xref: /linux/arch/x86/xen/p2m.c (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
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 #include <xen/grant_table.h>
165 
166 #include "multicalls.h"
167 #include "xen-ops.h"
168 
169 static void __init m2p_override_init(void);
170 
171 unsigned long xen_max_p2m_pfn __read_mostly;
172 
173 #define P2M_PER_PAGE		(PAGE_SIZE / sizeof(unsigned long))
174 #define P2M_MID_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long *))
175 #define P2M_TOP_PER_PAGE	(PAGE_SIZE / sizeof(unsigned long **))
176 
177 #define MAX_P2M_PFN		(P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE)
178 
179 /* Placeholders for holes in the address space */
180 static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE);
181 static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE);
182 static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn, P2M_MID_PER_PAGE);
183 
184 static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE);
185 static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, P2M_TOP_PER_PAGE);
186 static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_TOP_PER_PAGE);
187 
188 static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE);
189 
190 RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
191 RESERVE_BRK(p2m_mid_mfn, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
192 
193 /* We might hit two boundary violations at the start and end, at max each
194  * boundary violation will require three middle nodes. */
195 RESERVE_BRK(p2m_mid_identity, PAGE_SIZE * 2 * 3);
196 
197 static inline unsigned p2m_top_index(unsigned long pfn)
198 {
199 	BUG_ON(pfn >= MAX_P2M_PFN);
200 	return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
201 }
202 
203 static inline unsigned p2m_mid_index(unsigned long pfn)
204 {
205 	return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
206 }
207 
208 static inline unsigned p2m_index(unsigned long pfn)
209 {
210 	return pfn % P2M_PER_PAGE;
211 }
212 
213 static void p2m_top_init(unsigned long ***top)
214 {
215 	unsigned i;
216 
217 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
218 		top[i] = p2m_mid_missing;
219 }
220 
221 static void p2m_top_mfn_init(unsigned long *top)
222 {
223 	unsigned i;
224 
225 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
226 		top[i] = virt_to_mfn(p2m_mid_missing_mfn);
227 }
228 
229 static void p2m_top_mfn_p_init(unsigned long **top)
230 {
231 	unsigned i;
232 
233 	for (i = 0; i < P2M_TOP_PER_PAGE; i++)
234 		top[i] = p2m_mid_missing_mfn;
235 }
236 
237 static void p2m_mid_init(unsigned long **mid)
238 {
239 	unsigned i;
240 
241 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
242 		mid[i] = p2m_missing;
243 }
244 
245 static void p2m_mid_mfn_init(unsigned long *mid)
246 {
247 	unsigned i;
248 
249 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
250 		mid[i] = virt_to_mfn(p2m_missing);
251 }
252 
253 static void p2m_init(unsigned long *p2m)
254 {
255 	unsigned i;
256 
257 	for (i = 0; i < P2M_MID_PER_PAGE; i++)
258 		p2m[i] = INVALID_P2M_ENTRY;
259 }
260 
261 /*
262  * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
263  *
264  * This is called both at boot time, and after resuming from suspend:
265  * - At boot time we're called very early, and must use extend_brk()
266  *   to allocate memory.
267  *
268  * - After resume we're called from within stop_machine, but the mfn
269  *   tree should alreay be completely allocated.
270  */
271 void __ref xen_build_mfn_list_list(void)
272 {
273 	unsigned long pfn;
274 
275 	/* Pre-initialize p2m_top_mfn to be completely missing */
276 	if (p2m_top_mfn == NULL) {
277 		p2m_mid_missing_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
278 		p2m_mid_mfn_init(p2m_mid_missing_mfn);
279 
280 		p2m_top_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
281 		p2m_top_mfn_p_init(p2m_top_mfn_p);
282 
283 		p2m_top_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
284 		p2m_top_mfn_init(p2m_top_mfn);
285 	} else {
286 		/* Reinitialise, mfn's all change after migration */
287 		p2m_mid_mfn_init(p2m_mid_missing_mfn);
288 	}
289 
290 	for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) {
291 		unsigned topidx = p2m_top_index(pfn);
292 		unsigned mididx = p2m_mid_index(pfn);
293 		unsigned long **mid;
294 		unsigned long *mid_mfn_p;
295 
296 		mid = p2m_top[topidx];
297 		mid_mfn_p = p2m_top_mfn_p[topidx];
298 
299 		/* Don't bother allocating any mfn mid levels if
300 		 * they're just missing, just update the stored mfn,
301 		 * since all could have changed over a migrate.
302 		 */
303 		if (mid == p2m_mid_missing) {
304 			BUG_ON(mididx);
305 			BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
306 			p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
307 			pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
308 			continue;
309 		}
310 
311 		if (mid_mfn_p == p2m_mid_missing_mfn) {
312 			/*
313 			 * XXX boot-time only!  We should never find
314 			 * missing parts of the mfn tree after
315 			 * runtime.  extend_brk() will BUG if we call
316 			 * it too late.
317 			 */
318 			mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
319 			p2m_mid_mfn_init(mid_mfn_p);
320 
321 			p2m_top_mfn_p[topidx] = mid_mfn_p;
322 		}
323 
324 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
325 		mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]);
326 	}
327 }
328 
329 void xen_setup_mfn_list_list(void)
330 {
331 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
332 
333 	HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
334 		virt_to_mfn(p2m_top_mfn);
335 	HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
336 }
337 
338 /* Set up p2m_top to point to the domain-builder provided p2m pages */
339 void __init xen_build_dynamic_phys_to_machine(void)
340 {
341 	unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
342 	unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
343 	unsigned long pfn;
344 
345 	xen_max_p2m_pfn = max_pfn;
346 
347 	p2m_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
348 	p2m_init(p2m_missing);
349 
350 	p2m_mid_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
351 	p2m_mid_init(p2m_mid_missing);
352 
353 	p2m_top = extend_brk(PAGE_SIZE, PAGE_SIZE);
354 	p2m_top_init(p2m_top);
355 
356 	p2m_identity = extend_brk(PAGE_SIZE, PAGE_SIZE);
357 	p2m_init(p2m_identity);
358 
359 	/*
360 	 * The domain builder gives us a pre-constructed p2m array in
361 	 * mfn_list for all the pages initially given to us, so we just
362 	 * need to graft that into our tree structure.
363 	 */
364 	for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) {
365 		unsigned topidx = p2m_top_index(pfn);
366 		unsigned mididx = p2m_mid_index(pfn);
367 
368 		if (p2m_top[topidx] == p2m_mid_missing) {
369 			unsigned long **mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
370 			p2m_mid_init(mid);
371 
372 			p2m_top[topidx] = mid;
373 		}
374 
375 		/*
376 		 * As long as the mfn_list has enough entries to completely
377 		 * fill a p2m page, pointing into the array is ok. But if
378 		 * not the entries beyond the last pfn will be undefined.
379 		 */
380 		if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) {
381 			unsigned long p2midx;
382 
383 			p2midx = max_pfn % P2M_PER_PAGE;
384 			for ( ; p2midx < P2M_PER_PAGE; p2midx++)
385 				mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY;
386 		}
387 		p2m_top[topidx][mididx] = &mfn_list[pfn];
388 	}
389 
390 	m2p_override_init();
391 }
392 
393 unsigned long get_phys_to_machine(unsigned long pfn)
394 {
395 	unsigned topidx, mididx, idx;
396 
397 	if (unlikely(pfn >= MAX_P2M_PFN))
398 		return INVALID_P2M_ENTRY;
399 
400 	topidx = p2m_top_index(pfn);
401 	mididx = p2m_mid_index(pfn);
402 	idx = p2m_index(pfn);
403 
404 	/*
405 	 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
406 	 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
407 	 * would be wrong.
408 	 */
409 	if (p2m_top[topidx][mididx] == p2m_identity)
410 		return IDENTITY_FRAME(pfn);
411 
412 	return p2m_top[topidx][mididx][idx];
413 }
414 EXPORT_SYMBOL_GPL(get_phys_to_machine);
415 
416 static void *alloc_p2m_page(void)
417 {
418 	return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
419 }
420 
421 static void free_p2m_page(void *p)
422 {
423 	free_page((unsigned long)p);
424 }
425 
426 /*
427  * Fully allocate the p2m structure for a given pfn.  We need to check
428  * that both the top and mid levels are allocated, and make sure the
429  * parallel mfn tree is kept in sync.  We may race with other cpus, so
430  * the new pages are installed with cmpxchg; if we lose the race then
431  * simply free the page we allocated and use the one that's there.
432  */
433 static bool alloc_p2m(unsigned long pfn)
434 {
435 	unsigned topidx, mididx;
436 	unsigned long ***top_p, **mid;
437 	unsigned long *top_mfn_p, *mid_mfn;
438 
439 	topidx = p2m_top_index(pfn);
440 	mididx = p2m_mid_index(pfn);
441 
442 	top_p = &p2m_top[topidx];
443 	mid = *top_p;
444 
445 	if (mid == p2m_mid_missing) {
446 		/* Mid level is missing, allocate a new one */
447 		mid = alloc_p2m_page();
448 		if (!mid)
449 			return false;
450 
451 		p2m_mid_init(mid);
452 
453 		if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing)
454 			free_p2m_page(mid);
455 	}
456 
457 	top_mfn_p = &p2m_top_mfn[topidx];
458 	mid_mfn = p2m_top_mfn_p[topidx];
459 
460 	BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
461 
462 	if (mid_mfn == p2m_mid_missing_mfn) {
463 		/* Separately check the mid mfn level */
464 		unsigned long missing_mfn;
465 		unsigned long mid_mfn_mfn;
466 
467 		mid_mfn = alloc_p2m_page();
468 		if (!mid_mfn)
469 			return false;
470 
471 		p2m_mid_mfn_init(mid_mfn);
472 
473 		missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
474 		mid_mfn_mfn = virt_to_mfn(mid_mfn);
475 		if (cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn) != missing_mfn)
476 			free_p2m_page(mid_mfn);
477 		else
478 			p2m_top_mfn_p[topidx] = mid_mfn;
479 	}
480 
481 	if (p2m_top[topidx][mididx] == p2m_identity ||
482 	    p2m_top[topidx][mididx] == p2m_missing) {
483 		/* p2m leaf page is missing */
484 		unsigned long *p2m;
485 		unsigned long *p2m_orig = p2m_top[topidx][mididx];
486 
487 		p2m = alloc_p2m_page();
488 		if (!p2m)
489 			return false;
490 
491 		p2m_init(p2m);
492 
493 		if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig)
494 			free_p2m_page(p2m);
495 		else
496 			mid_mfn[mididx] = virt_to_mfn(p2m);
497 	}
498 
499 	return true;
500 }
501 
502 static bool __init early_alloc_p2m_middle(unsigned long pfn, bool check_boundary)
503 {
504 	unsigned topidx, mididx, idx;
505 	unsigned long *p2m;
506 	unsigned long *mid_mfn_p;
507 
508 	topidx = p2m_top_index(pfn);
509 	mididx = p2m_mid_index(pfn);
510 	idx = p2m_index(pfn);
511 
512 	/* Pfff.. No boundary cross-over, lets get out. */
513 	if (!idx && check_boundary)
514 		return false;
515 
516 	WARN(p2m_top[topidx][mididx] == p2m_identity,
517 		"P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
518 		topidx, mididx);
519 
520 	/*
521 	 * Could be done by xen_build_dynamic_phys_to_machine..
522 	 */
523 	if (p2m_top[topidx][mididx] != p2m_missing)
524 		return false;
525 
526 	/* Boundary cross-over for the edges: */
527 	p2m = extend_brk(PAGE_SIZE, PAGE_SIZE);
528 
529 	p2m_init(p2m);
530 
531 	p2m_top[topidx][mididx] = p2m;
532 
533 	/* For save/restore we need to MFN of the P2M saved */
534 
535 	mid_mfn_p = p2m_top_mfn_p[topidx];
536 	WARN(mid_mfn_p[mididx] != virt_to_mfn(p2m_missing),
537 		"P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n",
538 		topidx, mididx);
539 	mid_mfn_p[mididx] = virt_to_mfn(p2m);
540 
541 	return true;
542 }
543 
544 static bool __init early_alloc_p2m(unsigned long pfn)
545 {
546 	unsigned topidx = p2m_top_index(pfn);
547 	unsigned long *mid_mfn_p;
548 	unsigned long **mid;
549 
550 	mid = p2m_top[topidx];
551 	mid_mfn_p = p2m_top_mfn_p[topidx];
552 	if (mid == p2m_mid_missing) {
553 		mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
554 
555 		p2m_mid_init(mid);
556 
557 		p2m_top[topidx] = mid;
558 
559 		BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
560 	}
561 	/* And the save/restore P2M tables.. */
562 	if (mid_mfn_p == p2m_mid_missing_mfn) {
563 		mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
564 		p2m_mid_mfn_init(mid_mfn_p);
565 
566 		p2m_top_mfn_p[topidx] = mid_mfn_p;
567 		p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
568 		/* Note: we don't set mid_mfn_p[midix] here,
569 		 * look in early_alloc_p2m_middle */
570 	}
571 	return true;
572 }
573 bool __init early_set_phys_to_machine(unsigned long pfn, unsigned long mfn)
574 {
575 	if (unlikely(!__set_phys_to_machine(pfn, mfn)))  {
576 		if (!early_alloc_p2m(pfn))
577 			return false;
578 
579 		if (!early_alloc_p2m_middle(pfn, false /* boundary crossover OK!*/))
580 			return false;
581 
582 		if (!__set_phys_to_machine(pfn, mfn))
583 			return false;
584 	}
585 
586 	return true;
587 }
588 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
589 				      unsigned long pfn_e)
590 {
591 	unsigned long pfn;
592 
593 	if (unlikely(pfn_s >= MAX_P2M_PFN || pfn_e >= MAX_P2M_PFN))
594 		return 0;
595 
596 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
597 		return pfn_e - pfn_s;
598 
599 	if (pfn_s > pfn_e)
600 		return 0;
601 
602 	for (pfn = (pfn_s & ~(P2M_MID_PER_PAGE * P2M_PER_PAGE - 1));
603 		pfn < ALIGN(pfn_e, (P2M_MID_PER_PAGE * P2M_PER_PAGE));
604 		pfn += P2M_MID_PER_PAGE * P2M_PER_PAGE)
605 	{
606 		WARN_ON(!early_alloc_p2m(pfn));
607 	}
608 
609 	early_alloc_p2m_middle(pfn_s, true);
610 	early_alloc_p2m_middle(pfn_e, true);
611 
612 	for (pfn = pfn_s; pfn < pfn_e; pfn++)
613 		if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn)))
614 			break;
615 
616 	if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s),
617 		"Identity mapping failed. We are %ld short of 1-1 mappings!\n",
618 		(pfn_e - pfn_s) - (pfn - pfn_s)))
619 		printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn);
620 
621 	return pfn - pfn_s;
622 }
623 
624 /* Try to install p2m mapping; fail if intermediate bits missing */
625 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
626 {
627 	unsigned topidx, mididx, idx;
628 
629 	if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
630 		BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
631 		return true;
632 	}
633 	if (unlikely(pfn >= MAX_P2M_PFN)) {
634 		BUG_ON(mfn != INVALID_P2M_ENTRY);
635 		return true;
636 	}
637 
638 	topidx = p2m_top_index(pfn);
639 	mididx = p2m_mid_index(pfn);
640 	idx = p2m_index(pfn);
641 
642 	/* For sparse holes were the p2m leaf has real PFN along with
643 	 * PCI holes, stick in the PFN as the MFN value.
644 	 */
645 	if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) {
646 		if (p2m_top[topidx][mididx] == p2m_identity)
647 			return true;
648 
649 		/* Swap over from MISSING to IDENTITY if needed. */
650 		if (p2m_top[topidx][mididx] == p2m_missing) {
651 			WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing,
652 				p2m_identity) != p2m_missing);
653 			return true;
654 		}
655 	}
656 
657 	if (p2m_top[topidx][mididx] == p2m_missing)
658 		return mfn == INVALID_P2M_ENTRY;
659 
660 	p2m_top[topidx][mididx][idx] = mfn;
661 
662 	return true;
663 }
664 
665 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
666 {
667 	if (unlikely(!__set_phys_to_machine(pfn, mfn)))  {
668 		if (!alloc_p2m(pfn))
669 			return false;
670 
671 		if (!__set_phys_to_machine(pfn, mfn))
672 			return false;
673 	}
674 
675 	return true;
676 }
677 
678 #define M2P_OVERRIDE_HASH_SHIFT	10
679 #define M2P_OVERRIDE_HASH	(1 << M2P_OVERRIDE_HASH_SHIFT)
680 
681 static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH);
682 static DEFINE_SPINLOCK(m2p_override_lock);
683 
684 static void __init m2p_override_init(void)
685 {
686 	unsigned i;
687 
688 	m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH,
689 				   sizeof(unsigned long));
690 
691 	for (i = 0; i < M2P_OVERRIDE_HASH; i++)
692 		INIT_LIST_HEAD(&m2p_overrides[i]);
693 }
694 
695 static unsigned long mfn_hash(unsigned long mfn)
696 {
697 	return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT);
698 }
699 
700 /* Add an MFN override for a particular page */
701 int m2p_add_override(unsigned long mfn, struct page *page,
702 		struct gnttab_map_grant_ref *kmap_op)
703 {
704 	unsigned long flags;
705 	unsigned long pfn;
706 	unsigned long uninitialized_var(address);
707 	unsigned level;
708 	pte_t *ptep = NULL;
709 	int ret = 0;
710 
711 	pfn = page_to_pfn(page);
712 	if (!PageHighMem(page)) {
713 		address = (unsigned long)__va(pfn << PAGE_SHIFT);
714 		ptep = lookup_address(address, &level);
715 		if (WARN(ptep == NULL || level != PG_LEVEL_4K,
716 					"m2p_add_override: pfn %lx not mapped", pfn))
717 			return -EINVAL;
718 	}
719 	WARN_ON(PagePrivate(page));
720 	SetPagePrivate(page);
721 	set_page_private(page, mfn);
722 	page->index = pfn_to_mfn(pfn);
723 
724 	if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn))))
725 		return -ENOMEM;
726 
727 	if (kmap_op != NULL) {
728 		if (!PageHighMem(page)) {
729 			struct multicall_space mcs =
730 				xen_mc_entry(sizeof(*kmap_op));
731 
732 			MULTI_grant_table_op(mcs.mc,
733 					GNTTABOP_map_grant_ref, kmap_op, 1);
734 
735 			xen_mc_issue(PARAVIRT_LAZY_MMU);
736 		}
737 		/* let's use dev_bus_addr to record the old mfn instead */
738 		kmap_op->dev_bus_addr = page->index;
739 		page->index = (unsigned long) kmap_op;
740 	}
741 	spin_lock_irqsave(&m2p_override_lock, flags);
742 	list_add(&page->lru,  &m2p_overrides[mfn_hash(mfn)]);
743 	spin_unlock_irqrestore(&m2p_override_lock, flags);
744 
745 	/* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
746 	 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
747 	 * pfn so that the following mfn_to_pfn(mfn) calls will return the
748 	 * pfn from the m2p_override (the backend pfn) instead.
749 	 * We need to do this because the pages shared by the frontend
750 	 * (xen-blkfront) can be already locked (lock_page, called by
751 	 * do_read_cache_page); when the userspace backend tries to use them
752 	 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
753 	 * do_blockdev_direct_IO is going to try to lock the same pages
754 	 * again resulting in a deadlock.
755 	 * As a side effect get_user_pages_fast might not be safe on the
756 	 * frontend pages while they are being shared with the backend,
757 	 * because mfn_to_pfn (that ends up being called by GUPF) will
758 	 * return the backend pfn rather than the frontend pfn. */
759 	ret = __get_user(pfn, &machine_to_phys_mapping[mfn]);
760 	if (ret == 0 && get_phys_to_machine(pfn) == mfn)
761 		set_phys_to_machine(pfn, FOREIGN_FRAME(mfn));
762 
763 	return 0;
764 }
765 EXPORT_SYMBOL_GPL(m2p_add_override);
766 int m2p_remove_override(struct page *page, bool clear_pte)
767 {
768 	unsigned long flags;
769 	unsigned long mfn;
770 	unsigned long pfn;
771 	unsigned long uninitialized_var(address);
772 	unsigned level;
773 	pte_t *ptep = NULL;
774 	int ret = 0;
775 
776 	pfn = page_to_pfn(page);
777 	mfn = get_phys_to_machine(pfn);
778 	if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT))
779 		return -EINVAL;
780 
781 	if (!PageHighMem(page)) {
782 		address = (unsigned long)__va(pfn << PAGE_SHIFT);
783 		ptep = lookup_address(address, &level);
784 
785 		if (WARN(ptep == NULL || level != PG_LEVEL_4K,
786 					"m2p_remove_override: pfn %lx not mapped", pfn))
787 			return -EINVAL;
788 	}
789 
790 	spin_lock_irqsave(&m2p_override_lock, flags);
791 	list_del(&page->lru);
792 	spin_unlock_irqrestore(&m2p_override_lock, flags);
793 	WARN_ON(!PagePrivate(page));
794 	ClearPagePrivate(page);
795 
796 	if (clear_pte) {
797 		struct gnttab_map_grant_ref *map_op =
798 			(struct gnttab_map_grant_ref *) page->index;
799 		set_phys_to_machine(pfn, map_op->dev_bus_addr);
800 		if (!PageHighMem(page)) {
801 			struct multicall_space mcs;
802 			struct gnttab_unmap_grant_ref *unmap_op;
803 
804 			/*
805 			 * It might be that we queued all the m2p grant table
806 			 * hypercalls in a multicall, then m2p_remove_override
807 			 * get called before the multicall has actually been
808 			 * issued. In this case handle is going to -1 because
809 			 * it hasn't been modified yet.
810 			 */
811 			if (map_op->handle == -1)
812 				xen_mc_flush();
813 			/*
814 			 * Now if map_op->handle is negative it means that the
815 			 * hypercall actually returned an error.
816 			 */
817 			if (map_op->handle == GNTST_general_error) {
818 				printk(KERN_WARNING "m2p_remove_override: "
819 						"pfn %lx mfn %lx, failed to modify kernel mappings",
820 						pfn, mfn);
821 				return -1;
822 			}
823 
824 			mcs = xen_mc_entry(
825 					sizeof(struct gnttab_unmap_grant_ref));
826 			unmap_op = mcs.args;
827 			unmap_op->host_addr = map_op->host_addr;
828 			unmap_op->handle = map_op->handle;
829 			unmap_op->dev_bus_addr = 0;
830 
831 			MULTI_grant_table_op(mcs.mc,
832 					GNTTABOP_unmap_grant_ref, unmap_op, 1);
833 
834 			xen_mc_issue(PARAVIRT_LAZY_MMU);
835 
836 			set_pte_at(&init_mm, address, ptep,
837 					pfn_pte(pfn, PAGE_KERNEL));
838 			__flush_tlb_single(address);
839 			map_op->host_addr = 0;
840 		}
841 	} else
842 		set_phys_to_machine(pfn, page->index);
843 
844 	/* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
845 	 * somewhere in this domain, even before being added to the
846 	 * m2p_override (see comment above in m2p_add_override).
847 	 * If there are no other entries in the m2p_override corresponding
848 	 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
849 	 * the original pfn (the one shared by the frontend): the backend
850 	 * cannot do any IO on this page anymore because it has been
851 	 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
852 	 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
853 	 * pfn again. */
854 	mfn &= ~FOREIGN_FRAME_BIT;
855 	ret = __get_user(pfn, &machine_to_phys_mapping[mfn]);
856 	if (ret == 0 && get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) &&
857 			m2p_find_override(mfn) == NULL)
858 		set_phys_to_machine(pfn, mfn);
859 
860 	return 0;
861 }
862 EXPORT_SYMBOL_GPL(m2p_remove_override);
863 
864 struct page *m2p_find_override(unsigned long mfn)
865 {
866 	unsigned long flags;
867 	struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)];
868 	struct page *p, *ret;
869 
870 	ret = NULL;
871 
872 	spin_lock_irqsave(&m2p_override_lock, flags);
873 
874 	list_for_each_entry(p, bucket, lru) {
875 		if (page_private(p) == mfn) {
876 			ret = p;
877 			break;
878 		}
879 	}
880 
881 	spin_unlock_irqrestore(&m2p_override_lock, flags);
882 
883 	return ret;
884 }
885 
886 unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn)
887 {
888 	struct page *p = m2p_find_override(mfn);
889 	unsigned long ret = pfn;
890 
891 	if (p)
892 		ret = page_to_pfn(p);
893 
894 	return ret;
895 }
896 EXPORT_SYMBOL_GPL(m2p_find_override_pfn);
897 
898 #ifdef CONFIG_XEN_DEBUG_FS
899 #include <linux/debugfs.h>
900 #include "debugfs.h"
901 static int p2m_dump_show(struct seq_file *m, void *v)
902 {
903 	static const char * const level_name[] = { "top", "middle",
904 						"entry", "abnormal", "error"};
905 #define TYPE_IDENTITY 0
906 #define TYPE_MISSING 1
907 #define TYPE_PFN 2
908 #define TYPE_UNKNOWN 3
909 	static const char * const type_name[] = {
910 				[TYPE_IDENTITY] = "identity",
911 				[TYPE_MISSING] = "missing",
912 				[TYPE_PFN] = "pfn",
913 				[TYPE_UNKNOWN] = "abnormal"};
914 	unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0;
915 	unsigned int uninitialized_var(prev_level);
916 	unsigned int uninitialized_var(prev_type);
917 
918 	if (!p2m_top)
919 		return 0;
920 
921 	for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) {
922 		unsigned topidx = p2m_top_index(pfn);
923 		unsigned mididx = p2m_mid_index(pfn);
924 		unsigned idx = p2m_index(pfn);
925 		unsigned lvl, type;
926 
927 		lvl = 4;
928 		type = TYPE_UNKNOWN;
929 		if (p2m_top[topidx] == p2m_mid_missing) {
930 			lvl = 0; type = TYPE_MISSING;
931 		} else if (p2m_top[topidx] == NULL) {
932 			lvl = 0; type = TYPE_UNKNOWN;
933 		} else if (p2m_top[topidx][mididx] == NULL) {
934 			lvl = 1; type = TYPE_UNKNOWN;
935 		} else if (p2m_top[topidx][mididx] == p2m_identity) {
936 			lvl = 1; type = TYPE_IDENTITY;
937 		} else if (p2m_top[topidx][mididx] == p2m_missing) {
938 			lvl = 1; type = TYPE_MISSING;
939 		} else if (p2m_top[topidx][mididx][idx] == 0) {
940 			lvl = 2; type = TYPE_UNKNOWN;
941 		} else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) {
942 			lvl = 2; type = TYPE_IDENTITY;
943 		} else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) {
944 			lvl = 2; type = TYPE_MISSING;
945 		} else if (p2m_top[topidx][mididx][idx] == pfn) {
946 			lvl = 2; type = TYPE_PFN;
947 		} else if (p2m_top[topidx][mididx][idx] != pfn) {
948 			lvl = 2; type = TYPE_PFN;
949 		}
950 		if (pfn == 0) {
951 			prev_level = lvl;
952 			prev_type = type;
953 		}
954 		if (pfn == MAX_DOMAIN_PAGES-1) {
955 			lvl = 3;
956 			type = TYPE_UNKNOWN;
957 		}
958 		if (prev_type != type) {
959 			seq_printf(m, " [0x%lx->0x%lx] %s\n",
960 				prev_pfn_type, pfn, type_name[prev_type]);
961 			prev_pfn_type = pfn;
962 			prev_type = type;
963 		}
964 		if (prev_level != lvl) {
965 			seq_printf(m, " [0x%lx->0x%lx] level %s\n",
966 				prev_pfn_level, pfn, level_name[prev_level]);
967 			prev_pfn_level = pfn;
968 			prev_level = lvl;
969 		}
970 	}
971 	return 0;
972 #undef TYPE_IDENTITY
973 #undef TYPE_MISSING
974 #undef TYPE_PFN
975 #undef TYPE_UNKNOWN
976 }
977 
978 static int p2m_dump_open(struct inode *inode, struct file *filp)
979 {
980 	return single_open(filp, p2m_dump_show, NULL);
981 }
982 
983 static const struct file_operations p2m_dump_fops = {
984 	.open		= p2m_dump_open,
985 	.read		= seq_read,
986 	.llseek		= seq_lseek,
987 	.release	= single_release,
988 };
989 
990 static struct dentry *d_mmu_debug;
991 
992 static int __init xen_p2m_debugfs(void)
993 {
994 	struct dentry *d_xen = xen_init_debugfs();
995 
996 	if (d_xen == NULL)
997 		return -ENOMEM;
998 
999 	d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1000 
1001 	debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
1002 	return 0;
1003 }
1004 fs_initcall(xen_p2m_debugfs);
1005 #endif /* CONFIG_XEN_DEBUG_FS */
1006