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> | |
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 | ||
dc326fca PA |
69 | /* |
70 | * Each GENERIC_NOPX is of X bytes, and defined as an array of bytes | |
71 | * that correspond to that nop. Getting from one nop to the next, we | |
72 | * add to the array the offset that is equal to the sum of all sizes of | |
73 | * nops preceding the one we are after. | |
74 | * | |
75 | * Note: The GENERIC_NOP5_ATOMIC is at the end, as it breaks the | |
76 | * nice symmetry of sizes of the previous nops. | |
77 | */ | |
8b5a10fc | 78 | #if defined(GENERIC_NOP1) && !defined(CONFIG_X86_64) |
dc326fca PA |
79 | static const unsigned char intelnops[] = |
80 | { | |
81 | GENERIC_NOP1, | |
82 | GENERIC_NOP2, | |
83 | GENERIC_NOP3, | |
84 | GENERIC_NOP4, | |
85 | GENERIC_NOP5, | |
86 | GENERIC_NOP6, | |
87 | GENERIC_NOP7, | |
88 | GENERIC_NOP8, | |
89 | GENERIC_NOP5_ATOMIC | |
90 | }; | |
91 | static const unsigned char * const intel_nops[ASM_NOP_MAX+2] = | |
92 | { | |
9a0b5817 GH |
93 | NULL, |
94 | intelnops, | |
95 | intelnops + 1, | |
96 | intelnops + 1 + 2, | |
97 | intelnops + 1 + 2 + 3, | |
98 | intelnops + 1 + 2 + 3 + 4, | |
99 | intelnops + 1 + 2 + 3 + 4 + 5, | |
100 | intelnops + 1 + 2 + 3 + 4 + 5 + 6, | |
101 | intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
dc326fca | 102 | intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8, |
9a0b5817 | 103 | }; |
d167a518 GH |
104 | #endif |
105 | ||
106 | #ifdef K8_NOP1 | |
dc326fca PA |
107 | static const unsigned char k8nops[] = |
108 | { | |
109 | K8_NOP1, | |
110 | K8_NOP2, | |
111 | K8_NOP3, | |
112 | K8_NOP4, | |
113 | K8_NOP5, | |
114 | K8_NOP6, | |
115 | K8_NOP7, | |
116 | K8_NOP8, | |
117 | K8_NOP5_ATOMIC | |
118 | }; | |
119 | static const unsigned char * const k8_nops[ASM_NOP_MAX+2] = | |
120 | { | |
9a0b5817 GH |
121 | NULL, |
122 | k8nops, | |
123 | k8nops + 1, | |
124 | k8nops + 1 + 2, | |
125 | k8nops + 1 + 2 + 3, | |
126 | k8nops + 1 + 2 + 3 + 4, | |
127 | k8nops + 1 + 2 + 3 + 4 + 5, | |
128 | k8nops + 1 + 2 + 3 + 4 + 5 + 6, | |
129 | k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
dc326fca | 130 | k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8, |
9a0b5817 | 131 | }; |
d167a518 GH |
132 | #endif |
133 | ||
8b5a10fc | 134 | #if defined(K7_NOP1) && !defined(CONFIG_X86_64) |
dc326fca PA |
135 | static const unsigned char k7nops[] = |
136 | { | |
137 | K7_NOP1, | |
138 | K7_NOP2, | |
139 | K7_NOP3, | |
140 | K7_NOP4, | |
141 | K7_NOP5, | |
142 | K7_NOP6, | |
143 | K7_NOP7, | |
144 | K7_NOP8, | |
145 | K7_NOP5_ATOMIC | |
146 | }; | |
147 | static const unsigned char * const k7_nops[ASM_NOP_MAX+2] = | |
148 | { | |
9a0b5817 GH |
149 | NULL, |
150 | k7nops, | |
151 | k7nops + 1, | |
152 | k7nops + 1 + 2, | |
153 | k7nops + 1 + 2 + 3, | |
154 | k7nops + 1 + 2 + 3 + 4, | |
155 | k7nops + 1 + 2 + 3 + 4 + 5, | |
156 | k7nops + 1 + 2 + 3 + 4 + 5 + 6, | |
157 | k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
dc326fca | 158 | k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8, |
9a0b5817 | 159 | }; |
d167a518 GH |
160 | #endif |
161 | ||
32c464f5 | 162 | #ifdef P6_NOP1 |
dc326fca PA |
163 | static const unsigned char __initconst_or_module p6nops[] = |
164 | { | |
165 | P6_NOP1, | |
166 | P6_NOP2, | |
167 | P6_NOP3, | |
168 | P6_NOP4, | |
169 | P6_NOP5, | |
170 | P6_NOP6, | |
171 | P6_NOP7, | |
172 | P6_NOP8, | |
173 | P6_NOP5_ATOMIC | |
174 | }; | |
175 | static const unsigned char * const p6_nops[ASM_NOP_MAX+2] = | |
176 | { | |
32c464f5 JB |
177 | NULL, |
178 | p6nops, | |
179 | p6nops + 1, | |
180 | p6nops + 1 + 2, | |
181 | p6nops + 1 + 2 + 3, | |
182 | p6nops + 1 + 2 + 3 + 4, | |
183 | p6nops + 1 + 2 + 3 + 4 + 5, | |
184 | p6nops + 1 + 2 + 3 + 4 + 5 + 6, | |
185 | p6nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
dc326fca | 186 | p6nops + 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8, |
32c464f5 JB |
187 | }; |
188 | #endif | |
189 | ||
dc326fca | 190 | /* Initialize these to a safe default */ |
d167a518 | 191 | #ifdef CONFIG_X86_64 |
dc326fca PA |
192 | const unsigned char * const *ideal_nops = p6_nops; |
193 | #else | |
194 | const unsigned char * const *ideal_nops = intel_nops; | |
195 | #endif | |
d167a518 | 196 | |
dc326fca | 197 | void __init arch_init_ideal_nops(void) |
d167a518 | 198 | { |
dc326fca PA |
199 | switch (boot_cpu_data.x86_vendor) { |
200 | case X86_VENDOR_INTEL: | |
d8d9766c PA |
201 | /* |
202 | * Due to a decoder implementation quirk, some | |
203 | * specific Intel CPUs actually perform better with | |
204 | * the "k8_nops" than with the SDM-recommended NOPs. | |
205 | */ | |
206 | if (boot_cpu_data.x86 == 6 && | |
207 | boot_cpu_data.x86_model >= 0x0f && | |
208 | boot_cpu_data.x86_model != 0x1c && | |
209 | boot_cpu_data.x86_model != 0x26 && | |
210 | boot_cpu_data.x86_model != 0x27 && | |
211 | boot_cpu_data.x86_model < 0x30) { | |
212 | ideal_nops = k8_nops; | |
213 | } else if (boot_cpu_has(X86_FEATURE_NOPL)) { | |
dc326fca PA |
214 | ideal_nops = p6_nops; |
215 | } else { | |
216 | #ifdef CONFIG_X86_64 | |
217 | ideal_nops = k8_nops; | |
218 | #else | |
219 | ideal_nops = intel_nops; | |
220 | #endif | |
221 | } | |
d167a518 | 222 | |
dc326fca PA |
223 | default: |
224 | #ifdef CONFIG_X86_64 | |
225 | ideal_nops = k8_nops; | |
226 | #else | |
227 | if (boot_cpu_has(X86_FEATURE_K8)) | |
228 | ideal_nops = k8_nops; | |
229 | else if (boot_cpu_has(X86_FEATURE_K7)) | |
230 | ideal_nops = k7_nops; | |
231 | else | |
232 | ideal_nops = intel_nops; | |
233 | #endif | |
234 | } | |
9a0b5817 GH |
235 | } |
236 | ||
ab144f5e | 237 | /* Use this to add nops to a buffer, then text_poke the whole buffer. */ |
8b5a10fc | 238 | static void __init_or_module add_nops(void *insns, unsigned int len) |
139ec7c4 | 239 | { |
139ec7c4 RR |
240 | while (len > 0) { |
241 | unsigned int noplen = len; | |
242 | if (noplen > ASM_NOP_MAX) | |
243 | noplen = ASM_NOP_MAX; | |
dc326fca | 244 | memcpy(insns, ideal_nops[noplen], noplen); |
139ec7c4 RR |
245 | insns += noplen; |
246 | len -= noplen; | |
247 | } | |
248 | } | |
249 | ||
d167a518 | 250 | extern struct alt_instr __alt_instructions[], __alt_instructions_end[]; |
5967ed87 | 251 | extern s32 __smp_locks[], __smp_locks_end[]; |
fa6f2cc7 | 252 | void *text_poke_early(void *addr, const void *opcode, size_t len); |
d167a518 | 253 | |
9a0b5817 GH |
254 | /* Replace instructions with better alternatives for this CPU type. |
255 | This runs before SMP is initialized to avoid SMP problems with | |
0d2eb44f | 256 | self modifying code. This implies that asymmetric systems where |
9a0b5817 GH |
257 | APs have less capabilities than the boot processor are not handled. |
258 | Tough. Make sure you disable such features by hand. */ | |
259 | ||
8b5a10fc JB |
260 | void __init_or_module apply_alternatives(struct alt_instr *start, |
261 | struct alt_instr *end) | |
9a0b5817 | 262 | { |
9a0b5817 | 263 | struct alt_instr *a; |
59e97e4d | 264 | u8 *instr, *replacement; |
1b1d9258 | 265 | u8 insnbuf[MAX_PATCH_LEN]; |
9a0b5817 | 266 | |
77bf90ed | 267 | DPRINTK("%s: alt table %p -> %p\n", __func__, start, end); |
50973133 FY |
268 | /* |
269 | * The scan order should be from start to end. A later scanned | |
270 | * alternative code can overwrite a previous scanned alternative code. | |
271 | * Some kernel functions (e.g. memcpy, memset, etc) use this order to | |
272 | * patch code. | |
273 | * | |
274 | * So be careful if you want to change the scan order to any other | |
275 | * order. | |
276 | */ | |
9a0b5817 | 277 | for (a = start; a < end; a++) { |
59e97e4d AL |
278 | instr = (u8 *)&a->instr_offset + a->instr_offset; |
279 | replacement = (u8 *)&a->repl_offset + a->repl_offset; | |
9a0b5817 | 280 | BUG_ON(a->replacementlen > a->instrlen); |
ab144f5e | 281 | BUG_ON(a->instrlen > sizeof(insnbuf)); |
3b770a21 | 282 | BUG_ON(a->cpuid >= NCAPINTS*32); |
9a0b5817 GH |
283 | if (!boot_cpu_has(a->cpuid)) |
284 | continue; | |
59e97e4d AL |
285 | |
286 | memcpy(insnbuf, replacement, a->replacementlen); | |
287 | ||
288 | /* 0xe8 is a relative jump; fix the offset. */ | |
289 | if (*insnbuf == 0xe8 && a->replacementlen == 5) | |
290 | *(s32 *)(insnbuf + 1) += replacement - instr; | |
291 | ||
292 | add_nops(insnbuf + a->replacementlen, | |
293 | a->instrlen - a->replacementlen); | |
294 | ||
e587cadd | 295 | text_poke_early(instr, insnbuf, a->instrlen); |
9a0b5817 GH |
296 | } |
297 | } | |
298 | ||
8ec4d41f GH |
299 | #ifdef CONFIG_SMP |
300 | ||
5967ed87 JB |
301 | static void alternatives_smp_lock(const s32 *start, const s32 *end, |
302 | u8 *text, u8 *text_end) | |
9a0b5817 | 303 | { |
5967ed87 | 304 | const s32 *poff; |
9a0b5817 | 305 | |
3945dab4 | 306 | mutex_lock(&text_mutex); |
5967ed87 JB |
307 | for (poff = start; poff < end; poff++) { |
308 | u8 *ptr = (u8 *)poff + *poff; | |
309 | ||
310 | if (!*poff || ptr < text || ptr >= text_end) | |
9a0b5817 | 311 | continue; |
f88f07e0 | 312 | /* turn DS segment override prefix into lock prefix */ |
d9c5841e PA |
313 | if (*ptr == 0x3e) |
314 | text_poke(ptr, ((unsigned char []){0xf0}), 1); | |
9a0b5817 | 315 | }; |
3945dab4 | 316 | mutex_unlock(&text_mutex); |
9a0b5817 GH |
317 | } |
318 | ||
5967ed87 JB |
319 | static void alternatives_smp_unlock(const s32 *start, const s32 *end, |
320 | u8 *text, u8 *text_end) | |
9a0b5817 | 321 | { |
5967ed87 | 322 | const s32 *poff; |
9a0b5817 | 323 | |
b7fb4af0 JF |
324 | if (noreplace_smp) |
325 | return; | |
326 | ||
3945dab4 | 327 | mutex_lock(&text_mutex); |
5967ed87 JB |
328 | for (poff = start; poff < end; poff++) { |
329 | u8 *ptr = (u8 *)poff + *poff; | |
330 | ||
331 | if (!*poff || ptr < text || ptr >= text_end) | |
9a0b5817 | 332 | continue; |
f88f07e0 | 333 | /* turn lock prefix into DS segment override prefix */ |
d9c5841e PA |
334 | if (*ptr == 0xf0) |
335 | text_poke(ptr, ((unsigned char []){0x3E}), 1); | |
9a0b5817 | 336 | }; |
3945dab4 | 337 | mutex_unlock(&text_mutex); |
9a0b5817 GH |
338 | } |
339 | ||
340 | struct smp_alt_module { | |
341 | /* what is this ??? */ | |
342 | struct module *mod; | |
343 | char *name; | |
344 | ||
345 | /* ptrs to lock prefixes */ | |
5967ed87 JB |
346 | const s32 *locks; |
347 | const s32 *locks_end; | |
9a0b5817 GH |
348 | |
349 | /* .text segment, needed to avoid patching init code ;) */ | |
350 | u8 *text; | |
351 | u8 *text_end; | |
352 | ||
353 | struct list_head next; | |
354 | }; | |
355 | static LIST_HEAD(smp_alt_modules); | |
2f1dafe5 | 356 | static DEFINE_MUTEX(smp_alt); |
ca74a6f8 | 357 | static int smp_mode = 1; /* protected by smp_alt */ |
9a0b5817 | 358 | |
8b5a10fc JB |
359 | void __init_or_module alternatives_smp_module_add(struct module *mod, |
360 | char *name, | |
361 | void *locks, void *locks_end, | |
362 | void *text, void *text_end) | |
9a0b5817 GH |
363 | { |
364 | struct smp_alt_module *smp; | |
9a0b5817 | 365 | |
b7fb4af0 JF |
366 | if (noreplace_smp) |
367 | return; | |
368 | ||
9a0b5817 GH |
369 | if (smp_alt_once) { |
370 | if (boot_cpu_has(X86_FEATURE_UP)) | |
371 | alternatives_smp_unlock(locks, locks_end, | |
372 | text, text_end); | |
373 | return; | |
374 | } | |
375 | ||
376 | smp = kzalloc(sizeof(*smp), GFP_KERNEL); | |
377 | if (NULL == smp) | |
378 | return; /* we'll run the (safe but slow) SMP code then ... */ | |
379 | ||
380 | smp->mod = mod; | |
381 | smp->name = name; | |
382 | smp->locks = locks; | |
383 | smp->locks_end = locks_end; | |
384 | smp->text = text; | |
385 | smp->text_end = text_end; | |
386 | DPRINTK("%s: locks %p -> %p, text %p -> %p, name %s\n", | |
77bf90ed | 387 | __func__, smp->locks, smp->locks_end, |
9a0b5817 GH |
388 | smp->text, smp->text_end, smp->name); |
389 | ||
2f1dafe5 | 390 | mutex_lock(&smp_alt); |
9a0b5817 GH |
391 | list_add_tail(&smp->next, &smp_alt_modules); |
392 | if (boot_cpu_has(X86_FEATURE_UP)) | |
393 | alternatives_smp_unlock(smp->locks, smp->locks_end, | |
394 | smp->text, smp->text_end); | |
2f1dafe5 | 395 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
396 | } |
397 | ||
8b5a10fc | 398 | void __init_or_module alternatives_smp_module_del(struct module *mod) |
9a0b5817 GH |
399 | { |
400 | struct smp_alt_module *item; | |
9a0b5817 | 401 | |
b7fb4af0 | 402 | if (smp_alt_once || noreplace_smp) |
9a0b5817 GH |
403 | return; |
404 | ||
2f1dafe5 | 405 | mutex_lock(&smp_alt); |
9a0b5817 GH |
406 | list_for_each_entry(item, &smp_alt_modules, next) { |
407 | if (mod != item->mod) | |
408 | continue; | |
409 | list_del(&item->next); | |
2f1dafe5 | 410 | mutex_unlock(&smp_alt); |
77bf90ed | 411 | DPRINTK("%s: %s\n", __func__, item->name); |
9a0b5817 GH |
412 | kfree(item); |
413 | return; | |
414 | } | |
2f1dafe5 | 415 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
416 | } |
417 | ||
3fb82d56 | 418 | bool skip_smp_alternatives; |
9a0b5817 GH |
419 | void alternatives_smp_switch(int smp) |
420 | { | |
421 | struct smp_alt_module *mod; | |
9a0b5817 | 422 | |
3047e99e IM |
423 | #ifdef CONFIG_LOCKDEP |
424 | /* | |
17abecfe IM |
425 | * Older binutils section handling bug prevented |
426 | * alternatives-replacement from working reliably. | |
427 | * | |
428 | * If this still occurs then you should see a hang | |
429 | * or crash shortly after this line: | |
3047e99e | 430 | */ |
17abecfe | 431 | printk("lockdep: fixing up alternatives.\n"); |
3047e99e IM |
432 | #endif |
433 | ||
3fb82d56 | 434 | if (noreplace_smp || smp_alt_once || skip_smp_alternatives) |
9a0b5817 GH |
435 | return; |
436 | BUG_ON(!smp && (num_online_cpus() > 1)); | |
437 | ||
2f1dafe5 | 438 | mutex_lock(&smp_alt); |
ca74a6f8 AK |
439 | |
440 | /* | |
441 | * Avoid unnecessary switches because it forces JIT based VMs to | |
442 | * throw away all cached translations, which can be quite costly. | |
443 | */ | |
444 | if (smp == smp_mode) { | |
445 | /* nothing */ | |
446 | } else if (smp) { | |
9a0b5817 | 447 | printk(KERN_INFO "SMP alternatives: switching to SMP code\n"); |
53756d37 JF |
448 | clear_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
449 | clear_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
9a0b5817 GH |
450 | list_for_each_entry(mod, &smp_alt_modules, next) |
451 | alternatives_smp_lock(mod->locks, mod->locks_end, | |
452 | mod->text, mod->text_end); | |
453 | } else { | |
454 | printk(KERN_INFO "SMP alternatives: switching to UP code\n"); | |
53756d37 JF |
455 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
456 | set_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
9a0b5817 GH |
457 | list_for_each_entry(mod, &smp_alt_modules, next) |
458 | alternatives_smp_unlock(mod->locks, mod->locks_end, | |
459 | mod->text, mod->text_end); | |
460 | } | |
ca74a6f8 | 461 | smp_mode = smp; |
2f1dafe5 | 462 | mutex_unlock(&smp_alt); |
9a0b5817 GH |
463 | } |
464 | ||
2cfa1978 MH |
465 | /* Return 1 if the address range is reserved for smp-alternatives */ |
466 | int alternatives_text_reserved(void *start, void *end) | |
467 | { | |
468 | struct smp_alt_module *mod; | |
5967ed87 | 469 | const s32 *poff; |
076dc4a6 MH |
470 | u8 *text_start = start; |
471 | u8 *text_end = end; | |
2cfa1978 MH |
472 | |
473 | list_for_each_entry(mod, &smp_alt_modules, next) { | |
076dc4a6 | 474 | if (mod->text > text_end || mod->text_end < text_start) |
2cfa1978 | 475 | continue; |
5967ed87 JB |
476 | for (poff = mod->locks; poff < mod->locks_end; poff++) { |
477 | const u8 *ptr = (const u8 *)poff + *poff; | |
478 | ||
479 | if (text_start <= ptr && text_end > ptr) | |
2cfa1978 | 480 | return 1; |
5967ed87 | 481 | } |
2cfa1978 MH |
482 | } |
483 | ||
484 | return 0; | |
485 | } | |
8ec4d41f GH |
486 | #endif |
487 | ||
139ec7c4 | 488 | #ifdef CONFIG_PARAVIRT |
8b5a10fc JB |
489 | void __init_or_module apply_paravirt(struct paravirt_patch_site *start, |
490 | struct paravirt_patch_site *end) | |
139ec7c4 | 491 | { |
98de032b | 492 | struct paravirt_patch_site *p; |
ab144f5e | 493 | char insnbuf[MAX_PATCH_LEN]; |
139ec7c4 | 494 | |
959b4fdf JF |
495 | if (noreplace_paravirt) |
496 | return; | |
497 | ||
139ec7c4 RR |
498 | for (p = start; p < end; p++) { |
499 | unsigned int used; | |
500 | ||
ab144f5e | 501 | BUG_ON(p->len > MAX_PATCH_LEN); |
d34fda4a CW |
502 | /* prep the buffer with the original instructions */ |
503 | memcpy(insnbuf, p->instr, p->len); | |
93b1eab3 JF |
504 | used = pv_init_ops.patch(p->instrtype, p->clobbers, insnbuf, |
505 | (unsigned long)p->instr, p->len); | |
7f63c41c | 506 | |
63f70270 JF |
507 | BUG_ON(used > p->len); |
508 | ||
139ec7c4 | 509 | /* Pad the rest with nops */ |
ab144f5e | 510 | add_nops(insnbuf + used, p->len - used); |
e587cadd | 511 | text_poke_early(p->instr, insnbuf, p->len); |
139ec7c4 | 512 | } |
139ec7c4 | 513 | } |
98de032b | 514 | extern struct paravirt_patch_site __start_parainstructions[], |
139ec7c4 RR |
515 | __stop_parainstructions[]; |
516 | #endif /* CONFIG_PARAVIRT */ | |
517 | ||
9a0b5817 GH |
518 | void __init alternative_instructions(void) |
519 | { | |
8f4e956b AK |
520 | /* The patching is not fully atomic, so try to avoid local interruptions |
521 | that might execute the to be patched code. | |
522 | Other CPUs are not running. */ | |
523 | stop_nmi(); | |
123aa76e AK |
524 | |
525 | /* | |
526 | * Don't stop machine check exceptions while patching. | |
527 | * MCEs only happen when something got corrupted and in this | |
528 | * case we must do something about the corruption. | |
529 | * Ignoring it is worse than a unlikely patching race. | |
530 | * Also machine checks tend to be broadcast and if one CPU | |
531 | * goes into machine check the others follow quickly, so we don't | |
532 | * expect a machine check to cause undue problems during to code | |
533 | * patching. | |
534 | */ | |
8f4e956b | 535 | |
9a0b5817 GH |
536 | apply_alternatives(__alt_instructions, __alt_instructions_end); |
537 | ||
538 | /* switch to patch-once-at-boottime-only mode and free the | |
539 | * tables in case we know the number of CPUs will never ever | |
540 | * change */ | |
541 | #ifdef CONFIG_HOTPLUG_CPU | |
542 | if (num_possible_cpus() < 2) | |
543 | smp_alt_once = 1; | |
9a0b5817 GH |
544 | #endif |
545 | ||
8ec4d41f | 546 | #ifdef CONFIG_SMP |
9a0b5817 GH |
547 | if (smp_alt_once) { |
548 | if (1 == num_possible_cpus()) { | |
549 | printk(KERN_INFO "SMP alternatives: switching to UP code\n"); | |
53756d37 JF |
550 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP); |
551 | set_cpu_cap(&cpu_data(0), X86_FEATURE_UP); | |
552 | ||
9a0b5817 GH |
553 | alternatives_smp_unlock(__smp_locks, __smp_locks_end, |
554 | _text, _etext); | |
555 | } | |
9a0b5817 | 556 | } else { |
9a0b5817 GH |
557 | alternatives_smp_module_add(NULL, "core kernel", |
558 | __smp_locks, __smp_locks_end, | |
559 | _text, _etext); | |
ca74a6f8 AK |
560 | |
561 | /* Only switch to UP mode if we don't immediately boot others */ | |
649c6653 | 562 | if (num_present_cpus() == 1 || setup_max_cpus <= 1) |
ca74a6f8 | 563 | alternatives_smp_switch(0); |
9a0b5817 | 564 | } |
8ec4d41f | 565 | #endif |
441d40dc | 566 | apply_paravirt(__parainstructions, __parainstructions_end); |
8f4e956b | 567 | |
f68fd5f4 FW |
568 | if (smp_alt_once) |
569 | free_init_pages("SMP alternatives", | |
570 | (unsigned long)__smp_locks, | |
571 | (unsigned long)__smp_locks_end); | |
572 | ||
8f4e956b | 573 | restart_nmi(); |
9a0b5817 | 574 | } |
19d36ccd | 575 | |
e587cadd MD |
576 | /** |
577 | * text_poke_early - Update instructions on a live kernel at boot time | |
578 | * @addr: address to modify | |
579 | * @opcode: source of the copy | |
580 | * @len: length to copy | |
581 | * | |
19d36ccd AK |
582 | * When you use this code to patch more than one byte of an instruction |
583 | * you need to make sure that other CPUs cannot execute this code in parallel. | |
e587cadd MD |
584 | * Also no thread must be currently preempted in the middle of these |
585 | * instructions. And on the local CPU you need to be protected again NMI or MCE | |
586 | * handlers seeing an inconsistent instruction while you patch. | |
19d36ccd | 587 | */ |
fa6f2cc7 | 588 | void *__init_or_module text_poke_early(void *addr, const void *opcode, |
8b5a10fc | 589 | size_t len) |
19d36ccd | 590 | { |
e587cadd MD |
591 | unsigned long flags; |
592 | local_irq_save(flags); | |
19d36ccd | 593 | memcpy(addr, opcode, len); |
e587cadd | 594 | sync_core(); |
5367b688 | 595 | local_irq_restore(flags); |
e587cadd MD |
596 | /* Could also do a CLFLUSH here to speed up CPU recovery; but |
597 | that causes hangs on some VIA CPUs. */ | |
598 | return addr; | |
599 | } | |
600 | ||
601 | /** | |
602 | * text_poke - Update instructions on a live kernel | |
603 | * @addr: address to modify | |
604 | * @opcode: source of the copy | |
605 | * @len: length to copy | |
606 | * | |
607 | * Only atomic text poke/set should be allowed when not doing early patching. | |
608 | * It means the size must be writable atomically and the address must be aligned | |
609 | * in a way that permits an atomic write. It also makes sure we fit on a single | |
610 | * page. | |
78ff7fae MH |
611 | * |
612 | * Note: Must be called under text_mutex. | |
e587cadd MD |
613 | */ |
614 | void *__kprobes text_poke(void *addr, const void *opcode, size_t len) | |
615 | { | |
78ff7fae | 616 | unsigned long flags; |
e587cadd | 617 | char *vaddr; |
b7b66baa MD |
618 | struct page *pages[2]; |
619 | int i; | |
e587cadd | 620 | |
b7b66baa MD |
621 | if (!core_kernel_text((unsigned long)addr)) { |
622 | pages[0] = vmalloc_to_page(addr); | |
623 | pages[1] = vmalloc_to_page(addr + PAGE_SIZE); | |
15a601eb | 624 | } else { |
b7b66baa | 625 | pages[0] = virt_to_page(addr); |
00c6b2d5 | 626 | WARN_ON(!PageReserved(pages[0])); |
b7b66baa | 627 | pages[1] = virt_to_page(addr + PAGE_SIZE); |
e587cadd | 628 | } |
b7b66baa | 629 | BUG_ON(!pages[0]); |
7cf49427 | 630 | local_irq_save(flags); |
78ff7fae MH |
631 | set_fixmap(FIX_TEXT_POKE0, page_to_phys(pages[0])); |
632 | if (pages[1]) | |
633 | set_fixmap(FIX_TEXT_POKE1, page_to_phys(pages[1])); | |
634 | vaddr = (char *)fix_to_virt(FIX_TEXT_POKE0); | |
b7b66baa | 635 | memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len); |
78ff7fae MH |
636 | clear_fixmap(FIX_TEXT_POKE0); |
637 | if (pages[1]) | |
638 | clear_fixmap(FIX_TEXT_POKE1); | |
639 | local_flush_tlb(); | |
19d36ccd | 640 | sync_core(); |
a534b679 AK |
641 | /* Could also do a CLFLUSH here to speed up CPU recovery; but |
642 | that causes hangs on some VIA CPUs. */ | |
b7b66baa MD |
643 | for (i = 0; i < len; i++) |
644 | BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]); | |
7cf49427 | 645 | local_irq_restore(flags); |
e587cadd | 646 | return addr; |
19d36ccd | 647 | } |
3d55cc8a MH |
648 | |
649 | /* | |
650 | * Cross-modifying kernel text with stop_machine(). | |
651 | * This code originally comes from immediate value. | |
652 | */ | |
653 | static atomic_t stop_machine_first; | |
654 | static int wrote_text; | |
655 | ||
656 | struct text_poke_params { | |
7deb18dc MH |
657 | struct text_poke_param *params; |
658 | int nparams; | |
3d55cc8a MH |
659 | }; |
660 | ||
661 | static int __kprobes stop_machine_text_poke(void *data) | |
662 | { | |
663 | struct text_poke_params *tpp = data; | |
7deb18dc MH |
664 | struct text_poke_param *p; |
665 | int i; | |
3d55cc8a MH |
666 | |
667 | if (atomic_dec_and_test(&stop_machine_first)) { | |
7deb18dc MH |
668 | for (i = 0; i < tpp->nparams; i++) { |
669 | p = &tpp->params[i]; | |
670 | text_poke(p->addr, p->opcode, p->len); | |
671 | } | |
3d55cc8a MH |
672 | smp_wmb(); /* Make sure other cpus see that this has run */ |
673 | wrote_text = 1; | |
674 | } else { | |
675 | while (!wrote_text) | |
e5a11016 MH |
676 | cpu_relax(); |
677 | smp_mb(); /* Load wrote_text before following execution */ | |
3d55cc8a MH |
678 | } |
679 | ||
7deb18dc MH |
680 | for (i = 0; i < tpp->nparams; i++) { |
681 | p = &tpp->params[i]; | |
682 | flush_icache_range((unsigned long)p->addr, | |
683 | (unsigned long)p->addr + p->len); | |
684 | } | |
0e00f7ae MD |
685 | /* |
686 | * Intel Archiecture Software Developer's Manual section 7.1.3 specifies | |
687 | * that a core serializing instruction such as "cpuid" should be | |
688 | * executed on _each_ core before the new instruction is made visible. | |
689 | */ | |
690 | sync_core(); | |
3d55cc8a MH |
691 | return 0; |
692 | } | |
693 | ||
694 | /** | |
695 | * text_poke_smp - Update instructions on a live kernel on SMP | |
696 | * @addr: address to modify | |
697 | * @opcode: source of the copy | |
698 | * @len: length to copy | |
699 | * | |
700 | * Modify multi-byte instruction by using stop_machine() on SMP. This allows | |
701 | * user to poke/set multi-byte text on SMP. Only non-NMI/MCE code modifying | |
702 | * should be allowed, since stop_machine() does _not_ protect code against | |
703 | * NMI and MCE. | |
704 | * | |
705 | * Note: Must be called under get_online_cpus() and text_mutex. | |
706 | */ | |
707 | void *__kprobes text_poke_smp(void *addr, const void *opcode, size_t len) | |
708 | { | |
709 | struct text_poke_params tpp; | |
7deb18dc | 710 | struct text_poke_param p; |
3d55cc8a | 711 | |
7deb18dc MH |
712 | p.addr = addr; |
713 | p.opcode = opcode; | |
714 | p.len = len; | |
715 | tpp.params = &p; | |
716 | tpp.nparams = 1; | |
3d55cc8a MH |
717 | atomic_set(&stop_machine_first, 1); |
718 | wrote_text = 0; | |
3caa3751 | 719 | /* Use __stop_machine() because the caller already got online_cpus. */ |
404ba5d7 | 720 | __stop_machine(stop_machine_text_poke, (void *)&tpp, cpu_online_mask); |
3d55cc8a MH |
721 | return addr; |
722 | } | |
723 | ||
7deb18dc MH |
724 | /** |
725 | * text_poke_smp_batch - Update instructions on a live kernel on SMP | |
726 | * @params: an array of text_poke parameters | |
727 | * @n: the number of elements in params. | |
728 | * | |
729 | * Modify multi-byte instruction by using stop_machine() on SMP. Since the | |
730 | * stop_machine() is heavy task, it is better to aggregate text_poke requests | |
731 | * and do it once if possible. | |
732 | * | |
733 | * Note: Must be called under get_online_cpus() and text_mutex. | |
734 | */ | |
735 | void __kprobes text_poke_smp_batch(struct text_poke_param *params, int n) | |
736 | { | |
737 | struct text_poke_params tpp = {.params = params, .nparams = n}; | |
738 | ||
739 | atomic_set(&stop_machine_first, 1); | |
740 | wrote_text = 0; | |
78345d2e | 741 | __stop_machine(stop_machine_text_poke, (void *)&tpp, cpu_online_mask); |
7deb18dc | 742 | } |