1 #include <linux/module.h>
2 #include <linux/sched.h>
3 #include <linux/mutex.h>
4 #include <linux/list.h>
5 #include <linux/kprobes.h>
7 #include <linux/vmalloc.h>
8 #include <asm/alternative.h>
9 #include <asm/sections.h>
10 #include <asm/pgtable.h>
13 #include <asm/vsyscall.h>
14 #include <asm/cacheflush.h>
17 #define MAX_PATCH_LEN (255-1)
19 #ifdef CONFIG_HOTPLUG_CPU
20 static int smp_alt_once;
22 static int __init bootonly(char *str)
27 __setup("smp-alt-boot", bootonly);
29 #define smp_alt_once 1
32 static int debug_alternative;
34 static int __init debug_alt(char *str)
36 debug_alternative = 1;
39 __setup("debug-alternative", debug_alt);
41 static int noreplace_smp;
43 static int __init setup_noreplace_smp(char *str)
48 __setup("noreplace-smp", setup_noreplace_smp);
50 #ifdef CONFIG_PARAVIRT
51 static int noreplace_paravirt = 0;
53 static int __init setup_noreplace_paravirt(char *str)
55 noreplace_paravirt = 1;
58 __setup("noreplace-paravirt", setup_noreplace_paravirt);
61 #define DPRINTK(fmt, args...) if (debug_alternative) \
62 printk(KERN_DEBUG fmt, args)
65 /* Use inline assembly to define this because the nops are defined
66 as inline assembly strings in the include files and we cannot
67 get them easily into strings. */
68 asm("\t.section .rodata, \"a\"\nintelnops: "
69 GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6
70 GENERIC_NOP7 GENERIC_NOP8
72 extern const unsigned char intelnops[];
73 static const unsigned char *const intel_nops[ASM_NOP_MAX+1] = {
78 intelnops + 1 + 2 + 3,
79 intelnops + 1 + 2 + 3 + 4,
80 intelnops + 1 + 2 + 3 + 4 + 5,
81 intelnops + 1 + 2 + 3 + 4 + 5 + 6,
82 intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
87 asm("\t.section .rodata, \"a\"\nk8nops: "
88 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
91 extern const unsigned char k8nops[];
92 static const unsigned char *const k8_nops[ASM_NOP_MAX+1] = {
98 k8nops + 1 + 2 + 3 + 4,
99 k8nops + 1 + 2 + 3 + 4 + 5,
100 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
101 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
106 asm("\t.section .rodata, \"a\"\nk7nops: "
107 K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6
110 extern const unsigned char k7nops[];
111 static const unsigned char *const k7_nops[ASM_NOP_MAX+1] = {
117 k7nops + 1 + 2 + 3 + 4,
118 k7nops + 1 + 2 + 3 + 4 + 5,
119 k7nops + 1 + 2 + 3 + 4 + 5 + 6,
120 k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
125 asm("\t.section .rodata, \"a\"\np6nops: "
126 P6_NOP1 P6_NOP2 P6_NOP3 P6_NOP4 P6_NOP5 P6_NOP6
129 extern const unsigned char p6nops[];
130 static const unsigned char *const p6_nops[ASM_NOP_MAX+1] = {
136 p6nops + 1 + 2 + 3 + 4,
137 p6nops + 1 + 2 + 3 + 4 + 5,
138 p6nops + 1 + 2 + 3 + 4 + 5 + 6,
139 p6nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
145 extern char __vsyscall_0;
146 const unsigned char *const *find_nop_table(void)
148 return boot_cpu_data.x86_vendor != X86_VENDOR_INTEL ||
149 boot_cpu_data.x86 < 6 ? k8_nops : p6_nops;
152 #else /* CONFIG_X86_64 */
154 static const struct nop {
156 const unsigned char *const *noptable;
158 { X86_FEATURE_K8, k8_nops },
159 { X86_FEATURE_K7, k7_nops },
160 { X86_FEATURE_P4, p6_nops },
161 { X86_FEATURE_P3, p6_nops },
165 const unsigned char *const *find_nop_table(void)
167 const unsigned char *const *noptable = intel_nops;
170 for (i = 0; noptypes[i].cpuid >= 0; i++) {
171 if (boot_cpu_has(noptypes[i].cpuid)) {
172 noptable = noptypes[i].noptable;
179 #endif /* CONFIG_X86_64 */
181 /* Use this to add nops to a buffer, then text_poke the whole buffer. */
182 void add_nops(void *insns, unsigned int len)
184 const unsigned char *const *noptable = find_nop_table();
187 unsigned int noplen = len;
188 if (noplen > ASM_NOP_MAX)
189 noplen = ASM_NOP_MAX;
190 memcpy(insns, noptable[noplen], noplen);
195 EXPORT_SYMBOL_GPL(add_nops);
197 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
198 extern u8 *__smp_locks[], *__smp_locks_end[];
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 apply_alternatives(struct alt_instr *start, struct alt_instr *end)
209 char insnbuf[MAX_PATCH_LEN];
211 DPRINTK("%s: alt table %p -> %p\n", __func__, start, end);
212 for (a = start; a < end; a++) {
213 u8 *instr = a->instr;
214 BUG_ON(a->replacementlen > a->instrlen);
215 BUG_ON(a->instrlen > sizeof(insnbuf));
216 if (!boot_cpu_has(a->cpuid))
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",
223 __func__, a->instr, instr);
226 memcpy(insnbuf, a->replacement, a->replacementlen);
227 add_nops(insnbuf + a->replacementlen,
228 a->instrlen - a->replacementlen);
229 text_poke_early(instr, insnbuf, a->instrlen);
235 static void alternatives_smp_lock(u8 **start, u8 **end, u8 *text, u8 *text_end)
239 for (ptr = start; ptr < end; ptr++) {
244 text_poke(*ptr, ((unsigned char []){0xf0}), 1); /* add lock prefix */
248 static void alternatives_smp_unlock(u8 **start, u8 **end, u8 *text, u8 *text_end)
257 for (ptr = start; ptr < end; ptr++) {
262 text_poke(*ptr, insn, 1);
266 struct smp_alt_module {
267 /* what is this ??? */
271 /* ptrs to lock prefixes */
275 /* .text segment, needed to avoid patching init code ;) */
279 struct list_head next;
281 static LIST_HEAD(smp_alt_modules);
282 static DEFINE_MUTEX(smp_alt);
283 static int smp_mode = 1; /* protected by smp_alt */
285 void alternatives_smp_module_add(struct module *mod, char *name,
286 void *locks, void *locks_end,
287 void *text, void *text_end)
289 struct smp_alt_module *smp;
295 if (boot_cpu_has(X86_FEATURE_UP))
296 alternatives_smp_unlock(locks, locks_end,
301 smp = kzalloc(sizeof(*smp), GFP_KERNEL);
303 return; /* we'll run the (safe but slow) SMP code then ... */
308 smp->locks_end = locks_end;
310 smp->text_end = text_end;
311 DPRINTK("%s: locks %p -> %p, text %p -> %p, name %s\n",
312 __func__, smp->locks, smp->locks_end,
313 smp->text, smp->text_end, smp->name);
315 mutex_lock(&smp_alt);
316 list_add_tail(&smp->next, &smp_alt_modules);
317 if (boot_cpu_has(X86_FEATURE_UP))
318 alternatives_smp_unlock(smp->locks, smp->locks_end,
319 smp->text, smp->text_end);
320 mutex_unlock(&smp_alt);
323 void alternatives_smp_module_del(struct module *mod)
325 struct smp_alt_module *item;
327 if (smp_alt_once || noreplace_smp)
330 mutex_lock(&smp_alt);
331 list_for_each_entry(item, &smp_alt_modules, next) {
332 if (mod != item->mod)
334 list_del(&item->next);
335 mutex_unlock(&smp_alt);
336 DPRINTK("%s: %s\n", __func__, item->name);
340 mutex_unlock(&smp_alt);
343 void alternatives_smp_switch(int smp)
345 struct smp_alt_module *mod;
347 #ifdef CONFIG_LOCKDEP
349 * Older binutils section handling bug prevented
350 * alternatives-replacement from working reliably.
352 * If this still occurs then you should see a hang
353 * or crash shortly after this line:
355 printk("lockdep: fixing up alternatives.\n");
358 if (noreplace_smp || smp_alt_once)
360 BUG_ON(!smp && (num_online_cpus() > 1));
362 mutex_lock(&smp_alt);
365 * Avoid unnecessary switches because it forces JIT based VMs to
366 * throw away all cached translations, which can be quite costly.
368 if (smp == smp_mode) {
371 printk(KERN_INFO "SMP alternatives: switching to SMP code\n");
372 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
373 clear_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
374 list_for_each_entry(mod, &smp_alt_modules, next)
375 alternatives_smp_lock(mod->locks, mod->locks_end,
376 mod->text, mod->text_end);
378 printk(KERN_INFO "SMP alternatives: switching to UP code\n");
379 set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
380 set_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
381 list_for_each_entry(mod, &smp_alt_modules, next)
382 alternatives_smp_unlock(mod->locks, mod->locks_end,
383 mod->text, mod->text_end);
386 mutex_unlock(&smp_alt);
391 #ifdef CONFIG_PARAVIRT
392 void apply_paravirt(struct paravirt_patch_site *start,
393 struct paravirt_patch_site *end)
395 struct paravirt_patch_site *p;
396 char insnbuf[MAX_PATCH_LEN];
398 if (noreplace_paravirt)
401 for (p = start; p < end; p++) {
404 BUG_ON(p->len > MAX_PATCH_LEN);
405 /* prep the buffer with the original instructions */
406 memcpy(insnbuf, p->instr, p->len);
407 used = pv_init_ops.patch(p->instrtype, p->clobbers, insnbuf,
408 (unsigned long)p->instr, p->len);
410 BUG_ON(used > p->len);
412 /* Pad the rest with nops */
413 add_nops(insnbuf + used, p->len - used);
414 text_poke_early(p->instr, insnbuf, p->len);
417 extern struct paravirt_patch_site __start_parainstructions[],
418 __stop_parainstructions[];
419 #endif /* CONFIG_PARAVIRT */
421 void __init alternative_instructions(void)
423 /* The patching is not fully atomic, so try to avoid local interruptions
424 that might execute the to be patched code.
425 Other CPUs are not running. */
427 #ifdef CONFIG_X86_MCE
431 apply_alternatives(__alt_instructions, __alt_instructions_end);
433 /* switch to patch-once-at-boottime-only mode and free the
434 * tables in case we know the number of CPUs will never ever
436 #ifdef CONFIG_HOTPLUG_CPU
437 if (num_possible_cpus() < 2)
443 if (1 == num_possible_cpus()) {
444 printk(KERN_INFO "SMP alternatives: switching to UP code\n");
445 set_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
446 set_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
448 alternatives_smp_unlock(__smp_locks, __smp_locks_end,
452 alternatives_smp_module_add(NULL, "core kernel",
453 __smp_locks, __smp_locks_end,
456 /* Only switch to UP mode if we don't immediately boot others */
457 if (num_possible_cpus() == 1 || setup_max_cpus <= 1)
458 alternatives_smp_switch(0);
461 apply_paravirt(__parainstructions, __parainstructions_end);
464 free_init_pages("SMP alternatives",
465 (unsigned long)__smp_locks,
466 (unsigned long)__smp_locks_end);
469 #ifdef CONFIG_X86_MCE
475 * text_poke_early - Update instructions on a live kernel at boot time
476 * @addr: address to modify
477 * @opcode: source of the copy
478 * @len: length to copy
480 * When you use this code to patch more than one byte of an instruction
481 * you need to make sure that other CPUs cannot execute this code in parallel.
482 * Also no thread must be currently preempted in the middle of these
483 * instructions. And on the local CPU you need to be protected again NMI or MCE
484 * handlers seeing an inconsistent instruction while you patch.
486 void *text_poke_early(void *addr, const void *opcode, size_t len)
489 local_irq_save(flags);
490 memcpy(addr, opcode, len);
491 local_irq_restore(flags);
493 /* Could also do a CLFLUSH here to speed up CPU recovery; but
494 that causes hangs on some VIA CPUs. */
499 * text_poke - Update instructions on a live kernel
500 * @addr: address to modify
501 * @opcode: source of the copy
502 * @len: length to copy
504 * Only atomic text poke/set should be allowed when not doing early patching.
505 * It means the size must be writable atomically and the address must be aligned
506 * in a way that permits an atomic write. It also makes sure we fit on a single
509 void *__kprobes text_poke(void *addr, const void *opcode, size_t len)
514 struct page *pages[2];
517 if (!core_kernel_text((unsigned long)addr)) {
518 pages[0] = vmalloc_to_page(addr);
519 pages[1] = vmalloc_to_page(addr + PAGE_SIZE);
521 pages[0] = virt_to_page(addr);
522 WARN_ON(!PageReserved(pages[0]));
523 pages[1] = virt_to_page(addr + PAGE_SIZE);
528 vaddr = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL);
530 local_irq_save(flags);
531 memcpy(&vaddr[(unsigned long)addr & ~PAGE_MASK], opcode, len);
532 local_irq_restore(flags);
535 /* Could also do a CLFLUSH here to speed up CPU recovery; but
536 that causes hangs on some VIA CPUs. */
537 for (i = 0; i < len; i++)
538 BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]);