516 			paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
517 			free_p2m_page(pte_newpg[i]);
518 		}
519 
520 		vaddr += PMD_SIZE;
521 	}
522 
523 	return lookup_address(addr, &level);
524 }
525 
526 /*
527  * Fully allocate the p2m structure for a given pfn.  We need to check
528  * that both the top and mid levels are allocated, and make sure the
529  * parallel mfn tree is kept in sync.  We may race with other cpus, so
530  * the new pages are installed with cmpxchg; if we lose the race then
531  * simply free the page we allocated and use the one that's there.
532  */
533 int xen_alloc_p2m_entry(unsigned long pfn)
534 {
535 	unsigned topidx;
536 	unsigned long *top_mfn_p, *mid_mfn;
537 	pte_t *ptep, *pte_pg;
538 	unsigned int level;
539 	unsigned long flags;
540 	unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
541 	unsigned long p2m_pfn;
542 
543 	if (xen_feature(XENFEAT_auto_translated_physmap))
544 		return 0;
545 
546 	ptep = lookup_address(addr, &level);
547 	BUG_ON(!ptep || level != PG_LEVEL_4K);
548 	pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
549 
550 	if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
551 		/* PMD level is missing, allocate a new one */
552 		ptep = alloc_p2m_pmd(addr, pte_pg);
553 		if (!ptep)
554 			return -ENOMEM;
555 	}
556 
557 	if (p2m_top_mfn && pfn < MAX_P2M_PFN) {
558 		topidx = p2m_top_index(pfn);
559 		top_mfn_p = &p2m_top_mfn[topidx];
560 		mid_mfn = ACCESS_ONCE(p2m_top_mfn_p[topidx]);
561 
562 		BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
563 
564 		if (mid_mfn == p2m_mid_missing_mfn) {
565 			/* Separately check the mid mfn level */
566 			unsigned long missing_mfn;
567 			unsigned long mid_mfn_mfn;
568 			unsigned long old_mfn;
569 
570 			mid_mfn = alloc_p2m_page();
571 			if (!mid_mfn)
572 				return -ENOMEM;
573 
574 			p2m_mid_mfn_init(mid_mfn, p2m_missing);
575 
576 			missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
577 			mid_mfn_mfn = virt_to_mfn(mid_mfn);
578 			old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
579 			if (old_mfn != missing_mfn) {
580 				free_p2m_page(mid_mfn);
581 				mid_mfn = mfn_to_virt(old_mfn);
582 			} else {
583 				p2m_top_mfn_p[topidx] = mid_mfn;
584 			}
585 		}
586 	} else {
587 		mid_mfn = NULL;
588 	}
589 
590 	p2m_pfn = pte_pfn(READ_ONCE(*ptep));
591 	if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
592 	    p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
593 		/* p2m leaf page is missing */
594 		unsigned long *p2m;
595 
596 		p2m = alloc_p2m_page();
597 		if (!p2m)
598 			return -ENOMEM;
599 
600 		if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
601 			p2m_init(p2m);
602 		else
603 			p2m_init_identity(p2m, pfn & ~(P2M_PER_PAGE - 1));
604 
605 		spin_lock_irqsave(&p2m_update_lock, flags);
606 
607 		if (pte_pfn(*ptep) == p2m_pfn) {
608 			HYPERVISOR_shared_info->arch.p2m_generation++;
609 			wmb(); /* Tools are synchronizing via p2m_generation. */
610 			set_pte(ptep,
611 				pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
612 			wmb(); /* Tools are synchronizing via p2m_generation. */
613 			HYPERVISOR_shared_info->arch.p2m_generation++;
614 			if (mid_mfn)
615 				mid_mfn[p2m_mid_index(pfn)] = virt_to_mfn(p2m);
616 			p2m = NULL;
617 		}
618 
619 		spin_unlock_irqrestore(&p2m_update_lock, flags);
620 
621 		if (p2m)
622 			free_p2m_page(p2m);
623 	}
624 
625 	/* Expanded the p2m? */
626 	if (pfn > xen_p2m_last_pfn) {
627 		xen_p2m_last_pfn = pfn;
628 		HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
629 	}
630 
631 	return 0;
632 }
633 EXPORT_SYMBOL(xen_alloc_p2m_entry);
634 
635 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
636 				      unsigned long pfn_e)
637 {
638 	unsigned long pfn;
639 
640 	if (unlikely(pfn_s >= xen_p2m_size))
641 		return 0;
642 
643 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
644 		return pfn_e - pfn_s;
645 
646 	if (pfn_s > pfn_e)
647 		return 0;
648 
649 	if (pfn_e > xen_p2m_size)
650 		pfn_e = xen_p2m_size;
651 
652 	for (pfn = pfn_s; pfn < pfn_e; pfn++)
653 		xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
654 
655 	return pfn - pfn_s;
656 }
657 
658 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
659 {
660 	pte_t *ptep;
661 	unsigned int level;
662 
663 	/* don't track P2M changes in autotranslate guests */
664 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
665 		return true;
666 
667 	if (unlikely(pfn >= xen_p2m_size)) {
668 		BUG_ON(mfn != INVALID_P2M_ENTRY);
669 		return true;
670 	}
671 
672 	/*
673 	 * The interface requires atomic updates on p2m elements.
674 	 * xen_safe_write_ulong() is using __put_user which does an atomic
675 	 * store via asm().
676 	 */
677 	if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
678 		return true;
679 
680 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
681 	BUG_ON(!ptep || level != PG_LEVEL_4K);
682 
683 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
684 		return mfn == INVALID_P2M_ENTRY;
685 
686 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
687 		return mfn == IDENTITY_FRAME(pfn);
688 
689 	return false;
690 }
691 
692 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
693 {
694 	if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
695 		int ret;
696 
697 		ret = xen_alloc_p2m_entry(pfn);
698 		if (ret < 0)
699 			return false;
700 
701 		return __set_phys_to_machine(pfn, mfn);
702 	}
703 
704 	return true;
705 }
706 
707 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
708 			    struct gnttab_map_grant_ref *kmap_ops,
709 			    struct page **pages, unsigned int count)
710 {
711 	int i, ret = 0;
712 	pte_t *pte;
713 
714 	if (xen_feature(XENFEAT_auto_translated_physmap))
715 		return 0;
716 
717 	if (kmap_ops) {
718 		ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
719 						kmap_ops, count);
720 		if (ret)
721 			goto out;
722 	}
723 
724 	for (i = 0; i < count; i++) {
725 		unsigned long mfn, pfn;
726 
727 		/* Do not add to override if the map failed. */
728 		if (map_ops[i].status)
729 			continue;
730 
731 		if (map_ops[i].flags & GNTMAP_contains_pte) {
732 			pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
733 				(map_ops[i].host_addr & ~PAGE_MASK));
734 			mfn = pte_mfn(*pte);
735 		} else {
736 			mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
737 		}
738 		pfn = page_to_pfn(pages[i]);
739 
740 		WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
741 
742 		if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
743 			ret = -ENOMEM;
744 			goto out;
745 		}
746 	}
747 
748 out:
749 	return ret;
750 }
751 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
752 
753 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
754 			      struct gnttab_unmap_grant_ref *kunmap_ops,
755 			      struct page **pages, unsigned int count)
756 {
757 	int i, ret = 0;
758 
759 	if (xen_feature(XENFEAT_auto_translated_physmap))
760 		return 0;
761 
762 	for (i = 0; i < count; i++) {
763 		unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
764 		unsigned long pfn = page_to_pfn(pages[i]);
765 
766 		if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
767 			ret = -EINVAL;
768 			goto out;
769 		}
770 
771 		set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
772 	}
773 	if (kunmap_ops)
774 		ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
775 						kunmap_ops, count);
776 out:
777 	return ret;
778 }
779 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
780 
781 #ifdef CONFIG_XEN_DEBUG_FS
782 #include <linux/debugfs.h>
783 #include "debugfs.h"
784 static int p2m_dump_show(struct seq_file *m, void *v)
785 {
786 	static const char * const type_name[] = {
787 				[P2M_TYPE_IDENTITY] = "identity",
788 				[P2M_TYPE_MISSING] = "missing",
789 				[P2M_TYPE_PFN] = "pfn",
790 				[P2M_TYPE_UNKNOWN] = "abnormal"};
791 	unsigned long pfn, first_pfn;
792 	int type, prev_type;
793 
794 	prev_type = xen_p2m_elem_type(0);
795 	first_pfn = 0;
796 
797 	for (pfn = 0; pfn < xen_p2m_size; pfn++) {
798 		type = xen_p2m_elem_type(pfn);
799 		if (type != prev_type) {
800 			seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
801 				   type_name[prev_type]);
802 			prev_type = type;
803 			first_pfn = pfn;
804 		}
805 	}
806 	seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
807 		   type_name[prev_type]);
808 	return 0;
809 }
810 
811 static int p2m_dump_open(struct inode *inode, struct file *filp)
812 {
813 	return single_open(filp, p2m_dump_show, NULL);
814 }
815 
816 static const struct file_operations p2m_dump_fops = {
817 	.open		= p2m_dump_open,
818 	.read		= seq_read,
819 	.llseek		= seq_lseek,
820 	.release	= single_release,
821 };
822 
823 static struct dentry *d_mmu_debug;
824 
825 static int __init xen_p2m_debugfs(void)
826 {
827 	struct dentry *d_xen = xen_init_debugfs();
828 
829 	if (d_xen == NULL)
830 		return -ENOMEM;
831 
832 	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
833 
834 	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
835 	return 0;
836 }
837 fs_initcall(xen_p2m_debugfs);
838 #endif /* CONFIG_XEN_DEBUG_FS */
839