xref: /openbmc/linux/arch/x86/xen/p2m.c (revision f71a261a)
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_MEMORY_HOTPLUG_LIMIT
102 #define P2M_LIMIT CONFIG_XEN_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(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 			mfns = alloc_p2m_page();
383 			copy_page(mfns, xen_p2m_addr + pfn);
384 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
385 			set_pte(ptep,
386 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
387 			continue;
388 		}
389 
390 		if (chunk == P2M_PER_PAGE) {
391 			/* Map complete missing or identity p2m-page. */
392 			mfns = (type == P2M_TYPE_MISSING) ?
393 				p2m_missing : p2m_identity;
394 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
395 			set_pte(ptep,
396 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
397 			continue;
398 		}
399 
400 		/* Complete missing or identity PMD(s) can be mapped. */
401 		ptep = (type == P2M_TYPE_MISSING) ?
402 			p2m_missing_pte : p2m_identity_pte;
403 		for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
404 			pmdp = populate_extra_pmd(
405 				(unsigned long)(p2m + pfn) + i * PMD_SIZE);
406 			set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
407 		}
408 	}
409 }
410 
411 void __init xen_vmalloc_p2m_tree(void)
412 {
413 	static struct vm_struct vm;
414 	unsigned long p2m_limit;
415 
416 	xen_p2m_last_pfn = xen_max_p2m_pfn;
417 
418 	p2m_limit = (phys_addr_t)P2M_LIMIT * 1024 * 1024 * 1024 / PAGE_SIZE;
419 	vm.flags = VM_ALLOC;
420 	vm.size = ALIGN(sizeof(unsigned long) * max(xen_max_p2m_pfn, p2m_limit),
421 			PMD_SIZE * PMDS_PER_MID_PAGE);
422 	vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
423 	pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
424 
425 	xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
426 
427 	xen_rebuild_p2m_list(vm.addr);
428 
429 	xen_p2m_addr = vm.addr;
430 	xen_p2m_size = xen_max_p2m_pfn;
431 
432 	xen_inv_extra_mem();
433 }
434 
435 unsigned long get_phys_to_machine(unsigned long pfn)
436 {
437 	pte_t *ptep;
438 	unsigned int level;
439 
440 	if (unlikely(pfn >= xen_p2m_size)) {
441 		if (pfn < xen_max_p2m_pfn)
442 			return xen_chk_extra_mem(pfn);
443 
444 		return IDENTITY_FRAME(pfn);
445 	}
446 
447 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
448 	BUG_ON(!ptep || level != PG_LEVEL_4K);
449 
450 	/*
451 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
452 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
453 	 * would be wrong.
454 	 */
455 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
456 		return IDENTITY_FRAME(pfn);
457 
458 	return xen_p2m_addr[pfn];
459 }
460 EXPORT_SYMBOL_GPL(get_phys_to_machine);
461 
462 /*
463  * Allocate new pmd(s). It is checked whether the old pmd is still in place.
464  * If not, nothing is changed. This is okay as the only reason for allocating
465  * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
466  * pmd.
467  */
468 static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
469 {
470 	pte_t *ptechk;
471 	pte_t *pte_newpg[PMDS_PER_MID_PAGE];
472 	pmd_t *pmdp;
473 	unsigned int level;
474 	unsigned long flags;
475 	unsigned long vaddr;
476 	int i;
477 
478 	/* Do all allocations first to bail out in error case. */
479 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
480 		pte_newpg[i] = alloc_p2m_page();
481 		if (!pte_newpg[i]) {
482 			for (i--; i >= 0; i--)
483 				free_p2m_page(pte_newpg[i]);
484 
485 			return NULL;
486 		}
487 	}
488 
489 	vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
490 
491 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
492 		copy_page(pte_newpg[i], pte_pg);
493 		paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
494 
495 		pmdp = lookup_pmd_address(vaddr);
496 		BUG_ON(!pmdp);
497 
498 		spin_lock_irqsave(&p2m_update_lock, flags);
499 
500 		ptechk = lookup_address(vaddr, &level);
501 		if (ptechk == pte_pg) {
502 			HYPERVISOR_shared_info->arch.p2m_generation++;
503 			wmb(); /* Tools are synchronizing via p2m_generation. */
504 			set_pmd(pmdp,
505 				__pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
506 			wmb(); /* Tools are synchronizing via p2m_generation. */
507 			HYPERVISOR_shared_info->arch.p2m_generation++;
508 			pte_newpg[i] = NULL;
509 		}
510 
511 		spin_unlock_irqrestore(&p2m_update_lock, flags);
512 
513 		if (pte_newpg[i]) {
514 			paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
515 			free_p2m_page(pte_newpg[i]);
516 		}
517 
518 		vaddr += PMD_SIZE;
519 	}
520 
521 	return lookup_address(addr, &level);
522 }
523 
524 /*
525  * Fully allocate the p2m structure for a given pfn.  We need to check
526  * that both the top and mid levels are allocated, and make sure the
527  * parallel mfn tree is kept in sync.  We may race with other cpus, so
528  * the new pages are installed with cmpxchg; if we lose the race then
529  * simply free the page we allocated and use the one that's there.
530  */
531 int xen_alloc_p2m_entry(unsigned long pfn)
532 {
533 	unsigned topidx;
534 	unsigned long *top_mfn_p, *mid_mfn;
535 	pte_t *ptep, *pte_pg;
536 	unsigned int level;
537 	unsigned long flags;
538 	unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
539 	unsigned long p2m_pfn;
540 
541 	ptep = lookup_address(addr, &level);
542 	BUG_ON(!ptep || level != PG_LEVEL_4K);
543 	pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
544 
545 	if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
546 		/* PMD level is missing, allocate a new one */
547 		ptep = alloc_p2m_pmd(addr, pte_pg);
548 		if (!ptep)
549 			return -ENOMEM;
550 	}
551 
552 	if (p2m_top_mfn && pfn < MAX_P2M_PFN) {
553 		topidx = p2m_top_index(pfn);
554 		top_mfn_p = &p2m_top_mfn[topidx];
555 		mid_mfn = READ_ONCE(p2m_top_mfn_p[topidx]);
556 
557 		BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
558 
559 		if (mid_mfn == p2m_mid_missing_mfn) {
560 			/* Separately check the mid mfn level */
561 			unsigned long missing_mfn;
562 			unsigned long mid_mfn_mfn;
563 			unsigned long old_mfn;
564 
565 			mid_mfn = alloc_p2m_page();
566 			if (!mid_mfn)
567 				return -ENOMEM;
568 
569 			p2m_mid_mfn_init(mid_mfn, p2m_missing);
570 
571 			missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
572 			mid_mfn_mfn = virt_to_mfn(mid_mfn);
573 			old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
574 			if (old_mfn != missing_mfn) {
575 				free_p2m_page(mid_mfn);
576 				mid_mfn = mfn_to_virt(old_mfn);
577 			} else {
578 				p2m_top_mfn_p[topidx] = mid_mfn;
579 			}
580 		}
581 	} else {
582 		mid_mfn = NULL;
583 	}
584 
585 	p2m_pfn = pte_pfn(READ_ONCE(*ptep));
586 	if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
587 	    p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
588 		/* p2m leaf page is missing */
589 		unsigned long *p2m;
590 
591 		p2m = alloc_p2m_page();
592 		if (!p2m)
593 			return -ENOMEM;
594 
595 		if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
596 			p2m_init(p2m);
597 		else
598 			p2m_init_identity(p2m, pfn & ~(P2M_PER_PAGE - 1));
599 
600 		spin_lock_irqsave(&p2m_update_lock, flags);
601 
602 		if (pte_pfn(*ptep) == p2m_pfn) {
603 			HYPERVISOR_shared_info->arch.p2m_generation++;
604 			wmb(); /* Tools are synchronizing via p2m_generation. */
605 			set_pte(ptep,
606 				pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
607 			wmb(); /* Tools are synchronizing via p2m_generation. */
608 			HYPERVISOR_shared_info->arch.p2m_generation++;
609 			if (mid_mfn)
610 				mid_mfn[p2m_mid_index(pfn)] = virt_to_mfn(p2m);
611 			p2m = NULL;
612 		}
613 
614 		spin_unlock_irqrestore(&p2m_update_lock, flags);
615 
616 		if (p2m)
617 			free_p2m_page(p2m);
618 	}
619 
620 	/* Expanded the p2m? */
621 	if (pfn >= xen_p2m_last_pfn) {
622 		xen_p2m_last_pfn = ALIGN(pfn + 1, P2M_PER_PAGE);
623 		HYPERVISOR_shared_info->arch.max_pfn = xen_p2m_last_pfn;
624 	}
625 
626 	return 0;
627 }
628 EXPORT_SYMBOL(xen_alloc_p2m_entry);
629 
630 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
631 				      unsigned long pfn_e)
632 {
633 	unsigned long pfn;
634 
635 	if (unlikely(pfn_s >= xen_p2m_size))
636 		return 0;
637 
638 	if (pfn_s > pfn_e)
639 		return 0;
640 
641 	if (pfn_e > xen_p2m_size)
642 		pfn_e = xen_p2m_size;
643 
644 	for (pfn = pfn_s; pfn < pfn_e; pfn++)
645 		xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
646 
647 	return pfn - pfn_s;
648 }
649 
650 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
651 {
652 	pte_t *ptep;
653 	unsigned int level;
654 
655 	/* Only invalid entries allowed above the highest p2m covered frame. */
656 	if (unlikely(pfn >= xen_p2m_size))
657 		return mfn == INVALID_P2M_ENTRY;
658 
659 	/*
660 	 * The interface requires atomic updates on p2m elements.
661 	 * xen_safe_write_ulong() is using an atomic store via asm().
662 	 */
663 	if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
664 		return true;
665 
666 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
667 	BUG_ON(!ptep || level != PG_LEVEL_4K);
668 
669 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
670 		return mfn == INVALID_P2M_ENTRY;
671 
672 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
673 		return mfn == IDENTITY_FRAME(pfn);
674 
675 	return false;
676 }
677 
678 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
679 {
680 	if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
681 		int ret;
682 
683 		ret = xen_alloc_p2m_entry(pfn);
684 		if (ret < 0)
685 			return false;
686 
687 		return __set_phys_to_machine(pfn, mfn);
688 	}
689 
690 	return true;
691 }
692 
693 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
694 			    struct gnttab_map_grant_ref *kmap_ops,
695 			    struct page **pages, unsigned int count)
696 {
697 	int i, ret = 0;
698 	pte_t *pte;
699 
700 	if (xen_feature(XENFEAT_auto_translated_physmap))
701 		return 0;
702 
703 	if (kmap_ops) {
704 		ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
705 						kmap_ops, count);
706 		if (ret)
707 			goto out;
708 	}
709 
710 	for (i = 0; i < count; i++) {
711 		unsigned long mfn, pfn;
712 		struct gnttab_unmap_grant_ref unmap[2];
713 		int rc;
714 
715 		/* Do not add to override if the map failed. */
716 		if (map_ops[i].status != GNTST_okay ||
717 		    (kmap_ops && kmap_ops[i].status != GNTST_okay))
718 			continue;
719 
720 		if (map_ops[i].flags & GNTMAP_contains_pte) {
721 			pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
722 				(map_ops[i].host_addr & ~PAGE_MASK));
723 			mfn = pte_mfn(*pte);
724 		} else {
725 			mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
726 		}
727 		pfn = page_to_pfn(pages[i]);
728 
729 		WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
730 
731 		if (likely(set_phys_to_machine(pfn, FOREIGN_FRAME(mfn))))
732 			continue;
733 
734 		/*
735 		 * Signal an error for this slot. This in turn requires
736 		 * immediate unmapping.
737 		 */
738 		map_ops[i].status = GNTST_general_error;
739 		unmap[0].host_addr = map_ops[i].host_addr,
740 		unmap[0].handle = map_ops[i].handle;
741 		map_ops[i].handle = INVALID_GRANT_HANDLE;
742 		if (map_ops[i].flags & GNTMAP_device_map)
743 			unmap[0].dev_bus_addr = map_ops[i].dev_bus_addr;
744 		else
745 			unmap[0].dev_bus_addr = 0;
746 
747 		if (kmap_ops) {
748 			kmap_ops[i].status = GNTST_general_error;
749 			unmap[1].host_addr = kmap_ops[i].host_addr,
750 			unmap[1].handle = kmap_ops[i].handle;
751 			kmap_ops[i].handle = INVALID_GRANT_HANDLE;
752 			if (kmap_ops[i].flags & GNTMAP_device_map)
753 				unmap[1].dev_bus_addr = kmap_ops[i].dev_bus_addr;
754 			else
755 				unmap[1].dev_bus_addr = 0;
756 		}
757 
758 		/*
759 		 * Pre-populate both status fields, to be recognizable in
760 		 * the log message below.
761 		 */
762 		unmap[0].status = 1;
763 		unmap[1].status = 1;
764 
765 		rc = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
766 					       unmap, 1 + !!kmap_ops);
767 		if (rc || unmap[0].status != GNTST_okay ||
768 		    unmap[1].status != GNTST_okay)
769 			pr_err_once("gnttab unmap failed: rc=%d st0=%d st1=%d\n",
770 				    rc, unmap[0].status, unmap[1].status);
771 	}
772 
773 out:
774 	return ret;
775 }
776 
777 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
778 			      struct gnttab_unmap_grant_ref *kunmap_ops,
779 			      struct page **pages, unsigned int count)
780 {
781 	int i, ret = 0;
782 
783 	if (xen_feature(XENFEAT_auto_translated_physmap))
784 		return 0;
785 
786 	for (i = 0; i < count; i++) {
787 		unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
788 		unsigned long pfn = page_to_pfn(pages[i]);
789 
790 		if (mfn != INVALID_P2M_ENTRY && (mfn & FOREIGN_FRAME_BIT))
791 			set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
792 		else
793 			ret = -EINVAL;
794 	}
795 	if (kunmap_ops)
796 		ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
797 						kunmap_ops, count) ?: ret;
798 
799 	return ret;
800 }
801 
802 #ifdef CONFIG_XEN_DEBUG_FS
803 #include <linux/debugfs.h>
804 #include "debugfs.h"
805 static int p2m_dump_show(struct seq_file *m, void *v)
806 {
807 	static const char * const type_name[] = {
808 				[P2M_TYPE_IDENTITY] = "identity",
809 				[P2M_TYPE_MISSING] = "missing",
810 				[P2M_TYPE_PFN] = "pfn",
811 				[P2M_TYPE_UNKNOWN] = "abnormal"};
812 	unsigned long pfn, first_pfn;
813 	int type, prev_type;
814 
815 	prev_type = xen_p2m_elem_type(0);
816 	first_pfn = 0;
817 
818 	for (pfn = 0; pfn < xen_p2m_size; pfn++) {
819 		type = xen_p2m_elem_type(pfn);
820 		if (type != prev_type) {
821 			seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
822 				   type_name[prev_type]);
823 			prev_type = type;
824 			first_pfn = pfn;
825 		}
826 	}
827 	seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
828 		   type_name[prev_type]);
829 	return 0;
830 }
831 
832 DEFINE_SHOW_ATTRIBUTE(p2m_dump);
833 
834 static struct dentry *d_mmu_debug;
835 
836 static int __init xen_p2m_debugfs(void)
837 {
838 	struct dentry *d_xen = xen_init_debugfs();
839 
840 	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
841 
842 	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
843 	return 0;
844 }
845 fs_initcall(xen_p2m_debugfs);
846 #endif /* CONFIG_XEN_DEBUG_FS */
847