dm snapshot: avoid storing private suspended state
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kernel / alternative.c
blob123608531c8f933b819a3fc7c135748137c8eb5b
1 #include <linux/module.h>
2 #include <linux/sched.h>
3 #include <linux/mutex.h>
4 #include <linux/list.h>
5 #include <linux/stringify.h>
6 #include <linux/kprobes.h>
7 #include <linux/mm.h>
8 #include <linux/vmalloc.h>
9 #include <linux/memory.h>
10 #include <linux/stop_machine.h>
11 #include <linux/slab.h>
12 #include <asm/alternative.h>
13 #include <asm/sections.h>
14 #include <asm/pgtable.h>
15 #include <asm/mce.h>
16 #include <asm/nmi.h>
17 #include <asm/vsyscall.h>
18 #include <asm/cacheflush.h>
19 #include <asm/tlbflush.h>
20 #include <asm/io.h>
21 #include <asm/fixmap.h>
23 #define MAX_PATCH_LEN (255-1)
25 #ifdef CONFIG_HOTPLUG_CPU
26 static int smp_alt_once;
28 static int __init bootonly(char *str)
30 smp_alt_once = 1;
31 return 1;
33 __setup("smp-alt-boot", bootonly);
34 #else
35 #define smp_alt_once 1
36 #endif
38 static int __initdata_or_module debug_alternative;
40 static int __init debug_alt(char *str)
42 debug_alternative = 1;
43 return 1;
45 __setup("debug-alternative", debug_alt);
47 static int noreplace_smp;
49 static int __init setup_noreplace_smp(char *str)
51 noreplace_smp = 1;
52 return 1;
54 __setup("noreplace-smp", setup_noreplace_smp);
56 #ifdef CONFIG_PARAVIRT
57 static int __initdata_or_module noreplace_paravirt = 0;
59 static int __init setup_noreplace_paravirt(char *str)
61 noreplace_paravirt = 1;
62 return 1;
64 __setup("noreplace-paravirt", setup_noreplace_paravirt);
65 #endif
67 #define DPRINTK(fmt, args...) if (debug_alternative) \
68 printk(KERN_DEBUG fmt, args)
70 #if defined(GENERIC_NOP1) && !defined(CONFIG_X86_64)
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. */
74 asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nintelnops: "
75 GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6
76 GENERIC_NOP7 GENERIC_NOP8
77 "\t.previous");
78 extern const unsigned char intelnops[];
79 static const unsigned char *const __initconst_or_module
80 intel_nops[ASM_NOP_MAX+1] = {
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,
91 #endif
93 #ifdef K8_NOP1
94 asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nk8nops: "
95 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
96 K8_NOP7 K8_NOP8
97 "\t.previous");
98 extern const unsigned char k8nops[];
99 static const unsigned char *const __initconst_or_module
100 k8_nops[ASM_NOP_MAX+1] = {
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,
111 #endif
113 #if defined(K7_NOP1) && !defined(CONFIG_X86_64)
114 asm("\t" __stringify(__INITRODATA_OR_MODULE) "\nk7nops: "
115 K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6
116 K7_NOP7 K7_NOP8
117 "\t.previous");
118 extern const unsigned char k7nops[];
119 static const unsigned char *const __initconst_or_module
120 k7_nops[ASM_NOP_MAX+1] = {
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,
131 #endif
133 #ifdef P6_NOP1
134 asm("\t" __stringify(__INITRODATA_OR_MODULE) "\np6nops: "
135 P6_NOP1 P6_NOP2 P6_NOP3 P6_NOP4 P6_NOP5 P6_NOP6
136 P6_NOP7 P6_NOP8
137 "\t.previous");
138 extern const unsigned char p6nops[];
139 static const unsigned char *const __initconst_or_module
140 p6_nops[ASM_NOP_MAX+1] = {
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,
151 #endif
153 #ifdef CONFIG_X86_64
155 extern char __vsyscall_0;
156 static const unsigned char *const *__init_or_module find_nop_table(void)
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;
165 #else /* CONFIG_X86_64 */
167 static const unsigned char *const *__init_or_module find_nop_table(void)
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;
179 #endif /* CONFIG_X86_64 */
181 /* Use this to add nops to a buffer, then text_poke the whole buffer. */
182 static void __init_or_module add_nops(void *insns, unsigned int len)
184 const unsigned char *const *noptable = find_nop_table();
186 while (len > 0) {
187 unsigned int noplen = len;
188 if (noplen > ASM_NOP_MAX)
189 noplen = ASM_NOP_MAX;
190 memcpy(insns, noptable[noplen], noplen);
191 insns += noplen;
192 len -= noplen;
196 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
197 extern s32 __smp_locks[], __smp_locks_end[];
198 void *text_poke_early(void *addr, const void *opcode, size_t len);
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. */
206 void __init_or_module apply_alternatives(struct alt_instr *start,
207 struct alt_instr *end)
209 struct alt_instr *a;
210 u8 insnbuf[MAX_PATCH_LEN];
212 DPRINTK("%s: alt table %p -> %p\n", __func__, start, end);
213 for (a = start; a < end; a++) {
214 u8 *instr = a->instr;
215 BUG_ON(a->replacementlen > a->instrlen);
216 BUG_ON(a->instrlen > sizeof(insnbuf));
217 BUG_ON(a->cpuid >= NCAPINTS*32);
218 if (!boot_cpu_has(a->cpuid))
219 continue;
220 #ifdef CONFIG_X86_64
221 /* vsyscall code is not mapped yet. resolve it manually. */
222 if (instr >= (u8 *)VSYSCALL_START && instr < (u8*)VSYSCALL_END) {
223 instr = __va(instr - (u8*)VSYSCALL_START + (u8*)__pa_symbol(&__vsyscall_0));
224 DPRINTK("%s: vsyscall fixup: %p => %p\n",
225 __func__, a->instr, instr);
227 #endif
228 memcpy(insnbuf, a->replacement, a->replacementlen);
229 if (*insnbuf == 0xe8 && a->replacementlen == 5)
230 *(s32 *)(insnbuf + 1) += a->replacement - a->instr;
231 add_nops(insnbuf + a->replacementlen,
232 a->instrlen - a->replacementlen);
233 text_poke_early(instr, insnbuf, a->instrlen);
237 #ifdef CONFIG_SMP
239 static void alternatives_smp_lock(const s32 *start, const s32 *end,
240 u8 *text, u8 *text_end)
242 const s32 *poff;
244 mutex_lock(&text_mutex);
245 for (poff = start; poff < end; poff++) {
246 u8 *ptr = (u8 *)poff + *poff;
248 if (!*poff || ptr < text || ptr >= text_end)
249 continue;
250 /* turn DS segment override prefix into lock prefix */
251 if (*ptr == 0x3e)
252 text_poke(ptr, ((unsigned char []){0xf0}), 1);
254 mutex_unlock(&text_mutex);
257 static void alternatives_smp_unlock(const s32 *start, const s32 *end,
258 u8 *text, u8 *text_end)
260 const s32 *poff;
262 if (noreplace_smp)
263 return;
265 mutex_lock(&text_mutex);
266 for (poff = start; poff < end; poff++) {
267 u8 *ptr = (u8 *)poff + *poff;
269 if (!*poff || ptr < text || ptr >= text_end)
270 continue;
271 /* turn lock prefix into DS segment override prefix */
272 if (*ptr == 0xf0)
273 text_poke(ptr, ((unsigned char []){0x3E}), 1);
275 mutex_unlock(&text_mutex);
278 struct smp_alt_module {
279 /* what is this ??? */
280 struct module *mod;
281 char *name;
283 /* ptrs to lock prefixes */
284 const s32 *locks;
285 const s32 *locks_end;
287 /* .text segment, needed to avoid patching init code ;) */
288 u8 *text;
289 u8 *text_end;
291 struct list_head next;
293 static LIST_HEAD(smp_alt_modules);
294 static DEFINE_MUTEX(smp_alt);
295 static int smp_mode = 1; /* protected by smp_alt */
297 void __init_or_module alternatives_smp_module_add(struct module *mod,
298 char *name,
299 void *locks, void *locks_end,
300 void *text, void *text_end)
302 struct smp_alt_module *smp;
304 if (noreplace_smp)
305 return;
307 if (smp_alt_once) {
308 if (boot_cpu_has(X86_FEATURE_UP))
309 alternatives_smp_unlock(locks, locks_end,
310 text, text_end);
311 return;
314 smp = kzalloc(sizeof(*smp), GFP_KERNEL);
315 if (NULL == smp)
316 return; /* we'll run the (safe but slow) SMP code then ... */
318 smp->mod = mod;
319 smp->name = name;
320 smp->locks = locks;
321 smp->locks_end = locks_end;
322 smp->text = text;
323 smp->text_end = text_end;
324 DPRINTK("%s: locks %p -> %p, text %p -> %p, name %s\n",
325 __func__, smp->locks, smp->locks_end,
326 smp->text, smp->text_end, smp->name);
328 mutex_lock(&smp_alt);
329 list_add_tail(&smp->next, &smp_alt_modules);
330 if (boot_cpu_has(X86_FEATURE_UP))
331 alternatives_smp_unlock(smp->locks, smp->locks_end,
332 smp->text, smp->text_end);
333 mutex_unlock(&smp_alt);
336 void __init_or_module alternatives_smp_module_del(struct module *mod)
338 struct smp_alt_module *item;
340 if (smp_alt_once || noreplace_smp)
341 return;
343 mutex_lock(&smp_alt);
344 list_for_each_entry(item, &smp_alt_modules, next) {
345 if (mod != item->mod)
346 continue;
347 list_del(&item->next);
348 mutex_unlock(&smp_alt);
349 DPRINTK("%s: %s\n", __func__, item->name);
350 kfree(item);
351 return;
353 mutex_unlock(&smp_alt);
356 bool skip_smp_alternatives;
357 void alternatives_smp_switch(int smp)
359 struct smp_alt_module *mod;
361 #ifdef CONFIG_LOCKDEP
363 * Older binutils section handling bug prevented
364 * alternatives-replacement from working reliably.
366 * If this still occurs then you should see a hang
367 * or crash shortly after this line:
369 printk("lockdep: fixing up alternatives.\n");
370 #endif
372 if (noreplace_smp || smp_alt_once || skip_smp_alternatives)
373 return;
374 BUG_ON(!smp && (num_online_cpus() > 1));
376 mutex_lock(&smp_alt);
379 * Avoid unnecessary switches because it forces JIT based VMs to
380 * throw away all cached translations, which can be quite costly.
382 if (smp == smp_mode) {
383 /* nothing */
384 } else if (smp) {
385 printk(KERN_INFO "SMP alternatives: switching to SMP code\n");
386 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
387 clear_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
388 list_for_each_entry(mod, &smp_alt_modules, next)
389 alternatives_smp_lock(mod->locks, mod->locks_end,
390 mod->text, mod->text_end);
391 } else {
392 printk(KERN_INFO "SMP alternatives: switching to UP code\n");
393 set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
394 set_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
395 list_for_each_entry(mod, &smp_alt_modules, next)
396 alternatives_smp_unlock(mod->locks, mod->locks_end,
397 mod->text, mod->text_end);
399 smp_mode = smp;
400 mutex_unlock(&smp_alt);
403 /* Return 1 if the address range is reserved for smp-alternatives */
404 int alternatives_text_reserved(void *start, void *end)
406 struct smp_alt_module *mod;
407 const s32 *poff;
408 u8 *text_start = start;
409 u8 *text_end = end;
411 list_for_each_entry(mod, &smp_alt_modules, next) {
412 if (mod->text > text_end || mod->text_end < text_start)
413 continue;
414 for (poff = mod->locks; poff < mod->locks_end; poff++) {
415 const u8 *ptr = (const u8 *)poff + *poff;
417 if (text_start <= ptr && text_end > ptr)
418 return 1;
422 return 0;
424 #endif
426 #ifdef CONFIG_PARAVIRT
427 void __init_or_module apply_paravirt(struct paravirt_patch_site *start,
428 struct paravirt_patch_site *end)
430 struct paravirt_patch_site *p;
431 char insnbuf[MAX_PATCH_LEN];
433 if (noreplace_paravirt)
434 return;
436 for (p = start; p < end; p++) {
437 unsigned int used;
439 BUG_ON(p->len > MAX_PATCH_LEN);
440 /* prep the buffer with the original instructions */
441 memcpy(insnbuf, p->instr, p->len);
442 used = pv_init_ops.patch(p->instrtype, p->clobbers, insnbuf,
443 (unsigned long)p->instr, p->len);
445 BUG_ON(used > p->len);
447 /* Pad the rest with nops */
448 add_nops(insnbuf + used, p->len - used);
449 text_poke_early(p->instr, insnbuf, p->len);
452 extern struct paravirt_patch_site __start_parainstructions[],
453 __stop_parainstructions[];
454 #endif /* CONFIG_PARAVIRT */
456 void __init alternative_instructions(void)
458 /* The patching is not fully atomic, so try to avoid local interruptions
459 that might execute the to be patched code.
460 Other CPUs are not running. */
461 stop_nmi();
464 * Don't stop machine check exceptions while patching.
465 * MCEs only happen when something got corrupted and in this
466 * case we must do something about the corruption.
467 * Ignoring it is worse than a unlikely patching race.
468 * Also machine checks tend to be broadcast and if one CPU
469 * goes into machine check the others follow quickly, so we don't
470 * expect a machine check to cause undue problems during to code
471 * patching.
474 apply_alternatives(__alt_instructions, __alt_instructions_end);
476 /* switch to patch-once-at-boottime-only mode and free the
477 * tables in case we know the number of CPUs will never ever
478 * change */
479 #ifdef CONFIG_HOTPLUG_CPU
480 if (num_possible_cpus() < 2)
481 smp_alt_once = 1;
482 #endif
484 #ifdef CONFIG_SMP
485 if (smp_alt_once) {
486 if (1 == num_possible_cpus()) {
487 printk(KERN_INFO "SMP alternatives: switching to UP code\n");
488 set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
489 set_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
491 alternatives_smp_unlock(__smp_locks, __smp_locks_end,
492 _text, _etext);
494 } else {
495 alternatives_smp_module_add(NULL, "core kernel",
496 __smp_locks, __smp_locks_end,
497 _text, _etext);
499 /* Only switch to UP mode if we don't immediately boot others */
500 if (num_present_cpus() == 1 || setup_max_cpus <= 1)
501 alternatives_smp_switch(0);
503 #endif
504 apply_paravirt(__parainstructions, __parainstructions_end);
506 if (smp_alt_once)
507 free_init_pages("SMP alternatives",
508 (unsigned long)__smp_locks,
509 (unsigned long)__smp_locks_end);
511 restart_nmi();
515 * text_poke_early - Update instructions on a live kernel at boot time
516 * @addr: address to modify
517 * @opcode: source of the copy
518 * @len: length to copy
520 * When you use this code to patch more than one byte of an instruction
521 * you need to make sure that other CPUs cannot execute this code in parallel.
522 * Also no thread must be currently preempted in the middle of these
523 * instructions. And on the local CPU you need to be protected again NMI or MCE
524 * handlers seeing an inconsistent instruction while you patch.
526 void *__init_or_module text_poke_early(void *addr, const void *opcode,
527 size_t len)
529 unsigned long flags;
530 local_irq_save(flags);
531 memcpy(addr, opcode, len);
532 sync_core();
533 local_irq_restore(flags);
534 /* Could also do a CLFLUSH here to speed up CPU recovery; but
535 that causes hangs on some VIA CPUs. */
536 return addr;
540 * text_poke - Update instructions on a live kernel
541 * @addr: address to modify
542 * @opcode: source of the copy
543 * @len: length to copy
545 * Only atomic text poke/set should be allowed when not doing early patching.
546 * It means the size must be writable atomically and the address must be aligned
547 * in a way that permits an atomic write. It also makes sure we fit on a single
548 * page.
550 * Note: Must be called under text_mutex.
552 void *__kprobes text_poke(void *addr, const void *opcode, size_t len)
554 unsigned long flags;
555 char *vaddr;
556 struct page *pages[2];
557 int i;
559 if (!core_kernel_text((unsigned long)addr)) {
560 pages[0] = vmalloc_to_page(addr);
561 pages[1] = vmalloc_to_page(addr + PAGE_SIZE);
562 } else {
563 pages[0] = virt_to_page(addr);
564 WARN_ON(!PageReserved(pages[0]));
565 pages[1] = virt_to_page(addr + PAGE_SIZE);
567 BUG_ON(!pages[0]);
568 local_irq_save(flags);
569 set_fixmap(FIX_TEXT_POKE0, page_to_phys(pages[0]));
570 if (pages[1])
571 set_fixmap(FIX_TEXT_POKE1, page_to_phys(pages[1]));
572 vaddr = (char *)fix_to_virt(FIX_TEXT_POKE0);
573 memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len);
574 clear_fixmap(FIX_TEXT_POKE0);
575 if (pages[1])
576 clear_fixmap(FIX_TEXT_POKE1);
577 local_flush_tlb();
578 sync_core();
579 /* Could also do a CLFLUSH here to speed up CPU recovery; but
580 that causes hangs on some VIA CPUs. */
581 for (i = 0; i < len; i++)
582 BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]);
583 local_irq_restore(flags);
584 return addr;
588 * Cross-modifying kernel text with stop_machine().
589 * This code originally comes from immediate value.
591 static atomic_t stop_machine_first;
592 static int wrote_text;
594 struct text_poke_params {
595 struct text_poke_param *params;
596 int nparams;
599 static int __kprobes stop_machine_text_poke(void *data)
601 struct text_poke_params *tpp = data;
602 struct text_poke_param *p;
603 int i;
605 if (atomic_dec_and_test(&stop_machine_first)) {
606 for (i = 0; i < tpp->nparams; i++) {
607 p = &tpp->params[i];
608 text_poke(p->addr, p->opcode, p->len);
610 smp_wmb(); /* Make sure other cpus see that this has run */
611 wrote_text = 1;
612 } else {
613 while (!wrote_text)
614 cpu_relax();
615 smp_mb(); /* Load wrote_text before following execution */
618 for (i = 0; i < tpp->nparams; i++) {
619 p = &tpp->params[i];
620 flush_icache_range((unsigned long)p->addr,
621 (unsigned long)p->addr + p->len);
624 return 0;
628 * text_poke_smp - Update instructions on a live kernel on SMP
629 * @addr: address to modify
630 * @opcode: source of the copy
631 * @len: length to copy
633 * Modify multi-byte instruction by using stop_machine() on SMP. This allows
634 * user to poke/set multi-byte text on SMP. Only non-NMI/MCE code modifying
635 * should be allowed, since stop_machine() does _not_ protect code against
636 * NMI and MCE.
638 * Note: Must be called under get_online_cpus() and text_mutex.
640 void *__kprobes text_poke_smp(void *addr, const void *opcode, size_t len)
642 struct text_poke_params tpp;
643 struct text_poke_param p;
645 p.addr = addr;
646 p.opcode = opcode;
647 p.len = len;
648 tpp.params = &p;
649 tpp.nparams = 1;
650 atomic_set(&stop_machine_first, 1);
651 wrote_text = 0;
652 /* Use __stop_machine() because the caller already got online_cpus. */
653 __stop_machine(stop_machine_text_poke, (void *)&tpp, cpu_online_mask);
654 return addr;
658 * text_poke_smp_batch - Update instructions on a live kernel on SMP
659 * @params: an array of text_poke parameters
660 * @n: the number of elements in params.
662 * Modify multi-byte instruction by using stop_machine() on SMP. Since the
663 * stop_machine() is heavy task, it is better to aggregate text_poke requests
664 * and do it once if possible.
666 * Note: Must be called under get_online_cpus() and text_mutex.
668 void __kprobes text_poke_smp_batch(struct text_poke_param *params, int n)
670 struct text_poke_params tpp = {.params = params, .nparams = n};
672 atomic_set(&stop_machine_first, 1);
673 wrote_text = 0;
674 stop_machine(stop_machine_text_poke, (void *)&tpp, NULL);
677 #if defined(CONFIG_DYNAMIC_FTRACE) || defined(HAVE_JUMP_LABEL)
679 #ifdef CONFIG_X86_64
680 unsigned char ideal_nop5[5] = { 0x66, 0x66, 0x66, 0x66, 0x90 };
681 #else
682 unsigned char ideal_nop5[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
683 #endif
685 void __init arch_init_ideal_nop5(void)
688 * There is no good nop for all x86 archs. This selection
689 * algorithm should be unified with the one in find_nop_table(),
690 * but this should be good enough for now.
692 * For cases other than the ones below, use the safe (as in
693 * always functional) defaults above.
695 #ifdef CONFIG_X86_64
696 /* Don't use these on 32 bits due to broken virtualizers */
697 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
698 memcpy(ideal_nop5, p6_nops[5], 5);
699 #endif
701 #endif