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> | |
808fdb71 | 30 | #include <xen/hvc-console.h> |
5ead97c8 | 31 | #include "xen-ops.h" |
d2eea68e | 32 | #include "vdso.h" |
1f3ac86b | 33 | #include "mmu.h" |
5ead97c8 | 34 | |
c70727a5 JG |
35 | #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024) |
36 | ||
42ee1471 | 37 | /* Amount of extra memory space we add to the e820 ranges */ |
8b5d44a5 | 38 | struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; |
42ee1471 | 39 | |
aa24411b DV |
40 | /* Number of pages released from the initial allocation. */ |
41 | unsigned long xen_released_pages; | |
42 | ||
69632ecf JG |
43 | /* E820 map used during setting up memory. */ |
44 | static struct e820entry xen_e820_map[E820MAX] __initdata; | |
45 | static u32 xen_e820_map_entries __initdata; | |
46 | ||
1f3ac86b JG |
47 | /* |
48 | * Buffer used to remap identity mapped pages. We only need the virtual space. | |
49 | * The physical page behind this address is remapped as needed to different | |
50 | * buffer pages. | |
51 | */ | |
52 | #define REMAP_SIZE (P2M_PER_PAGE - 3) | |
53 | static struct { | |
54 | unsigned long next_area_mfn; | |
55 | unsigned long target_pfn; | |
56 | unsigned long size; | |
57 | unsigned long mfns[REMAP_SIZE]; | |
58 | } xen_remap_buf __initdata __aligned(PAGE_SIZE); | |
59 | static unsigned long xen_remap_mfn __initdata = INVALID_P2M_ENTRY; | |
4fbb67e3 | 60 | |
698bb8d1 JF |
61 | /* |
62 | * The maximum amount of extra memory compared to the base size. The | |
63 | * main scaling factor is the size of struct page. At extreme ratios | |
64 | * of base:extra, all the base memory can be filled with page | |
65 | * structures for the extra memory, leaving no space for anything | |
66 | * else. | |
67 | * | |
68 | * 10x seems like a reasonable balance between scaling flexibility and | |
69 | * leaving a practically usable system. | |
70 | */ | |
71 | #define EXTRA_MEM_RATIO (10) | |
72 | ||
c70727a5 JG |
73 | static bool xen_512gb_limit __initdata = IS_ENABLED(CONFIG_XEN_512GB); |
74 | ||
75 | static void __init xen_parse_512gb(void) | |
76 | { | |
77 | bool val = false; | |
78 | char *arg; | |
79 | ||
80 | arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit"); | |
81 | if (!arg) | |
82 | return; | |
83 | ||
84 | arg = strstr(xen_start_info->cmd_line, "xen_512gb_limit="); | |
85 | if (!arg) | |
86 | val = true; | |
87 | else if (strtobool(arg + strlen("xen_512gb_limit="), &val)) | |
88 | return; | |
89 | ||
90 | xen_512gb_limit = val; | |
91 | } | |
92 | ||
626d7508 JG |
93 | static void __init xen_add_extra_mem(unsigned long start_pfn, |
94 | unsigned long n_pfns) | |
42ee1471 | 95 | { |
dc91c728 | 96 | int i; |
6eaa412f | 97 | |
626d7508 JG |
98 | /* |
99 | * No need to check for zero size, should happen rarely and will only | |
100 | * write a new entry regarded to be unused due to zero size. | |
101 | */ | |
dc91c728 DV |
102 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
103 | /* Add new region. */ | |
626d7508 JG |
104 | if (xen_extra_mem[i].n_pfns == 0) { |
105 | xen_extra_mem[i].start_pfn = start_pfn; | |
106 | xen_extra_mem[i].n_pfns = n_pfns; | |
dc91c728 DV |
107 | break; |
108 | } | |
109 | /* Append to existing region. */ | |
626d7508 JG |
110 | if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns == |
111 | start_pfn) { | |
112 | xen_extra_mem[i].n_pfns += n_pfns; | |
dc91c728 DV |
113 | break; |
114 | } | |
115 | } | |
116 | if (i == XEN_EXTRA_MEM_MAX_REGIONS) | |
117 | printk(KERN_WARNING "Warning: not enough extra memory regions\n"); | |
42ee1471 | 118 | |
626d7508 | 119 | memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns)); |
5b8e7d80 | 120 | } |
2f7acb20 | 121 | |
626d7508 JG |
122 | static void __init xen_del_extra_mem(unsigned long start_pfn, |
123 | unsigned long n_pfns) | |
5b8e7d80 JG |
124 | { |
125 | int i; | |
626d7508 | 126 | unsigned long start_r, size_r; |
c96aae1f | 127 | |
5b8e7d80 | 128 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
626d7508 JG |
129 | start_r = xen_extra_mem[i].start_pfn; |
130 | size_r = xen_extra_mem[i].n_pfns; | |
5b8e7d80 JG |
131 | |
132 | /* Start of region. */ | |
626d7508 JG |
133 | if (start_r == start_pfn) { |
134 | BUG_ON(n_pfns > size_r); | |
135 | xen_extra_mem[i].start_pfn += n_pfns; | |
136 | xen_extra_mem[i].n_pfns -= n_pfns; | |
5b8e7d80 JG |
137 | break; |
138 | } | |
139 | /* End of region. */ | |
626d7508 JG |
140 | if (start_r + size_r == start_pfn + n_pfns) { |
141 | BUG_ON(n_pfns > size_r); | |
142 | xen_extra_mem[i].n_pfns -= n_pfns; | |
5b8e7d80 JG |
143 | break; |
144 | } | |
145 | /* Mid of region. */ | |
626d7508 JG |
146 | if (start_pfn > start_r && start_pfn < start_r + size_r) { |
147 | BUG_ON(start_pfn + n_pfns > start_r + size_r); | |
148 | xen_extra_mem[i].n_pfns = start_pfn - start_r; | |
5b8e7d80 | 149 | /* Calling memblock_reserve() again is okay. */ |
626d7508 JG |
150 | xen_add_extra_mem(start_pfn + n_pfns, start_r + size_r - |
151 | (start_pfn + n_pfns)); | |
5b8e7d80 JG |
152 | break; |
153 | } | |
154 | } | |
626d7508 | 155 | memblock_free(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns)); |
5b8e7d80 JG |
156 | } |
157 | ||
158 | /* | |
159 | * Called during boot before the p2m list can take entries beyond the | |
160 | * hypervisor supplied p2m list. Entries in extra mem are to be regarded as | |
161 | * invalid. | |
162 | */ | |
163 | unsigned long __ref xen_chk_extra_mem(unsigned long pfn) | |
164 | { | |
165 | int i; | |
6eaa412f | 166 | |
5b8e7d80 | 167 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
626d7508 JG |
168 | if (pfn >= xen_extra_mem[i].start_pfn && |
169 | pfn < xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns) | |
5b8e7d80 JG |
170 | return INVALID_P2M_ENTRY; |
171 | } | |
172 | ||
173 | return IDENTITY_FRAME(pfn); | |
174 | } | |
175 | ||
176 | /* | |
177 | * Mark all pfns of extra mem as invalid in p2m list. | |
178 | */ | |
179 | void __init xen_inv_extra_mem(void) | |
180 | { | |
181 | unsigned long pfn, pfn_s, pfn_e; | |
182 | int i; | |
183 | ||
184 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { | |
626d7508 | 185 | if (!xen_extra_mem[i].n_pfns) |
9a17ad7f | 186 | continue; |
626d7508 JG |
187 | pfn_s = xen_extra_mem[i].start_pfn; |
188 | pfn_e = pfn_s + xen_extra_mem[i].n_pfns; | |
5b8e7d80 JG |
189 | for (pfn = pfn_s; pfn < pfn_e; pfn++) |
190 | set_phys_to_machine(pfn, INVALID_P2M_ENTRY); | |
c96aae1f | 191 | } |
42ee1471 JF |
192 | } |
193 | ||
4fbb67e3 MR |
194 | /* |
195 | * Finds the next RAM pfn available in the E820 map after min_pfn. | |
196 | * This function updates min_pfn with the pfn found and returns | |
197 | * the size of that range or zero if not found. | |
198 | */ | |
69632ecf | 199 | static unsigned long __init xen_find_pfn_range(unsigned long *min_pfn) |
2e2fb754 | 200 | { |
69632ecf | 201 | const struct e820entry *entry = xen_e820_map; |
2e2fb754 KRW |
202 | unsigned int i; |
203 | unsigned long done = 0; | |
2e2fb754 | 204 | |
69632ecf | 205 | for (i = 0; i < xen_e820_map_entries; i++, entry++) { |
2e2fb754 KRW |
206 | unsigned long s_pfn; |
207 | unsigned long e_pfn; | |
2e2fb754 KRW |
208 | |
209 | if (entry->type != E820_RAM) | |
210 | continue; | |
211 | ||
c3d93f88 | 212 | e_pfn = PFN_DOWN(entry->addr + entry->size); |
2e2fb754 | 213 | |
4fbb67e3 MR |
214 | /* We only care about E820 after this */ |
215 | if (e_pfn < *min_pfn) | |
2e2fb754 KRW |
216 | continue; |
217 | ||
c3d93f88 | 218 | s_pfn = PFN_UP(entry->addr); |
4fbb67e3 MR |
219 | |
220 | /* If min_pfn falls within the E820 entry, we want to start | |
221 | * at the min_pfn PFN. | |
2e2fb754 | 222 | */ |
4fbb67e3 MR |
223 | if (s_pfn <= *min_pfn) { |
224 | done = e_pfn - *min_pfn; | |
2e2fb754 | 225 | } else { |
4fbb67e3 MR |
226 | done = e_pfn - s_pfn; |
227 | *min_pfn = s_pfn; | |
2e2fb754 | 228 | } |
4fbb67e3 MR |
229 | break; |
230 | } | |
2e2fb754 | 231 | |
4fbb67e3 MR |
232 | return done; |
233 | } | |
2e2fb754 | 234 | |
1f3ac86b JG |
235 | static int __init xen_free_mfn(unsigned long mfn) |
236 | { | |
237 | struct xen_memory_reservation reservation = { | |
238 | .address_bits = 0, | |
239 | .extent_order = 0, | |
240 | .domid = DOMID_SELF | |
241 | }; | |
242 | ||
243 | set_xen_guest_handle(reservation.extent_start, &mfn); | |
244 | reservation.nr_extents = 1; | |
245 | ||
246 | return HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); | |
247 | } | |
248 | ||
4fbb67e3 | 249 | /* |
1f3ac86b | 250 | * This releases a chunk of memory and then does the identity map. It's used |
4fbb67e3 MR |
251 | * as a fallback if the remapping fails. |
252 | */ | |
253 | static void __init xen_set_identity_and_release_chunk(unsigned long start_pfn, | |
5097cdf6 | 254 | unsigned long end_pfn, unsigned long nr_pages) |
4fbb67e3 | 255 | { |
1f3ac86b JG |
256 | unsigned long pfn, end; |
257 | int ret; | |
258 | ||
4fbb67e3 MR |
259 | WARN_ON(start_pfn > end_pfn); |
260 | ||
bc7142cf | 261 | /* Release pages first. */ |
1f3ac86b JG |
262 | end = min(end_pfn, nr_pages); |
263 | for (pfn = start_pfn; pfn < end; pfn++) { | |
264 | unsigned long mfn = pfn_to_mfn(pfn); | |
265 | ||
266 | /* Make sure pfn exists to start with */ | |
267 | if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn) | |
268 | continue; | |
269 | ||
270 | ret = xen_free_mfn(mfn); | |
271 | WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret); | |
272 | ||
273 | if (ret == 1) { | |
5097cdf6 | 274 | xen_released_pages++; |
1f3ac86b JG |
275 | if (!__set_phys_to_machine(pfn, INVALID_P2M_ENTRY)) |
276 | break; | |
1f3ac86b JG |
277 | } else |
278 | break; | |
279 | } | |
280 | ||
bc7142cf | 281 | set_phys_range_identity(start_pfn, end_pfn); |
4fbb67e3 MR |
282 | } |
283 | ||
284 | /* | |
1f3ac86b | 285 | * Helper function to update the p2m and m2p tables and kernel mapping. |
4fbb67e3 | 286 | */ |
1f3ac86b | 287 | static void __init xen_update_mem_tables(unsigned long pfn, unsigned long mfn) |
4fbb67e3 MR |
288 | { |
289 | struct mmu_update update = { | |
3ba5c867 | 290 | .ptr = ((uint64_t)mfn << PAGE_SHIFT) | MMU_MACHPHYS_UPDATE, |
4fbb67e3 MR |
291 | .val = pfn |
292 | }; | |
293 | ||
294 | /* Update p2m */ | |
1f3ac86b | 295 | if (!set_phys_to_machine(pfn, mfn)) { |
4fbb67e3 MR |
296 | WARN(1, "Failed to set p2m mapping for pfn=%ld mfn=%ld\n", |
297 | pfn, mfn); | |
1f3ac86b | 298 | BUG(); |
2e2fb754 | 299 | } |
4fbb67e3 MR |
300 | |
301 | /* Update m2p */ | |
302 | if (HYPERVISOR_mmu_update(&update, 1, NULL, DOMID_SELF) < 0) { | |
303 | WARN(1, "Failed to set m2p mapping for mfn=%ld pfn=%ld\n", | |
304 | mfn, pfn); | |
1f3ac86b | 305 | BUG(); |
4fbb67e3 MR |
306 | } |
307 | ||
1f3ac86b | 308 | /* Update kernel mapping, but not for highmem. */ |
e86f9496 | 309 | if (pfn >= PFN_UP(__pa(high_memory - 1))) |
1f3ac86b JG |
310 | return; |
311 | ||
312 | if (HYPERVISOR_update_va_mapping((unsigned long)__va(pfn << PAGE_SHIFT), | |
313 | mfn_pte(mfn, PAGE_KERNEL), 0)) { | |
314 | WARN(1, "Failed to update kernel mapping for mfn=%ld pfn=%ld\n", | |
315 | mfn, pfn); | |
316 | BUG(); | |
317 | } | |
2e2fb754 | 318 | } |
83d51ab4 | 319 | |
4fbb67e3 MR |
320 | /* |
321 | * This function updates the p2m and m2p tables with an identity map from | |
1f3ac86b JG |
322 | * start_pfn to start_pfn+size and prepares remapping the underlying RAM of the |
323 | * original allocation at remap_pfn. The information needed for remapping is | |
324 | * saved in the memory itself to avoid the need for allocating buffers. The | |
325 | * complete remap information is contained in a list of MFNs each containing | |
326 | * up to REMAP_SIZE MFNs and the start target PFN for doing the remap. | |
327 | * This enables us to preserve the original mfn sequence while doing the | |
328 | * remapping at a time when the memory management is capable of allocating | |
329 | * virtual and physical memory in arbitrary amounts, see 'xen_remap_memory' and | |
330 | * its callers. | |
4fbb67e3 | 331 | */ |
1f3ac86b | 332 | static void __init xen_do_set_identity_and_remap_chunk( |
4fbb67e3 | 333 | unsigned long start_pfn, unsigned long size, unsigned long remap_pfn) |
83d51ab4 | 334 | { |
1f3ac86b JG |
335 | unsigned long buf = (unsigned long)&xen_remap_buf; |
336 | unsigned long mfn_save, mfn; | |
4fbb67e3 | 337 | unsigned long ident_pfn_iter, remap_pfn_iter; |
1f3ac86b | 338 | unsigned long ident_end_pfn = start_pfn + size; |
4fbb67e3 | 339 | unsigned long left = size; |
1f3ac86b | 340 | unsigned int i, chunk; |
4fbb67e3 MR |
341 | |
342 | WARN_ON(size == 0); | |
343 | ||
344 | BUG_ON(xen_feature(XENFEAT_auto_translated_physmap)); | |
83d51ab4 | 345 | |
1f3ac86b | 346 | mfn_save = virt_to_mfn(buf); |
e201bfcc | 347 | |
1f3ac86b JG |
348 | for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn; |
349 | ident_pfn_iter < ident_end_pfn; | |
350 | ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) { | |
351 | chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE; | |
4fbb67e3 | 352 | |
1f3ac86b JG |
353 | /* Map first pfn to xen_remap_buf */ |
354 | mfn = pfn_to_mfn(ident_pfn_iter); | |
355 | set_pte_mfn(buf, mfn, PAGE_KERNEL); | |
4fbb67e3 | 356 | |
1f3ac86b JG |
357 | /* Save mapping information in page */ |
358 | xen_remap_buf.next_area_mfn = xen_remap_mfn; | |
359 | xen_remap_buf.target_pfn = remap_pfn_iter; | |
360 | xen_remap_buf.size = chunk; | |
361 | for (i = 0; i < chunk; i++) | |
362 | xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i); | |
4fbb67e3 | 363 | |
1f3ac86b JG |
364 | /* Put remap buf into list. */ |
365 | xen_remap_mfn = mfn; | |
4fbb67e3 | 366 | |
1f3ac86b | 367 | /* Set identity map */ |
bc7142cf | 368 | set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk); |
83d51ab4 | 369 | |
1f3ac86b | 370 | left -= chunk; |
4fbb67e3 | 371 | } |
83d51ab4 | 372 | |
1f3ac86b JG |
373 | /* Restore old xen_remap_buf mapping */ |
374 | set_pte_mfn(buf, mfn_save, PAGE_KERNEL); | |
83d51ab4 DV |
375 | } |
376 | ||
4fbb67e3 MR |
377 | /* |
378 | * This function takes a contiguous pfn range that needs to be identity mapped | |
379 | * and: | |
380 | * | |
381 | * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn. | |
382 | * 2) Calls the do_ function to actually do the mapping/remapping work. | |
383 | * | |
384 | * The goal is to not allocate additional memory but to remap the existing | |
385 | * pages. In the case of an error the underlying memory is simply released back | |
386 | * to Xen and not remapped. | |
387 | */ | |
76f0a486 | 388 | static unsigned long __init xen_set_identity_and_remap_chunk( |
69632ecf | 389 | unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages, |
5097cdf6 | 390 | unsigned long remap_pfn) |
4fbb67e3 MR |
391 | { |
392 | unsigned long pfn; | |
393 | unsigned long i = 0; | |
394 | unsigned long n = end_pfn - start_pfn; | |
395 | ||
396 | while (i < n) { | |
397 | unsigned long cur_pfn = start_pfn + i; | |
398 | unsigned long left = n - i; | |
399 | unsigned long size = left; | |
400 | unsigned long remap_range_size; | |
401 | ||
402 | /* Do not remap pages beyond the current allocation */ | |
403 | if (cur_pfn >= nr_pages) { | |
404 | /* Identity map remaining pages */ | |
bc7142cf | 405 | set_phys_range_identity(cur_pfn, cur_pfn + size); |
4fbb67e3 MR |
406 | break; |
407 | } | |
408 | if (cur_pfn + size > nr_pages) | |
409 | size = nr_pages - cur_pfn; | |
410 | ||
69632ecf | 411 | remap_range_size = xen_find_pfn_range(&remap_pfn); |
4fbb67e3 MR |
412 | if (!remap_range_size) { |
413 | pr_warning("Unable to find available pfn range, not remapping identity pages\n"); | |
414 | xen_set_identity_and_release_chunk(cur_pfn, | |
5097cdf6 | 415 | cur_pfn + left, nr_pages); |
4fbb67e3 MR |
416 | break; |
417 | } | |
418 | /* Adjust size to fit in current e820 RAM region */ | |
419 | if (size > remap_range_size) | |
420 | size = remap_range_size; | |
421 | ||
1f3ac86b | 422 | xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn); |
4fbb67e3 MR |
423 | |
424 | /* Update variables to reflect new mappings. */ | |
425 | i += size; | |
426 | remap_pfn += size; | |
4fbb67e3 MR |
427 | } |
428 | ||
429 | /* | |
430 | * If the PFNs are currently mapped, the VA mapping also needs | |
431 | * to be updated to be 1:1. | |
432 | */ | |
433 | for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++) | |
434 | (void)HYPERVISOR_update_va_mapping( | |
435 | (unsigned long)__va(pfn << PAGE_SHIFT), | |
436 | mfn_pte(pfn, PAGE_KERNEL_IO), 0); | |
437 | ||
438 | return remap_pfn; | |
439 | } | |
440 | ||
5097cdf6 | 441 | static void __init xen_set_identity_and_remap(unsigned long nr_pages) |
093d7b46 | 442 | { |
f3f436e3 | 443 | phys_addr_t start = 0; |
4fbb67e3 | 444 | unsigned long last_pfn = nr_pages; |
69632ecf | 445 | const struct e820entry *entry = xen_e820_map; |
68df0da7 KRW |
446 | int i; |
447 | ||
f3f436e3 DV |
448 | /* |
449 | * Combine non-RAM regions and gaps until a RAM region (or the | |
450 | * end of the map) is reached, then set the 1:1 map and | |
4fbb67e3 | 451 | * remap the memory in those non-RAM regions. |
f3f436e3 DV |
452 | * |
453 | * The combined non-RAM regions are rounded to a whole number | |
454 | * of pages so any partial pages are accessible via the 1:1 | |
455 | * mapping. This is needed for some BIOSes that put (for | |
456 | * example) the DMI tables in a reserved region that begins on | |
457 | * a non-page boundary. | |
458 | */ | |
69632ecf | 459 | for (i = 0; i < xen_e820_map_entries; i++, entry++) { |
f3f436e3 | 460 | phys_addr_t end = entry->addr + entry->size; |
69632ecf | 461 | if (entry->type == E820_RAM || i == xen_e820_map_entries - 1) { |
f3f436e3 DV |
462 | unsigned long start_pfn = PFN_DOWN(start); |
463 | unsigned long end_pfn = PFN_UP(end); | |
68df0da7 | 464 | |
f3f436e3 DV |
465 | if (entry->type == E820_RAM) |
466 | end_pfn = PFN_UP(entry->addr); | |
68df0da7 | 467 | |
83d51ab4 | 468 | if (start_pfn < end_pfn) |
4fbb67e3 | 469 | last_pfn = xen_set_identity_and_remap_chunk( |
69632ecf | 470 | start_pfn, end_pfn, nr_pages, |
5097cdf6 | 471 | last_pfn); |
f3f436e3 | 472 | start = end; |
68df0da7 | 473 | } |
68df0da7 | 474 | } |
f3f436e3 | 475 | |
5097cdf6 | 476 | pr_info("Released %ld page(s)\n", xen_released_pages); |
4fbb67e3 | 477 | } |
1f3ac86b JG |
478 | |
479 | /* | |
480 | * Remap the memory prepared in xen_do_set_identity_and_remap_chunk(). | |
481 | * The remap information (which mfn remap to which pfn) is contained in the | |
482 | * to be remapped memory itself in a linked list anchored at xen_remap_mfn. | |
483 | * This scheme allows to remap the different chunks in arbitrary order while | |
484 | * the resulting mapping will be independant from the order. | |
485 | */ | |
486 | void __init xen_remap_memory(void) | |
487 | { | |
488 | unsigned long buf = (unsigned long)&xen_remap_buf; | |
489 | unsigned long mfn_save, mfn, pfn; | |
490 | unsigned long remapped = 0; | |
491 | unsigned int i; | |
492 | unsigned long pfn_s = ~0UL; | |
493 | unsigned long len = 0; | |
494 | ||
495 | mfn_save = virt_to_mfn(buf); | |
496 | ||
497 | while (xen_remap_mfn != INVALID_P2M_ENTRY) { | |
498 | /* Map the remap information */ | |
499 | set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL); | |
500 | ||
501 | BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]); | |
502 | ||
503 | pfn = xen_remap_buf.target_pfn; | |
504 | for (i = 0; i < xen_remap_buf.size; i++) { | |
505 | mfn = xen_remap_buf.mfns[i]; | |
506 | xen_update_mem_tables(pfn, mfn); | |
507 | remapped++; | |
508 | pfn++; | |
509 | } | |
510 | if (pfn_s == ~0UL || pfn == pfn_s) { | |
511 | pfn_s = xen_remap_buf.target_pfn; | |
512 | len += xen_remap_buf.size; | |
513 | } else if (pfn_s + len == xen_remap_buf.target_pfn) { | |
514 | len += xen_remap_buf.size; | |
515 | } else { | |
626d7508 | 516 | xen_del_extra_mem(pfn_s, len); |
1f3ac86b JG |
517 | pfn_s = xen_remap_buf.target_pfn; |
518 | len = xen_remap_buf.size; | |
519 | } | |
520 | ||
521 | mfn = xen_remap_mfn; | |
522 | xen_remap_mfn = xen_remap_buf.next_area_mfn; | |
523 | } | |
524 | ||
525 | if (pfn_s != ~0UL && len) | |
626d7508 | 526 | xen_del_extra_mem(pfn_s, len); |
1f3ac86b JG |
527 | |
528 | set_pte_mfn(buf, mfn_save, PAGE_KERNEL); | |
529 | ||
530 | pr_info("Remapped %ld page(s)\n", remapped); | |
531 | } | |
532 | ||
c70727a5 JG |
533 | static unsigned long __init xen_get_pages_limit(void) |
534 | { | |
535 | unsigned long limit; | |
536 | ||
537 | #ifdef CONFIG_X86_32 | |
538 | limit = GB(64) / PAGE_SIZE; | |
539 | #else | |
cb9e444b | 540 | limit = MAXMEM / PAGE_SIZE; |
c70727a5 JG |
541 | if (!xen_initial_domain() && xen_512gb_limit) |
542 | limit = GB(512) / PAGE_SIZE; | |
543 | #endif | |
544 | return limit; | |
545 | } | |
546 | ||
d312ae87 DV |
547 | static unsigned long __init xen_get_max_pages(void) |
548 | { | |
c70727a5 | 549 | unsigned long max_pages, limit; |
d312ae87 | 550 | domid_t domid = DOMID_SELF; |
24f775a6 | 551 | long ret; |
d312ae87 | 552 | |
c70727a5 JG |
553 | limit = xen_get_pages_limit(); |
554 | max_pages = limit; | |
555 | ||
d3db7281 IC |
556 | /* |
557 | * For the initial domain we use the maximum reservation as | |
558 | * the maximum page. | |
559 | * | |
560 | * For guest domains the current maximum reservation reflects | |
561 | * the current maximum rather than the static maximum. In this | |
562 | * case the e820 map provided to us will cover the static | |
563 | * maximum region. | |
564 | */ | |
565 | if (xen_initial_domain()) { | |
566 | ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); | |
567 | if (ret > 0) | |
568 | max_pages = ret; | |
569 | } | |
570 | ||
c70727a5 | 571 | return min(max_pages, limit); |
d312ae87 DV |
572 | } |
573 | ||
a3f52396 JG |
574 | static void __init xen_align_and_add_e820_region(phys_addr_t start, |
575 | phys_addr_t size, int type) | |
dc91c728 | 576 | { |
3ba5c867 | 577 | phys_addr_t end = start + size; |
dc91c728 DV |
578 | |
579 | /* Align RAM regions to page boundaries. */ | |
580 | if (type == E820_RAM) { | |
581 | start = PAGE_ALIGN(start); | |
3ba5c867 | 582 | end &= ~((phys_addr_t)PAGE_SIZE - 1); |
dc91c728 DV |
583 | } |
584 | ||
585 | e820_add_region(start, end - start, type); | |
586 | } | |
587 | ||
69632ecf | 588 | static void __init xen_ignore_unusable(void) |
3bc38cbc | 589 | { |
69632ecf | 590 | struct e820entry *entry = xen_e820_map; |
3bc38cbc DV |
591 | unsigned int i; |
592 | ||
69632ecf | 593 | for (i = 0; i < xen_e820_map_entries; i++, entry++) { |
3bc38cbc DV |
594 | if (entry->type == E820_UNUSABLE) |
595 | entry->type = E820_RAM; | |
596 | } | |
597 | } | |
598 | ||
5097cdf6 JG |
599 | static unsigned long __init xen_count_remap_pages(unsigned long max_pfn) |
600 | { | |
601 | unsigned long extra = 0; | |
ab24507c | 602 | unsigned long start_pfn, end_pfn; |
5097cdf6 JG |
603 | const struct e820entry *entry = xen_e820_map; |
604 | int i; | |
605 | ||
ab24507c | 606 | end_pfn = 0; |
5097cdf6 | 607 | for (i = 0; i < xen_e820_map_entries; i++, entry++) { |
ab24507c JG |
608 | start_pfn = PFN_DOWN(entry->addr); |
609 | /* Adjacent regions on non-page boundaries handling! */ | |
610 | end_pfn = min(end_pfn, start_pfn); | |
5097cdf6 JG |
611 | |
612 | if (start_pfn >= max_pfn) | |
ab24507c JG |
613 | return extra + max_pfn - end_pfn; |
614 | ||
615 | /* Add any holes in map to result. */ | |
616 | extra += start_pfn - end_pfn; | |
617 | ||
618 | end_pfn = PFN_UP(entry->addr + entry->size); | |
619 | end_pfn = min(end_pfn, max_pfn); | |
620 | ||
621 | if (entry->type != E820_RAM) | |
622 | extra += end_pfn - start_pfn; | |
5097cdf6 JG |
623 | } |
624 | ||
625 | return extra; | |
626 | } | |
627 | ||
e612b4a7 JG |
628 | bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size) |
629 | { | |
630 | struct e820entry *entry; | |
631 | unsigned mapcnt; | |
632 | phys_addr_t end; | |
633 | ||
634 | if (!size) | |
635 | return false; | |
636 | ||
637 | end = start + size; | |
638 | entry = xen_e820_map; | |
639 | ||
640 | for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++) { | |
641 | if (entry->type == E820_RAM && entry->addr <= start && | |
642 | (entry->addr + entry->size) >= end) | |
643 | return false; | |
644 | ||
645 | entry++; | |
646 | } | |
647 | ||
648 | return true; | |
649 | } | |
650 | ||
9ddac5b7 JG |
651 | /* |
652 | * Find a free area in physical memory not yet reserved and compliant with | |
653 | * E820 map. | |
654 | * Used to relocate pre-allocated areas like initrd or p2m list which are in | |
655 | * conflict with the to be used E820 map. | |
656 | * In case no area is found, return 0. Otherwise return the physical address | |
657 | * of the area which is already reserved for convenience. | |
658 | */ | |
659 | phys_addr_t __init xen_find_free_area(phys_addr_t size) | |
660 | { | |
661 | unsigned mapcnt; | |
662 | phys_addr_t addr, start; | |
663 | struct e820entry *entry = xen_e820_map; | |
664 | ||
665 | for (mapcnt = 0; mapcnt < xen_e820_map_entries; mapcnt++, entry++) { | |
666 | if (entry->type != E820_RAM || entry->size < size) | |
667 | continue; | |
668 | start = entry->addr; | |
669 | for (addr = start; addr < start + size; addr += PAGE_SIZE) { | |
670 | if (!memblock_is_reserved(addr)) | |
671 | continue; | |
672 | start = addr + PAGE_SIZE; | |
673 | if (start + size > entry->addr + entry->size) | |
674 | break; | |
675 | } | |
676 | if (addr >= start + size) { | |
677 | memblock_reserve(start, size); | |
678 | return start; | |
679 | } | |
680 | } | |
681 | ||
682 | return 0; | |
683 | } | |
684 | ||
4b9c1537 JG |
685 | /* |
686 | * Like memcpy, but with physical addresses for dest and src. | |
687 | */ | |
688 | static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src, | |
689 | phys_addr_t n) | |
690 | { | |
691 | phys_addr_t dest_off, src_off, dest_len, src_len, len; | |
692 | void *from, *to; | |
693 | ||
694 | while (n) { | |
695 | dest_off = dest & ~PAGE_MASK; | |
696 | src_off = src & ~PAGE_MASK; | |
697 | dest_len = n; | |
698 | if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off) | |
699 | dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off; | |
700 | src_len = n; | |
701 | if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off) | |
702 | src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off; | |
703 | len = min(dest_len, src_len); | |
704 | to = early_memremap(dest - dest_off, dest_len + dest_off); | |
705 | from = early_memremap(src - src_off, src_len + src_off); | |
706 | memcpy(to, from, len); | |
707 | early_memunmap(to, dest_len + dest_off); | |
708 | early_memunmap(from, src_len + src_off); | |
709 | n -= len; | |
710 | dest += len; | |
711 | src += len; | |
712 | } | |
713 | } | |
714 | ||
8f5b0c63 JG |
715 | /* |
716 | * Reserve Xen mfn_list. | |
8f5b0c63 JG |
717 | */ |
718 | static void __init xen_reserve_xen_mfnlist(void) | |
719 | { | |
70e61199 JG |
720 | phys_addr_t start, size; |
721 | ||
8f5b0c63 | 722 | if (xen_start_info->mfn_list >= __START_KERNEL_map) { |
70e61199 JG |
723 | start = __pa(xen_start_info->mfn_list); |
724 | size = PFN_ALIGN(xen_start_info->nr_pages * | |
725 | sizeof(unsigned long)); | |
726 | } else { | |
727 | start = PFN_PHYS(xen_start_info->first_p2m_pfn); | |
728 | size = PFN_PHYS(xen_start_info->nr_p2m_frames); | |
729 | } | |
730 | ||
731 | if (!xen_is_e820_reserved(start, size)) { | |
732 | memblock_reserve(start, size); | |
8f5b0c63 JG |
733 | return; |
734 | } | |
735 | ||
70e61199 JG |
736 | #ifdef CONFIG_X86_32 |
737 | /* | |
738 | * Relocating the p2m on 32 bit system to an arbitrary virtual address | |
739 | * is not supported, so just give up. | |
740 | */ | |
741 | xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n"); | |
742 | BUG(); | |
743 | #else | |
744 | xen_relocate_p2m(); | |
745 | #endif | |
8f5b0c63 JG |
746 | } |
747 | ||
5ead97c8 JF |
748 | /** |
749 | * machine_specific_memory_setup - Hook for machine specific memory setup. | |
750 | **/ | |
5ead97c8 JF |
751 | char * __init xen_memory_setup(void) |
752 | { | |
626d7508 | 753 | unsigned long max_pfn, pfn_s, n_pfns; |
5097cdf6 JG |
754 | phys_addr_t mem_end, addr, size, chunk_size; |
755 | u32 type; | |
35ae11fd IC |
756 | int rc; |
757 | struct xen_memory_map memmap; | |
dc91c728 | 758 | unsigned long max_pages; |
42ee1471 | 759 | unsigned long extra_pages = 0; |
35ae11fd | 760 | int i; |
9e9a5fcb | 761 | int op; |
5ead97c8 | 762 | |
c70727a5 JG |
763 | xen_parse_512gb(); |
764 | max_pfn = xen_get_pages_limit(); | |
765 | max_pfn = min(max_pfn, xen_start_info->nr_pages); | |
35ae11fd IC |
766 | mem_end = PFN_PHYS(max_pfn); |
767 | ||
768 | memmap.nr_entries = E820MAX; | |
69632ecf | 769 | set_xen_guest_handle(memmap.buffer, xen_e820_map); |
35ae11fd | 770 | |
9e9a5fcb IC |
771 | op = xen_initial_domain() ? |
772 | XENMEM_machine_memory_map : | |
773 | XENMEM_memory_map; | |
774 | rc = HYPERVISOR_memory_op(op, &memmap); | |
35ae11fd | 775 | if (rc == -ENOSYS) { |
9ec23a7f | 776 | BUG_ON(xen_initial_domain()); |
35ae11fd | 777 | memmap.nr_entries = 1; |
69632ecf JG |
778 | xen_e820_map[0].addr = 0ULL; |
779 | xen_e820_map[0].size = mem_end; | |
35ae11fd | 780 | /* 8MB slack (to balance backend allocations). */ |
69632ecf JG |
781 | xen_e820_map[0].size += 8ULL << 20; |
782 | xen_e820_map[0].type = E820_RAM; | |
35ae11fd IC |
783 | rc = 0; |
784 | } | |
785 | BUG_ON(rc); | |
1ea644c8 | 786 | BUG_ON(memmap.nr_entries == 0); |
69632ecf | 787 | xen_e820_map_entries = memmap.nr_entries; |
8006ec3e | 788 | |
3bc38cbc DV |
789 | /* |
790 | * Xen won't allow a 1:1 mapping to be created to UNUSABLE | |
791 | * regions, so if we're using the machine memory map leave the | |
792 | * region as RAM as it is in the pseudo-physical map. | |
793 | * | |
794 | * UNUSABLE regions in domUs are not handled and will need | |
795 | * a patch in the future. | |
796 | */ | |
797 | if (xen_initial_domain()) | |
69632ecf | 798 | xen_ignore_unusable(); |
3bc38cbc | 799 | |
dc91c728 | 800 | /* Make sure the Xen-supplied memory map is well-ordered. */ |
64c98e7f | 801 | sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map), |
69632ecf | 802 | &xen_e820_map_entries); |
dc91c728 DV |
803 | |
804 | max_pages = xen_get_max_pages(); | |
dc91c728 | 805 | |
5097cdf6 | 806 | /* How many extra pages do we need due to remapping? */ |
eafd72e0 JG |
807 | max_pages += xen_count_remap_pages(max_pfn); |
808 | ||
809 | if (max_pages > max_pfn) | |
810 | extra_pages += max_pages - max_pfn; | |
2e2fb754 | 811 | |
dc91c728 DV |
812 | /* |
813 | * Clamp the amount of extra memory to a EXTRA_MEM_RATIO | |
814 | * factor the base size. On non-highmem systems, the base | |
815 | * size is the full initial memory allocation; on highmem it | |
816 | * is limited to the max size of lowmem, so that it doesn't | |
817 | * get completely filled. | |
818 | * | |
c70727a5 JG |
819 | * Make sure we have no memory above max_pages, as this area |
820 | * isn't handled by the p2m management. | |
821 | * | |
dc91c728 DV |
822 | * In principle there could be a problem in lowmem systems if |
823 | * the initial memory is also very large with respect to | |
824 | * lowmem, but we won't try to deal with that here. | |
825 | */ | |
c70727a5 JG |
826 | extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)), |
827 | extra_pages, max_pages - max_pfn); | |
dc91c728 | 828 | i = 0; |
5097cdf6 JG |
829 | addr = xen_e820_map[0].addr; |
830 | size = xen_e820_map[0].size; | |
69632ecf | 831 | while (i < xen_e820_map_entries) { |
f5775e0b DV |
832 | bool discard = false; |
833 | ||
5097cdf6 JG |
834 | chunk_size = size; |
835 | type = xen_e820_map[i].type; | |
dc91c728 DV |
836 | |
837 | if (type == E820_RAM) { | |
838 | if (addr < mem_end) { | |
5097cdf6 | 839 | chunk_size = min(size, mem_end - addr); |
dc91c728 | 840 | } else if (extra_pages) { |
5097cdf6 | 841 | chunk_size = min(size, PFN_PHYS(extra_pages)); |
626d7508 JG |
842 | pfn_s = PFN_UP(addr); |
843 | n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s; | |
844 | extra_pages -= n_pfns; | |
845 | xen_add_extra_mem(pfn_s, n_pfns); | |
846 | xen_max_p2m_pfn = pfn_s + n_pfns; | |
dc91c728 | 847 | } else |
f5775e0b | 848 | discard = true; |
3654581e JF |
849 | } |
850 | ||
f5775e0b DV |
851 | if (!discard) |
852 | xen_align_and_add_e820_region(addr, chunk_size, type); | |
b5b43ced | 853 | |
5097cdf6 JG |
854 | addr += chunk_size; |
855 | size -= chunk_size; | |
856 | if (size == 0) { | |
dc91c728 | 857 | i++; |
5097cdf6 JG |
858 | if (i < xen_e820_map_entries) { |
859 | addr = xen_e820_map[i].addr; | |
860 | size = xen_e820_map[i].size; | |
861 | } | |
862 | } | |
35ae11fd | 863 | } |
b792c755 | 864 | |
25b884a8 DV |
865 | /* |
866 | * Set the rest as identity mapped, in case PCI BARs are | |
867 | * located here. | |
25b884a8 | 868 | */ |
5097cdf6 | 869 | set_phys_range_identity(addr / PAGE_SIZE, ~0ul); |
25b884a8 | 870 | |
b792c755 | 871 | /* |
9ec23a7f IC |
872 | * In domU, the ISA region is normal, usable memory, but we |
873 | * reserve ISA memory anyway because too many things poke | |
b792c755 JF |
874 | * about in there. |
875 | */ | |
876 | e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, | |
877 | E820_RESERVED); | |
5ead97c8 | 878 | |
be5bf9fa JF |
879 | sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); |
880 | ||
808fdb71 JG |
881 | /* |
882 | * Check whether the kernel itself conflicts with the target E820 map. | |
883 | * Failing now is better than running into weird problems later due | |
884 | * to relocating (and even reusing) pages with kernel text or data. | |
885 | */ | |
886 | if (xen_is_e820_reserved(__pa_symbol(_text), | |
887 | __pa_symbol(__bss_stop) - __pa_symbol(_text))) { | |
888 | xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n"); | |
889 | BUG(); | |
890 | } | |
891 | ||
04414baa JG |
892 | /* |
893 | * Check for a conflict of the hypervisor supplied page tables with | |
894 | * the target E820 map. | |
895 | */ | |
896 | xen_pt_check_e820(); | |
897 | ||
8f5b0c63 JG |
898 | xen_reserve_xen_mfnlist(); |
899 | ||
4b9c1537 JG |
900 | /* Check for a conflict of the initrd with the target E820 map. */ |
901 | if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image, | |
902 | boot_params.hdr.ramdisk_size)) { | |
903 | phys_addr_t new_area, start, size; | |
904 | ||
905 | new_area = xen_find_free_area(boot_params.hdr.ramdisk_size); | |
906 | if (!new_area) { | |
907 | xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n"); | |
908 | BUG(); | |
909 | } | |
910 | ||
911 | start = boot_params.hdr.ramdisk_image; | |
912 | size = boot_params.hdr.ramdisk_size; | |
913 | xen_phys_memcpy(new_area, start, size); | |
914 | pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n", | |
915 | start, start + size, new_area, new_area + size); | |
916 | memblock_free(start, size); | |
917 | boot_params.hdr.ramdisk_image = new_area; | |
918 | boot_params.ext_ramdisk_image = new_area >> 32; | |
919 | } | |
920 | ||
5097cdf6 JG |
921 | /* |
922 | * Set identity map on non-RAM pages and prepare remapping the | |
923 | * underlying RAM. | |
924 | */ | |
925 | xen_set_identity_and_remap(max_pfn); | |
926 | ||
5ead97c8 JF |
927 | return "Xen"; |
928 | } | |
929 | ||
abacaadc DV |
930 | /* |
931 | * Machine specific memory setup for auto-translated guests. | |
932 | */ | |
933 | char * __init xen_auto_xlated_memory_setup(void) | |
934 | { | |
abacaadc DV |
935 | struct xen_memory_map memmap; |
936 | int i; | |
937 | int rc; | |
938 | ||
939 | memmap.nr_entries = E820MAX; | |
69632ecf | 940 | set_xen_guest_handle(memmap.buffer, xen_e820_map); |
abacaadc DV |
941 | |
942 | rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap); | |
943 | if (rc < 0) | |
944 | panic("No memory map (%d)\n", rc); | |
945 | ||
69632ecf JG |
946 | xen_e820_map_entries = memmap.nr_entries; |
947 | ||
948 | sanitize_e820_map(xen_e820_map, ARRAY_SIZE(xen_e820_map), | |
949 | &xen_e820_map_entries); | |
abacaadc | 950 | |
69632ecf JG |
951 | for (i = 0; i < xen_e820_map_entries; i++) |
952 | e820_add_region(xen_e820_map[i].addr, xen_e820_map[i].size, | |
953 | xen_e820_map[i].type); | |
abacaadc | 954 | |
70e61199 JG |
955 | /* Remove p2m info, it is not needed. */ |
956 | xen_start_info->mfn_list = 0; | |
957 | xen_start_info->first_p2m_pfn = 0; | |
958 | xen_start_info->nr_p2m_frames = 0; | |
abacaadc DV |
959 | |
960 | return "Xen"; | |
961 | } | |
962 | ||
d2eea68e RM |
963 | /* |
964 | * Set the bit indicating "nosegneg" library variants should be used. | |
6a52e4b1 JF |
965 | * We only need to bother in pure 32-bit mode; compat 32-bit processes |
966 | * can have un-truncated segments, so wrapping around is allowed. | |
d2eea68e | 967 | */ |
08b6d290 | 968 | static void __init fiddle_vdso(void) |
d2eea68e | 969 | { |
6a52e4b1 | 970 | #ifdef CONFIG_X86_32 |
6f121e54 AL |
971 | /* |
972 | * This could be called before selected_vdso32 is initialized, so | |
973 | * just fiddle with both possible images. vdso_image_32_syscall | |
974 | * can't be selected, since it only exists on 64-bit systems. | |
975 | */ | |
6a52e4b1 | 976 | u32 *mask; |
6f121e54 AL |
977 | mask = vdso_image_32_int80.data + |
978 | vdso_image_32_int80.sym_VDSO32_NOTE_MASK; | |
6a52e4b1 | 979 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; |
6f121e54 AL |
980 | mask = vdso_image_32_sysenter.data + |
981 | vdso_image_32_sysenter.sym_VDSO32_NOTE_MASK; | |
d2eea68e | 982 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; |
6fcac6d3 | 983 | #endif |
d2eea68e RM |
984 | } |
985 | ||
148f9bb8 | 986 | static int register_callback(unsigned type, const void *func) |
e2a81baf | 987 | { |
88459d4c JF |
988 | struct callback_register callback = { |
989 | .type = type, | |
990 | .address = XEN_CALLBACK(__KERNEL_CS, func), | |
e2a81baf JF |
991 | .flags = CALLBACKF_mask_events, |
992 | }; | |
993 | ||
88459d4c JF |
994 | return HYPERVISOR_callback_op(CALLBACKOP_register, &callback); |
995 | } | |
996 | ||
148f9bb8 | 997 | void xen_enable_sysenter(void) |
88459d4c | 998 | { |
6fcac6d3 | 999 | int ret; |
62541c37 | 1000 | unsigned sysenter_feature; |
6fcac6d3 JF |
1001 | |
1002 | #ifdef CONFIG_X86_32 | |
62541c37 | 1003 | sysenter_feature = X86_FEATURE_SEP; |
6fcac6d3 | 1004 | #else |
62541c37 | 1005 | sysenter_feature = X86_FEATURE_SYSENTER32; |
6fcac6d3 | 1006 | #endif |
88459d4c | 1007 | |
62541c37 JF |
1008 | if (!boot_cpu_has(sysenter_feature)) |
1009 | return; | |
1010 | ||
6fcac6d3 | 1011 | ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target); |
62541c37 JF |
1012 | if(ret != 0) |
1013 | setup_clear_cpu_cap(sysenter_feature); | |
e2a81baf JF |
1014 | } |
1015 | ||
148f9bb8 | 1016 | void xen_enable_syscall(void) |
6fcac6d3 JF |
1017 | { |
1018 | #ifdef CONFIG_X86_64 | |
6fcac6d3 | 1019 | int ret; |
6fcac6d3 JF |
1020 | |
1021 | ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target); | |
1022 | if (ret != 0) { | |
d5303b81 | 1023 | printk(KERN_ERR "Failed to set syscall callback: %d\n", ret); |
62541c37 JF |
1024 | /* Pretty fatal; 64-bit userspace has no other |
1025 | mechanism for syscalls. */ | |
1026 | } | |
1027 | ||
1028 | if (boot_cpu_has(X86_FEATURE_SYSCALL32)) { | |
6fcac6d3 JF |
1029 | ret = register_callback(CALLBACKTYPE_syscall32, |
1030 | xen_syscall32_target); | |
d5303b81 | 1031 | if (ret != 0) |
62541c37 | 1032 | setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); |
6fcac6d3 JF |
1033 | } |
1034 | #endif /* CONFIG_X86_64 */ | |
1035 | } | |
ea9f9274 | 1036 | |
d285d683 | 1037 | void __init xen_pvmmu_arch_setup(void) |
5ead97c8 | 1038 | { |
5ead97c8 JF |
1039 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments); |
1040 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables); | |
1041 | ||
d285d683 MR |
1042 | HYPERVISOR_vm_assist(VMASST_CMD_enable, |
1043 | VMASST_TYPE_pae_extended_cr3); | |
5ead97c8 | 1044 | |
88459d4c JF |
1045 | if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) || |
1046 | register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback)) | |
1047 | BUG(); | |
5ead97c8 | 1048 | |
e2a81baf | 1049 | xen_enable_sysenter(); |
6fcac6d3 | 1050 | xen_enable_syscall(); |
d285d683 MR |
1051 | } |
1052 | ||
1053 | /* This function is not called for HVM domains */ | |
1054 | void __init xen_arch_setup(void) | |
1055 | { | |
1056 | xen_panic_handler_init(); | |
1057 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
1058 | xen_pvmmu_arch_setup(); | |
1059 | ||
5ead97c8 JF |
1060 | #ifdef CONFIG_ACPI |
1061 | if (!(xen_start_info->flags & SIF_INITDOMAIN)) { | |
1062 | printk(KERN_INFO "ACPI in unprivileged domain disabled\n"); | |
1063 | disable_acpi(); | |
1064 | } | |
1065 | #endif | |
1066 | ||
1067 | memcpy(boot_command_line, xen_start_info->cmd_line, | |
1068 | MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ? | |
1069 | COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE); | |
1070 | ||
bc15fde7 | 1071 | /* Set up idle, making sure it calls safe_halt() pvop */ |
d91ee586 | 1072 | disable_cpuidle(); |
48cdd828 | 1073 | disable_cpufreq(); |
6a377ddc | 1074 | WARN_ON(xen_set_default_idle()); |
d2eea68e | 1075 | fiddle_vdso(); |
8d54db79 KRW |
1076 | #ifdef CONFIG_NUMA |
1077 | numa_off = 1; | |
1078 | #endif | |
5ead97c8 | 1079 | } |