]>
Commit | Line | Data |
---|---|---|
5ead97c8 JF |
1 | /* |
2 | * Machine specific setup for xen | |
3 | * | |
4 | * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 | |
5 | */ | |
6 | ||
7 | #include <linux/module.h> | |
8 | #include <linux/sched.h> | |
9 | #include <linux/mm.h> | |
10 | #include <linux/pm.h> | |
a9ce6bc1 | 11 | #include <linux/memblock.h> |
d91ee586 | 12 | #include <linux/cpuidle.h> |
48cdd828 | 13 | #include <linux/cpufreq.h> |
5ead97c8 JF |
14 | |
15 | #include <asm/elf.h> | |
6c3652ef | 16 | #include <asm/vdso.h> |
5ead97c8 JF |
17 | #include <asm/e820.h> |
18 | #include <asm/setup.h> | |
b792c755 | 19 | #include <asm/acpi.h> |
8d54db79 | 20 | #include <asm/numa.h> |
5ead97c8 JF |
21 | #include <asm/xen/hypervisor.h> |
22 | #include <asm/xen/hypercall.h> | |
23 | ||
45263cb0 | 24 | #include <xen/xen.h> |
8006ec3e | 25 | #include <xen/page.h> |
e2a81baf | 26 | #include <xen/interface/callback.h> |
35ae11fd | 27 | #include <xen/interface/memory.h> |
5ead97c8 JF |
28 | #include <xen/interface/physdev.h> |
29 | #include <xen/features.h> | |
5ead97c8 | 30 | #include "xen-ops.h" |
d2eea68e | 31 | #include "vdso.h" |
4fbb67e3 | 32 | #include "p2m.h" |
1f3ac86b | 33 | #include "mmu.h" |
5ead97c8 JF |
34 | |
35 | /* These are code, but not functions. Defined in entry.S */ | |
36 | extern const char xen_hypervisor_callback[]; | |
37 | extern const char xen_failsafe_callback[]; | |
6efa20e4 | 38 | #ifdef CONFIG_X86_64 |
07ba06d9 | 39 | extern asmlinkage void nmi(void); |
6efa20e4 | 40 | #endif |
f63c2f24 T |
41 | extern void xen_sysenter_target(void); |
42 | extern void xen_syscall_target(void); | |
43 | extern void xen_syscall32_target(void); | |
5ead97c8 | 44 | |
42ee1471 | 45 | /* Amount of extra memory space we add to the e820 ranges */ |
8b5d44a5 | 46 | struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; |
42ee1471 | 47 | |
aa24411b DV |
48 | /* Number of pages released from the initial allocation. */ |
49 | unsigned long xen_released_pages; | |
50 | ||
1f3ac86b JG |
51 | /* |
52 | * Buffer used to remap identity mapped pages. We only need the virtual space. | |
53 | * The physical page behind this address is remapped as needed to different | |
54 | * buffer pages. | |
55 | */ | |
56 | #define REMAP_SIZE (P2M_PER_PAGE - 3) | |
57 | static struct { | |
58 | unsigned long next_area_mfn; | |
59 | unsigned long target_pfn; | |
60 | unsigned long size; | |
61 | unsigned long mfns[REMAP_SIZE]; | |
62 | } xen_remap_buf __initdata __aligned(PAGE_SIZE); | |
63 | static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY; | |
4fbb67e3 | 64 | |
698bb8d1 JF |
65 | /* |
66 | * The maximum amount of extra memory compared to the base size. The | |
67 | * main scaling factor is the size of struct page. At extreme ratios | |
68 | * of base:extra, all the base memory can be filled with page | |
69 | * structures for the extra memory, leaving no space for anything | |
70 | * else. | |
71 | * | |
72 | * 10x seems like a reasonable balance between scaling flexibility and | |
73 | * leaving a practically usable system. | |
74 | */ | |
75 | #define EXTRA_MEM_RATIO (10) | |
76 | ||
dc91c728 | 77 | static void __init xen_add_extra_mem(u64 start, u64 size) |
42ee1471 | 78 | { |
dc91c728 | 79 | int i; |
6eaa412f | 80 | |
dc91c728 DV |
81 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
82 | /* Add new region. */ | |
83 | if (xen_extra_mem[i].size == 0) { | |
84 | xen_extra_mem[i].start = start; | |
85 | xen_extra_mem[i].size = size; | |
86 | break; | |
87 | } | |
88 | /* Append to existing region. */ | |
89 | if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) { | |
90 | xen_extra_mem[i].size += size; | |
91 | break; | |
92 | } | |
93 | } | |
94 | if (i == XEN_EXTRA_MEM_MAX_REGIONS) | |
95 | printk(KERN_WARNING "Warning: not enough extra memory regions\n"); | |
42ee1471 | 96 | |
d4bbf7e7 | 97 | memblock_reserve(start, size); |
5b8e7d80 | 98 | } |
2f7acb20 | 99 | |
5b8e7d80 JG |
100 | static void __init xen_del_extra_mem(u64 start, u64 size) |
101 | { | |
102 | int i; | |
103 | u64 start_r, size_r; | |
c96aae1f | 104 | |
5b8e7d80 JG |
105 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
106 | start_r = xen_extra_mem[i].start; | |
107 | size_r = xen_extra_mem[i].size; | |
108 | ||
109 | /* Start of region. */ | |
110 | if (start_r == start) { | |
111 | BUG_ON(size > size_r); | |
112 | xen_extra_mem[i].start += size; | |
113 | xen_extra_mem[i].size -= size; | |
114 | break; | |
115 | } | |
116 | /* End of region. */ | |
117 | if (start_r + size_r == start + size) { | |
118 | BUG_ON(size > size_r); | |
119 | xen_extra_mem[i].size -= size; | |
120 | break; | |
121 | } | |
122 | /* Mid of region. */ | |
123 | if (start > start_r && start < start_r + size_r) { | |
124 | BUG_ON(start + size > start_r + size_r); | |
125 | xen_extra_mem[i].size = start - start_r; | |
126 | /* Calling memblock_reserve() again is okay. */ | |
127 | xen_add_extra_mem(start + size, start_r + size_r - | |
128 | (start + size)); | |
129 | break; | |
130 | } | |
131 | } | |
132 | memblock_free(start, size); | |
133 | } | |
134 | ||
135 | /* | |
136 | * Called during boot before the p2m list can take entries beyond the | |
137 | * hypervisor supplied p2m list. Entries in extra mem are to be regarded as | |
138 | * invalid. | |
139 | */ | |
140 | unsigned long __ref xen_chk_extra_mem(unsigned long pfn) | |
141 | { | |
142 | int i; | |
143 | unsigned long addr = PFN_PHYS(pfn); | |
6eaa412f | 144 | |
5b8e7d80 JG |
145 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
146 | if (addr >= xen_extra_mem[i].start && | |
147 | addr < xen_extra_mem[i].start + xen_extra_mem[i].size) | |
148 | return INVALID_P2M_ENTRY; | |
149 | } | |
150 | ||
151 | return IDENTITY_FRAME(pfn); | |
152 | } | |
153 | ||
154 | /* | |
155 | * Mark all pfns of extra mem as invalid in p2m list. | |
156 | */ | |
157 | void __init xen_inv_extra_mem(void) | |
158 | { | |
159 | unsigned long pfn, pfn_s, pfn_e; | |
160 | int i; | |
161 | ||
162 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { | |
163 | pfn_s = PFN_DOWN(xen_extra_mem[i].start); | |
164 | pfn_e = PFN_UP(xen_extra_mem[i].start + xen_extra_mem[i].size); | |
165 | for (pfn = pfn_s; pfn < pfn_e; pfn++) | |
166 | set_phys_to_machine(pfn, INVALID_P2M_ENTRY); | |
c96aae1f | 167 | } |
42ee1471 JF |
168 | } |
169 | ||
4fbb67e3 MR |
170 | /* |
171 | * Finds the next RAM pfn available in the E820 map after min_pfn. | |
172 | * This function updates min_pfn with the pfn found and returns | |
173 | * the size of that range or zero if not found. | |
174 | */ | |
175 | static unsigned long __init xen_find_pfn_range( | |
2e2fb754 | 176 | const struct e820entry *list, size_t map_size, |
4fbb67e3 | 177 | unsigned long *min_pfn) |
2e2fb754 KRW |
178 | { |
179 | const struct e820entry *entry; | |
180 | unsigned int i; | |
181 | unsigned long done = 0; | |
2e2fb754 KRW |
182 | |
183 | for (i = 0, entry = list; i < map_size; i++, entry++) { | |
2e2fb754 KRW |
184 | unsigned long s_pfn; |
185 | unsigned long e_pfn; | |
2e2fb754 KRW |
186 | |
187 | if (entry->type != E820_RAM) | |
188 | continue; | |
189 | ||
c3d93f88 | 190 | e_pfn = PFN_DOWN(entry->addr + entry->size); |
2e2fb754 | 191 | |
4fbb67e3 MR |
192 | /* We only care about E820 after this */ |
193 | if (e_pfn < *min_pfn) | |
2e2fb754 KRW |
194 | continue; |
195 | ||
c3d93f88 | 196 | s_pfn = PFN_UP(entry->addr); |
4fbb67e3 MR |
197 | |
198 | /* If min_pfn falls within the E820 entry, we want to start | |
199 | * at the min_pfn PFN. | |
2e2fb754 | 200 | */ |
4fbb67e3 MR |
201 | if (s_pfn <= *min_pfn) { |
202 | done = e_pfn - *min_pfn; | |
2e2fb754 | 203 | } else { |
4fbb67e3 MR |
204 | done = e_pfn - s_pfn; |
205 | *min_pfn = s_pfn; | |
2e2fb754 | 206 | } |
4fbb67e3 MR |
207 | break; |
208 | } | |
2e2fb754 | 209 | |
4fbb67e3 MR |
210 | return done; |
211 | } | |
2e2fb754 | 212 | |
1f3ac86b JG |
213 | static int __init xen_free_mfn(unsigned long mfn) |
214 | { | |
215 | struct xen_memory_reservation reservation = { | |
216 | .address_bits = 0, | |
217 | .extent_order = 0, | |
218 | .domid = DOMID_SELF | |
219 | }; | |
220 | ||
221 | set_xen_guest_handle(reservation.extent_start, &mfn); | |
222 | reservation.nr_extents = 1; | |
223 | ||
224 | return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); | |
225 | } | |
226 | ||
4fbb67e3 | 227 | /* |
1f3ac86b | 228 | * This releases a chunk of memory and then does the identity map. It's used |
4fbb67e3 MR |
229 | * as a fallback if the remapping fails. |
230 | */ | |
231 | static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn, | |
232 | unsigned long end_pfn, unsigned long nr_pages, unsigned long *identity, | |
233 | unsigned long *released) | |
234 | { | |
1f3ac86b JG |
235 | unsigned long len = 0; |
236 | unsigned long pfn, end; | |
237 | int ret; | |
238 | ||
4fbb67e3 MR |
239 | WARN_ON(start_pfn > end_pfn); |
240 | ||
1f3ac86b JG |
241 | end = min(end_pfn, nr_pages); |
242 | for (pfn = start_pfn; pfn < end; pfn++) { | |
243 | unsigned long mfn = pfn_to_mfn(pfn); | |
244 | ||
245 | /* Make sure pfn exists to start with */ | |
246 | if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn) | |
247 | continue; | |
248 | ||
249 | ret = xen_free_mfn(mfn); | |
250 | WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret); | |
251 | ||
252 | if (ret == 1) { | |
253 | if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY)) | |
254 | break; | |
255 | len++; | |
256 | } else | |
257 | break; | |
258 | } | |
259 | ||
4fbb67e3 | 260 | /* Need to release pages first */ |
1f3ac86b | 261 | *released += len; |
4fbb67e3 MR |
262 | *identity += set_phys_range_identity(start_pfn, end_pfn); |
263 | } | |
264 | ||
265 | /* | |
1f3ac86b | 266 | * Helper function to update the p2m and m2p tables and kernel mapping. |
4fbb67e3 | 267 | */ |
1f3ac86b | 268 | static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn) |
4fbb67e3 MR |
269 | { |
270 | struct mmu_update update = { | |
271 | .ptr = ((unsigned long long)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE, | |
272 | .val = pfn | |
273 | }; | |
274 | ||
275 | /* Update p2m */ | |
1f3ac86b | 276 | if (!set_phys_to_machine(pfn, mfn)) { |
4fbb67e3 MR |
277 | WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n", |
278 | pfn, mfn); | |
1f3ac86b | 279 | BUG(); |
2e2fb754 | 280 | } |
4fbb67e3 MR |
281 | |
282 | /* Update m2p */ | |
283 | if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) { | |
284 | WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n", | |
285 | mfn, pfn); | |
1f3ac86b | 286 | BUG(); |
4fbb67e3 MR |
287 | } |
288 | ||
1f3ac86b JG |
289 | /* Update kernel mapping, but not for highmem. */ |
290 | if ((pfn << PAGE_SHIFT) >= __pa(high_memory)) | |
291 | return; | |
292 | ||
293 | if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT), | |
294 | mfn_pte(mfn, PAGE_KERNEL), 0)) { | |
295 | WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n", | |
296 | mfn, pfn); | |
297 | BUG(); | |
298 | } | |
2e2fb754 | 299 | } |
83d51ab4 | 300 | |
4fbb67e3 MR |
301 | /* |
302 | * This function updates the p2m and m2p tables with an identity map from | |
1f3ac86b JG |
303 | * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the |
304 | * original allocation at remap_pfn. The information needed for remapping is | |
305 | * saved in the memory itself to avoid the need for allocating buffers. The | |
306 | * complete remap information is contained in a list of MFNs each containing | |
307 | * up to REMAP_SIZE MFNs and the start target PFN for doing the remap. | |
308 | * This enables us to preserve the original mfn sequence while doing the | |
309 | * remapping at a time when the memory management is capable of allocating | |
310 | * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and | |
311 | * its callers. | |
4fbb67e3 | 312 | */ |
1f3ac86b | 313 | static void __init xen_do_set_identity_and_remap_chunk( |
4fbb67e3 | 314 | unsigned long start_pfn, unsigned long size, unsigned long remap_pfn) |
83d51ab4 | 315 | { |
1f3ac86b JG |
316 | unsigned long buf = (unsigned long)&xen_remap_buf; |
317 | unsigned long mfn_save, mfn; | |
4fbb67e3 | 318 | unsigned long ident_pfn_iter, remap_pfn_iter; |
1f3ac86b | 319 | unsigned long ident_end_pfn = start_pfn + size; |
4fbb67e3 | 320 | unsigned long left = size; |
1f3ac86b JG |
321 | unsigned long ident_cnt = 0; |
322 | unsigned int i, chunk; | |
4fbb67e3 MR |
323 | |
324 | WARN_ON(size == 0); | |
325 | ||
326 | BUG_ON(xen_feature(XENFEAT_auto_translated_physmap)); | |
83d51ab4 | 327 | |
1f3ac86b | 328 | mfn_save = virt_to_mfn(buf); |
e201bfcc | 329 | |
1f3ac86b JG |
330 | for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn; |
331 | ident_pfn_iter < ident_end_pfn; | |
332 | ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) { | |
333 | chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE; | |
4fbb67e3 | 334 | |
1f3ac86b JG |
335 | /* Map first pfn to xen_remap_buf */ |
336 | mfn = pfn_to_mfn(ident_pfn_iter); | |
337 | set_pte_mfn(buf, mfn, PAGE_KERNEL); | |
4fbb67e3 | 338 | |
1f3ac86b JG |
339 | /* Save mapping information in page */ |
340 | xen_remap_buf.next_area_mfn = xen_remap_mfn; | |
341 | xen_remap_buf.target_pfn = remap_pfn_iter; | |
342 | xen_remap_buf.size = chunk; | |
343 | for (i = 0; i < chunk; i++) | |
344 | xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i); | |
4fbb67e3 | 345 | |
1f3ac86b JG |
346 | /* Put remap buf into list. */ |
347 | xen_remap_mfn = mfn; | |
4fbb67e3 | 348 | |
1f3ac86b JG |
349 | /* Set identity map */ |
350 | ident_cnt += set_phys_range_identity(ident_pfn_iter, | |
351 | ident_pfn_iter + chunk); | |
83d51ab4 | 352 | |
1f3ac86b | 353 | left -= chunk; |
4fbb67e3 | 354 | } |
83d51ab4 | 355 | |
1f3ac86b JG |
356 | /* Restore old xen_remap_buf mapping */ |
357 | set_pte_mfn(buf, mfn_save, PAGE_KERNEL); | |
83d51ab4 DV |
358 | } |
359 | ||
4fbb67e3 MR |
360 | /* |
361 | * This function takes a contiguous pfn range that needs to be identity mapped | |
362 | * and: | |
363 | * | |
364 | * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn. | |
365 | * 2) Calls the do_ function to actually do the mapping/remapping work. | |
366 | * | |
367 | * The goal is to not allocate additional memory but to remap the existing | |
368 | * pages. In the case of an error the underlying memory is simply released back | |
369 | * to Xen and not remapped. | |
370 | */ | |
5b8e7d80 | 371 | static unsigned long xen_set_identity_and_remap_chunk( |
4fbb67e3 MR |
372 | const struct e820entry *list, size_t map_size, unsigned long start_pfn, |
373 | unsigned long end_pfn, unsigned long nr_pages, unsigned long remap_pfn, | |
1f3ac86b | 374 | unsigned long *identity, unsigned long *released) |
4fbb67e3 MR |
375 | { |
376 | unsigned long pfn; | |
377 | unsigned long i = 0; | |
378 | unsigned long n = end_pfn - start_pfn; | |
379 | ||
380 | while (i < n) { | |
381 | unsigned long cur_pfn = start_pfn + i; | |
382 | unsigned long left = n - i; | |
383 | unsigned long size = left; | |
384 | unsigned long remap_range_size; | |
385 | ||
386 | /* Do not remap pages beyond the current allocation */ | |
387 | if (cur_pfn >= nr_pages) { | |
388 | /* Identity map remaining pages */ | |
389 | *identity += set_phys_range_identity(cur_pfn, | |
390 | cur_pfn + size); | |
391 | break; | |
392 | } | |
393 | if (cur_pfn + size > nr_pages) | |
394 | size = nr_pages - cur_pfn; | |
395 | ||
396 | remap_range_size = xen_find_pfn_range(list, map_size, | |
397 | &remap_pfn); | |
398 | if (!remap_range_size) { | |
399 | pr_warning("Unable to find available pfn range, not remapping identity pages\n"); | |
400 | xen_set_identity_and_release_chunk(cur_pfn, | |
401 | cur_pfn + left, nr_pages, identity, released); | |
402 | break; | |
403 | } | |
404 | /* Adjust size to fit in current e820 RAM region */ | |
405 | if (size > remap_range_size) | |
406 | size = remap_range_size; | |
407 | ||
1f3ac86b | 408 | xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn); |
4fbb67e3 MR |
409 | |
410 | /* Update variables to reflect new mappings. */ | |
411 | i += size; | |
412 | remap_pfn += size; | |
413 | *identity += size; | |
4fbb67e3 MR |
414 | } |
415 | ||
416 | /* | |
417 | * If the PFNs are currently mapped, the VA mapping also needs | |
418 | * to be updated to be 1:1. | |
419 | */ | |
420 | for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++) | |
421 | (void)HYPERVISOR_update_va_mapping( | |
422 | (unsigned long)__va(pfn << PAGE_SHIFT), | |
423 | mfn_pte(pfn, PAGE_KERNEL_IO), 0); | |
424 | ||
425 | return remap_pfn; | |
426 | } | |
427 | ||
5b8e7d80 | 428 | static void __init xen_set_identity_and_remap( |
4fbb67e3 MR |
429 | const struct e820entry *list, size_t map_size, unsigned long nr_pages, |
430 | unsigned long *released) | |
093d7b46 | 431 | { |
f3f436e3 | 432 | phys_addr_t start = 0; |
68df0da7 | 433 | unsigned long identity = 0; |
4fbb67e3 | 434 | unsigned long last_pfn = nr_pages; |
f3f436e3 | 435 | const struct e820entry *entry; |
4fbb67e3 | 436 | unsigned long num_released = 0; |
68df0da7 KRW |
437 | int i; |
438 | ||
f3f436e3 DV |
439 | /* |
440 | * Combine non-RAM regions and gaps until a RAM region (or the | |
441 | * end of the map) is reached, then set the 1:1 map and | |
4fbb67e3 | 442 | * remap the memory in those non-RAM regions. |
f3f436e3 DV |
443 | * |
444 | * The combined non-RAM regions are rounded to a whole number | |
445 | * of pages so any partial pages are accessible via the 1:1 | |
446 | * mapping. This is needed for some BIOSes that put (for | |
447 | * example) the DMI tables in a reserved region that begins on | |
448 | * a non-page boundary. | |
449 | */ | |
68df0da7 | 450 | for (i = 0, entry = list; i < map_size; i++, entry++) { |
f3f436e3 | 451 | phys_addr_t end = entry->addr + entry->size; |
f3f436e3 DV |
452 | if (entry->type == E820_RAM || i == map_size - 1) { |
453 | unsigned long start_pfn = PFN_DOWN(start); | |
454 | unsigned long end_pfn = PFN_UP(end); | |
68df0da7 | 455 | |
f3f436e3 DV |
456 | if (entry->type == E820_RAM) |
457 | end_pfn = PFN_UP(entry->addr); | |
68df0da7 | 458 | |
83d51ab4 | 459 | if (start_pfn < end_pfn) |
4fbb67e3 MR |
460 | last_pfn = xen_set_identity_and_remap_chunk( |
461 | list, map_size, start_pfn, | |
462 | end_pfn, nr_pages, last_pfn, | |
1f3ac86b | 463 | &identity, &num_released); |
f3f436e3 | 464 | start = end; |
68df0da7 | 465 | } |
68df0da7 | 466 | } |
f3f436e3 | 467 | |
4fbb67e3 | 468 | *released = num_released; |
f3f436e3 | 469 | |
4fbb67e3 | 470 | pr_info("Set %ld page(s) to 1-1 mapping\n", identity); |
4fbb67e3 | 471 | pr_info("Released %ld page(s)\n", num_released); |
4fbb67e3 | 472 | } |
1f3ac86b JG |
473 | |
474 | /* | |
475 | * Remap the memory prepared in xen_do_set_identity_and_remap_chunk(). | |
476 | * The remap information (which mfn remap to which pfn) is contained in the | |
477 | * to be remapped memory itself in a linked list anchored at xen_remap_mfn. | |
478 | * This scheme allows to remap the different chunks in arbitrary order while | |
479 | * the resulting mapping will be independant from the order. | |
480 | */ | |
481 | void __init xen_remap_memory(void) | |
482 | { | |
483 | unsigned long buf = (unsigned long)&xen_remap_buf; | |
484 | unsigned long mfn_save, mfn, pfn; | |
485 | unsigned long remapped = 0; | |
486 | unsigned int i; | |
487 | unsigned long pfn_s = ~0UL; | |
488 | unsigned long len = 0; | |
489 | ||
490 | mfn_save = virt_to_mfn(buf); | |
491 | ||
492 | while (xen_remap_mfn != INVALID_P2M_ENTRY) { | |
493 | /* Map the remap information */ | |
494 | set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL); | |
495 | ||
496 | BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]); | |
497 | ||
498 | pfn = xen_remap_buf.target_pfn; | |
499 | for (i = 0; i < xen_remap_buf.size; i++) { | |
500 | mfn = xen_remap_buf.mfns[i]; | |
501 | xen_update_mem_tables(pfn, mfn); | |
502 | remapped++; | |
503 | pfn++; | |
504 | } | |
505 | if (pfn_s == ~0UL || pfn == pfn_s) { | |
506 | pfn_s = xen_remap_buf.target_pfn; | |
507 | len += xen_remap_buf.size; | |
508 | } else if (pfn_s + len == xen_remap_buf.target_pfn) { | |
509 | len += xen_remap_buf.size; | |
510 | } else { | |
5b8e7d80 | 511 | xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len)); |
1f3ac86b JG |
512 | pfn_s = xen_remap_buf.target_pfn; |
513 | len = xen_remap_buf.size; | |
514 | } | |
515 | ||
516 | mfn = xen_remap_mfn; | |
517 | xen_remap_mfn = xen_remap_buf.next_area_mfn; | |
518 | } | |
519 | ||
520 | if (pfn_s != ~0UL && len) | |
5b8e7d80 | 521 | xen_del_extra_mem(PFN_PHYS(pfn_s), PFN_PHYS(len)); |
1f3ac86b JG |
522 | |
523 | set_pte_mfn(buf, mfn_save, PAGE_KERNEL); | |
524 | ||
525 | pr_info("Remapped %ld page(s)\n", remapped); | |
526 | } | |
527 | ||
d312ae87 DV |
528 | static unsigned long __init xen_get_max_pages(void) |
529 | { | |
530 | unsigned long max_pages = MAX_DOMAIN_PAGES; | |
531 | domid_t domid = DOMID_SELF; | |
532 | int ret; | |
533 | ||
d3db7281 IC |
534 | /* |
535 | * For the initial domain we use the maximum reservation as | |
536 | * the maximum page. | |
537 | * | |
538 | * For guest domains the current maximum reservation reflects | |
539 | * the current maximum rather than the static maximum. In this | |
540 | * case the e820 map provided to us will cover the static | |
541 | * maximum region. | |
542 | */ | |
543 | if (xen_initial_domain()) { | |
544 | ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); | |
545 | if (ret > 0) | |
546 | max_pages = ret; | |
547 | } | |
548 | ||
d312ae87 DV |
549 | return min(max_pages, MAX_DOMAIN_PAGES); |
550 | } | |
551 | ||
dc91c728 DV |
552 | static void xen_align_and_add_e820_region(u64 start, u64 size, int type) |
553 | { | |
554 | u64 end = start + size; | |
555 | ||
556 | /* Align RAM regions to page boundaries. */ | |
557 | if (type == E820_RAM) { | |
558 | start = PAGE_ALIGN(start); | |
559 | end &= ~((u64)PAGE_SIZE - 1); | |
560 | } | |
561 | ||
562 | e820_add_region(start, end - start, type); | |
563 | } | |
564 | ||
3bc38cbc DV |
565 | void xen_ignore_unusable(struct e820entry *list, size_t map_size) |
566 | { | |
567 | struct e820entry *entry; | |
568 | unsigned int i; | |
569 | ||
570 | for (i = 0, entry = list; i < map_size; i++, entry++) { | |
571 | if (entry->type == E820_UNUSABLE) | |
572 | entry->type = E820_RAM; | |
573 | } | |
574 | } | |
575 | ||
5ead97c8 JF |
576 | /** |
577 | * machine_specific_memory_setup - Hook for machine specific memory setup. | |
578 | **/ | |
5ead97c8 JF |
579 | char * __init xen_memory_setup(void) |
580 | { | |
35ae11fd IC |
581 | static struct e820entry map[E820MAX] __initdata; |
582 | ||
5ead97c8 | 583 | unsigned long max_pfn = xen_start_info->nr_pages; |
35ae11fd IC |
584 | unsigned long long mem_end; |
585 | int rc; | |
586 | struct xen_memory_map memmap; | |
dc91c728 | 587 | unsigned long max_pages; |
42ee1471 | 588 | unsigned long extra_pages = 0; |
35ae11fd | 589 | int i; |
9e9a5fcb | 590 | int op; |
5ead97c8 | 591 | |
8006ec3e | 592 | max_pfn = min(MAX_DOMAIN_PAGES, max_pfn); |
35ae11fd IC |
593 | mem_end = PFN_PHYS(max_pfn); |
594 | ||
595 | memmap.nr_entries = E820MAX; | |
596 | set_xen_guest_handle(memmap.buffer, map); | |
597 | ||
9e9a5fcb IC |
598 | op = xen_initial_domain() ? |
599 | XENMEM_machine_memory_map : | |
600 | XENMEM_memory_map; | |
601 | rc = HYPERVISOR_memory_op(op, &memmap); | |
35ae11fd | 602 | if (rc == -ENOSYS) { |
9ec23a7f | 603 | BUG_ON(xen_initial_domain()); |
35ae11fd IC |
604 | memmap.nr_entries = 1; |
605 | map[0].addr = 0ULL; | |
606 | map[0].size = mem_end; | |
607 | /* 8MB slack (to balance backend allocations). */ | |
608 | map[0].size += 8ULL << 20; | |
609 | map[0].type = E820_RAM; | |
610 | rc = 0; | |
611 | } | |
612 | BUG_ON(rc); | |
1ea644c8 | 613 | BUG_ON(memmap.nr_entries == 0); |
8006ec3e | 614 | |
3bc38cbc DV |
615 | /* |
616 | * Xen won't allow a 1:1 mapping to be created to UNUSABLE | |
617 | * regions, so if we're using the machine memory map leave the | |
618 | * region as RAM as it is in the pseudo-physical map. | |
619 | * | |
620 | * UNUSABLE regions in domUs are not handled and will need | |
621 | * a patch in the future. | |
622 | */ | |
623 | if (xen_initial_domain()) | |
624 | xen_ignore_unusable(map, memmap.nr_entries); | |
625 | ||
dc91c728 DV |
626 | /* Make sure the Xen-supplied memory map is well-ordered. */ |
627 | sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries); | |
628 | ||
629 | max_pages = xen_get_max_pages(); | |
630 | if (max_pages > max_pfn) | |
631 | extra_pages += max_pages - max_pfn; | |
632 | ||
f3f436e3 | 633 | /* |
1f3ac86b JG |
634 | * Set identity map on non-RAM pages and prepare remapping the |
635 | * underlying RAM. | |
f3f436e3 | 636 | */ |
5b8e7d80 JG |
637 | xen_set_identity_and_remap(map, memmap.nr_entries, max_pfn, |
638 | &xen_released_pages); | |
2e2fb754 | 639 | |
58b7b53a | 640 | extra_pages += xen_released_pages; |
2e2fb754 | 641 | |
dc91c728 DV |
642 | /* |
643 | * Clamp the amount of extra memory to a EXTRA_MEM_RATIO | |
644 | * factor the base size. On non-highmem systems, the base | |
645 | * size is the full initial memory allocation; on highmem it | |
646 | * is limited to the max size of lowmem, so that it doesn't | |
647 | * get completely filled. | |
648 | * | |
649 | * In principle there could be a problem in lowmem systems if | |
650 | * the initial memory is also very large with respect to | |
651 | * lowmem, but we won't try to deal with that here. | |
652 | */ | |
653 | extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)), | |
654 | extra_pages); | |
dc91c728 DV |
655 | i = 0; |
656 | while (i < memmap.nr_entries) { | |
657 | u64 addr = map[i].addr; | |
658 | u64 size = map[i].size; | |
659 | u32 type = map[i].type; | |
660 | ||
661 | if (type == E820_RAM) { | |
662 | if (addr < mem_end) { | |
663 | size = min(size, mem_end - addr); | |
664 | } else if (extra_pages) { | |
665 | size = min(size, (u64)extra_pages * PAGE_SIZE); | |
666 | extra_pages -= size / PAGE_SIZE; | |
667 | xen_add_extra_mem(addr, size); | |
5b8e7d80 | 668 | xen_max_p2m_pfn = PFN_DOWN(addr + size); |
dc91c728 DV |
669 | } else |
670 | type = E820_UNUSABLE; | |
3654581e JF |
671 | } |
672 | ||
dc91c728 | 673 | xen_align_and_add_e820_region(addr, size, type); |
b5b43ced | 674 | |
dc91c728 DV |
675 | map[i].addr += size; |
676 | map[i].size -= size; | |
677 | if (map[i].size == 0) | |
678 | i++; | |
35ae11fd | 679 | } |
b792c755 | 680 | |
25b884a8 DV |
681 | /* |
682 | * Set the rest as identity mapped, in case PCI BARs are | |
683 | * located here. | |
684 | * | |
685 | * PFNs above MAX_P2M_PFN are considered identity mapped as | |
686 | * well. | |
687 | */ | |
688 | set_phys_range_identity(map[i-1].addr / PAGE_SIZE, ~0ul); | |
689 | ||
b792c755 | 690 | /* |
9ec23a7f IC |
691 | * In domU, the ISA region is normal, usable memory, but we |
692 | * reserve ISA memory anyway because too many things poke | |
b792c755 JF |
693 | * about in there. |
694 | */ | |
695 | e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, | |
696 | E820_RESERVED); | |
5ead97c8 | 697 | |
be5bf9fa JF |
698 | /* |
699 | * Reserve Xen bits: | |
700 | * - mfn_list | |
701 | * - xen_start_info | |
702 | * See comment above "struct start_info" in <xen/interface/xen.h> | |
51faaf2b KRW |
703 | * We tried to make the the memblock_reserve more selective so |
704 | * that it would be clear what region is reserved. Sadly we ran | |
705 | * in the problem wherein on a 64-bit hypervisor with a 32-bit | |
706 | * initial domain, the pt_base has the cr3 value which is not | |
707 | * neccessarily where the pagetable starts! As Jan put it: " | |
708 | * Actually, the adjustment turns out to be correct: The page | |
709 | * tables for a 32-on-64 dom0 get allocated in the order "first L1", | |
710 | * "first L2", "first L3", so the offset to the page table base is | |
711 | * indeed 2. When reading xen/include/public/xen.h's comment | |
712 | * very strictly, this is not a violation (since there nothing is said | |
713 | * that the first thing in the page table space is pointed to by | |
714 | * pt_base; I admit that this seems to be implied though, namely | |
715 | * do I think that it is implied that the page table space is the | |
716 | * range [pt_base, pt_base + nt_pt_frames), whereas that | |
717 | * range here indeed is [pt_base - 2, pt_base - 2 + nt_pt_frames), | |
718 | * which - without a priori knowledge - the kernel would have | |
719 | * difficulty to figure out)." - so lets just fall back to the | |
720 | * easy way and reserve the whole region. | |
be5bf9fa | 721 | */ |
24aa0788 TH |
722 | memblock_reserve(__pa(xen_start_info->mfn_list), |
723 | xen_start_info->pt_base - xen_start_info->mfn_list); | |
be5bf9fa JF |
724 | |
725 | sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); | |
726 | ||
5ead97c8 JF |
727 | return "Xen"; |
728 | } | |
729 | ||
abacaadc DV |
730 | /* |
731 | * Machine specific memory setup for auto-translated guests. | |
732 | */ | |
733 | char * __init xen_auto_xlated_memory_setup(void) | |
734 | { | |
735 | static struct e820entry map[E820MAX] __initdata; | |
736 | ||
737 | struct xen_memory_map memmap; | |
738 | int i; | |
739 | int rc; | |
740 | ||
741 | memmap.nr_entries = E820MAX; | |
742 | set_xen_guest_handle(memmap.buffer, map); | |
743 | ||
744 | rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap); | |
745 | if (rc < 0) | |
746 | panic("No memory map (%d)\n", rc); | |
747 | ||
748 | sanitize_e820_map(map, ARRAY_SIZE(map), &memmap.nr_entries); | |
749 | ||
750 | for (i = 0; i < memmap.nr_entries; i++) | |
751 | e820_add_region(map[i].addr, map[i].size, map[i].type); | |
752 | ||
753 | memblock_reserve(__pa(xen_start_info->mfn_list), | |
754 | xen_start_info->pt_base - xen_start_info->mfn_list); | |
755 | ||
756 | return "Xen"; | |
757 | } | |
758 | ||
d2eea68e RM |
759 | /* |
760 | * Set the bit indicating "nosegneg" library variants should be used. | |
6a52e4b1 JF |
761 | * We only need to bother in pure 32-bit mode; compat 32-bit processes |
762 | * can have un-truncated segments, so wrapping around is allowed. | |
d2eea68e | 763 | */ |
08b6d290 | 764 | static void __init fiddle_vdso(void) |
d2eea68e | 765 | { |
6a52e4b1 | 766 | #ifdef CONFIG_X86_32 |
6f121e54 AL |
767 | /* |
768 | * This could be called before selected_vdso32 is initialized, so | |
769 | * just fiddle with both possible images. vdso_image_32_syscall | |
770 | * can't be selected, since it only exists on 64-bit systems. | |
771 | */ | |
6a52e4b1 | 772 | u32 *mask; |
6f121e54 AL |
773 | mask = vdso_image_32_int80.data + |
774 | vdso_image_32_int80.sym_VDSO32_NOTE_MASK; | |
6a52e4b1 | 775 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; |
6f121e54 AL |
776 | mask = vdso_image_32_sysenter.data + |
777 | vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK; | |
d2eea68e | 778 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; |
6fcac6d3 | 779 | #endif |
d2eea68e RM |
780 | } |
781 | ||
148f9bb8 | 782 | static int register_callback(unsigned type, const void *func) |
e2a81baf | 783 | { |
88459d4c JF |
784 | struct callback_register callback = { |
785 | .type = type, | |
786 | .address = XEN_CALLBACK(__KERNEL_CS, func), | |
e2a81baf JF |
787 | .flags = CALLBACKF_mask_events, |
788 | }; | |
789 | ||
88459d4c JF |
790 | return HYPERVISOR_callback_op(CALLBACKOP_register, &callback); |
791 | } | |
792 | ||
148f9bb8 | 793 | void xen_enable_sysenter(void) |
88459d4c | 794 | { |
6fcac6d3 | 795 | int ret; |
62541c37 | 796 | unsigned sysenter_feature; |
6fcac6d3 JF |
797 | |
798 | #ifdef CONFIG_X86_32 | |
62541c37 | 799 | sysenter_feature = X86_FEATURE_SEP; |
6fcac6d3 | 800 | #else |
62541c37 | 801 | sysenter_feature = X86_FEATURE_SYSENTER32; |
6fcac6d3 | 802 | #endif |
88459d4c | 803 | |
62541c37 JF |
804 | if (!boot_cpu_has(sysenter_feature)) |
805 | return; | |
806 | ||
6fcac6d3 | 807 | ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target); |
62541c37 JF |
808 | if(ret != 0) |
809 | setup_clear_cpu_cap(sysenter_feature); | |
e2a81baf JF |
810 | } |
811 | ||
148f9bb8 | 812 | void xen_enable_syscall(void) |
6fcac6d3 JF |
813 | { |
814 | #ifdef CONFIG_X86_64 | |
6fcac6d3 | 815 | int ret; |
6fcac6d3 JF |
816 | |
817 | ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target); | |
818 | if (ret != 0) { | |
d5303b81 | 819 | printk(KERN_ERR "Failed to set syscall callback: %d\n", ret); |
62541c37 JF |
820 | /* Pretty fatal; 64-bit userspace has no other |
821 | mechanism for syscalls. */ | |
822 | } | |
823 | ||
824 | if (boot_cpu_has(X86_FEATURE_SYSCALL32)) { | |
6fcac6d3 JF |
825 | ret = register_callback(CALLBACKTYPE_syscall32, |
826 | xen_syscall32_target); | |
d5303b81 | 827 | if (ret != 0) |
62541c37 | 828 | setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); |
6fcac6d3 JF |
829 | } |
830 | #endif /* CONFIG_X86_64 */ | |
831 | } | |
ea9f9274 | 832 | |
d285d683 | 833 | void __init xen_pvmmu_arch_setup(void) |
5ead97c8 | 834 | { |
5ead97c8 JF |
835 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments); |
836 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables); | |
837 | ||
d285d683 MR |
838 | HYPERVISOR_vm_assist(VMASST_CMD_enable, |
839 | VMASST_TYPE_pae_extended_cr3); | |
5ead97c8 | 840 | |
88459d4c JF |
841 | if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) || |
842 | register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback)) | |
843 | BUG(); | |
5ead97c8 | 844 | |
e2a81baf | 845 | xen_enable_sysenter(); |
6fcac6d3 | 846 | xen_enable_syscall(); |
d285d683 MR |
847 | } |
848 | ||
849 | /* This function is not called for HVM domains */ | |
850 | void __init xen_arch_setup(void) | |
851 | { | |
852 | xen_panic_handler_init(); | |
853 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
854 | xen_pvmmu_arch_setup(); | |
855 | ||
5ead97c8 JF |
856 | #ifdef CONFIG_ACPI |
857 | if (!(xen_start_info->flags & SIF_INITDOMAIN)) { | |
858 | printk(KERN_INFO "ACPI in unprivileged domain disabled\n"); | |
859 | disable_acpi(); | |
860 | } | |
861 | #endif | |
862 | ||
863 | memcpy(boot_command_line, xen_start_info->cmd_line, | |
864 | MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ? | |
865 | COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE); | |
866 | ||
bc15fde7 | 867 | /* Set up idle, making sure it calls safe_halt() pvop */ |
d91ee586 | 868 | disable_cpuidle(); |
48cdd828 | 869 | disable_cpufreq(); |
6a377ddc | 870 | WARN_ON(xen_set_default_idle()); |
d2eea68e | 871 | fiddle_vdso(); |
8d54db79 KRW |
872 | #ifdef CONFIG_NUMA |
873 | numa_off = 1; | |
874 | #endif | |
5ead97c8 | 875 | } |