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> |
5ead97c8 JF |
13 | |
14 | #include <asm/elf.h> | |
6c3652ef | 15 | #include <asm/vdso.h> |
5ead97c8 JF |
16 | #include <asm/e820.h> |
17 | #include <asm/setup.h> | |
b792c755 | 18 | #include <asm/acpi.h> |
5ead97c8 JF |
19 | #include <asm/xen/hypervisor.h> |
20 | #include <asm/xen/hypercall.h> | |
21 | ||
45263cb0 | 22 | #include <xen/xen.h> |
8006ec3e | 23 | #include <xen/page.h> |
e2a81baf | 24 | #include <xen/interface/callback.h> |
35ae11fd | 25 | #include <xen/interface/memory.h> |
5ead97c8 JF |
26 | #include <xen/interface/physdev.h> |
27 | #include <xen/features.h> | |
28 | ||
29 | #include "xen-ops.h" | |
d2eea68e | 30 | #include "vdso.h" |
5ead97c8 JF |
31 | |
32 | /* These are code, but not functions. Defined in entry.S */ | |
33 | extern const char xen_hypervisor_callback[]; | |
34 | extern const char xen_failsafe_callback[]; | |
f63c2f24 T |
35 | extern void xen_sysenter_target(void); |
36 | extern void xen_syscall_target(void); | |
37 | extern void xen_syscall32_target(void); | |
5ead97c8 | 38 | |
42ee1471 | 39 | /* Amount of extra memory space we add to the e820 ranges */ |
8b5d44a5 | 40 | struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata; |
42ee1471 | 41 | |
aa24411b DV |
42 | /* Number of pages released from the initial allocation. */ |
43 | unsigned long xen_released_pages; | |
44 | ||
698bb8d1 JF |
45 | /* |
46 | * The maximum amount of extra memory compared to the base size. The | |
47 | * main scaling factor is the size of struct page. At extreme ratios | |
48 | * of base:extra, all the base memory can be filled with page | |
49 | * structures for the extra memory, leaving no space for anything | |
50 | * else. | |
51 | * | |
52 | * 10x seems like a reasonable balance between scaling flexibility and | |
53 | * leaving a practically usable system. | |
54 | */ | |
55 | #define EXTRA_MEM_RATIO (10) | |
56 | ||
dc91c728 | 57 | static void __init xen_add_extra_mem(u64 start, u64 size) |
42ee1471 | 58 | { |
6eaa412f | 59 | unsigned long pfn; |
dc91c728 | 60 | int i; |
6eaa412f | 61 | |
dc91c728 DV |
62 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) { |
63 | /* Add new region. */ | |
64 | if (xen_extra_mem[i].size == 0) { | |
65 | xen_extra_mem[i].start = start; | |
66 | xen_extra_mem[i].size = size; | |
67 | break; | |
68 | } | |
69 | /* Append to existing region. */ | |
70 | if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) { | |
71 | xen_extra_mem[i].size += size; | |
72 | break; | |
73 | } | |
74 | } | |
75 | if (i == XEN_EXTRA_MEM_MAX_REGIONS) | |
76 | printk(KERN_WARNING "Warning: not enough extra memory regions\n"); | |
42ee1471 | 77 | |
d4bbf7e7 | 78 | memblock_reserve(start, size); |
2f7acb20 | 79 | |
dc91c728 | 80 | xen_max_p2m_pfn = PFN_DOWN(start + size); |
6eaa412f | 81 | |
dc91c728 | 82 | for (pfn = PFN_DOWN(start); pfn <= xen_max_p2m_pfn; pfn++) |
6eaa412f | 83 | __set_phys_to_machine(pfn, INVALID_P2M_ENTRY); |
42ee1471 JF |
84 | } |
85 | ||
f3f436e3 DV |
86 | static unsigned long __init xen_release_chunk(unsigned long start, |
87 | unsigned long end) | |
093d7b46 MR |
88 | { |
89 | struct xen_memory_reservation reservation = { | |
90 | .address_bits = 0, | |
91 | .extent_order = 0, | |
92 | .domid = DOMID_SELF | |
93 | }; | |
f89e048e | 94 | unsigned long len = 0; |
093d7b46 MR |
95 | unsigned long pfn; |
96 | int ret; | |
97 | ||
f89e048e JF |
98 | for(pfn = start; pfn < end; pfn++) { |
99 | unsigned long mfn = pfn_to_mfn(pfn); | |
100 | ||
101 | /* Make sure pfn exists to start with */ | |
102 | if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn) | |
103 | continue; | |
104 | ||
105 | set_xen_guest_handle(reservation.extent_start, &mfn); | |
106 | reservation.nr_extents = 1; | |
107 | ||
108 | ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, | |
109 | &reservation); | |
98f531da | 110 | WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret); |
f89e048e | 111 | if (ret == 1) { |
6eaa412f | 112 | __set_phys_to_machine(pfn, INVALID_P2M_ENTRY); |
f89e048e JF |
113 | len++; |
114 | } | |
115 | } | |
98f531da IM |
116 | printk(KERN_INFO "Freeing %lx-%lx pfn range: %lu pages freed\n", |
117 | start, end, len); | |
093d7b46 MR |
118 | |
119 | return len; | |
120 | } | |
121 | ||
f3f436e3 DV |
122 | static unsigned long __init xen_set_identity_and_release( |
123 | const struct e820entry *list, size_t map_size, unsigned long nr_pages) | |
093d7b46 | 124 | { |
f3f436e3 | 125 | phys_addr_t start = 0; |
093d7b46 | 126 | unsigned long released = 0; |
68df0da7 | 127 | unsigned long identity = 0; |
f3f436e3 | 128 | const struct e820entry *entry; |
68df0da7 KRW |
129 | int i; |
130 | ||
f3f436e3 DV |
131 | /* |
132 | * Combine non-RAM regions and gaps until a RAM region (or the | |
133 | * end of the map) is reached, then set the 1:1 map and | |
134 | * release the pages (if available) in those non-RAM regions. | |
135 | * | |
136 | * The combined non-RAM regions are rounded to a whole number | |
137 | * of pages so any partial pages are accessible via the 1:1 | |
138 | * mapping. This is needed for some BIOSes that put (for | |
139 | * example) the DMI tables in a reserved region that begins on | |
140 | * a non-page boundary. | |
141 | */ | |
68df0da7 | 142 | for (i = 0, entry = list; i < map_size; i++, entry++) { |
f3f436e3 | 143 | phys_addr_t end = entry->addr + entry->size; |
68df0da7 | 144 | |
f3f436e3 DV |
145 | if (entry->type == E820_RAM || i == map_size - 1) { |
146 | unsigned long start_pfn = PFN_DOWN(start); | |
147 | unsigned long end_pfn = PFN_UP(end); | |
68df0da7 | 148 | |
f3f436e3 DV |
149 | if (entry->type == E820_RAM) |
150 | end_pfn = PFN_UP(entry->addr); | |
68df0da7 | 151 | |
f3f436e3 DV |
152 | if (start_pfn < end_pfn) { |
153 | if (start_pfn < nr_pages) | |
154 | released += xen_release_chunk( | |
155 | start_pfn, min(end_pfn, nr_pages)); | |
68df0da7 | 156 | |
68df0da7 | 157 | identity += set_phys_range_identity( |
f3f436e3 DV |
158 | start_pfn, end_pfn); |
159 | } | |
160 | start = end; | |
68df0da7 | 161 | } |
68df0da7 | 162 | } |
f3f436e3 DV |
163 | |
164 | printk(KERN_INFO "Released %lu pages of unused memory\n", released); | |
165 | printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity); | |
166 | ||
167 | return released; | |
68df0da7 | 168 | } |
d312ae87 DV |
169 | |
170 | static unsigned long __init xen_get_max_pages(void) | |
171 | { | |
172 | unsigned long max_pages = MAX_DOMAIN_PAGES; | |
173 | domid_t domid = DOMID_SELF; | |
174 | int ret; | |
175 | ||
d3db7281 IC |
176 | /* |
177 | * For the initial domain we use the maximum reservation as | |
178 | * the maximum page. | |
179 | * | |
180 | * For guest domains the current maximum reservation reflects | |
181 | * the current maximum rather than the static maximum. In this | |
182 | * case the e820 map provided to us will cover the static | |
183 | * maximum region. | |
184 | */ | |
185 | if (xen_initial_domain()) { | |
186 | ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid); | |
187 | if (ret > 0) | |
188 | max_pages = ret; | |
189 | } | |
190 | ||
d312ae87 DV |
191 | return min(max_pages, MAX_DOMAIN_PAGES); |
192 | } | |
193 | ||
dc91c728 DV |
194 | static void xen_align_and_add_e820_region(u64 start, u64 size, int type) |
195 | { | |
196 | u64 end = start + size; | |
197 | ||
198 | /* Align RAM regions to page boundaries. */ | |
199 | if (type == E820_RAM) { | |
200 | start = PAGE_ALIGN(start); | |
201 | end &= ~((u64)PAGE_SIZE - 1); | |
202 | } | |
203 | ||
204 | e820_add_region(start, end - start, type); | |
205 | } | |
206 | ||
5ead97c8 JF |
207 | /** |
208 | * machine_specific_memory_setup - Hook for machine specific memory setup. | |
209 | **/ | |
5ead97c8 JF |
210 | char * __init xen_memory_setup(void) |
211 | { | |
35ae11fd IC |
212 | static struct e820entry map[E820MAX] __initdata; |
213 | ||
5ead97c8 | 214 | unsigned long max_pfn = xen_start_info->nr_pages; |
35ae11fd IC |
215 | unsigned long long mem_end; |
216 | int rc; | |
217 | struct xen_memory_map memmap; | |
dc91c728 | 218 | unsigned long max_pages; |
42ee1471 | 219 | unsigned long extra_pages = 0; |
35ae11fd | 220 | int i; |
9e9a5fcb | 221 | int op; |
5ead97c8 | 222 | |
8006ec3e | 223 | max_pfn = min(MAX_DOMAIN_PAGES, max_pfn); |
35ae11fd IC |
224 | mem_end = PFN_PHYS(max_pfn); |
225 | ||
226 | memmap.nr_entries = E820MAX; | |
227 | set_xen_guest_handle(memmap.buffer, map); | |
228 | ||
9e9a5fcb IC |
229 | op = xen_initial_domain() ? |
230 | XENMEM_machine_memory_map : | |
231 | XENMEM_memory_map; | |
232 | rc = HYPERVISOR_memory_op(op, &memmap); | |
35ae11fd | 233 | if (rc == -ENOSYS) { |
9ec23a7f | 234 | BUG_ON(xen_initial_domain()); |
35ae11fd IC |
235 | memmap.nr_entries = 1; |
236 | map[0].addr = 0ULL; | |
237 | map[0].size = mem_end; | |
238 | /* 8MB slack (to balance backend allocations). */ | |
239 | map[0].size += 8ULL << 20; | |
240 | map[0].type = E820_RAM; | |
241 | rc = 0; | |
242 | } | |
243 | BUG_ON(rc); | |
8006ec3e | 244 | |
dc91c728 DV |
245 | /* Make sure the Xen-supplied memory map is well-ordered. */ |
246 | sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries); | |
247 | ||
248 | max_pages = xen_get_max_pages(); | |
249 | if (max_pages > max_pfn) | |
250 | extra_pages += max_pages - max_pfn; | |
251 | ||
f3f436e3 DV |
252 | /* |
253 | * Set P2M for all non-RAM pages and E820 gaps to be identity | |
254 | * type PFNs. Any RAM pages that would be made inaccesible by | |
255 | * this are first released. | |
256 | */ | |
257 | xen_released_pages = xen_set_identity_and_release( | |
258 | map, memmap.nr_entries, max_pfn); | |
dc91c728 DV |
259 | extra_pages += xen_released_pages; |
260 | ||
261 | /* | |
262 | * Clamp the amount of extra memory to a EXTRA_MEM_RATIO | |
263 | * factor the base size. On non-highmem systems, the base | |
264 | * size is the full initial memory allocation; on highmem it | |
265 | * is limited to the max size of lowmem, so that it doesn't | |
266 | * get completely filled. | |
267 | * | |
268 | * In principle there could be a problem in lowmem systems if | |
269 | * the initial memory is also very large with respect to | |
270 | * lowmem, but we won't try to deal with that here. | |
271 | */ | |
272 | extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)), | |
273 | extra_pages); | |
274 | ||
275 | i = 0; | |
276 | while (i < memmap.nr_entries) { | |
277 | u64 addr = map[i].addr; | |
278 | u64 size = map[i].size; | |
279 | u32 type = map[i].type; | |
280 | ||
281 | if (type == E820_RAM) { | |
282 | if (addr < mem_end) { | |
283 | size = min(size, mem_end - addr); | |
284 | } else if (extra_pages) { | |
285 | size = min(size, (u64)extra_pages * PAGE_SIZE); | |
286 | extra_pages -= size / PAGE_SIZE; | |
287 | xen_add_extra_mem(addr, size); | |
288 | } else | |
289 | type = E820_UNUSABLE; | |
3654581e JF |
290 | } |
291 | ||
dc91c728 | 292 | xen_align_and_add_e820_region(addr, size, type); |
b5b43ced | 293 | |
dc91c728 DV |
294 | map[i].addr += size; |
295 | map[i].size -= size; | |
296 | if (map[i].size == 0) | |
297 | i++; | |
35ae11fd | 298 | } |
b792c755 JF |
299 | |
300 | /* | |
9ec23a7f IC |
301 | * In domU, the ISA region is normal, usable memory, but we |
302 | * reserve ISA memory anyway because too many things poke | |
b792c755 JF |
303 | * about in there. |
304 | */ | |
305 | e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS, | |
306 | E820_RESERVED); | |
5ead97c8 | 307 | |
be5bf9fa JF |
308 | /* |
309 | * Reserve Xen bits: | |
310 | * - mfn_list | |
311 | * - xen_start_info | |
312 | * See comment above "struct start_info" in <xen/interface/xen.h> | |
313 | */ | |
24aa0788 TH |
314 | memblock_reserve(__pa(xen_start_info->mfn_list), |
315 | xen_start_info->pt_base - xen_start_info->mfn_list); | |
be5bf9fa JF |
316 | |
317 | sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map); | |
318 | ||
5ead97c8 JF |
319 | return "Xen"; |
320 | } | |
321 | ||
d2eea68e RM |
322 | /* |
323 | * Set the bit indicating "nosegneg" library variants should be used. | |
6a52e4b1 JF |
324 | * We only need to bother in pure 32-bit mode; compat 32-bit processes |
325 | * can have un-truncated segments, so wrapping around is allowed. | |
d2eea68e | 326 | */ |
08b6d290 | 327 | static void __init fiddle_vdso(void) |
d2eea68e | 328 | { |
6a52e4b1 JF |
329 | #ifdef CONFIG_X86_32 |
330 | u32 *mask; | |
331 | mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK); | |
332 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; | |
333 | mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK); | |
d2eea68e | 334 | *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT; |
6fcac6d3 | 335 | #endif |
d2eea68e RM |
336 | } |
337 | ||
ae15a3b4 | 338 | static int __cpuinit register_callback(unsigned type, const void *func) |
e2a81baf | 339 | { |
88459d4c JF |
340 | struct callback_register callback = { |
341 | .type = type, | |
342 | .address = XEN_CALLBACK(__KERNEL_CS, func), | |
e2a81baf JF |
343 | .flags = CALLBACKF_mask_events, |
344 | }; | |
345 | ||
88459d4c JF |
346 | return HYPERVISOR_callback_op(CALLBACKOP_register, &callback); |
347 | } | |
348 | ||
349 | void __cpuinit xen_enable_sysenter(void) | |
350 | { | |
6fcac6d3 | 351 | int ret; |
62541c37 | 352 | unsigned sysenter_feature; |
6fcac6d3 JF |
353 | |
354 | #ifdef CONFIG_X86_32 | |
62541c37 | 355 | sysenter_feature = X86_FEATURE_SEP; |
6fcac6d3 | 356 | #else |
62541c37 | 357 | sysenter_feature = X86_FEATURE_SYSENTER32; |
6fcac6d3 | 358 | #endif |
88459d4c | 359 | |
62541c37 JF |
360 | if (!boot_cpu_has(sysenter_feature)) |
361 | return; | |
362 | ||
6fcac6d3 | 363 | ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target); |
62541c37 JF |
364 | if(ret != 0) |
365 | setup_clear_cpu_cap(sysenter_feature); | |
e2a81baf JF |
366 | } |
367 | ||
6fcac6d3 JF |
368 | void __cpuinit xen_enable_syscall(void) |
369 | { | |
370 | #ifdef CONFIG_X86_64 | |
6fcac6d3 | 371 | int ret; |
6fcac6d3 JF |
372 | |
373 | ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target); | |
374 | if (ret != 0) { | |
d5303b81 | 375 | printk(KERN_ERR "Failed to set syscall callback: %d\n", ret); |
62541c37 JF |
376 | /* Pretty fatal; 64-bit userspace has no other |
377 | mechanism for syscalls. */ | |
378 | } | |
379 | ||
380 | if (boot_cpu_has(X86_FEATURE_SYSCALL32)) { | |
6fcac6d3 JF |
381 | ret = register_callback(CALLBACKTYPE_syscall32, |
382 | xen_syscall32_target); | |
d5303b81 | 383 | if (ret != 0) |
62541c37 | 384 | setup_clear_cpu_cap(X86_FEATURE_SYSCALL32); |
6fcac6d3 JF |
385 | } |
386 | #endif /* CONFIG_X86_64 */ | |
387 | } | |
388 | ||
5ead97c8 JF |
389 | void __init xen_arch_setup(void) |
390 | { | |
f09f6d19 DD |
391 | xen_panic_handler_init(); |
392 | ||
5ead97c8 JF |
393 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments); |
394 | HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables); | |
395 | ||
396 | if (!xen_feature(XENFEAT_auto_translated_physmap)) | |
f63c2f24 T |
397 | HYPERVISOR_vm_assist(VMASST_CMD_enable, |
398 | VMASST_TYPE_pae_extended_cr3); | |
5ead97c8 | 399 | |
88459d4c JF |
400 | if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) || |
401 | register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback)) | |
402 | BUG(); | |
5ead97c8 | 403 | |
e2a81baf | 404 | xen_enable_sysenter(); |
6fcac6d3 | 405 | xen_enable_syscall(); |
e2a81baf | 406 | |
5ead97c8 JF |
407 | #ifdef CONFIG_ACPI |
408 | if (!(xen_start_info->flags & SIF_INITDOMAIN)) { | |
409 | printk(KERN_INFO "ACPI in unprivileged domain disabled\n"); | |
410 | disable_acpi(); | |
411 | } | |
412 | #endif | |
413 | ||
414 | memcpy(boot_command_line, xen_start_info->cmd_line, | |
415 | MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ? | |
416 | COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE); | |
417 | ||
bc15fde7 JF |
418 | /* Set up idle, making sure it calls safe_halt() pvop */ |
419 | #ifdef CONFIG_X86_32 | |
420 | boot_cpu_data.hlt_works_ok = 1; | |
421 | #endif | |
d91ee586 | 422 | disable_cpuidle(); |
e5fd47bf | 423 | WARN_ON(set_pm_idle_to_default()); |
d2eea68e | 424 | fiddle_vdso(); |
5ead97c8 | 425 | } |