]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
5ead97c8 JF |
2 | /* |
3 | * Machine specific setup for xen | |
4 | * | |
5 | * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 | |
6 | */ | |
7 | ||
7a2463dc | 8 | #include <linux/init.h> |
5ead97c8 JF |
9 | #include <linux/sched.h> |
10 | #include <linux/mm.h> | |
11 | #include <linux/pm.h> | |
a9ce6bc1 | 12 | #include <linux/memblock.h> |
d91ee586 | 13 | #include <linux/cpuidle.h> |
48cdd828 | 14 | #include <linux/cpufreq.h> |
5ead97c8 JF |
15 | |
16 | #include <asm/elf.h> | |
6c3652ef | 17 | #include <asm/vdso.h> |
66441bd3 | 18 | #include <asm/e820/api.h> |
5ead97c8 | 19 | #include <asm/setup.h> |
b792c755 | 20 | #include <asm/acpi.h> |
8d54db79 | 21 | #include <asm/numa.h> |
5ead97c8 JF |
22 | #include <asm/xen/hypervisor.h> |
23 | #include <asm/xen/hypercall.h> | |
24 | ||
45263cb0 | 25 | #include <xen/xen.h> |
8006ec3e | 26 | #include <xen/page.h> |
e2a81baf | 27 | #include <xen/interface/callback.h> |
35ae11fd | 28 | #include <xen/interface/memory.h> |
5ead97c8 JF |
29 | #include <xen/interface/physdev.h> |
30 | #include <xen/features.h> | |
808fdb71 | 31 | #include <xen/hvc-console.h> |
5ead97c8 | 32 | #include "xen-ops.h" |
d2eea68e | 33 | #include "vdso.h" |
1f3ac86b | 34 | #include "mmu.h" |
5ead97c8 | 35 | |
c70727a5 JG |
36 | #define GB(x) ((uint64_t)(x) * 1024 * 1024 * 1024) |
37 | ||
42ee1471 | 38 | /* Amount of extra memory space we add to the e820 ranges */ |
8b5d44a5 | 39 | struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; |
42ee1471 | 40 | |
aa24411b DV |
41 | /* Number of pages released from the initial allocation. */ |
42 | unsigned long xen_released_pages; | |
43 | ||
69632ecf | 44 | /* E820 map used during setting up memory. */ |
e7dbf7ad | 45 | static struct e820_table xen_e820_table __initdata; |
69632ecf | 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 | { |
e7dbf7ad | 201 | const struct e820_entry *entry = xen_e820_table.entries; |
2e2fb754 KRW |
202 | unsigned int i; |
203 | unsigned long done = 0; | |
2e2fb754 | 204 | |
e7dbf7ad | 205 | for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) { |
2e2fb754 KRW |
206 | unsigned long s_pfn; |
207 | unsigned long e_pfn; | |
2e2fb754 | 208 | |
09821ff1 | 209 | if (entry->type != E820_TYPE_RAM) |
2e2fb754 KRW |
210 | continue; |
211 | ||
c3d93f88 | 212 | e_pfn = PFN_DOWN(entry->addr + entry->size); |
2e2fb754 | 213 | |
4fbb67e3 | 214 | /* We only care about E820 after this */ |
abed7d07 | 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 | ||
1f3ac86b | 344 | mfn_save = virt_to_mfn(buf); |
e201bfcc | 345 | |
1f3ac86b JG |
346 | for (ident_pfn_iter = start_pfn, remap_pfn_iter = remap_pfn; |
347 | ident_pfn_iter < ident_end_pfn; | |
348 | ident_pfn_iter += REMAP_SIZE, remap_pfn_iter += REMAP_SIZE) { | |
349 | chunk = (left < REMAP_SIZE) ? left : REMAP_SIZE; | |
4fbb67e3 | 350 | |
1f3ac86b JG |
351 | /* Map first pfn to xen_remap_buf */ |
352 | mfn = pfn_to_mfn(ident_pfn_iter); | |
353 | set_pte_mfn(buf, mfn, PAGE_KERNEL); | |
4fbb67e3 | 354 | |
1f3ac86b JG |
355 | /* Save mapping information in page */ |
356 | xen_remap_buf.next_area_mfn = xen_remap_mfn; | |
357 | xen_remap_buf.target_pfn = remap_pfn_iter; | |
358 | xen_remap_buf.size = chunk; | |
359 | for (i = 0; i < chunk; i++) | |
360 | xen_remap_buf.mfns[i] = pfn_to_mfn(ident_pfn_iter + i); | |
4fbb67e3 | 361 | |
1f3ac86b JG |
362 | /* Put remap buf into list. */ |
363 | xen_remap_mfn = mfn; | |
4fbb67e3 | 364 | |
1f3ac86b | 365 | /* Set identity map */ |
bc7142cf | 366 | set_phys_range_identity(ident_pfn_iter, ident_pfn_iter + chunk); |
83d51ab4 | 367 | |
1f3ac86b | 368 | left -= chunk; |
4fbb67e3 | 369 | } |
83d51ab4 | 370 | |
1f3ac86b JG |
371 | /* Restore old xen_remap_buf mapping */ |
372 | set_pte_mfn(buf, mfn_save, PAGE_KERNEL); | |
83d51ab4 DV |
373 | } |
374 | ||
4fbb67e3 MR |
375 | /* |
376 | * This function takes a contiguous pfn range that needs to be identity mapped | |
377 | * and: | |
378 | * | |
379 | * 1) Finds a new range of pfns to use to remap based on E820 and remap_pfn. | |
380 | * 2) Calls the do_ function to actually do the mapping/remapping work. | |
381 | * | |
382 | * The goal is to not allocate additional memory but to remap the existing | |
383 | * pages. In the case of an error the underlying memory is simply released back | |
384 | * to Xen and not remapped. | |
385 | */ | |
76f0a486 | 386 | static unsigned long __init xen_set_identity_and_remap_chunk( |
69632ecf | 387 | unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages, |
5097cdf6 | 388 | unsigned long remap_pfn) |
4fbb67e3 MR |
389 | { |
390 | unsigned long pfn; | |
391 | unsigned long i = 0; | |
392 | unsigned long n = end_pfn - start_pfn; | |
393 | ||
dd14be92 JG |
394 | if (remap_pfn == 0) |
395 | remap_pfn = nr_pages; | |
396 | ||
4fbb67e3 MR |
397 | while (i < n) { |
398 | unsigned long cur_pfn = start_pfn + i; | |
399 | unsigned long left = n - i; | |
400 | unsigned long size = left; | |
401 | unsigned long remap_range_size; | |
402 | ||
403 | /* Do not remap pages beyond the current allocation */ | |
404 | if (cur_pfn >= nr_pages) { | |
405 | /* Identity map remaining pages */ | |
bc7142cf | 406 | set_phys_range_identity(cur_pfn, cur_pfn + size); |
4fbb67e3 MR |
407 | break; |
408 | } | |
409 | if (cur_pfn + size > nr_pages) | |
410 | size = nr_pages - cur_pfn; | |
411 | ||
69632ecf | 412 | remap_range_size = xen_find_pfn_range(&remap_pfn); |
4fbb67e3 MR |
413 | if (!remap_range_size) { |
414 | pr_warning("Unable to find available pfn range, not remapping identity pages\n"); | |
415 | xen_set_identity_and_release_chunk(cur_pfn, | |
5097cdf6 | 416 | cur_pfn + left, nr_pages); |
4fbb67e3 MR |
417 | break; |
418 | } | |
419 | /* Adjust size to fit in current e820 RAM region */ | |
420 | if (size > remap_range_size) | |
421 | size = remap_range_size; | |
422 | ||
1f3ac86b | 423 | xen_do_set_identity_and_remap_chunk(cur_pfn, size, remap_pfn); |
4fbb67e3 MR |
424 | |
425 | /* Update variables to reflect new mappings. */ | |
426 | i += size; | |
427 | remap_pfn += size; | |
4fbb67e3 MR |
428 | } |
429 | ||
430 | /* | |
431 | * If the PFNs are currently mapped, the VA mapping also needs | |
432 | * to be updated to be 1:1. | |
433 | */ | |
434 | for (pfn = start_pfn; pfn <= max_pfn_mapped && pfn < end_pfn; pfn++) | |
435 | (void)HYPERVISOR_update_va_mapping( | |
436 | (unsigned long)__va(pfn << PAGE_SHIFT), | |
437 | mfn_pte(pfn, PAGE_KERNEL_IO), 0); | |
438 | ||
439 | return remap_pfn; | |
440 | } | |
441 | ||
dd14be92 JG |
442 | static unsigned long __init xen_count_remap_pages( |
443 | unsigned long start_pfn, unsigned long end_pfn, unsigned long nr_pages, | |
444 | unsigned long remap_pages) | |
445 | { | |
446 | if (start_pfn >= nr_pages) | |
447 | return remap_pages; | |
448 | ||
449 | return remap_pages + min(end_pfn, nr_pages) - start_pfn; | |
450 | } | |
451 | ||
452 | static unsigned long __init xen_foreach_remap_area(unsigned long nr_pages, | |
453 | unsigned long (*func)(unsigned long start_pfn, unsigned long end_pfn, | |
454 | unsigned long nr_pages, unsigned long last_val)) | |
093d7b46 | 455 | { |
f3f436e3 | 456 | phys_addr_t start = 0; |
dd14be92 | 457 | unsigned long ret_val = 0; |
e7dbf7ad | 458 | const struct e820_entry *entry = xen_e820_table.entries; |
68df0da7 KRW |
459 | int i; |
460 | ||
f3f436e3 DV |
461 | /* |
462 | * Combine non-RAM regions and gaps until a RAM region (or the | |
dd14be92 JG |
463 | * end of the map) is reached, then call the provided function |
464 | * to perform its duty on the non-RAM region. | |
f3f436e3 DV |
465 | * |
466 | * The combined non-RAM regions are rounded to a whole number | |
467 | * of pages so any partial pages are accessible via the 1:1 | |
468 | * mapping. This is needed for some BIOSes that put (for | |
469 | * example) the DMI tables in a reserved region that begins on | |
470 | * a non-page boundary. | |
471 | */ | |
e7dbf7ad | 472 | for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) { |
f3f436e3 | 473 | phys_addr_t end = entry->addr + entry->size; |
e7dbf7ad | 474 | if (entry->type == E820_TYPE_RAM || i == xen_e820_table.nr_entries - 1) { |
f3f436e3 DV |
475 | unsigned long start_pfn = PFN_DOWN(start); |
476 | unsigned long end_pfn = PFN_UP(end); | |
68df0da7 | 477 | |
09821ff1 | 478 | if (entry->type == E820_TYPE_RAM) |
f3f436e3 | 479 | end_pfn = PFN_UP(entry->addr); |
68df0da7 | 480 | |
83d51ab4 | 481 | if (start_pfn < end_pfn) |
dd14be92 JG |
482 | ret_val = func(start_pfn, end_pfn, nr_pages, |
483 | ret_val); | |
f3f436e3 | 484 | start = end; |
68df0da7 | 485 | } |
68df0da7 | 486 | } |
f3f436e3 | 487 | |
dd14be92 | 488 | return ret_val; |
4fbb67e3 | 489 | } |
1f3ac86b JG |
490 | |
491 | /* | |
492 | * Remap the memory prepared in xen_do_set_identity_and_remap_chunk(). | |
493 | * The remap information (which mfn remap to which pfn) is contained in the | |
494 | * to be remapped memory itself in a linked list anchored at xen_remap_mfn. | |
495 | * This scheme allows to remap the different chunks in arbitrary order while | |
496 | * the resulting mapping will be independant from the order. | |
497 | */ | |
498 | void __init xen_remap_memory(void) | |
499 | { | |
500 | unsigned long buf = (unsigned long)&xen_remap_buf; | |
bf1b9ddf | 501 | unsigned long mfn_save, pfn; |
1f3ac86b JG |
502 | unsigned long remapped = 0; |
503 | unsigned int i; | |
504 | unsigned long pfn_s = ~0UL; | |
505 | unsigned long len = 0; | |
506 | ||
507 | mfn_save = virt_to_mfn(buf); | |
508 | ||
509 | while (xen_remap_mfn != INVALID_P2M_ENTRY) { | |
510 | /* Map the remap information */ | |
511 | set_pte_mfn(buf, xen_remap_mfn, PAGE_KERNEL); | |
512 | ||
513 | BUG_ON(xen_remap_mfn != xen_remap_buf.mfns[0]); | |
514 | ||
515 | pfn = xen_remap_buf.target_pfn; | |
516 | for (i = 0; i < xen_remap_buf.size; i++) { | |
bf1b9ddf | 517 | xen_update_mem_tables(pfn, xen_remap_buf.mfns[i]); |
1f3ac86b JG |
518 | remapped++; |
519 | pfn++; | |
520 | } | |
521 | if (pfn_s == ~0UL || pfn == pfn_s) { | |
522 | pfn_s = xen_remap_buf.target_pfn; | |
523 | len += xen_remap_buf.size; | |
524 | } else if (pfn_s + len == xen_remap_buf.target_pfn) { | |
525 | len += xen_remap_buf.size; | |
526 | } else { | |
626d7508 | 527 | xen_del_extra_mem(pfn_s, len); |
1f3ac86b JG |
528 | pfn_s = xen_remap_buf.target_pfn; |
529 | len = xen_remap_buf.size; | |
530 | } | |
1f3ac86b JG |
531 | xen_remap_mfn = xen_remap_buf.next_area_mfn; |
532 | } | |
533 | ||
534 | if (pfn_s != ~0UL && len) | |
626d7508 | 535 | xen_del_extra_mem(pfn_s, len); |
1f3ac86b JG |
536 | |
537 | set_pte_mfn(buf, mfn_save, PAGE_KERNEL); | |
538 | ||
539 | pr_info("Remapped %ld page(s)\n", remapped); | |
540 | } | |
541 | ||
c70727a5 JG |
542 | static unsigned long __init xen_get_pages_limit(void) |
543 | { | |
544 | unsigned long limit; | |
545 | ||
546 | #ifdef CONFIG_X86_32 | |
547 | limit = GB(64) / PAGE_SIZE; | |
548 | #else | |
cb9e444b | 549 | limit = MAXMEM / PAGE_SIZE; |
c70727a5 JG |
550 | if (!xen_initial_domain() && xen_512gb_limit) |
551 | limit = GB(512) / PAGE_SIZE; | |
552 | #endif | |
553 | return limit; | |
554 | } | |
555 | ||
d312ae87 DV |
556 | static unsigned long __init xen_get_max_pages(void) |
557 | { | |
c70727a5 | 558 | unsigned long max_pages, limit; |
d312ae87 | 559 | domid_t domid = DOMID_SELF; |
24f775a6 | 560 | long ret; |
d312ae87 | 561 | |
c70727a5 JG |
562 | limit = xen_get_pages_limit(); |
563 | max_pages = limit; | |
564 | ||
d3db7281 IC |
565 | /* |
566 | * For the initial domain we use the maximum reservation as | |
567 | * the maximum page. | |
568 | * | |
569 | * For guest domains the current maximum reservation reflects | |
570 | * the current maximum rather than the static maximum. In this | |
571 | * case the e820 map provided to us will cover the static | |
572 | * maximum region. | |
573 | */ | |
574 | if (xen_initial_domain()) { | |
575 | ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); | |
576 | if (ret > 0) | |
577 | max_pages = ret; | |
578 | } | |
579 | ||
c70727a5 | 580 | return min(max_pages, limit); |
d312ae87 DV |
581 | } |
582 | ||
a3f52396 JG |
583 | static void __init xen_align_and_add_e820_region(phys_addr_t start, |
584 | phys_addr_t size, int type) | |
dc91c728 | 585 | { |
3ba5c867 | 586 | phys_addr_t end = start + size; |
dc91c728 DV |
587 | |
588 | /* Align RAM regions to page boundaries. */ | |
09821ff1 | 589 | if (type == E820_TYPE_RAM) { |
dc91c728 | 590 | start = PAGE_ALIGN(start); |
3ba5c867 | 591 | end &= ~((phys_addr_t)PAGE_SIZE - 1); |
dc91c728 DV |
592 | } |
593 | ||
ab6bc04c | 594 | e820__range_add(start, end - start, type); |
dc91c728 DV |
595 | } |
596 | ||
69632ecf | 597 | static void __init xen_ignore_unusable(void) |
3bc38cbc | 598 | { |
e7dbf7ad | 599 | struct e820_entry *entry = xen_e820_table.entries; |
3bc38cbc DV |
600 | unsigned int i; |
601 | ||
e7dbf7ad | 602 | for (i = 0; i < xen_e820_table.nr_entries; i++, entry++) { |
09821ff1 IM |
603 | if (entry->type == E820_TYPE_UNUSABLE) |
604 | entry->type = E820_TYPE_RAM; | |
3bc38cbc DV |
605 | } |
606 | } | |
607 | ||
e612b4a7 JG |
608 | bool __init xen_is_e820_reserved(phys_addr_t start, phys_addr_t size) |
609 | { | |
8ec67d97 | 610 | struct e820_entry *entry; |
e612b4a7 JG |
611 | unsigned mapcnt; |
612 | phys_addr_t end; | |
613 | ||
614 | if (!size) | |
615 | return false; | |
616 | ||
617 | end = start + size; | |
e7dbf7ad | 618 | entry = xen_e820_table.entries; |
e612b4a7 | 619 | |
e7dbf7ad | 620 | for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++) { |
09821ff1 | 621 | if (entry->type == E820_TYPE_RAM && entry->addr <= start && |
e612b4a7 JG |
622 | (entry->addr + entry->size) >= end) |
623 | return false; | |
624 | ||
625 | entry++; | |
626 | } | |
627 | ||
628 | return true; | |
629 | } | |
630 | ||
9ddac5b7 JG |
631 | /* |
632 | * Find a free area in physical memory not yet reserved and compliant with | |
633 | * E820 map. | |
634 | * Used to relocate pre-allocated areas like initrd or p2m list which are in | |
635 | * conflict with the to be used E820 map. | |
636 | * In case no area is found, return 0. Otherwise return the physical address | |
637 | * of the area which is already reserved for convenience. | |
638 | */ | |
639 | phys_addr_t __init xen_find_free_area(phys_addr_t size) | |
640 | { | |
641 | unsigned mapcnt; | |
642 | phys_addr_t addr, start; | |
e7dbf7ad | 643 | struct e820_entry *entry = xen_e820_table.entries; |
9ddac5b7 | 644 | |
e7dbf7ad | 645 | for (mapcnt = 0; mapcnt < xen_e820_table.nr_entries; mapcnt++, entry++) { |
09821ff1 | 646 | if (entry->type != E820_TYPE_RAM || entry->size < size) |
9ddac5b7 JG |
647 | continue; |
648 | start = entry->addr; | |
649 | for (addr = start; addr < start + size; addr += PAGE_SIZE) { | |
650 | if (!memblock_is_reserved(addr)) | |
651 | continue; | |
652 | start = addr + PAGE_SIZE; | |
653 | if (start + size > entry->addr + entry->size) | |
654 | break; | |
655 | } | |
656 | if (addr >= start + size) { | |
657 | memblock_reserve(start, size); | |
658 | return start; | |
659 | } | |
660 | } | |
661 | ||
662 | return 0; | |
663 | } | |
664 | ||
4b9c1537 JG |
665 | /* |
666 | * Like memcpy, but with physical addresses for dest and src. | |
667 | */ | |
668 | static void __init xen_phys_memcpy(phys_addr_t dest, phys_addr_t src, | |
669 | phys_addr_t n) | |
670 | { | |
671 | phys_addr_t dest_off, src_off, dest_len, src_len, len; | |
672 | void *from, *to; | |
673 | ||
674 | while (n) { | |
675 | dest_off = dest & ~PAGE_MASK; | |
676 | src_off = src & ~PAGE_MASK; | |
677 | dest_len = n; | |
678 | if (dest_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off) | |
679 | dest_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - dest_off; | |
680 | src_len = n; | |
681 | if (src_len > (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off) | |
682 | src_len = (NR_FIX_BTMAPS << PAGE_SHIFT) - src_off; | |
683 | len = min(dest_len, src_len); | |
684 | to = early_memremap(dest - dest_off, dest_len + dest_off); | |
685 | from = early_memremap(src - src_off, src_len + src_off); | |
686 | memcpy(to, from, len); | |
687 | early_memunmap(to, dest_len + dest_off); | |
688 | early_memunmap(from, src_len + src_off); | |
689 | n -= len; | |
690 | dest += len; | |
691 | src += len; | |
692 | } | |
693 | } | |
694 | ||
8f5b0c63 JG |
695 | /* |
696 | * Reserve Xen mfn_list. | |
8f5b0c63 JG |
697 | */ |
698 | static void __init xen_reserve_xen_mfnlist(void) | |
699 | { | |
70e61199 JG |
700 | phys_addr_t start, size; |
701 | ||
8f5b0c63 | 702 | if (xen_start_info->mfn_list >= __START_KERNEL_map) { |
70e61199 JG |
703 | start = __pa(xen_start_info->mfn_list); |
704 | size = PFN_ALIGN(xen_start_info->nr_pages * | |
705 | sizeof(unsigned long)); | |
706 | } else { | |
707 | start = PFN_PHYS(xen_start_info->first_p2m_pfn); | |
708 | size = PFN_PHYS(xen_start_info->nr_p2m_frames); | |
709 | } | |
710 | ||
7ecec850 RL |
711 | memblock_reserve(start, size); |
712 | if (!xen_is_e820_reserved(start, size)) | |
8f5b0c63 | 713 | return; |
8f5b0c63 | 714 | |
70e61199 JG |
715 | #ifdef CONFIG_X86_32 |
716 | /* | |
717 | * Relocating the p2m on 32 bit system to an arbitrary virtual address | |
718 | * is not supported, so just give up. | |
719 | */ | |
720 | xen_raw_console_write("Xen hypervisor allocated p2m list conflicts with E820 map\n"); | |
721 | BUG(); | |
722 | #else | |
723 | xen_relocate_p2m(); | |
7ecec850 | 724 | memblock_free(start, size); |
70e61199 | 725 | #endif |
8f5b0c63 JG |
726 | } |
727 | ||
5ead97c8 JF |
728 | /** |
729 | * machine_specific_memory_setup - Hook for machine specific memory setup. | |
730 | **/ | |
5ead97c8 JF |
731 | char * __init xen_memory_setup(void) |
732 | { | |
626d7508 | 733 | unsigned long max_pfn, pfn_s, n_pfns; |
5097cdf6 JG |
734 | phys_addr_t mem_end, addr, size, chunk_size; |
735 | u32 type; | |
35ae11fd IC |
736 | int rc; |
737 | struct xen_memory_map memmap; | |
dc91c728 | 738 | unsigned long max_pages; |
42ee1471 | 739 | unsigned long extra_pages = 0; |
35ae11fd | 740 | int i; |
9e9a5fcb | 741 | int op; |
5ead97c8 | 742 | |
c70727a5 JG |
743 | xen_parse_512gb(); |
744 | max_pfn = xen_get_pages_limit(); | |
745 | max_pfn = min(max_pfn, xen_start_info->nr_pages); | |
35ae11fd IC |
746 | mem_end = PFN_PHYS(max_pfn); |
747 | ||
e7dbf7ad IM |
748 | memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries); |
749 | set_xen_guest_handle(memmap.buffer, xen_e820_table.entries); | |
35ae11fd | 750 | |
9e9a5fcb IC |
751 | op = xen_initial_domain() ? |
752 | XENMEM_machine_memory_map : | |
753 | XENMEM_memory_map; | |
754 | rc = HYPERVISOR_memory_op(op, &memmap); | |
35ae11fd | 755 | if (rc == -ENOSYS) { |
9ec23a7f | 756 | BUG_ON(xen_initial_domain()); |
35ae11fd | 757 | memmap.nr_entries = 1; |
e7dbf7ad IM |
758 | xen_e820_table.entries[0].addr = 0ULL; |
759 | xen_e820_table.entries[0].size = mem_end; | |
35ae11fd | 760 | /* 8MB slack (to balance backend allocations). */ |
e7dbf7ad IM |
761 | xen_e820_table.entries[0].size += 8ULL << 20; |
762 | xen_e820_table.entries[0].type = E820_TYPE_RAM; | |
35ae11fd IC |
763 | rc = 0; |
764 | } | |
765 | BUG_ON(rc); | |
1ea644c8 | 766 | BUG_ON(memmap.nr_entries == 0); |
e7dbf7ad | 767 | xen_e820_table.nr_entries = memmap.nr_entries; |
8006ec3e | 768 | |
3bc38cbc DV |
769 | /* |
770 | * Xen won't allow a 1:1 mapping to be created to UNUSABLE | |
771 | * regions, so if we're using the machine memory map leave the | |
772 | * region as RAM as it is in the pseudo-physical map. | |
773 | * | |
774 | * UNUSABLE regions in domUs are not handled and will need | |
775 | * a patch in the future. | |
776 | */ | |
777 | if (xen_initial_domain()) | |
69632ecf | 778 | xen_ignore_unusable(); |
3bc38cbc | 779 | |
dc91c728 | 780 | /* Make sure the Xen-supplied memory map is well-ordered. */ |
f9748fa0 | 781 | e820__update_table(&xen_e820_table); |
dc91c728 DV |
782 | |
783 | max_pages = xen_get_max_pages(); | |
dc91c728 | 784 | |
5097cdf6 | 785 | /* How many extra pages do we need due to remapping? */ |
dd14be92 | 786 | max_pages += xen_foreach_remap_area(max_pfn, xen_count_remap_pages); |
eafd72e0 JG |
787 | |
788 | if (max_pages > max_pfn) | |
789 | extra_pages += max_pages - max_pfn; | |
2e2fb754 | 790 | |
dc91c728 DV |
791 | /* |
792 | * Clamp the amount of extra memory to a EXTRA_MEM_RATIO | |
793 | * factor the base size. On non-highmem systems, the base | |
794 | * size is the full initial memory allocation; on highmem it | |
795 | * is limited to the max size of lowmem, so that it doesn't | |
796 | * get completely filled. | |
797 | * | |
c70727a5 JG |
798 | * Make sure we have no memory above max_pages, as this area |
799 | * isn't handled by the p2m management. | |
800 | * | |
dc91c728 DV |
801 | * In principle there could be a problem in lowmem systems if |
802 | * the initial memory is also very large with respect to | |
803 | * lowmem, but we won't try to deal with that here. | |
804 | */ | |
c70727a5 JG |
805 | extra_pages = min3(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)), |
806 | extra_pages, max_pages - max_pfn); | |
dc91c728 | 807 | i = 0; |
e7dbf7ad IM |
808 | addr = xen_e820_table.entries[0].addr; |
809 | size = xen_e820_table.entries[0].size; | |
810 | while (i < xen_e820_table.nr_entries) { | |
f5775e0b | 811 | |
5097cdf6 | 812 | chunk_size = size; |
e7dbf7ad | 813 | type = xen_e820_table.entries[i].type; |
dc91c728 | 814 | |
09821ff1 | 815 | if (type == E820_TYPE_RAM) { |
dc91c728 | 816 | if (addr < mem_end) { |
5097cdf6 | 817 | chunk_size = min(size, mem_end - addr); |
dc91c728 | 818 | } else if (extra_pages) { |
5097cdf6 | 819 | chunk_size = min(size, PFN_PHYS(extra_pages)); |
626d7508 JG |
820 | pfn_s = PFN_UP(addr); |
821 | n_pfns = PFN_DOWN(addr + chunk_size) - pfn_s; | |
822 | extra_pages -= n_pfns; | |
823 | xen_add_extra_mem(pfn_s, n_pfns); | |
824 | xen_max_p2m_pfn = pfn_s + n_pfns; | |
dc91c728 | 825 | } else |
b3cf8528 | 826 | type = E820_TYPE_UNUSABLE; |
3654581e JF |
827 | } |
828 | ||
b3cf8528 | 829 | xen_align_and_add_e820_region(addr, chunk_size, type); |
b5b43ced | 830 | |
5097cdf6 JG |
831 | addr += chunk_size; |
832 | size -= chunk_size; | |
833 | if (size == 0) { | |
dc91c728 | 834 | i++; |
e7dbf7ad IM |
835 | if (i < xen_e820_table.nr_entries) { |
836 | addr = xen_e820_table.entries[i].addr; | |
837 | size = xen_e820_table.entries[i].size; | |
5097cdf6 JG |
838 | } |
839 | } | |
35ae11fd | 840 | } |
b792c755 | 841 | |
25b884a8 DV |
842 | /* |
843 | * Set the rest as identity mapped, in case PCI BARs are | |
844 | * located here. | |
25b884a8 | 845 | */ |
5097cdf6 | 846 | set_phys_range_identity(addr / PAGE_SIZE, ~0ul); |
25b884a8 | 847 | |
b792c755 | 848 | /* |
9ec23a7f IC |
849 | * In domU, the ISA region is normal, usable memory, but we |
850 | * reserve ISA memory anyway because too many things poke | |
b792c755 JF |
851 | * about in there. |
852 | */ | |
f9748fa0 | 853 | e820__range_add(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, E820_TYPE_RESERVED); |
5ead97c8 | 854 | |
f9748fa0 | 855 | e820__update_table(e820_table); |
be5bf9fa | 856 | |
808fdb71 JG |
857 | /* |
858 | * Check whether the kernel itself conflicts with the target E820 map. | |
859 | * Failing now is better than running into weird problems later due | |
860 | * to relocating (and even reusing) pages with kernel text or data. | |
861 | */ | |
862 | if (xen_is_e820_reserved(__pa_symbol(_text), | |
863 | __pa_symbol(__bss_stop) - __pa_symbol(_text))) { | |
864 | xen_raw_console_write("Xen hypervisor allocated kernel memory conflicts with E820 map\n"); | |
865 | BUG(); | |
866 | } | |
867 | ||
04414baa JG |
868 | /* |
869 | * Check for a conflict of the hypervisor supplied page tables with | |
870 | * the target E820 map. | |
871 | */ | |
872 | xen_pt_check_e820(); | |
873 | ||
8f5b0c63 JG |
874 | xen_reserve_xen_mfnlist(); |
875 | ||
4b9c1537 JG |
876 | /* Check for a conflict of the initrd with the target E820 map. */ |
877 | if (xen_is_e820_reserved(boot_params.hdr.ramdisk_image, | |
878 | boot_params.hdr.ramdisk_size)) { | |
879 | phys_addr_t new_area, start, size; | |
880 | ||
881 | new_area = xen_find_free_area(boot_params.hdr.ramdisk_size); | |
882 | if (!new_area) { | |
883 | xen_raw_console_write("Can't find new memory area for initrd needed due to E820 map conflict\n"); | |
884 | BUG(); | |
885 | } | |
886 | ||
887 | start = boot_params.hdr.ramdisk_image; | |
888 | size = boot_params.hdr.ramdisk_size; | |
889 | xen_phys_memcpy(new_area, start, size); | |
890 | pr_info("initrd moved from [mem %#010llx-%#010llx] to [mem %#010llx-%#010llx]\n", | |
891 | start, start + size, new_area, new_area + size); | |
892 | memblock_free(start, size); | |
893 | boot_params.hdr.ramdisk_image = new_area; | |
894 | boot_params.ext_ramdisk_image = new_area >> 32; | |
895 | } | |
896 | ||
5097cdf6 JG |
897 | /* |
898 | * Set identity map on non-RAM pages and prepare remapping the | |
899 | * underlying RAM. | |
900 | */ | |
dd14be92 JG |
901 | xen_foreach_remap_area(max_pfn, xen_set_identity_and_remap_chunk); |
902 | ||
903 | pr_info("Released %ld page(s)\n", xen_released_pages); | |
5097cdf6 | 904 | |
5ead97c8 JF |
905 | return "Xen"; |
906 | } | |
907 | ||
abacaadc DV |
908 | /* |
909 | * Machine specific memory setup for auto-translated guests. | |
910 | */ | |
911 | char * __init xen_auto_xlated_memory_setup(void) | |
912 | { | |
abacaadc DV |
913 | struct xen_memory_map memmap; |
914 | int i; | |
915 | int rc; | |
916 | ||
e7dbf7ad IM |
917 | memmap.nr_entries = ARRAY_SIZE(xen_e820_table.entries); |
918 | set_xen_guest_handle(memmap.buffer, xen_e820_table.entries); | |
abacaadc DV |
919 | |
920 | rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap); | |
921 | if (rc < 0) | |
922 | panic("No memory map (%d)\n", rc); | |
923 | ||
e7dbf7ad | 924 | xen_e820_table.nr_entries = memmap.nr_entries; |
69632ecf | 925 | |
f9748fa0 | 926 | e820__update_table(&xen_e820_table); |
abacaadc | 927 | |
e7dbf7ad IM |
928 | for (i = 0; i < xen_e820_table.nr_entries; i++) |
929 | e820__range_add(xen_e820_table.entries[i].addr, xen_e820_table.entries[i].size, xen_e820_table.entries[i].type); | |
abacaadc | 930 | |
70e61199 JG |
931 | /* Remove p2m info, it is not needed. */ |
932 | xen_start_info->mfn_list = 0; | |
933 | xen_start_info->first_p2m_pfn = 0; | |
934 | xen_start_info->nr_p2m_frames = 0; | |
abacaadc DV |
935 | |
936 | return "Xen"; | |
937 | } | |
938 | ||
d2eea68e RM |
939 | /* |
940 | * Set the bit indicating "nosegneg" library variants should be used. | |
6a52e4b1 JF |
941 | * We only need to bother in pure 32-bit mode; compat 32-bit processes |
942 | * can have un-truncated segments, so wrapping around is allowed. | |
d2eea68e | 943 | */ |
08b6d290 | 944 | static void __init fiddle_vdso(void) |
d2eea68e | 945 | { |
6a52e4b1 | 946 | #ifdef CONFIG_X86_32 |
0a6d1fa0 AL |
947 | u32 *mask = vdso_image_32.data + |
948 | vdso_image_32.sym_VDSO32_NOTE_MASK; | |
d2eea68e | 949 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; |
6fcac6d3 | 950 | #endif |
d2eea68e RM |
951 | } |
952 | ||
148f9bb8 | 953 | static int register_callback(unsigned type, const void *func) |
e2a81baf | 954 | { |
88459d4c JF |
955 | struct callback_register callback = { |
956 | .type = type, | |
957 | .address = XEN_CALLBACK(__KERNEL_CS, func), | |
e2a81baf JF |
958 | .flags = CALLBACKF_mask_events, |
959 | }; | |
960 | ||
88459d4c JF |
961 | return HYPERVISOR_callback_op(CALLBACKOP_register, &callback); |
962 | } | |
963 | ||
148f9bb8 | 964 | void xen_enable_sysenter(void) |
88459d4c | 965 | { |
6fcac6d3 | 966 | int ret; |
62541c37 | 967 | unsigned sysenter_feature; |
6fcac6d3 JF |
968 | |
969 | #ifdef CONFIG_X86_32 | |
62541c37 | 970 | sysenter_feature = X86_FEATURE_SEP; |
6fcac6d3 | 971 | #else |
62541c37 | 972 | sysenter_feature = X86_FEATURE_SYSENTER32; |
6fcac6d3 | 973 | #endif |
88459d4c | 974 | |
62541c37 JF |
975 | if (!boot_cpu_has(sysenter_feature)) |
976 | return; | |
977 | ||
6fcac6d3 | 978 | ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target); |
62541c37 JF |
979 | if(ret != 0) |
980 | setup_clear_cpu_cap(sysenter_feature); | |
e2a81baf JF |
981 | } |
982 | ||
148f9bb8 | 983 | void xen_enable_syscall(void) |
6fcac6d3 JF |
984 | { |
985 | #ifdef CONFIG_X86_64 | |
6fcac6d3 | 986 | int ret; |
6fcac6d3 JF |
987 | |
988 | ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target); | |
989 | if (ret != 0) { | |
d5303b81 | 990 | printk(KERN_ERR "Failed to set syscall callback: %d\n", ret); |
62541c37 JF |
991 | /* Pretty fatal; 64-bit userspace has no other |
992 | mechanism for syscalls. */ | |
993 | } | |
994 | ||
995 | if (boot_cpu_has(X86_FEATURE_SYSCALL32)) { | |
6fcac6d3 JF |
996 | ret = register_callback(CALLBACKTYPE_syscall32, |
997 | xen_syscall32_target); | |
d5303b81 | 998 | if (ret != 0) |
62541c37 | 999 | setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); |
6fcac6d3 JF |
1000 | } |
1001 | #endif /* CONFIG_X86_64 */ | |
1002 | } | |
ea9f9274 | 1003 | |
d285d683 | 1004 | void __init xen_pvmmu_arch_setup(void) |
5ead97c8 | 1005 | { |
5ead97c8 JF |
1006 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments); |
1007 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables); | |
1008 | ||
d285d683 MR |
1009 | HYPERVISOR_vm_assist(VMASST_CMD_enable, |
1010 | VMASST_TYPE_pae_extended_cr3); | |
5ead97c8 | 1011 | |
88459d4c JF |
1012 | if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) || |
1013 | register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback)) | |
1014 | BUG(); | |
5ead97c8 | 1015 | |
e2a81baf | 1016 | xen_enable_sysenter(); |
6fcac6d3 | 1017 | xen_enable_syscall(); |
d285d683 MR |
1018 | } |
1019 | ||
1020 | /* This function is not called for HVM domains */ | |
1021 | void __init xen_arch_setup(void) | |
1022 | { | |
1023 | xen_panic_handler_init(); | |
82616f95 | 1024 | xen_pvmmu_arch_setup(); |
d285d683 | 1025 | |
5ead97c8 JF |
1026 | #ifdef CONFIG_ACPI |
1027 | if (!(xen_start_info->flags & SIF_INITDOMAIN)) { | |
1028 | printk(KERN_INFO "ACPI in unprivileged domain disabled\n"); | |
1029 | disable_acpi(); | |
1030 | } | |
1031 | #endif | |
1032 | ||
1033 | memcpy(boot_command_line, xen_start_info->cmd_line, | |
1034 | MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ? | |
1035 | COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE); | |
1036 | ||
bc15fde7 | 1037 | /* Set up idle, making sure it calls safe_halt() pvop */ |
d91ee586 | 1038 | disable_cpuidle(); |
48cdd828 | 1039 | disable_cpufreq(); |
6a377ddc | 1040 | WARN_ON(xen_set_default_idle()); |
d2eea68e | 1041 | fiddle_vdso(); |
8d54db79 KRW |
1042 | #ifdef CONFIG_NUMA |
1043 | numa_off = 1; | |
1044 | #endif | |
5ead97c8 | 1045 | } |