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
p2m_top_index(unsigned long pfn)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
p2m_mid_index(unsigned long pfn)203 static inline unsigned p2m_mid_index(unsigned long pfn)
204 {
205 return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
206 }
207
p2m_index(unsigned long pfn)208 static inline unsigned p2m_index(unsigned long pfn)
209 {
210 return pfn % P2M_PER_PAGE;
211 }
212
p2m_top_init(unsigned long *** top)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
p2m_top_mfn_init(unsigned long * top)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
p2m_top_mfn_p_init(unsigned long ** top)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
p2m_mid_init(unsigned long ** mid)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
p2m_mid_mfn_init(unsigned long * mid)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
p2m_init(unsigned long * p2m)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 */
xen_build_mfn_list_list(void)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
xen_setup_mfn_list_list(void)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 */
xen_build_dynamic_phys_to_machine(void)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
get_phys_to_machine(unsigned long pfn)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
alloc_p2m_page(void)416 static void *alloc_p2m_page(void)
417 {
418 return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
419 }
420
free_p2m_page(void * p)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 */
alloc_p2m(unsigned long pfn)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
__early_alloc_p2m(unsigned long pfn)502 static bool __init __early_alloc_p2m(unsigned long pfn)
503 {
504 unsigned topidx, mididx, idx;
505
506 topidx = p2m_top_index(pfn);
507 mididx = p2m_mid_index(pfn);
508 idx = p2m_index(pfn);
509
510 /* Pfff.. No boundary cross-over, lets get out. */
511 if (!idx)
512 return false;
513
514 WARN(p2m_top[topidx][mididx] == p2m_identity,
515 "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
516 topidx, mididx);
517
518 /*
519 * Could be done by xen_build_dynamic_phys_to_machine..
520 */
521 if (p2m_top[topidx][mididx] != p2m_missing)
522 return false;
523
524 /* Boundary cross-over for the edges: */
525 if (idx) {
526 unsigned long *p2m = extend_brk(PAGE_SIZE, PAGE_SIZE);
527 unsigned long *mid_mfn_p;
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 }
542 return idx != 0;
543 }
set_phys_range_identity(unsigned long pfn_s,unsigned long pfn_e)544 unsigned long __init set_phys_range_identity(unsigned long pfn_s,
545 unsigned long pfn_e)
546 {
547 unsigned long pfn;
548
549 if (unlikely(pfn_s >= MAX_P2M_PFN || pfn_e >= MAX_P2M_PFN))
550 return 0;
551
552 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
553 return pfn_e - pfn_s;
554
555 if (pfn_s > pfn_e)
556 return 0;
557
558 for (pfn = (pfn_s & ~(P2M_MID_PER_PAGE * P2M_PER_PAGE - 1));
559 pfn < ALIGN(pfn_e, (P2M_MID_PER_PAGE * P2M_PER_PAGE));
560 pfn += P2M_MID_PER_PAGE * P2M_PER_PAGE)
561 {
562 unsigned topidx = p2m_top_index(pfn);
563 unsigned long *mid_mfn_p;
564 unsigned long **mid;
565
566 mid = p2m_top[topidx];
567 mid_mfn_p = p2m_top_mfn_p[topidx];
568 if (mid == p2m_mid_missing) {
569 mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
570
571 p2m_mid_init(mid);
572
573 p2m_top[topidx] = mid;
574
575 BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
576 }
577 /* And the save/restore P2M tables.. */
578 if (mid_mfn_p == p2m_mid_missing_mfn) {
579 mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
580 p2m_mid_mfn_init(mid_mfn_p);
581
582 p2m_top_mfn_p[topidx] = mid_mfn_p;
583 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
584 /* Note: we don't set mid_mfn_p[midix] here,
585 * look in __early_alloc_p2m */
586 }
587 }
588
589 __early_alloc_p2m(pfn_s);
590 __early_alloc_p2m(pfn_e);
591
592 for (pfn = pfn_s; pfn < pfn_e; pfn++)
593 if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn)))
594 break;
595
596 if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s),
597 "Identity mapping failed. We are %ld short of 1-1 mappings!\n",
598 (pfn_e - pfn_s) - (pfn - pfn_s)))
599 printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn);
600
601 return pfn - pfn_s;
602 }
603
604 /* Try to install p2m mapping; fail if intermediate bits missing */
__set_phys_to_machine(unsigned long pfn,unsigned long mfn)605 bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
606 {
607 unsigned topidx, mididx, idx;
608
609 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
610 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
611 return true;
612 }
613 if (unlikely(pfn >= MAX_P2M_PFN)) {
614 BUG_ON(mfn != INVALID_P2M_ENTRY);
615 return true;
616 }
617
618 topidx = p2m_top_index(pfn);
619 mididx = p2m_mid_index(pfn);
620 idx = p2m_index(pfn);
621
622 /* For sparse holes were the p2m leaf has real PFN along with
623 * PCI holes, stick in the PFN as the MFN value.
624 */
625 if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) {
626 if (p2m_top[topidx][mididx] == p2m_identity)
627 return true;
628
629 /* Swap over from MISSING to IDENTITY if needed. */
630 if (p2m_top[topidx][mididx] == p2m_missing) {
631 WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing,
632 p2m_identity) != p2m_missing);
633 return true;
634 }
635 }
636
637 if (p2m_top[topidx][mididx] == p2m_missing)
638 return mfn == INVALID_P2M_ENTRY;
639
640 p2m_top[topidx][mididx][idx] = mfn;
641
642 return true;
643 }
644
set_phys_to_machine(unsigned long pfn,unsigned long mfn)645 bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
646 {
647 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
648 if (!alloc_p2m(pfn))
649 return false;
650
651 if (!__set_phys_to_machine(pfn, mfn))
652 return false;
653 }
654
655 return true;
656 }
657
658 #define M2P_OVERRIDE_HASH_SHIFT 10
659 #define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT)
660
661 static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH);
662 static DEFINE_SPINLOCK(m2p_override_lock);
663
m2p_override_init(void)664 static void __init m2p_override_init(void)
665 {
666 unsigned i;
667
668 m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH,
669 sizeof(unsigned long));
670
671 for (i = 0; i < M2P_OVERRIDE_HASH; i++)
672 INIT_LIST_HEAD(&m2p_overrides[i]);
673 }
674
mfn_hash(unsigned long mfn)675 static unsigned long mfn_hash(unsigned long mfn)
676 {
677 return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT);
678 }
679
680 /* Add an MFN override for a particular page */
m2p_add_override(unsigned long mfn,struct page * page,struct gnttab_map_grant_ref * kmap_op)681 int m2p_add_override(unsigned long mfn, struct page *page,
682 struct gnttab_map_grant_ref *kmap_op)
683 {
684 unsigned long flags;
685 unsigned long pfn;
686 unsigned long uninitialized_var(address);
687 unsigned level;
688 pte_t *ptep = NULL;
689 int ret = 0;
690
691 pfn = page_to_pfn(page);
692 if (!PageHighMem(page)) {
693 address = (unsigned long)__va(pfn << PAGE_SHIFT);
694 ptep = lookup_address(address, &level);
695 if (WARN(ptep == NULL || level != PG_LEVEL_4K,
696 "m2p_add_override: pfn %lx not mapped", pfn))
697 return -EINVAL;
698 }
699 WARN_ON(PagePrivate(page));
700 SetPagePrivate(page);
701 set_page_private(page, mfn);
702 page->index = pfn_to_mfn(pfn);
703
704 if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn))))
705 return -ENOMEM;
706
707 if (kmap_op != NULL) {
708 if (!PageHighMem(page)) {
709 struct multicall_space mcs =
710 xen_mc_entry(sizeof(*kmap_op));
711
712 MULTI_grant_table_op(mcs.mc,
713 GNTTABOP_map_grant_ref, kmap_op, 1);
714
715 xen_mc_issue(PARAVIRT_LAZY_MMU);
716 }
717 }
718 spin_lock_irqsave(&m2p_override_lock, flags);
719 list_add(&page->lru, &m2p_overrides[mfn_hash(mfn)]);
720 spin_unlock_irqrestore(&m2p_override_lock, flags);
721
722 /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
723 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
724 * pfn so that the following mfn_to_pfn(mfn) calls will return the
725 * pfn from the m2p_override (the backend pfn) instead.
726 * We need to do this because the pages shared by the frontend
727 * (xen-blkfront) can be already locked (lock_page, called by
728 * do_read_cache_page); when the userspace backend tries to use them
729 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
730 * do_blockdev_direct_IO is going to try to lock the same pages
731 * again resulting in a deadlock.
732 * As a side effect get_user_pages_fast might not be safe on the
733 * frontend pages while they are being shared with the backend,
734 * because mfn_to_pfn (that ends up being called by GUPF) will
735 * return the backend pfn rather than the frontend pfn. */
736 ret = __get_user(pfn, &machine_to_phys_mapping[mfn]);
737 if (ret == 0 && get_phys_to_machine(pfn) == mfn)
738 set_phys_to_machine(pfn, FOREIGN_FRAME(mfn));
739
740 return 0;
741 }
742 EXPORT_SYMBOL_GPL(m2p_add_override);
m2p_remove_override(struct page * page,struct gnttab_map_grant_ref * kmap_op)743 int m2p_remove_override(struct page *page,
744 struct gnttab_map_grant_ref *kmap_op)
745 {
746 unsigned long flags;
747 unsigned long mfn;
748 unsigned long pfn;
749 unsigned long uninitialized_var(address);
750 unsigned level;
751 pte_t *ptep = NULL;
752 int ret = 0;
753
754 pfn = page_to_pfn(page);
755 mfn = get_phys_to_machine(pfn);
756 if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT))
757 return -EINVAL;
758
759 if (!PageHighMem(page)) {
760 address = (unsigned long)__va(pfn << PAGE_SHIFT);
761 ptep = lookup_address(address, &level);
762
763 if (WARN(ptep == NULL || level != PG_LEVEL_4K,
764 "m2p_remove_override: pfn %lx not mapped", pfn))
765 return -EINVAL;
766 }
767
768 spin_lock_irqsave(&m2p_override_lock, flags);
769 list_del(&page->lru);
770 spin_unlock_irqrestore(&m2p_override_lock, flags);
771 WARN_ON(!PagePrivate(page));
772 ClearPagePrivate(page);
773
774 set_phys_to_machine(pfn, page->index);
775 if (kmap_op != NULL) {
776 if (!PageHighMem(page)) {
777 struct multicall_space mcs;
778 struct gnttab_unmap_grant_ref *unmap_op;
779
780 /*
781 * It might be that we queued all the m2p grant table
782 * hypercalls in a multicall, then m2p_remove_override
783 * get called before the multicall has actually been
784 * issued. In this case handle is going to -1 because
785 * it hasn't been modified yet.
786 */
787 if (kmap_op->handle == -1)
788 xen_mc_flush();
789 /*
790 * Now if kmap_op->handle is negative it means that the
791 * hypercall actually returned an error.
792 */
793 if (kmap_op->handle == GNTST_general_error) {
794 printk(KERN_WARNING "m2p_remove_override: "
795 "pfn %lx mfn %lx, failed to modify kernel mappings",
796 pfn, mfn);
797 return -1;
798 }
799
800 mcs = xen_mc_entry(
801 sizeof(struct gnttab_unmap_grant_ref));
802 unmap_op = mcs.args;
803 unmap_op->host_addr = kmap_op->host_addr;
804 unmap_op->handle = kmap_op->handle;
805 unmap_op->dev_bus_addr = 0;
806
807 MULTI_grant_table_op(mcs.mc,
808 GNTTABOP_unmap_grant_ref, unmap_op, 1);
809
810 xen_mc_issue(PARAVIRT_LAZY_MMU);
811
812 set_pte_at(&init_mm, address, ptep,
813 pfn_pte(pfn, PAGE_KERNEL));
814 __flush_tlb_single(address);
815 kmap_op->host_addr = 0;
816 }
817 }
818
819 /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
820 * somewhere in this domain, even before being added to the
821 * m2p_override (see comment above in m2p_add_override).
822 * If there are no other entries in the m2p_override corresponding
823 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
824 * the original pfn (the one shared by the frontend): the backend
825 * cannot do any IO on this page anymore because it has been
826 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
827 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
828 * pfn again. */
829 mfn &= ~FOREIGN_FRAME_BIT;
830 ret = __get_user(pfn, &machine_to_phys_mapping[mfn]);
831 if (ret == 0 && get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) &&
832 m2p_find_override(mfn) == NULL)
833 set_phys_to_machine(pfn, mfn);
834
835 return 0;
836 }
837 EXPORT_SYMBOL_GPL(m2p_remove_override);
838
m2p_find_override(unsigned long mfn)839 struct page *m2p_find_override(unsigned long mfn)
840 {
841 unsigned long flags;
842 struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)];
843 struct page *p, *ret;
844
845 ret = NULL;
846
847 spin_lock_irqsave(&m2p_override_lock, flags);
848
849 list_for_each_entry(p, bucket, lru) {
850 if (page_private(p) == mfn) {
851 ret = p;
852 break;
853 }
854 }
855
856 spin_unlock_irqrestore(&m2p_override_lock, flags);
857
858 return ret;
859 }
860
m2p_find_override_pfn(unsigned long mfn,unsigned long pfn)861 unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn)
862 {
863 struct page *p = m2p_find_override(mfn);
864 unsigned long ret = pfn;
865
866 if (p)
867 ret = page_to_pfn(p);
868
869 return ret;
870 }
871 EXPORT_SYMBOL_GPL(m2p_find_override_pfn);
872
873 #ifdef CONFIG_XEN_DEBUG_FS
874 #include <linux/debugfs.h>
875 #include "debugfs.h"
p2m_dump_show(struct seq_file * m,void * v)876 static int p2m_dump_show(struct seq_file *m, void *v)
877 {
878 static const char * const level_name[] = { "top", "middle",
879 "entry", "abnormal", "error"};
880 #define TYPE_IDENTITY 0
881 #define TYPE_MISSING 1
882 #define TYPE_PFN 2
883 #define TYPE_UNKNOWN 3
884 static const char * const type_name[] = {
885 [TYPE_IDENTITY] = "identity",
886 [TYPE_MISSING] = "missing",
887 [TYPE_PFN] = "pfn",
888 [TYPE_UNKNOWN] = "abnormal"};
889 unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0;
890 unsigned int uninitialized_var(prev_level);
891 unsigned int uninitialized_var(prev_type);
892
893 if (!p2m_top)
894 return 0;
895
896 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) {
897 unsigned topidx = p2m_top_index(pfn);
898 unsigned mididx = p2m_mid_index(pfn);
899 unsigned idx = p2m_index(pfn);
900 unsigned lvl, type;
901
902 lvl = 4;
903 type = TYPE_UNKNOWN;
904 if (p2m_top[topidx] == p2m_mid_missing) {
905 lvl = 0; type = TYPE_MISSING;
906 } else if (p2m_top[topidx] == NULL) {
907 lvl = 0; type = TYPE_UNKNOWN;
908 } else if (p2m_top[topidx][mididx] == NULL) {
909 lvl = 1; type = TYPE_UNKNOWN;
910 } else if (p2m_top[topidx][mididx] == p2m_identity) {
911 lvl = 1; type = TYPE_IDENTITY;
912 } else if (p2m_top[topidx][mididx] == p2m_missing) {
913 lvl = 1; type = TYPE_MISSING;
914 } else if (p2m_top[topidx][mididx][idx] == 0) {
915 lvl = 2; type = TYPE_UNKNOWN;
916 } else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) {
917 lvl = 2; type = TYPE_IDENTITY;
918 } else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) {
919 lvl = 2; type = TYPE_MISSING;
920 } else if (p2m_top[topidx][mididx][idx] == pfn) {
921 lvl = 2; type = TYPE_PFN;
922 } else if (p2m_top[topidx][mididx][idx] != pfn) {
923 lvl = 2; type = TYPE_PFN;
924 }
925 if (pfn == 0) {
926 prev_level = lvl;
927 prev_type = type;
928 }
929 if (pfn == MAX_DOMAIN_PAGES-1) {
930 lvl = 3;
931 type = TYPE_UNKNOWN;
932 }
933 if (prev_type != type) {
934 seq_printf(m, " [0x%lx->0x%lx] %s\n",
935 prev_pfn_type, pfn, type_name[prev_type]);
936 prev_pfn_type = pfn;
937 prev_type = type;
938 }
939 if (prev_level != lvl) {
940 seq_printf(m, " [0x%lx->0x%lx] level %s\n",
941 prev_pfn_level, pfn, level_name[prev_level]);
942 prev_pfn_level = pfn;
943 prev_level = lvl;
944 }
945 }
946 return 0;
947 #undef TYPE_IDENTITY
948 #undef TYPE_MISSING
949 #undef TYPE_PFN
950 #undef TYPE_UNKNOWN
951 }
952
p2m_dump_open(struct inode * inode,struct file * filp)953 static int p2m_dump_open(struct inode *inode, struct file *filp)
954 {
955 return single_open(filp, p2m_dump_show, NULL);
956 }
957
958 static const struct file_operations p2m_dump_fops = {
959 .open = p2m_dump_open,
960 .read = seq_read,
961 .llseek = seq_lseek,
962 .release = single_release,
963 };
964
965 static struct dentry *d_mmu_debug;
966
xen_p2m_debugfs(void)967 static int __init xen_p2m_debugfs(void)
968 {
969 struct dentry *d_xen = xen_init_debugfs();
970
971 if (d_xen == NULL)
972 return -ENOMEM;
973
974 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
975
976 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
977 return 0;
978 }
979 fs_initcall(xen_p2m_debugfs);
980 #endif /* CONFIG_XEN_DEBUG_FS */
981