1 #ifndef __ASM_X86_PROCESSOR_H
2 #define __ASM_X86_PROCESSOR_H
4 #include <asm/processor-flags.h>
6 /* migration helpers, for KVM - will be removed in 2.6.25: */
8 #define Xgt_desc_struct desc_ptr
10 /* Forward declaration, a strange C thing */
15 #include <asm/math_emu.h>
16 #include <asm/segment.h>
17 #include <asm/types.h>
18 #include <asm/sigcontext.h>
19 #include <asm/current.h>
20 #include <asm/cpufeature.h>
21 #include <asm/system.h>
23 #include <asm/percpu.h>
25 #include <asm/desc_defs.h>
28 #include <linux/personality.h>
29 #include <linux/cpumask.h>
30 #include <linux/cache.h>
31 #include <linux/threads.h>
32 #include <linux/init.h>
35 * Default implementation of macro that returns current
36 * instruction pointer ("program counter").
38 static inline void *current_text_addr(void)
42 asm volatile("mov $1f, %0; 1:":"=r" (pc));
47 #ifdef CONFIG_X86_VSMP
48 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
49 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
51 # define ARCH_MIN_TASKALIGN 16
52 # define ARCH_MIN_MMSTRUCT_ALIGN 0
56 * CPU type and hardware bug flags. Kept separately for each CPU.
57 * Members of this structure are referenced in head.S, so think twice
58 * before touching them. [mj]
62 __u8 x86; /* CPU family */
63 __u8 x86_vendor; /* CPU vendor */
67 char wp_works_ok; /* It doesn't on 386's */
69 /* Problems on some 486Dx4's and old 386's: */
78 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
82 /* CPUID returned core id bits: */
84 /* Max extended CPUID function supported: */
85 __u32 extended_cpuid_level;
87 /* Maximum supported CPUID level, -1=no CPUID: */
89 __u32 x86_capability[NCAPINTS];
90 char x86_vendor_id[16];
91 char x86_model_id[64];
92 /* in KB - valid for CPUS which support this call: */
94 int x86_cache_alignment; /* In bytes */
96 unsigned long loops_per_jiffy;
98 /* cpus sharing the last level cache: */
99 cpumask_t llc_shared_map;
101 /* cpuid returned max cores value: */
105 u16 x86_clflush_size;
107 /* number of cores as seen by the OS: */
109 /* Physical processor id: */
113 /* Index into per_cpu list: */
116 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
118 #define X86_VENDOR_INTEL 0
119 #define X86_VENDOR_CYRIX 1
120 #define X86_VENDOR_AMD 2
121 #define X86_VENDOR_UMC 3
122 #define X86_VENDOR_NEXGEN 4
123 #define X86_VENDOR_CENTAUR 5
124 #define X86_VENDOR_TRANSMETA 7
125 #define X86_VENDOR_NSC 8
126 #define X86_VENDOR_NUM 9
128 #define X86_VENDOR_UNKNOWN 0xff
131 * capabilities of CPUs
133 extern struct cpuinfo_x86 boot_cpu_data;
134 extern struct cpuinfo_x86 new_cpu_data;
136 extern struct tss_struct doublefault_tss;
137 extern __u32 cleared_cpu_caps[NCAPINTS];
140 DECLARE_PER_CPU(struct cpuinfo_x86, cpu_info);
141 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
142 #define current_cpu_data cpu_data(smp_processor_id())
144 #define cpu_data(cpu) boot_cpu_data
145 #define current_cpu_data boot_cpu_data
148 static inline int hlt_works(int cpu)
151 return cpu_data(cpu).hlt_works_ok;
157 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
159 extern void cpu_detect(struct cpuinfo_x86 *c);
161 extern void identify_cpu(struct cpuinfo_x86 *);
162 extern void identify_boot_cpu(void);
163 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
164 extern void print_cpu_info(struct cpuinfo_x86 *);
165 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
166 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
167 extern unsigned short num_cache_leaves;
169 #if defined(CONFIG_X86_HT) || defined(CONFIG_X86_64)
170 extern void detect_ht(struct cpuinfo_x86 *c);
172 static inline void detect_ht(struct cpuinfo_x86 *c) {}
175 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
176 unsigned int *ecx, unsigned int *edx)
178 /* ecx is often an input as well as an output. */
184 : "0" (*eax), "2" (*ecx));
187 static inline void load_cr3(pgd_t *pgdir)
189 write_cr3(__pa(pgdir));
193 /* This is the TSS defined by the hardware. */
195 unsigned short back_link, __blh;
197 unsigned short ss0, __ss0h;
199 /* ss1 caches MSR_IA32_SYSENTER_CS: */
200 unsigned short ss1, __ss1h;
202 unsigned short ss2, __ss2h;
214 unsigned short es, __esh;
215 unsigned short cs, __csh;
216 unsigned short ss, __ssh;
217 unsigned short ds, __dsh;
218 unsigned short fs, __fsh;
219 unsigned short gs, __gsh;
220 unsigned short ldt, __ldth;
221 unsigned short trace;
222 unsigned short io_bitmap_base;
224 } __attribute__((packed));
238 } __attribute__((packed)) ____cacheline_aligned;
244 #define IO_BITMAP_BITS 65536
245 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
246 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
247 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
248 #define INVALID_IO_BITMAP_OFFSET 0x8000
249 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
253 * The hardware state:
255 struct x86_hw_tss x86_tss;
258 * The extra 1 is there because the CPU will access an
259 * additional byte beyond the end of the IO permission
260 * bitmap. The extra byte must be all 1 bits, and must
261 * be within the limit.
263 unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
265 * Cache the current maximum and the last task that used the bitmap:
267 unsigned long io_bitmap_max;
268 struct thread_struct *io_bitmap_owner;
271 * Pad the TSS to be cacheline-aligned (size is 0x100):
273 unsigned long __cacheline_filler[35];
275 * .. and then another 0x100 bytes for the emergency kernel stack:
277 unsigned long stack[64];
279 } __attribute__((packed));
281 DECLARE_PER_CPU(struct tss_struct, init_tss);
284 * Save the original ist values for checking stack pointers during debugging
287 unsigned long ist[7];
290 #define MXCSR_DEFAULT 0x1f80
292 struct i387_fsave_struct {
293 u32 cwd; /* FPU Control Word */
294 u32 swd; /* FPU Status Word */
295 u32 twd; /* FPU Tag Word */
296 u32 fip; /* FPU IP Offset */
297 u32 fcs; /* FPU IP Selector */
298 u32 foo; /* FPU Operand Pointer Offset */
299 u32 fos; /* FPU Operand Pointer Selector */
301 /* 8*10 bytes for each FP-reg = 80 bytes: */
304 /* Software status information [not touched by FSAVE ]: */
308 struct i387_fxsave_struct {
309 u16 cwd; /* Control Word */
310 u16 swd; /* Status Word */
311 u16 twd; /* Tag Word */
312 u16 fop; /* Last Instruction Opcode */
315 u64 rip; /* Instruction Pointer */
316 u64 rdp; /* Data Pointer */
319 u32 fip; /* FPU IP Offset */
320 u32 fcs; /* FPU IP Selector */
321 u32 foo; /* FPU Operand Offset */
322 u32 fos; /* FPU Operand Selector */
325 u32 mxcsr; /* MXCSR Register State */
326 u32 mxcsr_mask; /* MXCSR Mask */
328 /* 8*16 bytes for each FP-reg = 128 bytes: */
331 /* 16*16 bytes for each XMM-reg = 256 bytes: */
336 } __attribute__((aligned(16)));
338 struct i387_soft_struct {
346 /* 8*10 bytes for each FP-reg = 80 bytes: */
359 struct i387_fsave_struct fsave;
360 struct i387_fxsave_struct fxsave;
361 struct i387_soft_struct soft;
365 DECLARE_PER_CPU(struct orig_ist, orig_ist);
368 extern void print_cpu_info(struct cpuinfo_x86 *);
369 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
370 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
371 extern unsigned short num_cache_leaves;
373 struct thread_struct {
374 /* Cached TLS descriptors: */
375 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
379 unsigned long sysenter_cs;
381 unsigned long usersp; /* Copy from PDA */
384 unsigned short fsindex;
385 unsigned short gsindex;
390 /* Hardware debugging registers: */
391 unsigned long debugreg0;
392 unsigned long debugreg1;
393 unsigned long debugreg2;
394 unsigned long debugreg3;
395 unsigned long debugreg6;
396 unsigned long debugreg7;
399 unsigned long trap_no;
400 unsigned long error_code;
401 /* Floating point info: */
402 union i387_union i387 __attribute__((aligned(16)));;
404 /* Virtual 86 mode info */
405 struct vm86_struct __user *vm86_info;
406 unsigned long screen_bitmap;
407 unsigned long v86flags;
408 unsigned long v86mask;
409 unsigned long saved_sp0;
410 unsigned int saved_fs;
411 unsigned int saved_gs;
413 /* IO permissions: */
414 unsigned long *io_bitmap_ptr;
416 /* Max allowed port in the bitmap, in bytes: */
417 unsigned io_bitmap_max;
418 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
419 unsigned long debugctlmsr;
420 /* Debug Store - if not 0 points to a DS Save Area configuration;
421 * goes into MSR_IA32_DS_AREA */
422 unsigned long ds_area_msr;
425 static inline unsigned long native_get_debugreg(int regno)
427 unsigned long val = 0; /* Damn you, gcc! */
431 asm("mov %%db0, %0" :"=r" (val)); break;
433 asm("mov %%db1, %0" :"=r" (val)); break;
435 asm("mov %%db2, %0" :"=r" (val)); break;
437 asm("mov %%db3, %0" :"=r" (val)); break;
439 asm("mov %%db6, %0" :"=r" (val)); break;
441 asm("mov %%db7, %0" :"=r" (val)); break;
448 static inline void native_set_debugreg(int regno, unsigned long value)
452 asm("mov %0, %%db0" ::"r" (value));
455 asm("mov %0, %%db1" ::"r" (value));
458 asm("mov %0, %%db2" ::"r" (value));
461 asm("mov %0, %%db3" ::"r" (value));
464 asm("mov %0, %%db6" ::"r" (value));
467 asm("mov %0, %%db7" ::"r" (value));
475 * Set IOPL bits in EFLAGS from given mask
477 static inline void native_set_iopl_mask(unsigned mask)
482 __asm__ __volatile__ ("pushfl;"
489 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
494 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
496 tss->x86_tss.sp0 = thread->sp0;
498 /* Only happens when SEP is enabled, no need to test "SEP"arately: */
499 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
500 tss->x86_tss.ss1 = thread->sysenter_cs;
501 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
506 static inline void native_swapgs(void)
509 asm volatile("swapgs" ::: "memory");
513 #ifdef CONFIG_PARAVIRT
514 #include <asm/paravirt.h>
516 #define __cpuid native_cpuid
517 #define paravirt_enabled() 0
520 * These special macros can be used to get or set a debugging register
522 #define get_debugreg(var, register) \
523 (var) = native_get_debugreg(register)
524 #define set_debugreg(value, register) \
525 native_set_debugreg(register, value)
528 load_sp0(struct tss_struct *tss, struct thread_struct *thread)
530 native_load_sp0(tss, thread);
533 #define set_iopl_mask native_set_iopl_mask
534 #define SWAPGS swapgs
535 #endif /* CONFIG_PARAVIRT */
538 * Save the cr4 feature set we're using (ie
539 * Pentium 4MB enable and PPro Global page
540 * enable), so that any CPU's that boot up
541 * after us can get the correct flags.
543 extern unsigned long mmu_cr4_features;
545 static inline void set_in_cr4(unsigned long mask)
549 mmu_cr4_features |= mask;
555 static inline void clear_in_cr4(unsigned long mask)
559 mmu_cr4_features &= ~mask;
565 struct microcode_header {
573 unsigned int datasize;
574 unsigned int totalsize;
575 unsigned int reserved[3];
579 struct microcode_header hdr;
580 unsigned int bits[0];
583 typedef struct microcode microcode_t;
584 typedef struct microcode_header microcode_header_t;
586 /* microcode format is extended from prescott processors */
587 struct extended_signature {
593 struct extended_sigtable {
596 unsigned int reserved[3];
597 struct extended_signature sigs[0];
606 * create a kernel thread without removing it from tasklists
608 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
610 /* Free all resources held by a thread. */
611 extern void release_thread(struct task_struct *);
613 /* Prepare to copy thread state - unlazy all lazy state */
614 extern void prepare_to_copy(struct task_struct *tsk);
616 unsigned long get_wchan(struct task_struct *p);
619 * Generic CPUID function
620 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
621 * resulting in stale register contents being returned.
623 static inline void cpuid(unsigned int op,
624 unsigned int *eax, unsigned int *ebx,
625 unsigned int *ecx, unsigned int *edx)
629 __cpuid(eax, ebx, ecx, edx);
632 /* Some CPUID calls want 'count' to be placed in ecx */
633 static inline void cpuid_count(unsigned int op, int count,
634 unsigned int *eax, unsigned int *ebx,
635 unsigned int *ecx, unsigned int *edx)
639 __cpuid(eax, ebx, ecx, edx);
643 * CPUID functions returning a single datum
645 static inline unsigned int cpuid_eax(unsigned int op)
647 unsigned int eax, ebx, ecx, edx;
649 cpuid(op, &eax, &ebx, &ecx, &edx);
654 static inline unsigned int cpuid_ebx(unsigned int op)
656 unsigned int eax, ebx, ecx, edx;
658 cpuid(op, &eax, &ebx, &ecx, &edx);
663 static inline unsigned int cpuid_ecx(unsigned int op)
665 unsigned int eax, ebx, ecx, edx;
667 cpuid(op, &eax, &ebx, &ecx, &edx);
672 static inline unsigned int cpuid_edx(unsigned int op)
674 unsigned int eax, ebx, ecx, edx;
676 cpuid(op, &eax, &ebx, &ecx, &edx);
681 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
682 static inline void rep_nop(void)
684 __asm__ __volatile__("rep; nop" ::: "memory");
687 static inline void cpu_relax(void)
692 /* Stop speculative execution: */
693 static inline void sync_core(void)
697 asm volatile("cpuid" : "=a" (tmp) : "0" (1)
698 : "ebx", "ecx", "edx", "memory");
702 __monitor(const void *eax, unsigned long ecx, unsigned long edx)
704 /* "monitor %eax, %ecx, %edx;" */
706 ".byte 0x0f, 0x01, 0xc8;"
707 :: "a" (eax), "c" (ecx), "d"(edx));
710 static inline void __mwait(unsigned long eax, unsigned long ecx)
712 /* "mwait %eax, %ecx;" */
714 ".byte 0x0f, 0x01, 0xc9;"
715 :: "a" (eax), "c" (ecx));
718 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
720 /* "mwait %eax, %ecx;" */
722 "sti; .byte 0x0f, 0x01, 0xc9;"
723 :: "a" (eax), "c" (ecx));
726 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
728 extern int force_mwait;
730 extern void select_idle_routine(const struct cpuinfo_x86 *c);
732 extern unsigned long boot_option_idle_override;
734 extern void enable_sep_cpu(void);
735 extern int sysenter_setup(void);
737 /* Defined in head.S */
738 extern struct desc_ptr early_gdt_descr;
740 extern void cpu_set_gdt(int);
741 extern void switch_to_new_gdt(void);
742 extern void cpu_init(void);
743 extern void init_gdt(int cpu);
746 * from system description table in BIOS. Mostly for MCA use, but
747 * others may find it useful:
749 extern unsigned int machine_id;
750 extern unsigned int machine_submodel_id;
751 extern unsigned int BIOS_revision;
753 /* Boot loader type from the setup header: */
754 extern int bootloader_type;
756 extern char ignore_fpu_irq;
758 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
759 #define ARCH_HAS_PREFETCHW
760 #define ARCH_HAS_SPINLOCK_PREFETCH
763 # define BASE_PREFETCH ASM_NOP4
764 # define ARCH_HAS_PREFETCH
766 # define BASE_PREFETCH "prefetcht0 (%1)"
770 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
772 * It's not worth to care about 3dnow prefetches for the K6
773 * because they are microcoded there and very slow.
775 static inline void prefetch(const void *x)
777 alternative_input(BASE_PREFETCH,
784 * 3dnow prefetch to get an exclusive cache line.
785 * Useful for spinlocks to avoid one state transition in the
786 * cache coherency protocol:
788 static inline void prefetchw(const void *x)
790 alternative_input(BASE_PREFETCH,
796 static inline void spin_lock_prefetch(const void *x)
803 * User space process size: 3GB (default).
805 #define TASK_SIZE PAGE_OFFSET
806 #define STACK_TOP TASK_SIZE
807 #define STACK_TOP_MAX STACK_TOP
809 #define INIT_THREAD { \
810 .sp0 = sizeof(init_stack) + (long)&init_stack, \
812 .sysenter_cs = __KERNEL_CS, \
813 .io_bitmap_ptr = NULL, \
814 .fs = __KERNEL_PERCPU, \
818 * Note that the .io_bitmap member must be extra-big. This is because
819 * the CPU will access an additional byte beyond the end of the IO
820 * permission bitmap. The extra byte must be all 1 bits, and must
821 * be within the limit.
825 .sp0 = sizeof(init_stack) + (long)&init_stack, \
826 .ss0 = __KERNEL_DS, \
827 .ss1 = __KERNEL_CS, \
828 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
830 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 }, \
833 extern unsigned long thread_saved_pc(struct task_struct *tsk);
835 #define THREAD_SIZE_LONGS (THREAD_SIZE/sizeof(unsigned long))
836 #define KSTK_TOP(info) \
838 unsigned long *__ptr = (unsigned long *)(info); \
839 (unsigned long)(&__ptr[THREAD_SIZE_LONGS]); \
843 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
844 * This is necessary to guarantee that the entire "struct pt_regs"
845 * is accessable even if the CPU haven't stored the SS/ESP registers
846 * on the stack (interrupt gate does not save these registers
847 * when switching to the same priv ring).
848 * Therefore beware: accessing the ss/esp fields of the
849 * "struct pt_regs" is possible, but they may contain the
850 * completely wrong values.
852 #define task_pt_regs(task) \
854 struct pt_regs *__regs__; \
855 __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
859 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
863 * User space process size. 47bits minus one guard page.
865 #define TASK_SIZE64 (0x800000000000UL - 4096)
867 /* This decides where the kernel will search for a free chunk of vm
868 * space during mmap's.
870 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
871 0xc0000000 : 0xFFFFe000)
873 #define TASK_SIZE (test_thread_flag(TIF_IA32) ? \
874 IA32_PAGE_OFFSET : TASK_SIZE64)
875 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_IA32)) ? \
876 IA32_PAGE_OFFSET : TASK_SIZE64)
878 #define STACK_TOP TASK_SIZE
879 #define STACK_TOP_MAX TASK_SIZE64
881 #define INIT_THREAD { \
882 .sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
886 .x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
890 * Return saved PC of a blocked thread.
891 * What is this good for? it will be always the scheduler or ret_from_fork.
893 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
895 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
896 #define KSTK_ESP(tsk) -1 /* sorry. doesn't work for syscall. */
897 #endif /* CONFIG_X86_64 */
899 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
900 unsigned long new_sp);
903 * This decides where the kernel will search for a free chunk of vm
904 * space during mmap's.
906 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
908 #define KSTK_EIP(task) (task_pt_regs(task)->ip)