arch/x86/kernel/alternative.c

   1 #include <linux/module.h>
   2 #include <linux/sched.h>
   3 #include <linux/spinlock.h>
   4 #include <linux/list.h>
   5 #include <linux/kprobes.h>
   6 #include <linux/mm.h>
   7 #include <linux/vmalloc.h>
   8 #include <asm/alternative.h>
   9 #include <asm/sections.h>
  10 #include <asm/pgtable.h>
  11 #include <asm/mce.h>
  12 #include <asm/nmi.h>
  13 #include <asm/vsyscall.h>
  14
  15 #define MAX_PATCH_LEN (255-1)
  16
  17 #ifdef CONFIG_HOTPLUG_CPU
  18 static int smp_alt_once;
  19
  20 static int __init bootonly(char *str)
  21 {
  22         smp_alt_once = 1;
  23         return 1;
  24 }
  25 __setup("smp-alt-boot", bootonly);
  26 #else
  27 #define smp_alt_once 1
  28 #endif
  29
  30 static int debug_alternative;
  31
  32 static int __init debug_alt(char *str)
  33 {
  34         debug_alternative = 1;
  35         return 1;
  36 }
  37 __setup("debug-alternative", debug_alt);
  38
  39 static int noreplace_smp;
  40
  41 static int __init setup_noreplace_smp(char *str)
  42 {
  43         noreplace_smp = 1;
  44         return 1;
  45 }
  46 __setup("noreplace-smp", setup_noreplace_smp);
  47
  48 #ifdef CONFIG_PARAVIRT
  49 static int noreplace_paravirt = 0;
  50
  51 static int __init setup_noreplace_paravirt(char *str)
  52 {
  53         noreplace_paravirt = 1;
  54         return 1;
  55 }
  56 __setup("noreplace-paravirt", setup_noreplace_paravirt);
  57 #endif
  58
  59 #define DPRINTK(fmt, args...) if (debug_alternative) \
  60         printk(KERN_DEBUG fmt, args)
  61
  62 #ifdef GENERIC_NOP1
  63 /* Use inline assembly to define this because the nops are defined
  64    as inline assembly strings in the include files and we cannot
  65    get them easily into strings. */
  66 asm("\t.section .rodata, \"a\"\nintelnops: "
  67         GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6
  68         GENERIC_NOP7 GENERIC_NOP8);
  69 extern const unsigned char intelnops[];
  70 static const unsigned char *const intel_nops[ASM_NOP_MAX+1] = {
  71         NULL,
  72         intelnops,
  73         intelnops + 1,
  74         intelnops + 1 + 2,
  75         intelnops + 1 + 2 + 3,
  76         intelnops + 1 + 2 + 3 + 4,
  77         intelnops + 1 + 2 + 3 + 4 + 5,
  78         intelnops + 1 + 2 + 3 + 4 + 5 + 6,
  79         intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
  80 };
  81 #endif
  82
  83 #ifdef K8_NOP1
  84 asm("\t.section .rodata, \"a\"\nk8nops: "
  85         K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
  86         K8_NOP7 K8_NOP8);
  87 extern const unsigned char k8nops[];
  88 static const unsigned char *const k8_nops[ASM_NOP_MAX+1] = {
  89         NULL,
  90         k8nops,
  91         k8nops + 1,
  92         k8nops + 1 + 2,
  93         k8nops + 1 + 2 + 3,
  94         k8nops + 1 + 2 + 3 + 4,
  95         k8nops + 1 + 2 + 3 + 4 + 5,
  96         k8nops + 1 + 2 + 3 + 4 + 5 + 6,
  97         k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
  98 };
  99 #endif
 100
 101 #ifdef K7_NOP1
 102 asm("\t.section .rodata, \"a\"\nk7nops: "
 103         K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6
 104         K7_NOP7 K7_NOP8);
 105 extern const unsigned char k7nops[];
 106 static const unsigned char *const k7_nops[ASM_NOP_MAX+1] = {
 107         NULL,
 108         k7nops,
 109         k7nops + 1,
 110         k7nops + 1 + 2,
 111         k7nops + 1 + 2 + 3,
 112         k7nops + 1 + 2 + 3 + 4,
 113         k7nops + 1 + 2 + 3 + 4 + 5,
 114         k7nops + 1 + 2 + 3 + 4 + 5 + 6,
 115         k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
 116 };
 117 #endif
 118
 119 #ifdef P6_NOP1
 120 asm("\t.section .rodata, \"a\"\np6nops: "
 121         P6_NOP1 P6_NOP2 P6_NOP3 P6_NOP4 P6_NOP5 P6_NOP6
 122         P6_NOP7 P6_NOP8);
 123 extern const unsigned char p6nops[];
 124 static const unsigned char *const p6_nops[ASM_NOP_MAX+1] = {
 125         NULL,
 126         p6nops,
 127         p6nops + 1,
 128         p6nops + 1 + 2,
 129         p6nops + 1 + 2 + 3,
 130         p6nops + 1 + 2 + 3 + 4,
 131         p6nops + 1 + 2 + 3 + 4 + 5,
 132         p6nops + 1 + 2 + 3 + 4 + 5 + 6,
 133         p6nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
 134 };
 135 #endif
 136
 137 #ifdef CONFIG_X86_64
 138
 139 extern char __vsyscall_0;
 140 static inline const unsigned char*const * find_nop_table(void)
 141 {
 142         return boot_cpu_data.x86_vendor != X86_VENDOR_INTEL ||
 143                boot_cpu_data.x86 < 6 ? k8_nops : p6_nops;
 144 }
 145
 146 #else /* CONFIG_X86_64 */
 147
 148 static const struct nop {
 149         int cpuid;
 150         const unsigned char *const *noptable;
 151 } noptypes[] = {
 152         { X86_FEATURE_K8, k8_nops },
 153         { X86_FEATURE_K7, k7_nops },
 154         { X86_FEATURE_P4, p6_nops },
 155         { X86_FEATURE_P3, p6_nops },
 156         { -1, NULL }
 157 };
 158
 159 static const unsigned char*const * find_nop_table(void)
 160 {
 161         const unsigned char *const *noptable = intel_nops;
 162         int i;
 163
 164         for (i = 0; noptypes[i].cpuid >= 0; i++) {
 165                 if (boot_cpu_has(noptypes[i].cpuid)) {
 166                         noptable = noptypes[i].noptable;
 167                         break;
 168                 }
 169         }
 170         return noptable;
 171 }
 172
 173 #endif /* CONFIG_X86_64 */
 174
 175 /* Use this to add nops to a buffer, then text_poke the whole buffer. */
 176 static void add_nops(void *insns, unsigned int len)
 177 {
 178         const unsigned char *const *noptable = find_nop_table();
 179
 180         while (len > 0) {
 181                 unsigned int noplen = len;
 182                 if (noplen > ASM_NOP_MAX)
 183                         noplen = ASM_NOP_MAX;
 184                 memcpy(insns, noptable[noplen], noplen);
 185                 insns += noplen;
 186                 len -= noplen;
 187         }
 188 }
 189
 190 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
 191 extern u8 *__smp_locks[], *__smp_locks_end[];
 192
 193 /* Replace instructions with better alternatives for this CPU type.
 194    This runs before SMP is initialized to avoid SMP problems with
 195    self modifying code. This implies that assymetric systems where
 196    APs have less capabilities than the boot processor are not handled.
 197    Tough. Make sure you disable such features by hand. */
 198
 199 void apply_alternatives(struct alt_instr *start, struct alt_instr *end)
 200 {
 201         struct alt_instr *a;
 202         char insnbuf[MAX_PATCH_LEN];
 203
 204         DPRINTK("%s: alt table %p -> %p\n", __FUNCTION__, start, end);
 205         for (a = start; a < end; a++) {
 206                 u8 *instr = a->instr;
 207                 BUG_ON(a->replacementlen > a->instrlen);
 208                 BUG_ON(a->instrlen > sizeof(insnbuf));
 209                 if (!boot_cpu_has(a->cpuid))
 210                         continue;
 211 #ifdef CONFIG_X86_64
 212                 /* vsyscall code is not mapped yet. resolve it manually. */
 213                 if (instr >= (u8 *)VSYSCALL_START && instr < (u8*)VSYSCALL_END) {
 214                         instr = __va(instr - (u8*)VSYSCALL_START + (u8*)__pa_symbol(&__vsyscall_0));
 215                         DPRINTK("%s: vsyscall fixup: %p => %p\n",
 216                                 __FUNCTION__, a->instr, instr);
 217                 }
 218 #endif
 219                 memcpy(insnbuf, a->replacement, a->replacementlen);
 220                 add_nops(insnbuf + a->replacementlen,
 221                          a->instrlen - a->replacementlen);
 222                 text_poke(instr, insnbuf, a->instrlen);
 223         }
 224 }
 225
 226 #ifdef CONFIG_SMP
 227
 228 static void alternatives_smp_lock(u8 **start, u8 **end, u8 *text, u8 *text_end)
 229 {
 230         u8 **ptr;
 231
 232         for (ptr = start; ptr < end; ptr++) {
 233                 if (*ptr < text)
 234                         continue;
 235                 if (*ptr > text_end)
 236                         continue;
 237                 text_poke(*ptr, ((unsigned char []){0xf0}), 1); /* add lock prefix */
 238         };
 239 }
 240
 241 static void alternatives_smp_unlock(u8 **start, u8 **end, u8 *text, u8 *text_end)
 242 {
 243         u8 **ptr;
 244         char insn[1];
 245
 246         if (noreplace_smp)
 247                 return;
 248
 249         add_nops(insn, 1);
 250         for (ptr = start; ptr < end; ptr++) {
 251                 if (*ptr < text)
 252                         continue;
 253                 if (*ptr > text_end)
 254                         continue;
 255                 text_poke(*ptr, insn, 1);
 256         };
 257 }
 258
 259 struct smp_alt_module {
 260         /* what is this ??? */
 261         struct module   *mod;
 262         char            *name;
 263
 264         /* ptrs to lock prefixes */
 265         u8              **locks;
 266         u8              **locks_end;
 267
 268         /* .text segment, needed to avoid patching init code ;) */
 269         u8              *text;
 270         u8              *text_end;
 271
 272         struct list_head next;
 273 };
 274 static LIST_HEAD(smp_alt_modules);
 275 static DEFINE_SPINLOCK(smp_alt);
 276
 277 void alternatives_smp_module_add(struct module *mod, char *name,
 278                                  void *locks, void *locks_end,
 279                                  void *text,  void *text_end)
 280 {
 281         struct smp_alt_module *smp;
 282         unsigned long flags;
 283
 284         if (noreplace_smp)
 285                 return;
 286
 287         if (smp_alt_once) {
 288                 if (boot_cpu_has(X86_FEATURE_UP))
 289                         alternatives_smp_unlock(locks, locks_end,
 290                                                 text, text_end);
 291                 return;
 292         }
 293
 294         smp = kzalloc(sizeof(*smp), GFP_KERNEL);
 295         if (NULL == smp)
 296                 return; /* we'll run the (safe but slow) SMP code then ... */
 297
 298         smp->mod        = mod;
 299         smp->name       = name;
 300         smp->locks      = locks;
 301         smp->locks_end  = locks_end;
 302         smp->text       = text;
 303         smp->text_end   = text_end;
 304         DPRINTK("%s: locks %p -> %p, text %p -> %p, name %s\n",
 305                 __FUNCTION__, smp->locks, smp->locks_end,
 306                 smp->text, smp->text_end, smp->name);
 307
 308         spin_lock_irqsave(&smp_alt, flags);
 309         list_add_tail(&smp->next, &smp_alt_modules);
 310         if (boot_cpu_has(X86_FEATURE_UP))
 311                 alternatives_smp_unlock(smp->locks, smp->locks_end,
 312                                         smp->text, smp->text_end);
 313         spin_unlock_irqrestore(&smp_alt, flags);
 314 }
 315
 316 void alternatives_smp_module_del(struct module *mod)
 317 {
 318         struct smp_alt_module *item;
 319         unsigned long flags;
 320
 321         if (smp_alt_once || noreplace_smp)
 322                 return;
 323
 324         spin_lock_irqsave(&smp_alt, flags);
 325         list_for_each_entry(item, &smp_alt_modules, next) {
 326                 if (mod != item->mod)
 327                         continue;
 328                 list_del(&item->next);
 329                 spin_unlock_irqrestore(&smp_alt, flags);
 330                 DPRINTK("%s: %s\n", __FUNCTION__, item->name);
 331                 kfree(item);
 332                 return;
 333         }
 334         spin_unlock_irqrestore(&smp_alt, flags);
 335 }
 336
 337 void alternatives_smp_switch(int smp)
 338 {
 339         struct smp_alt_module *mod;
 340         unsigned long flags;
 341
 342 #ifdef CONFIG_LOCKDEP
 343         /*
 344          * A not yet fixed binutils section handling bug prevents
 345          * alternatives-replacement from working reliably, so turn
 346          * it off:
 347          */
 348         printk("lockdep: not fixing up alternatives.\n");
 349         return;
 350 #endif
 351
 352         if (noreplace_smp || smp_alt_once)
 353                 return;
 354         BUG_ON(!smp && (num_online_cpus() > 1));
 355
 356         spin_lock_irqsave(&smp_alt, flags);
 357         if (smp) {
 358                 printk(KERN_INFO "SMP alternatives: switching to SMP code\n");
 359                 clear_bit(X86_FEATURE_UP, boot_cpu_data.x86_capability);
 360                 clear_bit(X86_FEATURE_UP, cpu_data(0).x86_capability);
 361                 list_for_each_entry(mod, &smp_alt_modules, next)
 362                         alternatives_smp_lock(mod->locks, mod->locks_end,
 363                                               mod->text, mod->text_end);
 364         } else {
 365                 printk(KERN_INFO "SMP alternatives: switching to UP code\n");
 366                 set_bit(X86_FEATURE_UP, boot_cpu_data.x86_capability);
 367                 set_bit(X86_FEATURE_UP, cpu_data(0).x86_capability);
 368                 list_for_each_entry(mod, &smp_alt_modules, next)
 369                         alternatives_smp_unlock(mod->locks, mod->locks_end,
 370                                                 mod->text, mod->text_end);
 371         }
 372         spin_unlock_irqrestore(&smp_alt, flags);
 373 }
 374
 375 #endif
 376
 377 #ifdef CONFIG_PARAVIRT
 378 void apply_paravirt(struct paravirt_patch_site *start,
 379                     struct paravirt_patch_site *end)
 380 {
 381         struct paravirt_patch_site *p;
 382         char insnbuf[MAX_PATCH_LEN];
 383
 384         if (noreplace_paravirt)
 385                 return;
 386
 387         for (p = start; p < end; p++) {
 388                 unsigned int used;
 389
 390                 BUG_ON(p->len > MAX_PATCH_LEN);
 391                 /* prep the buffer with the original instructions */
 392                 memcpy(insnbuf, p->instr, p->len);
 393                 used = pv_init_ops.patch(p->instrtype, p->clobbers, insnbuf,
 394                                          (unsigned long)p->instr, p->len);
 395
 396                 BUG_ON(used > p->len);
 397
 398                 /* Pad the rest with nops */
 399                 add_nops(insnbuf + used, p->len - used);
 400                 text_poke(p->instr, insnbuf, p->len);
 401         }
 402 }
 403 extern struct paravirt_patch_site __start_parainstructions[],
 404         __stop_parainstructions[];
 405 #endif  /* CONFIG_PARAVIRT */
 406
 407 void __init alternative_instructions(void)
 408 {
 409         unsigned long flags;
 410
 411         /* The patching is not fully atomic, so try to avoid local interruptions
 412            that might execute the to be patched code.
 413            Other CPUs are not running. */
 414         stop_nmi();
 415 #ifdef CONFIG_X86_MCE
 416         stop_mce();
 417 #endif
 418
 419         local_irq_save(flags);
 420         apply_alternatives(__alt_instructions, __alt_instructions_end);
 421
 422         /* switch to patch-once-at-boottime-only mode and free the
 423          * tables in case we know the number of CPUs will never ever
 424          * change */
 425 #ifdef CONFIG_HOTPLUG_CPU
 426         if (num_possible_cpus() < 2)
 427                 smp_alt_once = 1;
 428 #endif
 429
 430 #ifdef CONFIG_SMP
 431         if (smp_alt_once) {
 432                 if (1 == num_possible_cpus()) {
 433                         printk(KERN_INFO "SMP alternatives: switching to UP code\n");
 434                         set_bit(X86_FEATURE_UP, boot_cpu_data.x86_capability);
 435                         set_bit(X86_FEATURE_UP, cpu_data(0).x86_capability);
 436                         alternatives_smp_unlock(__smp_locks, __smp_locks_end,
 437                                                 _text, _etext);
 438                 }
 439         } else {
 440                 alternatives_smp_module_add(NULL, "core kernel",
 441                                             __smp_locks, __smp_locks_end,
 442                                             _text, _etext);
 443                 alternatives_smp_switch(0);
 444         }
 445 #endif
 446         apply_paravirt(__parainstructions, __parainstructions_end);
 447         local_irq_restore(flags);
 448
 449         if (smp_alt_once)
 450                 free_init_pages("SMP alternatives",
 451                                 (unsigned long)__smp_locks,
 452                                 (unsigned long)__smp_locks_end);
 453
 454         restart_nmi();
 455 #ifdef CONFIG_X86_MCE
 456         restart_mce();
 457 #endif
 458 }
 459
 460 /*
 461  * Warning:
 462  * When you use this code to patch more than one byte of an instruction
 463  * you need to make sure that other CPUs cannot execute this code in parallel.
 464  * Also no thread must be currently preempted in the middle of these instructions.
 465  * And on the local CPU you need to be protected again NMI or MCE handlers
 466  * seeing an inconsistent instruction while you patch.
 467  */
 468 void __kprobes text_poke(void *addr, unsigned char *opcode, int len)
 469 {
 470         memcpy(addr, opcode, len);
 471         sync_core();
 472         /* Could also do a CLFLUSH here to speed up CPU recovery; but
 473            that causes hangs on some VIA CPUs. */
 474 }