Loading...
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
169static void __init m2p_override_init(void);
170
171unsigned 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 */
180static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE);
181static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE);
182static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn, P2M_MID_PER_PAGE);
183
184static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE);
185static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, P2M_TOP_PER_PAGE);
186static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_TOP_PER_PAGE);
187
188static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE);
189
190RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE)));
191RESERVE_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. */
195RESERVE_BRK(p2m_mid_identity, PAGE_SIZE * 2 * 3);
196
197/* When we populate back during bootup, the amount of pages can vary. The
198 * max we have is seen is 395979, but that does not mean it can't be more.
199 * Some machines can have 3GB I/O holes even. With early_can_reuse_p2m_middle
200 * it can re-use Xen provided mfn_list array, so we only need to allocate at
201 * most three P2M top nodes. */
202RESERVE_BRK(p2m_populated, PAGE_SIZE * 3);
203
204static inline unsigned p2m_top_index(unsigned long pfn)
205{
206 BUG_ON(pfn >= MAX_P2M_PFN);
207 return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE);
208}
209
210static inline unsigned p2m_mid_index(unsigned long pfn)
211{
212 return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE;
213}
214
215static inline unsigned p2m_index(unsigned long pfn)
216{
217 return pfn % P2M_PER_PAGE;
218}
219
220static void p2m_top_init(unsigned long ***top)
221{
222 unsigned i;
223
224 for (i = 0; i < P2M_TOP_PER_PAGE; i++)
225 top[i] = p2m_mid_missing;
226}
227
228static void p2m_top_mfn_init(unsigned long *top)
229{
230 unsigned i;
231
232 for (i = 0; i < P2M_TOP_PER_PAGE; i++)
233 top[i] = virt_to_mfn(p2m_mid_missing_mfn);
234}
235
236static void p2m_top_mfn_p_init(unsigned long **top)
237{
238 unsigned i;
239
240 for (i = 0; i < P2M_TOP_PER_PAGE; i++)
241 top[i] = p2m_mid_missing_mfn;
242}
243
244static void p2m_mid_init(unsigned long **mid)
245{
246 unsigned i;
247
248 for (i = 0; i < P2M_MID_PER_PAGE; i++)
249 mid[i] = p2m_missing;
250}
251
252static void p2m_mid_mfn_init(unsigned long *mid)
253{
254 unsigned i;
255
256 for (i = 0; i < P2M_MID_PER_PAGE; i++)
257 mid[i] = virt_to_mfn(p2m_missing);
258}
259
260static void p2m_init(unsigned long *p2m)
261{
262 unsigned i;
263
264 for (i = 0; i < P2M_MID_PER_PAGE; i++)
265 p2m[i] = INVALID_P2M_ENTRY;
266}
267
268/*
269 * Build the parallel p2m_top_mfn and p2m_mid_mfn structures
270 *
271 * This is called both at boot time, and after resuming from suspend:
272 * - At boot time we're called very early, and must use extend_brk()
273 * to allocate memory.
274 *
275 * - After resume we're called from within stop_machine, but the mfn
276 * tree should alreay be completely allocated.
277 */
278void __ref xen_build_mfn_list_list(void)
279{
280 unsigned long pfn;
281
282 /* Pre-initialize p2m_top_mfn to be completely missing */
283 if (p2m_top_mfn == NULL) {
284 p2m_mid_missing_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
285 p2m_mid_mfn_init(p2m_mid_missing_mfn);
286
287 p2m_top_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
288 p2m_top_mfn_p_init(p2m_top_mfn_p);
289
290 p2m_top_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE);
291 p2m_top_mfn_init(p2m_top_mfn);
292 } else {
293 /* Reinitialise, mfn's all change after migration */
294 p2m_mid_mfn_init(p2m_mid_missing_mfn);
295 }
296
297 for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) {
298 unsigned topidx = p2m_top_index(pfn);
299 unsigned mididx = p2m_mid_index(pfn);
300 unsigned long **mid;
301 unsigned long *mid_mfn_p;
302
303 mid = p2m_top[topidx];
304 mid_mfn_p = p2m_top_mfn_p[topidx];
305
306 /* Don't bother allocating any mfn mid levels if
307 * they're just missing, just update the stored mfn,
308 * since all could have changed over a migrate.
309 */
310 if (mid == p2m_mid_missing) {
311 BUG_ON(mididx);
312 BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
313 p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn);
314 pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE;
315 continue;
316 }
317
318 if (mid_mfn_p == p2m_mid_missing_mfn) {
319 /*
320 * XXX boot-time only! We should never find
321 * missing parts of the mfn tree after
322 * runtime. extend_brk() will BUG if we call
323 * it too late.
324 */
325 mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
326 p2m_mid_mfn_init(mid_mfn_p);
327
328 p2m_top_mfn_p[topidx] = mid_mfn_p;
329 }
330
331 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
332 mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]);
333 }
334}
335
336void xen_setup_mfn_list_list(void)
337{
338 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
339
340 HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list =
341 virt_to_mfn(p2m_top_mfn);
342 HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn;
343}
344
345/* Set up p2m_top to point to the domain-builder provided p2m pages */
346void __init xen_build_dynamic_phys_to_machine(void)
347{
348 unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list;
349 unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages);
350 unsigned long pfn;
351
352 xen_max_p2m_pfn = max_pfn;
353
354 p2m_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
355 p2m_init(p2m_missing);
356
357 p2m_mid_missing = extend_brk(PAGE_SIZE, PAGE_SIZE);
358 p2m_mid_init(p2m_mid_missing);
359
360 p2m_top = extend_brk(PAGE_SIZE, PAGE_SIZE);
361 p2m_top_init(p2m_top);
362
363 p2m_identity = extend_brk(PAGE_SIZE, PAGE_SIZE);
364 p2m_init(p2m_identity);
365
366 /*
367 * The domain builder gives us a pre-constructed p2m array in
368 * mfn_list for all the pages initially given to us, so we just
369 * need to graft that into our tree structure.
370 */
371 for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) {
372 unsigned topidx = p2m_top_index(pfn);
373 unsigned mididx = p2m_mid_index(pfn);
374
375 if (p2m_top[topidx] == p2m_mid_missing) {
376 unsigned long **mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
377 p2m_mid_init(mid);
378
379 p2m_top[topidx] = mid;
380 }
381
382 /*
383 * As long as the mfn_list has enough entries to completely
384 * fill a p2m page, pointing into the array is ok. But if
385 * not the entries beyond the last pfn will be undefined.
386 */
387 if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) {
388 unsigned long p2midx;
389
390 p2midx = max_pfn % P2M_PER_PAGE;
391 for ( ; p2midx < P2M_PER_PAGE; p2midx++)
392 mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY;
393 }
394 p2m_top[topidx][mididx] = &mfn_list[pfn];
395 }
396
397 m2p_override_init();
398}
399
400unsigned long get_phys_to_machine(unsigned long pfn)
401{
402 unsigned topidx, mididx, idx;
403
404 if (unlikely(pfn >= MAX_P2M_PFN))
405 return INVALID_P2M_ENTRY;
406
407 topidx = p2m_top_index(pfn);
408 mididx = p2m_mid_index(pfn);
409 idx = p2m_index(pfn);
410
411 /*
412 * The INVALID_P2M_ENTRY is filled in both p2m_*identity
413 * and in p2m_*missing, so returning the INVALID_P2M_ENTRY
414 * would be wrong.
415 */
416 if (p2m_top[topidx][mididx] == p2m_identity)
417 return IDENTITY_FRAME(pfn);
418
419 return p2m_top[topidx][mididx][idx];
420}
421EXPORT_SYMBOL_GPL(get_phys_to_machine);
422
423static void *alloc_p2m_page(void)
424{
425 return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT);
426}
427
428static void free_p2m_page(void *p)
429{
430 free_page((unsigned long)p);
431}
432
433/*
434 * Fully allocate the p2m structure for a given pfn. We need to check
435 * that both the top and mid levels are allocated, and make sure the
436 * parallel mfn tree is kept in sync. We may race with other cpus, so
437 * the new pages are installed with cmpxchg; if we lose the race then
438 * simply free the page we allocated and use the one that's there.
439 */
440static bool alloc_p2m(unsigned long pfn)
441{
442 unsigned topidx, mididx;
443 unsigned long ***top_p, **mid;
444 unsigned long *top_mfn_p, *mid_mfn;
445
446 topidx = p2m_top_index(pfn);
447 mididx = p2m_mid_index(pfn);
448
449 top_p = &p2m_top[topidx];
450 mid = *top_p;
451
452 if (mid == p2m_mid_missing) {
453 /* Mid level is missing, allocate a new one */
454 mid = alloc_p2m_page();
455 if (!mid)
456 return false;
457
458 p2m_mid_init(mid);
459
460 if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing)
461 free_p2m_page(mid);
462 }
463
464 top_mfn_p = &p2m_top_mfn[topidx];
465 mid_mfn = p2m_top_mfn_p[topidx];
466
467 BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p);
468
469 if (mid_mfn == p2m_mid_missing_mfn) {
470 /* Separately check the mid mfn level */
471 unsigned long missing_mfn;
472 unsigned long mid_mfn_mfn;
473
474 mid_mfn = alloc_p2m_page();
475 if (!mid_mfn)
476 return false;
477
478 p2m_mid_mfn_init(mid_mfn);
479
480 missing_mfn = virt_to_mfn(p2m_mid_missing_mfn);
481 mid_mfn_mfn = virt_to_mfn(mid_mfn);
482 if (cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn) != missing_mfn)
483 free_p2m_page(mid_mfn);
484 else
485 p2m_top_mfn_p[topidx] = mid_mfn;
486 }
487
488 if (p2m_top[topidx][mididx] == p2m_identity ||
489 p2m_top[topidx][mididx] == p2m_missing) {
490 /* p2m leaf page is missing */
491 unsigned long *p2m;
492 unsigned long *p2m_orig = p2m_top[topidx][mididx];
493
494 p2m = alloc_p2m_page();
495 if (!p2m)
496 return false;
497
498 p2m_init(p2m);
499
500 if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig)
501 free_p2m_page(p2m);
502 else
503 mid_mfn[mididx] = virt_to_mfn(p2m);
504 }
505
506 return true;
507}
508
509static bool __init early_alloc_p2m_middle(unsigned long pfn, bool check_boundary)
510{
511 unsigned topidx, mididx, idx;
512 unsigned long *p2m;
513 unsigned long *mid_mfn_p;
514
515 topidx = p2m_top_index(pfn);
516 mididx = p2m_mid_index(pfn);
517 idx = p2m_index(pfn);
518
519 /* Pfff.. No boundary cross-over, lets get out. */
520 if (!idx && check_boundary)
521 return false;
522
523 WARN(p2m_top[topidx][mididx] == p2m_identity,
524 "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n",
525 topidx, mididx);
526
527 /*
528 * Could be done by xen_build_dynamic_phys_to_machine..
529 */
530 if (p2m_top[topidx][mididx] != p2m_missing)
531 return false;
532
533 /* Boundary cross-over for the edges: */
534 p2m = extend_brk(PAGE_SIZE, PAGE_SIZE);
535
536 p2m_init(p2m);
537
538 p2m_top[topidx][mididx] = p2m;
539
540 /* For save/restore we need to MFN of the P2M saved */
541
542 mid_mfn_p = p2m_top_mfn_p[topidx];
543 WARN(mid_mfn_p[mididx] != virt_to_mfn(p2m_missing),
544 "P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n",
545 topidx, mididx);
546 mid_mfn_p[mididx] = virt_to_mfn(p2m);
547
548 return true;
549}
550
551static bool __init early_alloc_p2m(unsigned long pfn)
552{
553 unsigned topidx = p2m_top_index(pfn);
554 unsigned long *mid_mfn_p;
555 unsigned long **mid;
556
557 mid = p2m_top[topidx];
558 mid_mfn_p = p2m_top_mfn_p[topidx];
559 if (mid == p2m_mid_missing) {
560 mid = extend_brk(PAGE_SIZE, PAGE_SIZE);
561
562 p2m_mid_init(mid);
563
564 p2m_top[topidx] = mid;
565
566 BUG_ON(mid_mfn_p != p2m_mid_missing_mfn);
567 }
568 /* And the save/restore P2M tables.. */
569 if (mid_mfn_p == p2m_mid_missing_mfn) {
570 mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE);
571 p2m_mid_mfn_init(mid_mfn_p);
572
573 p2m_top_mfn_p[topidx] = mid_mfn_p;
574 p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p);
575 /* Note: we don't set mid_mfn_p[midix] here,
576 * look in early_alloc_p2m_middle */
577 }
578 return true;
579}
580
581/*
582 * Skim over the P2M tree looking at pages that are either filled with
583 * INVALID_P2M_ENTRY or with 1:1 PFNs. If found, re-use that page and
584 * replace the P2M leaf with a p2m_missing or p2m_identity.
585 * Stick the old page in the new P2M tree location.
586 */
587bool __init early_can_reuse_p2m_middle(unsigned long set_pfn, unsigned long set_mfn)
588{
589 unsigned topidx;
590 unsigned mididx;
591 unsigned ident_pfns;
592 unsigned inv_pfns;
593 unsigned long *p2m;
594 unsigned long *mid_mfn_p;
595 unsigned idx;
596 unsigned long pfn;
597
598 /* We only look when this entails a P2M middle layer */
599 if (p2m_index(set_pfn))
600 return false;
601
602 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_PER_PAGE) {
603 topidx = p2m_top_index(pfn);
604
605 if (!p2m_top[topidx])
606 continue;
607
608 if (p2m_top[topidx] == p2m_mid_missing)
609 continue;
610
611 mididx = p2m_mid_index(pfn);
612 p2m = p2m_top[topidx][mididx];
613 if (!p2m)
614 continue;
615
616 if ((p2m == p2m_missing) || (p2m == p2m_identity))
617 continue;
618
619 if ((unsigned long)p2m == INVALID_P2M_ENTRY)
620 continue;
621
622 ident_pfns = 0;
623 inv_pfns = 0;
624 for (idx = 0; idx < P2M_PER_PAGE; idx++) {
625 /* IDENTITY_PFNs are 1:1 */
626 if (p2m[idx] == IDENTITY_FRAME(pfn + idx))
627 ident_pfns++;
628 else if (p2m[idx] == INVALID_P2M_ENTRY)
629 inv_pfns++;
630 else
631 break;
632 }
633 if ((ident_pfns == P2M_PER_PAGE) || (inv_pfns == P2M_PER_PAGE))
634 goto found;
635 }
636 return false;
637found:
638 /* Found one, replace old with p2m_identity or p2m_missing */
639 p2m_top[topidx][mididx] = (ident_pfns ? p2m_identity : p2m_missing);
640 /* And the other for save/restore.. */
641 mid_mfn_p = p2m_top_mfn_p[topidx];
642 /* NOTE: Even if it is a p2m_identity it should still be point to
643 * a page filled with INVALID_P2M_ENTRY entries. */
644 mid_mfn_p[mididx] = virt_to_mfn(p2m_missing);
645
646 /* Reset where we want to stick the old page in. */
647 topidx = p2m_top_index(set_pfn);
648 mididx = p2m_mid_index(set_pfn);
649
650 /* This shouldn't happen */
651 if (WARN_ON(p2m_top[topidx] == p2m_mid_missing))
652 early_alloc_p2m(set_pfn);
653
654 if (WARN_ON(p2m_top[topidx][mididx] != p2m_missing))
655 return false;
656
657 p2m_init(p2m);
658 p2m_top[topidx][mididx] = p2m;
659 mid_mfn_p = p2m_top_mfn_p[topidx];
660 mid_mfn_p[mididx] = virt_to_mfn(p2m);
661
662 return true;
663}
664bool __init early_set_phys_to_machine(unsigned long pfn, unsigned long mfn)
665{
666 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
667 if (!early_alloc_p2m(pfn))
668 return false;
669
670 if (early_can_reuse_p2m_middle(pfn, mfn))
671 return __set_phys_to_machine(pfn, mfn);
672
673 if (!early_alloc_p2m_middle(pfn, false /* boundary crossover OK!*/))
674 return false;
675
676 if (!__set_phys_to_machine(pfn, mfn))
677 return false;
678 }
679
680 return true;
681}
682unsigned long __init set_phys_range_identity(unsigned long pfn_s,
683 unsigned long pfn_e)
684{
685 unsigned long pfn;
686
687 if (unlikely(pfn_s >= MAX_P2M_PFN || pfn_e >= MAX_P2M_PFN))
688 return 0;
689
690 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap)))
691 return pfn_e - pfn_s;
692
693 if (pfn_s > pfn_e)
694 return 0;
695
696 for (pfn = (pfn_s & ~(P2M_MID_PER_PAGE * P2M_PER_PAGE - 1));
697 pfn < ALIGN(pfn_e, (P2M_MID_PER_PAGE * P2M_PER_PAGE));
698 pfn += P2M_MID_PER_PAGE * P2M_PER_PAGE)
699 {
700 WARN_ON(!early_alloc_p2m(pfn));
701 }
702
703 early_alloc_p2m_middle(pfn_s, true);
704 early_alloc_p2m_middle(pfn_e, true);
705
706 for (pfn = pfn_s; pfn < pfn_e; pfn++)
707 if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn)))
708 break;
709
710 if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s),
711 "Identity mapping failed. We are %ld short of 1-1 mappings!\n",
712 (pfn_e - pfn_s) - (pfn - pfn_s)))
713 printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn);
714
715 return pfn - pfn_s;
716}
717
718/* Try to install p2m mapping; fail if intermediate bits missing */
719bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn)
720{
721 unsigned topidx, mididx, idx;
722
723 if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) {
724 BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY);
725 return true;
726 }
727 if (unlikely(pfn >= MAX_P2M_PFN)) {
728 BUG_ON(mfn != INVALID_P2M_ENTRY);
729 return true;
730 }
731
732 topidx = p2m_top_index(pfn);
733 mididx = p2m_mid_index(pfn);
734 idx = p2m_index(pfn);
735
736 /* For sparse holes were the p2m leaf has real PFN along with
737 * PCI holes, stick in the PFN as the MFN value.
738 */
739 if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) {
740 if (p2m_top[topidx][mididx] == p2m_identity)
741 return true;
742
743 /* Swap over from MISSING to IDENTITY if needed. */
744 if (p2m_top[topidx][mididx] == p2m_missing) {
745 WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing,
746 p2m_identity) != p2m_missing);
747 return true;
748 }
749 }
750
751 if (p2m_top[topidx][mididx] == p2m_missing)
752 return mfn == INVALID_P2M_ENTRY;
753
754 p2m_top[topidx][mididx][idx] = mfn;
755
756 return true;
757}
758
759bool set_phys_to_machine(unsigned long pfn, unsigned long mfn)
760{
761 if (unlikely(!__set_phys_to_machine(pfn, mfn))) {
762 if (!alloc_p2m(pfn))
763 return false;
764
765 if (!__set_phys_to_machine(pfn, mfn))
766 return false;
767 }
768
769 return true;
770}
771
772#define M2P_OVERRIDE_HASH_SHIFT 10
773#define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT)
774
775static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH);
776static DEFINE_SPINLOCK(m2p_override_lock);
777
778static void __init m2p_override_init(void)
779{
780 unsigned i;
781
782 m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH,
783 sizeof(unsigned long));
784
785 for (i = 0; i < M2P_OVERRIDE_HASH; i++)
786 INIT_LIST_HEAD(&m2p_overrides[i]);
787}
788
789static unsigned long mfn_hash(unsigned long mfn)
790{
791 return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT);
792}
793
794/* Add an MFN override for a particular page */
795int m2p_add_override(unsigned long mfn, struct page *page,
796 struct gnttab_map_grant_ref *kmap_op)
797{
798 unsigned long flags;
799 unsigned long pfn;
800 unsigned long uninitialized_var(address);
801 unsigned level;
802 pte_t *ptep = NULL;
803 int ret = 0;
804
805 pfn = page_to_pfn(page);
806 if (!PageHighMem(page)) {
807 address = (unsigned long)__va(pfn << PAGE_SHIFT);
808 ptep = lookup_address(address, &level);
809 if (WARN(ptep == NULL || level != PG_LEVEL_4K,
810 "m2p_add_override: pfn %lx not mapped", pfn))
811 return -EINVAL;
812 }
813 WARN_ON(PagePrivate(page));
814 SetPagePrivate(page);
815 set_page_private(page, mfn);
816 page->index = pfn_to_mfn(pfn);
817
818 if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn))))
819 return -ENOMEM;
820
821 if (kmap_op != NULL) {
822 if (!PageHighMem(page)) {
823 struct multicall_space mcs =
824 xen_mc_entry(sizeof(*kmap_op));
825
826 MULTI_grant_table_op(mcs.mc,
827 GNTTABOP_map_grant_ref, kmap_op, 1);
828
829 xen_mc_issue(PARAVIRT_LAZY_MMU);
830 }
831 }
832 spin_lock_irqsave(&m2p_override_lock, flags);
833 list_add(&page->lru, &m2p_overrides[mfn_hash(mfn)]);
834 spin_unlock_irqrestore(&m2p_override_lock, flags);
835
836 /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in
837 * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other
838 * pfn so that the following mfn_to_pfn(mfn) calls will return the
839 * pfn from the m2p_override (the backend pfn) instead.
840 * We need to do this because the pages shared by the frontend
841 * (xen-blkfront) can be already locked (lock_page, called by
842 * do_read_cache_page); when the userspace backend tries to use them
843 * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so
844 * do_blockdev_direct_IO is going to try to lock the same pages
845 * again resulting in a deadlock.
846 * As a side effect get_user_pages_fast might not be safe on the
847 * frontend pages while they are being shared with the backend,
848 * because mfn_to_pfn (that ends up being called by GUPF) will
849 * return the backend pfn rather than the frontend pfn. */
850 ret = __get_user(pfn, &machine_to_phys_mapping[mfn]);
851 if (ret == 0 && get_phys_to_machine(pfn) == mfn)
852 set_phys_to_machine(pfn, FOREIGN_FRAME(mfn));
853
854 return 0;
855}
856EXPORT_SYMBOL_GPL(m2p_add_override);
857int m2p_remove_override(struct page *page,
858 struct gnttab_map_grant_ref *kmap_op)
859{
860 unsigned long flags;
861 unsigned long mfn;
862 unsigned long pfn;
863 unsigned long uninitialized_var(address);
864 unsigned level;
865 pte_t *ptep = NULL;
866 int ret = 0;
867
868 pfn = page_to_pfn(page);
869 mfn = get_phys_to_machine(pfn);
870 if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT))
871 return -EINVAL;
872
873 if (!PageHighMem(page)) {
874 address = (unsigned long)__va(pfn << PAGE_SHIFT);
875 ptep = lookup_address(address, &level);
876
877 if (WARN(ptep == NULL || level != PG_LEVEL_4K,
878 "m2p_remove_override: pfn %lx not mapped", pfn))
879 return -EINVAL;
880 }
881
882 spin_lock_irqsave(&m2p_override_lock, flags);
883 list_del(&page->lru);
884 spin_unlock_irqrestore(&m2p_override_lock, flags);
885 WARN_ON(!PagePrivate(page));
886 ClearPagePrivate(page);
887
888 set_phys_to_machine(pfn, page->index);
889 if (kmap_op != NULL) {
890 if (!PageHighMem(page)) {
891 struct multicall_space mcs;
892 struct gnttab_unmap_grant_ref *unmap_op;
893
894 /*
895 * It might be that we queued all the m2p grant table
896 * hypercalls in a multicall, then m2p_remove_override
897 * get called before the multicall has actually been
898 * issued. In this case handle is going to -1 because
899 * it hasn't been modified yet.
900 */
901 if (kmap_op->handle == -1)
902 xen_mc_flush();
903 /*
904 * Now if kmap_op->handle is negative it means that the
905 * hypercall actually returned an error.
906 */
907 if (kmap_op->handle == GNTST_general_error) {
908 printk(KERN_WARNING "m2p_remove_override: "
909 "pfn %lx mfn %lx, failed to modify kernel mappings",
910 pfn, mfn);
911 return -1;
912 }
913
914 mcs = xen_mc_entry(
915 sizeof(struct gnttab_unmap_grant_ref));
916 unmap_op = mcs.args;
917 unmap_op->host_addr = kmap_op->host_addr;
918 unmap_op->handle = kmap_op->handle;
919 unmap_op->dev_bus_addr = 0;
920
921 MULTI_grant_table_op(mcs.mc,
922 GNTTABOP_unmap_grant_ref, unmap_op, 1);
923
924 xen_mc_issue(PARAVIRT_LAZY_MMU);
925
926 set_pte_at(&init_mm, address, ptep,
927 pfn_pte(pfn, PAGE_KERNEL));
928 __flush_tlb_single(address);
929 kmap_op->host_addr = 0;
930 }
931 }
932
933 /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present
934 * somewhere in this domain, even before being added to the
935 * m2p_override (see comment above in m2p_add_override).
936 * If there are no other entries in the m2p_override corresponding
937 * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for
938 * the original pfn (the one shared by the frontend): the backend
939 * cannot do any IO on this page anymore because it has been
940 * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of
941 * the original pfn causes mfn_to_pfn(mfn) to return the frontend
942 * pfn again. */
943 mfn &= ~FOREIGN_FRAME_BIT;
944 ret = __get_user(pfn, &machine_to_phys_mapping[mfn]);
945 if (ret == 0 && get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) &&
946 m2p_find_override(mfn) == NULL)
947 set_phys_to_machine(pfn, mfn);
948
949 return 0;
950}
951EXPORT_SYMBOL_GPL(m2p_remove_override);
952
953struct page *m2p_find_override(unsigned long mfn)
954{
955 unsigned long flags;
956 struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)];
957 struct page *p, *ret;
958
959 ret = NULL;
960
961 spin_lock_irqsave(&m2p_override_lock, flags);
962
963 list_for_each_entry(p, bucket, lru) {
964 if (page_private(p) == mfn) {
965 ret = p;
966 break;
967 }
968 }
969
970 spin_unlock_irqrestore(&m2p_override_lock, flags);
971
972 return ret;
973}
974
975unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn)
976{
977 struct page *p = m2p_find_override(mfn);
978 unsigned long ret = pfn;
979
980 if (p)
981 ret = page_to_pfn(p);
982
983 return ret;
984}
985EXPORT_SYMBOL_GPL(m2p_find_override_pfn);
986
987#ifdef CONFIG_XEN_DEBUG_FS
988#include <linux/debugfs.h>
989#include "debugfs.h"
990static int p2m_dump_show(struct seq_file *m, void *v)
991{
992 static const char * const level_name[] = { "top", "middle",
993 "entry", "abnormal", "error"};
994#define TYPE_IDENTITY 0
995#define TYPE_MISSING 1
996#define TYPE_PFN 2
997#define TYPE_UNKNOWN 3
998 static const char * const type_name[] = {
999 [TYPE_IDENTITY] = "identity",
1000 [TYPE_MISSING] = "missing",
1001 [TYPE_PFN] = "pfn",
1002 [TYPE_UNKNOWN] = "abnormal"};
1003 unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0;
1004 unsigned int uninitialized_var(prev_level);
1005 unsigned int uninitialized_var(prev_type);
1006
1007 if (!p2m_top)
1008 return 0;
1009
1010 for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) {
1011 unsigned topidx = p2m_top_index(pfn);
1012 unsigned mididx = p2m_mid_index(pfn);
1013 unsigned idx = p2m_index(pfn);
1014 unsigned lvl, type;
1015
1016 lvl = 4;
1017 type = TYPE_UNKNOWN;
1018 if (p2m_top[topidx] == p2m_mid_missing) {
1019 lvl = 0; type = TYPE_MISSING;
1020 } else if (p2m_top[topidx] == NULL) {
1021 lvl = 0; type = TYPE_UNKNOWN;
1022 } else if (p2m_top[topidx][mididx] == NULL) {
1023 lvl = 1; type = TYPE_UNKNOWN;
1024 } else if (p2m_top[topidx][mididx] == p2m_identity) {
1025 lvl = 1; type = TYPE_IDENTITY;
1026 } else if (p2m_top[topidx][mididx] == p2m_missing) {
1027 lvl = 1; type = TYPE_MISSING;
1028 } else if (p2m_top[topidx][mididx][idx] == 0) {
1029 lvl = 2; type = TYPE_UNKNOWN;
1030 } else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) {
1031 lvl = 2; type = TYPE_IDENTITY;
1032 } else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) {
1033 lvl = 2; type = TYPE_MISSING;
1034 } else if (p2m_top[topidx][mididx][idx] == pfn) {
1035 lvl = 2; type = TYPE_PFN;
1036 } else if (p2m_top[topidx][mididx][idx] != pfn) {
1037 lvl = 2; type = TYPE_PFN;
1038 }
1039 if (pfn == 0) {
1040 prev_level = lvl;
1041 prev_type = type;
1042 }
1043 if (pfn == MAX_DOMAIN_PAGES-1) {
1044 lvl = 3;
1045 type = TYPE_UNKNOWN;
1046 }
1047 if (prev_type != type) {
1048 seq_printf(m, " [0x%lx->0x%lx] %s\n",
1049 prev_pfn_type, pfn, type_name[prev_type]);
1050 prev_pfn_type = pfn;
1051 prev_type = type;
1052 }
1053 if (prev_level != lvl) {
1054 seq_printf(m, " [0x%lx->0x%lx] level %s\n",
1055 prev_pfn_level, pfn, level_name[prev_level]);
1056 prev_pfn_level = pfn;
1057 prev_level = lvl;
1058 }
1059 }
1060 return 0;
1061#undef TYPE_IDENTITY
1062#undef TYPE_MISSING
1063#undef TYPE_PFN
1064#undef TYPE_UNKNOWN
1065}
1066
1067static int p2m_dump_open(struct inode *inode, struct file *filp)
1068{
1069 return single_open(filp, p2m_dump_show, NULL);
1070}
1071
1072static const struct file_operations p2m_dump_fops = {
1073 .open = p2m_dump_open,
1074 .read = seq_read,
1075 .llseek = seq_lseek,
1076 .release = single_release,
1077};
1078
1079static struct dentry *d_mmu_debug;
1080
1081static int __init xen_p2m_debugfs(void)
1082{
1083 struct dentry *d_xen = xen_init_debugfs();
1084
1085 if (d_xen == NULL)
1086 return -ENOMEM;
1087
1088 d_mmu_debug = debugfs_create_dir("mmu", d_xen);
1089
1090 debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops);
1091 return 0;
1092}
1093fs_initcall(xen_p2m_debugfs);
1094#endif /* CONFIG_XEN_DEBUG_FS */