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 <uapi/asm/sigcontext.h>
15 #include <asm/current.h>
16 #include <asm/cpufeatures.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/cache.h>
28 #include <linux/threads.h>
29 #include <linux/math64.h>
30 #include <linux/err.h>
31 #include <linux/irqflags.h>
34 * We handle most unaligned accesses in hardware. On the other hand
35 * unaligned DMA can be quite expensive on some Nehalem processors.
37 * Based on this we disable the IP header alignment in network drivers.
39 #define NET_IP_ALIGN 0
43 * Default implementation of macro that returns current
44 * instruction pointer ("program counter").
46 static inline void *current_text_addr(void)
50 asm volatile("mov $1f, %0; 1:":"=r" (pc));
56 * These alignment constraints are for performance in the vSMP case,
57 * but in the task_struct case we must also meet hardware imposed
58 * alignment requirements of the FPU state:
60 #ifdef CONFIG_X86_VSMP
61 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
62 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
64 # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
65 # define ARCH_MIN_MMSTRUCT_ALIGN 0
73 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
74 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
75 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
76 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
77 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
78 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
79 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
82 * CPU type and hardware bug flags. Kept separately for each CPU.
83 * Members of this structure are referenced in head.S, so think twice
84 * before touching them. [mj]
88 __u8 x86; /* CPU family */
89 __u8 x86_vendor; /* CPU vendor */
93 char wp_works_ok; /* It doesn't on 386's */
95 /* Problems on some 486Dx4's and old 386's: */
100 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
105 /* CPUID returned core id bits: */
106 __u8 x86_coreid_bits;
107 /* Max extended CPUID function supported: */
108 __u32 extended_cpuid_level;
109 /* Maximum supported CPUID level, -1=no CPUID: */
111 __u32 x86_capability[NCAPINTS + NBUGINTS];
112 char x86_vendor_id[16];
113 char x86_model_id[64];
114 /* in KB - valid for CPUS which support this call: */
116 int x86_cache_alignment; /* In bytes */
117 /* Cache QoS architectural values: */
118 int x86_cache_max_rmid; /* max index */
119 int x86_cache_occ_scale; /* scale to bytes */
121 unsigned long loops_per_jiffy;
122 /* cpuid returned max cores value: */
126 u16 x86_clflush_size;
127 /* number of cores as seen by the OS: */
129 /* Physical processor id: */
131 /* Logical processor id: */
135 /* Index into per_cpu list: */
140 #define X86_VENDOR_INTEL 0
141 #define X86_VENDOR_CYRIX 1
142 #define X86_VENDOR_AMD 2
143 #define X86_VENDOR_UMC 3
144 #define X86_VENDOR_CENTAUR 5
145 #define X86_VENDOR_TRANSMETA 7
146 #define X86_VENDOR_NSC 8
147 #define X86_VENDOR_NUM 9
149 #define X86_VENDOR_UNKNOWN 0xff
152 * capabilities of CPUs
154 extern struct cpuinfo_x86 boot_cpu_data;
155 extern struct cpuinfo_x86 new_cpu_data;
157 extern struct tss_struct doublefault_tss;
158 extern __u32 cpu_caps_cleared[NCAPINTS];
159 extern __u32 cpu_caps_set[NCAPINTS];
162 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
163 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
165 #define cpu_info boot_cpu_data
166 #define cpu_data(cpu) boot_cpu_data
169 extern const struct seq_operations cpuinfo_op;
171 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
173 extern void cpu_detect(struct cpuinfo_x86 *c);
175 extern void early_cpu_init(void);
176 extern void identify_boot_cpu(void);
177 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
178 extern void print_cpu_info(struct cpuinfo_x86 *);
179 void print_cpu_msr(struct cpuinfo_x86 *);
180 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
181 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
182 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
184 extern void detect_extended_topology(struct cpuinfo_x86 *c);
185 extern void detect_ht(struct cpuinfo_x86 *c);
188 extern int have_cpuid_p(void);
190 static inline int have_cpuid_p(void)
195 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
196 unsigned int *ecx, unsigned int *edx)
198 /* ecx is often an input as well as an output. */
204 : "0" (*eax), "2" (*ecx)
208 static inline void load_cr3(pgd_t *pgdir)
210 write_cr3(__pa(pgdir));
214 /* This is the TSS defined by the hardware. */
216 unsigned short back_link, __blh;
218 unsigned short ss0, __ss0h;
222 * We don't use ring 1, so ss1 is a convenient scratch space in
223 * the same cacheline as sp0. We use ss1 to cache the value in
224 * MSR_IA32_SYSENTER_CS. When we context switch
225 * MSR_IA32_SYSENTER_CS, we first check if the new value being
226 * written matches ss1, and, if it's not, then we wrmsr the new
227 * value and update ss1.
229 * The only reason we context switch MSR_IA32_SYSENTER_CS is
230 * that we set it to zero in vm86 tasks to avoid corrupting the
231 * stack if we were to go through the sysenter path from vm86
234 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
236 unsigned short __ss1h;
238 unsigned short ss2, __ss2h;
250 unsigned short es, __esh;
251 unsigned short cs, __csh;
252 unsigned short ss, __ssh;
253 unsigned short ds, __dsh;
254 unsigned short fs, __fsh;
255 unsigned short gs, __gsh;
256 unsigned short ldt, __ldth;
257 unsigned short trace;
258 unsigned short io_bitmap_base;
260 } __attribute__((packed));
274 } __attribute__((packed)) ____cacheline_aligned;
280 #define IO_BITMAP_BITS 65536
281 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
282 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
283 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
284 #define INVALID_IO_BITMAP_OFFSET 0x8000
288 * The hardware state:
290 struct x86_hw_tss x86_tss;
293 * The extra 1 is there because the CPU will access an
294 * additional byte beyond the end of the IO permission
295 * bitmap. The extra byte must be all 1 bits, and must
296 * be within the limit.
298 unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
302 * Space for the temporary SYSENTER stack.
304 unsigned long SYSENTER_stack_canary;
305 unsigned long SYSENTER_stack[64];
308 } ____cacheline_aligned;
310 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss);
313 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
317 * Save the original ist values for checking stack pointers during debugging
320 unsigned long ist[7];
324 DECLARE_PER_CPU(struct orig_ist, orig_ist);
326 union irq_stack_union {
327 char irq_stack[IRQ_STACK_SIZE];
329 * GCC hardcodes the stack canary as %gs:40. Since the
330 * irq_stack is the object at %gs:0, we reserve the bottom
331 * 48 bytes of the irq stack for the canary.
335 unsigned long stack_canary;
339 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
340 DECLARE_INIT_PER_CPU(irq_stack_union);
342 DECLARE_PER_CPU(char *, irq_stack_ptr);
343 DECLARE_PER_CPU(unsigned int, irq_count);
344 extern asmlinkage void ignore_sysret(void);
346 #ifdef CONFIG_CC_STACKPROTECTOR
348 * Make sure stack canary segment base is cached-aligned:
349 * "For Intel Atom processors, avoid non zero segment base address
350 * that is not aligned to cache line boundary at all cost."
351 * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
353 struct stack_canary {
354 char __pad[20]; /* canary at %gs:20 */
355 unsigned long canary;
357 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
360 * per-CPU IRQ handling stacks
363 u32 stack[THREAD_SIZE/sizeof(u32)];
364 } __aligned(THREAD_SIZE);
366 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
367 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
370 extern unsigned int fpu_kernel_xstate_size;
371 extern unsigned int fpu_user_xstate_size;
379 struct thread_struct {
380 /* Cached TLS descriptors: */
381 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
385 unsigned long sysenter_cs;
389 unsigned short fsindex;
390 unsigned short gsindex;
393 u32 status; /* thread synchronous flags */
396 unsigned long fsbase;
397 unsigned long gsbase;
400 * XXX: this could presumably be unsigned short. Alternatively,
401 * 32-bit kernels could be taught to use fsindex instead.
407 /* Save middle states of ptrace breakpoints */
408 struct perf_event *ptrace_bps[HBP_NUM];
409 /* Debug status used for traps, single steps, etc... */
410 unsigned long debugreg6;
411 /* Keep track of the exact dr7 value set by the user */
412 unsigned long ptrace_dr7;
415 unsigned long trap_nr;
416 unsigned long error_code;
418 /* Virtual 86 mode info */
421 /* IO permissions: */
422 unsigned long *io_bitmap_ptr;
424 /* Max allowed port in the bitmap, in bytes: */
425 unsigned io_bitmap_max;
427 mm_segment_t addr_limit;
429 unsigned int sig_on_uaccess_err:1;
430 unsigned int uaccess_err:1; /* uaccess failed */
432 /* Floating point and extended processor state */
435 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
441 * Thread-synchronous status.
443 * This is different from the flags in that nobody else
444 * ever touches our thread-synchronous status, so we don't
445 * have to worry about atomic accesses.
447 #define TS_COMPAT 0x0002 /* 32bit syscall active (64BIT)*/
450 * Set IOPL bits in EFLAGS from given mask
452 static inline void native_set_iopl_mask(unsigned mask)
457 asm volatile ("pushfl;"
464 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
469 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
471 tss->x86_tss.sp0 = thread->sp0;
473 /* Only happens when SEP is enabled, no need to test "SEP"arately: */
474 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
475 tss->x86_tss.ss1 = thread->sysenter_cs;
476 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
481 static inline void native_swapgs(void)
484 asm volatile("swapgs" ::: "memory");
488 static inline unsigned long current_top_of_stack(void)
491 return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
493 /* sp0 on x86_32 is special in and around vm86 mode. */
494 return this_cpu_read_stable(cpu_current_top_of_stack);
498 #ifdef CONFIG_PARAVIRT
499 #include <asm/paravirt.h>
501 #define __cpuid native_cpuid
503 static inline void load_sp0(struct tss_struct *tss,
504 struct thread_struct *thread)
506 native_load_sp0(tss, thread);
509 #define set_iopl_mask native_set_iopl_mask
510 #endif /* CONFIG_PARAVIRT */
512 /* Free all resources held by a thread. */
513 extern void release_thread(struct task_struct *);
515 unsigned long get_wchan(struct task_struct *p);
518 * Generic CPUID function
519 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
520 * resulting in stale register contents being returned.
522 static inline void cpuid(unsigned int op,
523 unsigned int *eax, unsigned int *ebx,
524 unsigned int *ecx, unsigned int *edx)
528 __cpuid(eax, ebx, ecx, edx);
531 /* Some CPUID calls want 'count' to be placed in ecx */
532 static inline void cpuid_count(unsigned int op, int count,
533 unsigned int *eax, unsigned int *ebx,
534 unsigned int *ecx, unsigned int *edx)
538 __cpuid(eax, ebx, ecx, edx);
542 * CPUID functions returning a single datum
544 static inline unsigned int cpuid_eax(unsigned int op)
546 unsigned int eax, ebx, ecx, edx;
548 cpuid(op, &eax, &ebx, &ecx, &edx);
553 static inline unsigned int cpuid_ebx(unsigned int op)
555 unsigned int eax, ebx, ecx, edx;
557 cpuid(op, &eax, &ebx, &ecx, &edx);
562 static inline unsigned int cpuid_ecx(unsigned int op)
564 unsigned int eax, ebx, ecx, edx;
566 cpuid(op, &eax, &ebx, &ecx, &edx);
571 static inline unsigned int cpuid_edx(unsigned int op)
573 unsigned int eax, ebx, ecx, edx;
575 cpuid(op, &eax, &ebx, &ecx, &edx);
580 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
581 static __always_inline void rep_nop(void)
583 asm volatile("rep; nop" ::: "memory");
586 static __always_inline void cpu_relax(void)
591 #define cpu_relax_lowlatency() cpu_relax()
593 /* Stop speculative execution and prefetching of modified code. */
594 static inline void sync_core(void)
600 * Do a CPUID if available, otherwise do a jump. The jump
601 * can conveniently enough be the jump around CPUID.
603 asm volatile("cmpl %2,%1\n\t"
608 : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
609 : "ebx", "ecx", "edx", "memory");
612 * CPUID is a barrier to speculative execution.
613 * Prefetched instructions are automatically
614 * invalidated when modified.
619 : "ebx", "ecx", "edx", "memory");
623 extern void select_idle_routine(const struct cpuinfo_x86 *c);
624 extern void init_amd_e400_c1e_mask(void);
626 extern unsigned long boot_option_idle_override;
627 extern bool amd_e400_c1e_detected;
629 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
632 extern void enable_sep_cpu(void);
633 extern int sysenter_setup(void);
635 extern void early_trap_init(void);
636 void early_trap_pf_init(void);
638 /* Defined in head.S */
639 extern struct desc_ptr early_gdt_descr;
641 extern void cpu_set_gdt(int);
642 extern void switch_to_new_gdt(int);
643 extern void load_percpu_segment(int);
644 extern void cpu_init(void);
646 static inline unsigned long get_debugctlmsr(void)
648 unsigned long debugctlmsr = 0;
650 #ifndef CONFIG_X86_DEBUGCTLMSR
651 if (boot_cpu_data.x86 < 6)
654 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
659 static inline void update_debugctlmsr(unsigned long debugctlmsr)
661 #ifndef CONFIG_X86_DEBUGCTLMSR
662 if (boot_cpu_data.x86 < 6)
665 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
668 extern void set_task_blockstep(struct task_struct *task, bool on);
670 /* Boot loader type from the setup header: */
671 extern int bootloader_type;
672 extern int bootloader_version;
674 extern char ignore_fpu_irq;
676 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
677 #define ARCH_HAS_PREFETCHW
678 #define ARCH_HAS_SPINLOCK_PREFETCH
681 # define BASE_PREFETCH ""
682 # define ARCH_HAS_PREFETCH
684 # define BASE_PREFETCH "prefetcht0 %P1"
688 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
690 * It's not worth to care about 3dnow prefetches for the K6
691 * because they are microcoded there and very slow.
693 static inline void prefetch(const void *x)
695 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
697 "m" (*(const char *)x));
701 * 3dnow prefetch to get an exclusive cache line.
702 * Useful for spinlocks to avoid one state transition in the
703 * cache coherency protocol:
705 static inline void prefetchw(const void *x)
707 alternative_input(BASE_PREFETCH, "prefetchw %P1",
708 X86_FEATURE_3DNOWPREFETCH,
709 "m" (*(const char *)x));
712 static inline void spin_lock_prefetch(const void *x)
717 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
718 TOP_OF_KERNEL_STACK_PADDING)
722 * User space process size: 3GB (default).
724 #define TASK_SIZE PAGE_OFFSET
725 #define TASK_SIZE_MAX TASK_SIZE
726 #define STACK_TOP TASK_SIZE
727 #define STACK_TOP_MAX STACK_TOP
729 #define INIT_THREAD { \
730 .sp0 = TOP_OF_INIT_STACK, \
731 .sysenter_cs = __KERNEL_CS, \
732 .io_bitmap_ptr = NULL, \
733 .addr_limit = KERNEL_DS, \
737 * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
738 * This is necessary to guarantee that the entire "struct pt_regs"
739 * is accessible even if the CPU haven't stored the SS/ESP registers
740 * on the stack (interrupt gate does not save these registers
741 * when switching to the same priv ring).
742 * Therefore beware: accessing the ss/esp fields of the
743 * "struct pt_regs" is possible, but they may contain the
744 * completely wrong values.
746 #define task_pt_regs(task) \
748 unsigned long __ptr = (unsigned long)task_stack_page(task); \
749 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
750 ((struct pt_regs *)__ptr) - 1; \
753 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
757 * User space process size. 47bits minus one guard page. The guard
758 * page is necessary on Intel CPUs: if a SYSCALL instruction is at
759 * the highest possible canonical userspace address, then that
760 * syscall will enter the kernel with a non-canonical return
761 * address, and SYSRET will explode dangerously. We avoid this
762 * particular problem by preventing anything from being mapped
763 * at the maximum canonical address.
765 #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE)
767 /* This decides where the kernel will search for a free chunk of vm
768 * space during mmap's.
770 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
771 0xc0000000 : 0xFFFFe000)
773 #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \
774 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
775 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
776 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
778 #define STACK_TOP TASK_SIZE
779 #define STACK_TOP_MAX TASK_SIZE_MAX
781 #define INIT_THREAD { \
782 .sp0 = TOP_OF_INIT_STACK, \
783 .addr_limit = KERNEL_DS, \
786 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
787 extern unsigned long KSTK_ESP(struct task_struct *task);
789 #endif /* CONFIG_X86_64 */
791 extern unsigned long thread_saved_pc(struct task_struct *tsk);
793 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
794 unsigned long new_sp);
797 * This decides where the kernel will search for a free chunk of vm
798 * space during mmap's.
800 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
802 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
804 /* Get/set a process' ability to use the timestamp counter instruction */
805 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
806 #define SET_TSC_CTL(val) set_tsc_mode((val))
808 extern int get_tsc_mode(unsigned long adr);
809 extern int set_tsc_mode(unsigned int val);
811 /* Register/unregister a process' MPX related resource */
812 #define MPX_ENABLE_MANAGEMENT() mpx_enable_management()
813 #define MPX_DISABLE_MANAGEMENT() mpx_disable_management()
815 #ifdef CONFIG_X86_INTEL_MPX
816 extern int mpx_enable_management(void);
817 extern int mpx_disable_management(void);
819 static inline int mpx_enable_management(void)
823 static inline int mpx_disable_management(void)
827 #endif /* CONFIG_X86_INTEL_MPX */
829 extern u16 amd_get_nb_id(int cpu);
830 extern u32 amd_get_nodes_per_socket(void);
832 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
834 uint32_t base, eax, signature[3];
836 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
837 cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
839 if (!memcmp(sig, signature, 12) &&
840 (leaves == 0 || ((eax - base) >= leaves)))
847 extern unsigned long arch_align_stack(unsigned long sp);
848 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
850 void default_idle(void);
852 bool xen_set_default_idle(void);
854 #define xen_set_default_idle 0
857 void stop_this_cpu(void *dummy);
858 void df_debug(struct pt_regs *regs, long error_code);
859 #endif /* _ASM_X86_PROCESSOR_H */