1 #ifndef __ASM_X86_PROCESSOR_H
2 #define __ASM_X86_PROCESSOR_H
4 #include <asm/processor-flags.h>
6 /* Forward declaration, a strange C thing */
11 #include <asm/math_emu.h>
12 #include <asm/segment.h>
13 #include <asm/types.h>
14 #include <asm/sigcontext.h>
15 #include <asm/current.h>
16 #include <asm/cpufeature.h>
17 #include <asm/system.h>
19 #include <asm/percpu.h>
21 #include <asm/desc_defs.h>
24 #include <linux/personality.h>
25 #include <linux/cpumask.h>
26 #include <linux/cache.h>
27 #include <linux/threads.h>
28 #include <linux/init.h>
31 * Default implementation of macro that returns current
32 * instruction pointer ("program counter").
34 static inline void *current_text_addr(void)
38 asm volatile("mov $1f, %0; 1:":"=r" (pc
));
43 #ifdef CONFIG_X86_VSMP
44 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
45 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
47 # define ARCH_MIN_TASKALIGN 16
48 # define ARCH_MIN_MMSTRUCT_ALIGN 0
52 * CPU type and hardware bug flags. Kept separately for each CPU.
53 * Members of this structure are referenced in head.S, so think twice
54 * before touching them. [mj]
58 __u8 x86
; /* CPU family */
59 __u8 x86_vendor
; /* CPU vendor */
63 char wp_works_ok
; /* It doesn't on 386's */
65 /* Problems on some 486Dx4's and old 386's: */
74 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
78 /* CPUID returned core id bits: */
80 /* Max extended CPUID function supported: */
81 __u32 extended_cpuid_level
;
83 /* Maximum supported CPUID level, -1=no CPUID: */
85 __u32 x86_capability
[NCAPINTS
];
86 char x86_vendor_id
[16];
87 char x86_model_id
[64];
88 /* in KB - valid for CPUS which support this call: */
90 int x86_cache_alignment
; /* In bytes */
92 unsigned long loops_per_jiffy
;
94 /* cpus sharing the last level cache: */
95 cpumask_t llc_shared_map
;
97 /* cpuid returned max cores value: */
101 u16 x86_clflush_size
;
103 /* number of cores as seen by the OS: */
105 /* Physical processor id: */
109 /* Index into per_cpu list: */
112 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
114 #define X86_VENDOR_INTEL 0
115 #define X86_VENDOR_CYRIX 1
116 #define X86_VENDOR_AMD 2
117 #define X86_VENDOR_UMC 3
118 #define X86_VENDOR_CENTAUR 5
119 #define X86_VENDOR_TRANSMETA 7
120 #define X86_VENDOR_NSC 8
121 #define X86_VENDOR_NUM 9
123 #define X86_VENDOR_UNKNOWN 0xff
126 * capabilities of CPUs
128 extern struct cpuinfo_x86 boot_cpu_data
;
129 extern struct cpuinfo_x86 new_cpu_data
;
131 extern struct tss_struct doublefault_tss
;
132 extern __u32 cleared_cpu_caps
[NCAPINTS
];
135 DECLARE_PER_CPU(struct cpuinfo_x86
, cpu_info
);
136 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
137 #define current_cpu_data __get_cpu_var(cpu_info)
139 #define cpu_data(cpu) boot_cpu_data
140 #define current_cpu_data boot_cpu_data
143 static inline int hlt_works(int cpu
)
146 return cpu_data(cpu
).hlt_works_ok
;
152 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
154 extern void cpu_detect(struct cpuinfo_x86
*c
);
156 extern void early_cpu_init(void);
157 extern void identify_boot_cpu(void);
158 extern void identify_secondary_cpu(struct cpuinfo_x86
*);
159 extern void print_cpu_info(struct cpuinfo_x86
*);
160 extern void init_scattered_cpuid_features(struct cpuinfo_x86
*c
);
161 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86
*c
);
162 extern unsigned short num_cache_leaves
;
164 #if defined(CONFIG_X86_HT) || defined(CONFIG_X86_64)
165 extern void detect_ht(struct cpuinfo_x86
*c
);
167 static inline void detect_ht(struct cpuinfo_x86
*c
) {}
170 static inline void native_cpuid(unsigned int *eax
, unsigned int *ebx
,
171 unsigned int *ecx
, unsigned int *edx
)
173 /* ecx is often an input as well as an output. */
179 : "0" (*eax
), "2" (*ecx
));
182 static inline void load_cr3(pgd_t
*pgdir
)
184 write_cr3(__pa(pgdir
));
188 /* This is the TSS defined by the hardware. */
190 unsigned short back_link
, __blh
;
192 unsigned short ss0
, __ss0h
;
194 /* ss1 caches MSR_IA32_SYSENTER_CS: */
195 unsigned short ss1
, __ss1h
;
197 unsigned short ss2
, __ss2h
;
209 unsigned short es
, __esh
;
210 unsigned short cs
, __csh
;
211 unsigned short ss
, __ssh
;
212 unsigned short ds
, __dsh
;
213 unsigned short fs
, __fsh
;
214 unsigned short gs
, __gsh
;
215 unsigned short ldt
, __ldth
;
216 unsigned short trace
;
217 unsigned short io_bitmap_base
;
219 } __attribute__((packed
));
233 } __attribute__((packed
)) ____cacheline_aligned
;
239 #define IO_BITMAP_BITS 65536
240 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
241 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
242 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
243 #define INVALID_IO_BITMAP_OFFSET 0x8000
244 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
248 * The hardware state:
250 struct x86_hw_tss x86_tss
;
253 * The extra 1 is there because the CPU will access an
254 * additional byte beyond the end of the IO permission
255 * bitmap. The extra byte must be all 1 bits, and must
256 * be within the limit.
258 unsigned long io_bitmap
[IO_BITMAP_LONGS
+ 1];
260 * Cache the current maximum and the last task that used the bitmap:
262 unsigned long io_bitmap_max
;
263 struct thread_struct
*io_bitmap_owner
;
266 * .. and then another 0x100 bytes for the emergency kernel stack:
268 unsigned long stack
[64];
270 } ____cacheline_aligned
;
272 DECLARE_PER_CPU(struct tss_struct
, init_tss
);
275 * Save the original ist values for checking stack pointers during debugging
278 unsigned long ist
[7];
281 #define MXCSR_DEFAULT 0x1f80
283 struct i387_fsave_struct
{
284 u32 cwd
; /* FPU Control Word */
285 u32 swd
; /* FPU Status Word */
286 u32 twd
; /* FPU Tag Word */
287 u32 fip
; /* FPU IP Offset */
288 u32 fcs
; /* FPU IP Selector */
289 u32 foo
; /* FPU Operand Pointer Offset */
290 u32 fos
; /* FPU Operand Pointer Selector */
292 /* 8*10 bytes for each FP-reg = 80 bytes: */
295 /* Software status information [not touched by FSAVE ]: */
299 struct i387_fxsave_struct
{
300 u16 cwd
; /* Control Word */
301 u16 swd
; /* Status Word */
302 u16 twd
; /* Tag Word */
303 u16 fop
; /* Last Instruction Opcode */
306 u64 rip
; /* Instruction Pointer */
307 u64 rdp
; /* Data Pointer */
310 u32 fip
; /* FPU IP Offset */
311 u32 fcs
; /* FPU IP Selector */
312 u32 foo
; /* FPU Operand Offset */
313 u32 fos
; /* FPU Operand Selector */
316 u32 mxcsr
; /* MXCSR Register State */
317 u32 mxcsr_mask
; /* MXCSR Mask */
319 /* 8*16 bytes for each FP-reg = 128 bytes: */
322 /* 16*16 bytes for each XMM-reg = 256 bytes: */
327 } __attribute__((aligned(16)));
329 struct i387_soft_struct
{
337 /* 8*10 bytes for each FP-reg = 80 bytes: */
349 union thread_xstate
{
350 struct i387_fsave_struct fsave
;
351 struct i387_fxsave_struct fxsave
;
352 struct i387_soft_struct soft
;
356 DECLARE_PER_CPU(struct orig_ist
, orig_ist
);
359 extern void print_cpu_info(struct cpuinfo_x86
*);
360 extern unsigned int xstate_size
;
361 extern void free_thread_xstate(struct task_struct
*);
362 extern struct kmem_cache
*task_xstate_cachep
;
363 extern void init_scattered_cpuid_features(struct cpuinfo_x86
*c
);
364 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86
*c
);
365 extern unsigned short num_cache_leaves
;
367 struct thread_struct
{
368 /* Cached TLS descriptors: */
369 struct desc_struct tls_array
[GDT_ENTRY_TLS_ENTRIES
];
373 unsigned long sysenter_cs
;
375 unsigned long usersp
; /* Copy from PDA */
378 unsigned short fsindex
;
379 unsigned short gsindex
;
384 /* Hardware debugging registers: */
385 unsigned long debugreg0
;
386 unsigned long debugreg1
;
387 unsigned long debugreg2
;
388 unsigned long debugreg3
;
389 unsigned long debugreg6
;
390 unsigned long debugreg7
;
393 unsigned long trap_no
;
394 unsigned long error_code
;
395 /* floating point and extended processor state */
396 union thread_xstate
*xstate
;
398 /* Virtual 86 mode info */
399 struct vm86_struct __user
*vm86_info
;
400 unsigned long screen_bitmap
;
401 unsigned long v86flags
;
402 unsigned long v86mask
;
403 unsigned long saved_sp0
;
404 unsigned int saved_fs
;
405 unsigned int saved_gs
;
407 /* IO permissions: */
408 unsigned long *io_bitmap_ptr
;
410 /* Max allowed port in the bitmap, in bytes: */
411 unsigned io_bitmap_max
;
412 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
413 unsigned long debugctlmsr
;
414 /* Debug Store - if not 0 points to a DS Save Area configuration;
415 * goes into MSR_IA32_DS_AREA */
416 unsigned long ds_area_msr
;
419 static inline unsigned long native_get_debugreg(int regno
)
421 unsigned long val
= 0; /* Damn you, gcc! */
425 asm("mov %%db0, %0" :"=r" (val
));
428 asm("mov %%db1, %0" :"=r" (val
));
431 asm("mov %%db2, %0" :"=r" (val
));
434 asm("mov %%db3, %0" :"=r" (val
));
437 asm("mov %%db6, %0" :"=r" (val
));
440 asm("mov %%db7, %0" :"=r" (val
));
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)
527 static inline void load_sp0(struct tss_struct
*tss
,
528 struct thread_struct
*thread
)
530 native_load_sp0(tss
, thread
);
533 #define set_iopl_mask native_set_iopl_mask
534 #endif /* CONFIG_PARAVIRT */
537 * Save the cr4 feature set we're using (ie
538 * Pentium 4MB enable and PPro Global page
539 * enable), so that any CPU's that boot up
540 * after us can get the correct flags.
542 extern unsigned long mmu_cr4_features
;
544 static inline void set_in_cr4(unsigned long mask
)
548 mmu_cr4_features
|= mask
;
554 static inline void clear_in_cr4(unsigned long mask
)
558 mmu_cr4_features
&= ~mask
;
564 struct microcode_header
{
572 unsigned int datasize
;
573 unsigned int totalsize
;
574 unsigned int reserved
[3];
578 struct microcode_header hdr
;
579 unsigned int bits
[0];
582 typedef struct microcode microcode_t
;
583 typedef struct microcode_header microcode_header_t
;
585 /* microcode format is extended from prescott processors */
586 struct extended_signature
{
592 struct extended_sigtable
{
595 unsigned int reserved
[3];
596 struct extended_signature sigs
[0];
605 * create a kernel thread without removing it from tasklists
607 extern int kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
);
609 /* Free all resources held by a thread. */
610 extern void release_thread(struct task_struct
*);
612 /* Prepare to copy thread state - unlazy all lazy state */
613 extern void prepare_to_copy(struct task_struct
*tsk
);
615 unsigned long get_wchan(struct task_struct
*p
);
618 * Generic CPUID function
619 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
620 * resulting in stale register contents being returned.
622 static inline void cpuid(unsigned int op
,
623 unsigned int *eax
, unsigned int *ebx
,
624 unsigned int *ecx
, unsigned int *edx
)
628 __cpuid(eax
, ebx
, ecx
, edx
);
631 /* Some CPUID calls want 'count' to be placed in ecx */
632 static inline void cpuid_count(unsigned int op
, int count
,
633 unsigned int *eax
, unsigned int *ebx
,
634 unsigned int *ecx
, unsigned int *edx
)
638 __cpuid(eax
, ebx
, ecx
, edx
);
642 * CPUID functions returning a single datum
644 static inline unsigned int cpuid_eax(unsigned int op
)
646 unsigned int eax
, ebx
, ecx
, edx
;
648 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
653 static inline unsigned int cpuid_ebx(unsigned int op
)
655 unsigned int eax
, ebx
, ecx
, edx
;
657 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
662 static inline unsigned int cpuid_ecx(unsigned int op
)
664 unsigned int eax
, ebx
, ecx
, edx
;
666 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
671 static inline unsigned int cpuid_edx(unsigned int op
)
673 unsigned int eax
, ebx
, ecx
, edx
;
675 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
680 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
681 static inline void rep_nop(void)
683 asm volatile("rep; nop" ::: "memory");
686 static inline void cpu_relax(void)
691 /* Stop speculative execution: */
692 static inline void sync_core(void)
696 asm volatile("cpuid" : "=a" (tmp
) : "0" (1)
697 : "ebx", "ecx", "edx", "memory");
700 static inline void __monitor(const void *eax
, unsigned long ecx
,
703 /* "monitor %eax, %ecx, %edx;" */
704 asm volatile(".byte 0x0f, 0x01, 0xc8;"
705 :: "a" (eax
), "c" (ecx
), "d"(edx
));
708 static inline void __mwait(unsigned long eax
, unsigned long ecx
)
710 /* "mwait %eax, %ecx;" */
711 asm volatile(".byte 0x0f, 0x01, 0xc9;"
712 :: "a" (eax
), "c" (ecx
));
715 static inline void __sti_mwait(unsigned long eax
, unsigned long ecx
)
718 /* "mwait %eax, %ecx;" */
719 asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
720 :: "a" (eax
), "c" (ecx
));
723 extern void mwait_idle_with_hints(unsigned long eax
, unsigned long ecx
);
725 extern void select_idle_routine(const struct cpuinfo_x86
*c
);
727 extern unsigned long boot_option_idle_override
;
728 extern unsigned long idle_halt
;
729 extern unsigned long idle_nomwait
;
732 * on systems with caches, caches must be flashed as the absolute
733 * last instruction before going into a suspended halt. Otherwise,
734 * dirty data can linger in the cache and become stale on resume,
735 * leading to strange errors.
737 * perform a variety of operations to guarantee that the compiler
738 * will not reorder instructions. wbinvd itself is serializing
739 * so the processor will not reorder.
741 * Systems without cache can just go into halt.
743 static inline void wbinvd_halt(void)
746 /* check for clflush to determine if wbinvd is legal */
748 asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
754 extern void enable_sep_cpu(void);
755 extern int sysenter_setup(void);
757 /* Defined in head.S */
758 extern struct desc_ptr early_gdt_descr
;
760 extern void cpu_set_gdt(int);
761 extern void switch_to_new_gdt(void);
762 extern void cpu_init(void);
763 extern void init_gdt(int cpu
);
765 static inline void update_debugctlmsr(unsigned long debugctlmsr
)
767 #ifndef CONFIG_X86_DEBUGCTLMSR
768 if (boot_cpu_data
.x86
< 6)
771 wrmsrl(MSR_IA32_DEBUGCTLMSR
, debugctlmsr
);
775 * from system description table in BIOS. Mostly for MCA use, but
776 * others may find it useful:
778 extern unsigned int machine_id
;
779 extern unsigned int machine_submodel_id
;
780 extern unsigned int BIOS_revision
;
782 /* Boot loader type from the setup header: */
783 extern int bootloader_type
;
785 extern char ignore_fpu_irq
;
787 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
788 #define ARCH_HAS_PREFETCHW
789 #define ARCH_HAS_SPINLOCK_PREFETCH
792 # define BASE_PREFETCH ASM_NOP4
793 # define ARCH_HAS_PREFETCH
795 # define BASE_PREFETCH "prefetcht0 (%1)"
799 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
801 * It's not worth to care about 3dnow prefetches for the K6
802 * because they are microcoded there and very slow.
804 static inline void prefetch(const void *x
)
806 alternative_input(BASE_PREFETCH
,
813 * 3dnow prefetch to get an exclusive cache line.
814 * Useful for spinlocks to avoid one state transition in the
815 * cache coherency protocol:
817 static inline void prefetchw(const void *x
)
819 alternative_input(BASE_PREFETCH
,
825 static inline void spin_lock_prefetch(const void *x
)
832 * User space process size: 3GB (default).
834 #define TASK_SIZE PAGE_OFFSET
835 #define STACK_TOP TASK_SIZE
836 #define STACK_TOP_MAX STACK_TOP
838 #define INIT_THREAD { \
839 .sp0 = sizeof(init_stack) + (long)&init_stack, \
841 .sysenter_cs = __KERNEL_CS, \
842 .io_bitmap_ptr = NULL, \
843 .fs = __KERNEL_PERCPU, \
847 * Note that the .io_bitmap member must be extra-big. This is because
848 * the CPU will access an additional byte beyond the end of the IO
849 * permission bitmap. The extra byte must be all 1 bits, and must
850 * be within the limit.
854 .sp0 = sizeof(init_stack) + (long)&init_stack, \
855 .ss0 = __KERNEL_DS, \
856 .ss1 = __KERNEL_CS, \
857 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
859 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 }, \
862 extern unsigned long thread_saved_pc(struct task_struct
*tsk
);
864 #define THREAD_SIZE_LONGS (THREAD_SIZE/sizeof(unsigned long))
865 #define KSTK_TOP(info) \
867 unsigned long *__ptr = (unsigned long *)(info); \
868 (unsigned long)(&__ptr[THREAD_SIZE_LONGS]); \
872 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
873 * This is necessary to guarantee that the entire "struct pt_regs"
874 * is accessable even if the CPU haven't stored the SS/ESP registers
875 * on the stack (interrupt gate does not save these registers
876 * when switching to the same priv ring).
877 * Therefore beware: accessing the ss/esp fields of the
878 * "struct pt_regs" is possible, but they may contain the
879 * completely wrong values.
881 #define task_pt_regs(task) \
883 struct pt_regs *__regs__; \
884 __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
888 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
892 * User space process size. 47bits minus one guard page.
894 #define TASK_SIZE64 ((1UL << 47) - PAGE_SIZE)
896 /* This decides where the kernel will search for a free chunk of vm
897 * space during mmap's.
899 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
900 0xc0000000 : 0xFFFFe000)
902 #define TASK_SIZE (test_thread_flag(TIF_IA32) ? \
903 IA32_PAGE_OFFSET : TASK_SIZE64)
904 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_IA32)) ? \
905 IA32_PAGE_OFFSET : TASK_SIZE64)
907 #define STACK_TOP TASK_SIZE
908 #define STACK_TOP_MAX TASK_SIZE64
910 #define INIT_THREAD { \
911 .sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
915 .x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
919 * Return saved PC of a blocked thread.
920 * What is this good for? it will be always the scheduler or ret_from_fork.
922 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
924 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
925 #define KSTK_ESP(tsk) -1 /* sorry. doesn't work for syscall. */
926 #endif /* CONFIG_X86_64 */
928 extern void start_thread(struct pt_regs
*regs
, unsigned long new_ip
,
929 unsigned long new_sp
);
932 * This decides where the kernel will search for a free chunk of vm
933 * space during mmap's.
935 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
937 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
939 /* Get/set a process' ability to use the timestamp counter instruction */
940 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
941 #define SET_TSC_CTL(val) set_tsc_mode((val))
943 extern int get_tsc_mode(unsigned long adr
);
944 extern int set_tsc_mode(unsigned int val
);