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;
109 /* Max extended CPUID function supported: */
110 __u32 extended_cpuid_level;
111 /* Maximum supported CPUID level, -1=no CPUID: */
113 __u32 x86_capability[NCAPINTS + NBUGINTS];
114 char x86_vendor_id[16];
115 char x86_model_id[64];
116 /* in KB - valid for CPUS which support this call: */
117 unsigned int x86_cache_size;
118 int x86_cache_alignment; /* In bytes */
119 /* Cache QoS architectural values: */
120 int x86_cache_max_rmid; /* max index */
121 int x86_cache_occ_scale; /* scale to bytes */
123 unsigned long loops_per_jiffy;
124 /* cpuid returned max cores value: */
128 u16 x86_clflush_size;
129 /* number of cores as seen by the OS: */
131 /* Physical processor id: */
135 /* Compute unit id */
137 /* Index into per_cpu list: */
142 #define X86_VENDOR_INTEL 0
143 #define X86_VENDOR_CYRIX 1
144 #define X86_VENDOR_AMD 2
145 #define X86_VENDOR_UMC 3
146 #define X86_VENDOR_CENTAUR 5
147 #define X86_VENDOR_TRANSMETA 7
148 #define X86_VENDOR_NSC 8
149 #define X86_VENDOR_NUM 9
151 #define X86_VENDOR_UNKNOWN 0xff
154 * capabilities of CPUs
156 extern struct cpuinfo_x86 boot_cpu_data;
157 extern struct cpuinfo_x86 new_cpu_data;
159 extern struct tss_struct doublefault_tss;
160 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
161 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
164 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
165 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
167 #define cpu_info boot_cpu_data
168 #define cpu_data(cpu) boot_cpu_data
171 extern const struct seq_operations cpuinfo_op;
173 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
175 extern void cpu_detect(struct cpuinfo_x86 *c);
177 static inline unsigned long long l1tf_pfn_limit(void)
179 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
182 extern void early_cpu_init(void);
183 extern void identify_boot_cpu(void);
184 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
185 extern void print_cpu_info(struct cpuinfo_x86 *);
186 void print_cpu_msr(struct cpuinfo_x86 *);
187 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
188 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
189 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
191 extern void detect_extended_topology(struct cpuinfo_x86 *c);
192 extern void detect_ht(struct cpuinfo_x86 *c);
195 extern int have_cpuid_p(void);
197 static inline int have_cpuid_p(void)
202 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
203 unsigned int *ecx, unsigned int *edx)
205 /* ecx is often an input as well as an output. */
211 : "0" (*eax), "2" (*ecx)
215 #define native_cpuid_reg(reg) \
216 static inline unsigned int native_cpuid_##reg(unsigned int op) \
218 unsigned int eax = op, ebx, ecx = 0, edx; \
220 native_cpuid(&eax, &ebx, &ecx, &edx); \
226 * Native CPUID functions returning a single datum.
228 native_cpuid_reg(eax)
229 native_cpuid_reg(ebx)
230 native_cpuid_reg(ecx)
231 native_cpuid_reg(edx)
233 static inline void load_cr3(pgd_t *pgdir)
235 write_cr3(__pa(pgdir));
239 /* This is the TSS defined by the hardware. */
241 unsigned short back_link, __blh;
243 unsigned short ss0, __ss0h;
247 * We don't use ring 1, so ss1 is a convenient scratch space in
248 * the same cacheline as sp0. We use ss1 to cache the value in
249 * MSR_IA32_SYSENTER_CS. When we context switch
250 * MSR_IA32_SYSENTER_CS, we first check if the new value being
251 * written matches ss1, and, if it's not, then we wrmsr the new
252 * value and update ss1.
254 * The only reason we context switch MSR_IA32_SYSENTER_CS is
255 * that we set it to zero in vm86 tasks to avoid corrupting the
256 * stack if we were to go through the sysenter path from vm86
259 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
261 unsigned short __ss1h;
263 unsigned short ss2, __ss2h;
275 unsigned short es, __esh;
276 unsigned short cs, __csh;
277 unsigned short ss, __ssh;
278 unsigned short ds, __dsh;
279 unsigned short fs, __fsh;
280 unsigned short gs, __gsh;
281 unsigned short ldt, __ldth;
282 unsigned short trace;
283 unsigned short io_bitmap_base;
285 } __attribute__((packed));
299 } __attribute__((packed)) ____cacheline_aligned;
305 #define IO_BITMAP_BITS 65536
306 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
307 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
308 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
309 #define INVALID_IO_BITMAP_OFFSET 0x8000
313 * The hardware state:
315 struct x86_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];
326 * Space for the temporary SYSENTER stack:
328 unsigned long SYSENTER_stack[64];
330 } ____cacheline_aligned;
332 DECLARE_PER_CPU_SHARED_ALIGNED_USER_MAPPED(struct tss_struct, cpu_tss);
335 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
339 * Save the original ist values for checking stack pointers during debugging
342 unsigned long ist[7];
346 DECLARE_PER_CPU(struct orig_ist, orig_ist);
348 union irq_stack_union {
349 char irq_stack[IRQ_STACK_SIZE];
351 * GCC hardcodes the stack canary as %gs:40. Since the
352 * irq_stack is the object at %gs:0, we reserve the bottom
353 * 48 bytes of the irq stack for the canary.
357 unsigned long stack_canary;
361 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
362 DECLARE_INIT_PER_CPU(irq_stack_union);
364 DECLARE_PER_CPU(char *, irq_stack_ptr);
365 DECLARE_PER_CPU(unsigned int, irq_count);
366 extern asmlinkage void ignore_sysret(void);
368 #ifdef CONFIG_CC_STACKPROTECTOR
370 * Make sure stack canary segment base is cached-aligned:
371 * "For Intel Atom processors, avoid non zero segment base address
372 * that is not aligned to cache line boundary at all cost."
373 * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
375 struct stack_canary {
376 char __pad[20]; /* canary at %gs:20 */
377 unsigned long canary;
379 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
382 * per-CPU IRQ handling stacks
385 u32 stack[THREAD_SIZE/sizeof(u32)];
386 } __aligned(THREAD_SIZE);
388 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
389 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
392 extern unsigned int xstate_size;
396 struct thread_struct {
397 /* Cached TLS descriptors: */
398 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
402 unsigned long sysenter_cs;
406 unsigned short fsindex;
407 unsigned short gsindex;
417 /* Save middle states of ptrace breakpoints */
418 struct perf_event *ptrace_bps[HBP_NUM];
419 /* Debug status used for traps, single steps, etc... */
420 unsigned long debugreg6;
421 /* Keep track of the exact dr7 value set by the user */
422 unsigned long ptrace_dr7;
425 unsigned long trap_nr;
426 unsigned long error_code;
428 /* Virtual 86 mode info */
431 /* IO permissions: */
432 unsigned long *io_bitmap_ptr;
434 /* Max allowed port in the bitmap, in bytes: */
435 unsigned io_bitmap_max;
437 /* Floating point and extended processor state */
440 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
446 * Set IOPL bits in EFLAGS from given mask
448 static inline void native_set_iopl_mask(unsigned mask)
453 asm volatile ("pushfl;"
460 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
465 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
467 tss->x86_tss.sp0 = thread->sp0;
469 /* Only happens when SEP is enabled, no need to test "SEP"arately: */
470 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
471 tss->x86_tss.ss1 = thread->sysenter_cs;
472 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
477 static inline void native_swapgs(void)
480 asm volatile("swapgs" ::: "memory");
484 static inline unsigned long current_top_of_stack(void)
487 return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
489 /* sp0 on x86_32 is special in and around vm86 mode. */
490 return this_cpu_read_stable(cpu_current_top_of_stack);
494 #ifdef CONFIG_PARAVIRT
495 #include <asm/paravirt.h>
497 #define __cpuid native_cpuid
498 #define paravirt_enabled() 0
499 #define paravirt_has(x) 0
501 static inline void load_sp0(struct tss_struct *tss,
502 struct thread_struct *thread)
504 native_load_sp0(tss, thread);
507 #define set_iopl_mask native_set_iopl_mask
508 #endif /* CONFIG_PARAVIRT */
515 /* Free all resources held by a thread. */
516 extern void release_thread(struct task_struct *);
518 unsigned long get_wchan(struct task_struct *p);
521 * Generic CPUID function
522 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
523 * resulting in stale register contents being returned.
525 static inline void cpuid(unsigned int op,
526 unsigned int *eax, unsigned int *ebx,
527 unsigned int *ecx, unsigned int *edx)
531 __cpuid(eax, ebx, ecx, edx);
534 /* Some CPUID calls want 'count' to be placed in ecx */
535 static inline void cpuid_count(unsigned int op, int count,
536 unsigned int *eax, unsigned int *ebx,
537 unsigned int *ecx, unsigned int *edx)
541 __cpuid(eax, ebx, ecx, edx);
545 * CPUID functions returning a single datum
547 static inline unsigned int cpuid_eax(unsigned int op)
549 unsigned int eax, ebx, ecx, edx;
551 cpuid(op, &eax, &ebx, &ecx, &edx);
556 static inline unsigned int cpuid_ebx(unsigned int op)
558 unsigned int eax, ebx, ecx, edx;
560 cpuid(op, &eax, &ebx, &ecx, &edx);
565 static inline unsigned int cpuid_ecx(unsigned int op)
567 unsigned int eax, ebx, ecx, edx;
569 cpuid(op, &eax, &ebx, &ecx, &edx);
574 static inline unsigned int cpuid_edx(unsigned int op)
576 unsigned int eax, ebx, ecx, edx;
578 cpuid(op, &eax, &ebx, &ecx, &edx);
583 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
584 static __always_inline void rep_nop(void)
586 asm volatile("rep; nop" ::: "memory");
589 static __always_inline void cpu_relax(void)
594 #define cpu_relax_lowlatency() cpu_relax()
596 /* Stop speculative execution and prefetching of modified code. */
597 static inline void sync_core(void)
603 * Do a CPUID if available, otherwise do a jump. The jump
604 * can conveniently enough be the jump around CPUID.
606 asm volatile("cmpl %2,%1\n\t"
611 : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
612 : "ebx", "ecx", "edx", "memory");
615 * CPUID is a barrier to speculative execution.
616 * Prefetched instructions are automatically
617 * invalidated when modified.
622 : "ebx", "ecx", "edx", "memory");
626 extern void select_idle_routine(const struct cpuinfo_x86 *c);
627 extern void init_amd_e400_c1e_mask(void);
629 extern unsigned long boot_option_idle_override;
630 extern bool amd_e400_c1e_detected;
632 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
635 extern void enable_sep_cpu(void);
636 extern int sysenter_setup(void);
638 extern void early_trap_init(void);
639 void early_trap_pf_init(void);
641 /* Defined in head.S */
642 extern struct desc_ptr early_gdt_descr;
644 extern void cpu_set_gdt(int);
645 extern void switch_to_new_gdt(int);
646 extern void load_percpu_segment(int);
647 extern void cpu_init(void);
649 static inline unsigned long get_debugctlmsr(void)
651 unsigned long debugctlmsr = 0;
653 #ifndef CONFIG_X86_DEBUGCTLMSR
654 if (boot_cpu_data.x86 < 6)
657 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
662 static inline void update_debugctlmsr(unsigned long debugctlmsr)
664 #ifndef CONFIG_X86_DEBUGCTLMSR
665 if (boot_cpu_data.x86 < 6)
668 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
671 extern void set_task_blockstep(struct task_struct *task, bool on);
673 /* Boot loader type from the setup header: */
674 extern int bootloader_type;
675 extern int bootloader_version;
677 extern char ignore_fpu_irq;
679 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
680 #define ARCH_HAS_PREFETCHW
681 #define ARCH_HAS_SPINLOCK_PREFETCH
684 # define BASE_PREFETCH ""
685 # define ARCH_HAS_PREFETCH
687 # define BASE_PREFETCH "prefetcht0 %P1"
691 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
693 * It's not worth to care about 3dnow prefetches for the K6
694 * because they are microcoded there and very slow.
696 static inline void prefetch(const void *x)
698 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
700 "m" (*(const char *)x));
704 * 3dnow prefetch to get an exclusive cache line.
705 * Useful for spinlocks to avoid one state transition in the
706 * cache coherency protocol:
708 static inline void prefetchw(const void *x)
710 alternative_input(BASE_PREFETCH, "prefetchw %P1",
711 X86_FEATURE_3DNOWPREFETCH,
712 "m" (*(const char *)x));
715 static inline void spin_lock_prefetch(const void *x)
720 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
721 TOP_OF_KERNEL_STACK_PADDING)
725 * User space process size: 3GB (default).
727 #define TASK_SIZE PAGE_OFFSET
728 #define TASK_SIZE_MAX TASK_SIZE
729 #define STACK_TOP TASK_SIZE
730 #define STACK_TOP_MAX STACK_TOP
732 #define INIT_THREAD { \
733 .sp0 = TOP_OF_INIT_STACK, \
734 .sysenter_cs = __KERNEL_CS, \
735 .io_bitmap_ptr = NULL, \
738 extern unsigned long thread_saved_pc(struct task_struct *tsk);
741 * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
742 * This is necessary to guarantee that the entire "struct pt_regs"
743 * is accessible even if the CPU haven't stored the SS/ESP registers
744 * on the stack (interrupt gate does not save these registers
745 * when switching to the same priv ring).
746 * Therefore beware: accessing the ss/esp fields of the
747 * "struct pt_regs" is possible, but they may contain the
748 * completely wrong values.
750 #define task_pt_regs(task) \
752 unsigned long __ptr = (unsigned long)task_stack_page(task); \
753 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
754 ((struct pt_regs *)__ptr) - 1; \
757 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
761 * User space process size. 47bits minus one guard page. The guard
762 * page is necessary on Intel CPUs: if a SYSCALL instruction is at
763 * the highest possible canonical userspace address, then that
764 * syscall will enter the kernel with a non-canonical return
765 * address, and SYSRET will explode dangerously. We avoid this
766 * particular problem by preventing anything from being mapped
767 * at the maximum canonical address.
769 #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE)
771 /* This decides where the kernel will search for a free chunk of vm
772 * space during mmap's.
774 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
775 0xc0000000 : 0xFFFFe000)
777 #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \
778 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
779 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
780 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
782 #define STACK_TOP TASK_SIZE
783 #define STACK_TOP_MAX TASK_SIZE_MAX
785 #define INIT_THREAD { \
786 .sp0 = TOP_OF_INIT_STACK \
790 * Return saved PC of a blocked thread.
791 * What is this good for? it will be always the scheduler or ret_from_fork.
793 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
795 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
796 extern unsigned long KSTK_ESP(struct task_struct *task);
798 #endif /* CONFIG_X86_64 */
800 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
801 unsigned long new_sp);
804 * This decides where the kernel will search for a free chunk of vm
805 * space during mmap's.
807 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
809 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
811 /* Get/set a process' ability to use the timestamp counter instruction */
812 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
813 #define SET_TSC_CTL(val) set_tsc_mode((val))
815 extern int get_tsc_mode(unsigned long adr);
816 extern int set_tsc_mode(unsigned int val);
818 /* Register/unregister a process' MPX related resource */
819 #define MPX_ENABLE_MANAGEMENT() mpx_enable_management()
820 #define MPX_DISABLE_MANAGEMENT() mpx_disable_management()
822 #ifdef CONFIG_X86_INTEL_MPX
823 extern int mpx_enable_management(void);
824 extern int mpx_disable_management(void);
826 static inline int mpx_enable_management(void)
830 static inline int mpx_disable_management(void)
834 #endif /* CONFIG_X86_INTEL_MPX */
836 extern u16 amd_get_nb_id(int cpu);
837 extern u32 amd_get_nodes_per_socket(void);
839 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
841 uint32_t base, eax, signature[3];
843 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
844 cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
846 if (!memcmp(sig, signature, 12) &&
847 (leaves == 0 || ((eax - base) >= leaves)))
854 extern unsigned long arch_align_stack(unsigned long sp);
855 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
857 void default_idle(void);
859 bool xen_set_default_idle(void);
861 #define xen_set_default_idle 0
864 void stop_this_cpu(void *dummy);
865 void df_debug(struct pt_regs *regs, long error_code);
867 enum mds_mitigations {
870 MDS_MITIGATION_VMWERV,
873 enum taa_mitigations {
875 TAA_MITIGATION_UCODE_NEEDED,
877 TAA_MITIGATION_TSX_DISABLED,
880 #endif /* _ASM_X86_PROCESSOR_H */