2 * include/asm-i386/processor.h
4 * Copyright (C) 1994 Linus Torvalds
7 #ifndef __ASM_I386_PROCESSOR_H
8 #define __ASM_I386_PROCESSOR_H
11 #include <asm/math_emu.h>
12 #include <asm/segment.h>
14 #include <asm/types.h>
15 #include <asm/sigcontext.h>
16 #include <asm/cpufeature.h>
18 #include <asm/system.h>
19 #include <linux/cache.h>
20 #include <linux/threads.h>
21 #include <asm/percpu.h>
22 #include <linux/cpumask.h>
24 /* flag for disabling the tsc */
25 extern int tsc_disable
;
31 #define desc_empty(desc) \
32 (!((desc)->a | (desc)->b))
34 #define desc_equal(desc1, desc2) \
35 (((desc1)->a == (desc2)->a) && ((desc1)->b == (desc2)->b))
37 * Default implementation of macro that returns current
38 * instruction pointer ("program counter").
40 #define current_text_addr() ({ void *pc; __asm__("movl $1f,%0\n1:":"=g" (pc)); pc; })
43 * CPU type and hardware bug flags. Kept separately for each CPU.
44 * Members of this structure are referenced in head.S, so think twice
45 * before touching them. [mj]
49 __u8 x86
; /* CPU family */
50 __u8 x86_vendor
; /* CPU vendor */
53 char wp_works_ok
; /* It doesn't on 386's */
54 char hlt_works_ok
; /* Problems on some 486Dx4's and old 386's */
57 int cpuid_level
; /* Maximum supported CPUID level, -1=no CPUID */
58 unsigned long x86_capability
[NCAPINTS
];
59 char x86_vendor_id
[16];
60 char x86_model_id
[64];
61 int x86_cache_size
; /* in KB - valid for CPUS which support this
63 int x86_cache_alignment
; /* In bytes */
69 unsigned long loops_per_jiffy
;
71 cpumask_t llc_shared_map
; /* cpus sharing the last level cache */
73 unsigned char x86_max_cores
; /* cpuid returned max cores value */
76 unsigned char booted_cores
; /* number of cores as seen by OS */
77 __u8 phys_proc_id
; /* Physical processor id. */
78 __u8 cpu_core_id
; /* Core id */
80 } __attribute__((__aligned__(SMP_CACHE_BYTES
)));
82 #define X86_VENDOR_INTEL 0
83 #define X86_VENDOR_CYRIX 1
84 #define X86_VENDOR_AMD 2
85 #define X86_VENDOR_UMC 3
86 #define X86_VENDOR_NEXGEN 4
87 #define X86_VENDOR_CENTAUR 5
88 #define X86_VENDOR_RISE 6
89 #define X86_VENDOR_TRANSMETA 7
90 #define X86_VENDOR_NSC 8
91 #define X86_VENDOR_NUM 9
92 #define X86_VENDOR_UNKNOWN 0xff
95 * capabilities of CPUs
98 extern struct cpuinfo_x86 boot_cpu_data
;
99 extern struct cpuinfo_x86 new_cpu_data
;
100 extern struct tss_struct doublefault_tss
;
101 DECLARE_PER_CPU(struct tss_struct
, init_tss
);
104 extern struct cpuinfo_x86 cpu_data
[];
105 #define current_cpu_data cpu_data[smp_processor_id()]
107 #define cpu_data (&boot_cpu_data)
108 #define current_cpu_data boot_cpu_data
111 extern int cpu_llc_id
[NR_CPUS
];
112 extern char ignore_fpu_irq
;
114 extern void identify_cpu(struct cpuinfo_x86
*);
115 extern void print_cpu_info(struct cpuinfo_x86
*);
116 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86
*c
);
117 extern unsigned short num_cache_leaves
;
120 extern void detect_ht(struct cpuinfo_x86
*c
);
122 static inline void detect_ht(struct cpuinfo_x86
*c
) {}
128 #define X86_EFLAGS_CF 0x00000001 /* Carry Flag */
129 #define X86_EFLAGS_PF 0x00000004 /* Parity Flag */
130 #define X86_EFLAGS_AF 0x00000010 /* Auxillary carry Flag */
131 #define X86_EFLAGS_ZF 0x00000040 /* Zero Flag */
132 #define X86_EFLAGS_SF 0x00000080 /* Sign Flag */
133 #define X86_EFLAGS_TF 0x00000100 /* Trap Flag */
134 #define X86_EFLAGS_IF 0x00000200 /* Interrupt Flag */
135 #define X86_EFLAGS_DF 0x00000400 /* Direction Flag */
136 #define X86_EFLAGS_OF 0x00000800 /* Overflow Flag */
137 #define X86_EFLAGS_IOPL 0x00003000 /* IOPL mask */
138 #define X86_EFLAGS_NT 0x00004000 /* Nested Task */
139 #define X86_EFLAGS_RF 0x00010000 /* Resume Flag */
140 #define X86_EFLAGS_VM 0x00020000 /* Virtual Mode */
141 #define X86_EFLAGS_AC 0x00040000 /* Alignment Check */
142 #define X86_EFLAGS_VIF 0x00080000 /* Virtual Interrupt Flag */
143 #define X86_EFLAGS_VIP 0x00100000 /* Virtual Interrupt Pending */
144 #define X86_EFLAGS_ID 0x00200000 /* CPUID detection flag */
146 static inline void __cpuid(unsigned int *eax
, unsigned int *ebx
,
147 unsigned int *ecx
, unsigned int *edx
)
149 /* ecx is often an input as well as an output. */
155 : "0" (*eax
), "2" (*ecx
));
159 * Generic CPUID function
160 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
161 * resulting in stale register contents being returned.
163 static inline void cpuid(unsigned int op
, unsigned int *eax
, unsigned int *ebx
, unsigned int *ecx
, unsigned int *edx
)
167 __cpuid(eax
, ebx
, ecx
, edx
);
170 /* Some CPUID calls want 'count' to be placed in ecx */
171 static inline void cpuid_count(int op
, int count
, int *eax
, int *ebx
, int *ecx
,
176 __cpuid(eax
, ebx
, ecx
, edx
);
180 * CPUID functions returning a single datum
182 static inline unsigned int cpuid_eax(unsigned int op
)
184 unsigned int eax
, ebx
, ecx
, edx
;
186 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
189 static inline unsigned int cpuid_ebx(unsigned int op
)
191 unsigned int eax
, ebx
, ecx
, edx
;
193 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
196 static inline unsigned int cpuid_ecx(unsigned int op
)
198 unsigned int eax
, ebx
, ecx
, edx
;
200 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
203 static inline unsigned int cpuid_edx(unsigned int op
)
205 unsigned int eax
, ebx
, ecx
, edx
;
207 cpuid(op
, &eax
, &ebx
, &ecx
, &edx
);
211 #define load_cr3(pgdir) write_cr3(__pa(pgdir))
214 * Intel CPU features in CR4
216 #define X86_CR4_VME 0x0001 /* enable vm86 extensions */
217 #define X86_CR4_PVI 0x0002 /* virtual interrupts flag enable */
218 #define X86_CR4_TSD 0x0004 /* disable time stamp at ipl 3 */
219 #define X86_CR4_DE 0x0008 /* enable debugging extensions */
220 #define X86_CR4_PSE 0x0010 /* enable page size extensions */
221 #define X86_CR4_PAE 0x0020 /* enable physical address extensions */
222 #define X86_CR4_MCE 0x0040 /* Machine check enable */
223 #define X86_CR4_PGE 0x0080 /* enable global pages */
224 #define X86_CR4_PCE 0x0100 /* enable performance counters at ipl 3 */
225 #define X86_CR4_OSFXSR 0x0200 /* enable fast FPU save and restore */
226 #define X86_CR4_OSXMMEXCPT 0x0400 /* enable unmasked SSE exceptions */
229 * Save the cr4 feature set we're using (ie
230 * Pentium 4MB enable and PPro Global page
231 * enable), so that any CPU's that boot up
232 * after us can get the correct flags.
234 extern unsigned long mmu_cr4_features
;
236 static inline void set_in_cr4 (unsigned long mask
)
239 mmu_cr4_features
|= mask
;
245 static inline void clear_in_cr4 (unsigned long mask
)
248 mmu_cr4_features
&= ~mask
;
255 * NSC/Cyrix CPU configuration register indexes
258 #define CX86_PCR0 0x20
259 #define CX86_GCR 0xb8
260 #define CX86_CCR0 0xc0
261 #define CX86_CCR1 0xc1
262 #define CX86_CCR2 0xc2
263 #define CX86_CCR3 0xc3
264 #define CX86_CCR4 0xe8
265 #define CX86_CCR5 0xe9
266 #define CX86_CCR6 0xea
267 #define CX86_CCR7 0xeb
268 #define CX86_PCR1 0xf0
269 #define CX86_DIR0 0xfe
270 #define CX86_DIR1 0xff
271 #define CX86_ARR_BASE 0xc4
272 #define CX86_RCR_BASE 0xdc
275 * NSC/Cyrix CPU indexed register access macros
278 #define getCx86(reg) ({ outb((reg), 0x22); inb(0x23); })
280 #define setCx86(reg, data) do { \
282 outb((data), 0x23); \
285 /* Stop speculative execution */
286 static inline void sync_core(void)
289 asm volatile("cpuid" : "=a" (tmp
) : "0" (1) : "ebx","ecx","edx","memory");
292 static inline void __monitor(const void *eax
, unsigned long ecx
,
295 /* "monitor %eax,%ecx,%edx;" */
297 ".byte 0x0f,0x01,0xc8;"
298 : :"a" (eax
), "c" (ecx
), "d"(edx
));
301 static inline void __mwait(unsigned long eax
, unsigned long ecx
)
303 /* "mwait %eax,%ecx;" */
305 ".byte 0x0f,0x01,0xc9;"
306 : :"a" (eax
), "c" (ecx
));
309 /* from system description table in BIOS. Mostly for MCA use, but
310 others may find it useful. */
311 extern unsigned int machine_id
;
312 extern unsigned int machine_submodel_id
;
313 extern unsigned int BIOS_revision
;
314 extern unsigned int mca_pentium_flag
;
316 /* Boot loader type from the setup header */
317 extern int bootloader_type
;
320 * User space process size: 3GB (default).
322 #define TASK_SIZE (PAGE_OFFSET)
324 /* This decides where the kernel will search for a free chunk of vm
325 * space during mmap's.
327 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
329 #define HAVE_ARCH_PICK_MMAP_LAYOUT
334 #define IO_BITMAP_BITS 65536
335 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
336 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
337 #define IO_BITMAP_OFFSET offsetof(struct tss_struct,io_bitmap)
338 #define INVALID_IO_BITMAP_OFFSET 0x8000
339 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
341 struct i387_fsave_struct
{
349 long st_space
[20]; /* 8*10 bytes for each FP-reg = 80 bytes */
350 long status
; /* software status information */
353 struct i387_fxsave_struct
{
364 long st_space
[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
365 long xmm_space
[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
367 } __attribute__ ((aligned (16)));
369 struct i387_soft_struct
{
377 long st_space
[20]; /* 8*10 bytes for each FP-reg = 80 bytes */
378 unsigned char ftop
, changed
, lookahead
, no_update
, rm
, alimit
;
380 unsigned long entry_eip
;
384 struct i387_fsave_struct fsave
;
385 struct i387_fxsave_struct fxsave
;
386 struct i387_soft_struct soft
;
393 struct thread_struct
;
396 unsigned short back_link
,__blh
;
398 unsigned short ss0
,__ss0h
;
400 unsigned short ss1
,__ss1h
; /* ss1 is used to cache MSR_IA32_SYSENTER_CS */
402 unsigned short ss2
,__ss2h
;
405 unsigned long eflags
;
406 unsigned long eax
,ecx
,edx
,ebx
;
411 unsigned short es
, __esh
;
412 unsigned short cs
, __csh
;
413 unsigned short ss
, __ssh
;
414 unsigned short ds
, __dsh
;
415 unsigned short fs
, __fsh
;
416 unsigned short gs
, __gsh
;
417 unsigned short ldt
, __ldth
;
418 unsigned short trace
, io_bitmap_base
;
420 * The extra 1 is there because the CPU will access an
421 * additional byte beyond the end of the IO permission
422 * bitmap. The extra byte must be all 1 bits, and must
423 * be within the limit.
425 unsigned long io_bitmap
[IO_BITMAP_LONGS
+ 1];
427 * Cache the current maximum and the last task that used the bitmap:
429 unsigned long io_bitmap_max
;
430 struct thread_struct
*io_bitmap_owner
;
432 * pads the TSS to be cacheline-aligned (size is 0x100)
434 unsigned long __cacheline_filler
[35];
436 * .. and then another 0x100 bytes for emergency kernel stack
438 unsigned long stack
[64];
439 } __attribute__((packed
));
441 #define ARCH_MIN_TASKALIGN 16
443 struct thread_struct
{
444 /* cached TLS descriptors. */
445 struct desc_struct tls_array
[GDT_ENTRY_TLS_ENTRIES
];
447 unsigned long sysenter_cs
;
452 /* Hardware debugging registers */
453 unsigned long debugreg
[8]; /* %%db0-7 debug registers */
455 unsigned long cr2
, trap_no
, error_code
;
456 /* floating point info */
457 union i387_union i387
;
458 /* virtual 86 mode info */
459 struct vm86_struct __user
* vm86_info
;
460 unsigned long screen_bitmap
;
461 unsigned long v86flags
, v86mask
, saved_esp0
;
462 unsigned int saved_fs
, saved_gs
;
464 unsigned long *io_bitmap_ptr
;
466 /* max allowed port in the bitmap, in bytes: */
467 unsigned long io_bitmap_max
;
470 #define INIT_THREAD { \
472 .sysenter_cs = __KERNEL_CS, \
473 .io_bitmap_ptr = NULL, \
477 * Note that the .io_bitmap member must be extra-big. This is because
478 * the CPU will access an additional byte beyond the end of the IO
479 * permission bitmap. The extra byte must be all 1 bits, and must
480 * be within the limit.
483 .esp0 = sizeof(init_stack) + (long)&init_stack, \
484 .ss0 = __KERNEL_DS, \
485 .ss1 = __KERNEL_CS, \
486 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
487 .io_bitmap = { [ 0 ... IO_BITMAP_LONGS] = ~0 }, \
490 static inline void load_esp0(struct tss_struct
*tss
, struct thread_struct
*thread
)
492 tss
->esp0
= thread
->esp0
;
493 /* This can only happen when SEP is enabled, no need to test "SEP"arately */
494 if (unlikely(tss
->ss1
!= thread
->sysenter_cs
)) {
495 tss
->ss1
= thread
->sysenter_cs
;
496 wrmsr(MSR_IA32_SYSENTER_CS
, thread
->sysenter_cs
, 0);
500 #define start_thread(regs, new_eip, new_esp) do { \
501 __asm__("movl %0,%%fs ; movl %0,%%gs": :"r" (0)); \
503 regs->xds = __USER_DS; \
504 regs->xes = __USER_DS; \
505 regs->xss = __USER_DS; \
506 regs->xcs = __USER_CS; \
507 regs->eip = new_eip; \
508 regs->esp = new_esp; \
512 * These special macros can be used to get or set a debugging register
514 #define get_debugreg(var, register) \
515 __asm__("movl %%db" #register ", %0" \
517 #define set_debugreg(value, register) \
518 __asm__("movl %0,%%db" #register \
523 * Set IOPL bits in EFLAGS from given mask
525 static inline void set_iopl_mask(unsigned mask
)
528 __asm__
__volatile__ ("pushfl;"
535 : "i" (~X86_EFLAGS_IOPL
), "r" (mask
));
538 /* Forward declaration, a strange C thing */
542 /* Free all resources held by a thread. */
543 extern void release_thread(struct task_struct
*);
545 /* Prepare to copy thread state - unlazy all lazy status */
546 extern void prepare_to_copy(struct task_struct
*tsk
);
549 * create a kernel thread without removing it from tasklists
551 extern int kernel_thread(int (*fn
)(void *), void * arg
, unsigned long flags
);
553 extern unsigned long thread_saved_pc(struct task_struct
*tsk
);
554 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
, unsigned long *stack
);
556 unsigned long get_wchan(struct task_struct
*p
);
558 #define THREAD_SIZE_LONGS (THREAD_SIZE/sizeof(unsigned long))
559 #define KSTK_TOP(info) \
561 unsigned long *__ptr = (unsigned long *)(info); \
562 (unsigned long)(&__ptr[THREAD_SIZE_LONGS]); \
566 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
567 * This is necessary to guarantee that the entire "struct pt_regs"
568 * is accessable even if the CPU haven't stored the SS/ESP registers
569 * on the stack (interrupt gate does not save these registers
570 * when switching to the same priv ring).
571 * Therefore beware: accessing the xss/esp fields of the
572 * "struct pt_regs" is possible, but they may contain the
573 * completely wrong values.
575 #define task_pt_regs(task) \
577 struct pt_regs *__regs__; \
578 __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
582 #define KSTK_EIP(task) (task_pt_regs(task)->eip)
583 #define KSTK_ESP(task) (task_pt_regs(task)->esp)
586 struct microcode_header
{
594 unsigned int datasize
;
595 unsigned int totalsize
;
596 unsigned int reserved
[3];
600 struct microcode_header hdr
;
601 unsigned int bits
[0];
604 typedef struct microcode microcode_t
;
605 typedef struct microcode_header microcode_header_t
;
607 /* microcode format is extended from prescott processors */
608 struct extended_signature
{
614 struct extended_sigtable
{
617 unsigned int reserved
[3];
618 struct extended_signature sigs
[0];
621 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
622 static inline void rep_nop(void)
624 __asm__
__volatile__("rep;nop": : :"memory");
627 #define cpu_relax() rep_nop()
629 /* generic versions from gas */
630 #define GENERIC_NOP1 ".byte 0x90\n"
631 #define GENERIC_NOP2 ".byte 0x89,0xf6\n"
632 #define GENERIC_NOP3 ".byte 0x8d,0x76,0x00\n"
633 #define GENERIC_NOP4 ".byte 0x8d,0x74,0x26,0x00\n"
634 #define GENERIC_NOP5 GENERIC_NOP1 GENERIC_NOP4
635 #define GENERIC_NOP6 ".byte 0x8d,0xb6,0x00,0x00,0x00,0x00\n"
636 #define GENERIC_NOP7 ".byte 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00\n"
637 #define GENERIC_NOP8 GENERIC_NOP1 GENERIC_NOP7
640 #define K8_NOP1 GENERIC_NOP1
641 #define K8_NOP2 ".byte 0x66,0x90\n"
642 #define K8_NOP3 ".byte 0x66,0x66,0x90\n"
643 #define K8_NOP4 ".byte 0x66,0x66,0x66,0x90\n"
644 #define K8_NOP5 K8_NOP3 K8_NOP2
645 #define K8_NOP6 K8_NOP3 K8_NOP3
646 #define K8_NOP7 K8_NOP4 K8_NOP3
647 #define K8_NOP8 K8_NOP4 K8_NOP4
650 /* uses eax dependencies (arbitary choice) */
651 #define K7_NOP1 GENERIC_NOP1
652 #define K7_NOP2 ".byte 0x8b,0xc0\n"
653 #define K7_NOP3 ".byte 0x8d,0x04,0x20\n"
654 #define K7_NOP4 ".byte 0x8d,0x44,0x20,0x00\n"
655 #define K7_NOP5 K7_NOP4 ASM_NOP1
656 #define K7_NOP6 ".byte 0x8d,0x80,0,0,0,0\n"
657 #define K7_NOP7 ".byte 0x8D,0x04,0x05,0,0,0,0\n"
658 #define K7_NOP8 K7_NOP7 ASM_NOP1
661 #define ASM_NOP1 K8_NOP1
662 #define ASM_NOP2 K8_NOP2
663 #define ASM_NOP3 K8_NOP3
664 #define ASM_NOP4 K8_NOP4
665 #define ASM_NOP5 K8_NOP5
666 #define ASM_NOP6 K8_NOP6
667 #define ASM_NOP7 K8_NOP7
668 #define ASM_NOP8 K8_NOP8
669 #elif defined(CONFIG_MK7)
670 #define ASM_NOP1 K7_NOP1
671 #define ASM_NOP2 K7_NOP2
672 #define ASM_NOP3 K7_NOP3
673 #define ASM_NOP4 K7_NOP4
674 #define ASM_NOP5 K7_NOP5
675 #define ASM_NOP6 K7_NOP6
676 #define ASM_NOP7 K7_NOP7
677 #define ASM_NOP8 K7_NOP8
679 #define ASM_NOP1 GENERIC_NOP1
680 #define ASM_NOP2 GENERIC_NOP2
681 #define ASM_NOP3 GENERIC_NOP3
682 #define ASM_NOP4 GENERIC_NOP4
683 #define ASM_NOP5 GENERIC_NOP5
684 #define ASM_NOP6 GENERIC_NOP6
685 #define ASM_NOP7 GENERIC_NOP7
686 #define ASM_NOP8 GENERIC_NOP8
689 #define ASM_NOP_MAX 8
691 /* Prefetch instructions for Pentium III and AMD Athlon */
692 /* It's not worth to care about 3dnow! prefetches for the K6
693 because they are microcoded there and very slow.
694 However we don't do prefetches for pre XP Athlons currently
695 That should be fixed. */
696 #define ARCH_HAS_PREFETCH
697 static inline void prefetch(const void *x
)
699 alternative_input(ASM_NOP4
,
705 #define ARCH_HAS_PREFETCH
706 #define ARCH_HAS_PREFETCHW
707 #define ARCH_HAS_SPINLOCK_PREFETCH
709 /* 3dnow! prefetch to get an exclusive cache line. Useful for
710 spinlocks to avoid one state transition in the cache coherency protocol. */
711 static inline void prefetchw(const void *x
)
713 alternative_input(ASM_NOP4
,
718 #define spin_lock_prefetch(x) prefetchw(x)
720 extern void select_idle_routine(const struct cpuinfo_x86
*c
);
722 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
724 extern unsigned long boot_option_idle_override
;
725 extern void enable_sep_cpu(void);
726 extern int sysenter_setup(void);
728 #endif /* __ASM_I386_PROCESSOR_H */