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>
23 #include <linux/init.h>
24 #include <asm/processor-flags.h>
26 /* flag for disabling the tsc */
27 extern int tsc_disable;
33 #define desc_empty(desc) \
34 (!((desc)->a | (desc)->b))
36 #define desc_equal(desc1, desc2) \
37 (((desc1)->a == (desc2)->a) && ((desc1)->b == (desc2)->b))
39 * Default implementation of macro that returns current
40 * instruction pointer ("program counter").
42 #define current_text_addr() ({ void *pc; __asm__("movl $1f,%0\n1:":"=g" (pc)); pc; })
45 * CPU type and hardware bug flags. Kept separately for each CPU.
46 * Members of this structure are referenced in head.S, so think twice
47 * before touching them. [mj]
51 __u8 x86; /* CPU family */
52 __u8 x86_vendor; /* CPU vendor */
55 char wp_works_ok; /* It doesn't on 386's */
56 char hlt_works_ok; /* Problems on some 486Dx4's and old 386's */
59 int cpuid_level; /* Maximum supported CPUID level, -1=no CPUID */
60 unsigned long x86_capability[NCAPINTS];
61 char x86_vendor_id[16];
62 char x86_model_id[64];
63 int x86_cache_size; /* in KB - valid for CPUS which support this
65 int x86_cache_alignment; /* In bytes */
71 unsigned long loops_per_jiffy;
73 cpumask_t llc_shared_map; /* cpus sharing the last level cache */
75 unsigned char x86_max_cores; /* cpuid returned max cores value */
77 unsigned short x86_clflush_size;
79 unsigned char booted_cores; /* number of cores as seen by OS */
80 __u8 phys_proc_id; /* Physical processor id. */
81 __u8 cpu_core_id; /* Core id */
83 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
85 #define X86_VENDOR_INTEL 0
86 #define X86_VENDOR_CYRIX 1
87 #define X86_VENDOR_AMD 2
88 #define X86_VENDOR_UMC 3
89 #define X86_VENDOR_NEXGEN 4
90 #define X86_VENDOR_CENTAUR 5
91 #define X86_VENDOR_TRANSMETA 7
92 #define X86_VENDOR_NSC 8
93 #define X86_VENDOR_NUM 9
94 #define X86_VENDOR_UNKNOWN 0xff
97 * capabilities of CPUs
100 extern struct cpuinfo_x86 boot_cpu_data;
101 extern struct cpuinfo_x86 new_cpu_data;
102 extern struct tss_struct doublefault_tss;
103 DECLARE_PER_CPU(struct tss_struct, init_tss);
106 extern struct cpuinfo_x86 cpu_data[];
107 #define current_cpu_data cpu_data[smp_processor_id()]
109 #define cpu_data (&boot_cpu_data)
110 #define current_cpu_data boot_cpu_data
113 extern int cpu_llc_id[NR_CPUS];
114 extern char ignore_fpu_irq;
116 void __init cpu_detect(struct cpuinfo_x86 *c);
118 extern void identify_boot_cpu(void);
119 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
120 extern void print_cpu_info(struct cpuinfo_x86 *);
121 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
122 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
123 extern unsigned short num_cache_leaves;
126 extern void detect_ht(struct cpuinfo_x86 *c);
128 static inline void detect_ht(struct cpuinfo_x86 *c) {}
131 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
132 unsigned int *ecx, unsigned int *edx)
134 /* ecx is often an input as well as an output. */
140 : "0" (*eax), "2" (*ecx));
143 #define load_cr3(pgdir) write_cr3(__pa(pgdir))
146 * Save the cr4 feature set we're using (ie
147 * Pentium 4MB enable and PPro Global page
148 * enable), so that any CPU's that boot up
149 * after us can get the correct flags.
151 extern unsigned long mmu_cr4_features;
153 static inline void set_in_cr4 (unsigned long mask)
156 mmu_cr4_features |= mask;
162 static inline void clear_in_cr4 (unsigned long mask)
165 mmu_cr4_features &= ~mask;
171 /* Stop speculative execution */
172 static inline void sync_core(void)
175 asm volatile("cpuid" : "=a" (tmp) : "0" (1) : "ebx","ecx","edx","memory");
178 static inline void __monitor(const void *eax, unsigned long ecx,
181 /* "monitor %eax,%ecx,%edx;" */
183 ".byte 0x0f,0x01,0xc8;"
184 : :"a" (eax), "c" (ecx), "d"(edx));
187 static inline void __mwait(unsigned long eax, unsigned long ecx)
189 /* "mwait %eax,%ecx;" */
191 ".byte 0x0f,0x01,0xc9;"
192 : :"a" (eax), "c" (ecx));
195 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
197 /* from system description table in BIOS. Mostly for MCA use, but
198 others may find it useful. */
199 extern unsigned int machine_id;
200 extern unsigned int machine_submodel_id;
201 extern unsigned int BIOS_revision;
202 extern unsigned int mca_pentium_flag;
204 /* Boot loader type from the setup header */
205 extern int bootloader_type;
208 * User space process size: 3GB (default).
210 #define TASK_SIZE (PAGE_OFFSET)
212 /* This decides where the kernel will search for a free chunk of vm
213 * space during mmap's.
215 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
217 #define HAVE_ARCH_PICK_MMAP_LAYOUT
219 extern void hard_disable_TSC(void);
220 extern void disable_TSC(void);
221 extern void hard_enable_TSC(void);
226 #define IO_BITMAP_BITS 65536
227 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
228 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
229 #define IO_BITMAP_OFFSET offsetof(struct tss_struct,io_bitmap)
230 #define INVALID_IO_BITMAP_OFFSET 0x8000
231 #define INVALID_IO_BITMAP_OFFSET_LAZY 0x9000
233 struct i387_fsave_struct {
241 long st_space[20]; /* 8*10 bytes for each FP-reg = 80 bytes */
242 long status; /* software status information */
245 struct i387_fxsave_struct {
256 long st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
257 long xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
259 } __attribute__ ((aligned (16)));
261 struct i387_soft_struct {
269 long st_space[20]; /* 8*10 bytes for each FP-reg = 80 bytes */
270 unsigned char ftop, changed, lookahead, no_update, rm, alimit;
272 unsigned long entry_eip;
276 struct i387_fsave_struct fsave;
277 struct i387_fxsave_struct fxsave;
278 struct i387_soft_struct soft;
285 struct thread_struct;
287 /* This is the TSS defined by the hardware. */
289 unsigned short back_link,__blh;
291 unsigned short ss0,__ss0h;
293 unsigned short ss1,__ss1h; /* ss1 is used to cache MSR_IA32_SYSENTER_CS */
295 unsigned short ss2,__ss2h;
298 unsigned long eflags;
299 unsigned long eax,ecx,edx,ebx;
304 unsigned short es, __esh;
305 unsigned short cs, __csh;
306 unsigned short ss, __ssh;
307 unsigned short ds, __dsh;
308 unsigned short fs, __fsh;
309 unsigned short gs, __gsh;
310 unsigned short ldt, __ldth;
311 unsigned short trace, io_bitmap_base;
312 } __attribute__((packed));
315 struct i386_hw_tss x86_tss;
318 * The extra 1 is there because the CPU will access an
319 * additional byte beyond the end of the IO permission
320 * bitmap. The extra byte must be all 1 bits, and must
321 * be within the limit.
323 unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
325 * Cache the current maximum and the last task that used the bitmap:
327 unsigned long io_bitmap_max;
328 struct thread_struct *io_bitmap_owner;
330 * pads the TSS to be cacheline-aligned (size is 0x100)
332 unsigned long __cacheline_filler[35];
334 * .. and then another 0x100 bytes for emergency kernel stack
336 unsigned long stack[64];
337 } __attribute__((packed));
339 #define ARCH_MIN_TASKALIGN 16
341 struct thread_struct {
342 /* cached TLS descriptors. */
343 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
345 unsigned long sysenter_cs;
350 /* Hardware debugging registers */
351 unsigned long debugreg[8]; /* %%db0-7 debug registers */
353 unsigned long cr2, trap_no, error_code;
354 /* floating point info */
355 union i387_union i387;
356 /* virtual 86 mode info */
357 struct vm86_struct __user * vm86_info;
358 unsigned long screen_bitmap;
359 unsigned long v86flags, v86mask, saved_esp0;
360 unsigned int saved_fs, saved_gs;
362 unsigned long *io_bitmap_ptr;
364 /* max allowed port in the bitmap, in bytes: */
365 unsigned long io_bitmap_max;
368 #define INIT_THREAD { \
369 .esp0 = sizeof(init_stack) + (long)&init_stack, \
371 .sysenter_cs = __KERNEL_CS, \
372 .io_bitmap_ptr = NULL, \
373 .fs = __KERNEL_PERCPU, \
377 * Note that the .io_bitmap member must be extra-big. This is because
378 * the CPU will access an additional byte beyond the end of the IO
379 * permission bitmap. The extra byte must be all 1 bits, and must
380 * be within the limit.
384 .esp0 = sizeof(init_stack) + (long)&init_stack, \
385 .ss0 = __KERNEL_DS, \
386 .ss1 = __KERNEL_CS, \
387 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
389 .io_bitmap = { [ 0 ... IO_BITMAP_LONGS] = ~0 }, \
392 #define start_thread(regs, new_eip, new_esp) do { \
393 __asm__("movl %0,%%gs": :"r" (0)); \
396 regs->xds = __USER_DS; \
397 regs->xes = __USER_DS; \
398 regs->xss = __USER_DS; \
399 regs->xcs = __USER_CS; \
400 regs->eip = new_eip; \
401 regs->esp = new_esp; \
404 /* Forward declaration, a strange C thing */
408 /* Free all resources held by a thread. */
409 extern void release_thread(struct task_struct *);
411 /* Prepare to copy thread state - unlazy all lazy status */
412 extern void prepare_to_copy(struct task_struct *tsk);
415 * create a kernel thread without removing it from tasklists
417 extern int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags);
419 extern unsigned long thread_saved_pc(struct task_struct *tsk);
420 void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long *stack);
422 unsigned long get_wchan(struct task_struct *p);
424 #define THREAD_SIZE_LONGS (THREAD_SIZE/sizeof(unsigned long))
425 #define KSTK_TOP(info) \
427 unsigned long *__ptr = (unsigned long *)(info); \
428 (unsigned long)(&__ptr[THREAD_SIZE_LONGS]); \
432 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
433 * This is necessary to guarantee that the entire "struct pt_regs"
434 * is accessable even if the CPU haven't stored the SS/ESP registers
435 * on the stack (interrupt gate does not save these registers
436 * when switching to the same priv ring).
437 * Therefore beware: accessing the xss/esp fields of the
438 * "struct pt_regs" is possible, but they may contain the
439 * completely wrong values.
441 #define task_pt_regs(task) \
443 struct pt_regs *__regs__; \
444 __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
448 #define KSTK_EIP(task) (task_pt_regs(task)->eip)
449 #define KSTK_ESP(task) (task_pt_regs(task)->esp)
452 struct microcode_header {
460 unsigned int datasize;
461 unsigned int totalsize;
462 unsigned int reserved[3];
466 struct microcode_header hdr;
467 unsigned int bits[0];
470 typedef struct microcode microcode_t;
471 typedef struct microcode_header microcode_header_t;
473 /* microcode format is extended from prescott processors */
474 struct extended_signature {
480 struct extended_sigtable {
483 unsigned int reserved[3];
484 struct extended_signature sigs[0];
487 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
488 static inline void rep_nop(void)
490 __asm__ __volatile__("rep;nop": : :"memory");
493 #define cpu_relax() rep_nop()
495 static inline void native_load_esp0(struct tss_struct *tss, struct thread_struct *thread)
497 tss->x86_tss.esp0 = thread->esp0;
498 /* This can only happen 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 unsigned long native_get_debugreg(int regno)
508 unsigned long val = 0; /* Damn you, gcc! */
512 asm("movl %%db0, %0" :"=r" (val)); break;
514 asm("movl %%db1, %0" :"=r" (val)); break;
516 asm("movl %%db2, %0" :"=r" (val)); break;
518 asm("movl %%db3, %0" :"=r" (val)); break;
520 asm("movl %%db6, %0" :"=r" (val)); break;
522 asm("movl %%db7, %0" :"=r" (val)); break;
529 static inline void native_set_debugreg(int regno, unsigned long value)
533 asm("movl %0,%%db0" : /* no output */ :"r" (value));
536 asm("movl %0,%%db1" : /* no output */ :"r" (value));
539 asm("movl %0,%%db2" : /* no output */ :"r" (value));
542 asm("movl %0,%%db3" : /* no output */ :"r" (value));
545 asm("movl %0,%%db6" : /* no output */ :"r" (value));
548 asm("movl %0,%%db7" : /* no output */ :"r" (value));
556 * Set IOPL bits in EFLAGS from given mask
558 static inline void native_set_iopl_mask(unsigned mask)
561 __asm__ __volatile__ ("pushfl;"
568 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
571 #ifdef CONFIG_PARAVIRT
572 #include <asm/paravirt.h>
574 #define paravirt_enabled() 0
575 #define __cpuid native_cpuid
577 static inline void load_esp0(struct tss_struct *tss, struct thread_struct *thread)
579 native_load_esp0(tss, thread);
583 * These special macros can be used to get or set a debugging register
585 #define get_debugreg(var, register) \
586 (var) = native_get_debugreg(register)
587 #define set_debugreg(value, register) \
588 native_set_debugreg(register, value)
590 #define set_iopl_mask native_set_iopl_mask
591 #endif /* CONFIG_PARAVIRT */
594 * Generic CPUID function
595 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
596 * resulting in stale register contents being returned.
598 static inline void cpuid(unsigned int op, unsigned int *eax, unsigned int *ebx, unsigned int *ecx, unsigned int *edx)
602 __cpuid(eax, ebx, ecx, edx);
605 /* Some CPUID calls want 'count' to be placed in ecx */
606 static inline void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx,
611 __cpuid(eax, ebx, ecx, edx);
615 * CPUID functions returning a single datum
617 static inline unsigned int cpuid_eax(unsigned int op)
619 unsigned int eax, ebx, ecx, edx;
621 cpuid(op, &eax, &ebx, &ecx, &edx);
624 static inline unsigned int cpuid_ebx(unsigned int op)
626 unsigned int eax, ebx, ecx, edx;
628 cpuid(op, &eax, &ebx, &ecx, &edx);
631 static inline unsigned int cpuid_ecx(unsigned int op)
633 unsigned int eax, ebx, ecx, edx;
635 cpuid(op, &eax, &ebx, &ecx, &edx);
638 static inline unsigned int cpuid_edx(unsigned int op)
640 unsigned int eax, ebx, ecx, edx;
642 cpuid(op, &eax, &ebx, &ecx, &edx);
646 /* generic versions from gas */
647 #define GENERIC_NOP1 ".byte 0x90\n"
648 #define GENERIC_NOP2 ".byte 0x89,0xf6\n"
649 #define GENERIC_NOP3 ".byte 0x8d,0x76,0x00\n"
650 #define GENERIC_NOP4 ".byte 0x8d,0x74,0x26,0x00\n"
651 #define GENERIC_NOP5 GENERIC_NOP1 GENERIC_NOP4
652 #define GENERIC_NOP6 ".byte 0x8d,0xb6,0x00,0x00,0x00,0x00\n"
653 #define GENERIC_NOP7 ".byte 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00\n"
654 #define GENERIC_NOP8 GENERIC_NOP1 GENERIC_NOP7
657 #define K8_NOP1 GENERIC_NOP1
658 #define K8_NOP2 ".byte 0x66,0x90\n"
659 #define K8_NOP3 ".byte 0x66,0x66,0x90\n"
660 #define K8_NOP4 ".byte 0x66,0x66,0x66,0x90\n"
661 #define K8_NOP5 K8_NOP3 K8_NOP2
662 #define K8_NOP6 K8_NOP3 K8_NOP3
663 #define K8_NOP7 K8_NOP4 K8_NOP3
664 #define K8_NOP8 K8_NOP4 K8_NOP4
667 /* uses eax dependencies (arbitary choice) */
668 #define K7_NOP1 GENERIC_NOP1
669 #define K7_NOP2 ".byte 0x8b,0xc0\n"
670 #define K7_NOP3 ".byte 0x8d,0x04,0x20\n"
671 #define K7_NOP4 ".byte 0x8d,0x44,0x20,0x00\n"
672 #define K7_NOP5 K7_NOP4 ASM_NOP1
673 #define K7_NOP6 ".byte 0x8d,0x80,0,0,0,0\n"
674 #define K7_NOP7 ".byte 0x8D,0x04,0x05,0,0,0,0\n"
675 #define K7_NOP8 K7_NOP7 ASM_NOP1
678 #define ASM_NOP1 K8_NOP1
679 #define ASM_NOP2 K8_NOP2
680 #define ASM_NOP3 K8_NOP3
681 #define ASM_NOP4 K8_NOP4
682 #define ASM_NOP5 K8_NOP5
683 #define ASM_NOP6 K8_NOP6
684 #define ASM_NOP7 K8_NOP7
685 #define ASM_NOP8 K8_NOP8
686 #elif defined(CONFIG_MK7)
687 #define ASM_NOP1 K7_NOP1
688 #define ASM_NOP2 K7_NOP2
689 #define ASM_NOP3 K7_NOP3
690 #define ASM_NOP4 K7_NOP4
691 #define ASM_NOP5 K7_NOP5
692 #define ASM_NOP6 K7_NOP6
693 #define ASM_NOP7 K7_NOP7
694 #define ASM_NOP8 K7_NOP8
696 #define ASM_NOP1 GENERIC_NOP1
697 #define ASM_NOP2 GENERIC_NOP2
698 #define ASM_NOP3 GENERIC_NOP3
699 #define ASM_NOP4 GENERIC_NOP4
700 #define ASM_NOP5 GENERIC_NOP5
701 #define ASM_NOP6 GENERIC_NOP6
702 #define ASM_NOP7 GENERIC_NOP7
703 #define ASM_NOP8 GENERIC_NOP8
706 #define ASM_NOP_MAX 8
708 /* Prefetch instructions for Pentium III and AMD Athlon */
709 /* It's not worth to care about 3dnow! prefetches for the K6
710 because they are microcoded there and very slow.
711 However we don't do prefetches for pre XP Athlons currently
712 That should be fixed. */
713 #define ARCH_HAS_PREFETCH
714 static inline void prefetch(const void *x)
716 alternative_input(ASM_NOP4,
722 #define ARCH_HAS_PREFETCH
723 #define ARCH_HAS_PREFETCHW
724 #define ARCH_HAS_SPINLOCK_PREFETCH
726 /* 3dnow! prefetch to get an exclusive cache line. Useful for
727 spinlocks to avoid one state transition in the cache coherency protocol. */
728 static inline void prefetchw(const void *x)
730 alternative_input(ASM_NOP4,
735 #define spin_lock_prefetch(x) prefetchw(x)
737 extern void select_idle_routine(const struct cpuinfo_x86 *c);
739 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
741 extern unsigned long boot_option_idle_override;
742 extern void enable_sep_cpu(void);
743 extern int sysenter_setup(void);
745 /* Defined in head.S */
746 extern struct Xgt_desc_struct early_gdt_descr;
748 extern void cpu_set_gdt(int);
749 extern void switch_to_new_gdt(void);
750 extern void cpu_init(void);
751 extern void init_gdt(int cpu);
753 extern int force_mwait;
755 #endif /* __ASM_I386_PROCESSOR_H */