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>
18 #include <asm/pgtable_types.h>
19 #include <asm/percpu.h>
21 #include <asm/desc_defs.h>
23 #include <asm/special_insns.h>
24 #include <asm/fpu/types.h>
26 #include <linux/personality.h>
27 #include <linux/cpumask.h>
28 #include <linux/cache.h>
29 #include <linux/threads.h>
30 #include <linux/math64.h>
31 #include <linux/err.h>
32 #include <linux/irqflags.h>
35 * We handle most unaligned accesses in hardware. On the other hand
36 * unaligned DMA can be quite expensive on some Nehalem processors.
38 * Based on this we disable the IP header alignment in network drivers.
40 #define NET_IP_ALIGN 0
44 * Default implementation of macro that returns current
45 * instruction pointer ("program counter").
47 static inline void *current_text_addr(void)
51 asm volatile("mov $1f, %0; 1:":"=r" (pc));
56 #ifdef CONFIG_X86_VSMP
57 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
58 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
60 # define ARCH_MIN_TASKALIGN 16
61 # define ARCH_MIN_MMSTRUCT_ALIGN 0
69 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
70 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
71 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
72 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
73 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
75 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
78 * CPU type and hardware bug flags. Kept separately for each CPU.
79 * Members of this structure are referenced in head.S, so think twice
80 * before touching them. [mj]
84 __u8 x86; /* CPU family */
85 __u8 x86_vendor; /* CPU vendor */
89 char wp_works_ok; /* It doesn't on 386's */
91 /* Problems on some 486Dx4's and old 386's: */
96 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
101 /* CPUID returned core id bits: */
102 __u8 x86_coreid_bits;
103 /* Max extended CPUID function supported: */
104 __u32 extended_cpuid_level;
105 /* Maximum supported CPUID level, -1=no CPUID: */
107 __u32 x86_capability[NCAPINTS + NBUGINTS];
108 char x86_vendor_id[16];
109 char x86_model_id[64];
110 /* in KB - valid for CPUS which support this call: */
112 int x86_cache_alignment; /* In bytes */
113 /* Cache QoS architectural values: */
114 int x86_cache_max_rmid; /* max index */
115 int x86_cache_occ_scale; /* scale to bytes */
117 unsigned long loops_per_jiffy;
118 /* cpuid returned max cores value: */
122 u16 x86_clflush_size;
123 /* number of cores as seen by the OS: */
125 /* Physical processor id: */
129 /* Compute unit id */
131 /* Index into per_cpu list: */
136 #define X86_VENDOR_INTEL 0
137 #define X86_VENDOR_CYRIX 1
138 #define X86_VENDOR_AMD 2
139 #define X86_VENDOR_UMC 3
140 #define X86_VENDOR_CENTAUR 5
141 #define X86_VENDOR_TRANSMETA 7
142 #define X86_VENDOR_NSC 8
143 #define X86_VENDOR_NUM 9
145 #define X86_VENDOR_UNKNOWN 0xff
148 * capabilities of CPUs
150 extern struct cpuinfo_x86 boot_cpu_data;
151 extern struct cpuinfo_x86 new_cpu_data;
153 extern struct tss_struct doublefault_tss;
154 extern __u32 cpu_caps_cleared[NCAPINTS];
155 extern __u32 cpu_caps_set[NCAPINTS];
158 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
159 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
161 #define cpu_info boot_cpu_data
162 #define cpu_data(cpu) boot_cpu_data
165 extern const struct seq_operations cpuinfo_op;
167 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
169 extern void cpu_detect(struct cpuinfo_x86 *c);
171 extern void early_cpu_init(void);
172 extern void identify_boot_cpu(void);
173 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
174 extern void print_cpu_info(struct cpuinfo_x86 *);
175 void print_cpu_msr(struct cpuinfo_x86 *);
176 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
177 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
178 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
180 extern void detect_extended_topology(struct cpuinfo_x86 *c);
181 extern void detect_ht(struct cpuinfo_x86 *c);
184 extern int have_cpuid_p(void);
186 static inline int have_cpuid_p(void)
191 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
192 unsigned int *ecx, unsigned int *edx)
194 /* ecx is often an input as well as an output. */
200 : "0" (*eax), "2" (*ecx)
204 static inline void load_cr3(pgd_t *pgdir)
206 write_cr3(__pa(pgdir));
210 /* This is the TSS defined by the hardware. */
212 unsigned short back_link, __blh;
214 unsigned short ss0, __ss0h;
218 * We don't use ring 1, so ss1 is a convenient scratch space in
219 * the same cacheline as sp0. We use ss1 to cache the value in
220 * MSR_IA32_SYSENTER_CS. When we context switch
221 * MSR_IA32_SYSENTER_CS, we first check if the new value being
222 * written matches ss1, and, if it's not, then we wrmsr the new
223 * value and update ss1.
225 * The only reason we context switch MSR_IA32_SYSENTER_CS is
226 * that we set it to zero in vm86 tasks to avoid corrupting the
227 * stack if we were to go through the sysenter path from vm86
230 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
232 unsigned short __ss1h;
234 unsigned short ss2, __ss2h;
246 unsigned short es, __esh;
247 unsigned short cs, __csh;
248 unsigned short ss, __ssh;
249 unsigned short ds, __dsh;
250 unsigned short fs, __fsh;
251 unsigned short gs, __gsh;
252 unsigned short ldt, __ldth;
253 unsigned short trace;
254 unsigned short io_bitmap_base;
256 } __attribute__((packed));
270 } __attribute__((packed)) ____cacheline_aligned;
276 #define IO_BITMAP_BITS 65536
277 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
278 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
279 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
280 #define INVALID_IO_BITMAP_OFFSET 0x8000
284 * The hardware state:
286 struct x86_hw_tss x86_tss;
289 * The extra 1 is there because the CPU will access an
290 * additional byte beyond the end of the IO permission
291 * bitmap. The extra byte must be all 1 bits, and must
292 * be within the limit.
294 unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
297 * Space for the temporary SYSENTER stack:
299 unsigned long SYSENTER_stack[64];
301 } ____cacheline_aligned;
303 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss);
306 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
310 * Save the original ist values for checking stack pointers during debugging
313 unsigned long ist[7];
317 DECLARE_PER_CPU(struct orig_ist, orig_ist);
319 union irq_stack_union {
320 char irq_stack[IRQ_STACK_SIZE];
322 * GCC hardcodes the stack canary as %gs:40. Since the
323 * irq_stack is the object at %gs:0, we reserve the bottom
324 * 48 bytes of the irq stack for the canary.
328 unsigned long stack_canary;
332 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
333 DECLARE_INIT_PER_CPU(irq_stack_union);
335 DECLARE_PER_CPU(char *, irq_stack_ptr);
336 DECLARE_PER_CPU(unsigned int, irq_count);
337 extern asmlinkage void ignore_sysret(void);
339 #ifdef CONFIG_CC_STACKPROTECTOR
341 * Make sure stack canary segment base is cached-aligned:
342 * "For Intel Atom processors, avoid non zero segment base address
343 * that is not aligned to cache line boundary at all cost."
344 * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
346 struct stack_canary {
347 char __pad[20]; /* canary at %gs:20 */
348 unsigned long canary;
350 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
353 * per-CPU IRQ handling stacks
356 u32 stack[THREAD_SIZE/sizeof(u32)];
357 } __aligned(THREAD_SIZE);
359 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
360 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
363 extern unsigned int xstate_size;
367 struct thread_struct {
368 /* Cached TLS descriptors: */
369 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
373 unsigned long sysenter_cs;
377 unsigned short fsindex;
378 unsigned short gsindex;
388 /* Floating point and extended processor state */
391 /* Save middle states of ptrace breakpoints */
392 struct perf_event *ptrace_bps[HBP_NUM];
393 /* Debug status used for traps, single steps, etc... */
394 unsigned long debugreg6;
395 /* Keep track of the exact dr7 value set by the user */
396 unsigned long ptrace_dr7;
399 unsigned long trap_nr;
400 unsigned long error_code;
402 /* Virtual 86 mode info */
403 struct vm86_struct __user *vm86_info;
404 unsigned long screen_bitmap;
405 unsigned long v86flags;
406 unsigned long v86mask;
407 unsigned long saved_sp0;
408 unsigned int saved_fs;
409 unsigned int saved_gs;
411 /* IO permissions: */
412 unsigned long *io_bitmap_ptr;
414 /* Max allowed port in the bitmap, in bytes: */
415 unsigned io_bitmap_max;
419 * Set IOPL bits in EFLAGS from given mask
421 static inline void native_set_iopl_mask(unsigned mask)
426 asm volatile ("pushfl;"
433 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
438 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
440 tss->x86_tss.sp0 = thread->sp0;
442 /* Only happens when SEP is enabled, no need to test "SEP"arately: */
443 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
444 tss->x86_tss.ss1 = thread->sysenter_cs;
445 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
450 static inline void native_swapgs(void)
453 asm volatile("swapgs" ::: "memory");
457 static inline unsigned long current_top_of_stack(void)
460 return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
462 /* sp0 on x86_32 is special in and around vm86 mode. */
463 return this_cpu_read_stable(cpu_current_top_of_stack);
467 #ifdef CONFIG_PARAVIRT
468 #include <asm/paravirt.h>
470 #define __cpuid native_cpuid
471 #define paravirt_enabled() 0
473 static inline void load_sp0(struct tss_struct *tss,
474 struct thread_struct *thread)
476 native_load_sp0(tss, thread);
479 #define set_iopl_mask native_set_iopl_mask
480 #endif /* CONFIG_PARAVIRT */
487 /* Free all resources held by a thread. */
488 extern void release_thread(struct task_struct *);
490 unsigned long get_wchan(struct task_struct *p);
493 * Generic CPUID function
494 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
495 * resulting in stale register contents being returned.
497 static inline void cpuid(unsigned int op,
498 unsigned int *eax, unsigned int *ebx,
499 unsigned int *ecx, unsigned int *edx)
503 __cpuid(eax, ebx, ecx, edx);
506 /* Some CPUID calls want 'count' to be placed in ecx */
507 static inline void cpuid_count(unsigned int op, int count,
508 unsigned int *eax, unsigned int *ebx,
509 unsigned int *ecx, unsigned int *edx)
513 __cpuid(eax, ebx, ecx, edx);
517 * CPUID functions returning a single datum
519 static inline unsigned int cpuid_eax(unsigned int op)
521 unsigned int eax, ebx, ecx, edx;
523 cpuid(op, &eax, &ebx, &ecx, &edx);
528 static inline unsigned int cpuid_ebx(unsigned int op)
530 unsigned int eax, ebx, ecx, edx;
532 cpuid(op, &eax, &ebx, &ecx, &edx);
537 static inline unsigned int cpuid_ecx(unsigned int op)
539 unsigned int eax, ebx, ecx, edx;
541 cpuid(op, &eax, &ebx, &ecx, &edx);
546 static inline unsigned int cpuid_edx(unsigned int op)
548 unsigned int eax, ebx, ecx, edx;
550 cpuid(op, &eax, &ebx, &ecx, &edx);
555 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
556 static inline void rep_nop(void)
558 asm volatile("rep; nop" ::: "memory");
561 static inline void cpu_relax(void)
566 #define cpu_relax_lowlatency() cpu_relax()
568 /* Stop speculative execution and prefetching of modified code. */
569 static inline void sync_core(void)
575 * Do a CPUID if available, otherwise do a jump. The jump
576 * can conveniently enough be the jump around CPUID.
578 asm volatile("cmpl %2,%1\n\t"
583 : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
584 : "ebx", "ecx", "edx", "memory");
587 * CPUID is a barrier to speculative execution.
588 * Prefetched instructions are automatically
589 * invalidated when modified.
594 : "ebx", "ecx", "edx", "memory");
598 extern void select_idle_routine(const struct cpuinfo_x86 *c);
599 extern void init_amd_e400_c1e_mask(void);
601 extern unsigned long boot_option_idle_override;
602 extern bool amd_e400_c1e_detected;
604 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
607 extern void enable_sep_cpu(void);
608 extern int sysenter_setup(void);
610 extern void early_trap_init(void);
611 void early_trap_pf_init(void);
613 /* Defined in head.S */
614 extern struct desc_ptr early_gdt_descr;
616 extern void cpu_set_gdt(int);
617 extern void switch_to_new_gdt(int);
618 extern void load_percpu_segment(int);
619 extern void cpu_init(void);
621 static inline unsigned long get_debugctlmsr(void)
623 unsigned long debugctlmsr = 0;
625 #ifndef CONFIG_X86_DEBUGCTLMSR
626 if (boot_cpu_data.x86 < 6)
629 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
634 static inline void update_debugctlmsr(unsigned long debugctlmsr)
636 #ifndef CONFIG_X86_DEBUGCTLMSR
637 if (boot_cpu_data.x86 < 6)
640 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
643 extern void set_task_blockstep(struct task_struct *task, bool on);
646 * from system description table in BIOS. Mostly for MCA use, but
647 * others may find it useful:
649 extern unsigned int machine_id;
650 extern unsigned int machine_submodel_id;
651 extern unsigned int BIOS_revision;
653 /* Boot loader type from the setup header: */
654 extern int bootloader_type;
655 extern int bootloader_version;
657 extern char ignore_fpu_irq;
659 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
660 #define ARCH_HAS_PREFETCHW
661 #define ARCH_HAS_SPINLOCK_PREFETCH
664 # define BASE_PREFETCH ""
665 # define ARCH_HAS_PREFETCH
667 # define BASE_PREFETCH "prefetcht0 %P1"
671 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
673 * It's not worth to care about 3dnow prefetches for the K6
674 * because they are microcoded there and very slow.
676 static inline void prefetch(const void *x)
678 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
680 "m" (*(const char *)x));
684 * 3dnow prefetch to get an exclusive cache line.
685 * Useful for spinlocks to avoid one state transition in the
686 * cache coherency protocol:
688 static inline void prefetchw(const void *x)
690 alternative_input(BASE_PREFETCH, "prefetchw %P1",
691 X86_FEATURE_3DNOWPREFETCH,
692 "m" (*(const char *)x));
695 static inline void spin_lock_prefetch(const void *x)
700 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
701 TOP_OF_KERNEL_STACK_PADDING)
705 * User space process size: 3GB (default).
707 #define TASK_SIZE PAGE_OFFSET
708 #define TASK_SIZE_MAX TASK_SIZE
709 #define STACK_TOP TASK_SIZE
710 #define STACK_TOP_MAX STACK_TOP
712 #define INIT_THREAD { \
713 .sp0 = TOP_OF_INIT_STACK, \
715 .sysenter_cs = __KERNEL_CS, \
716 .io_bitmap_ptr = NULL, \
719 extern unsigned long thread_saved_pc(struct task_struct *tsk);
722 * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
723 * This is necessary to guarantee that the entire "struct pt_regs"
724 * is accessible even if the CPU haven't stored the SS/ESP registers
725 * on the stack (interrupt gate does not save these registers
726 * when switching to the same priv ring).
727 * Therefore beware: accessing the ss/esp fields of the
728 * "struct pt_regs" is possible, but they may contain the
729 * completely wrong values.
731 #define task_pt_regs(task) \
733 unsigned long __ptr = (unsigned long)task_stack_page(task); \
734 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
735 ((struct pt_regs *)__ptr) - 1; \
738 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
742 * User space process size. 47bits minus one guard page. The guard
743 * page is necessary on Intel CPUs: if a SYSCALL instruction is at
744 * the highest possible canonical userspace address, then that
745 * syscall will enter the kernel with a non-canonical return
746 * address, and SYSRET will explode dangerously. We avoid this
747 * particular problem by preventing anything from being mapped
748 * at the maximum canonical address.
750 #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE)
752 /* This decides where the kernel will search for a free chunk of vm
753 * space during mmap's.
755 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
756 0xc0000000 : 0xFFFFe000)
758 #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \
759 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
760 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
761 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
763 #define STACK_TOP TASK_SIZE
764 #define STACK_TOP_MAX TASK_SIZE_MAX
766 #define INIT_THREAD { \
767 .sp0 = TOP_OF_INIT_STACK \
771 * Return saved PC of a blocked thread.
772 * What is this good for? it will be always the scheduler or ret_from_fork.
774 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
776 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
777 extern unsigned long KSTK_ESP(struct task_struct *task);
779 #endif /* CONFIG_X86_64 */
781 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
782 unsigned long new_sp);
785 * This decides where the kernel will search for a free chunk of vm
786 * space during mmap's.
788 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
790 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
792 /* Get/set a process' ability to use the timestamp counter instruction */
793 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
794 #define SET_TSC_CTL(val) set_tsc_mode((val))
796 extern int get_tsc_mode(unsigned long adr);
797 extern int set_tsc_mode(unsigned int val);
799 /* Register/unregister a process' MPX related resource */
800 #define MPX_ENABLE_MANAGEMENT(tsk) mpx_enable_management((tsk))
801 #define MPX_DISABLE_MANAGEMENT(tsk) mpx_disable_management((tsk))
803 #ifdef CONFIG_X86_INTEL_MPX
804 extern int mpx_enable_management(struct task_struct *tsk);
805 extern int mpx_disable_management(struct task_struct *tsk);
807 static inline int mpx_enable_management(struct task_struct *tsk)
811 static inline int mpx_disable_management(struct task_struct *tsk)
815 #endif /* CONFIG_X86_INTEL_MPX */
817 extern u16 amd_get_nb_id(int cpu);
819 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
821 uint32_t base, eax, signature[3];
823 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
824 cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
826 if (!memcmp(sig, signature, 12) &&
827 (leaves == 0 || ((eax - base) >= leaves)))
834 extern unsigned long arch_align_stack(unsigned long sp);
835 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
837 void default_idle(void);
839 bool xen_set_default_idle(void);
841 #define xen_set_default_idle 0
844 void stop_this_cpu(void *dummy);
845 void df_debug(struct pt_regs *regs, long error_code);
846 #endif /* _ASM_X86_PROCESSOR_H */