processor.h 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841
  1. #ifndef _ASM_X86_PROCESSOR_H
  2. #define _ASM_X86_PROCESSOR_H
  3. #include <asm/processor-flags.h>
  4. /* Forward declaration, a strange C thing */
  5. struct task_struct;
  6. struct mm_struct;
  7. struct vm86;
  8. #include <asm/math_emu.h>
  9. #include <asm/segment.h>
  10. #include <asm/types.h>
  11. #include <uapi/asm/sigcontext.h>
  12. #include <asm/current.h>
  13. #include <asm/cpufeatures.h>
  14. #include <asm/page.h>
  15. #include <asm/pgtable_types.h>
  16. #include <asm/percpu.h>
  17. #include <asm/msr.h>
  18. #include <asm/desc_defs.h>
  19. #include <asm/nops.h>
  20. #include <asm/special_insns.h>
  21. #include <asm/fpu/types.h>
  22. #include <linux/personality.h>
  23. #include <linux/cache.h>
  24. #include <linux/threads.h>
  25. #include <linux/math64.h>
  26. #include <linux/err.h>
  27. #include <linux/irqflags.h>
  28. /*
  29. * We handle most unaligned accesses in hardware. On the other hand
  30. * unaligned DMA can be quite expensive on some Nehalem processors.
  31. *
  32. * Based on this we disable the IP header alignment in network drivers.
  33. */
  34. #define NET_IP_ALIGN 0
  35. #define HBP_NUM 4
  36. /*
  37. * Default implementation of macro that returns current
  38. * instruction pointer ("program counter").
  39. */
  40. static inline void *current_text_addr(void)
  41. {
  42. void *pc;
  43. asm volatile("mov $1f, %0; 1:":"=r" (pc));
  44. return pc;
  45. }
  46. /*
  47. * These alignment constraints are for performance in the vSMP case,
  48. * but in the task_struct case we must also meet hardware imposed
  49. * alignment requirements of the FPU state:
  50. */
  51. #ifdef CONFIG_X86_VSMP
  52. # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
  53. # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
  54. #else
  55. # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
  56. # define ARCH_MIN_MMSTRUCT_ALIGN 0
  57. #endif
  58. enum tlb_infos {
  59. ENTRIES,
  60. NR_INFO
  61. };
  62. extern u16 __read_mostly tlb_lli_4k[NR_INFO];
  63. extern u16 __read_mostly tlb_lli_2m[NR_INFO];
  64. extern u16 __read_mostly tlb_lli_4m[NR_INFO];
  65. extern u16 __read_mostly tlb_lld_4k[NR_INFO];
  66. extern u16 __read_mostly tlb_lld_2m[NR_INFO];
  67. extern u16 __read_mostly tlb_lld_4m[NR_INFO];
  68. extern u16 __read_mostly tlb_lld_1g[NR_INFO];
  69. /*
  70. * CPU type and hardware bug flags. Kept separately for each CPU.
  71. * Members of this structure are referenced in head.S, so think twice
  72. * before touching them. [mj]
  73. */
  74. struct cpuinfo_x86 {
  75. __u8 x86; /* CPU family */
  76. __u8 x86_vendor; /* CPU vendor */
  77. __u8 x86_model;
  78. __u8 x86_mask;
  79. #ifdef CONFIG_X86_32
  80. char wp_works_ok; /* It doesn't on 386's */
  81. /* Problems on some 486Dx4's and old 386's: */
  82. char rfu;
  83. char pad0;
  84. char pad1;
  85. #else
  86. /* Number of 4K pages in DTLB/ITLB combined(in pages): */
  87. int x86_tlbsize;
  88. #endif
  89. __u8 x86_virt_bits;
  90. __u8 x86_phys_bits;
  91. /* CPUID returned core id bits: */
  92. __u8 x86_coreid_bits;
  93. /* Max extended CPUID function supported: */
  94. __u32 extended_cpuid_level;
  95. /* Maximum supported CPUID level, -1=no CPUID: */
  96. int cpuid_level;
  97. __u32 x86_capability[NCAPINTS + NBUGINTS];
  98. char x86_vendor_id[16];
  99. char x86_model_id[64];
  100. /* in KB - valid for CPUS which support this call: */
  101. int x86_cache_size;
  102. int x86_cache_alignment; /* In bytes */
  103. /* Cache QoS architectural values: */
  104. int x86_cache_max_rmid; /* max index */
  105. int x86_cache_occ_scale; /* scale to bytes */
  106. int x86_power;
  107. unsigned long loops_per_jiffy;
  108. /* cpuid returned max cores value: */
  109. u16 x86_max_cores;
  110. u16 apicid;
  111. u16 initial_apicid;
  112. u16 x86_clflush_size;
  113. /* number of cores as seen by the OS: */
  114. u16 booted_cores;
  115. /* Physical processor id: */
  116. u16 phys_proc_id;
  117. /* Core id: */
  118. u16 cpu_core_id;
  119. /* Compute unit id */
  120. u8 compute_unit_id;
  121. /* Index into per_cpu list: */
  122. u16 cpu_index;
  123. u32 microcode;
  124. };
  125. #define X86_VENDOR_INTEL 0
  126. #define X86_VENDOR_CYRIX 1
  127. #define X86_VENDOR_AMD 2
  128. #define X86_VENDOR_UMC 3
  129. #define X86_VENDOR_CENTAUR 5
  130. #define X86_VENDOR_TRANSMETA 7
  131. #define X86_VENDOR_NSC 8
  132. #define X86_VENDOR_NUM 9
  133. #define X86_VENDOR_UNKNOWN 0xff
  134. /*
  135. * capabilities of CPUs
  136. */
  137. extern struct cpuinfo_x86 boot_cpu_data;
  138. extern struct cpuinfo_x86 new_cpu_data;
  139. extern struct tss_struct doublefault_tss;
  140. extern __u32 cpu_caps_cleared[NCAPINTS];
  141. extern __u32 cpu_caps_set[NCAPINTS];
  142. #ifdef CONFIG_SMP
  143. DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
  144. #define cpu_data(cpu) per_cpu(cpu_info, cpu)
  145. #else
  146. #define cpu_info boot_cpu_data
  147. #define cpu_data(cpu) boot_cpu_data
  148. #endif
  149. extern const struct seq_operations cpuinfo_op;
  150. #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
  151. extern void cpu_detect(struct cpuinfo_x86 *c);
  152. extern void early_cpu_init(void);
  153. extern void identify_boot_cpu(void);
  154. extern void identify_secondary_cpu(struct cpuinfo_x86 *);
  155. extern void print_cpu_info(struct cpuinfo_x86 *);
  156. void print_cpu_msr(struct cpuinfo_x86 *);
  157. extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
  158. extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
  159. extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
  160. extern void detect_extended_topology(struct cpuinfo_x86 *c);
  161. extern void detect_ht(struct cpuinfo_x86 *c);
  162. #ifdef CONFIG_X86_32
  163. extern int have_cpuid_p(void);
  164. #else
  165. static inline int have_cpuid_p(void)
  166. {
  167. return 1;
  168. }
  169. #endif
  170. static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
  171. unsigned int *ecx, unsigned int *edx)
  172. {
  173. /* ecx is often an input as well as an output. */
  174. asm volatile("cpuid"
  175. : "=a" (*eax),
  176. "=b" (*ebx),
  177. "=c" (*ecx),
  178. "=d" (*edx)
  179. : "0" (*eax), "2" (*ecx)
  180. : "memory");
  181. }
  182. static inline void load_cr3(pgd_t *pgdir)
  183. {
  184. write_cr3(__pa(pgdir));
  185. }
  186. #ifdef CONFIG_X86_32
  187. /* This is the TSS defined by the hardware. */
  188. struct x86_hw_tss {
  189. unsigned short back_link, __blh;
  190. unsigned long sp0;
  191. unsigned short ss0, __ss0h;
  192. unsigned long sp1;
  193. /*
  194. * We don't use ring 1, so ss1 is a convenient scratch space in
  195. * the same cacheline as sp0. We use ss1 to cache the value in
  196. * MSR_IA32_SYSENTER_CS. When we context switch
  197. * MSR_IA32_SYSENTER_CS, we first check if the new value being
  198. * written matches ss1, and, if it's not, then we wrmsr the new
  199. * value and update ss1.
  200. *
  201. * The only reason we context switch MSR_IA32_SYSENTER_CS is
  202. * that we set it to zero in vm86 tasks to avoid corrupting the
  203. * stack if we were to go through the sysenter path from vm86
  204. * mode.
  205. */
  206. unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
  207. unsigned short __ss1h;
  208. unsigned long sp2;
  209. unsigned short ss2, __ss2h;
  210. unsigned long __cr3;
  211. unsigned long ip;
  212. unsigned long flags;
  213. unsigned long ax;
  214. unsigned long cx;
  215. unsigned long dx;
  216. unsigned long bx;
  217. unsigned long sp;
  218. unsigned long bp;
  219. unsigned long si;
  220. unsigned long di;
  221. unsigned short es, __esh;
  222. unsigned short cs, __csh;
  223. unsigned short ss, __ssh;
  224. unsigned short ds, __dsh;
  225. unsigned short fs, __fsh;
  226. unsigned short gs, __gsh;
  227. unsigned short ldt, __ldth;
  228. unsigned short trace;
  229. unsigned short io_bitmap_base;
  230. } __attribute__((packed));
  231. #else
  232. struct x86_hw_tss {
  233. u32 reserved1;
  234. u64 sp0;
  235. u64 sp1;
  236. u64 sp2;
  237. u64 reserved2;
  238. u64 ist[7];
  239. u32 reserved3;
  240. u32 reserved4;
  241. u16 reserved5;
  242. u16 io_bitmap_base;
  243. } __attribute__((packed)) ____cacheline_aligned;
  244. #endif
  245. /*
  246. * IO-bitmap sizes:
  247. */
  248. #define IO_BITMAP_BITS 65536
  249. #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
  250. #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
  251. #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
  252. #define INVALID_IO_BITMAP_OFFSET 0x8000
  253. struct tss_struct {
  254. /*
  255. * The hardware state:
  256. */
  257. struct x86_hw_tss x86_tss;
  258. /*
  259. * The extra 1 is there because the CPU will access an
  260. * additional byte beyond the end of the IO permission
  261. * bitmap. The extra byte must be all 1 bits, and must
  262. * be within the limit.
  263. */
  264. unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
  265. /*
  266. * Space for the temporary SYSENTER stack:
  267. */
  268. unsigned long SYSENTER_stack[64];
  269. } ____cacheline_aligned;
  270. DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss);
  271. #ifdef CONFIG_X86_32
  272. DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
  273. #endif
  274. /*
  275. * Save the original ist values for checking stack pointers during debugging
  276. */
  277. struct orig_ist {
  278. unsigned long ist[7];
  279. };
  280. #ifdef CONFIG_X86_64
  281. DECLARE_PER_CPU(struct orig_ist, orig_ist);
  282. union irq_stack_union {
  283. char irq_stack[IRQ_STACK_SIZE];
  284. /*
  285. * GCC hardcodes the stack canary as %gs:40. Since the
  286. * irq_stack is the object at %gs:0, we reserve the bottom
  287. * 48 bytes of the irq stack for the canary.
  288. */
  289. struct {
  290. char gs_base[40];
  291. unsigned long stack_canary;
  292. };
  293. };
  294. DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
  295. DECLARE_INIT_PER_CPU(irq_stack_union);
  296. DECLARE_PER_CPU(char *, irq_stack_ptr);
  297. DECLARE_PER_CPU(unsigned int, irq_count);
  298. extern asmlinkage void ignore_sysret(void);
  299. #else /* X86_64 */
  300. #ifdef CONFIG_CC_STACKPROTECTOR
  301. /*
  302. * Make sure stack canary segment base is cached-aligned:
  303. * "For Intel Atom processors, avoid non zero segment base address
  304. * that is not aligned to cache line boundary at all cost."
  305. * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
  306. */
  307. struct stack_canary {
  308. char __pad[20]; /* canary at %gs:20 */
  309. unsigned long canary;
  310. };
  311. DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
  312. #endif
  313. /*
  314. * per-CPU IRQ handling stacks
  315. */
  316. struct irq_stack {
  317. u32 stack[THREAD_SIZE/sizeof(u32)];
  318. } __aligned(THREAD_SIZE);
  319. DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
  320. DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
  321. #endif /* X86_64 */
  322. extern unsigned int xstate_size;
  323. struct perf_event;
  324. struct thread_struct {
  325. /* Cached TLS descriptors: */
  326. struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
  327. unsigned long sp0;
  328. unsigned long sp;
  329. #ifdef CONFIG_X86_32
  330. unsigned long sysenter_cs;
  331. #else
  332. unsigned short es;
  333. unsigned short ds;
  334. unsigned short fsindex;
  335. unsigned short gsindex;
  336. #endif
  337. #ifdef CONFIG_X86_32
  338. unsigned long ip;
  339. #endif
  340. #ifdef CONFIG_X86_64
  341. unsigned long fs;
  342. #endif
  343. unsigned long gs;
  344. /* Save middle states of ptrace breakpoints */
  345. struct perf_event *ptrace_bps[HBP_NUM];
  346. /* Debug status used for traps, single steps, etc... */
  347. unsigned long debugreg6;
  348. /* Keep track of the exact dr7 value set by the user */
  349. unsigned long ptrace_dr7;
  350. /* Fault info: */
  351. unsigned long cr2;
  352. unsigned long trap_nr;
  353. unsigned long error_code;
  354. #ifdef CONFIG_VM86
  355. /* Virtual 86 mode info */
  356. struct vm86 *vm86;
  357. #endif
  358. /* IO permissions: */
  359. unsigned long *io_bitmap_ptr;
  360. unsigned long iopl;
  361. /* Max allowed port in the bitmap, in bytes: */
  362. unsigned io_bitmap_max;
  363. /* Floating point and extended processor state */
  364. struct fpu fpu;
  365. /*
  366. * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
  367. * the end.
  368. */
  369. };
  370. /*
  371. * Set IOPL bits in EFLAGS from given mask
  372. */
  373. static inline void native_set_iopl_mask(unsigned mask)
  374. {
  375. #ifdef CONFIG_X86_32
  376. unsigned int reg;
  377. asm volatile ("pushfl;"
  378. "popl %0;"
  379. "andl %1, %0;"
  380. "orl %2, %0;"
  381. "pushl %0;"
  382. "popfl"
  383. : "=&r" (reg)
  384. : "i" (~X86_EFLAGS_IOPL), "r" (mask));
  385. #endif
  386. }
  387. static inline void
  388. native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
  389. {
  390. tss->x86_tss.sp0 = thread->sp0;
  391. #ifdef CONFIG_X86_32
  392. /* Only happens when SEP is enabled, no need to test "SEP"arately: */
  393. if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
  394. tss->x86_tss.ss1 = thread->sysenter_cs;
  395. wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
  396. }
  397. #endif
  398. }
  399. static inline void native_swapgs(void)
  400. {
  401. #ifdef CONFIG_X86_64
  402. asm volatile("swapgs" ::: "memory");
  403. #endif
  404. }
  405. static inline unsigned long current_top_of_stack(void)
  406. {
  407. #ifdef CONFIG_X86_64
  408. return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
  409. #else
  410. /* sp0 on x86_32 is special in and around vm86 mode. */
  411. return this_cpu_read_stable(cpu_current_top_of_stack);
  412. #endif
  413. }
  414. #ifdef CONFIG_PARAVIRT
  415. #include <asm/paravirt.h>
  416. #else
  417. #define __cpuid native_cpuid
  418. #define paravirt_enabled() 0
  419. #define paravirt_has(x) 0
  420. static inline void load_sp0(struct tss_struct *tss,
  421. struct thread_struct *thread)
  422. {
  423. native_load_sp0(tss, thread);
  424. }
  425. #define set_iopl_mask native_set_iopl_mask
  426. #endif /* CONFIG_PARAVIRT */
  427. typedef struct {
  428. unsigned long seg;
  429. } mm_segment_t;
  430. /* Free all resources held by a thread. */
  431. extern void release_thread(struct task_struct *);
  432. unsigned long get_wchan(struct task_struct *p);
  433. /*
  434. * Generic CPUID function
  435. * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
  436. * resulting in stale register contents being returned.
  437. */
  438. static inline void cpuid(unsigned int op,
  439. unsigned int *eax, unsigned int *ebx,
  440. unsigned int *ecx, unsigned int *edx)
  441. {
  442. *eax = op;
  443. *ecx = 0;
  444. __cpuid(eax, ebx, ecx, edx);
  445. }
  446. /* Some CPUID calls want 'count' to be placed in ecx */
  447. static inline void cpuid_count(unsigned int op, int count,
  448. unsigned int *eax, unsigned int *ebx,
  449. unsigned int *ecx, unsigned int *edx)
  450. {
  451. *eax = op;
  452. *ecx = count;
  453. __cpuid(eax, ebx, ecx, edx);
  454. }
  455. /*
  456. * CPUID functions returning a single datum
  457. */
  458. static inline unsigned int cpuid_eax(unsigned int op)
  459. {
  460. unsigned int eax, ebx, ecx, edx;
  461. cpuid(op, &eax, &ebx, &ecx, &edx);
  462. return eax;
  463. }
  464. static inline unsigned int cpuid_ebx(unsigned int op)
  465. {
  466. unsigned int eax, ebx, ecx, edx;
  467. cpuid(op, &eax, &ebx, &ecx, &edx);
  468. return ebx;
  469. }
  470. static inline unsigned int cpuid_ecx(unsigned int op)
  471. {
  472. unsigned int eax, ebx, ecx, edx;
  473. cpuid(op, &eax, &ebx, &ecx, &edx);
  474. return ecx;
  475. }
  476. static inline unsigned int cpuid_edx(unsigned int op)
  477. {
  478. unsigned int eax, ebx, ecx, edx;
  479. cpuid(op, &eax, &ebx, &ecx, &edx);
  480. return edx;
  481. }
  482. /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
  483. static __always_inline void rep_nop(void)
  484. {
  485. asm volatile("rep; nop" ::: "memory");
  486. }
  487. static __always_inline void cpu_relax(void)
  488. {
  489. rep_nop();
  490. }
  491. #define cpu_relax_lowlatency() cpu_relax()
  492. /* Stop speculative execution and prefetching of modified code. */
  493. static inline void sync_core(void)
  494. {
  495. int tmp;
  496. #ifdef CONFIG_M486
  497. /*
  498. * Do a CPUID if available, otherwise do a jump. The jump
  499. * can conveniently enough be the jump around CPUID.
  500. */
  501. asm volatile("cmpl %2,%1\n\t"
  502. "jl 1f\n\t"
  503. "cpuid\n"
  504. "1:"
  505. : "=a" (tmp)
  506. : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
  507. : "ebx", "ecx", "edx", "memory");
  508. #else
  509. /*
  510. * CPUID is a barrier to speculative execution.
  511. * Prefetched instructions are automatically
  512. * invalidated when modified.
  513. */
  514. asm volatile("cpuid"
  515. : "=a" (tmp)
  516. : "0" (1)
  517. : "ebx", "ecx", "edx", "memory");
  518. #endif
  519. }
  520. extern void select_idle_routine(const struct cpuinfo_x86 *c);
  521. extern void init_amd_e400_c1e_mask(void);
  522. extern unsigned long boot_option_idle_override;
  523. extern bool amd_e400_c1e_detected;
  524. enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
  525. IDLE_POLL};
  526. extern void enable_sep_cpu(void);
  527. extern int sysenter_setup(void);
  528. extern void early_trap_init(void);
  529. void early_trap_pf_init(void);
  530. /* Defined in head.S */
  531. extern struct desc_ptr early_gdt_descr;
  532. extern void cpu_set_gdt(int);
  533. extern void switch_to_new_gdt(int);
  534. extern void load_percpu_segment(int);
  535. extern void cpu_init(void);
  536. static inline unsigned long get_debugctlmsr(void)
  537. {
  538. unsigned long debugctlmsr = 0;
  539. #ifndef CONFIG_X86_DEBUGCTLMSR
  540. if (boot_cpu_data.x86 < 6)
  541. return 0;
  542. #endif
  543. rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
  544. return debugctlmsr;
  545. }
  546. static inline void update_debugctlmsr(unsigned long debugctlmsr)
  547. {
  548. #ifndef CONFIG_X86_DEBUGCTLMSR
  549. if (boot_cpu_data.x86 < 6)
  550. return;
  551. #endif
  552. wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
  553. }
  554. extern void set_task_blockstep(struct task_struct *task, bool on);
  555. /* Boot loader type from the setup header: */
  556. extern int bootloader_type;
  557. extern int bootloader_version;
  558. extern char ignore_fpu_irq;
  559. #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
  560. #define ARCH_HAS_PREFETCHW
  561. #define ARCH_HAS_SPINLOCK_PREFETCH
  562. #ifdef CONFIG_X86_32
  563. # define BASE_PREFETCH ""
  564. # define ARCH_HAS_PREFETCH
  565. #else
  566. # define BASE_PREFETCH "prefetcht0 %P1"
  567. #endif
  568. /*
  569. * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
  570. *
  571. * It's not worth to care about 3dnow prefetches for the K6
  572. * because they are microcoded there and very slow.
  573. */
  574. static inline void prefetch(const void *x)
  575. {
  576. alternative_input(BASE_PREFETCH, "prefetchnta %P1",
  577. X86_FEATURE_XMM,
  578. "m" (*(const char *)x));
  579. }
  580. /*
  581. * 3dnow prefetch to get an exclusive cache line.
  582. * Useful for spinlocks to avoid one state transition in the
  583. * cache coherency protocol:
  584. */
  585. static inline void prefetchw(const void *x)
  586. {
  587. alternative_input(BASE_PREFETCH, "prefetchw %P1",
  588. X86_FEATURE_3DNOWPREFETCH,
  589. "m" (*(const char *)x));
  590. }
  591. static inline void spin_lock_prefetch(const void *x)
  592. {
  593. prefetchw(x);
  594. }
  595. #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
  596. TOP_OF_KERNEL_STACK_PADDING)
  597. #ifdef CONFIG_X86_32
  598. /*
  599. * User space process size: 3GB (default).
  600. */
  601. #define TASK_SIZE PAGE_OFFSET
  602. #define TASK_SIZE_MAX TASK_SIZE
  603. #define STACK_TOP TASK_SIZE
  604. #define STACK_TOP_MAX STACK_TOP
  605. #define INIT_THREAD { \
  606. .sp0 = TOP_OF_INIT_STACK, \
  607. .sysenter_cs = __KERNEL_CS, \
  608. .io_bitmap_ptr = NULL, \
  609. }
  610. extern unsigned long thread_saved_pc(struct task_struct *tsk);
  611. /*
  612. * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
  613. * This is necessary to guarantee that the entire "struct pt_regs"
  614. * is accessible even if the CPU haven't stored the SS/ESP registers
  615. * on the stack (interrupt gate does not save these registers
  616. * when switching to the same priv ring).
  617. * Therefore beware: accessing the ss/esp fields of the
  618. * "struct pt_regs" is possible, but they may contain the
  619. * completely wrong values.
  620. */
  621. #define task_pt_regs(task) \
  622. ({ \
  623. unsigned long __ptr = (unsigned long)task_stack_page(task); \
  624. __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
  625. ((struct pt_regs *)__ptr) - 1; \
  626. })
  627. #define KSTK_ESP(task) (task_pt_regs(task)->sp)
  628. #else
  629. /*
  630. * User space process size. 47bits minus one guard page. The guard
  631. * page is necessary on Intel CPUs: if a SYSCALL instruction is at
  632. * the highest possible canonical userspace address, then that
  633. * syscall will enter the kernel with a non-canonical return
  634. * address, and SYSRET will explode dangerously. We avoid this
  635. * particular problem by preventing anything from being mapped
  636. * at the maximum canonical address.
  637. */
  638. #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE)
  639. /* This decides where the kernel will search for a free chunk of vm
  640. * space during mmap's.
  641. */
  642. #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
  643. 0xc0000000 : 0xFFFFe000)
  644. #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \
  645. IA32_PAGE_OFFSET : TASK_SIZE_MAX)
  646. #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
  647. IA32_PAGE_OFFSET : TASK_SIZE_MAX)
  648. #define STACK_TOP TASK_SIZE
  649. #define STACK_TOP_MAX TASK_SIZE_MAX
  650. #define INIT_THREAD { \
  651. .sp0 = TOP_OF_INIT_STACK \
  652. }
  653. /*
  654. * Return saved PC of a blocked thread.
  655. * What is this good for? it will be always the scheduler or ret_from_fork.
  656. */
  657. #define thread_saved_pc(t) READ_ONCE_NOCHECK(*(unsigned long *)((t)->thread.sp - 8))
  658. #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
  659. extern unsigned long KSTK_ESP(struct task_struct *task);
  660. #endif /* CONFIG_X86_64 */
  661. extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
  662. unsigned long new_sp);
  663. /*
  664. * This decides where the kernel will search for a free chunk of vm
  665. * space during mmap's.
  666. */
  667. #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
  668. #define KSTK_EIP(task) (task_pt_regs(task)->ip)
  669. /* Get/set a process' ability to use the timestamp counter instruction */
  670. #define GET_TSC_CTL(adr) get_tsc_mode((adr))
  671. #define SET_TSC_CTL(val) set_tsc_mode((val))
  672. extern int get_tsc_mode(unsigned long adr);
  673. extern int set_tsc_mode(unsigned int val);
  674. /* Register/unregister a process' MPX related resource */
  675. #define MPX_ENABLE_MANAGEMENT() mpx_enable_management()
  676. #define MPX_DISABLE_MANAGEMENT() mpx_disable_management()
  677. #ifdef CONFIG_X86_INTEL_MPX
  678. extern int mpx_enable_management(void);
  679. extern int mpx_disable_management(void);
  680. #else
  681. static inline int mpx_enable_management(void)
  682. {
  683. return -EINVAL;
  684. }
  685. static inline int mpx_disable_management(void)
  686. {
  687. return -EINVAL;
  688. }
  689. #endif /* CONFIG_X86_INTEL_MPX */
  690. extern u16 amd_get_nb_id(int cpu);
  691. extern u32 amd_get_nodes_per_socket(void);
  692. static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
  693. {
  694. uint32_t base, eax, signature[3];
  695. for (base = 0x40000000; base < 0x40010000; base += 0x100) {
  696. cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
  697. if (!memcmp(sig, signature, 12) &&
  698. (leaves == 0 || ((eax - base) >= leaves)))
  699. return base;
  700. }
  701. return 0;
  702. }
  703. extern unsigned long arch_align_stack(unsigned long sp);
  704. extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
  705. void default_idle(void);
  706. #ifdef CONFIG_XEN
  707. bool xen_set_default_idle(void);
  708. #else
  709. #define xen_set_default_idle 0
  710. #endif
  711. void stop_this_cpu(void *dummy);
  712. void df_debug(struct pt_regs *regs, long error_code);
  713. #endif /* _ASM_X86_PROCESSOR_H */