Commit | Line | Data |
---|---|---|
9a0b5817 | 1 | #include <linux/module.h> |
f6a57033 | 2 | #include <linux/sched.h> |
2f1dafe5 | 3 | #include <linux/mutex.h> |
9a0b5817 | 4 | #include <linux/list.h> |
8b5a10fc | 5 | #include <linux/stringify.h> |
19d36ccd AK |
6 | #include <linux/kprobes.h> |
7 | #include <linux/mm.h> | |
8 | #include <linux/vmalloc.h> | |
3945dab4 | 9 | #include <linux/memory.h> |
3d55cc8a | 10 | #include <linux/stop_machine.h> |
9a0b5817 GH |
11 | #include <asm/alternative.h> |
12 | #include <asm/sections.h> | |
19d36ccd | 13 | #include <asm/pgtable.h> |
8f4e956b AK |
14 | #include <asm/mce.h> |
15 | #include <asm/nmi.h> | |
b097976e | 16 | #include <asm/vsyscall.h> |
e587cadd | 17 | #include <asm/cacheflush.h> |
78ff7fae | 18 | #include <asm/tlbflush.h> |
e587cadd | 19 | #include <asm/io.h> |
78ff7fae | 20 | #include <asm/fixmap.h> |
9a0b5817 | 21 | |
ab144f5e AK |
22 | #define MAX_PATCH_LEN (255-1) |
23 | ||
09488165 JB |
24 | #ifdef CONFIG_HOTPLUG_CPU |
25 | static int smp_alt_once; | |
9a0b5817 | 26 | |
d167a518 GH |
27 | static int __init bootonly(char *str) |
28 | { | |
29 | smp_alt_once = 1; | |
30 | return 1; | |
31 | } | |
b7fb4af0 | 32 | __setup("smp-alt-boot", bootonly); |
09488165 JB |
33 | #else |
34 | #define smp_alt_once 1 | |
35 | #endif | |
36 | ||
8b5a10fc | 37 | static int __initdata_or_module debug_alternative; |
b7fb4af0 | 38 | |
d167a518 GH |
39 | static int __init debug_alt(char *str) |
40 | { | |
41 | debug_alternative = 1; | |
42 | return 1; | |
43 | } | |
d167a518 GH |
44 | __setup("debug-alternative", debug_alt); |
45 | ||
09488165 JB |
46 | static int noreplace_smp; |
47 | ||
b7fb4af0 JF |
48 | static int __init setup_noreplace_smp(char *str) |
49 | { | |
50 | noreplace_smp = 1; | |
51 | return 1; | |
52 | } | |
53 | __setup("noreplace-smp", setup_noreplace_smp); | |
54 | ||
959b4fdf | 55 | #ifdef CONFIG_PARAVIRT |
8b5a10fc | 56 | static int __initdata_or_module noreplace_paravirt = 0; |
959b4fdf JF |
57 | |
58 | static int __init setup_noreplace_paravirt(char *str) | |
59 | { | |
60 | noreplace_paravirt = 1; | |
61 | return 1; | |
62 | } | |
63 | __setup("noreplace-paravirt", setup_noreplace_paravirt); | |
64 | #endif | |
b7fb4af0 | 65 | |
d167a518 GH |
66 | #define DPRINTK(fmt, args...) if (debug_alternative) \ |
67 | printk(KERN_DEBUG fmt, args) | |
68 | ||
8b5a10fc | 69 | #if defined(GENERIC_NOP1) && !defined(CONFIG_X86_64) |
9a0b5817 GH |
70 | /* Use inline assembly to define this because the nops are defined |
71 | as inline assembly strings in the include files and we cannot | |
72 | get them easily into strings. */ | |
8b5a10fc | 73 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nintelnops: " |
9a0b5817 | 74 | GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6 |
f4be31ec SR |
75 | GENERIC_NOP7 GENERIC_NOP8 |
76 | "\t.previous"); | |
121d7bf5 | 77 | extern const unsigned char intelnops[]; |
8b5a10fc JB |
78 | static const unsigned char *const __initconst_or_module |
79 | intel_nops[ASM_NOP_MAX+1] = { | |
9a0b5817 GH |
80 | NULL, |
81 | intelnops, | |
82 | intelnops + 1, | |
83 | intelnops + 1 + 2, | |
84 | intelnops + 1 + 2 + 3, | |
85 | intelnops + 1 + 2 + 3 + 4, | |
86 | intelnops + 1 + 2 + 3 + 4 + 5, | |
87 | intelnops + 1 + 2 + 3 + 4 + 5 + 6, | |
88 | intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
89 | }; | |
d167a518 GH |
90 | #endif |
91 | ||
92 | #ifdef K8_NOP1 | |
8b5a10fc | 93 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nk8nops: " |
d167a518 | 94 | K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6 |
f4be31ec SR |
95 | K8_NOP7 K8_NOP8 |
96 | "\t.previous"); | |
121d7bf5 | 97 | extern const unsigned char k8nops[]; |
8b5a10fc JB |
98 | static const unsigned char *const __initconst_or_module |
99 | k8_nops[ASM_NOP_MAX+1] = { | |
9a0b5817 GH |
100 | NULL, |
101 | k8nops, | |
102 | k8nops + 1, | |
103 | k8nops + 1 + 2, | |
104 | k8nops + 1 + 2 + 3, | |
105 | k8nops + 1 + 2 + 3 + 4, | |
106 | k8nops + 1 + 2 + 3 + 4 + 5, | |
107 | k8nops + 1 + 2 + 3 + 4 + 5 + 6, | |
108 | k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
109 | }; | |
d167a518 GH |
110 | #endif |
111 | ||
8b5a10fc JB |
112 | #if defined(K7_NOP1) && !defined(CONFIG_X86_64) |
113 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nk7nops: " | |
d167a518 | 114 | K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6 |
f4be31ec SR |
115 | K7_NOP7 K7_NOP8 |
116 | "\t.previous"); | |
121d7bf5 | 117 | extern const unsigned char k7nops[]; |
8b5a10fc JB |
118 | static const unsigned char *const __initconst_or_module |
119 | k7_nops[ASM_NOP_MAX+1] = { | |
9a0b5817 GH |
120 | NULL, |
121 | k7nops, | |
122 | k7nops + 1, | |
123 | k7nops + 1 + 2, | |
124 | k7nops + 1 + 2 + 3, | |
125 | k7nops + 1 + 2 + 3 + 4, | |
126 | k7nops + 1 + 2 + 3 + 4 + 5, | |
127 | k7nops + 1 + 2 + 3 + 4 + 5 + 6, | |
128 | k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
129 | }; | |
d167a518 GH |
130 | #endif |
131 | ||
32c464f5 | 132 | #ifdef P6_NOP1 |
8b5a10fc | 133 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\np6nops: " |
32c464f5 | 134 | P6_NOP1 P6_NOP2 P6_NOP3 P6_NOP4 P6_NOP5 P6_NOP6 |
f4be31ec SR |
135 | P6_NOP7 P6_NOP8 |
136 | "\t.previous"); | |
32c464f5 | 137 | extern const unsigned char p6nops[]; |
8b5a10fc JB |
138 | static const unsigned char *const __initconst_or_module |
139 | p6_nops[ASM_NOP_MAX+1] = { | |
32c464f5 JB |
140 | NULL, |
141 | p6nops, | |
142 | p6nops + 1, | |
143 | p6nops + 1 + 2, | |
144 | p6nops + 1 + 2 + 3, | |
145 | p6nops + 1 + 2 + 3 + 4, | |
146 | p6nops + 1 + 2 + 3 + 4 + 5, | |
147 | p6nops + 1 + 2 + 3 + 4 + 5 + 6, | |
148 | p6nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
149 | }; | |
150 | #endif | |
151 | ||
d167a518 GH |
152 | #ifdef CONFIG_X86_64 |
153 | ||
154 | extern char __vsyscall_0; | |
8b5a10fc | 155 | static const unsigned char *const *__init_or_module find_nop_table(void) |
d167a518 | 156 | { |
f31d731e PA |
157 | if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && |
158 | boot_cpu_has(X86_FEATURE_NOPL)) | |
159 | return p6_nops; | |
160 | else | |
161 | return k8_nops; | |
d167a518 GH |
162 | } |
163 | ||
164 | #else /* CONFIG_X86_64 */ | |
165 | ||
8b5a10fc | 166 | static const unsigned char *const *__init_or_module find_nop_table(void) |
9a0b5817 | 167 | { |
f31d731e PA |
168 | if (boot_cpu_has(X86_FEATURE_K8)) |
169 | return k8_nops; | |
170 | else if (boot_cpu_has(X86_FEATURE_K7)) | |
171 | return k7_nops; | |
172 | else if (boot_cpu_has(X86_FEATURE_NOPL)) | |
173 | return p6_nops; | |
174 | else | |
175 | return intel_nops; | |
9a0b5817 GH |
176 | } |
177 | ||
d167a518 GH |
178 | #endif /* CONFIG_X86_64 */ |
179 | ||
ab144f5e | 180 | /* Use this to add nops to a buffer, then text_poke the whole buffer. */ |
8b5a10fc | 181 | static void __init_or_module add_nops(void *insns, unsigned int len) |
139ec7c4 | 182 | { |
121d7bf5 | 183 | const unsigned char *const *noptable = find_nop_table(); |
139ec7c4 RR |
184 | |
185 | while (len > 0) { | |
186 | unsigned int noplen = len; | |
187 | if (noplen > ASM_NOP_MAX) | |
188 | noplen = ASM_NOP_MAX; | |
ab144f5e | 189 | memcpy(insns, noptable[noplen], noplen); |
139ec7c4 RR |
190 | insns += noplen; |
191 | len -= noplen; | |
192 | } | |
193 | } | |
194 | ||
d167a518 | 195 | extern struct alt_instr __alt_instructions[], __alt_instructions_end[]; |
d167a518 | 196 | extern u8 *__smp_locks[], *__smp_locks_end[]; |
8b5a10fc | 197 | static void *text_poke_early(void *addr, const void *opcode, size_t len); |
d167a518 | 198 | |
9a0b5817 GH |
199 | /* Replace instructions with better alternatives for this CPU type. |
200 | This runs before SMP is initialized to avoid SMP problems with | |
201 | self modifying code. This implies that assymetric systems where | |
202 | APs have less capabilities than the boot processor are not handled. | |
203 | Tough. Make sure you disable such features by hand. */ | |
204 | ||
8b5a10fc JB |
205 | void __init_or_module apply_alternatives(struct alt_instr *start, |
206 | struct alt_instr *end) | |
9a0b5817 | 207 | { |
9a0b5817 | 208 | struct alt_instr *a; |
1b1d9258 | 209 | u8 insnbuf[MAX_PATCH_LEN]; |
9a0b5817 | 210 | |
77bf90ed | 211 | DPRINTK("%s: alt table %p -> %p\n", __func__, start, end); |
9a0b5817 | 212 | for (a = start; a < end; a++) { |
ab144f5e | 213 | u8 *instr = a->instr; |
9a0b5817 | 214 | BUG_ON(a->replacementlen > a->instrlen); |
ab144f5e | 215 | BUG_ON(a->instrlen > sizeof(insnbuf)); |
9a0b5817 GH |
216 | if (!boot_cpu_has(a->cpuid)) |
217 | continue; | |
d167a518 GH |
218 | #ifdef CONFIG_X86_64 |
219 | /* vsyscall code is not mapped yet. resolve it manually. */ | |
220 | if (instr >= (u8 *)VSYSCALL_START && instr < (u8*)VSYSCALL_END) { | |
221 | instr = __va(instr - (u8*)VSYSCALL_START + (u8*)__pa_symbol(&__vsyscall_0)); | |
222 | DPRINTK("%s: vsyscall fixup: %p => %p\n", | |
77bf90ed | 223 | __func__, a->instr, instr); |
d167a518 GH |
224 | } |
225 | #endif | |
ab144f5e | 226 | memcpy(insnbuf, a->replacement, a->replacementlen); |
1b1d9258 JB |
227 | if (*insnbuf == 0xe8 && a->replacementlen == 5) |
228 | *(s32 *)(insnbuf + 1) += a->replacement - a->instr; | |
ab144f5e AK |
229 | add_nops(insnbuf + a->replacementlen, |
230 | a->instrlen - a->replacementlen); | |
e587cadd | 231 | text_poke_early(instr, insnbuf, a->instrlen); |
9a0b5817 GH |
232 | } |
233 | } | |
234 | ||
8ec4d41f GH |
235 | #ifdef CONFIG_SMP |
236 | ||
9a0b5817 GH |
237 | static void alternatives_smp_lock(u8 **start, u8 **end, u8 *text, u8 *text_end) |
238 | { | |
239 | u8 **ptr; | |
240 | ||
3945dab4 | 241 | mutex_lock(&text_mutex); |
9a0b5817 GH |
242 | for (ptr = start; ptr < end; ptr++) { |
243 | if (*ptr < text) | |
244 | continue; | |
245 | if (*ptr > text_end) | |
246 | continue; | |
f88f07e0 MD |
247 | /* turn DS segment override prefix into lock prefix */ |
248 | text_poke(*ptr, ((unsigned char []){0xf0}), 1); | |
9a0b5817 | 249 | }; |
3945dab4 | 250 | mutex_unlock(&text_mutex); |
9a0b5817 GH |
251 | } |
252 | ||
253 | static void alternatives_smp_unlock(u8 **start, u8 **end, u8 *text, u8 *text_end) | |
254 | { | |
9a0b5817 GH |
255 | u8 **ptr; |
256 | ||
b7fb4af0 JF |
257 | if (noreplace_smp) |
258 | return; | |
259 | ||
3945dab4 | 260 | mutex_lock(&text_mutex); |
9a0b5817 GH |
261 | for (ptr = start; ptr < end; ptr++) { |
262 | if (*ptr < text) | |
263 | continue; | |
264 | if (*ptr > text_end) | |
265 | continue; | |
f88f07e0 MD |
266 | /* turn lock prefix into DS segment override prefix */ |
267 | text_poke(*ptr, ((unsigned char []){0x3E}), 1); | |
9a0b5817 | 268 | }; |
3945dab4 | 269 | mutex_unlock(&text_mutex); |
9a0b5817 GH |
270 | } |
271 | ||
272 | struct smp_alt_module { | |
273 | /* what is this ??? */ | |
274 | struct module *mod; | |
275 | char *name; | |
276 | ||
277 | /* ptrs to lock prefixes */ | |
278 | u8 **locks; | |
279 | u8 **locks_end; | |
280 | ||
281 | /* .text segment, needed to avoid patching init code ;) */ | |
282 | u8 *text; | |
283 | u8 *text_end; | |
284 | ||
285 | struct list_head next; | |
286 | }; | |
287 | static LIST_HEAD(smp_alt_modules); | |
2f1dafe5 | 288 | static DEFINE_MUTEX(smp_alt); |
ca74a6f8 | 289 | static int smp_mode = 1; /* protected by smp_alt */ |
9a0b5817 | 290 | |
8b5a10fc JB |
291 | void __init_or_module alternatives_smp_module_add(struct module *mod, |
292 | char *name, | |
293 | void *locks, void *locks_end, | |
294 | void *text, void *text_end) | |
9a0b5817 GH |
295 | { |
296 | struct smp_alt_module *smp; | |
9a0b5817 | 297 | |
b7fb4af0 JF |
298 | if (noreplace_smp) |
299 | return; | |
300 | ||
9a0b5817 GH |
301 | if (smp_alt_once) { |
302 | if (boot_cpu_has(X86_FEATURE_UP)) | |
303 | alternatives_smp_unlock(locks, locks_end, | |
304 | text, text_end); | |
305 | return; | |
306 | } | |
307 | ||
308 | smp = kzalloc(sizeof(*smp), GFP_KERNEL); | |
309 | if (NULL == smp) | |
310 | return; /* we'll run the (safe but slow) SMP code then ... */ | |
311 | ||
312 | smp->mod = mod; | |
313 | smp->name = name; | |
314 | smp->locks = locks; | |
315 | smp->locks_end = locks_end; | |
316 | smp->text = text; | |
317 | smp->text_end = text_end; | |
318 | DPRINTK("%s: locks %p -> %p, text %p -> %p, name %s\n", | |
77bf90ed | 319 | __func__, smp->locks, smp->locks_end, |
9a0b5817 GH |
320 | smp->text, smp->text_end, smp->name); |
321 | ||
2f1dafe5 | 322 | mutex_lock(&smp_alt); |
9a0b5817 GH |
323 | list_add_tail(&smp->next, &smp_alt_modules); |
324 | if (boot_cpu_has(X86_FEATURE_UP)) | |
325 | alternatives_smp_unlock(smp->locks, smp->locks_end, | |
326 | smp->text, smp->text_end); | |
2f1dafe5 | 327 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
328 | } |
329 | ||
8b5a10fc | 330 | void __init_or_module alternatives_smp_module_del(struct module *mod) |
9a0b5817 GH |
331 | { |
332 | struct smp_alt_module *item; | |
9a0b5817 | 333 | |
b7fb4af0 | 334 | if (smp_alt_once || noreplace_smp) |
9a0b5817 GH |
335 | return; |
336 | ||
2f1dafe5 | 337 | mutex_lock(&smp_alt); |
9a0b5817 GH |
338 | list_for_each_entry(item, &smp_alt_modules, next) { |
339 | if (mod != item->mod) | |
340 | continue; | |
341 | list_del(&item->next); | |
2f1dafe5 | 342 | mutex_unlock(&smp_alt); |
77bf90ed | 343 | DPRINTK("%s: %s\n", __func__, item->name); |
9a0b5817 GH |
344 | kfree(item); |
345 | return; | |
346 | } | |
2f1dafe5 | 347 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
348 | } |
349 | ||
350 | void alternatives_smp_switch(int smp) | |
351 | { | |
352 | struct smp_alt_module *mod; | |
9a0b5817 | 353 | |
3047e99e IM |
354 | #ifdef CONFIG_LOCKDEP |
355 | /* | |
17abecfe IM |
356 | * Older binutils section handling bug prevented |
357 | * alternatives-replacement from working reliably. | |
358 | * | |
359 | * If this still occurs then you should see a hang | |
360 | * or crash shortly after this line: | |
3047e99e | 361 | */ |
17abecfe | 362 | printk("lockdep: fixing up alternatives.\n"); |
3047e99e IM |
363 | #endif |
364 | ||
b7fb4af0 | 365 | if (noreplace_smp || smp_alt_once) |
9a0b5817 GH |
366 | return; |
367 | BUG_ON(!smp && (num_online_cpus() > 1)); | |
368 | ||
2f1dafe5 | 369 | mutex_lock(&smp_alt); |
ca74a6f8 AK |
370 | |
371 | /* | |
372 | * Avoid unnecessary switches because it forces JIT based VMs to | |
373 | * throw away all cached translations, which can be quite costly. | |
374 | */ | |
375 | if (smp == smp_mode) { | |
376 | /* nothing */ | |
377 | } else if (smp) { | |
9a0b5817 | 378 | printk(KERN_INFO "SMP alternatives: switching to SMP code\n"); |
53756d37 JF |
379 | clear_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
380 | clear_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
9a0b5817 GH |
381 | list_for_each_entry(mod, &smp_alt_modules, next) |
382 | alternatives_smp_lock(mod->locks, mod->locks_end, | |
383 | mod->text, mod->text_end); | |
384 | } else { | |
385 | printk(KERN_INFO "SMP alternatives: switching to UP code\n"); | |
53756d37 JF |
386 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
387 | set_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
9a0b5817 GH |
388 | list_for_each_entry(mod, &smp_alt_modules, next) |
389 | alternatives_smp_unlock(mod->locks, mod->locks_end, | |
390 | mod->text, mod->text_end); | |
391 | } | |
ca74a6f8 | 392 | smp_mode = smp; |
2f1dafe5 | 393 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
394 | } |
395 | ||
2cfa1978 MH |
396 | /* Return 1 if the address range is reserved for smp-alternatives */ |
397 | int alternatives_text_reserved(void *start, void *end) | |
398 | { | |
399 | struct smp_alt_module *mod; | |
400 | u8 **ptr; | |
076dc4a6 MH |
401 | u8 *text_start = start; |
402 | u8 *text_end = end; | |
2cfa1978 MH |
403 | |
404 | list_for_each_entry(mod, &smp_alt_modules, next) { | |
076dc4a6 | 405 | if (mod->text > text_end || mod->text_end < text_start) |
2cfa1978 MH |
406 | continue; |
407 | for (ptr = mod->locks; ptr < mod->locks_end; ptr++) | |
076dc4a6 | 408 | if (text_start <= *ptr && text_end >= *ptr) |
2cfa1978 MH |
409 | return 1; |
410 | } | |
411 | ||
412 | return 0; | |
413 | } | |
8ec4d41f GH |
414 | #endif |
415 | ||
139ec7c4 | 416 | #ifdef CONFIG_PARAVIRT |
8b5a10fc JB |
417 | void __init_or_module apply_paravirt(struct paravirt_patch_site *start, |
418 | struct paravirt_patch_site *end) | |
139ec7c4 | 419 | { |
98de032b | 420 | struct paravirt_patch_site *p; |
ab144f5e | 421 | char insnbuf[MAX_PATCH_LEN]; |
139ec7c4 | 422 | |
959b4fdf JF |
423 | if (noreplace_paravirt) |
424 | return; | |
425 | ||
139ec7c4 RR |
426 | for (p = start; p < end; p++) { |
427 | unsigned int used; | |
428 | ||
ab144f5e | 429 | BUG_ON(p->len > MAX_PATCH_LEN); |
d34fda4a CW |
430 | /* prep the buffer with the original instructions */ |
431 | memcpy(insnbuf, p->instr, p->len); | |
93b1eab3 JF |
432 | used = pv_init_ops.patch(p->instrtype, p->clobbers, insnbuf, |
433 | (unsigned long)p->instr, p->len); | |
7f63c41c | 434 | |
63f70270 JF |
435 | BUG_ON(used > p->len); |
436 | ||
139ec7c4 | 437 | /* Pad the rest with nops */ |
ab144f5e | 438 | add_nops(insnbuf + used, p->len - used); |
e587cadd | 439 | text_poke_early(p->instr, insnbuf, p->len); |
139ec7c4 | 440 | } |
139ec7c4 | 441 | } |
98de032b | 442 | extern struct paravirt_patch_site __start_parainstructions[], |
139ec7c4 RR |
443 | __stop_parainstructions[]; |
444 | #endif /* CONFIG_PARAVIRT */ | |
445 | ||
9a0b5817 GH |
446 | void __init alternative_instructions(void) |
447 | { | |
8f4e956b AK |
448 | /* The patching is not fully atomic, so try to avoid local interruptions |
449 | that might execute the to be patched code. | |
450 | Other CPUs are not running. */ | |
451 | stop_nmi(); | |
123aa76e AK |
452 | |
453 | /* | |
454 | * Don't stop machine check exceptions while patching. | |
455 | * MCEs only happen when something got corrupted and in this | |
456 | * case we must do something about the corruption. | |
457 | * Ignoring it is worse than a unlikely patching race. | |
458 | * Also machine checks tend to be broadcast and if one CPU | |
459 | * goes into machine check the others follow quickly, so we don't | |
460 | * expect a machine check to cause undue problems during to code | |
461 | * patching. | |
462 | */ | |
8f4e956b | 463 | |
9a0b5817 GH |
464 | apply_alternatives(__alt_instructions, __alt_instructions_end); |
465 | ||
466 | /* switch to patch-once-at-boottime-only mode and free the | |
467 | * tables in case we know the number of CPUs will never ever | |
468 | * change */ | |
469 | #ifdef CONFIG_HOTPLUG_CPU | |
470 | if (num_possible_cpus() < 2) | |
471 | smp_alt_once = 1; | |
9a0b5817 GH |
472 | #endif |
473 | ||
8ec4d41f | 474 | #ifdef CONFIG_SMP |
9a0b5817 GH |
475 | if (smp_alt_once) { |
476 | if (1 == num_possible_cpus()) { | |
477 | printk(KERN_INFO "SMP alternatives: switching to UP code\n"); | |
53756d37 JF |
478 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
479 | set_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
480 | ||
9a0b5817 GH |
481 | alternatives_smp_unlock(__smp_locks, __smp_locks_end, |
482 | _text, _etext); | |
483 | } | |
9a0b5817 | 484 | } else { |
9a0b5817 GH |
485 | alternatives_smp_module_add(NULL, "core kernel", |
486 | __smp_locks, __smp_locks_end, | |
487 | _text, _etext); | |
ca74a6f8 AK |
488 | |
489 | /* Only switch to UP mode if we don't immediately boot others */ | |
649c6653 | 490 | if (num_present_cpus() == 1 || setup_max_cpus <= 1) |
ca74a6f8 | 491 | alternatives_smp_switch(0); |
9a0b5817 | 492 | } |
8ec4d41f | 493 | #endif |
441d40dc | 494 | apply_paravirt(__parainstructions, __parainstructions_end); |
8f4e956b | 495 | |
f68fd5f4 FW |
496 | if (smp_alt_once) |
497 | free_init_pages("SMP alternatives", | |
498 | (unsigned long)__smp_locks, | |
499 | (unsigned long)__smp_locks_end); | |
500 | ||
8f4e956b | 501 | restart_nmi(); |
9a0b5817 | 502 | } |
19d36ccd | 503 | |
e587cadd MD |
504 | /** |
505 | * text_poke_early - Update instructions on a live kernel at boot time | |
506 | * @addr: address to modify | |
507 | * @opcode: source of the copy | |
508 | * @len: length to copy | |
509 | * | |
19d36ccd AK |
510 | * When you use this code to patch more than one byte of an instruction |
511 | * you need to make sure that other CPUs cannot execute this code in parallel. | |
e587cadd MD |
512 | * Also no thread must be currently preempted in the middle of these |
513 | * instructions. And on the local CPU you need to be protected again NMI or MCE | |
514 | * handlers seeing an inconsistent instruction while you patch. | |
19d36ccd | 515 | */ |
8b5a10fc JB |
516 | static void *__init_or_module text_poke_early(void *addr, const void *opcode, |
517 | size_t len) | |
19d36ccd | 518 | { |
e587cadd MD |
519 | unsigned long flags; |
520 | local_irq_save(flags); | |
19d36ccd | 521 | memcpy(addr, opcode, len); |
e587cadd | 522 | sync_core(); |
5367b688 | 523 | local_irq_restore(flags); |
e587cadd MD |
524 | /* Could also do a CLFLUSH here to speed up CPU recovery; but |
525 | that causes hangs on some VIA CPUs. */ | |
526 | return addr; | |
527 | } | |
528 | ||
529 | /** | |
530 | * text_poke - Update instructions on a live kernel | |
531 | * @addr: address to modify | |
532 | * @opcode: source of the copy | |
533 | * @len: length to copy | |
534 | * | |
535 | * Only atomic text poke/set should be allowed when not doing early patching. | |
536 | * It means the size must be writable atomically and the address must be aligned | |
537 | * in a way that permits an atomic write. It also makes sure we fit on a single | |
538 | * page. | |
78ff7fae MH |
539 | * |
540 | * Note: Must be called under text_mutex. | |
e587cadd MD |
541 | */ |
542 | void *__kprobes text_poke(void *addr, const void *opcode, size_t len) | |
543 | { | |
78ff7fae | 544 | unsigned long flags; |
e587cadd | 545 | char *vaddr; |
b7b66baa MD |
546 | struct page *pages[2]; |
547 | int i; | |
e587cadd | 548 | |
b7b66baa MD |
549 | if (!core_kernel_text((unsigned long)addr)) { |
550 | pages[0] = vmalloc_to_page(addr); | |
551 | pages[1] = vmalloc_to_page(addr + PAGE_SIZE); | |
15a601eb | 552 | } else { |
b7b66baa | 553 | pages[0] = virt_to_page(addr); |
00c6b2d5 | 554 | WARN_ON(!PageReserved(pages[0])); |
b7b66baa | 555 | pages[1] = virt_to_page(addr + PAGE_SIZE); |
e587cadd | 556 | } |
b7b66baa | 557 | BUG_ON(!pages[0]); |
7cf49427 | 558 | local_irq_save(flags); |
78ff7fae MH |
559 | set_fixmap(FIX_TEXT_POKE0, page_to_phys(pages[0])); |
560 | if (pages[1]) | |
561 | set_fixmap(FIX_TEXT_POKE1, page_to_phys(pages[1])); | |
562 | vaddr = (char *)fix_to_virt(FIX_TEXT_POKE0); | |
b7b66baa | 563 | memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len); |
78ff7fae MH |
564 | clear_fixmap(FIX_TEXT_POKE0); |
565 | if (pages[1]) | |
566 | clear_fixmap(FIX_TEXT_POKE1); | |
567 | local_flush_tlb(); | |
19d36ccd | 568 | sync_core(); |
a534b679 AK |
569 | /* Could also do a CLFLUSH here to speed up CPU recovery; but |
570 | that causes hangs on some VIA CPUs. */ | |
b7b66baa MD |
571 | for (i = 0; i < len; i++) |
572 | BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]); | |
7cf49427 | 573 | local_irq_restore(flags); |
e587cadd | 574 | return addr; |
19d36ccd | 575 | } |
3d55cc8a MH |
576 | |
577 | /* | |
578 | * Cross-modifying kernel text with stop_machine(). | |
579 | * This code originally comes from immediate value. | |
580 | */ | |
581 | static atomic_t stop_machine_first; | |
582 | static int wrote_text; | |
583 | ||
584 | struct text_poke_params { | |
585 | void *addr; | |
586 | const void *opcode; | |
587 | size_t len; | |
588 | }; | |
589 | ||
590 | static int __kprobes stop_machine_text_poke(void *data) | |
591 | { | |
592 | struct text_poke_params *tpp = data; | |
593 | ||
594 | if (atomic_dec_and_test(&stop_machine_first)) { | |
595 | text_poke(tpp->addr, tpp->opcode, tpp->len); | |
596 | smp_wmb(); /* Make sure other cpus see that this has run */ | |
597 | wrote_text = 1; | |
598 | } else { | |
599 | while (!wrote_text) | |
e5a11016 MH |
600 | cpu_relax(); |
601 | smp_mb(); /* Load wrote_text before following execution */ | |
3d55cc8a MH |
602 | } |
603 | ||
604 | flush_icache_range((unsigned long)tpp->addr, | |
605 | (unsigned long)tpp->addr + tpp->len); | |
606 | return 0; | |
607 | } | |
608 | ||
609 | /** | |
610 | * text_poke_smp - Update instructions on a live kernel on SMP | |
611 | * @addr: address to modify | |
612 | * @opcode: source of the copy | |
613 | * @len: length to copy | |
614 | * | |
615 | * Modify multi-byte instruction by using stop_machine() on SMP. This allows | |
616 | * user to poke/set multi-byte text on SMP. Only non-NMI/MCE code modifying | |
617 | * should be allowed, since stop_machine() does _not_ protect code against | |
618 | * NMI and MCE. | |
619 | * | |
620 | * Note: Must be called under get_online_cpus() and text_mutex. | |
621 | */ | |
622 | void *__kprobes text_poke_smp(void *addr, const void *opcode, size_t len) | |
623 | { | |
624 | struct text_poke_params tpp; | |
625 | ||
626 | tpp.addr = addr; | |
627 | tpp.opcode = opcode; | |
628 | tpp.len = len; | |
629 | atomic_set(&stop_machine_first, 1); | |
630 | wrote_text = 0; | |
631 | stop_machine(stop_machine_text_poke, (void *)&tpp, NULL); | |
632 | return addr; | |
633 | } | |
634 |