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 extern void detect_extended_topology(struct cpuinfo_x86
*c
);
165 #if defined(CONFIG_X86_HT) || defined(CONFIG_X86_64)
166 extern void detect_ht(struct cpuinfo_x86
*c
);
168 static inline void detect_ht(struct cpuinfo_x86
*c
) {}
171 static inline void native_cpuid(unsigned int *eax
, unsigned int *ebx
,
172 unsigned int *ecx
, unsigned int *edx
)
174 /* ecx is often an input as well as an output. */
180 : "0" (*eax
), "2" (*ecx
));
183 static inline void load_cr3(pgd_t
*pgdir
)
185 write_cr3(__pa(pgdir
));
189 /* This is the TSS defined by the hardware. */
191 unsigned short back_link
, __blh
;
193 unsigned short ss0
, __ss0h
;
195 /* ss1 caches MSR_IA32_SYSENTER_CS: */
196 unsigned short ss1
, __ss1h
;
198 unsigned short ss2
, __ss2h
;
210 unsigned short es
, __esh
;
211 unsigned short cs
, __csh
;
212 unsigned short ss
, __ssh
;
213 unsigned short ds
, __dsh
;
214 unsigned short fs
, __fsh
;
215 unsigned short gs
, __gsh
;
216 unsigned short ldt
, __ldth
;
217 unsigned short trace
;
218 unsigned short io_bitmap_base
;
220 } __attribute__((packed
));
234 } __attribute__((packed
)) ____cacheline_aligned
;
240 #define IO_BITMAP_BITS 65536
241 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
242 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
243 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
244 #define INVALID_IO_BITMAP_OFFSET 0x8000
245 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
249 * The hardware state:
251 struct x86_hw_tss x86_tss
;
254 * The extra 1 is there because the CPU will access an
255 * additional byte beyond the end of the IO permission
256 * bitmap. The extra byte must be all 1 bits, and must
257 * be within the limit.
259 unsigned long io_bitmap
[IO_BITMAP_LONGS
+ 1];
261 * Cache the current maximum and the last task that used the bitmap:
263 unsigned long io_bitmap_max
;
264 struct thread_struct
*io_bitmap_owner
;
267 * .. and then another 0x100 bytes for the emergency kernel stack:
269 unsigned long stack
[64];
271 } ____cacheline_aligned
;
273 DECLARE_PER_CPU(struct tss_struct
, init_tss
);
276 * Save the original ist values for checking stack pointers during debugging
279 unsigned long ist
[7];
282 #define MXCSR_DEFAULT 0x1f80
284 struct i387_fsave_struct
{
285 u32 cwd
; /* FPU Control Word */
286 u32 swd
; /* FPU Status Word */
287 u32 twd
; /* FPU Tag Word */
288 u32 fip
; /* FPU IP Offset */
289 u32 fcs
; /* FPU IP Selector */
290 u32 foo
; /* FPU Operand Pointer Offset */
291 u32 fos
; /* FPU Operand Pointer Selector */
293 /* 8*10 bytes for each FP-reg = 80 bytes: */
296 /* Software status information [not touched by FSAVE ]: */
300 struct i387_fxsave_struct
{
301 u16 cwd
; /* Control Word */
302 u16 swd
; /* Status Word */
303 u16 twd
; /* Tag Word */
304 u16 fop
; /* Last Instruction Opcode */
307 u64 rip
; /* Instruction Pointer */
308 u64 rdp
; /* Data Pointer */
311 u32 fip
; /* FPU IP Offset */
312 u32 fcs
; /* FPU IP Selector */
313 u32 foo
; /* FPU Operand Offset */
314 u32 fos
; /* FPU Operand Selector */
317 u32 mxcsr
; /* MXCSR Register State */
318 u32 mxcsr_mask
; /* MXCSR Mask */
320 /* 8*16 bytes for each FP-reg = 128 bytes: */
323 /* 16*16 bytes for each XMM-reg = 256 bytes: */
328 } __attribute__((aligned(16)));
330 struct i387_soft_struct
{
338 /* 8*10 bytes for each FP-reg = 80 bytes: */
350 union thread_xstate
{
351 struct i387_fsave_struct fsave
;
352 struct i387_fxsave_struct fxsave
;
353 struct i387_soft_struct soft
;
357 DECLARE_PER_CPU(struct orig_ist
, orig_ist
);
360 extern void print_cpu_info(struct cpuinfo_x86
*);
361 extern unsigned int xstate_size
;
362 extern void free_thread_xstate(struct task_struct
*);
363 extern struct kmem_cache
*task_xstate_cachep
;
364 extern void init_scattered_cpuid_features(struct cpuinfo_x86
*c
);
365 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86
*c
);
366 extern unsigned short num_cache_leaves
;
368 struct thread_struct
{
369 /* Cached TLS descriptors: */
370 struct desc_struct tls_array
[GDT_ENTRY_TLS_ENTRIES
];
374 unsigned long sysenter_cs
;
376 unsigned long usersp
; /* Copy from PDA */
379 unsigned short fsindex
;
380 unsigned short gsindex
;
385 /* Hardware debugging registers: */
386 unsigned long debugreg0
;
387 unsigned long debugreg1
;
388 unsigned long debugreg2
;
389 unsigned long debugreg3
;
390 unsigned long debugreg6
;
391 unsigned long debugreg7
;
394 unsigned long trap_no
;
395 unsigned long error_code
;
396 /* floating point and extended processor state */
397 union thread_xstate
*xstate
;
399 /* Virtual 86 mode info */
400 struct vm86_struct __user
*vm86_info
;
401 unsigned long screen_bitmap
;
402 unsigned long v86flags
;
403 unsigned long v86mask
;
404 unsigned long saved_sp0
;
405 unsigned int saved_fs
;
406 unsigned int saved_gs
;
408 /* IO permissions: */
409 unsigned long *io_bitmap_ptr
;
411 /* Max allowed port in the bitmap, in bytes: */
412 unsigned io_bitmap_max
;
413 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
414 unsigned long debugctlmsr
;
415 /* Debug Store - if not 0 points to a DS Save Area configuration;
416 * goes into MSR_IA32_DS_AREA */
417 unsigned long ds_area_msr
;
420 static inline unsigned long native_get_debugreg(int regno
)
422 unsigned long val
= 0; /* Damn you, gcc! */
426 asm("mov %%db0, %0" :"=r" (val
));
429 asm("mov %%db1, %0" :"=r" (val
));
432 asm("mov %%db2, %0" :"=r" (val
));
435 asm("mov %%db3, %0" :"=r" (val
));
438 asm("mov %%db6, %0" :"=r" (val
));
441 asm("mov %%db7, %0" :"=r" (val
));
449 static inline void native_set_debugreg(int regno
, unsigned long value
)
453 asm("mov %0, %%db0" ::"r" (value
));
456 asm("mov %0, %%db1" ::"r" (value
));
459 asm("mov %0, %%db2" ::"r" (value
));
462 asm("mov %0, %%db3" ::"r" (value
));
465 asm("mov %0, %%db6" ::"r" (value
));
468 asm("mov %0, %%db7" ::"r" (value
));
476 * Set IOPL bits in EFLAGS from given mask
478 static inline void native_set_iopl_mask(unsigned mask
)
483 asm volatile ("pushfl;"
490 : "i" (~X86_EFLAGS_IOPL
), "r" (mask
));
495 native_load_sp0(struct tss_struct
*tss
, struct thread_struct
*thread
)
497 tss
->x86_tss
.sp0
= thread
->sp0
;
499 /* Only happens when SEP is enabled, no need to test "SEP"arately: */
500 if (unlikely(tss
->x86_tss
.ss1
!= thread
->sysenter_cs
)) {
501 tss
->x86_tss
.ss1
= thread
->sysenter_cs
;
502 wrmsr(MSR_IA32_SYSENTER_CS
, thread
->sysenter_cs
, 0);
507 static inline void native_swapgs(void)
510 asm volatile("swapgs" ::: "memory");
514 #ifdef CONFIG_PARAVIRT
515 #include <asm/paravirt.h>
517 #define __cpuid native_cpuid
518 #define paravirt_enabled() 0
521 * These special macros can be used to get or set a debugging register
523 #define get_debugreg(var, register) \
524 (var) = native_get_debugreg(register)
525 #define set_debugreg(value, register) \
526 native_set_debugreg(register, value)
528 static inline void load_sp0(struct tss_struct
*tss
,
529 struct thread_struct
*thread
)
531 native_load_sp0(tss
, thread
);
534 #define set_iopl_mask native_set_iopl_mask
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");
701 static inline void __monitor(const void *eax
, unsigned long ecx
,
704 /* "monitor %eax, %ecx, %edx;" */
705 asm volatile(".byte 0x0f, 0x01, 0xc8;"
706 :: "a" (eax
), "c" (ecx
), "d"(edx
));
709 static inline void __mwait(unsigned long eax
, unsigned long ecx
)
711 /* "mwait %eax, %ecx;" */
712 asm volatile(".byte 0x0f, 0x01, 0xc9;"
713 :: "a" (eax
), "c" (ecx
));
716 static inline void __sti_mwait(unsigned long eax
, unsigned long ecx
)
719 /* "mwait %eax, %ecx;" */
720 asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
721 :: "a" (eax
), "c" (ecx
));
724 extern void mwait_idle_with_hints(unsigned long eax
, unsigned long ecx
);
726 extern void select_idle_routine(const struct cpuinfo_x86
*c
);
728 extern unsigned long boot_option_idle_override
;
729 extern unsigned long idle_halt
;
730 extern unsigned long idle_nomwait
;
733 * on systems with caches, caches must be flashed as the absolute
734 * last instruction before going into a suspended halt. Otherwise,
735 * dirty data can linger in the cache and become stale on resume,
736 * leading to strange errors.
738 * perform a variety of operations to guarantee that the compiler
739 * will not reorder instructions. wbinvd itself is serializing
740 * so the processor will not reorder.
742 * Systems without cache can just go into halt.
744 static inline void wbinvd_halt(void)
747 /* check for clflush to determine if wbinvd is legal */
749 asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
755 extern void enable_sep_cpu(void);
756 extern int sysenter_setup(void);
758 /* Defined in head.S */
759 extern struct desc_ptr early_gdt_descr
;
761 extern void cpu_set_gdt(int);
762 extern void switch_to_new_gdt(void);
763 extern void cpu_init(void);
764 extern void init_gdt(int cpu
);
766 static inline void update_debugctlmsr(unsigned long debugctlmsr
)
768 #ifndef CONFIG_X86_DEBUGCTLMSR
769 if (boot_cpu_data
.x86
< 6)
772 wrmsrl(MSR_IA32_DEBUGCTLMSR
, debugctlmsr
);
776 * from system description table in BIOS. Mostly for MCA use, but
777 * others may find it useful:
779 extern unsigned int machine_id
;
780 extern unsigned int machine_submodel_id
;
781 extern unsigned int BIOS_revision
;
783 /* Boot loader type from the setup header: */
784 extern int bootloader_type
;
786 extern char ignore_fpu_irq
;
788 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
789 #define ARCH_HAS_PREFETCHW
790 #define ARCH_HAS_SPINLOCK_PREFETCH
793 # define BASE_PREFETCH ASM_NOP4
794 # define ARCH_HAS_PREFETCH
796 # define BASE_PREFETCH "prefetcht0 (%1)"
800 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
802 * It's not worth to care about 3dnow prefetches for the K6
803 * because they are microcoded there and very slow.
805 static inline void prefetch(const void *x
)
807 alternative_input(BASE_PREFETCH
,
814 * 3dnow prefetch to get an exclusive cache line.
815 * Useful for spinlocks to avoid one state transition in the
816 * cache coherency protocol:
818 static inline void prefetchw(const void *x
)
820 alternative_input(BASE_PREFETCH
,
826 static inline void spin_lock_prefetch(const void *x
)
833 * User space process size: 3GB (default).
835 #define TASK_SIZE PAGE_OFFSET
836 #define STACK_TOP TASK_SIZE
837 #define STACK_TOP_MAX STACK_TOP
839 #define INIT_THREAD { \
840 .sp0 = sizeof(init_stack) + (long)&init_stack, \
842 .sysenter_cs = __KERNEL_CS, \
843 .io_bitmap_ptr = NULL, \
844 .fs = __KERNEL_PERCPU, \
848 * Note that the .io_bitmap member must be extra-big. This is because
849 * the CPU will access an additional byte beyond the end of the IO
850 * permission bitmap. The extra byte must be all 1 bits, and must
851 * be within the limit.
855 .sp0 = sizeof(init_stack) + (long)&init_stack, \
856 .ss0 = __KERNEL_DS, \
857 .ss1 = __KERNEL_CS, \
858 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
860 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 }, \
863 extern unsigned long thread_saved_pc(struct task_struct
*tsk
);
865 #define THREAD_SIZE_LONGS (THREAD_SIZE/sizeof(unsigned long))
866 #define KSTK_TOP(info) \
868 unsigned long *__ptr = (unsigned long *)(info); \
869 (unsigned long)(&__ptr[THREAD_SIZE_LONGS]); \
873 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
874 * This is necessary to guarantee that the entire "struct pt_regs"
875 * is accessable even if the CPU haven't stored the SS/ESP registers
876 * on the stack (interrupt gate does not save these registers
877 * when switching to the same priv ring).
878 * Therefore beware: accessing the ss/esp fields of the
879 * "struct pt_regs" is possible, but they may contain the
880 * completely wrong values.
882 #define task_pt_regs(task) \
884 struct pt_regs *__regs__; \
885 __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
889 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
893 * User space process size. 47bits minus one guard page.
895 #define TASK_SIZE64 ((1UL << 47) - PAGE_SIZE)
897 /* This decides where the kernel will search for a free chunk of vm
898 * space during mmap's.
900 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
901 0xc0000000 : 0xFFFFe000)
903 #define TASK_SIZE (test_thread_flag(TIF_IA32) ? \
904 IA32_PAGE_OFFSET : TASK_SIZE64)
905 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_IA32)) ? \
906 IA32_PAGE_OFFSET : TASK_SIZE64)
908 #define STACK_TOP TASK_SIZE
909 #define STACK_TOP_MAX TASK_SIZE64
911 #define INIT_THREAD { \
912 .sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
916 .x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
920 * Return saved PC of a blocked thread.
921 * What is this good for? it will be always the scheduler or ret_from_fork.
923 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
925 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
926 #define KSTK_ESP(tsk) -1 /* sorry. doesn't work for syscall. */
927 #endif /* CONFIG_X86_64 */
929 extern void start_thread(struct pt_regs
*regs
, unsigned long new_ip
,
930 unsigned long new_sp
);
933 * This decides where the kernel will search for a free chunk of vm
934 * space during mmap's.
936 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
938 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
940 /* Get/set a process' ability to use the timestamp counter instruction */
941 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
942 #define SET_TSC_CTL(val) set_tsc_mode((val))
944 extern int get_tsc_mode(unsigned long adr
);
945 extern int set_tsc_mode(unsigned int val
);