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> |
5a0e3ad6 | 11 | #include <linux/slab.h> |
9a0b5817 GH |
12 | #include <asm/alternative.h> |
13 | #include <asm/sections.h> | |
19d36ccd | 14 | #include <asm/pgtable.h> |
8f4e956b AK |
15 | #include <asm/mce.h> |
16 | #include <asm/nmi.h> | |
b097976e | 17 | #include <asm/vsyscall.h> |
e587cadd | 18 | #include <asm/cacheflush.h> |
78ff7fae | 19 | #include <asm/tlbflush.h> |
e587cadd | 20 | #include <asm/io.h> |
78ff7fae | 21 | #include <asm/fixmap.h> |
9a0b5817 | 22 | |
ab144f5e AK |
23 | #define MAX_PATCH_LEN (255-1) |
24 | ||
09488165 JB |
25 | #ifdef CONFIG_HOTPLUG_CPU |
26 | static int smp_alt_once; | |
9a0b5817 | 27 | |
d167a518 GH |
28 | static int __init bootonly(char *str) |
29 | { | |
30 | smp_alt_once = 1; | |
31 | return 1; | |
32 | } | |
b7fb4af0 | 33 | __setup("smp-alt-boot", bootonly); |
09488165 JB |
34 | #else |
35 | #define smp_alt_once 1 | |
36 | #endif | |
37 | ||
8b5a10fc | 38 | static int __initdata_or_module debug_alternative; |
b7fb4af0 | 39 | |
d167a518 GH |
40 | static int __init debug_alt(char *str) |
41 | { | |
42 | debug_alternative = 1; | |
43 | return 1; | |
44 | } | |
d167a518 GH |
45 | __setup("debug-alternative", debug_alt); |
46 | ||
09488165 JB |
47 | static int noreplace_smp; |
48 | ||
b7fb4af0 JF |
49 | static int __init setup_noreplace_smp(char *str) |
50 | { | |
51 | noreplace_smp = 1; | |
52 | return 1; | |
53 | } | |
54 | __setup("noreplace-smp", setup_noreplace_smp); | |
55 | ||
959b4fdf | 56 | #ifdef CONFIG_PARAVIRT |
8b5a10fc | 57 | static int __initdata_or_module noreplace_paravirt = 0; |
959b4fdf JF |
58 | |
59 | static int __init setup_noreplace_paravirt(char *str) | |
60 | { | |
61 | noreplace_paravirt = 1; | |
62 | return 1; | |
63 | } | |
64 | __setup("noreplace-paravirt", setup_noreplace_paravirt); | |
65 | #endif | |
b7fb4af0 | 66 | |
d167a518 GH |
67 | #define DPRINTK(fmt, args...) if (debug_alternative) \ |
68 | printk(KERN_DEBUG fmt, args) | |
69 | ||
8b5a10fc | 70 | #if defined(GENERIC_NOP1) && !defined(CONFIG_X86_64) |
9a0b5817 GH |
71 | /* Use inline assembly to define this because the nops are defined |
72 | as inline assembly strings in the include files and we cannot | |
73 | get them easily into strings. */ | |
8b5a10fc | 74 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nintelnops: " |
9a0b5817 | 75 | GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6 |
f4be31ec SR |
76 | GENERIC_NOP7 GENERIC_NOP8 |
77 | "\t.previous"); | |
121d7bf5 | 78 | extern const unsigned char intelnops[]; |
8b5a10fc JB |
79 | static const unsigned char *const __initconst_or_module |
80 | intel_nops[ASM_NOP_MAX+1] = { | |
9a0b5817 GH |
81 | NULL, |
82 | intelnops, | |
83 | intelnops + 1, | |
84 | intelnops + 1 + 2, | |
85 | intelnops + 1 + 2 + 3, | |
86 | intelnops + 1 + 2 + 3 + 4, | |
87 | intelnops + 1 + 2 + 3 + 4 + 5, | |
88 | intelnops + 1 + 2 + 3 + 4 + 5 + 6, | |
89 | intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
90 | }; | |
d167a518 GH |
91 | #endif |
92 | ||
93 | #ifdef K8_NOP1 | |
8b5a10fc | 94 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nk8nops: " |
d167a518 | 95 | K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6 |
f4be31ec SR |
96 | K8_NOP7 K8_NOP8 |
97 | "\t.previous"); | |
121d7bf5 | 98 | extern const unsigned char k8nops[]; |
8b5a10fc JB |
99 | static const unsigned char *const __initconst_or_module |
100 | k8_nops[ASM_NOP_MAX+1] = { | |
9a0b5817 GH |
101 | NULL, |
102 | k8nops, | |
103 | k8nops + 1, | |
104 | k8nops + 1 + 2, | |
105 | k8nops + 1 + 2 + 3, | |
106 | k8nops + 1 + 2 + 3 + 4, | |
107 | k8nops + 1 + 2 + 3 + 4 + 5, | |
108 | k8nops + 1 + 2 + 3 + 4 + 5 + 6, | |
109 | k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
110 | }; | |
d167a518 GH |
111 | #endif |
112 | ||
8b5a10fc JB |
113 | #if defined(K7_NOP1) && !defined(CONFIG_X86_64) |
114 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nk7nops: " | |
d167a518 | 115 | K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6 |
f4be31ec SR |
116 | K7_NOP7 K7_NOP8 |
117 | "\t.previous"); | |
121d7bf5 | 118 | extern const unsigned char k7nops[]; |
8b5a10fc JB |
119 | static const unsigned char *const __initconst_or_module |
120 | k7_nops[ASM_NOP_MAX+1] = { | |
9a0b5817 GH |
121 | NULL, |
122 | k7nops, | |
123 | k7nops + 1, | |
124 | k7nops + 1 + 2, | |
125 | k7nops + 1 + 2 + 3, | |
126 | k7nops + 1 + 2 + 3 + 4, | |
127 | k7nops + 1 + 2 + 3 + 4 + 5, | |
128 | k7nops + 1 + 2 + 3 + 4 + 5 + 6, | |
129 | k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
130 | }; | |
d167a518 GH |
131 | #endif |
132 | ||
32c464f5 | 133 | #ifdef P6_NOP1 |
8b5a10fc | 134 | asm("\t" __stringify(__INITRODATA_OR_MODULE) "\np6nops: " |
32c464f5 | 135 | P6_NOP1 P6_NOP2 P6_NOP3 P6_NOP4 P6_NOP5 P6_NOP6 |
f4be31ec SR |
136 | P6_NOP7 P6_NOP8 |
137 | "\t.previous"); | |
32c464f5 | 138 | extern const unsigned char p6nops[]; |
8b5a10fc JB |
139 | static const unsigned char *const __initconst_or_module |
140 | p6_nops[ASM_NOP_MAX+1] = { | |
32c464f5 JB |
141 | NULL, |
142 | p6nops, | |
143 | p6nops + 1, | |
144 | p6nops + 1 + 2, | |
145 | p6nops + 1 + 2 + 3, | |
146 | p6nops + 1 + 2 + 3 + 4, | |
147 | p6nops + 1 + 2 + 3 + 4 + 5, | |
148 | p6nops + 1 + 2 + 3 + 4 + 5 + 6, | |
149 | p6nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
150 | }; | |
151 | #endif | |
152 | ||
d167a518 GH |
153 | #ifdef CONFIG_X86_64 |
154 | ||
155 | extern char __vsyscall_0; | |
8b5a10fc | 156 | static const unsigned char *const *__init_or_module find_nop_table(void) |
d167a518 | 157 | { |
f31d731e PA |
158 | if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && |
159 | boot_cpu_has(X86_FEATURE_NOPL)) | |
160 | return p6_nops; | |
161 | else | |
162 | return k8_nops; | |
d167a518 GH |
163 | } |
164 | ||
165 | #else /* CONFIG_X86_64 */ | |
166 | ||
8b5a10fc | 167 | static const unsigned char *const *__init_or_module find_nop_table(void) |
9a0b5817 | 168 | { |
f31d731e PA |
169 | if (boot_cpu_has(X86_FEATURE_K8)) |
170 | return k8_nops; | |
171 | else if (boot_cpu_has(X86_FEATURE_K7)) | |
172 | return k7_nops; | |
173 | else if (boot_cpu_has(X86_FEATURE_NOPL)) | |
174 | return p6_nops; | |
175 | else | |
176 | return intel_nops; | |
9a0b5817 GH |
177 | } |
178 | ||
d167a518 GH |
179 | #endif /* CONFIG_X86_64 */ |
180 | ||
ab144f5e | 181 | /* Use this to add nops to a buffer, then text_poke the whole buffer. */ |
8b5a10fc | 182 | static void __init_or_module add_nops(void *insns, unsigned int len) |
139ec7c4 | 183 | { |
121d7bf5 | 184 | const unsigned char *const *noptable = find_nop_table(); |
139ec7c4 RR |
185 | |
186 | while (len > 0) { | |
187 | unsigned int noplen = len; | |
188 | if (noplen > ASM_NOP_MAX) | |
189 | noplen = ASM_NOP_MAX; | |
ab144f5e | 190 | memcpy(insns, noptable[noplen], noplen); |
139ec7c4 RR |
191 | insns += noplen; |
192 | len -= noplen; | |
193 | } | |
194 | } | |
195 | ||
d167a518 | 196 | extern struct alt_instr __alt_instructions[], __alt_instructions_end[]; |
5967ed87 | 197 | extern s32 __smp_locks[], __smp_locks_end[]; |
8b5a10fc | 198 | static void *text_poke_early(void *addr, const void *opcode, size_t len); |
d167a518 | 199 | |
9a0b5817 GH |
200 | /* Replace instructions with better alternatives for this CPU type. |
201 | This runs before SMP is initialized to avoid SMP problems with | |
202 | self modifying code. This implies that assymetric systems where | |
203 | APs have less capabilities than the boot processor are not handled. | |
204 | Tough. Make sure you disable such features by hand. */ | |
205 | ||
8b5a10fc JB |
206 | void __init_or_module apply_alternatives(struct alt_instr *start, |
207 | struct alt_instr *end) | |
9a0b5817 | 208 | { |
9a0b5817 | 209 | struct alt_instr *a; |
1b1d9258 | 210 | u8 insnbuf[MAX_PATCH_LEN]; |
9a0b5817 | 211 | |
77bf90ed | 212 | DPRINTK("%s: alt table %p -> %p\n", __func__, start, end); |
9a0b5817 | 213 | for (a = start; a < end; a++) { |
ab144f5e | 214 | u8 *instr = a->instr; |
9a0b5817 | 215 | BUG_ON(a->replacementlen > a->instrlen); |
ab144f5e | 216 | BUG_ON(a->instrlen > sizeof(insnbuf)); |
9a0b5817 GH |
217 | if (!boot_cpu_has(a->cpuid)) |
218 | continue; | |
d167a518 GH |
219 | #ifdef CONFIG_X86_64 |
220 | /* vsyscall code is not mapped yet. resolve it manually. */ | |
221 | if (instr >= (u8 *)VSYSCALL_START && instr < (u8*)VSYSCALL_END) { | |
222 | instr = __va(instr - (u8*)VSYSCALL_START + (u8*)__pa_symbol(&__vsyscall_0)); | |
223 | DPRINTK("%s: vsyscall fixup: %p => %p\n", | |
77bf90ed | 224 | __func__, a->instr, instr); |
d167a518 GH |
225 | } |
226 | #endif | |
ab144f5e | 227 | memcpy(insnbuf, a->replacement, a->replacementlen); |
1b1d9258 JB |
228 | if (*insnbuf == 0xe8 && a->replacementlen == 5) |
229 | *(s32 *)(insnbuf + 1) += a->replacement - a->instr; | |
ab144f5e AK |
230 | add_nops(insnbuf + a->replacementlen, |
231 | a->instrlen - a->replacementlen); | |
e587cadd | 232 | text_poke_early(instr, insnbuf, a->instrlen); |
9a0b5817 GH |
233 | } |
234 | } | |
235 | ||
8ec4d41f GH |
236 | #ifdef CONFIG_SMP |
237 | ||
5967ed87 JB |
238 | static void alternatives_smp_lock(const s32 *start, const s32 *end, |
239 | u8 *text, u8 *text_end) | |
9a0b5817 | 240 | { |
5967ed87 | 241 | const s32 *poff; |
9a0b5817 | 242 | |
3945dab4 | 243 | mutex_lock(&text_mutex); |
5967ed87 JB |
244 | for (poff = start; poff < end; poff++) { |
245 | u8 *ptr = (u8 *)poff + *poff; | |
246 | ||
247 | if (!*poff || ptr < text || ptr >= text_end) | |
9a0b5817 | 248 | continue; |
f88f07e0 | 249 | /* turn DS segment override prefix into lock prefix */ |
d9c5841e PA |
250 | if (*ptr == 0x3e) |
251 | text_poke(ptr, ((unsigned char []){0xf0}), 1); | |
9a0b5817 | 252 | }; |
3945dab4 | 253 | mutex_unlock(&text_mutex); |
9a0b5817 GH |
254 | } |
255 | ||
5967ed87 JB |
256 | static void alternatives_smp_unlock(const s32 *start, const s32 *end, |
257 | u8 *text, u8 *text_end) | |
9a0b5817 | 258 | { |
5967ed87 | 259 | const s32 *poff; |
9a0b5817 | 260 | |
b7fb4af0 JF |
261 | if (noreplace_smp) |
262 | return; | |
263 | ||
3945dab4 | 264 | mutex_lock(&text_mutex); |
5967ed87 JB |
265 | for (poff = start; poff < end; poff++) { |
266 | u8 *ptr = (u8 *)poff + *poff; | |
267 | ||
268 | if (!*poff || ptr < text || ptr >= text_end) | |
9a0b5817 | 269 | continue; |
f88f07e0 | 270 | /* turn lock prefix into DS segment override prefix */ |
d9c5841e PA |
271 | if (*ptr == 0xf0) |
272 | text_poke(ptr, ((unsigned char []){0x3E}), 1); | |
9a0b5817 | 273 | }; |
3945dab4 | 274 | mutex_unlock(&text_mutex); |
9a0b5817 GH |
275 | } |
276 | ||
277 | struct smp_alt_module { | |
278 | /* what is this ??? */ | |
279 | struct module *mod; | |
280 | char *name; | |
281 | ||
282 | /* ptrs to lock prefixes */ | |
5967ed87 JB |
283 | const s32 *locks; |
284 | const s32 *locks_end; | |
9a0b5817 GH |
285 | |
286 | /* .text segment, needed to avoid patching init code ;) */ | |
287 | u8 *text; | |
288 | u8 *text_end; | |
289 | ||
290 | struct list_head next; | |
291 | }; | |
292 | static LIST_HEAD(smp_alt_modules); | |
2f1dafe5 | 293 | static DEFINE_MUTEX(smp_alt); |
ca74a6f8 | 294 | static int smp_mode = 1; /* protected by smp_alt */ |
9a0b5817 | 295 | |
8b5a10fc JB |
296 | void __init_or_module alternatives_smp_module_add(struct module *mod, |
297 | char *name, | |
298 | void *locks, void *locks_end, | |
299 | void *text, void *text_end) | |
9a0b5817 GH |
300 | { |
301 | struct smp_alt_module *smp; | |
9a0b5817 | 302 | |
b7fb4af0 JF |
303 | if (noreplace_smp) |
304 | return; | |
305 | ||
9a0b5817 GH |
306 | if (smp_alt_once) { |
307 | if (boot_cpu_has(X86_FEATURE_UP)) | |
308 | alternatives_smp_unlock(locks, locks_end, | |
309 | text, text_end); | |
310 | return; | |
311 | } | |
312 | ||
313 | smp = kzalloc(sizeof(*smp), GFP_KERNEL); | |
314 | if (NULL == smp) | |
315 | return; /* we'll run the (safe but slow) SMP code then ... */ | |
316 | ||
317 | smp->mod = mod; | |
318 | smp->name = name; | |
319 | smp->locks = locks; | |
320 | smp->locks_end = locks_end; | |
321 | smp->text = text; | |
322 | smp->text_end = text_end; | |
323 | DPRINTK("%s: locks %p -> %p, text %p -> %p, name %s\n", | |
77bf90ed | 324 | __func__, smp->locks, smp->locks_end, |
9a0b5817 GH |
325 | smp->text, smp->text_end, smp->name); |
326 | ||
2f1dafe5 | 327 | mutex_lock(&smp_alt); |
9a0b5817 GH |
328 | list_add_tail(&smp->next, &smp_alt_modules); |
329 | if (boot_cpu_has(X86_FEATURE_UP)) | |
330 | alternatives_smp_unlock(smp->locks, smp->locks_end, | |
331 | smp->text, smp->text_end); | |
2f1dafe5 | 332 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
333 | } |
334 | ||
8b5a10fc | 335 | void __init_or_module alternatives_smp_module_del(struct module *mod) |
9a0b5817 GH |
336 | { |
337 | struct smp_alt_module *item; | |
9a0b5817 | 338 | |
b7fb4af0 | 339 | if (smp_alt_once || noreplace_smp) |
9a0b5817 GH |
340 | return; |
341 | ||
2f1dafe5 | 342 | mutex_lock(&smp_alt); |
9a0b5817 GH |
343 | list_for_each_entry(item, &smp_alt_modules, next) { |
344 | if (mod != item->mod) | |
345 | continue; | |
346 | list_del(&item->next); | |
2f1dafe5 | 347 | mutex_unlock(&smp_alt); |
77bf90ed | 348 | DPRINTK("%s: %s\n", __func__, item->name); |
9a0b5817 GH |
349 | kfree(item); |
350 | return; | |
351 | } | |
2f1dafe5 | 352 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
353 | } |
354 | ||
355 | void alternatives_smp_switch(int smp) | |
356 | { | |
357 | struct smp_alt_module *mod; | |
9a0b5817 | 358 | |
3047e99e IM |
359 | #ifdef CONFIG_LOCKDEP |
360 | /* | |
17abecfe IM |
361 | * Older binutils section handling bug prevented |
362 | * alternatives-replacement from working reliably. | |
363 | * | |
364 | * If this still occurs then you should see a hang | |
365 | * or crash shortly after this line: | |
3047e99e | 366 | */ |
17abecfe | 367 | printk("lockdep: fixing up alternatives.\n"); |
3047e99e IM |
368 | #endif |
369 | ||
b7fb4af0 | 370 | if (noreplace_smp || smp_alt_once) |
9a0b5817 GH |
371 | return; |
372 | BUG_ON(!smp && (num_online_cpus() > 1)); | |
373 | ||
2f1dafe5 | 374 | mutex_lock(&smp_alt); |
ca74a6f8 AK |
375 | |
376 | /* | |
377 | * Avoid unnecessary switches because it forces JIT based VMs to | |
378 | * throw away all cached translations, which can be quite costly. | |
379 | */ | |
380 | if (smp == smp_mode) { | |
381 | /* nothing */ | |
382 | } else if (smp) { | |
9a0b5817 | 383 | printk(KERN_INFO "SMP alternatives: switching to SMP code\n"); |
53756d37 JF |
384 | clear_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
385 | clear_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
9a0b5817 GH |
386 | list_for_each_entry(mod, &smp_alt_modules, next) |
387 | alternatives_smp_lock(mod->locks, mod->locks_end, | |
388 | mod->text, mod->text_end); | |
389 | } else { | |
390 | printk(KERN_INFO "SMP alternatives: switching to UP code\n"); | |
53756d37 JF |
391 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
392 | set_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
9a0b5817 GH |
393 | list_for_each_entry(mod, &smp_alt_modules, next) |
394 | alternatives_smp_unlock(mod->locks, mod->locks_end, | |
395 | mod->text, mod->text_end); | |
396 | } | |
ca74a6f8 | 397 | smp_mode = smp; |
2f1dafe5 | 398 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
399 | } |
400 | ||
2cfa1978 MH |
401 | /* Return 1 if the address range is reserved for smp-alternatives */ |
402 | int alternatives_text_reserved(void *start, void *end) | |
403 | { | |
404 | struct smp_alt_module *mod; | |
5967ed87 | 405 | const s32 *poff; |
076dc4a6 MH |
406 | u8 *text_start = start; |
407 | u8 *text_end = end; | |
2cfa1978 MH |
408 | |
409 | list_for_each_entry(mod, &smp_alt_modules, next) { | |
076dc4a6 | 410 | if (mod->text > text_end || mod->text_end < text_start) |
2cfa1978 | 411 | continue; |
5967ed87 JB |
412 | for (poff = mod->locks; poff < mod->locks_end; poff++) { |
413 | const u8 *ptr = (const u8 *)poff + *poff; | |
414 | ||
415 | if (text_start <= ptr && text_end > ptr) | |
2cfa1978 | 416 | return 1; |
5967ed87 | 417 | } |
2cfa1978 MH |
418 | } |
419 | ||
420 | return 0; | |
421 | } | |
8ec4d41f GH |
422 | #endif |
423 | ||
139ec7c4 | 424 | #ifdef CONFIG_PARAVIRT |
8b5a10fc JB |
425 | void __init_or_module apply_paravirt(struct paravirt_patch_site *start, |
426 | struct paravirt_patch_site *end) | |
139ec7c4 | 427 | { |
98de032b | 428 | struct paravirt_patch_site *p; |
ab144f5e | 429 | char insnbuf[MAX_PATCH_LEN]; |
139ec7c4 | 430 | |
959b4fdf JF |
431 | if (noreplace_paravirt) |
432 | return; | |
433 | ||
139ec7c4 RR |
434 | for (p = start; p < end; p++) { |
435 | unsigned int used; | |
436 | ||
ab144f5e | 437 | BUG_ON(p->len > MAX_PATCH_LEN); |
d34fda4a CW |
438 | /* prep the buffer with the original instructions */ |
439 | memcpy(insnbuf, p->instr, p->len); | |
93b1eab3 JF |
440 | used = pv_init_ops.patch(p->instrtype, p->clobbers, insnbuf, |
441 | (unsigned long)p->instr, p->len); | |
7f63c41c | 442 | |
63f70270 JF |
443 | BUG_ON(used > p->len); |
444 | ||
139ec7c4 | 445 | /* Pad the rest with nops */ |
ab144f5e | 446 | add_nops(insnbuf + used, p->len - used); |
e587cadd | 447 | text_poke_early(p->instr, insnbuf, p->len); |
139ec7c4 | 448 | } |
139ec7c4 | 449 | } |
98de032b | 450 | extern struct paravirt_patch_site __start_parainstructions[], |
139ec7c4 RR |
451 | __stop_parainstructions[]; |
452 | #endif /* CONFIG_PARAVIRT */ | |
453 | ||
9a0b5817 GH |
454 | void __init alternative_instructions(void) |
455 | { | |
8f4e956b AK |
456 | /* The patching is not fully atomic, so try to avoid local interruptions |
457 | that might execute the to be patched code. | |
458 | Other CPUs are not running. */ | |
459 | stop_nmi(); | |
123aa76e AK |
460 | |
461 | /* | |
462 | * Don't stop machine check exceptions while patching. | |
463 | * MCEs only happen when something got corrupted and in this | |
464 | * case we must do something about the corruption. | |
465 | * Ignoring it is worse than a unlikely patching race. | |
466 | * Also machine checks tend to be broadcast and if one CPU | |
467 | * goes into machine check the others follow quickly, so we don't | |
468 | * expect a machine check to cause undue problems during to code | |
469 | * patching. | |
470 | */ | |
8f4e956b | 471 | |
9a0b5817 GH |
472 | apply_alternatives(__alt_instructions, __alt_instructions_end); |
473 | ||
474 | /* switch to patch-once-at-boottime-only mode and free the | |
475 | * tables in case we know the number of CPUs will never ever | |
476 | * change */ | |
477 | #ifdef CONFIG_HOTPLUG_CPU | |
478 | if (num_possible_cpus() < 2) | |
479 | smp_alt_once = 1; | |
9a0b5817 GH |
480 | #endif |
481 | ||
8ec4d41f | 482 | #ifdef CONFIG_SMP |
9a0b5817 GH |
483 | if (smp_alt_once) { |
484 | if (1 == num_possible_cpus()) { | |
485 | printk(KERN_INFO "SMP alternatives: switching to UP code\n"); | |
53756d37 JF |
486 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
487 | set_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
488 | ||
9a0b5817 GH |
489 | alternatives_smp_unlock(__smp_locks, __smp_locks_end, |
490 | _text, _etext); | |
491 | } | |
9a0b5817 | 492 | } else { |
9a0b5817 GH |
493 | alternatives_smp_module_add(NULL, "core kernel", |
494 | __smp_locks, __smp_locks_end, | |
495 | _text, _etext); | |
ca74a6f8 AK |
496 | |
497 | /* Only switch to UP mode if we don't immediately boot others */ | |
649c6653 | 498 | if (num_present_cpus() == 1 || setup_max_cpus <= 1) |
ca74a6f8 | 499 | alternatives_smp_switch(0); |
9a0b5817 | 500 | } |
8ec4d41f | 501 | #endif |
441d40dc | 502 | apply_paravirt(__parainstructions, __parainstructions_end); |
8f4e956b | 503 | |
f68fd5f4 FW |
504 | if (smp_alt_once) |
505 | free_init_pages("SMP alternatives", | |
506 | (unsigned long)__smp_locks, | |
507 | (unsigned long)__smp_locks_end); | |
508 | ||
8f4e956b | 509 | restart_nmi(); |
9a0b5817 | 510 | } |
19d36ccd | 511 | |
e587cadd MD |
512 | /** |
513 | * text_poke_early - Update instructions on a live kernel at boot time | |
514 | * @addr: address to modify | |
515 | * @opcode: source of the copy | |
516 | * @len: length to copy | |
517 | * | |
19d36ccd AK |
518 | * When you use this code to patch more than one byte of an instruction |
519 | * you need to make sure that other CPUs cannot execute this code in parallel. | |
e587cadd MD |
520 | * Also no thread must be currently preempted in the middle of these |
521 | * instructions. And on the local CPU you need to be protected again NMI or MCE | |
522 | * handlers seeing an inconsistent instruction while you patch. | |
19d36ccd | 523 | */ |
8b5a10fc JB |
524 | static void *__init_or_module text_poke_early(void *addr, const void *opcode, |
525 | size_t len) | |
19d36ccd | 526 | { |
e587cadd MD |
527 | unsigned long flags; |
528 | local_irq_save(flags); | |
19d36ccd | 529 | memcpy(addr, opcode, len); |
e587cadd | 530 | sync_core(); |
5367b688 | 531 | local_irq_restore(flags); |
e587cadd MD |
532 | /* Could also do a CLFLUSH here to speed up CPU recovery; but |
533 | that causes hangs on some VIA CPUs. */ | |
534 | return addr; | |
535 | } | |
536 | ||
537 | /** | |
538 | * text_poke - Update instructions on a live kernel | |
539 | * @addr: address to modify | |
540 | * @opcode: source of the copy | |
541 | * @len: length to copy | |
542 | * | |
543 | * Only atomic text poke/set should be allowed when not doing early patching. | |
544 | * It means the size must be writable atomically and the address must be aligned | |
545 | * in a way that permits an atomic write. It also makes sure we fit on a single | |
546 | * page. | |
78ff7fae MH |
547 | * |
548 | * Note: Must be called under text_mutex. | |
e587cadd MD |
549 | */ |
550 | void *__kprobes text_poke(void *addr, const void *opcode, size_t len) | |
551 | { | |
78ff7fae | 552 | unsigned long flags; |
e587cadd | 553 | char *vaddr; |
b7b66baa MD |
554 | struct page *pages[2]; |
555 | int i; | |
e587cadd | 556 | |
b7b66baa MD |
557 | if (!core_kernel_text((unsigned long)addr)) { |
558 | pages[0] = vmalloc_to_page(addr); | |
559 | pages[1] = vmalloc_to_page(addr + PAGE_SIZE); | |
15a601eb | 560 | } else { |
b7b66baa | 561 | pages[0] = virt_to_page(addr); |
00c6b2d5 | 562 | WARN_ON(!PageReserved(pages[0])); |
b7b66baa | 563 | pages[1] = virt_to_page(addr + PAGE_SIZE); |
e587cadd | 564 | } |
b7b66baa | 565 | BUG_ON(!pages[0]); |
7cf49427 | 566 | local_irq_save(flags); |
78ff7fae MH |
567 | set_fixmap(FIX_TEXT_POKE0, page_to_phys(pages[0])); |
568 | if (pages[1]) | |
569 | set_fixmap(FIX_TEXT_POKE1, page_to_phys(pages[1])); | |
570 | vaddr = (char *)fix_to_virt(FIX_TEXT_POKE0); | |
b7b66baa | 571 | memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len); |
78ff7fae MH |
572 | clear_fixmap(FIX_TEXT_POKE0); |
573 | if (pages[1]) | |
574 | clear_fixmap(FIX_TEXT_POKE1); | |
575 | local_flush_tlb(); | |
19d36ccd | 576 | sync_core(); |
a534b679 AK |
577 | /* Could also do a CLFLUSH here to speed up CPU recovery; but |
578 | that causes hangs on some VIA CPUs. */ | |
b7b66baa MD |
579 | for (i = 0; i < len; i++) |
580 | BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]); | |
7cf49427 | 581 | local_irq_restore(flags); |
e587cadd | 582 | return addr; |
19d36ccd | 583 | } |
3d55cc8a MH |
584 | |
585 | /* | |
586 | * Cross-modifying kernel text with stop_machine(). | |
587 | * This code originally comes from immediate value. | |
588 | */ | |
589 | static atomic_t stop_machine_first; | |
590 | static int wrote_text; | |
591 | ||
592 | struct text_poke_params { | |
593 | void *addr; | |
594 | const void *opcode; | |
595 | size_t len; | |
596 | }; | |
597 | ||
598 | static int __kprobes stop_machine_text_poke(void *data) | |
599 | { | |
600 | struct text_poke_params *tpp = data; | |
601 | ||
602 | if (atomic_dec_and_test(&stop_machine_first)) { | |
603 | text_poke(tpp->addr, tpp->opcode, tpp->len); | |
604 | smp_wmb(); /* Make sure other cpus see that this has run */ | |
605 | wrote_text = 1; | |
606 | } else { | |
607 | while (!wrote_text) | |
e5a11016 MH |
608 | cpu_relax(); |
609 | smp_mb(); /* Load wrote_text before following execution */ | |
3d55cc8a MH |
610 | } |
611 | ||
612 | flush_icache_range((unsigned long)tpp->addr, | |
613 | (unsigned long)tpp->addr + tpp->len); | |
614 | return 0; | |
615 | } | |
616 | ||
617 | /** | |
618 | * text_poke_smp - Update instructions on a live kernel on SMP | |
619 | * @addr: address to modify | |
620 | * @opcode: source of the copy | |
621 | * @len: length to copy | |
622 | * | |
623 | * Modify multi-byte instruction by using stop_machine() on SMP. This allows | |
624 | * user to poke/set multi-byte text on SMP. Only non-NMI/MCE code modifying | |
625 | * should be allowed, since stop_machine() does _not_ protect code against | |
626 | * NMI and MCE. | |
627 | * | |
628 | * Note: Must be called under get_online_cpus() and text_mutex. | |
629 | */ | |
630 | void *__kprobes text_poke_smp(void *addr, const void *opcode, size_t len) | |
631 | { | |
632 | struct text_poke_params tpp; | |
633 | ||
634 | tpp.addr = addr; | |
635 | tpp.opcode = opcode; | |
636 | tpp.len = len; | |
637 | atomic_set(&stop_machine_first, 1); | |
638 | wrote_text = 0; | |
639 | stop_machine(stop_machine_text_poke, (void *)&tpp, NULL); | |
640 | return addr; | |
641 | } | |
642 |