xref: /linux/arch/x86/xen/p2m.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
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 b/c 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
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 loop on aligned 1GB PFNs from start pfn
81  * to end pfn.  We reserve_brk top leaf pages if they are missing (means they
82  * point to p2m_mid_missing).
83  *
84  * With the E820 example above, 263424 is not 1GB aligned so we allocate a
85  * reserve_brk page which will cover the PFNs estate from 0x40000 to 0x80000.
86  * Each entry in the allocate page is "missing" (points to p2m_missing).
87  *
88  * Next stage is to determine if we need to do a more granular boundary check
89  * on the 4MB (or 2MB depending on architecture) off the start and end pfn's.
90  * We check if the start pfn and end pfn violate that boundary check, and if
91  * so reserve_brk a middle (p2m[x][y]) leaf page. This way we have a much finer
92  * granularity of setting which PFNs are missing and which ones are identity.
93  * In our example 263424 and 512256 both fail the check so we reserve_brk two
94  * pages. Populate them with INVALID_P2M_ENTRY (so they both have "missing"
95  * values) and assign them to p2m[1][2] and p2m[1][488] respectively.
96  *
97  * At this point we would at minimum reserve_brk one page, but could be up to
98  * three. Each call to set_phys_range_identity has at maximum a three page
99  * cost. If we were to query the P2M at this stage, all those entries from
100  * start PFN through end PFN (so 1029MB -> 2001MB) would return
101  * INVALID_P2M_ENTRY ("missing").
102  *
103  * The next step is to walk from the start pfn to the end pfn setting
104  * the IDENTITY_FRAME_BIT on each PFN. This is done in set_phys_range_identity.
105  * If we find that the middle leaf is pointing to p2m_missing we can swap it
106  * over to p2m_identity - this way covering 4MB (or 2MB) PFN space.  At this
107  * 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  * This is what the p2m ends up looking (for the E820 above) with this
122  * fabulous drawing:
123  *
124  *    p2m         /--------------\
125  *  /-----\       | &mfn_list[0],|                           /-----------------\
126  *  |  0  |------>| &mfn_list[1],|    /---------------\      | ~0, ~0, ..      |
127  *  |-----|       |  ..., ~0, ~0 |    | ~0, ~0, [x]---+----->| IDENTITY [@256] |
128  *  |  1  |---\   \--------------/    | [p2m_identity]+\     | IDENTITY [@257] |
129  *  |-----|    \                      | [p2m_identity]+\\    | ....            |
130  *  |  2  |--\  \-------------------->|  ...          | \\   \----------------/
131  *  |-----|   \                       \---------------/  \\
132  *  |  3  |\   \                                          \\  p2m_identity
133  *  |-----| \   \-------------------->/---------------\   /-----------------\
134  *  | ..  +->+                        | [p2m_identity]+-->| ~0, ~0, ~0, ... |
135  *  \-----/ /                         | [p2m_identity]+-->| ..., ~0         |
136  *         / /---------------\        | ....          |   \-----------------/
137  *        /  | IDENTITY[@0]  |      /-+-[x], ~0, ~0.. |
138  *       /   | IDENTITY[@256]|<----/  \---------------/
139  *      /    | ~0, ~0, ....  |
140  *     |     \---------------/
141  *     |
142  *     p2m_missing             p2m_missing
143  * /------------------\     /------------\
144  * | [p2m_mid_missing]+---->| ~0, ~0, ~0 |
145  * | [p2m_mid_missing]+---->| ..., ~0    |
146  * \------------------/     \------------/
147  *
148  * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT)
149  */
150 
151 #include <linux/init.h>
152 #include <linux/module.h>
153 #include <linux/list.h>
154 #include <linux/hash.h>
155 #include <linux/sched.h>
156 #include <linux/seq_file.h>
157 
158 #include <asm/cache.h>
159 #include <asm/setup.h>
160 
161 #include <asm/xen/page.h>
162 #include <asm/xen/hypercall.h>
163 #include <asm/xen/hypervisor.h>
164 
165 #include "xen-ops.h"
166 
167 static void __init m2p_override_init(void);
168 
169 unsigned long xen_max_p2m_pfn __read_mostly;
170 
171 #define P2M_PER_PAGE		(PAGE_SIZE / sizeof(unsigned long))
172 #define P2M_MID_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long *))
173 #define P2M_TOP_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long **))
174 
175 #define MAX_P2M_PFN		(P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
176 
177 /* Placeholders for holes in the address space */
178 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE);
179 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE);
180 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn, P2M_MID_PER_PAGE);
181 
182 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE);
183 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, P2M_TOP_PER_PAGE);
184 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_TOP_PER_PAGE);
185 
186 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE);
187 
188 RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
189 RESERVE_BRK(p2m_mid_mfn, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
190 
191 /* We might hit two boundary violations at the start and end, at max each
192  * boundary violation will require three middle nodes. */
193 RESERVE_BRK(p2m_mid_identity, PAGE_SIZE * 2 * 3);
194 
195 static inline unsigned p2m_top_index(unsigned long pfn)
196 {
197 	BUG_ON(pfn >= MAX_P2M_PFN);
198 	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
199 }
200 
201 static inline unsigned p2m_mid_index(unsigned long pfn)
202 {
203 	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
204 }
205 
206 static inline unsigned p2m_index(unsigned long pfn)
207 {
208 	return pfn % P2M_PER_PAGE;
209 }
210 
211 static void p2m_top_init(unsigned long ***top)
212 {
213 	unsigned i;
214 
215 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
216 		top[i] = p2m_mid_missing;
217 }
218 
219 static void p2m_top_mfn_init(unsigned long *top)
220 {
221 	unsigned i;
222 
223 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
224 		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
225 }
226 
227 static void p2m_top_mfn_p_init(unsigned long **top)
228 {
229 	unsigned i;
230 
231 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
232 		top[i] = p2m_mid_missing_mfn;
233 }
234 
235 static void p2m_mid_init(unsigned long **mid)
236 {
237 	unsigned i;
238 
239 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
240 		mid[i] = p2m_missing;
241 }
242 
243 static void p2m_mid_mfn_init(unsigned long *mid)
244 {
245 	unsigned i;
246 
247 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
248 		mid[i] = virt_to_mfn(p2m_missing);
249 }
250 
251 static void p2m_init(unsigned long *p2m)
252 {
253 	unsigned i;
254 
255 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
256 		p2m[i] = INVALID_P2M_ENTRY;
257 }
258 
259 /*
260  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
261  *
262  * This is called both at boot time, and after resuming from suspend:
263  * - At boot time we're called very early, and must use extend_brk()
264  *   to allocate memory.
265  *
266  * - After resume we're called from within stop_machine, but the mfn
267  *   tree should alreay be completely allocated.
268  */
269 void __ref xen_build_mfn_list_list(void)
270 {
271 	unsigned long pfn;
272 
273 	/* Pre-initialize p2m_top_mfn to be completely missing */
274 	if (p2m_top_mfn == NULL) {
275 		p2m_mid_missing_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
276 		p2m_mid_mfn_init(p2m_mid_missing_mfn);
277 
278 		p2m_top_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
279 		p2m_top_mfn_p_init(p2m_top_mfn_p);
280 
281 		p2m_top_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
282 		p2m_top_mfn_init(p2m_top_mfn);
283 	} else {
284 		/* Reinitialise, mfn's all change after migration */
285 		p2m_mid_mfn_init(p2m_mid_missing_mfn);
286 	}
287 
288 	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) {
289 		unsigned topidx = p2m_top_index(pfn);
290 		unsigned mididx = p2m_mid_index(pfn);
291 		unsigned long **mid;
292 		unsigned long *mid_mfn_p;
293 
294 		mid = p2m_top[topidx];
295 		mid_mfn_p = p2m_top_mfn_p[topidx];
296 
297 		/* Don't bother allocating any mfn mid levels if
298 		 * they're just missing, just update the stored mfn,
299 		 * since all could have changed over a migrate.
300 		 */
301 		if (mid == p2m_mid_missing) {
302 			BUG_ON(mididx);
303 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
304 			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
305 			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
306 			continue;
307 		}
308 
309 		if (mid_mfn_p == p2m_mid_missing_mfn) {
310 			/*
311 			 * XXX boot-time only!  We should never find
312 			 * missing parts of the mfn tree after
313 			 * runtime.  extend_brk() will BUG if we call
314 			 * it too late.
315 			 */
316 			mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
317 			p2m_mid_mfn_init(mid_mfn_p);
318 
319 			p2m_top_mfn_p[topidx] = mid_mfn_p;
320 		}
321 
322 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
323 		mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]);
324 	}
325 }
326 
327 void xen_setup_mfn_list_list(void)
328 {
329 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
330 
331 	HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
332 		virt_to_mfn(p2m_top_mfn);
333 	HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
334 }
335 
336 /* Set up p2m_top to point to the domain-builder provided p2m pages */
337 void __init xen_build_dynamic_phys_to_machine(void)
338 {
339 	unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
340 	unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
341 	unsigned long pfn;
342 
343 	xen_max_p2m_pfn = max_pfn;
344 
345 	p2m_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
346 	p2m_init(p2m_missing);
347 
348 	p2m_mid_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
349 	p2m_mid_init(p2m_mid_missing);
350 
351 	p2m_top = extend_brk(PAGE_SIZE, PAGE_SIZE);
352 	p2m_top_init(p2m_top);
353 
354 	p2m_identity = extend_brk(PAGE_SIZE, PAGE_SIZE);
355 	p2m_init(p2m_identity);
356 
357 	/*
358 	 * The domain builder gives us a pre-constructed p2m array in
359 	 * mfn_list for all the pages initially given to us, so we just
360 	 * need to graft that into our tree structure.
361 	 */
362 	for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) {
363 		unsigned topidx = p2m_top_index(pfn);
364 		unsigned mididx = p2m_mid_index(pfn);
365 
366 		if (p2m_top[topidx] == p2m_mid_missing) {
367 			unsigned long **mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
368 			p2m_mid_init(mid);
369 
370 			p2m_top[topidx] = mid;
371 		}
372 
373 		/*
374 		 * As long as the mfn_list has enough entries to completely
375 		 * fill a p2m page, pointing into the array is ok. But if
376 		 * not the entries beyond the last pfn will be undefined.
377 		 */
378 		if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) {
379 			unsigned long p2midx;
380 
381 			p2midx = max_pfn % P2M_PER_PAGE;
382 			for ( ; p2midx < P2M_PER_PAGE; p2midx++)
383 				mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY;
384 		}
385 		p2m_top[topidx][mididx] = &mfn_list[pfn];
386 	}
387 
388 	m2p_override_init();
389 }
390 
391 unsigned long get_phys_to_machine(unsigned long pfn)
392 {
393 	unsigned topidx, mididx, idx;
394 
395 	if (unlikely(pfn >= MAX_P2M_PFN))
396 		return INVALID_P2M_ENTRY;
397 
398 	topidx = p2m_top_index(pfn);
399 	mididx = p2m_mid_index(pfn);
400 	idx = p2m_index(pfn);
401 
402 	/*
403 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
404 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
405 	 * would be wrong.
406 	 */
407 	if (p2m_top[topidx][mididx] == p2m_identity)
408 		return IDENTITY_FRAME(pfn);
409 
410 	return p2m_top[topidx][mididx][idx];
411 }
412 EXPORT_SYMBOL_GPL(get_phys_to_machine);
413 
414 static void *alloc_p2m_page(void)
415 {
416 	return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
417 }
418 
419 static void free_p2m_page(void *p)
420 {
421 	free_page((unsigned long)p);
422 }
423 
424 /*
425  * Fully allocate the p2m structure for a given pfn.  We need to check
426  * that both the top and mid levels are allocated, and make sure the
427  * parallel mfn tree is kept in sync.  We may race with other cpus, so
428  * the new pages are installed with cmpxchg; if we lose the race then
429  * simply free the page we allocated and use the one that's there.
430  */
431 static bool alloc_p2m(unsigned long pfn)
432 {
433 	unsigned topidx, mididx;
434 	unsigned long ***top_p, **mid;
435 	unsigned long *top_mfn_p, *mid_mfn;
436 
437 	topidx = p2m_top_index(pfn);
438 	mididx = p2m_mid_index(pfn);
439 
440 	top_p = &p2m_top[topidx];
441 	mid = *top_p;
442 
443 	if (mid == p2m_mid_missing) {
444 		/* Mid level is missing, allocate a new one */
445 		mid = alloc_p2m_page();
446 		if (!mid)
447 			return false;
448 
449 		p2m_mid_init(mid);
450 
451 		if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing)
452 			free_p2m_page(mid);
453 	}
454 
455 	top_mfn_p = &p2m_top_mfn[topidx];
456 	mid_mfn = p2m_top_mfn_p[topidx];
457 
458 	BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
459 
460 	if (mid_mfn == p2m_mid_missing_mfn) {
461 		/* Separately check the mid mfn level */
462 		unsigned long missing_mfn;
463 		unsigned long mid_mfn_mfn;
464 
465 		mid_mfn = alloc_p2m_page();
466 		if (!mid_mfn)
467 			return false;
468 
469 		p2m_mid_mfn_init(mid_mfn);
470 
471 		missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
472 		mid_mfn_mfn = virt_to_mfn(mid_mfn);
473 		if (cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn) != missing_mfn)
474 			free_p2m_page(mid_mfn);
475 		else
476 			p2m_top_mfn_p[topidx] = mid_mfn;
477 	}
478 
479 	if (p2m_top[topidx][mididx] == p2m_identity ||
480 	    p2m_top[topidx][mididx] == p2m_missing) {
481 		/* p2m leaf page is missing */
482 		unsigned long *p2m;
483 		unsigned long *p2m_orig = p2m_top[topidx][mididx];
484 
485 		p2m = alloc_p2m_page();
486 		if (!p2m)
487 			return false;
488 
489 		p2m_init(p2m);
490 
491 		if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig)
492 			free_p2m_page(p2m);
493 		else
494 			mid_mfn[mididx] = virt_to_mfn(p2m);
495 	}
496 
497 	return true;
498 }
499 
500 static bool __init __early_alloc_p2m(unsigned long pfn)
501 {
502 	unsigned topidx, mididx, idx;
503 
504 	topidx = p2m_top_index(pfn);
505 	mididx = p2m_mid_index(pfn);
506 	idx = p2m_index(pfn);
507 
508 	/* Pfff.. No boundary cross-over, lets get out. */
509 	if (!idx)
510 		return false;
511 
512 	WARN(p2m_top[topidx][mididx] == p2m_identity,
513 		"P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
514 		topidx, mididx);
515 
516 	/*
517 	 * Could be done by xen_build_dynamic_phys_to_machine..
518 	 */
519 	if (p2m_top[topidx][mididx] != p2m_missing)
520 		return false;
521 
522 	/* Boundary cross-over for the edges: */
523 	if (idx) {
524 		unsigned long *p2m = extend_brk(PAGE_SIZE, PAGE_SIZE);
525 		unsigned long *mid_mfn_p;
526 
527 		p2m_init(p2m);
528 
529 		p2m_top[topidx][mididx] = p2m;
530 
531 		/* For save/restore we need to MFN of the P2M saved */
532 
533 		mid_mfn_p = p2m_top_mfn_p[topidx];
534 		WARN(mid_mfn_p[mididx] != virt_to_mfn(p2m_missing),
535 			"P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n",
536 			topidx, mididx);
537 		mid_mfn_p[mididx] = virt_to_mfn(p2m);
538 
539 	}
540 	return idx != 0;
541 }
542 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
543 				      unsigned long pfn_e)
544 {
545 	unsigned long pfn;
546 
547 	if (unlikely(pfn_s >= MAX_P2M_PFN || pfn_e >= MAX_P2M_PFN))
548 		return 0;
549 
550 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
551 		return pfn_e - pfn_s;
552 
553 	if (pfn_s > pfn_e)
554 		return 0;
555 
556 	for (pfn = (pfn_s & ~(P2M_MID_PER_PAGE * P2M_PER_PAGE - 1));
557 		pfn < ALIGN(pfn_e, (P2M_MID_PER_PAGE * P2M_PER_PAGE));
558 		pfn += P2M_MID_PER_PAGE * P2M_PER_PAGE)
559 	{
560 		unsigned topidx = p2m_top_index(pfn);
561 		unsigned long *mid_mfn_p;
562 		unsigned long **mid;
563 
564 		mid = p2m_top[topidx];
565 		mid_mfn_p = p2m_top_mfn_p[topidx];
566 		if (mid == p2m_mid_missing) {
567 			mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
568 
569 			p2m_mid_init(mid);
570 
571 			p2m_top[topidx] = mid;
572 
573 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
574 		}
575 		/* And the save/restore P2M tables.. */
576 		if (mid_mfn_p == p2m_mid_missing_mfn) {
577 			mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
578 			p2m_mid_mfn_init(mid_mfn_p);
579 
580 			p2m_top_mfn_p[topidx] = mid_mfn_p;
581 			p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
582 			/* Note: we don't set mid_mfn_p[midix] here,
583 		 	 * look in __early_alloc_p2m */
584 		}
585 	}
586 
587 	__early_alloc_p2m(pfn_s);
588 	__early_alloc_p2m(pfn_e);
589 
590 	for (pfn = pfn_s; pfn < pfn_e; pfn++)
591 		if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn)))
592 			break;
593 
594 	if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s),
595 		"Identity mapping failed. We are %ld short of 1-1 mappings!\n",
596 		(pfn_e - pfn_s) - (pfn - pfn_s)))
597 		printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn);
598 
599 	return pfn - pfn_s;
600 }
601 
602 /* Try to install p2m mapping; fail if intermediate bits missing */
603 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
604 {
605 	unsigned topidx, mididx, idx;
606 
607 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
608 		BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
609 		return true;
610 	}
611 	if (unlikely(pfn >= MAX_P2M_PFN)) {
612 		BUG_ON(mfn != INVALID_P2M_ENTRY);
613 		return true;
614 	}
615 
616 	topidx = p2m_top_index(pfn);
617 	mididx = p2m_mid_index(pfn);
618 	idx = p2m_index(pfn);
619 
620 	/* For sparse holes were the p2m leaf has real PFN along with
621 	 * PCI holes, stick in the PFN as the MFN value.
622 	 */
623 	if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) {
624 		if (p2m_top[topidx][mididx] == p2m_identity)
625 			return true;
626 
627 		/* Swap over from MISSING to IDENTITY if needed. */
628 		if (p2m_top[topidx][mididx] == p2m_missing) {
629 			WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing,
630 				p2m_identity) != p2m_missing);
631 			return true;
632 		}
633 	}
634 
635 	if (p2m_top[topidx][mididx] == p2m_missing)
636 		return mfn == INVALID_P2M_ENTRY;
637 
638 	p2m_top[topidx][mididx][idx] = mfn;
639 
640 	return true;
641 }
642 
643 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
644 {
645 	if (unlikely(!__set_phys_to_machine(pfn, mfn)))  {
646 		if (!alloc_p2m(pfn))
647 			return false;
648 
649 		if (!__set_phys_to_machine(pfn, mfn))
650 			return false;
651 	}
652 
653 	return true;
654 }
655 
656 #define M2P_OVERRIDE_HASH_SHIFT	10
657 #define M2P_OVERRIDE_HASH	(1 << M2P_OVERRIDE_HASH_SHIFT)
658 
659 static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH);
660 static DEFINE_SPINLOCK(m2p_override_lock);
661 
662 static void __init m2p_override_init(void)
663 {
664 	unsigned i;
665 
666 	m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH,
667 				   sizeof(unsigned long));
668 
669 	for (i = 0; i < M2P_OVERRIDE_HASH; i++)
670 		INIT_LIST_HEAD(&m2p_overrides[i]);
671 }
672 
673 static unsigned long mfn_hash(unsigned long mfn)
674 {
675 	return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT);
676 }
677 
678 /* Add an MFN override for a particular page */
679 int m2p_add_override(unsigned long mfn, struct page *page, bool clear_pte)
680 {
681 	unsigned long flags;
682 	unsigned long pfn;
683 	unsigned long uninitialized_var(address);
684 	unsigned level;
685 	pte_t *ptep = NULL;
686 
687 	pfn = page_to_pfn(page);
688 	if (!PageHighMem(page)) {
689 		address = (unsigned long)__va(pfn << PAGE_SHIFT);
690 		ptep = lookup_address(address, &level);
691 		if (WARN(ptep == NULL || level != PG_LEVEL_4K,
692 					"m2p_add_override: pfn %lx not mapped", pfn))
693 			return -EINVAL;
694 	}
695 
696 	page->private = mfn;
697 	page->index = pfn_to_mfn(pfn);
698 
699 	if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn))))
700 		return -ENOMEM;
701 
702 	if (clear_pte && !PageHighMem(page))
703 		/* Just zap old mapping for now */
704 		pte_clear(&init_mm, address, ptep);
705 	spin_lock_irqsave(&m2p_override_lock, flags);
706 	list_add(&page->lru,  &m2p_overrides[mfn_hash(mfn)]);
707 	spin_unlock_irqrestore(&m2p_override_lock, flags);
708 
709 	return 0;
710 }
711 EXPORT_SYMBOL_GPL(m2p_add_override);
712 int m2p_remove_override(struct page *page, bool clear_pte)
713 {
714 	unsigned long flags;
715 	unsigned long mfn;
716 	unsigned long pfn;
717 	unsigned long uninitialized_var(address);
718 	unsigned level;
719 	pte_t *ptep = NULL;
720 
721 	pfn = page_to_pfn(page);
722 	mfn = get_phys_to_machine(pfn);
723 	if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT))
724 		return -EINVAL;
725 
726 	if (!PageHighMem(page)) {
727 		address = (unsigned long)__va(pfn << PAGE_SHIFT);
728 		ptep = lookup_address(address, &level);
729 
730 		if (WARN(ptep == NULL || level != PG_LEVEL_4K,
731 					"m2p_remove_override: pfn %lx not mapped", pfn))
732 			return -EINVAL;
733 	}
734 
735 	spin_lock_irqsave(&m2p_override_lock, flags);
736 	list_del(&page->lru);
737 	spin_unlock_irqrestore(&m2p_override_lock, flags);
738 	set_phys_to_machine(pfn, page->index);
739 
740 	if (clear_pte && !PageHighMem(page))
741 		set_pte_at(&init_mm, address, ptep,
742 				pfn_pte(pfn, PAGE_KERNEL));
743 		/* No tlb flush necessary because the caller already
744 		 * left the pte unmapped. */
745 
746 	return 0;
747 }
748 EXPORT_SYMBOL_GPL(m2p_remove_override);
749 
750 struct page *m2p_find_override(unsigned long mfn)
751 {
752 	unsigned long flags;
753 	struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)];
754 	struct page *p, *ret;
755 
756 	ret = NULL;
757 
758 	spin_lock_irqsave(&m2p_override_lock, flags);
759 
760 	list_for_each_entry(p, bucket, lru) {
761 		if (p->private == mfn) {
762 			ret = p;
763 			break;
764 		}
765 	}
766 
767 	spin_unlock_irqrestore(&m2p_override_lock, flags);
768 
769 	return ret;
770 }
771 
772 unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn)
773 {
774 	struct page *p = m2p_find_override(mfn);
775 	unsigned long ret = pfn;
776 
777 	if (p)
778 		ret = page_to_pfn(p);
779 
780 	return ret;
781 }
782 EXPORT_SYMBOL_GPL(m2p_find_override_pfn);
783 
784 #ifdef CONFIG_XEN_DEBUG_FS
785 
786 int p2m_dump_show(struct seq_file *m, void *v)
787 {
788 	static const char * const level_name[] = { "top", "middle",
789 						"entry", "abnormal" };
790 	static const char * const type_name[] = { "identity", "missing",
791 						"pfn", "abnormal"};
792 #define TYPE_IDENTITY 0
793 #define TYPE_MISSING 1
794 #define TYPE_PFN 2
795 #define TYPE_UNKNOWN 3
796 	unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0;
797 	unsigned int uninitialized_var(prev_level);
798 	unsigned int uninitialized_var(prev_type);
799 
800 	if (!p2m_top)
801 		return 0;
802 
803 	for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) {
804 		unsigned topidx = p2m_top_index(pfn);
805 		unsigned mididx = p2m_mid_index(pfn);
806 		unsigned idx = p2m_index(pfn);
807 		unsigned lvl, type;
808 
809 		lvl = 4;
810 		type = TYPE_UNKNOWN;
811 		if (p2m_top[topidx] == p2m_mid_missing) {
812 			lvl = 0; type = TYPE_MISSING;
813 		} else if (p2m_top[topidx] == NULL) {
814 			lvl = 0; type = TYPE_UNKNOWN;
815 		} else if (p2m_top[topidx][mididx] == NULL) {
816 			lvl = 1; type = TYPE_UNKNOWN;
817 		} else if (p2m_top[topidx][mididx] == p2m_identity) {
818 			lvl = 1; type = TYPE_IDENTITY;
819 		} else if (p2m_top[topidx][mididx] == p2m_missing) {
820 			lvl = 1; type = TYPE_MISSING;
821 		} else if (p2m_top[topidx][mididx][idx] == 0) {
822 			lvl = 2; type = TYPE_UNKNOWN;
823 		} else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) {
824 			lvl = 2; type = TYPE_IDENTITY;
825 		} else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) {
826 			lvl = 2; type = TYPE_MISSING;
827 		} else if (p2m_top[topidx][mididx][idx] == pfn) {
828 			lvl = 2; type = TYPE_PFN;
829 		} else if (p2m_top[topidx][mididx][idx] != pfn) {
830 			lvl = 2; type = TYPE_PFN;
831 		}
832 		if (pfn == 0) {
833 			prev_level = lvl;
834 			prev_type = type;
835 		}
836 		if (pfn == MAX_DOMAIN_PAGES-1) {
837 			lvl = 3;
838 			type = TYPE_UNKNOWN;
839 		}
840 		if (prev_type != type) {
841 			seq_printf(m, " [0x%lx->0x%lx] %s\n",
842 				prev_pfn_type, pfn, type_name[prev_type]);
843 			prev_pfn_type = pfn;
844 			prev_type = type;
845 		}
846 		if (prev_level != lvl) {
847 			seq_printf(m, " [0x%lx->0x%lx] level %s\n",
848 				prev_pfn_level, pfn, level_name[prev_level]);
849 			prev_pfn_level = pfn;
850 			prev_level = lvl;
851 		}
852 	}
853 	return 0;
854 #undef TYPE_IDENTITY
855 #undef TYPE_MISSING
856 #undef TYPE_PFN
857 #undef TYPE_UNKNOWN
858 }
859 #endif
860