xref: /openbmc/linux/arch/x86/xen/p2m.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Xen leaves the responsibility for maintaining p2m mappings to the
5  * guests themselves, but it must also access and update the p2m array
6  * during suspend/resume when all the pages are reallocated.
7  *
8  * The logical flat p2m table is mapped to a linear kernel memory area.
9  * For accesses by Xen a three-level tree linked via mfns only is set up to
10  * allow the address space to be sparse.
11  *
12  *               Xen
13  *                |
14  *          p2m_top_mfn
15  *              /   \
16  * p2m_mid_mfn p2m_mid_mfn
17  *         /           /
18  *  p2m p2m p2m ...
19  *
20  * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
21  *
22  * The p2m_top_mfn level is limited to 1 page, so the maximum representable
23  * pseudo-physical address space is:
24  *  P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
25  *
26  * P2M_PER_PAGE depends on the architecture, as a mfn is always
27  * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
28  * 512 and 1024 entries respectively.
29  *
30  * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
31  *
32  * However not all entries are filled with MFNs. Specifically for all other
33  * leaf entries, or for the top  root, or middle one, for which there is a void
34  * entry, we assume it is  "missing". So (for example)
35  *  pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
36  * We have a dedicated page p2m_missing with all entries being
37  * INVALID_P2M_ENTRY. This page may be referenced multiple times in the p2m
38  * list/tree in case there are multiple areas with P2M_PER_PAGE invalid pfns.
39  *
40  * We also have the possibility of setting 1-1 mappings on certain regions, so
41  * that:
42  *  pfn_to_mfn(0xc0000)=0xc0000
43  *
44  * The benefit of this is, that we can assume for non-RAM regions (think
45  * PCI BARs, or ACPI spaces), we can create mappings easily because we
46  * get the PFN value to match the MFN.
47  *
48  * For this to work efficiently we have one new page p2m_identity. All entries
49  * in p2m_identity are set to INVALID_P2M_ENTRY type (Xen toolstack only
50  * recognizes that and MFNs, no other fancy value).
51  *
52  * On lookup we spot that the entry points to p2m_identity and return the
53  * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
54  * If the entry points to an allocated page, we just proceed as before and
55  * return the PFN. If the PFN has IDENTITY_FRAME_BIT set we unmask that in
56  * appropriate functions (pfn_to_mfn).
57  *
58  * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
59  * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
60  * non-identity pfn. To protect ourselves against we elect to set (and get) the
61  * IDENTITY_FRAME_BIT on all identity mapped PFNs.
62  */
63 
64 #include <linux/init.h>
65 #include <linux/export.h>
66 #include <linux/list.h>
67 #include <linux/hash.h>
68 #include <linux/sched.h>
69 #include <linux/seq_file.h>
70 #include <linux/memblock.h>
71 #include <linux/slab.h>
72 #include <linux/vmalloc.h>
73 
74 #include <asm/cache.h>
75 #include <asm/setup.h>
76 #include <linux/uaccess.h>
77 
78 #include <asm/xen/page.h>
79 #include <asm/xen/hypercall.h>
80 #include <asm/xen/hypervisor.h>
81 #include <xen/balloon.h>
82 #include <xen/grant_table.h>
83 
84 #include "multicalls.h"
85 #include "xen-ops.h"
86 
87 #define P2M_MID_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long *))
88 #define P2M_TOP_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long **))
89 
90 #define MAX_P2M_PFN	(P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
91 
92 #define PMDS_PER_MID_PAGE	(P2M_MID_PER_PAGE / PTRS_PER_PTE)
93 
94 unsigned long *xen_p2m_addr __read_mostly;
95 EXPORT_SYMBOL_GPL(xen_p2m_addr);
96 unsigned long xen_p2m_size __read_mostly;
97 EXPORT_SYMBOL_GPL(xen_p2m_size);
98 unsigned long xen_max_p2m_pfn __read_mostly;
99 EXPORT_SYMBOL_GPL(xen_max_p2m_pfn);
100 
101 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
102 #define P2M_LIMIT CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
103 #else
104 #define P2M_LIMIT 0
105 #endif
106 
107 static DEFINE_SPINLOCK(p2m_update_lock);
108 
109 static unsigned long *p2m_mid_missing_mfn;
110 static unsigned long *p2m_top_mfn;
111 static unsigned long **p2m_top_mfn_p;
112 static unsigned long *p2m_missing;
113 static unsigned long *p2m_identity;
114 static pte_t *p2m_missing_pte;
115 static pte_t *p2m_identity_pte;
116 
117 /*
118  * Hint at last populated PFN.
119  *
120  * Used to set HYPERVISOR_shared_info->arch.max_pfn so the toolstack
121  * can avoid scanning the whole P2M (which may be sized to account for
122  * hotplugged memory).
123  */
124 static unsigned long xen_p2m_last_pfn;
125 
126 static inline unsigned p2m_top_index(unsigned long pfn)
127 {
128 	BUG_ON(pfn >= MAX_P2M_PFN);
129 	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
130 }
131 
132 static inline unsigned p2m_mid_index(unsigned long pfn)
133 {
134 	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
135 }
136 
137 static inline unsigned p2m_index(unsigned long pfn)
138 {
139 	return pfn % P2M_PER_PAGE;
140 }
141 
142 static void p2m_top_mfn_init(unsigned long *top)
143 {
144 	unsigned i;
145 
146 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
147 		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
148 }
149 
150 static void p2m_top_mfn_p_init(unsigned long **top)
151 {
152 	unsigned i;
153 
154 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
155 		top[i] = p2m_mid_missing_mfn;
156 }
157 
158 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
159 {
160 	unsigned i;
161 
162 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
163 		mid[i] = virt_to_mfn(leaf);
164 }
165 
166 static void p2m_init(unsigned long *p2m)
167 {
168 	unsigned i;
169 
170 	for (i = 0; i < P2M_PER_PAGE; i++)
171 		p2m[i] = INVALID_P2M_ENTRY;
172 }
173 
174 static void p2m_init_identity(unsigned long *p2m, unsigned long pfn)
175 {
176 	unsigned i;
177 
178 	for (i = 0; i < P2M_PER_PAGE; i++)
179 		p2m[i] = IDENTITY_FRAME(pfn + i);
180 }
181 
182 static void * __ref alloc_p2m_page(void)
183 {
184 	if (unlikely(!slab_is_available())) {
185 		void *ptr = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
186 
187 		if (!ptr)
188 			panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
189 			      __func__, PAGE_SIZE, PAGE_SIZE);
190 
191 		return ptr;
192 	}
193 
194 	return (void *)__get_free_page(GFP_KERNEL);
195 }
196 
197 static void __ref free_p2m_page(void *p)
198 {
199 	if (unlikely(!slab_is_available())) {
200 		memblock_free((unsigned long)p, PAGE_SIZE);
201 		return;
202 	}
203 
204 	free_page((unsigned long)p);
205 }
206 
207 /*
208  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
209  *
210  * This is called both at boot time, and after resuming from suspend:
211  * - At boot time we're called rather early, and must use alloc_bootmem*()
212  *   to allocate memory.
213  *
214  * - After resume we're called from within stop_machine, but the mfn
215  *   tree should already be completely allocated.
216  */
217 void __ref xen_build_mfn_list_list(void)
218 {
219 	unsigned long pfn, mfn;
220 	pte_t *ptep;
221 	unsigned int level, topidx, mididx;
222 	unsigned long *mid_mfn_p;
223 
224 	if (xen_start_info->flags & SIF_VIRT_P2M_4TOOLS)
225 		return;
226 
227 	/* Pre-initialize p2m_top_mfn to be completely missing */
228 	if (p2m_top_mfn == NULL) {
229 		p2m_mid_missing_mfn = alloc_p2m_page();
230 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
231 
232 		p2m_top_mfn_p = alloc_p2m_page();
233 		p2m_top_mfn_p_init(p2m_top_mfn_p);
234 
235 		p2m_top_mfn = alloc_p2m_page();
236 		p2m_top_mfn_init(p2m_top_mfn);
237 	} else {
238 		/* Reinitialise, mfn's all change after migration */
239 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
240 	}
241 
242 	for (pfn = 0; pfn < xen_max_p2m_pfn && pfn < MAX_P2M_PFN;
243 	     pfn += P2M_PER_PAGE) {
244 		topidx = p2m_top_index(pfn);
245 		mididx = p2m_mid_index(pfn);
246 
247 		mid_mfn_p = p2m_top_mfn_p[topidx];
248 		ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn),
249 				      &level);
250 		BUG_ON(!ptep || level != PG_LEVEL_4K);
251 		mfn = pte_mfn(*ptep);
252 		ptep = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
253 
254 		/* Don't bother allocating any mfn mid levels if
255 		 * they're just missing, just update the stored mfn,
256 		 * since all could have changed over a migrate.
257 		 */
258 		if (ptep == p2m_missing_pte || ptep == p2m_identity_pte) {
259 			BUG_ON(mididx);
260 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
261 			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
262 			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
263 			continue;
264 		}
265 
266 		if (mid_mfn_p == p2m_mid_missing_mfn) {
267 			mid_mfn_p = alloc_p2m_page();
268 			p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
269 
270 			p2m_top_mfn_p[topidx] = mid_mfn_p;
271 		}
272 
273 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
274 		mid_mfn_p[mididx] = mfn;
275 	}
276 }
277 
278 void xen_setup_mfn_list_list(void)
279 {
280 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
281 
282 	if (xen_start_info->flags & SIF_VIRT_P2M_4TOOLS)
283 		HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = ~0UL;
284 	else
285 		HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
286 			virt_to_mfn(p2m_top_mfn);
287 	HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
288 	HYPERVISOR_shared_info->arch.p2m_generation = 0;
289 	HYPERVISOR_shared_info->arch.p2m_vaddr = (unsigned long)xen_p2m_addr;
290 	HYPERVISOR_shared_info->arch.p2m_cr3 =
291 		xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
292 }
293 
294 /* Set up p2m_top to point to the domain-builder provided p2m pages */
295 void __init xen_build_dynamic_phys_to_machine(void)
296 {
297 	unsigned long pfn;
298 
299 	xen_p2m_addr = (unsigned long *)xen_start_info->mfn_list;
300 	xen_p2m_size = ALIGN(xen_start_info->nr_pages, P2M_PER_PAGE);
301 
302 	for (pfn = xen_start_info->nr_pages; pfn < xen_p2m_size; pfn++)
303 		xen_p2m_addr[pfn] = INVALID_P2M_ENTRY;
304 
305 	xen_max_p2m_pfn = xen_p2m_size;
306 }
307 
308 #define P2M_TYPE_IDENTITY	0
309 #define P2M_TYPE_MISSING	1
310 #define P2M_TYPE_PFN		2
311 #define P2M_TYPE_UNKNOWN	3
312 
313 static int xen_p2m_elem_type(unsigned long pfn)
314 {
315 	unsigned long mfn;
316 
317 	if (pfn >= xen_p2m_size)
318 		return P2M_TYPE_IDENTITY;
319 
320 	mfn = xen_p2m_addr[pfn];
321 
322 	if (mfn == INVALID_P2M_ENTRY)
323 		return P2M_TYPE_MISSING;
324 
325 	if (mfn & IDENTITY_FRAME_BIT)
326 		return P2M_TYPE_IDENTITY;
327 
328 	return P2M_TYPE_PFN;
329 }
330 
331 static void __init xen_rebuild_p2m_list(unsigned long *p2m)
332 {
333 	unsigned int i, chunk;
334 	unsigned long pfn;
335 	unsigned long *mfns;
336 	pte_t *ptep;
337 	pmd_t *pmdp;
338 	int type;
339 
340 	p2m_missing = alloc_p2m_page();
341 	p2m_init(p2m_missing);
342 	p2m_identity = alloc_p2m_page();
343 	p2m_init(p2m_identity);
344 
345 	p2m_missing_pte = alloc_p2m_page();
346 	paravirt_alloc_pte(&init_mm, __pa(p2m_missing_pte) >> PAGE_SHIFT);
347 	p2m_identity_pte = alloc_p2m_page();
348 	paravirt_alloc_pte(&init_mm, __pa(p2m_identity_pte) >> PAGE_SHIFT);
349 	for (i = 0; i < PTRS_PER_PTE; i++) {
350 		set_pte(p2m_missing_pte + i,
351 			pfn_pte(PFN_DOWN(__pa(p2m_missing)), PAGE_KERNEL_RO));
352 		set_pte(p2m_identity_pte + i,
353 			pfn_pte(PFN_DOWN(__pa(p2m_identity)), PAGE_KERNEL_RO));
354 	}
355 
356 	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += chunk) {
357 		/*
358 		 * Try to map missing/identity PMDs or p2m-pages if possible.
359 		 * We have to respect the structure of the mfn_list_list
360 		 * which will be built just afterwards.
361 		 * Chunk size to test is one p2m page if we are in the middle
362 		 * of a mfn_list_list mid page and the complete mid page area
363 		 * if we are at index 0 of the mid page. Please note that a
364 		 * mid page might cover more than one PMD, e.g. on 32 bit PAE
365 		 * kernels.
366 		 */
367 		chunk = (pfn & (P2M_PER_PAGE * P2M_MID_PER_PAGE - 1)) ?
368 			P2M_PER_PAGE : P2M_PER_PAGE * P2M_MID_PER_PAGE;
369 
370 		type = xen_p2m_elem_type(pfn);
371 		i = 0;
372 		if (type != P2M_TYPE_PFN)
373 			for (i = 1; i < chunk; i++)
374 				if (xen_p2m_elem_type(pfn + i) != type)
375 					break;
376 		if (i < chunk)
377 			/* Reset to minimal chunk size. */
378 			chunk = P2M_PER_PAGE;
379 
380 		if (type == P2M_TYPE_PFN || i < chunk) {
381 			/* Use initial p2m page contents. */
382 #ifdef CONFIG_X86_64
383 			mfns = alloc_p2m_page();
384 			copy_page(mfns, xen_p2m_addr + pfn);
385 #else
386 			mfns = xen_p2m_addr + pfn;
387 #endif
388 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
389 			set_pte(ptep,
390 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
391 			continue;
392 		}
393 
394 		if (chunk == P2M_PER_PAGE) {
395 			/* Map complete missing or identity p2m-page. */
396 			mfns = (type == P2M_TYPE_MISSING) ?
397 				p2m_missing : p2m_identity;
398 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
399 			set_pte(ptep,
400 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
401 			continue;
402 		}
403 
404 		/* Complete missing or identity PMD(s) can be mapped. */
405 		ptep = (type == P2M_TYPE_MISSING) ?
406 			p2m_missing_pte : p2m_identity_pte;
407 		for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
408 			pmdp = populate_extra_pmd(
409 				(unsigned long)(p2m + pfn) + i * PMD_SIZE);
410 			set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
411 		}
412 	}
413 }
414 
415 void __init xen_vmalloc_p2m_tree(void)
416 {
417 	static struct vm_struct vm;
418 	unsigned long p2m_limit;
419 
420 	xen_p2m_last_pfn = xen_max_p2m_pfn;
421 
422 	p2m_limit = (phys_addr_t)P2M_LIMIT * 1024 * 1024 * 1024 / PAGE_SIZE;
423 	vm.flags = VM_ALLOC;
424 	vm.size = ALIGN(sizeof(unsigned long) * max(xen_max_p2m_pfn, p2m_limit),
425 			PMD_SIZE * PMDS_PER_MID_PAGE);
426 	vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
427 	pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
428 
429 	xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
430 
431 	xen_rebuild_p2m_list(vm.addr);
432 
433 	xen_p2m_addr = vm.addr;
434 	xen_p2m_size = xen_max_p2m_pfn;
435 
436 	xen_inv_extra_mem();
437 }
438 
439 unsigned long get_phys_to_machine(unsigned long pfn)
440 {
441 	pte_t *ptep;
442 	unsigned int level;
443 
444 	if (unlikely(pfn >= xen_p2m_size)) {
445 		if (pfn < xen_max_p2m_pfn)
446 			return xen_chk_extra_mem(pfn);
447 
448 		return IDENTITY_FRAME(pfn);
449 	}
450 
451 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
452 	BUG_ON(!ptep || level != PG_LEVEL_4K);
453 
454 	/*
455 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
456 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
457 	 * would be wrong.
458 	 */
459 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
460 		return IDENTITY_FRAME(pfn);
461 
462 	return xen_p2m_addr[pfn];
463 }
464 EXPORT_SYMBOL_GPL(get_phys_to_machine);
465 
466 /*
467  * Allocate new pmd(s). It is checked whether the old pmd is still in place.
468  * If not, nothing is changed. This is okay as the only reason for allocating
469  * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
470  * pmd. In case of PAE/x86-32 there are multiple pmds to allocate!
471  */
472 static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
473 {
474 	pte_t *ptechk;
475 	pte_t *pte_newpg[PMDS_PER_MID_PAGE];
476 	pmd_t *pmdp;
477 	unsigned int level;
478 	unsigned long flags;
479 	unsigned long vaddr;
480 	int i;
481 
482 	/* Do all allocations first to bail out in error case. */
483 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
484 		pte_newpg[i] = alloc_p2m_page();
485 		if (!pte_newpg[i]) {
486 			for (i--; i >= 0; i--)
487 				free_p2m_page(pte_newpg[i]);
488 
489 			return NULL;
490 		}
491 	}
492 
493 	vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
494 
495 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
496 		copy_page(pte_newpg[i], pte_pg);
497 		paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
498 
499 		pmdp = lookup_pmd_address(vaddr);
500 		BUG_ON(!pmdp);
501 
502 		spin_lock_irqsave(&p2m_update_lock, flags);
503 
504 		ptechk = lookup_address(vaddr, &level);
505 		if (ptechk == pte_pg) {
506 			HYPERVISOR_shared_info->arch.p2m_generation++;
507 			wmb(); /* Tools are synchronizing via p2m_generation. */
508 			set_pmd(pmdp,
509 				__pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
510 			wmb(); /* Tools are synchronizing via p2m_generation. */
511 			HYPERVISOR_shared_info->arch.p2m_generation++;
512 			pte_newpg[i] = NULL;
513 		}
514 
515 		spin_unlock_irqrestore(&p2m_update_lock, flags);
516 
517 		if (pte_newpg[i]) {
518 			paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
519 			free_p2m_page(pte_newpg[i]);
520 		}
521 
522 		vaddr += PMD_SIZE;
523 	}
524 
525 	return lookup_address(addr, &level);
526 }
527 
528 /*
529  * Fully allocate the p2m structure for a given pfn.  We need to check
530  * that both the top and mid levels are allocated, and make sure the
531  * parallel mfn tree is kept in sync.  We may race with other cpus, so
532  * the new pages are installed with cmpxchg; if we lose the race then
533  * simply free the page we allocated and use the one that's there.
534  */
535 int xen_alloc_p2m_entry(unsigned long pfn)
536 {
537 	unsigned topidx;
538 	unsigned long *top_mfn_p, *mid_mfn;
539 	pte_t *ptep, *pte_pg;
540 	unsigned int level;
541 	unsigned long flags;
542 	unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
543 	unsigned long p2m_pfn;
544 
545 	ptep = lookup_address(addr, &level);
546 	BUG_ON(!ptep || level != PG_LEVEL_4K);
547 	pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
548 
549 	if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
550 		/* PMD level is missing, allocate a new one */
551 		ptep = alloc_p2m_pmd(addr, pte_pg);
552 		if (!ptep)
553 			return -ENOMEM;
554 	}
555 
556 	if (p2m_top_mfn && pfn < MAX_P2M_PFN) {
557 		topidx = p2m_top_index(pfn);
558 		top_mfn_p = &p2m_top_mfn[topidx];
559 		mid_mfn = READ_ONCE(p2m_top_mfn_p[topidx]);
560 
561 		BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
562 
563 		if (mid_mfn == p2m_mid_missing_mfn) {
564 			/* Separately check the mid mfn level */
565 			unsigned long missing_mfn;
566 			unsigned long mid_mfn_mfn;
567 			unsigned long old_mfn;
568 
569 			mid_mfn = alloc_p2m_page();
570 			if (!mid_mfn)
571 				return -ENOMEM;
572 
573 			p2m_mid_mfn_init(mid_mfn, p2m_missing);
574 
575 			missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
576 			mid_mfn_mfn = virt_to_mfn(mid_mfn);
577 			old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
578 			if (old_mfn != missing_mfn) {
579 				free_p2m_page(mid_mfn);
580 				mid_mfn = mfn_to_virt(old_mfn);
581 			} else {
582 				p2m_top_mfn_p[topidx] = mid_mfn;
583 			}
584 		}
585 	} else {
586 		mid_mfn = NULL;
587 	}
588 
589 	p2m_pfn = pte_pfn(READ_ONCE(*ptep));
590 	if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
591 	    p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
592 		/* p2m leaf page is missing */
593 		unsigned long *p2m;
594 
595 		p2m = alloc_p2m_page();
596 		if (!p2m)
597 			return -ENOMEM;
598 
599 		if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
600 			p2m_init(p2m);
601 		else
602 			p2m_init_identity(p2m, pfn & ~(P2M_PER_PAGE - 1));
603 
604 		spin_lock_irqsave(&p2m_update_lock, flags);
605 
606 		if (pte_pfn(*ptep) == p2m_pfn) {
607 			HYPERVISOR_shared_info->arch.p2m_generation++;
608 			wmb(); /* Tools are synchronizing via p2m_generation. */
609 			set_pte(ptep,
610 				pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
611 			wmb(); /* Tools are synchronizing via p2m_generation. */
612 			HYPERVISOR_shared_info->arch.p2m_generation++;
613 			if (mid_mfn)
614 				mid_mfn[p2m_mid_index(pfn)] = virt_to_mfn(p2m);
615 			p2m = NULL;
616 		}
617 
618 		spin_unlock_irqrestore(&p2m_update_lock, flags);
619 
620 		if (p2m)
621 			free_p2m_page(p2m);
622 	}
623 
624 	/* Expanded the p2m? */
625 	if (pfn > xen_p2m_last_pfn) {
626 		xen_p2m_last_pfn = pfn;
627 		HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
628 	}
629 
630 	return 0;
631 }
632 EXPORT_SYMBOL(xen_alloc_p2m_entry);
633 
634 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
635 				      unsigned long pfn_e)
636 {
637 	unsigned long pfn;
638 
639 	if (unlikely(pfn_s >= xen_p2m_size))
640 		return 0;
641 
642 	if (pfn_s > pfn_e)
643 		return 0;
644 
645 	if (pfn_e > xen_p2m_size)
646 		pfn_e = xen_p2m_size;
647 
648 	for (pfn = pfn_s; pfn < pfn_e; pfn++)
649 		xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
650 
651 	return pfn - pfn_s;
652 }
653 
654 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
655 {
656 	pte_t *ptep;
657 	unsigned int level;
658 
659 	if (unlikely(pfn >= xen_p2m_size)) {
660 		BUG_ON(mfn != INVALID_P2M_ENTRY);
661 		return true;
662 	}
663 
664 	/*
665 	 * The interface requires atomic updates on p2m elements.
666 	 * xen_safe_write_ulong() is using an atomic store via asm().
667 	 */
668 	if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
669 		return true;
670 
671 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
672 	BUG_ON(!ptep || level != PG_LEVEL_4K);
673 
674 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
675 		return mfn == INVALID_P2M_ENTRY;
676 
677 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
678 		return mfn == IDENTITY_FRAME(pfn);
679 
680 	return false;
681 }
682 
683 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
684 {
685 	if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
686 		int ret;
687 
688 		ret = xen_alloc_p2m_entry(pfn);
689 		if (ret < 0)
690 			return false;
691 
692 		return __set_phys_to_machine(pfn, mfn);
693 	}
694 
695 	return true;
696 }
697 
698 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
699 			    struct gnttab_map_grant_ref *kmap_ops,
700 			    struct page **pages, unsigned int count)
701 {
702 	int i, ret = 0;
703 	pte_t *pte;
704 
705 	if (xen_feature(XENFEAT_auto_translated_physmap))
706 		return 0;
707 
708 	if (kmap_ops) {
709 		ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
710 						kmap_ops, count);
711 		if (ret)
712 			goto out;
713 	}
714 
715 	for (i = 0; i < count; i++) {
716 		unsigned long mfn, pfn;
717 
718 		/* Do not add to override if the map failed. */
719 		if (map_ops[i].status)
720 			continue;
721 
722 		if (map_ops[i].flags & GNTMAP_contains_pte) {
723 			pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
724 				(map_ops[i].host_addr & ~PAGE_MASK));
725 			mfn = pte_mfn(*pte);
726 		} else {
727 			mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
728 		}
729 		pfn = page_to_pfn(pages[i]);
730 
731 		WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
732 
733 		if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
734 			ret = -ENOMEM;
735 			goto out;
736 		}
737 	}
738 
739 out:
740 	return ret;
741 }
742 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
743 
744 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
745 			      struct gnttab_unmap_grant_ref *kunmap_ops,
746 			      struct page **pages, unsigned int count)
747 {
748 	int i, ret = 0;
749 
750 	if (xen_feature(XENFEAT_auto_translated_physmap))
751 		return 0;
752 
753 	for (i = 0; i < count; i++) {
754 		unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
755 		unsigned long pfn = page_to_pfn(pages[i]);
756 
757 		if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
758 			ret = -EINVAL;
759 			goto out;
760 		}
761 
762 		set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
763 	}
764 	if (kunmap_ops)
765 		ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
766 						kunmap_ops, count);
767 out:
768 	return ret;
769 }
770 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
771 
772 #ifdef CONFIG_XEN_DEBUG_FS
773 #include <linux/debugfs.h>
774 #include "debugfs.h"
775 static int p2m_dump_show(struct seq_file *m, void *v)
776 {
777 	static const char * const type_name[] = {
778 				[P2M_TYPE_IDENTITY] = "identity",
779 				[P2M_TYPE_MISSING] = "missing",
780 				[P2M_TYPE_PFN] = "pfn",
781 				[P2M_TYPE_UNKNOWN] = "abnormal"};
782 	unsigned long pfn, first_pfn;
783 	int type, prev_type;
784 
785 	prev_type = xen_p2m_elem_type(0);
786 	first_pfn = 0;
787 
788 	for (pfn = 0; pfn < xen_p2m_size; pfn++) {
789 		type = xen_p2m_elem_type(pfn);
790 		if (type != prev_type) {
791 			seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
792 				   type_name[prev_type]);
793 			prev_type = type;
794 			first_pfn = pfn;
795 		}
796 	}
797 	seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
798 		   type_name[prev_type]);
799 	return 0;
800 }
801 
802 static int p2m_dump_open(struct inode *inode, struct file *filp)
803 {
804 	return single_open(filp, p2m_dump_show, NULL);
805 }
806 
807 static const struct file_operations p2m_dump_fops = {
808 	.open		= p2m_dump_open,
809 	.read		= seq_read,
810 	.llseek		= seq_lseek,
811 	.release	= single_release,
812 };
813 
814 static struct dentry *d_mmu_debug;
815 
816 static int __init xen_p2m_debugfs(void)
817 {
818 	struct dentry *d_xen = xen_init_debugfs();
819 
820 	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
821 
822 	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
823 	return 0;
824 }
825 fs_initcall(xen_p2m_debugfs);
826 #endif /* CONFIG_XEN_DEBUG_FS */
827