xref: /openbmc/linux/arch/x86/xen/p2m.c (revision d2999e1b)
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 p2m table is logically a flat array, but we implement it as a
7  * three-level tree to allow the address space to be sparse.
8  *
9  *                               Xen
10  *                                |
11  *     p2m_top              p2m_top_mfn
12  *       /  \                   /   \
13  * p2m_mid p2m_mid	p2m_mid_mfn p2m_mid_mfn
14  *    / \      / \         /           /
15  *  p2m p2m p2m p2m p2m p2m p2m ...
16  *
17  * The p2m_mid_mfn pages are mapped by p2m_top_mfn_p.
18  *
19  * The p2m_top and p2m_top_mfn levels are limited to 1 page, so the
20  * maximum representable pseudo-physical address space is:
21  *  P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE pages
22  *
23  * P2M_PER_PAGE depends on the architecture, as a mfn is always
24  * unsigned long (8 bytes on 64-bit, 4 bytes on 32), leading to
25  * 512 and 1024 entries respectively.
26  *
27  * In short, these structures contain the Machine Frame Number (MFN) of the PFN.
28  *
29  * However not all entries are filled with MFNs. Specifically for all other
30  * leaf entries, or for the top  root, or middle one, for which there is a void
31  * entry, we assume it is  "missing". So (for example)
32  *  pfn_to_mfn(0x90909090)=INVALID_P2M_ENTRY.
33  *
34  * We also have the possibility of setting 1-1 mappings on certain regions, so
35  * that:
36  *  pfn_to_mfn(0xc0000)=0xc0000
37  *
38  * The benefit of this is, that we can assume for non-RAM regions (think
39  * PCI BARs, or ACPI spaces), we can create mappings easily because we
40  * get the PFN value to match the MFN.
41  *
42  * For this to work efficiently we have one new page p2m_identity and
43  * allocate (via reserved_brk) any other pages we need to cover the sides
44  * (1GB or 4MB boundary violations). All entries in p2m_identity are set to
45  * INVALID_P2M_ENTRY type (Xen toolstack only recognizes that and MFNs,
46  * no other fancy value).
47  *
48  * On lookup we spot that the entry points to p2m_identity and return the
49  * identity value instead of dereferencing and returning INVALID_P2M_ENTRY.
50  * If the entry points to an allocated page, we just proceed as before and
51  * return the PFN.  If the PFN has IDENTITY_FRAME_BIT set we unmask that in
52  * appropriate functions (pfn_to_mfn).
53  *
54  * The reason for having the IDENTITY_FRAME_BIT instead of just returning the
55  * PFN is that we could find ourselves where pfn_to_mfn(pfn)==pfn for a
56  * non-identity pfn. To protect ourselves against we elect to set (and get) the
57  * IDENTITY_FRAME_BIT on all identity mapped PFNs.
58  *
59  * This simplistic diagram is used to explain the more subtle piece of code.
60  * There is also a digram of the P2M at the end that can help.
61  * Imagine your E820 looking as so:
62  *
63  *                    1GB                                           2GB    4GB
64  * /-------------------+---------\/----\         /----------\    /---+-----\
65  * | System RAM        | Sys RAM ||ACPI|         | reserved |    | Sys RAM |
66  * \-------------------+---------/\----/         \----------/    \---+-----/
67  *                               ^- 1029MB                       ^- 2001MB
68  *
69  * [1029MB = 263424 (0x40500), 2001MB = 512256 (0x7D100),
70  *  2048MB = 524288 (0x80000)]
71  *
72  * And dom0_mem=max:3GB,1GB is passed in to the guest, meaning memory past 1GB
73  * is actually not present (would have to kick the balloon driver to put it in).
74  *
75  * When we are told to set the PFNs for identity mapping (see patch: "xen/setup:
76  * Set identity mapping for non-RAM E820 and E820 gaps.") we pass in the start
77  * of the PFN and the end PFN (263424 and 512256 respectively). The first step
78  * is to reserve_brk a top leaf page if the p2m[1] is missing. The top leaf page
79  * covers 512^2 of page estate (1GB) and in case the start or end PFN is not
80  * aligned on 512^2*PAGE_SIZE (1GB) we reserve_brk new middle and leaf pages as
81  * required to split any existing p2m_mid_missing middle pages.
82  *
83  * With the E820 example above, 263424 is not 1GB aligned so we allocate a
84  * reserve_brk page which will cover the PFNs estate from 0x40000 to 0x80000.
85  * Each entry in the allocate page is "missing" (points to p2m_missing).
86  *
87  * Next stage is to determine if we need to do a more granular boundary check
88  * on the 4MB (or 2MB depending on architecture) off the start and end pfn's.
89  * We check if the start pfn and end pfn violate that boundary check, and if
90  * so reserve_brk a (p2m[x][y]) leaf page. This way we have a much finer
91  * granularity of setting which PFNs are missing and which ones are identity.
92  * In our example 263424 and 512256 both fail the check so we reserve_brk two
93  * pages. Populate them with INVALID_P2M_ENTRY (so they both have "missing"
94  * values) and assign them to p2m[1][2] and p2m[1][488] respectively.
95  *
96  * At this point we would at minimum reserve_brk one page, but could be up to
97  * three. Each call to set_phys_range_identity has at maximum a three page
98  * cost. If we were to query the P2M at this stage, all those entries from
99  * start PFN through end PFN (so 1029MB -> 2001MB) would return
100  * INVALID_P2M_ENTRY ("missing").
101  *
102  * The next step is to walk from the start pfn to the end pfn setting
103  * the IDENTITY_FRAME_BIT on each PFN. This is done in set_phys_range_identity.
104  * If we find that the middle entry is pointing to p2m_missing we can swap it
105  * over to p2m_identity - this way covering 4MB (or 2MB) PFN space (and
106  * similarly swapping p2m_mid_missing for p2m_mid_identity for larger regions).
107  * At this point we do not need to worry about boundary aligment (so no need to
108  * reserve_brk a middle page, figure out which PFNs are "missing" and which
109  * ones are identity), as that has been done earlier.  If we find that the
110  * middle leaf is not occupied by p2m_identity or p2m_missing, we dereference
111  * that page (which covers 512 PFNs) and set the appropriate PFN with
112  * IDENTITY_FRAME_BIT. In our example 263424 and 512256 end up there, and we
113  * set from p2m[1][2][256->511] and p2m[1][488][0->256] with
114  * IDENTITY_FRAME_BIT set.
115  *
116  * All other regions that are void (or not filled) either point to p2m_missing
117  * (considered missing) or have the default value of INVALID_P2M_ENTRY (also
118  * considered missing). In our case, p2m[1][2][0->255] and p2m[1][488][257->511]
119  * contain the INVALID_P2M_ENTRY value and are considered "missing."
120  *
121  * Finally, the region beyond the end of of the E820 (4 GB in this example)
122  * is set to be identity (in case there are MMIO regions placed here).
123  *
124  * This is what the p2m ends up looking (for the E820 above) with this
125  * fabulous drawing:
126  *
127  *    p2m         /--------------\
128  *  /-----\       | &mfn_list[0],|                           /-----------------\
129  *  |  0  |------>| &mfn_list[1],|    /---------------\      | ~0, ~0, ..      |
130  *  |-----|       |  ..., ~0, ~0 |    | ~0, ~0, [x]---+----->| IDENTITY [@256] |
131  *  |  1  |---\   \--------------/    | [p2m_identity]+\     | IDENTITY [@257] |
132  *  |-----|    \                      | [p2m_identity]+\\    | ....            |
133  *  |  2  |--\  \-------------------->|  ...          | \\   \----------------/
134  *  |-----|   \                       \---------------/  \\
135  *  |  3  |-\  \                                          \\  p2m_identity [1]
136  *  |-----|  \  \-------------------->/---------------\   /-----------------\
137  *  | ..  |\  |                       | [p2m_identity]+-->| ~0, ~0, ~0, ... |
138  *  \-----/ | |                       | [p2m_identity]+-->| ..., ~0         |
139  *          | |                       | ....          |   \-----------------/
140  *          | |                       +-[x], ~0, ~0.. +\
141  *          | |                       \---------------/ \
142  *          | |                                          \-> /---------------\
143  *          | V  p2m_mid_missing       p2m_missing           | IDENTITY[@0]  |
144  *          | /-----------------\     /------------\         | IDENTITY[@256]|
145  *          | | [p2m_missing]   +---->| ~0, ~0, ...|         | ~0, ~0, ....  |
146  *          | | [p2m_missing]   +---->| ..., ~0    |         \---------------/
147  *          | | ...             |     \------------/
148  *          | \-----------------/
149  *          |
150  *          |     p2m_mid_identity
151  *          |   /-----------------\
152  *          \-->| [p2m_identity]  +---->[1]
153  *              | [p2m_identity]  +---->[1]
154  *              | ...             |
155  *              \-----------------/
156  *
157  * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT)
158  */
159 
160 #include <linux/init.h>
161 #include <linux/module.h>
162 #include <linux/list.h>
163 #include <linux/hash.h>
164 #include <linux/sched.h>
165 #include <linux/seq_file.h>
166 
167 #include <asm/cache.h>
168 #include <asm/setup.h>
169 
170 #include <asm/xen/page.h>
171 #include <asm/xen/hypercall.h>
172 #include <asm/xen/hypervisor.h>
173 #include <xen/balloon.h>
174 #include <xen/grant_table.h>
175 
176 #include "multicalls.h"
177 #include "xen-ops.h"
178 
179 static void __init m2p_override_init(void);
180 
181 unsigned long xen_max_p2m_pfn __read_mostly;
182 
183 #define P2M_PER_PAGE		(PAGE_SIZE / sizeof(unsigned long))
184 #define P2M_MID_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long *))
185 #define P2M_TOP_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long **))
186 
187 #define MAX_P2M_PFN		(P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
188 
189 /* Placeholders for holes in the address space */
190 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE);
191 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE);
192 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn, P2M_MID_PER_PAGE);
193 
194 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE);
195 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, P2M_TOP_PER_PAGE);
196 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_TOP_PER_PAGE);
197 
198 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE);
199 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_identity, P2M_MID_PER_PAGE);
200 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_identity_mfn, P2M_MID_PER_PAGE);
201 
202 RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
203 RESERVE_BRK(p2m_mid_mfn, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
204 
205 /* We might hit two boundary violations at the start and end, at max each
206  * boundary violation will require three middle nodes. */
207 RESERVE_BRK(p2m_mid_extra, PAGE_SIZE * 2 * 3);
208 
209 /* When we populate back during bootup, the amount of pages can vary. The
210  * max we have is seen is 395979, but that does not mean it can't be more.
211  * Some machines can have 3GB I/O holes even. With early_can_reuse_p2m_middle
212  * it can re-use Xen provided mfn_list array, so we only need to allocate at
213  * most three P2M top nodes. */
214 RESERVE_BRK(p2m_populated, PAGE_SIZE * 3);
215 
216 static inline unsigned p2m_top_index(unsigned long pfn)
217 {
218 	BUG_ON(pfn >= MAX_P2M_PFN);
219 	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
220 }
221 
222 static inline unsigned p2m_mid_index(unsigned long pfn)
223 {
224 	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
225 }
226 
227 static inline unsigned p2m_index(unsigned long pfn)
228 {
229 	return pfn % P2M_PER_PAGE;
230 }
231 
232 static void p2m_top_init(unsigned long ***top)
233 {
234 	unsigned i;
235 
236 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
237 		top[i] = p2m_mid_missing;
238 }
239 
240 static void p2m_top_mfn_init(unsigned long *top)
241 {
242 	unsigned i;
243 
244 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
245 		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
246 }
247 
248 static void p2m_top_mfn_p_init(unsigned long **top)
249 {
250 	unsigned i;
251 
252 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
253 		top[i] = p2m_mid_missing_mfn;
254 }
255 
256 static void p2m_mid_init(unsigned long **mid, unsigned long *leaf)
257 {
258 	unsigned i;
259 
260 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
261 		mid[i] = leaf;
262 }
263 
264 static void p2m_mid_mfn_init(unsigned long *mid, unsigned long *leaf)
265 {
266 	unsigned i;
267 
268 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
269 		mid[i] = virt_to_mfn(leaf);
270 }
271 
272 static void p2m_init(unsigned long *p2m)
273 {
274 	unsigned i;
275 
276 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
277 		p2m[i] = INVALID_P2M_ENTRY;
278 }
279 
280 /*
281  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
282  *
283  * This is called both at boot time, and after resuming from suspend:
284  * - At boot time we're called very early, and must use extend_brk()
285  *   to allocate memory.
286  *
287  * - After resume we're called from within stop_machine, but the mfn
288  *   tree should alreay be completely allocated.
289  */
290 void __ref xen_build_mfn_list_list(void)
291 {
292 	unsigned long pfn;
293 
294 	if (xen_feature(XENFEAT_auto_translated_physmap))
295 		return;
296 
297 	/* Pre-initialize p2m_top_mfn to be completely missing */
298 	if (p2m_top_mfn == NULL) {
299 		p2m_mid_missing_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
300 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
301 		p2m_mid_identity_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
302 		p2m_mid_mfn_init(p2m_mid_identity_mfn, p2m_identity);
303 
304 		p2m_top_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
305 		p2m_top_mfn_p_init(p2m_top_mfn_p);
306 
307 		p2m_top_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
308 		p2m_top_mfn_init(p2m_top_mfn);
309 	} else {
310 		/* Reinitialise, mfn's all change after migration */
311 		p2m_mid_mfn_init(p2m_mid_missing_mfn, p2m_missing);
312 		p2m_mid_mfn_init(p2m_mid_identity_mfn, p2m_identity);
313 	}
314 
315 	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) {
316 		unsigned topidx = p2m_top_index(pfn);
317 		unsigned mididx = p2m_mid_index(pfn);
318 		unsigned long **mid;
319 		unsigned long *mid_mfn_p;
320 
321 		mid = p2m_top[topidx];
322 		mid_mfn_p = p2m_top_mfn_p[topidx];
323 
324 		/* Don't bother allocating any mfn mid levels if
325 		 * they're just missing, just update the stored mfn,
326 		 * since all could have changed over a migrate.
327 		 */
328 		if (mid == p2m_mid_missing) {
329 			BUG_ON(mididx);
330 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
331 			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
332 			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
333 			continue;
334 		}
335 
336 		if (mid_mfn_p == p2m_mid_missing_mfn) {
337 			/*
338 			 * XXX boot-time only!  We should never find
339 			 * missing parts of the mfn tree after
340 			 * runtime.  extend_brk() will BUG if we call
341 			 * it too late.
342 			 */
343 			mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
344 			p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
345 
346 			p2m_top_mfn_p[topidx] = mid_mfn_p;
347 		}
348 
349 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
350 		mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]);
351 	}
352 }
353 
354 void xen_setup_mfn_list_list(void)
355 {
356 	if (xen_feature(XENFEAT_auto_translated_physmap))
357 		return;
358 
359 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
360 
361 	HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
362 		virt_to_mfn(p2m_top_mfn);
363 	HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
364 }
365 
366 /* Set up p2m_top to point to the domain-builder provided p2m pages */
367 void __init xen_build_dynamic_phys_to_machine(void)
368 {
369 	unsigned long *mfn_list;
370 	unsigned long max_pfn;
371 	unsigned long pfn;
372 
373 	 if (xen_feature(XENFEAT_auto_translated_physmap))
374 		return;
375 
376 	mfn_list = (unsigned long *)xen_start_info->mfn_list;
377 	max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
378 	xen_max_p2m_pfn = max_pfn;
379 
380 	p2m_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
381 	p2m_init(p2m_missing);
382 	p2m_identity = extend_brk(PAGE_SIZE, PAGE_SIZE);
383 	p2m_init(p2m_identity);
384 
385 	p2m_mid_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
386 	p2m_mid_init(p2m_mid_missing, p2m_missing);
387 	p2m_mid_identity = extend_brk(PAGE_SIZE, PAGE_SIZE);
388 	p2m_mid_init(p2m_mid_identity, p2m_identity);
389 
390 	p2m_top = extend_brk(PAGE_SIZE, PAGE_SIZE);
391 	p2m_top_init(p2m_top);
392 
393 	/*
394 	 * The domain builder gives us a pre-constructed p2m array in
395 	 * mfn_list for all the pages initially given to us, so we just
396 	 * need to graft that into our tree structure.
397 	 */
398 	for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) {
399 		unsigned topidx = p2m_top_index(pfn);
400 		unsigned mididx = p2m_mid_index(pfn);
401 
402 		if (p2m_top[topidx] == p2m_mid_missing) {
403 			unsigned long **mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
404 			p2m_mid_init(mid, p2m_missing);
405 
406 			p2m_top[topidx] = mid;
407 		}
408 
409 		/*
410 		 * As long as the mfn_list has enough entries to completely
411 		 * fill a p2m page, pointing into the array is ok. But if
412 		 * not the entries beyond the last pfn will be undefined.
413 		 */
414 		if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) {
415 			unsigned long p2midx;
416 
417 			p2midx = max_pfn % P2M_PER_PAGE;
418 			for ( ; p2midx < P2M_PER_PAGE; p2midx++)
419 				mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY;
420 		}
421 		p2m_top[topidx][mididx] = &mfn_list[pfn];
422 	}
423 
424 	m2p_override_init();
425 }
426 #ifdef CONFIG_X86_64
427 #include <linux/bootmem.h>
428 unsigned long __init xen_revector_p2m_tree(void)
429 {
430 	unsigned long va_start;
431 	unsigned long va_end;
432 	unsigned long pfn;
433 	unsigned long pfn_free = 0;
434 	unsigned long *mfn_list = NULL;
435 	unsigned long size;
436 
437 	va_start = xen_start_info->mfn_list;
438 	/*We copy in increments of P2M_PER_PAGE * sizeof(unsigned long),
439 	 * so make sure it is rounded up to that */
440 	size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
441 	va_end = va_start + size;
442 
443 	/* If we were revectored already, don't do it again. */
444 	if (va_start <= __START_KERNEL_map && va_start >= __PAGE_OFFSET)
445 		return 0;
446 
447 	mfn_list = alloc_bootmem_align(size, PAGE_SIZE);
448 	if (!mfn_list) {
449 		pr_warn("Could not allocate space for a new P2M tree!\n");
450 		return xen_start_info->mfn_list;
451 	}
452 	/* Fill it out with INVALID_P2M_ENTRY value */
453 	memset(mfn_list, 0xFF, size);
454 
455 	for (pfn = 0; pfn < ALIGN(MAX_DOMAIN_PAGES, P2M_PER_PAGE); pfn += P2M_PER_PAGE) {
456 		unsigned topidx = p2m_top_index(pfn);
457 		unsigned mididx;
458 		unsigned long *mid_p;
459 
460 		if (!p2m_top[topidx])
461 			continue;
462 
463 		if (p2m_top[topidx] == p2m_mid_missing)
464 			continue;
465 
466 		mididx = p2m_mid_index(pfn);
467 		mid_p = p2m_top[topidx][mididx];
468 		if (!mid_p)
469 			continue;
470 		if ((mid_p == p2m_missing) || (mid_p == p2m_identity))
471 			continue;
472 
473 		if ((unsigned long)mid_p == INVALID_P2M_ENTRY)
474 			continue;
475 
476 		/* The old va. Rebase it on mfn_list */
477 		if (mid_p >= (unsigned long *)va_start && mid_p <= (unsigned long *)va_end) {
478 			unsigned long *new;
479 
480 			if (pfn_free  > (size / sizeof(unsigned long))) {
481 				WARN(1, "Only allocated for %ld pages, but we want %ld!\n",
482 				     size / sizeof(unsigned long), pfn_free);
483 				return 0;
484 			}
485 			new = &mfn_list[pfn_free];
486 
487 			copy_page(new, mid_p);
488 			p2m_top[topidx][mididx] = &mfn_list[pfn_free];
489 			p2m_top_mfn_p[topidx][mididx] = virt_to_mfn(&mfn_list[pfn_free]);
490 
491 			pfn_free += P2M_PER_PAGE;
492 
493 		}
494 		/* This should be the leafs allocated for identity from _brk. */
495 	}
496 	return (unsigned long)mfn_list;
497 
498 }
499 #else
500 unsigned long __init xen_revector_p2m_tree(void)
501 {
502 	return 0;
503 }
504 #endif
505 unsigned long get_phys_to_machine(unsigned long pfn)
506 {
507 	unsigned topidx, mididx, idx;
508 
509 	if (unlikely(pfn >= MAX_P2M_PFN))
510 		return IDENTITY_FRAME(pfn);
511 
512 	topidx = p2m_top_index(pfn);
513 	mididx = p2m_mid_index(pfn);
514 	idx = p2m_index(pfn);
515 
516 	/*
517 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
518 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
519 	 * would be wrong.
520 	 */
521 	if (p2m_top[topidx][mididx] == p2m_identity)
522 		return IDENTITY_FRAME(pfn);
523 
524 	return p2m_top[topidx][mididx][idx];
525 }
526 EXPORT_SYMBOL_GPL(get_phys_to_machine);
527 
528 static void *alloc_p2m_page(void)
529 {
530 	return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
531 }
532 
533 static void free_p2m_page(void *p)
534 {
535 	free_page((unsigned long)p);
536 }
537 
538 /*
539  * Fully allocate the p2m structure for a given pfn.  We need to check
540  * that both the top and mid levels are allocated, and make sure the
541  * parallel mfn tree is kept in sync.  We may race with other cpus, so
542  * the new pages are installed with cmpxchg; if we lose the race then
543  * simply free the page we allocated and use the one that's there.
544  */
545 static bool alloc_p2m(unsigned long pfn)
546 {
547 	unsigned topidx, mididx;
548 	unsigned long ***top_p, **mid;
549 	unsigned long *top_mfn_p, *mid_mfn;
550 
551 	topidx = p2m_top_index(pfn);
552 	mididx = p2m_mid_index(pfn);
553 
554 	top_p = &p2m_top[topidx];
555 	mid = *top_p;
556 
557 	if (mid == p2m_mid_missing) {
558 		/* Mid level is missing, allocate a new one */
559 		mid = alloc_p2m_page();
560 		if (!mid)
561 			return false;
562 
563 		p2m_mid_init(mid, p2m_missing);
564 
565 		if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing)
566 			free_p2m_page(mid);
567 	}
568 
569 	top_mfn_p = &p2m_top_mfn[topidx];
570 	mid_mfn = p2m_top_mfn_p[topidx];
571 
572 	BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
573 
574 	if (mid_mfn == p2m_mid_missing_mfn) {
575 		/* Separately check the mid mfn level */
576 		unsigned long missing_mfn;
577 		unsigned long mid_mfn_mfn;
578 
579 		mid_mfn = alloc_p2m_page();
580 		if (!mid_mfn)
581 			return false;
582 
583 		p2m_mid_mfn_init(mid_mfn, p2m_missing);
584 
585 		missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
586 		mid_mfn_mfn = virt_to_mfn(mid_mfn);
587 		if (cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn) != missing_mfn)
588 			free_p2m_page(mid_mfn);
589 		else
590 			p2m_top_mfn_p[topidx] = mid_mfn;
591 	}
592 
593 	if (p2m_top[topidx][mididx] == p2m_identity ||
594 	    p2m_top[topidx][mididx] == p2m_missing) {
595 		/* p2m leaf page is missing */
596 		unsigned long *p2m;
597 		unsigned long *p2m_orig = p2m_top[topidx][mididx];
598 
599 		p2m = alloc_p2m_page();
600 		if (!p2m)
601 			return false;
602 
603 		p2m_init(p2m);
604 
605 		if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig)
606 			free_p2m_page(p2m);
607 		else
608 			mid_mfn[mididx] = virt_to_mfn(p2m);
609 	}
610 
611 	return true;
612 }
613 
614 static bool __init early_alloc_p2m(unsigned long pfn, bool check_boundary)
615 {
616 	unsigned topidx, mididx, idx;
617 	unsigned long *p2m;
618 	unsigned long *mid_mfn_p;
619 
620 	topidx = p2m_top_index(pfn);
621 	mididx = p2m_mid_index(pfn);
622 	idx = p2m_index(pfn);
623 
624 	/* Pfff.. No boundary cross-over, lets get out. */
625 	if (!idx && check_boundary)
626 		return false;
627 
628 	WARN(p2m_top[topidx][mididx] == p2m_identity,
629 		"P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
630 		topidx, mididx);
631 
632 	/*
633 	 * Could be done by xen_build_dynamic_phys_to_machine..
634 	 */
635 	if (p2m_top[topidx][mididx] != p2m_missing)
636 		return false;
637 
638 	/* Boundary cross-over for the edges: */
639 	p2m = extend_brk(PAGE_SIZE, PAGE_SIZE);
640 
641 	p2m_init(p2m);
642 
643 	p2m_top[topidx][mididx] = p2m;
644 
645 	/* For save/restore we need to MFN of the P2M saved */
646 
647 	mid_mfn_p = p2m_top_mfn_p[topidx];
648 	WARN(mid_mfn_p[mididx] != virt_to_mfn(p2m_missing),
649 		"P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n",
650 		topidx, mididx);
651 	mid_mfn_p[mididx] = virt_to_mfn(p2m);
652 
653 	return true;
654 }
655 
656 static bool __init early_alloc_p2m_middle(unsigned long pfn)
657 {
658 	unsigned topidx = p2m_top_index(pfn);
659 	unsigned long *mid_mfn_p;
660 	unsigned long **mid;
661 
662 	mid = p2m_top[topidx];
663 	mid_mfn_p = p2m_top_mfn_p[topidx];
664 	if (mid == p2m_mid_missing) {
665 		mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
666 
667 		p2m_mid_init(mid, p2m_missing);
668 
669 		p2m_top[topidx] = mid;
670 
671 		BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
672 	}
673 	/* And the save/restore P2M tables.. */
674 	if (mid_mfn_p == p2m_mid_missing_mfn) {
675 		mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
676 		p2m_mid_mfn_init(mid_mfn_p, p2m_missing);
677 
678 		p2m_top_mfn_p[topidx] = mid_mfn_p;
679 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
680 		/* Note: we don't set mid_mfn_p[midix] here,
681 		 * look in early_alloc_p2m() */
682 	}
683 	return true;
684 }
685 
686 /*
687  * Skim over the P2M tree looking at pages that are either filled with
688  * INVALID_P2M_ENTRY or with 1:1 PFNs. If found, re-use that page and
689  * replace the P2M leaf with a p2m_missing or p2m_identity.
690  * Stick the old page in the new P2M tree location.
691  */
692 bool __init early_can_reuse_p2m_middle(unsigned long set_pfn, unsigned long set_mfn)
693 {
694 	unsigned topidx;
695 	unsigned mididx;
696 	unsigned ident_pfns;
697 	unsigned inv_pfns;
698 	unsigned long *p2m;
699 	unsigned long *mid_mfn_p;
700 	unsigned idx;
701 	unsigned long pfn;
702 
703 	/* We only look when this entails a P2M middle layer */
704 	if (p2m_index(set_pfn))
705 		return false;
706 
707 	for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_PER_PAGE) {
708 		topidx = p2m_top_index(pfn);
709 
710 		if (!p2m_top[topidx])
711 			continue;
712 
713 		if (p2m_top[topidx] == p2m_mid_missing)
714 			continue;
715 
716 		mididx = p2m_mid_index(pfn);
717 		p2m = p2m_top[topidx][mididx];
718 		if (!p2m)
719 			continue;
720 
721 		if ((p2m == p2m_missing) || (p2m == p2m_identity))
722 			continue;
723 
724 		if ((unsigned long)p2m == INVALID_P2M_ENTRY)
725 			continue;
726 
727 		ident_pfns = 0;
728 		inv_pfns = 0;
729 		for (idx = 0; idx < P2M_PER_PAGE; idx++) {
730 			/* IDENTITY_PFNs are 1:1 */
731 			if (p2m[idx] == IDENTITY_FRAME(pfn + idx))
732 				ident_pfns++;
733 			else if (p2m[idx] == INVALID_P2M_ENTRY)
734 				inv_pfns++;
735 			else
736 				break;
737 		}
738 		if ((ident_pfns == P2M_PER_PAGE) || (inv_pfns == P2M_PER_PAGE))
739 			goto found;
740 	}
741 	return false;
742 found:
743 	/* Found one, replace old with p2m_identity or p2m_missing */
744 	p2m_top[topidx][mididx] = (ident_pfns ? p2m_identity : p2m_missing);
745 	/* And the other for save/restore.. */
746 	mid_mfn_p = p2m_top_mfn_p[topidx];
747 	/* NOTE: Even if it is a p2m_identity it should still be point to
748 	 * a page filled with INVALID_P2M_ENTRY entries. */
749 	mid_mfn_p[mididx] = virt_to_mfn(p2m_missing);
750 
751 	/* Reset where we want to stick the old page in. */
752 	topidx = p2m_top_index(set_pfn);
753 	mididx = p2m_mid_index(set_pfn);
754 
755 	/* This shouldn't happen */
756 	if (WARN_ON(p2m_top[topidx] == p2m_mid_missing))
757 		early_alloc_p2m_middle(set_pfn);
758 
759 	if (WARN_ON(p2m_top[topidx][mididx] != p2m_missing))
760 		return false;
761 
762 	p2m_init(p2m);
763 	p2m_top[topidx][mididx] = p2m;
764 	mid_mfn_p = p2m_top_mfn_p[topidx];
765 	mid_mfn_p[mididx] = virt_to_mfn(p2m);
766 
767 	return true;
768 }
769 bool __init early_set_phys_to_machine(unsigned long pfn, unsigned long mfn)
770 {
771 	if (unlikely(!__set_phys_to_machine(pfn, mfn)))  {
772 		if (!early_alloc_p2m_middle(pfn))
773 			return false;
774 
775 		if (early_can_reuse_p2m_middle(pfn, mfn))
776 			return __set_phys_to_machine(pfn, mfn);
777 
778 		if (!early_alloc_p2m(pfn, false /* boundary crossover OK!*/))
779 			return false;
780 
781 		if (!__set_phys_to_machine(pfn, mfn))
782 			return false;
783 	}
784 
785 	return true;
786 }
787 
788 static void __init early_split_p2m(unsigned long pfn)
789 {
790 	unsigned long mididx, idx;
791 
792 	mididx = p2m_mid_index(pfn);
793 	idx = p2m_index(pfn);
794 
795 	/*
796 	 * Allocate new middle and leaf pages if this pfn lies in the
797 	 * middle of one.
798 	 */
799 	if (mididx || idx)
800 		early_alloc_p2m_middle(pfn);
801 	if (idx)
802 		early_alloc_p2m(pfn, false);
803 }
804 
805 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
806 				      unsigned long pfn_e)
807 {
808 	unsigned long pfn;
809 
810 	if (unlikely(pfn_s >= MAX_P2M_PFN))
811 		return 0;
812 
813 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
814 		return pfn_e - pfn_s;
815 
816 	if (pfn_s > pfn_e)
817 		return 0;
818 
819 	if (pfn_e > MAX_P2M_PFN)
820 		pfn_e = MAX_P2M_PFN;
821 
822 	early_split_p2m(pfn_s);
823 	early_split_p2m(pfn_e);
824 
825 	for (pfn = pfn_s; pfn < pfn_e;) {
826 		unsigned topidx = p2m_top_index(pfn);
827 		unsigned mididx = p2m_mid_index(pfn);
828 
829 		if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn)))
830 			break;
831 		pfn++;
832 
833 		/*
834 		 * If the PFN was set to a middle or leaf identity
835 		 * page the remainder must also be identity, so skip
836 		 * ahead to the next middle or leaf entry.
837 		 */
838 		if (p2m_top[topidx] == p2m_mid_identity)
839 			pfn = ALIGN(pfn, P2M_MID_PER_PAGE * P2M_PER_PAGE);
840 		else if (p2m_top[topidx][mididx] == p2m_identity)
841 			pfn = ALIGN(pfn, P2M_PER_PAGE);
842 	}
843 
844 	if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s),
845 		"Identity mapping failed. We are %ld short of 1-1 mappings!\n",
846 		(pfn_e - pfn_s) - (pfn - pfn_s)))
847 		printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn);
848 
849 	return pfn - pfn_s;
850 }
851 
852 /* Try to install p2m mapping; fail if intermediate bits missing */
853 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
854 {
855 	unsigned topidx, mididx, idx;
856 
857 	/* don't track P2M changes in autotranslate guests */
858 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
859 		return true;
860 
861 	if (unlikely(pfn >= MAX_P2M_PFN)) {
862 		BUG_ON(mfn != INVALID_P2M_ENTRY);
863 		return true;
864 	}
865 
866 	topidx = p2m_top_index(pfn);
867 	mididx = p2m_mid_index(pfn);
868 	idx = p2m_index(pfn);
869 
870 	/* For sparse holes were the p2m leaf has real PFN along with
871 	 * PCI holes, stick in the PFN as the MFN value.
872 	 *
873 	 * set_phys_range_identity() will have allocated new middle
874 	 * and leaf pages as required so an existing p2m_mid_missing
875 	 * or p2m_missing mean that whole range will be identity so
876 	 * these can be switched to p2m_mid_identity or p2m_identity.
877 	 */
878 	if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) {
879 		if (p2m_top[topidx] == p2m_mid_identity)
880 			return true;
881 
882 		if (p2m_top[topidx] == p2m_mid_missing) {
883 			WARN_ON(cmpxchg(&p2m_top[topidx], p2m_mid_missing,
884 					p2m_mid_identity) != p2m_mid_missing);
885 			return true;
886 		}
887 
888 		if (p2m_top[topidx][mididx] == p2m_identity)
889 			return true;
890 
891 		/* Swap over from MISSING to IDENTITY if needed. */
892 		if (p2m_top[topidx][mididx] == p2m_missing) {
893 			WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing,
894 				p2m_identity) != p2m_missing);
895 			return true;
896 		}
897 	}
898 
899 	if (p2m_top[topidx][mididx] == p2m_missing)
900 		return mfn == INVALID_P2M_ENTRY;
901 
902 	p2m_top[topidx][mididx][idx] = mfn;
903 
904 	return true;
905 }
906 
907 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
908 {
909 	if (unlikely(!__set_phys_to_machine(pfn, mfn)))  {
910 		if (!alloc_p2m(pfn))
911 			return false;
912 
913 		if (!__set_phys_to_machine(pfn, mfn))
914 			return false;
915 	}
916 
917 	return true;
918 }
919 
920 #define M2P_OVERRIDE_HASH_SHIFT	10
921 #define M2P_OVERRIDE_HASH	(1 << M2P_OVERRIDE_HASH_SHIFT)
922 
923 static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH);
924 static DEFINE_SPINLOCK(m2p_override_lock);
925 
926 static void __init m2p_override_init(void)
927 {
928 	unsigned i;
929 
930 	m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH,
931 				   sizeof(unsigned long));
932 
933 	for (i = 0; i < M2P_OVERRIDE_HASH; i++)
934 		INIT_LIST_HEAD(&m2p_overrides[i]);
935 }
936 
937 static unsigned long mfn_hash(unsigned long mfn)
938 {
939 	return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT);
940 }
941 
942 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
943 			    struct gnttab_map_grant_ref *kmap_ops,
944 			    struct page **pages, unsigned int count)
945 {
946 	int i, ret = 0;
947 	bool lazy = false;
948 	pte_t *pte;
949 
950 	if (xen_feature(XENFEAT_auto_translated_physmap))
951 		return 0;
952 
953 	if (kmap_ops &&
954 	    !in_interrupt() &&
955 	    paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE) {
956 		arch_enter_lazy_mmu_mode();
957 		lazy = true;
958 	}
959 
960 	for (i = 0; i < count; i++) {
961 		unsigned long mfn, pfn;
962 
963 		/* Do not add to override if the map failed. */
964 		if (map_ops[i].status)
965 			continue;
966 
967 		if (map_ops[i].flags & GNTMAP_contains_pte) {
968 			pte = (pte_t *) (mfn_to_virt(PFN_DOWN(map_ops[i].host_addr)) +
969 				(map_ops[i].host_addr & ~PAGE_MASK));
970 			mfn = pte_mfn(*pte);
971 		} else {
972 			mfn = PFN_DOWN(map_ops[i].dev_bus_addr);
973 		}
974 		pfn = page_to_pfn(pages[i]);
975 
976 		WARN_ON(PagePrivate(pages[i]));
977 		SetPagePrivate(pages[i]);
978 		set_page_private(pages[i], mfn);
979 		pages[i]->index = pfn_to_mfn(pfn);
980 
981 		if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) {
982 			ret = -ENOMEM;
983 			goto out;
984 		}
985 
986 		if (kmap_ops) {
987 			ret = m2p_add_override(mfn, pages[i], &kmap_ops[i]);
988 			if (ret)
989 				goto out;
990 		}
991 	}
992 
993 out:
994 	if (lazy)
995 		arch_leave_lazy_mmu_mode();
996 
997 	return ret;
998 }
999 EXPORT_SYMBOL_GPL(set_foreign_p2m_mapping);
1000 
1001 /* Add an MFN override for a particular page */
1002 int m2p_add_override(unsigned long mfn, struct page *page,
1003 		struct gnttab_map_grant_ref *kmap_op)
1004 {
1005 	unsigned long flags;
1006 	unsigned long pfn;
1007 	unsigned long uninitialized_var(address);
1008 	unsigned level;
1009 	pte_t *ptep = NULL;
1010 
1011 	pfn = page_to_pfn(page);
1012 	if (!PageHighMem(page)) {
1013 		address = (unsigned long)__va(pfn << PAGE_SHIFT);
1014 		ptep = lookup_address(address, &level);
1015 		if (WARN(ptep == NULL || level != PG_LEVEL_4K,
1016 					"m2p_add_override: pfn %lx not mapped", pfn))
1017 			return -EINVAL;
1018 	}
1019 
1020 	if (kmap_op != NULL) {
1021 		if (!PageHighMem(page)) {
1022 			struct multicall_space mcs =
1023 				xen_mc_entry(sizeof(*kmap_op));
1024 
1025 			MULTI_grant_table_op(mcs.mc,
1026 					GNTTABOP_map_grant_ref, kmap_op, 1);
1027 
1028 			xen_mc_issue(PARAVIRT_LAZY_MMU);
1029 		}
1030 	}
1031 	spin_lock_irqsave(&m2p_override_lock, flags);
1032 	list_add(&page->lru,  &m2p_overrides[mfn_hash(mfn)]);
1033 	spin_unlock_irqrestore(&m2p_override_lock, flags);
1034 
1035 	/* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
1036 	 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
1037 	 * pfn so that the following mfn_to_pfn(mfn) calls will return the
1038 	 * pfn from the m2p_override (the backend pfn) instead.
1039 	 * We need to do this because the pages shared by the frontend
1040 	 * (xen-blkfront) can be already locked (lock_page, called by
1041 	 * do_read_cache_page); when the userspace backend tries to use them
1042 	 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
1043 	 * do_blockdev_direct_IO is going to try to lock the same pages
1044 	 * again resulting in a deadlock.
1045 	 * As a side effect get_user_pages_fast might not be safe on the
1046 	 * frontend pages while they are being shared with the backend,
1047 	 * because mfn_to_pfn (that ends up being called by GUPF) will
1048 	 * return the backend pfn rather than the frontend pfn. */
1049 	pfn = mfn_to_pfn_no_overrides(mfn);
1050 	if (get_phys_to_machine(pfn) == mfn)
1051 		set_phys_to_machine(pfn, FOREIGN_FRAME(mfn));
1052 
1053 	return 0;
1054 }
1055 EXPORT_SYMBOL_GPL(m2p_add_override);
1056 
1057 int clear_foreign_p2m_mapping(struct gnttab_unmap_grant_ref *unmap_ops,
1058 			      struct gnttab_map_grant_ref *kmap_ops,
1059 			      struct page **pages, unsigned int count)
1060 {
1061 	int i, ret = 0;
1062 	bool lazy = false;
1063 
1064 	if (xen_feature(XENFEAT_auto_translated_physmap))
1065 		return 0;
1066 
1067 	if (kmap_ops &&
1068 	    !in_interrupt() &&
1069 	    paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE) {
1070 		arch_enter_lazy_mmu_mode();
1071 		lazy = true;
1072 	}
1073 
1074 	for (i = 0; i < count; i++) {
1075 		unsigned long mfn = get_phys_to_machine(page_to_pfn(pages[i]));
1076 		unsigned long pfn = page_to_pfn(pages[i]);
1077 
1078 		if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) {
1079 			ret = -EINVAL;
1080 			goto out;
1081 		}
1082 
1083 		set_page_private(pages[i], INVALID_P2M_ENTRY);
1084 		WARN_ON(!PagePrivate(pages[i]));
1085 		ClearPagePrivate(pages[i]);
1086 		set_phys_to_machine(pfn, pages[i]->index);
1087 
1088 		if (kmap_ops)
1089 			ret = m2p_remove_override(pages[i], &kmap_ops[i], mfn);
1090 		if (ret)
1091 			goto out;
1092 	}
1093 
1094 out:
1095 	if (lazy)
1096 		arch_leave_lazy_mmu_mode();
1097 	return ret;
1098 }
1099 EXPORT_SYMBOL_GPL(clear_foreign_p2m_mapping);
1100 
1101 int m2p_remove_override(struct page *page,
1102 			struct gnttab_map_grant_ref *kmap_op,
1103 			unsigned long mfn)
1104 {
1105 	unsigned long flags;
1106 	unsigned long pfn;
1107 	unsigned long uninitialized_var(address);
1108 	unsigned level;
1109 	pte_t *ptep = NULL;
1110 
1111 	pfn = page_to_pfn(page);
1112 
1113 	if (!PageHighMem(page)) {
1114 		address = (unsigned long)__va(pfn << PAGE_SHIFT);
1115 		ptep = lookup_address(address, &level);
1116 
1117 		if (WARN(ptep == NULL || level != PG_LEVEL_4K,
1118 					"m2p_remove_override: pfn %lx not mapped", pfn))
1119 			return -EINVAL;
1120 	}
1121 
1122 	spin_lock_irqsave(&m2p_override_lock, flags);
1123 	list_del(&page->lru);
1124 	spin_unlock_irqrestore(&m2p_override_lock, flags);
1125 
1126 	if (kmap_op != NULL) {
1127 		if (!PageHighMem(page)) {
1128 			struct multicall_space mcs;
1129 			struct gnttab_unmap_and_replace *unmap_op;
1130 			struct page *scratch_page = get_balloon_scratch_page();
1131 			unsigned long scratch_page_address = (unsigned long)
1132 				__va(page_to_pfn(scratch_page) << PAGE_SHIFT);
1133 
1134 			/*
1135 			 * It might be that we queued all the m2p grant table
1136 			 * hypercalls in a multicall, then m2p_remove_override
1137 			 * get called before the multicall has actually been
1138 			 * issued. In this case handle is going to -1 because
1139 			 * it hasn't been modified yet.
1140 			 */
1141 			if (kmap_op->handle == -1)
1142 				xen_mc_flush();
1143 			/*
1144 			 * Now if kmap_op->handle is negative it means that the
1145 			 * hypercall actually returned an error.
1146 			 */
1147 			if (kmap_op->handle == GNTST_general_error) {
1148 				printk(KERN_WARNING "m2p_remove_override: "
1149 						"pfn %lx mfn %lx, failed to modify kernel mappings",
1150 						pfn, mfn);
1151 				put_balloon_scratch_page();
1152 				return -1;
1153 			}
1154 
1155 			xen_mc_batch();
1156 
1157 			mcs = __xen_mc_entry(
1158 					sizeof(struct gnttab_unmap_and_replace));
1159 			unmap_op = mcs.args;
1160 			unmap_op->host_addr = kmap_op->host_addr;
1161 			unmap_op->new_addr = scratch_page_address;
1162 			unmap_op->handle = kmap_op->handle;
1163 
1164 			MULTI_grant_table_op(mcs.mc,
1165 					GNTTABOP_unmap_and_replace, unmap_op, 1);
1166 
1167 			mcs = __xen_mc_entry(0);
1168 			MULTI_update_va_mapping(mcs.mc, scratch_page_address,
1169 					pfn_pte(page_to_pfn(scratch_page),
1170 					PAGE_KERNEL_RO), 0);
1171 
1172 			xen_mc_issue(PARAVIRT_LAZY_MMU);
1173 
1174 			kmap_op->host_addr = 0;
1175 			put_balloon_scratch_page();
1176 		}
1177 	}
1178 
1179 	/* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
1180 	 * somewhere in this domain, even before being added to the
1181 	 * m2p_override (see comment above in m2p_add_override).
1182 	 * If there are no other entries in the m2p_override corresponding
1183 	 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
1184 	 * the original pfn (the one shared by the frontend): the backend
1185 	 * cannot do any IO on this page anymore because it has been
1186 	 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
1187 	 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
1188 	 * pfn again. */
1189 	mfn &= ~FOREIGN_FRAME_BIT;
1190 	pfn = mfn_to_pfn_no_overrides(mfn);
1191 	if (get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) &&
1192 			m2p_find_override(mfn) == NULL)
1193 		set_phys_to_machine(pfn, mfn);
1194 
1195 	return 0;
1196 }
1197 EXPORT_SYMBOL_GPL(m2p_remove_override);
1198 
1199 struct page *m2p_find_override(unsigned long mfn)
1200 {
1201 	unsigned long flags;
1202 	struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)];
1203 	struct page *p, *ret;
1204 
1205 	ret = NULL;
1206 
1207 	spin_lock_irqsave(&m2p_override_lock, flags);
1208 
1209 	list_for_each_entry(p, bucket, lru) {
1210 		if (page_private(p) == mfn) {
1211 			ret = p;
1212 			break;
1213 		}
1214 	}
1215 
1216 	spin_unlock_irqrestore(&m2p_override_lock, flags);
1217 
1218 	return ret;
1219 }
1220 
1221 unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn)
1222 {
1223 	struct page *p = m2p_find_override(mfn);
1224 	unsigned long ret = pfn;
1225 
1226 	if (p)
1227 		ret = page_to_pfn(p);
1228 
1229 	return ret;
1230 }
1231 EXPORT_SYMBOL_GPL(m2p_find_override_pfn);
1232 
1233 #ifdef CONFIG_XEN_DEBUG_FS
1234 #include <linux/debugfs.h>
1235 #include "debugfs.h"
1236 static int p2m_dump_show(struct seq_file *m, void *v)
1237 {
1238 	static const char * const level_name[] = { "top", "middle",
1239 						"entry", "abnormal", "error"};
1240 #define TYPE_IDENTITY 0
1241 #define TYPE_MISSING 1
1242 #define TYPE_PFN 2
1243 #define TYPE_UNKNOWN 3
1244 	static const char * const type_name[] = {
1245 				[TYPE_IDENTITY] = "identity",
1246 				[TYPE_MISSING] = "missing",
1247 				[TYPE_PFN] = "pfn",
1248 				[TYPE_UNKNOWN] = "abnormal"};
1249 	unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0;
1250 	unsigned int uninitialized_var(prev_level);
1251 	unsigned int uninitialized_var(prev_type);
1252 
1253 	if (!p2m_top)
1254 		return 0;
1255 
1256 	for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) {
1257 		unsigned topidx = p2m_top_index(pfn);
1258 		unsigned mididx = p2m_mid_index(pfn);
1259 		unsigned idx = p2m_index(pfn);
1260 		unsigned lvl, type;
1261 
1262 		lvl = 4;
1263 		type = TYPE_UNKNOWN;
1264 		if (p2m_top[topidx] == p2m_mid_missing) {
1265 			lvl = 0; type = TYPE_MISSING;
1266 		} else if (p2m_top[topidx] == NULL) {
1267 			lvl = 0; type = TYPE_UNKNOWN;
1268 		} else if (p2m_top[topidx][mididx] == NULL) {
1269 			lvl = 1; type = TYPE_UNKNOWN;
1270 		} else if (p2m_top[topidx][mididx] == p2m_identity) {
1271 			lvl = 1; type = TYPE_IDENTITY;
1272 		} else if (p2m_top[topidx][mididx] == p2m_missing) {
1273 			lvl = 1; type = TYPE_MISSING;
1274 		} else if (p2m_top[topidx][mididx][idx] == 0) {
1275 			lvl = 2; type = TYPE_UNKNOWN;
1276 		} else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) {
1277 			lvl = 2; type = TYPE_IDENTITY;
1278 		} else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) {
1279 			lvl = 2; type = TYPE_MISSING;
1280 		} else if (p2m_top[topidx][mididx][idx] == pfn) {
1281 			lvl = 2; type = TYPE_PFN;
1282 		} else if (p2m_top[topidx][mididx][idx] != pfn) {
1283 			lvl = 2; type = TYPE_PFN;
1284 		}
1285 		if (pfn == 0) {
1286 			prev_level = lvl;
1287 			prev_type = type;
1288 		}
1289 		if (pfn == MAX_DOMAIN_PAGES-1) {
1290 			lvl = 3;
1291 			type = TYPE_UNKNOWN;
1292 		}
1293 		if (prev_type != type) {
1294 			seq_printf(m, " [0x%lx->0x%lx] %s\n",
1295 				prev_pfn_type, pfn, type_name[prev_type]);
1296 			prev_pfn_type = pfn;
1297 			prev_type = type;
1298 		}
1299 		if (prev_level != lvl) {
1300 			seq_printf(m, " [0x%lx->0x%lx] level %s\n",
1301 				prev_pfn_level, pfn, level_name[prev_level]);
1302 			prev_pfn_level = pfn;
1303 			prev_level = lvl;
1304 		}
1305 	}
1306 	return 0;
1307 #undef TYPE_IDENTITY
1308 #undef TYPE_MISSING
1309 #undef TYPE_PFN
1310 #undef TYPE_UNKNOWN
1311 }
1312 
1313 static int p2m_dump_open(struct inode *inode, struct file *filp)
1314 {
1315 	return single_open(filp, p2m_dump_show, NULL);
1316 }
1317 
1318 static const struct file_operations p2m_dump_fops = {
1319 	.open		= p2m_dump_open,
1320 	.read		= seq_read,
1321 	.llseek		= seq_lseek,
1322 	.release	= single_release,
1323 };
1324 
1325 static struct dentry *d_mmu_debug;
1326 
1327 static int __init xen_p2m_debugfs(void)
1328 {
1329 	struct dentry *d_xen = xen_init_debugfs();
1330 
1331 	if (d_xen == NULL)
1332 		return -ENOMEM;
1333 
1334 	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1335 
1336 	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
1337 	return 0;
1338 }
1339 fs_initcall(xen_p2m_debugfs);
1340 #endif /* CONFIG_XEN_DEBUG_FS */
1341