xref: /openbmc/linux/arch/x86/xen/p2m.c (revision 92b19ff5)
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/module.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 "p2m.h"
83 #include "multicalls.h"
84 #include "xen-ops.h"
85 
86 #define PMDS_PER_MID_PAGE	(P2M_MID_PER_PAGE / PTRS_PER_PTE)
87 
88 unsigned long *xen_p2m_addr __read_mostly;
89 EXPORT_SYMBOL_GPL(xen_p2m_addr);
90 unsigned long xen_p2m_size __read_mostly;
91 EXPORT_SYMBOL_GPL(xen_p2m_size);
92 unsigned long xen_max_p2m_pfn __read_mostly;
93 EXPORT_SYMBOL_GPL(xen_max_p2m_pfn);
94 
95 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
96 #define P2M_LIMIT CONFIG_XEN_BALLOON_MEMORY_HOTPLUG_LIMIT
97 #else
98 #define P2M_LIMIT 0
99 #endif
100 
101 static DEFINE_SPINLOCK(p2m_update_lock);
102 
103 static unsigned long *p2m_mid_missing_mfn;
104 static unsigned long *p2m_top_mfn;
105 static unsigned long **p2m_top_mfn_p;
106 static unsigned long *p2m_missing;
107 static unsigned long *p2m_identity;
108 static pte_t *p2m_missing_pte;
109 static pte_t *p2m_identity_pte;
110 
111 static inline unsigned p2m_top_index(unsigned long pfn)
112 {
113 	BUG_ON(pfn >= MAX_P2M_PFN);
114 	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
115 }
116 
117 static inline unsigned p2m_mid_index(unsigned long pfn)
118 {
119 	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
120 }
121 
122 static inline unsigned p2m_index(unsigned long pfn)
123 {
124 	return pfn % P2M_PER_PAGE;
125 }
126 
127 static void p2m_top_mfn_init(unsigned long *top)
128 {
129 	unsigned i;
130 
131 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
132 		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
133 }
134 
135 static void p2m_top_mfn_p_init(unsigned long **top)
136 {
137 	unsigned i;
138 
139 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
140 		top[i] = p2m_mid_missing_mfn;
141 }
142 
143 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
144 {
145 	unsigned i;
146 
147 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
148 		mid[i] = virt_to_mfn(leaf);
149 }
150 
151 static void p2m_init(unsigned long *p2m)
152 {
153 	unsigned i;
154 
155 	for (i = 0; i < P2M_PER_PAGE; i++)
156 		p2m[i] = INVALID_P2M_ENTRY;
157 }
158 
159 static void p2m_init_identity(unsigned long *p2m, unsigned long pfn)
160 {
161 	unsigned i;
162 
163 	for (i = 0; i < P2M_PER_PAGE; i++)
164 		p2m[i] = IDENTITY_FRAME(pfn + i);
165 }
166 
167 static void * __ref alloc_p2m_page(void)
168 {
169 	if (unlikely(!slab_is_available()))
170 		return alloc_bootmem_align(PAGE_SIZE, PAGE_SIZE);
171 
172 	return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
173 }
174 
175 static void __ref free_p2m_page(void *p)
176 {
177 	if (unlikely(!slab_is_available())) {
178 		free_bootmem((unsigned long)p, PAGE_SIZE);
179 		return;
180 	}
181 
182 	free_page((unsigned long)p);
183 }
184 
185 /*
186  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
187  *
188  * This is called both at boot time, and after resuming from suspend:
189  * - At boot time we're called rather early, and must use alloc_bootmem*()
190  *   to allocate memory.
191  *
192  * - After resume we're called from within stop_machine, but the mfn
193  *   tree should already be completely allocated.
194  */
195 void __ref xen_build_mfn_list_list(void)
196 {
197 	unsigned long pfn, mfn;
198 	pte_t *ptep;
199 	unsigned int level, topidx, mididx;
200 	unsigned long *mid_mfn_p;
201 
202 	if (xen_feature(XENFEAT_auto_translated_physmap))
203 		return;
204 
205 	/* Pre-initialize p2m_top_mfn to be completely missing */
206 	if (p2m_top_mfn == NULL) {
207 		p2m_mid_missing_mfn = alloc_p2m_page();
208 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
209 
210 		p2m_top_mfn_p = alloc_p2m_page();
211 		p2m_top_mfn_p_init(p2m_top_mfn_p);
212 
213 		p2m_top_mfn = alloc_p2m_page();
214 		p2m_top_mfn_init(p2m_top_mfn);
215 	} else {
216 		/* Reinitialise, mfn's all change after migration */
217 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
218 	}
219 
220 	for (pfn = 0; pfn < xen_max_p2m_pfn && pfn < MAX_P2M_PFN;
221 	     pfn += P2M_PER_PAGE) {
222 		topidx = p2m_top_index(pfn);
223 		mididx = p2m_mid_index(pfn);
224 
225 		mid_mfn_p = p2m_top_mfn_p[topidx];
226 		ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn),
227 				      &level);
228 		BUG_ON(!ptep || level != PG_LEVEL_4K);
229 		mfn = pte_mfn(*ptep);
230 		ptep = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
231 
232 		/* Don't bother allocating any mfn mid levels if
233 		 * they're just missing, just update the stored mfn,
234 		 * since all could have changed over a migrate.
235 		 */
236 		if (ptep == p2m_missing_pte || ptep == p2m_identity_pte) {
237 			BUG_ON(mididx);
238 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
239 			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
240 			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
241 			continue;
242 		}
243 
244 		if (mid_mfn_p == p2m_mid_missing_mfn) {
245 			mid_mfn_p = alloc_p2m_page();
246 			p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
247 
248 			p2m_top_mfn_p[topidx] = mid_mfn_p;
249 		}
250 
251 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
252 		mid_mfn_p[mididx] = mfn;
253 	}
254 }
255 
256 void xen_setup_mfn_list_list(void)
257 {
258 	if (xen_feature(XENFEAT_auto_translated_physmap))
259 		return;
260 
261 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
262 
263 	HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
264 		virt_to_mfn(p2m_top_mfn);
265 	HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
266 }
267 
268 /* Set up p2m_top to point to the domain-builder provided p2m pages */
269 void __init xen_build_dynamic_phys_to_machine(void)
270 {
271 	unsigned long pfn;
272 
273 	 if (xen_feature(XENFEAT_auto_translated_physmap))
274 		return;
275 
276 	xen_p2m_addr = (unsigned long *)xen_start_info->mfn_list;
277 	xen_p2m_size = ALIGN(xen_start_info->nr_pages, P2M_PER_PAGE);
278 
279 	for (pfn = xen_start_info->nr_pages; pfn < xen_p2m_size; pfn++)
280 		xen_p2m_addr[pfn] = INVALID_P2M_ENTRY;
281 
282 	xen_max_p2m_pfn = xen_p2m_size;
283 }
284 
285 #define P2M_TYPE_IDENTITY	0
286 #define P2M_TYPE_MISSING	1
287 #define P2M_TYPE_PFN		2
288 #define P2M_TYPE_UNKNOWN	3
289 
290 static int xen_p2m_elem_type(unsigned long pfn)
291 {
292 	unsigned long mfn;
293 
294 	if (pfn >= xen_p2m_size)
295 		return P2M_TYPE_IDENTITY;
296 
297 	mfn = xen_p2m_addr[pfn];
298 
299 	if (mfn == INVALID_P2M_ENTRY)
300 		return P2M_TYPE_MISSING;
301 
302 	if (mfn & IDENTITY_FRAME_BIT)
303 		return P2M_TYPE_IDENTITY;
304 
305 	return P2M_TYPE_PFN;
306 }
307 
308 static void __init xen_rebuild_p2m_list(unsigned long *p2m)
309 {
310 	unsigned int i, chunk;
311 	unsigned long pfn;
312 	unsigned long *mfns;
313 	pte_t *ptep;
314 	pmd_t *pmdp;
315 	int type;
316 
317 	p2m_missing = alloc_p2m_page();
318 	p2m_init(p2m_missing);
319 	p2m_identity = alloc_p2m_page();
320 	p2m_init(p2m_identity);
321 
322 	p2m_missing_pte = alloc_p2m_page();
323 	paravirt_alloc_pte(&init_mm, __pa(p2m_missing_pte) >> PAGE_SHIFT);
324 	p2m_identity_pte = alloc_p2m_page();
325 	paravirt_alloc_pte(&init_mm, __pa(p2m_identity_pte) >> PAGE_SHIFT);
326 	for (i = 0; i < PTRS_PER_PTE; i++) {
327 		set_pte(p2m_missing_pte + i,
328 			pfn_pte(PFN_DOWN(__pa(p2m_missing)), PAGE_KERNEL_RO));
329 		set_pte(p2m_identity_pte + i,
330 			pfn_pte(PFN_DOWN(__pa(p2m_identity)), PAGE_KERNEL_RO));
331 	}
332 
333 	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += chunk) {
334 		/*
335 		 * Try to map missing/identity PMDs or p2m-pages if possible.
336 		 * We have to respect the structure of the mfn_list_list
337 		 * which will be built just afterwards.
338 		 * Chunk size to test is one p2m page if we are in the middle
339 		 * of a mfn_list_list mid page and the complete mid page area
340 		 * if we are at index 0 of the mid page. Please note that a
341 		 * mid page might cover more than one PMD, e.g. on 32 bit PAE
342 		 * kernels.
343 		 */
344 		chunk = (pfn & (P2M_PER_PAGE * P2M_MID_PER_PAGE - 1)) ?
345 			P2M_PER_PAGE : P2M_PER_PAGE * P2M_MID_PER_PAGE;
346 
347 		type = xen_p2m_elem_type(pfn);
348 		i = 0;
349 		if (type != P2M_TYPE_PFN)
350 			for (i = 1; i < chunk; i++)
351 				if (xen_p2m_elem_type(pfn + i) != type)
352 					break;
353 		if (i < chunk)
354 			/* Reset to minimal chunk size. */
355 			chunk = P2M_PER_PAGE;
356 
357 		if (type == P2M_TYPE_PFN || i < chunk) {
358 			/* Use initial p2m page contents. */
359 #ifdef CONFIG_X86_64
360 			mfns = alloc_p2m_page();
361 			copy_page(mfns, xen_p2m_addr + pfn);
362 #else
363 			mfns = xen_p2m_addr + pfn;
364 #endif
365 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
366 			set_pte(ptep,
367 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL));
368 			continue;
369 		}
370 
371 		if (chunk == P2M_PER_PAGE) {
372 			/* Map complete missing or identity p2m-page. */
373 			mfns = (type == P2M_TYPE_MISSING) ?
374 				p2m_missing : p2m_identity;
375 			ptep = populate_extra_pte((unsigned long)(p2m + pfn));
376 			set_pte(ptep,
377 				pfn_pte(PFN_DOWN(__pa(mfns)), PAGE_KERNEL_RO));
378 			continue;
379 		}
380 
381 		/* Complete missing or identity PMD(s) can be mapped. */
382 		ptep = (type == P2M_TYPE_MISSING) ?
383 			p2m_missing_pte : p2m_identity_pte;
384 		for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
385 			pmdp = populate_extra_pmd(
386 				(unsigned long)(p2m + pfn) + i * PMD_SIZE);
387 			set_pmd(pmdp, __pmd(__pa(ptep) | _KERNPG_TABLE));
388 		}
389 	}
390 }
391 
392 void __init xen_vmalloc_p2m_tree(void)
393 {
394 	static struct vm_struct vm;
395 	unsigned long p2m_limit;
396 
397 	p2m_limit = (phys_addr_t)P2M_LIMIT * 1024 * 1024 * 1024 / PAGE_SIZE;
398 	vm.flags = VM_ALLOC;
399 	vm.size = ALIGN(sizeof(unsigned long) * max(xen_max_p2m_pfn, p2m_limit),
400 			PMD_SIZE * PMDS_PER_MID_PAGE);
401 	vm_area_register_early(&vm, PMD_SIZE * PMDS_PER_MID_PAGE);
402 	pr_notice("p2m virtual area at %p, size is %lx\n", vm.addr, vm.size);
403 
404 	xen_max_p2m_pfn = vm.size / sizeof(unsigned long);
405 
406 	xen_rebuild_p2m_list(vm.addr);
407 
408 	xen_p2m_addr = vm.addr;
409 	xen_p2m_size = xen_max_p2m_pfn;
410 
411 	xen_inv_extra_mem();
412 }
413 
414 unsigned long get_phys_to_machine(unsigned long pfn)
415 {
416 	pte_t *ptep;
417 	unsigned int level;
418 
419 	if (unlikely(pfn >= xen_p2m_size)) {
420 		if (pfn < xen_max_p2m_pfn)
421 			return xen_chk_extra_mem(pfn);
422 
423 		return IDENTITY_FRAME(pfn);
424 	}
425 
426 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
427 	BUG_ON(!ptep || level != PG_LEVEL_4K);
428 
429 	/*
430 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
431 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
432 	 * would be wrong.
433 	 */
434 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
435 		return IDENTITY_FRAME(pfn);
436 
437 	return xen_p2m_addr[pfn];
438 }
439 EXPORT_SYMBOL_GPL(get_phys_to_machine);
440 
441 /*
442  * Allocate new pmd(s). It is checked whether the old pmd is still in place.
443  * If not, nothing is changed. This is okay as the only reason for allocating
444  * a new pmd is to replace p2m_missing_pte or p2m_identity_pte by a individual
445  * pmd. In case of PAE/x86-32 there are multiple pmds to allocate!
446  */
447 static pte_t *alloc_p2m_pmd(unsigned long addr, pte_t *pte_pg)
448 {
449 	pte_t *ptechk;
450 	pte_t *pte_newpg[PMDS_PER_MID_PAGE];
451 	pmd_t *pmdp;
452 	unsigned int level;
453 	unsigned long flags;
454 	unsigned long vaddr;
455 	int i;
456 
457 	/* Do all allocations first to bail out in error case. */
458 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
459 		pte_newpg[i] = alloc_p2m_page();
460 		if (!pte_newpg[i]) {
461 			for (i--; i >= 0; i--)
462 				free_p2m_page(pte_newpg[i]);
463 
464 			return NULL;
465 		}
466 	}
467 
468 	vaddr = addr & ~(PMD_SIZE * PMDS_PER_MID_PAGE - 1);
469 
470 	for (i = 0; i < PMDS_PER_MID_PAGE; i++) {
471 		copy_page(pte_newpg[i], pte_pg);
472 		paravirt_alloc_pte(&init_mm, __pa(pte_newpg[i]) >> PAGE_SHIFT);
473 
474 		pmdp = lookup_pmd_address(vaddr);
475 		BUG_ON(!pmdp);
476 
477 		spin_lock_irqsave(&p2m_update_lock, flags);
478 
479 		ptechk = lookup_address(vaddr, &level);
480 		if (ptechk == pte_pg) {
481 			set_pmd(pmdp,
482 				__pmd(__pa(pte_newpg[i]) | _KERNPG_TABLE));
483 			pte_newpg[i] = NULL;
484 		}
485 
486 		spin_unlock_irqrestore(&p2m_update_lock, flags);
487 
488 		if (pte_newpg[i]) {
489 			paravirt_release_pte(__pa(pte_newpg[i]) >> PAGE_SHIFT);
490 			free_p2m_page(pte_newpg[i]);
491 		}
492 
493 		vaddr += PMD_SIZE;
494 	}
495 
496 	return lookup_address(addr, &level);
497 }
498 
499 /*
500  * Fully allocate the p2m structure for a given pfn.  We need to check
501  * that both the top and mid levels are allocated, and make sure the
502  * parallel mfn tree is kept in sync.  We may race with other cpus, so
503  * the new pages are installed with cmpxchg; if we lose the race then
504  * simply free the page we allocated and use the one that's there.
505  */
506 static bool alloc_p2m(unsigned long pfn)
507 {
508 	unsigned topidx, mididx;
509 	unsigned long *top_mfn_p, *mid_mfn;
510 	pte_t *ptep, *pte_pg;
511 	unsigned int level;
512 	unsigned long flags;
513 	unsigned long addr = (unsigned long)(xen_p2m_addr + pfn);
514 	unsigned long p2m_pfn;
515 
516 	topidx = p2m_top_index(pfn);
517 	mididx = p2m_mid_index(pfn);
518 
519 	ptep = lookup_address(addr, &level);
520 	BUG_ON(!ptep || level != PG_LEVEL_4K);
521 	pte_pg = (pte_t *)((unsigned long)ptep & ~(PAGE_SIZE - 1));
522 
523 	if (pte_pg == p2m_missing_pte || pte_pg == p2m_identity_pte) {
524 		/* PMD level is missing, allocate a new one */
525 		ptep = alloc_p2m_pmd(addr, pte_pg);
526 		if (!ptep)
527 			return false;
528 	}
529 
530 	if (p2m_top_mfn) {
531 		top_mfn_p = &p2m_top_mfn[topidx];
532 		mid_mfn = ACCESS_ONCE(p2m_top_mfn_p[topidx]);
533 
534 		BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
535 
536 		if (mid_mfn == p2m_mid_missing_mfn) {
537 			/* Separately check the mid mfn level */
538 			unsigned long missing_mfn;
539 			unsigned long mid_mfn_mfn;
540 			unsigned long old_mfn;
541 
542 			mid_mfn = alloc_p2m_page();
543 			if (!mid_mfn)
544 				return false;
545 
546 			p2m_mid_mfn_init(mid_mfn, p2m_missing);
547 
548 			missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
549 			mid_mfn_mfn = virt_to_mfn(mid_mfn);
550 			old_mfn = cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn);
551 			if (old_mfn != missing_mfn) {
552 				free_p2m_page(mid_mfn);
553 				mid_mfn = mfn_to_virt(old_mfn);
554 			} else {
555 				p2m_top_mfn_p[topidx] = mid_mfn;
556 			}
557 		}
558 	} else {
559 		mid_mfn = NULL;
560 	}
561 
562 	p2m_pfn = pte_pfn(READ_ONCE(*ptep));
563 	if (p2m_pfn == PFN_DOWN(__pa(p2m_identity)) ||
564 	    p2m_pfn == PFN_DOWN(__pa(p2m_missing))) {
565 		/* p2m leaf page is missing */
566 		unsigned long *p2m;
567 
568 		p2m = alloc_p2m_page();
569 		if (!p2m)
570 			return false;
571 
572 		if (p2m_pfn == PFN_DOWN(__pa(p2m_missing)))
573 			p2m_init(p2m);
574 		else
575 			p2m_init_identity(p2m, pfn & ~(P2M_PER_PAGE - 1));
576 
577 		spin_lock_irqsave(&p2m_update_lock, flags);
578 
579 		if (pte_pfn(*ptep) == p2m_pfn) {
580 			set_pte(ptep,
581 				pfn_pte(PFN_DOWN(__pa(p2m)), PAGE_KERNEL));
582 			if (mid_mfn)
583 				mid_mfn[mididx] = virt_to_mfn(p2m);
584 			p2m = NULL;
585 		}
586 
587 		spin_unlock_irqrestore(&p2m_update_lock, flags);
588 
589 		if (p2m)
590 			free_p2m_page(p2m);
591 	}
592 
593 	return true;
594 }
595 
596 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
597 				      unsigned long pfn_e)
598 {
599 	unsigned long pfn;
600 
601 	if (unlikely(pfn_s >= xen_p2m_size))
602 		return 0;
603 
604 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
605 		return pfn_e - pfn_s;
606 
607 	if (pfn_s > pfn_e)
608 		return 0;
609 
610 	if (pfn_e > xen_p2m_size)
611 		pfn_e = xen_p2m_size;
612 
613 	for (pfn = pfn_s; pfn < pfn_e; pfn++)
614 		xen_p2m_addr[pfn] = IDENTITY_FRAME(pfn);
615 
616 	return pfn - pfn_s;
617 }
618 
619 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
620 {
621 	pte_t *ptep;
622 	unsigned int level;
623 
624 	/* don't track P2M changes in autotranslate guests */
625 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
626 		return true;
627 
628 	if (unlikely(pfn >= xen_p2m_size)) {
629 		BUG_ON(mfn != INVALID_P2M_ENTRY);
630 		return true;
631 	}
632 
633 	if (likely(!xen_safe_write_ulong(xen_p2m_addr + pfn, mfn)))
634 		return true;
635 
636 	ptep = lookup_address((unsigned long)(xen_p2m_addr + pfn), &level);
637 	BUG_ON(!ptep || level != PG_LEVEL_4K);
638 
639 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_missing)))
640 		return mfn == INVALID_P2M_ENTRY;
641 
642 	if (pte_pfn(*ptep) == PFN_DOWN(__pa(p2m_identity)))
643 		return mfn == IDENTITY_FRAME(pfn);
644 
645 	return false;
646 }
647 
648 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
649 {
650 	if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
651 		if (!alloc_p2m(pfn))
652 			return false;
653 
654 		return __set_phys_to_machine(pfn, mfn);
655 	}
656 
657 	return true;
658 }
659 
660 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
661 			    struct gnttab_map_grant_ref *kmap_ops,
662 			    struct page **pages, unsigned int count)
663 {
664 	int i, ret = 0;
665 	pte_t *pte;
666 
667 	if (xen_feature(XENFEAT_auto_translated_physmap))
668 		return 0;
669 
670 	if (kmap_ops) {
671 		ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
672 						kmap_ops, count);
673 		if (ret)
674 			goto out;
675 	}
676 
677 	for (i = 0; i < count; i++) {
678 		unsigned long mfn, pfn;
679 
680 		/* Do not add to override if the map failed. */
681 		if (map_ops[i].status)
682 			continue;
683 
684 		if (map_ops[i].flags & GNTMAP_contains_pte) {
685 			pte = (pte_t *)(mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
686 				(map_ops[i].host_addr & ~PAGE_MASK));
687 			mfn = pte_mfn(*pte);
688 		} else {
689 			mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
690 		}
691 		pfn = page_to_pfn(pages[i]);
692 
693 		WARN(pfn_to_mfn(pfn) != INVALID_P2M_ENTRY, "page must be ballooned");
694 
695 		if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
696 			ret = -ENOMEM;
697 			goto out;
698 		}
699 	}
700 
701 out:
702 	return ret;
703 }
704 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
705 
706 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
707 			      struct gnttab_unmap_grant_ref *kunmap_ops,
708 			      struct page **pages, unsigned int count)
709 {
710 	int i, ret = 0;
711 
712 	if (xen_feature(XENFEAT_auto_translated_physmap))
713 		return 0;
714 
715 	for (i = 0; i < count; i++) {
716 		unsigned long mfn = __pfn_to_mfn(page_to_pfn(pages[i]));
717 		unsigned long pfn = page_to_pfn(pages[i]);
718 
719 		if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
720 			ret = -EINVAL;
721 			goto out;
722 		}
723 
724 		set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
725 	}
726 	if (kunmap_ops)
727 		ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
728 						kunmap_ops, count);
729 out:
730 	return ret;
731 }
732 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
733 
734 #ifdef CONFIG_XEN_DEBUG_FS
735 #include <linux/debugfs.h>
736 #include "debugfs.h"
737 static int p2m_dump_show(struct seq_file *m, void *v)
738 {
739 	static const char * const type_name[] = {
740 				[P2M_TYPE_IDENTITY] = "identity",
741 				[P2M_TYPE_MISSING] = "missing",
742 				[P2M_TYPE_PFN] = "pfn",
743 				[P2M_TYPE_UNKNOWN] = "abnormal"};
744 	unsigned long pfn, first_pfn;
745 	int type, prev_type;
746 
747 	prev_type = xen_p2m_elem_type(0);
748 	first_pfn = 0;
749 
750 	for (pfn = 0; pfn < xen_p2m_size; pfn++) {
751 		type = xen_p2m_elem_type(pfn);
752 		if (type != prev_type) {
753 			seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
754 				   type_name[prev_type]);
755 			prev_type = type;
756 			first_pfn = pfn;
757 		}
758 	}
759 	seq_printf(m, " [0x%lx->0x%lx] %s\n", first_pfn, pfn,
760 		   type_name[prev_type]);
761 	return 0;
762 }
763 
764 static int p2m_dump_open(struct inode *inode, struct file *filp)
765 {
766 	return single_open(filp, p2m_dump_show, NULL);
767 }
768 
769 static const struct file_operations p2m_dump_fops = {
770 	.open		= p2m_dump_open,
771 	.read		= seq_read,
772 	.llseek		= seq_lseek,
773 	.release	= single_release,
774 };
775 
776 static struct dentry *d_mmu_debug;
777 
778 static int __init xen_p2m_debugfs(void)
779 {
780 	struct dentry *d_xen = xen_init_debugfs();
781 
782 	if (d_xen == NULL)
783 		return -ENOMEM;
784 
785 	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
786 
787 	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
788 	return 0;
789 }
790 fs_initcall(xen_p2m_debugfs);
791 #endif /* CONFIG_XEN_DEBUG_FS */
792