Commit | Line | Data |
---|---|---|
b5eafe92 JF |
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 | |
a3118beb | 25 | * 512 and 1024 entries respectively. |
f4cec35b KRW |
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 | * | | |
a3118beb KRW |
142 | * p2m_mid_missing p2m_missing |
143 | * /-----------------\ /------------\ | |
144 | * | [p2m_missing] +---->| ~0, ~0, ~0 | | |
145 | * | [p2m_missing] +---->| ..., ~0 | | |
146 | * \-----------------/ \------------/ | |
f4cec35b KRW |
147 | * |
148 | * where ~0 is INVALID_P2M_ENTRY. IDENTITY is (PFN | IDENTITY_BIT) | |
b5eafe92 JF |
149 | */ |
150 | ||
151 | #include <linux/init.h> | |
152 | #include <linux/module.h> | |
448f2831 JF |
153 | #include <linux/list.h> |
154 | #include <linux/hash.h> | |
87f1d40a | 155 | #include <linux/sched.h> |
2222e71b | 156 | #include <linux/seq_file.h> |
b5eafe92 JF |
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> | |
ee072640 | 164 | #include <xen/balloon.h> |
0930bba6 | 165 | #include <xen/grant_table.h> |
b5eafe92 | 166 | |
0930bba6 | 167 | #include "multicalls.h" |
b5eafe92 JF |
168 | #include "xen-ops.h" |
169 | ||
448f2831 JF |
170 | static void __init m2p_override_init(void); |
171 | ||
b5eafe92 JF |
172 | unsigned long xen_max_p2m_pfn __read_mostly; |
173 | ||
174 | #define P2M_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) | |
175 | #define P2M_MID_PER_PAGE (PAGE_SIZE / sizeof(unsigned long *)) | |
176 | #define P2M_TOP_PER_PAGE (PAGE_SIZE / sizeof(unsigned long **)) | |
177 | ||
178 | #define MAX_P2M_PFN (P2M_TOP_PER_PAGE * P2M_MID_PER_PAGE * P2M_PER_PAGE) | |
179 | ||
180 | /* Placeholders for holes in the address space */ | |
181 | static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_PER_PAGE); | |
182 | static RESERVE_BRK_ARRAY(unsigned long *, p2m_mid_missing, P2M_MID_PER_PAGE); | |
183 | static RESERVE_BRK_ARRAY(unsigned long, p2m_mid_missing_mfn, P2M_MID_PER_PAGE); | |
184 | ||
185 | static RESERVE_BRK_ARRAY(unsigned long **, p2m_top, P2M_TOP_PER_PAGE); | |
186 | static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, P2M_TOP_PER_PAGE); | |
187 | static RESERVE_BRK_ARRAY(unsigned long *, p2m_top_mfn_p, P2M_TOP_PER_PAGE); | |
188 | ||
f4cec35b KRW |
189 | static RESERVE_BRK_ARRAY(unsigned long, p2m_identity, P2M_PER_PAGE); |
190 | ||
b5eafe92 JF |
191 | RESERVE_BRK(p2m_mid, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE))); |
192 | RESERVE_BRK(p2m_mid_mfn, PAGE_SIZE * (MAX_DOMAIN_PAGES / (P2M_PER_PAGE * P2M_MID_PER_PAGE))); | |
193 | ||
f4cec35b KRW |
194 | /* We might hit two boundary violations at the start and end, at max each |
195 | * boundary violation will require three middle nodes. */ | |
196 | RESERVE_BRK(p2m_mid_identity, PAGE_SIZE * 2 * 3); | |
197 | ||
5bc6f988 KRW |
198 | /* When we populate back during bootup, the amount of pages can vary. The |
199 | * max we have is seen is 395979, but that does not mean it can't be more. | |
250a41e0 KRW |
200 | * Some machines can have 3GB I/O holes even. With early_can_reuse_p2m_middle |
201 | * it can re-use Xen provided mfn_list array, so we only need to allocate at | |
202 | * most three P2M top nodes. */ | |
203 | RESERVE_BRK(p2m_populated, PAGE_SIZE * 3); | |
204 | ||
b5eafe92 JF |
205 | static inline unsigned p2m_top_index(unsigned long pfn) |
206 | { | |
207 | BUG_ON(pfn >= MAX_P2M_PFN); | |
208 | return pfn / (P2M_MID_PER_PAGE * P2M_PER_PAGE); | |
209 | } | |
210 | ||
211 | static inline unsigned p2m_mid_index(unsigned long pfn) | |
212 | { | |
213 | return (pfn / P2M_PER_PAGE) % P2M_MID_PER_PAGE; | |
214 | } | |
215 | ||
216 | static inline unsigned p2m_index(unsigned long pfn) | |
217 | { | |
218 | return pfn % P2M_PER_PAGE; | |
219 | } | |
220 | ||
221 | static void p2m_top_init(unsigned long ***top) | |
222 | { | |
223 | unsigned i; | |
224 | ||
225 | for (i = 0; i < P2M_TOP_PER_PAGE; i++) | |
226 | top[i] = p2m_mid_missing; | |
227 | } | |
228 | ||
229 | static void p2m_top_mfn_init(unsigned long *top) | |
230 | { | |
231 | unsigned i; | |
232 | ||
233 | for (i = 0; i < P2M_TOP_PER_PAGE; i++) | |
234 | top[i] = virt_to_mfn(p2m_mid_missing_mfn); | |
235 | } | |
236 | ||
237 | static void p2m_top_mfn_p_init(unsigned long **top) | |
238 | { | |
239 | unsigned i; | |
240 | ||
241 | for (i = 0; i < P2M_TOP_PER_PAGE; i++) | |
242 | top[i] = p2m_mid_missing_mfn; | |
243 | } | |
244 | ||
245 | static void p2m_mid_init(unsigned long **mid) | |
246 | { | |
247 | unsigned i; | |
248 | ||
249 | for (i = 0; i < P2M_MID_PER_PAGE; i++) | |
250 | mid[i] = p2m_missing; | |
251 | } | |
252 | ||
253 | static void p2m_mid_mfn_init(unsigned long *mid) | |
254 | { | |
255 | unsigned i; | |
256 | ||
257 | for (i = 0; i < P2M_MID_PER_PAGE; i++) | |
258 | mid[i] = virt_to_mfn(p2m_missing); | |
259 | } | |
260 | ||
261 | static void p2m_init(unsigned long *p2m) | |
262 | { | |
263 | unsigned i; | |
264 | ||
265 | for (i = 0; i < P2M_MID_PER_PAGE; i++) | |
266 | p2m[i] = INVALID_P2M_ENTRY; | |
267 | } | |
268 | ||
269 | /* | |
270 | * Build the parallel p2m_top_mfn and p2m_mid_mfn structures | |
271 | * | |
272 | * This is called both at boot time, and after resuming from suspend: | |
273 | * - At boot time we're called very early, and must use extend_brk() | |
274 | * to allocate memory. | |
275 | * | |
276 | * - After resume we're called from within stop_machine, but the mfn | |
277 | * tree should alreay be completely allocated. | |
278 | */ | |
44b46c3e | 279 | void __ref xen_build_mfn_list_list(void) |
b5eafe92 JF |
280 | { |
281 | unsigned long pfn; | |
282 | ||
696fd7c5 KRW |
283 | if (xen_feature(XENFEAT_auto_translated_physmap)) |
284 | return; | |
285 | ||
b5eafe92 JF |
286 | /* Pre-initialize p2m_top_mfn to be completely missing */ |
287 | if (p2m_top_mfn == NULL) { | |
288 | p2m_mid_missing_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
289 | p2m_mid_mfn_init(p2m_mid_missing_mfn); | |
290 | ||
291 | p2m_top_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
292 | p2m_top_mfn_p_init(p2m_top_mfn_p); | |
293 | ||
294 | p2m_top_mfn = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
295 | p2m_top_mfn_init(p2m_top_mfn); | |
296 | } else { | |
297 | /* Reinitialise, mfn's all change after migration */ | |
298 | p2m_mid_mfn_init(p2m_mid_missing_mfn); | |
299 | } | |
300 | ||
301 | for (pfn = 0; pfn < xen_max_p2m_pfn; pfn += P2M_PER_PAGE) { | |
302 | unsigned topidx = p2m_top_index(pfn); | |
303 | unsigned mididx = p2m_mid_index(pfn); | |
304 | unsigned long **mid; | |
305 | unsigned long *mid_mfn_p; | |
306 | ||
307 | mid = p2m_top[topidx]; | |
308 | mid_mfn_p = p2m_top_mfn_p[topidx]; | |
309 | ||
310 | /* Don't bother allocating any mfn mid levels if | |
311 | * they're just missing, just update the stored mfn, | |
312 | * since all could have changed over a migrate. | |
313 | */ | |
314 | if (mid == p2m_mid_missing) { | |
315 | BUG_ON(mididx); | |
316 | BUG_ON(mid_mfn_p != p2m_mid_missing_mfn); | |
317 | p2m_top_mfn[topidx] = virt_to_mfn(p2m_mid_missing_mfn); | |
318 | pfn += (P2M_MID_PER_PAGE - 1) * P2M_PER_PAGE; | |
319 | continue; | |
320 | } | |
321 | ||
322 | if (mid_mfn_p == p2m_mid_missing_mfn) { | |
323 | /* | |
324 | * XXX boot-time only! We should never find | |
325 | * missing parts of the mfn tree after | |
326 | * runtime. extend_brk() will BUG if we call | |
327 | * it too late. | |
328 | */ | |
329 | mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
330 | p2m_mid_mfn_init(mid_mfn_p); | |
331 | ||
332 | p2m_top_mfn_p[topidx] = mid_mfn_p; | |
333 | } | |
334 | ||
335 | p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p); | |
336 | mid_mfn_p[mididx] = virt_to_mfn(mid[mididx]); | |
337 | } | |
338 | } | |
339 | ||
340 | void xen_setup_mfn_list_list(void) | |
341 | { | |
342 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); | |
343 | ||
344 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
345 | virt_to_mfn(p2m_top_mfn); | |
346 | HYPERVISOR_shared_info->arch.max_pfn = xen_max_p2m_pfn; | |
347 | } | |
348 | ||
349 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
350 | void __init xen_build_dynamic_phys_to_machine(void) | |
351 | { | |
696fd7c5 KRW |
352 | unsigned long *mfn_list; |
353 | unsigned long max_pfn; | |
b5eafe92 JF |
354 | unsigned long pfn; |
355 | ||
696fd7c5 KRW |
356 | if (xen_feature(XENFEAT_auto_translated_physmap)) |
357 | return; | |
358 | ||
359 | mfn_list = (unsigned long *)xen_start_info->mfn_list; | |
360 | max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); | |
b5eafe92 JF |
361 | xen_max_p2m_pfn = max_pfn; |
362 | ||
363 | p2m_missing = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
364 | p2m_init(p2m_missing); | |
365 | ||
366 | p2m_mid_missing = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
367 | p2m_mid_init(p2m_mid_missing); | |
368 | ||
369 | p2m_top = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
370 | p2m_top_init(p2m_top); | |
371 | ||
f4cec35b KRW |
372 | p2m_identity = extend_brk(PAGE_SIZE, PAGE_SIZE); |
373 | p2m_init(p2m_identity); | |
374 | ||
b5eafe92 JF |
375 | /* |
376 | * The domain builder gives us a pre-constructed p2m array in | |
377 | * mfn_list for all the pages initially given to us, so we just | |
378 | * need to graft that into our tree structure. | |
379 | */ | |
380 | for (pfn = 0; pfn < max_pfn; pfn += P2M_PER_PAGE) { | |
381 | unsigned topidx = p2m_top_index(pfn); | |
382 | unsigned mididx = p2m_mid_index(pfn); | |
383 | ||
384 | if (p2m_top[topidx] == p2m_mid_missing) { | |
385 | unsigned long **mid = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
386 | p2m_mid_init(mid); | |
387 | ||
388 | p2m_top[topidx] = mid; | |
389 | } | |
390 | ||
8e1b4cf2 SB |
391 | /* |
392 | * As long as the mfn_list has enough entries to completely | |
393 | * fill a p2m page, pointing into the array is ok. But if | |
394 | * not the entries beyond the last pfn will be undefined. | |
8e1b4cf2 SB |
395 | */ |
396 | if (unlikely(pfn + P2M_PER_PAGE > max_pfn)) { | |
397 | unsigned long p2midx; | |
cf04d120 SB |
398 | |
399 | p2midx = max_pfn % P2M_PER_PAGE; | |
400 | for ( ; p2midx < P2M_PER_PAGE; p2midx++) | |
401 | mfn_list[pfn + p2midx] = INVALID_P2M_ENTRY; | |
402 | } | |
403 | p2m_top[topidx][mididx] = &mfn_list[pfn]; | |
b5eafe92 | 404 | } |
448f2831 JF |
405 | |
406 | m2p_override_init(); | |
b5eafe92 | 407 | } |
357a3cfb KRW |
408 | #ifdef CONFIG_X86_64 |
409 | #include <linux/bootmem.h> | |
410 | unsigned long __init xen_revector_p2m_tree(void) | |
411 | { | |
412 | unsigned long va_start; | |
413 | unsigned long va_end; | |
414 | unsigned long pfn; | |
3fc509fc | 415 | unsigned long pfn_free = 0; |
357a3cfb KRW |
416 | unsigned long *mfn_list = NULL; |
417 | unsigned long size; | |
418 | ||
419 | va_start = xen_start_info->mfn_list; | |
420 | /*We copy in increments of P2M_PER_PAGE * sizeof(unsigned long), | |
421 | * so make sure it is rounded up to that */ | |
422 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
423 | va_end = va_start + size; | |
424 | ||
425 | /* If we were revectored already, don't do it again. */ | |
426 | if (va_start <= __START_KERNEL_map && va_start >= __PAGE_OFFSET) | |
427 | return 0; | |
428 | ||
429 | mfn_list = alloc_bootmem_align(size, PAGE_SIZE); | |
430 | if (!mfn_list) { | |
431 | pr_warn("Could not allocate space for a new P2M tree!\n"); | |
432 | return xen_start_info->mfn_list; | |
433 | } | |
434 | /* Fill it out with INVALID_P2M_ENTRY value */ | |
435 | memset(mfn_list, 0xFF, size); | |
436 | ||
437 | for (pfn = 0; pfn < ALIGN(MAX_DOMAIN_PAGES, P2M_PER_PAGE); pfn += P2M_PER_PAGE) { | |
438 | unsigned topidx = p2m_top_index(pfn); | |
439 | unsigned mididx; | |
440 | unsigned long *mid_p; | |
441 | ||
442 | if (!p2m_top[topidx]) | |
443 | continue; | |
b5eafe92 | 444 | |
357a3cfb KRW |
445 | if (p2m_top[topidx] == p2m_mid_missing) |
446 | continue; | |
447 | ||
448 | mididx = p2m_mid_index(pfn); | |
449 | mid_p = p2m_top[topidx][mididx]; | |
450 | if (!mid_p) | |
451 | continue; | |
452 | if ((mid_p == p2m_missing) || (mid_p == p2m_identity)) | |
453 | continue; | |
b5eafe92 | 454 | |
357a3cfb KRW |
455 | if ((unsigned long)mid_p == INVALID_P2M_ENTRY) |
456 | continue; | |
457 | ||
458 | /* The old va. Rebase it on mfn_list */ | |
459 | if (mid_p >= (unsigned long *)va_start && mid_p <= (unsigned long *)va_end) { | |
460 | unsigned long *new; | |
461 | ||
3fc509fc KRW |
462 | if (pfn_free > (size / sizeof(unsigned long))) { |
463 | WARN(1, "Only allocated for %ld pages, but we want %ld!\n", | |
464 | size / sizeof(unsigned long), pfn_free); | |
465 | return 0; | |
466 | } | |
467 | new = &mfn_list[pfn_free]; | |
357a3cfb KRW |
468 | |
469 | copy_page(new, mid_p); | |
3fc509fc KRW |
470 | p2m_top[topidx][mididx] = &mfn_list[pfn_free]; |
471 | p2m_top_mfn_p[topidx][mididx] = virt_to_mfn(&mfn_list[pfn_free]); | |
472 | ||
473 | pfn_free += P2M_PER_PAGE; | |
357a3cfb KRW |
474 | |
475 | } | |
476 | /* This should be the leafs allocated for identity from _brk. */ | |
477 | } | |
478 | return (unsigned long)mfn_list; | |
479 | ||
480 | } | |
481 | #else | |
482 | unsigned long __init xen_revector_p2m_tree(void) | |
483 | { | |
484 | return 0; | |
485 | } | |
486 | #endif | |
b5eafe92 JF |
487 | unsigned long get_phys_to_machine(unsigned long pfn) |
488 | { | |
489 | unsigned topidx, mididx, idx; | |
490 | ||
491 | if (unlikely(pfn >= MAX_P2M_PFN)) | |
492 | return INVALID_P2M_ENTRY; | |
493 | ||
494 | topidx = p2m_top_index(pfn); | |
495 | mididx = p2m_mid_index(pfn); | |
496 | idx = p2m_index(pfn); | |
497 | ||
f4cec35b KRW |
498 | /* |
499 | * The INVALID_P2M_ENTRY is filled in both p2m_*identity | |
500 | * and in p2m_*missing, so returning the INVALID_P2M_ENTRY | |
501 | * would be wrong. | |
502 | */ | |
503 | if (p2m_top[topidx][mididx] == p2m_identity) | |
504 | return IDENTITY_FRAME(pfn); | |
505 | ||
b5eafe92 JF |
506 | return p2m_top[topidx][mididx][idx]; |
507 | } | |
508 | EXPORT_SYMBOL_GPL(get_phys_to_machine); | |
509 | ||
510 | static void *alloc_p2m_page(void) | |
511 | { | |
512 | return (void *)__get_free_page(GFP_KERNEL | __GFP_REPEAT); | |
513 | } | |
514 | ||
515 | static void free_p2m_page(void *p) | |
516 | { | |
517 | free_page((unsigned long)p); | |
518 | } | |
519 | ||
a3118beb | 520 | /* |
b5eafe92 JF |
521 | * Fully allocate the p2m structure for a given pfn. We need to check |
522 | * that both the top and mid levels are allocated, and make sure the | |
523 | * parallel mfn tree is kept in sync. We may race with other cpus, so | |
524 | * the new pages are installed with cmpxchg; if we lose the race then | |
525 | * simply free the page we allocated and use the one that's there. | |
526 | */ | |
527 | static bool alloc_p2m(unsigned long pfn) | |
528 | { | |
529 | unsigned topidx, mididx; | |
530 | unsigned long ***top_p, **mid; | |
531 | unsigned long *top_mfn_p, *mid_mfn; | |
532 | ||
533 | topidx = p2m_top_index(pfn); | |
534 | mididx = p2m_mid_index(pfn); | |
535 | ||
536 | top_p = &p2m_top[topidx]; | |
537 | mid = *top_p; | |
538 | ||
539 | if (mid == p2m_mid_missing) { | |
540 | /* Mid level is missing, allocate a new one */ | |
541 | mid = alloc_p2m_page(); | |
542 | if (!mid) | |
543 | return false; | |
544 | ||
545 | p2m_mid_init(mid); | |
546 | ||
547 | if (cmpxchg(top_p, p2m_mid_missing, mid) != p2m_mid_missing) | |
548 | free_p2m_page(mid); | |
549 | } | |
550 | ||
551 | top_mfn_p = &p2m_top_mfn[topidx]; | |
552 | mid_mfn = p2m_top_mfn_p[topidx]; | |
553 | ||
554 | BUG_ON(virt_to_mfn(mid_mfn) != *top_mfn_p); | |
555 | ||
556 | if (mid_mfn == p2m_mid_missing_mfn) { | |
557 | /* Separately check the mid mfn level */ | |
558 | unsigned long missing_mfn; | |
559 | unsigned long mid_mfn_mfn; | |
560 | ||
561 | mid_mfn = alloc_p2m_page(); | |
562 | if (!mid_mfn) | |
563 | return false; | |
564 | ||
565 | p2m_mid_mfn_init(mid_mfn); | |
566 | ||
567 | missing_mfn = virt_to_mfn(p2m_mid_missing_mfn); | |
568 | mid_mfn_mfn = virt_to_mfn(mid_mfn); | |
569 | if (cmpxchg(top_mfn_p, missing_mfn, mid_mfn_mfn) != missing_mfn) | |
570 | free_p2m_page(mid_mfn); | |
571 | else | |
572 | p2m_top_mfn_p[topidx] = mid_mfn; | |
573 | } | |
574 | ||
f4cec35b KRW |
575 | if (p2m_top[topidx][mididx] == p2m_identity || |
576 | p2m_top[topidx][mididx] == p2m_missing) { | |
b5eafe92 JF |
577 | /* p2m leaf page is missing */ |
578 | unsigned long *p2m; | |
f4cec35b | 579 | unsigned long *p2m_orig = p2m_top[topidx][mididx]; |
b5eafe92 JF |
580 | |
581 | p2m = alloc_p2m_page(); | |
582 | if (!p2m) | |
583 | return false; | |
584 | ||
585 | p2m_init(p2m); | |
586 | ||
f4cec35b | 587 | if (cmpxchg(&mid[mididx], p2m_orig, p2m) != p2m_orig) |
b5eafe92 JF |
588 | free_p2m_page(p2m); |
589 | else | |
590 | mid_mfn[mididx] = virt_to_mfn(p2m); | |
591 | } | |
592 | ||
593 | return true; | |
594 | } | |
595 | ||
cef4cca5 | 596 | static bool __init early_alloc_p2m_middle(unsigned long pfn, bool check_boundary) |
f4cec35b KRW |
597 | { |
598 | unsigned topidx, mididx, idx; | |
d5096850 KRW |
599 | unsigned long *p2m; |
600 | unsigned long *mid_mfn_p; | |
f4cec35b KRW |
601 | |
602 | topidx = p2m_top_index(pfn); | |
603 | mididx = p2m_mid_index(pfn); | |
604 | idx = p2m_index(pfn); | |
605 | ||
606 | /* Pfff.. No boundary cross-over, lets get out. */ | |
cef4cca5 | 607 | if (!idx && check_boundary) |
f4cec35b KRW |
608 | return false; |
609 | ||
610 | WARN(p2m_top[topidx][mididx] == p2m_identity, | |
611 | "P2M[%d][%d] == IDENTITY, should be MISSING (or alloced)!\n", | |
612 | topidx, mididx); | |
613 | ||
614 | /* | |
615 | * Could be done by xen_build_dynamic_phys_to_machine.. | |
616 | */ | |
617 | if (p2m_top[topidx][mididx] != p2m_missing) | |
618 | return false; | |
619 | ||
620 | /* Boundary cross-over for the edges: */ | |
d5096850 | 621 | p2m = extend_brk(PAGE_SIZE, PAGE_SIZE); |
f4cec35b | 622 | |
d5096850 | 623 | p2m_init(p2m); |
f4cec35b | 624 | |
d5096850 | 625 | p2m_top[topidx][mididx] = p2m; |
f4cec35b | 626 | |
d5096850 | 627 | /* For save/restore we need to MFN of the P2M saved */ |
cef4cca5 | 628 | |
d5096850 KRW |
629 | mid_mfn_p = p2m_top_mfn_p[topidx]; |
630 | WARN(mid_mfn_p[mididx] != virt_to_mfn(p2m_missing), | |
631 | "P2M_TOP_P[%d][%d] != MFN of p2m_missing!\n", | |
632 | topidx, mididx); | |
633 | mid_mfn_p[mididx] = virt_to_mfn(p2m); | |
8c595088 | 634 | |
d5096850 | 635 | return true; |
f4cec35b | 636 | } |
3f3aaea2 KRW |
637 | |
638 | static bool __init early_alloc_p2m(unsigned long pfn) | |
639 | { | |
640 | unsigned topidx = p2m_top_index(pfn); | |
641 | unsigned long *mid_mfn_p; | |
642 | unsigned long **mid; | |
643 | ||
644 | mid = p2m_top[topidx]; | |
645 | mid_mfn_p = p2m_top_mfn_p[topidx]; | |
646 | if (mid == p2m_mid_missing) { | |
647 | mid = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
648 | ||
649 | p2m_mid_init(mid); | |
650 | ||
651 | p2m_top[topidx] = mid; | |
652 | ||
653 | BUG_ON(mid_mfn_p != p2m_mid_missing_mfn); | |
654 | } | |
655 | /* And the save/restore P2M tables.. */ | |
656 | if (mid_mfn_p == p2m_mid_missing_mfn) { | |
657 | mid_mfn_p = extend_brk(PAGE_SIZE, PAGE_SIZE); | |
658 | p2m_mid_mfn_init(mid_mfn_p); | |
659 | ||
660 | p2m_top_mfn_p[topidx] = mid_mfn_p; | |
661 | p2m_top_mfn[topidx] = virt_to_mfn(mid_mfn_p); | |
662 | /* Note: we don't set mid_mfn_p[midix] here, | |
663 | * look in early_alloc_p2m_middle */ | |
664 | } | |
665 | return true; | |
666 | } | |
250a41e0 KRW |
667 | |
668 | /* | |
669 | * Skim over the P2M tree looking at pages that are either filled with | |
670 | * INVALID_P2M_ENTRY or with 1:1 PFNs. If found, re-use that page and | |
671 | * replace the P2M leaf with a p2m_missing or p2m_identity. | |
672 | * Stick the old page in the new P2M tree location. | |
673 | */ | |
674 | bool __init early_can_reuse_p2m_middle(unsigned long set_pfn, unsigned long set_mfn) | |
675 | { | |
676 | unsigned topidx; | |
677 | unsigned mididx; | |
678 | unsigned ident_pfns; | |
679 | unsigned inv_pfns; | |
680 | unsigned long *p2m; | |
681 | unsigned long *mid_mfn_p; | |
682 | unsigned idx; | |
683 | unsigned long pfn; | |
684 | ||
685 | /* We only look when this entails a P2M middle layer */ | |
686 | if (p2m_index(set_pfn)) | |
687 | return false; | |
688 | ||
50e90041 | 689 | for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_PER_PAGE) { |
250a41e0 KRW |
690 | topidx = p2m_top_index(pfn); |
691 | ||
692 | if (!p2m_top[topidx]) | |
693 | continue; | |
694 | ||
695 | if (p2m_top[topidx] == p2m_mid_missing) | |
696 | continue; | |
697 | ||
698 | mididx = p2m_mid_index(pfn); | |
699 | p2m = p2m_top[topidx][mididx]; | |
700 | if (!p2m) | |
701 | continue; | |
702 | ||
703 | if ((p2m == p2m_missing) || (p2m == p2m_identity)) | |
704 | continue; | |
705 | ||
706 | if ((unsigned long)p2m == INVALID_P2M_ENTRY) | |
707 | continue; | |
708 | ||
709 | ident_pfns = 0; | |
710 | inv_pfns = 0; | |
711 | for (idx = 0; idx < P2M_PER_PAGE; idx++) { | |
712 | /* IDENTITY_PFNs are 1:1 */ | |
713 | if (p2m[idx] == IDENTITY_FRAME(pfn + idx)) | |
714 | ident_pfns++; | |
715 | else if (p2m[idx] == INVALID_P2M_ENTRY) | |
716 | inv_pfns++; | |
717 | else | |
718 | break; | |
719 | } | |
720 | if ((ident_pfns == P2M_PER_PAGE) || (inv_pfns == P2M_PER_PAGE)) | |
721 | goto found; | |
722 | } | |
723 | return false; | |
724 | found: | |
725 | /* Found one, replace old with p2m_identity or p2m_missing */ | |
726 | p2m_top[topidx][mididx] = (ident_pfns ? p2m_identity : p2m_missing); | |
727 | /* And the other for save/restore.. */ | |
728 | mid_mfn_p = p2m_top_mfn_p[topidx]; | |
729 | /* NOTE: Even if it is a p2m_identity it should still be point to | |
730 | * a page filled with INVALID_P2M_ENTRY entries. */ | |
731 | mid_mfn_p[mididx] = virt_to_mfn(p2m_missing); | |
732 | ||
733 | /* Reset where we want to stick the old page in. */ | |
734 | topidx = p2m_top_index(set_pfn); | |
735 | mididx = p2m_mid_index(set_pfn); | |
736 | ||
737 | /* This shouldn't happen */ | |
738 | if (WARN_ON(p2m_top[topidx] == p2m_mid_missing)) | |
739 | early_alloc_p2m(set_pfn); | |
740 | ||
741 | if (WARN_ON(p2m_top[topidx][mididx] != p2m_missing)) | |
742 | return false; | |
743 | ||
744 | p2m_init(p2m); | |
745 | p2m_top[topidx][mididx] = p2m; | |
746 | mid_mfn_p = p2m_top_mfn_p[topidx]; | |
747 | mid_mfn_p[mididx] = virt_to_mfn(p2m); | |
748 | ||
749 | return true; | |
750 | } | |
940713bb KRW |
751 | bool __init early_set_phys_to_machine(unsigned long pfn, unsigned long mfn) |
752 | { | |
753 | if (unlikely(!__set_phys_to_machine(pfn, mfn))) { | |
754 | if (!early_alloc_p2m(pfn)) | |
755 | return false; | |
756 | ||
250a41e0 KRW |
757 | if (early_can_reuse_p2m_middle(pfn, mfn)) |
758 | return __set_phys_to_machine(pfn, mfn); | |
759 | ||
940713bb KRW |
760 | if (!early_alloc_p2m_middle(pfn, false /* boundary crossover OK!*/)) |
761 | return false; | |
762 | ||
763 | if (!__set_phys_to_machine(pfn, mfn)) | |
764 | return false; | |
765 | } | |
766 | ||
767 | return true; | |
768 | } | |
b83c6e55 | 769 | unsigned long __init set_phys_range_identity(unsigned long pfn_s, |
f4cec35b KRW |
770 | unsigned long pfn_e) |
771 | { | |
772 | unsigned long pfn; | |
773 | ||
774 | if (unlikely(pfn_s >= MAX_P2M_PFN || pfn_e >= MAX_P2M_PFN)) | |
775 | return 0; | |
776 | ||
777 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) | |
778 | return pfn_e - pfn_s; | |
779 | ||
780 | if (pfn_s > pfn_e) | |
781 | return 0; | |
782 | ||
783 | for (pfn = (pfn_s & ~(P2M_MID_PER_PAGE * P2M_PER_PAGE - 1)); | |
784 | pfn < ALIGN(pfn_e, (P2M_MID_PER_PAGE * P2M_PER_PAGE)); | |
785 | pfn += P2M_MID_PER_PAGE * P2M_PER_PAGE) | |
786 | { | |
3f3aaea2 | 787 | WARN_ON(!early_alloc_p2m(pfn)); |
f4cec35b KRW |
788 | } |
789 | ||
cef4cca5 KRW |
790 | early_alloc_p2m_middle(pfn_s, true); |
791 | early_alloc_p2m_middle(pfn_e, true); | |
f4cec35b KRW |
792 | |
793 | for (pfn = pfn_s; pfn < pfn_e; pfn++) | |
794 | if (!__set_phys_to_machine(pfn, IDENTITY_FRAME(pfn))) | |
795 | break; | |
796 | ||
797 | if (!WARN((pfn - pfn_s) != (pfn_e - pfn_s), | |
798 | "Identity mapping failed. We are %ld short of 1-1 mappings!\n", | |
799 | (pfn_e - pfn_s) - (pfn - pfn_s))) | |
800 | printk(KERN_DEBUG "1-1 mapping on %lx->%lx\n", pfn_s, pfn); | |
801 | ||
802 | return pfn - pfn_s; | |
803 | } | |
804 | ||
b5eafe92 JF |
805 | /* Try to install p2m mapping; fail if intermediate bits missing */ |
806 | bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
807 | { | |
808 | unsigned topidx, mididx, idx; | |
809 | ||
2f558d40 SS |
810 | /* don't track P2M changes in autotranslate guests */ |
811 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) | |
6eaa412f | 812 | return true; |
2f558d40 | 813 | |
b5eafe92 JF |
814 | if (unlikely(pfn >= MAX_P2M_PFN)) { |
815 | BUG_ON(mfn != INVALID_P2M_ENTRY); | |
816 | return true; | |
817 | } | |
818 | ||
819 | topidx = p2m_top_index(pfn); | |
820 | mididx = p2m_mid_index(pfn); | |
821 | idx = p2m_index(pfn); | |
822 | ||
f4cec35b KRW |
823 | /* For sparse holes were the p2m leaf has real PFN along with |
824 | * PCI holes, stick in the PFN as the MFN value. | |
825 | */ | |
826 | if (mfn != INVALID_P2M_ENTRY && (mfn & IDENTITY_FRAME_BIT)) { | |
827 | if (p2m_top[topidx][mididx] == p2m_identity) | |
828 | return true; | |
829 | ||
830 | /* Swap over from MISSING to IDENTITY if needed. */ | |
831 | if (p2m_top[topidx][mididx] == p2m_missing) { | |
c7617798 KRW |
832 | WARN_ON(cmpxchg(&p2m_top[topidx][mididx], p2m_missing, |
833 | p2m_identity) != p2m_missing); | |
f4cec35b KRW |
834 | return true; |
835 | } | |
836 | } | |
837 | ||
b5eafe92 JF |
838 | if (p2m_top[topidx][mididx] == p2m_missing) |
839 | return mfn == INVALID_P2M_ENTRY; | |
840 | ||
841 | p2m_top[topidx][mididx][idx] = mfn; | |
842 | ||
843 | return true; | |
844 | } | |
845 | ||
846 | bool set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
847 | { | |
b5eafe92 JF |
848 | if (unlikely(!__set_phys_to_machine(pfn, mfn))) { |
849 | if (!alloc_p2m(pfn)) | |
850 | return false; | |
851 | ||
852 | if (!__set_phys_to_machine(pfn, mfn)) | |
853 | return false; | |
854 | } | |
855 | ||
856 | return true; | |
857 | } | |
448f2831 JF |
858 | |
859 | #define M2P_OVERRIDE_HASH_SHIFT 10 | |
860 | #define M2P_OVERRIDE_HASH (1 << M2P_OVERRIDE_HASH_SHIFT) | |
861 | ||
862 | static RESERVE_BRK_ARRAY(struct list_head, m2p_overrides, M2P_OVERRIDE_HASH); | |
863 | static DEFINE_SPINLOCK(m2p_override_lock); | |
864 | ||
865 | static void __init m2p_override_init(void) | |
866 | { | |
867 | unsigned i; | |
868 | ||
869 | m2p_overrides = extend_brk(sizeof(*m2p_overrides) * M2P_OVERRIDE_HASH, | |
870 | sizeof(unsigned long)); | |
871 | ||
872 | for (i = 0; i < M2P_OVERRIDE_HASH; i++) | |
873 | INIT_LIST_HEAD(&m2p_overrides[i]); | |
874 | } | |
875 | ||
876 | static unsigned long mfn_hash(unsigned long mfn) | |
877 | { | |
878 | return hash_long(mfn, M2P_OVERRIDE_HASH_SHIFT); | |
879 | } | |
880 | ||
881 | /* Add an MFN override for a particular page */ | |
0930bba6 SS |
882 | int m2p_add_override(unsigned long mfn, struct page *page, |
883 | struct gnttab_map_grant_ref *kmap_op) | |
448f2831 JF |
884 | { |
885 | unsigned long flags; | |
87f1d40a | 886 | unsigned long pfn; |
6b08cfeb | 887 | unsigned long uninitialized_var(address); |
87f1d40a JF |
888 | unsigned level; |
889 | pte_t *ptep = NULL; | |
890 | ||
891 | pfn = page_to_pfn(page); | |
892 | if (!PageHighMem(page)) { | |
893 | address = (unsigned long)__va(pfn << PAGE_SHIFT); | |
894 | ptep = lookup_address(address, &level); | |
87f1d40a JF |
895 | if (WARN(ptep == NULL || level != PG_LEVEL_4K, |
896 | "m2p_add_override: pfn %lx not mapped", pfn)) | |
897 | return -EINVAL; | |
898 | } | |
0f4b49ea KRW |
899 | WARN_ON(PagePrivate(page)); |
900 | SetPagePrivate(page); | |
901 | set_page_private(page, mfn); | |
9b705f0e | 902 | page->index = pfn_to_mfn(pfn); |
448f2831 | 903 | |
b254244d DDG |
904 | if (unlikely(!set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)))) |
905 | return -ENOMEM; | |
906 | ||
0930bba6 SS |
907 | if (kmap_op != NULL) { |
908 | if (!PageHighMem(page)) { | |
909 | struct multicall_space mcs = | |
910 | xen_mc_entry(sizeof(*kmap_op)); | |
911 | ||
912 | MULTI_grant_table_op(mcs.mc, | |
913 | GNTTABOP_map_grant_ref, kmap_op, 1); | |
914 | ||
915 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
916 | } | |
0930bba6 | 917 | } |
448f2831 JF |
918 | spin_lock_irqsave(&m2p_override_lock, flags); |
919 | list_add(&page->lru, &m2p_overrides[mfn_hash(mfn)]); | |
920 | spin_unlock_irqrestore(&m2p_override_lock, flags); | |
87f1d40a | 921 | |
b9e0d95c SS |
922 | /* p2m(m2p(mfn)) == mfn: the mfn is already present somewhere in |
923 | * this domain. Set the FOREIGN_FRAME_BIT in the p2m for the other | |
924 | * pfn so that the following mfn_to_pfn(mfn) calls will return the | |
925 | * pfn from the m2p_override (the backend pfn) instead. | |
926 | * We need to do this because the pages shared by the frontend | |
927 | * (xen-blkfront) can be already locked (lock_page, called by | |
928 | * do_read_cache_page); when the userspace backend tries to use them | |
929 | * with direct_IO, mfn_to_pfn returns the pfn of the frontend, so | |
930 | * do_blockdev_direct_IO is going to try to lock the same pages | |
931 | * again resulting in a deadlock. | |
932 | * As a side effect get_user_pages_fast might not be safe on the | |
933 | * frontend pages while they are being shared with the backend, | |
934 | * because mfn_to_pfn (that ends up being called by GUPF) will | |
935 | * return the backend pfn rather than the frontend pfn. */ | |
0160676b DV |
936 | pfn = mfn_to_pfn_no_overrides(mfn); |
937 | if (get_phys_to_machine(pfn) == mfn) | |
b9e0d95c SS |
938 | set_phys_to_machine(pfn, FOREIGN_FRAME(mfn)); |
939 | ||
87f1d40a | 940 | return 0; |
448f2831 | 941 | } |
8a91707d | 942 | EXPORT_SYMBOL_GPL(m2p_add_override); |
2fc136ee SS |
943 | int m2p_remove_override(struct page *page, |
944 | struct gnttab_map_grant_ref *kmap_op) | |
448f2831 JF |
945 | { |
946 | unsigned long flags; | |
9b705f0e SS |
947 | unsigned long mfn; |
948 | unsigned long pfn; | |
6b08cfeb | 949 | unsigned long uninitialized_var(address); |
87f1d40a JF |
950 | unsigned level; |
951 | pte_t *ptep = NULL; | |
9b705f0e SS |
952 | |
953 | pfn = page_to_pfn(page); | |
954 | mfn = get_phys_to_machine(pfn); | |
955 | if (mfn == INVALID_P2M_ENTRY || !(mfn & FOREIGN_FRAME_BIT)) | |
87f1d40a JF |
956 | return -EINVAL; |
957 | ||
958 | if (!PageHighMem(page)) { | |
959 | address = (unsigned long)__va(pfn << PAGE_SHIFT); | |
960 | ptep = lookup_address(address, &level); | |
961 | ||
962 | if (WARN(ptep == NULL || level != PG_LEVEL_4K, | |
963 | "m2p_remove_override: pfn %lx not mapped", pfn)) | |
964 | return -EINVAL; | |
965 | } | |
9b705f0e | 966 | |
448f2831 JF |
967 | spin_lock_irqsave(&m2p_override_lock, flags); |
968 | list_del(&page->lru); | |
969 | spin_unlock_irqrestore(&m2p_override_lock, flags); | |
0f4b49ea KRW |
970 | WARN_ON(!PagePrivate(page)); |
971 | ClearPagePrivate(page); | |
87f1d40a | 972 | |
2fc136ee SS |
973 | set_phys_to_machine(pfn, page->index); |
974 | if (kmap_op != NULL) { | |
0930bba6 SS |
975 | if (!PageHighMem(page)) { |
976 | struct multicall_space mcs; | |
ee072640 SS |
977 | struct gnttab_unmap_and_replace *unmap_op; |
978 | struct page *scratch_page = get_balloon_scratch_page(); | |
979 | unsigned long scratch_page_address = (unsigned long) | |
980 | __va(page_to_pfn(scratch_page) << PAGE_SHIFT); | |
0930bba6 SS |
981 | |
982 | /* | |
983 | * It might be that we queued all the m2p grant table | |
984 | * hypercalls in a multicall, then m2p_remove_override | |
985 | * get called before the multicall has actually been | |
986 | * issued. In this case handle is going to -1 because | |
987 | * it hasn't been modified yet. | |
988 | */ | |
2fc136ee | 989 | if (kmap_op->handle == -1) |
0930bba6 SS |
990 | xen_mc_flush(); |
991 | /* | |
2fc136ee | 992 | * Now if kmap_op->handle is negative it means that the |
0930bba6 SS |
993 | * hypercall actually returned an error. |
994 | */ | |
2fc136ee | 995 | if (kmap_op->handle == GNTST_general_error) { |
0930bba6 SS |
996 | printk(KERN_WARNING "m2p_remove_override: " |
997 | "pfn %lx mfn %lx, failed to modify kernel mappings", | |
998 | pfn, mfn); | |
d7f8f48d | 999 | put_balloon_scratch_page(); |
0930bba6 SS |
1000 | return -1; |
1001 | } | |
1002 | ||
d7f8f48d BO |
1003 | xen_mc_batch(); |
1004 | ||
1005 | mcs = __xen_mc_entry( | |
ee072640 | 1006 | sizeof(struct gnttab_unmap_and_replace)); |
0930bba6 | 1007 | unmap_op = mcs.args; |
2fc136ee | 1008 | unmap_op->host_addr = kmap_op->host_addr; |
ee072640 | 1009 | unmap_op->new_addr = scratch_page_address; |
2fc136ee | 1010 | unmap_op->handle = kmap_op->handle; |
0930bba6 SS |
1011 | |
1012 | MULTI_grant_table_op(mcs.mc, | |
ee072640 | 1013 | GNTTABOP_unmap_and_replace, unmap_op, 1); |
0930bba6 | 1014 | |
ee072640 SS |
1015 | mcs = __xen_mc_entry(0); |
1016 | MULTI_update_va_mapping(mcs.mc, scratch_page_address, | |
d7f8f48d | 1017 | pfn_pte(page_to_pfn(scratch_page), |
ee072640 | 1018 | PAGE_KERNEL_RO), 0); |
d7f8f48d | 1019 | |
ee072640 SS |
1020 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1021 | ||
2fc136ee | 1022 | kmap_op->host_addr = 0; |
ee072640 | 1023 | put_balloon_scratch_page(); |
0930bba6 | 1024 | } |
2fc136ee | 1025 | } |
87f1d40a | 1026 | |
b9e0d95c SS |
1027 | /* p2m(m2p(mfn)) == FOREIGN_FRAME(mfn): the mfn is already present |
1028 | * somewhere in this domain, even before being added to the | |
1029 | * m2p_override (see comment above in m2p_add_override). | |
1030 | * If there are no other entries in the m2p_override corresponding | |
1031 | * to this mfn, then remove the FOREIGN_FRAME_BIT from the p2m for | |
1032 | * the original pfn (the one shared by the frontend): the backend | |
1033 | * cannot do any IO on this page anymore because it has been | |
1034 | * unshared. Removing the FOREIGN_FRAME_BIT from the p2m entry of | |
1035 | * the original pfn causes mfn_to_pfn(mfn) to return the frontend | |
1036 | * pfn again. */ | |
1037 | mfn &= ~FOREIGN_FRAME_BIT; | |
0160676b DV |
1038 | pfn = mfn_to_pfn_no_overrides(mfn); |
1039 | if (get_phys_to_machine(pfn) == FOREIGN_FRAME(mfn) && | |
b9e0d95c SS |
1040 | m2p_find_override(mfn) == NULL) |
1041 | set_phys_to_machine(pfn, mfn); | |
1042 | ||
87f1d40a | 1043 | return 0; |
448f2831 | 1044 | } |
8a91707d | 1045 | EXPORT_SYMBOL_GPL(m2p_remove_override); |
448f2831 JF |
1046 | |
1047 | struct page *m2p_find_override(unsigned long mfn) | |
1048 | { | |
1049 | unsigned long flags; | |
1050 | struct list_head *bucket = &m2p_overrides[mfn_hash(mfn)]; | |
1051 | struct page *p, *ret; | |
1052 | ||
1053 | ret = NULL; | |
1054 | ||
1055 | spin_lock_irqsave(&m2p_override_lock, flags); | |
1056 | ||
1057 | list_for_each_entry(p, bucket, lru) { | |
0f4b49ea | 1058 | if (page_private(p) == mfn) { |
448f2831 JF |
1059 | ret = p; |
1060 | break; | |
1061 | } | |
1062 | } | |
1063 | ||
1064 | spin_unlock_irqrestore(&m2p_override_lock, flags); | |
1065 | ||
1066 | return ret; | |
1067 | } | |
1068 | ||
1069 | unsigned long m2p_find_override_pfn(unsigned long mfn, unsigned long pfn) | |
1070 | { | |
1071 | struct page *p = m2p_find_override(mfn); | |
1072 | unsigned long ret = pfn; | |
1073 | ||
1074 | if (p) | |
1075 | ret = page_to_pfn(p); | |
1076 | ||
1077 | return ret; | |
1078 | } | |
e1b478e4 | 1079 | EXPORT_SYMBOL_GPL(m2p_find_override_pfn); |
2222e71b KRW |
1080 | |
1081 | #ifdef CONFIG_XEN_DEBUG_FS | |
a867db10 KRW |
1082 | #include <linux/debugfs.h> |
1083 | #include "debugfs.h" | |
1084 | static int p2m_dump_show(struct seq_file *m, void *v) | |
2222e71b KRW |
1085 | { |
1086 | static const char * const level_name[] = { "top", "middle", | |
8404877e | 1087 | "entry", "abnormal", "error"}; |
2222e71b KRW |
1088 | #define TYPE_IDENTITY 0 |
1089 | #define TYPE_MISSING 1 | |
1090 | #define TYPE_PFN 2 | |
1091 | #define TYPE_UNKNOWN 3 | |
a491dbef KRW |
1092 | static const char * const type_name[] = { |
1093 | [TYPE_IDENTITY] = "identity", | |
1094 | [TYPE_MISSING] = "missing", | |
1095 | [TYPE_PFN] = "pfn", | |
1096 | [TYPE_UNKNOWN] = "abnormal"}; | |
2222e71b KRW |
1097 | unsigned long pfn, prev_pfn_type = 0, prev_pfn_level = 0; |
1098 | unsigned int uninitialized_var(prev_level); | |
1099 | unsigned int uninitialized_var(prev_type); | |
1100 | ||
1101 | if (!p2m_top) | |
1102 | return 0; | |
1103 | ||
1104 | for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn++) { | |
1105 | unsigned topidx = p2m_top_index(pfn); | |
1106 | unsigned mididx = p2m_mid_index(pfn); | |
1107 | unsigned idx = p2m_index(pfn); | |
1108 | unsigned lvl, type; | |
1109 | ||
1110 | lvl = 4; | |
1111 | type = TYPE_UNKNOWN; | |
1112 | if (p2m_top[topidx] == p2m_mid_missing) { | |
1113 | lvl = 0; type = TYPE_MISSING; | |
1114 | } else if (p2m_top[topidx] == NULL) { | |
1115 | lvl = 0; type = TYPE_UNKNOWN; | |
1116 | } else if (p2m_top[topidx][mididx] == NULL) { | |
1117 | lvl = 1; type = TYPE_UNKNOWN; | |
1118 | } else if (p2m_top[topidx][mididx] == p2m_identity) { | |
1119 | lvl = 1; type = TYPE_IDENTITY; | |
1120 | } else if (p2m_top[topidx][mididx] == p2m_missing) { | |
1121 | lvl = 1; type = TYPE_MISSING; | |
1122 | } else if (p2m_top[topidx][mididx][idx] == 0) { | |
1123 | lvl = 2; type = TYPE_UNKNOWN; | |
1124 | } else if (p2m_top[topidx][mididx][idx] == IDENTITY_FRAME(pfn)) { | |
1125 | lvl = 2; type = TYPE_IDENTITY; | |
1126 | } else if (p2m_top[topidx][mididx][idx] == INVALID_P2M_ENTRY) { | |
1127 | lvl = 2; type = TYPE_MISSING; | |
1128 | } else if (p2m_top[topidx][mididx][idx] == pfn) { | |
1129 | lvl = 2; type = TYPE_PFN; | |
1130 | } else if (p2m_top[topidx][mididx][idx] != pfn) { | |
1131 | lvl = 2; type = TYPE_PFN; | |
1132 | } | |
1133 | if (pfn == 0) { | |
1134 | prev_level = lvl; | |
1135 | prev_type = type; | |
1136 | } | |
1137 | if (pfn == MAX_DOMAIN_PAGES-1) { | |
1138 | lvl = 3; | |
1139 | type = TYPE_UNKNOWN; | |
1140 | } | |
1141 | if (prev_type != type) { | |
1142 | seq_printf(m, " [0x%lx->0x%lx] %s\n", | |
1143 | prev_pfn_type, pfn, type_name[prev_type]); | |
1144 | prev_pfn_type = pfn; | |
1145 | prev_type = type; | |
1146 | } | |
1147 | if (prev_level != lvl) { | |
1148 | seq_printf(m, " [0x%lx->0x%lx] level %s\n", | |
1149 | prev_pfn_level, pfn, level_name[prev_level]); | |
1150 | prev_pfn_level = pfn; | |
1151 | prev_level = lvl; | |
1152 | } | |
1153 | } | |
1154 | return 0; | |
1155 | #undef TYPE_IDENTITY | |
1156 | #undef TYPE_MISSING | |
1157 | #undef TYPE_PFN | |
1158 | #undef TYPE_UNKNOWN | |
1159 | } | |
a867db10 KRW |
1160 | |
1161 | static int p2m_dump_open(struct inode *inode, struct file *filp) | |
1162 | { | |
1163 | return single_open(filp, p2m_dump_show, NULL); | |
1164 | } | |
1165 | ||
1166 | static const struct file_operations p2m_dump_fops = { | |
1167 | .open = p2m_dump_open, | |
1168 | .read = seq_read, | |
1169 | .llseek = seq_lseek, | |
1170 | .release = single_release, | |
1171 | }; | |
1172 | ||
1173 | static struct dentry *d_mmu_debug; | |
1174 | ||
1175 | static int __init xen_p2m_debugfs(void) | |
1176 | { | |
1177 | struct dentry *d_xen = xen_init_debugfs(); | |
1178 | ||
1179 | if (d_xen == NULL) | |
1180 | return -ENOMEM; | |
1181 | ||
1182 | d_mmu_debug = debugfs_create_dir("mmu", d_xen); | |
1183 | ||
1184 | debugfs_create_file("p2m", 0600, d_mmu_debug, NULL, &p2m_dump_fops); | |
1185 | return 0; | |
1186 | } | |
1187 | fs_initcall(xen_p2m_debugfs); | |
1188 | #endif /* CONFIG_XEN_DEBUG_FS */ |