1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PROCESSOR_H
3 #define _ASM_X86_PROCESSOR_H
5 #include <asm/processor-flags.h>
7 /* Forward declaration, a strange C thing */
13 #include <asm/math_emu.h>
14 #include <asm/segment.h>
15 #include <asm/types.h>
16 #include <uapi/asm/sigcontext.h>
17 #include <asm/current.h>
18 #include <asm/cpufeatures.h>
19 #include <asm/cpuid.h>
21 #include <asm/pgtable_types.h>
22 #include <asm/percpu.h>
24 #include <asm/desc_defs.h>
26 #include <asm/special_insns.h>
27 #include <asm/fpu/types.h>
28 #include <asm/unwind_hints.h>
29 #include <asm/vmxfeatures.h>
30 #include <asm/vdso/processor.h>
32 #include <linux/personality.h>
33 #include <linux/cache.h>
34 #include <linux/threads.h>
35 #include <linux/math64.h>
36 #include <linux/err.h>
37 #include <linux/irqflags.h>
38 #include <linux/mem_encrypt.h>
41 * We handle most unaligned accesses in hardware. On the other hand
42 * unaligned DMA can be quite expensive on some Nehalem processors.
44 * Based on this we disable the IP header alignment in network drivers.
46 #define NET_IP_ALIGN 0
51 * These alignment constraints are for performance in the vSMP case,
52 * but in the task_struct case we must also meet hardware imposed
53 * alignment requirements of the FPU state:
55 #ifdef CONFIG_X86_VSMP
56 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
57 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
59 # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
60 # define ARCH_MIN_MMSTRUCT_ALIGN 0
68 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
69 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
70 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
71 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
72 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
73 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
77 * CPU type and hardware bug flags. Kept separately for each CPU.
78 * Members of this structure are referenced in head_32.S, so think twice
79 * before touching them. [mj]
83 __u8 x86; /* CPU family */
84 __u8 x86_vendor; /* CPU vendor */
88 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
91 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
92 __u32 vmx_capability[NVMXINTS];
96 /* CPUID returned core id bits: */
99 /* Max extended CPUID function supported: */
100 __u32 extended_cpuid_level;
101 /* Maximum supported CPUID level, -1=no CPUID: */
104 * Align to size of unsigned long because the x86_capability array
105 * is passed to bitops which require the alignment. Use unnamed
106 * union to enforce the array is aligned to size of unsigned long.
109 __u32 x86_capability[NCAPINTS + NBUGINTS];
110 unsigned long x86_capability_alignment;
112 char x86_vendor_id[16];
113 char x86_model_id[64];
114 /* in KB - valid for CPUS which support this call: */
115 unsigned int x86_cache_size;
116 int x86_cache_alignment; /* In bytes */
117 /* Cache QoS architectural values, valid only on the BSP: */
118 int x86_cache_max_rmid; /* max index */
119 int x86_cache_occ_scale; /* scale to bytes */
120 int x86_cache_mbm_width_offset;
122 unsigned long loops_per_jiffy;
123 /* protected processor identification number */
125 /* cpuid returned max cores value: */
129 u16 x86_clflush_size;
130 /* number of cores as seen by the OS: */
132 /* Physical processor id: */
134 /* Logical processor id: */
140 /* Index into per_cpu list: */
142 /* Is SMT active on this core? */
145 /* Address space bits used by the cache internally */
147 unsigned initialized : 1;
148 } __randomize_layout;
150 #define X86_VENDOR_INTEL 0
151 #define X86_VENDOR_CYRIX 1
152 #define X86_VENDOR_AMD 2
153 #define X86_VENDOR_UMC 3
154 #define X86_VENDOR_CENTAUR 5
155 #define X86_VENDOR_TRANSMETA 7
156 #define X86_VENDOR_NSC 8
157 #define X86_VENDOR_HYGON 9
158 #define X86_VENDOR_ZHAOXIN 10
159 #define X86_VENDOR_VORTEX 11
160 #define X86_VENDOR_NUM 12
162 #define X86_VENDOR_UNKNOWN 0xff
165 * capabilities of CPUs
167 extern struct cpuinfo_x86 boot_cpu_data;
168 extern struct cpuinfo_x86 new_cpu_data;
170 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
171 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
174 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
175 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
177 #define cpu_info boot_cpu_data
178 #define cpu_data(cpu) boot_cpu_data
181 extern const struct seq_operations cpuinfo_op;
183 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
185 extern void cpu_detect(struct cpuinfo_x86 *c);
187 static inline unsigned long long l1tf_pfn_limit(void)
189 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
192 extern void early_cpu_init(void);
193 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
194 extern void print_cpu_info(struct cpuinfo_x86 *);
195 void print_cpu_msr(struct cpuinfo_x86 *);
198 * Friendlier CR3 helpers.
200 static inline unsigned long read_cr3_pa(void)
202 return __read_cr3() & CR3_ADDR_MASK;
205 static inline unsigned long native_read_cr3_pa(void)
207 return __native_read_cr3() & CR3_ADDR_MASK;
210 static inline void load_cr3(pgd_t *pgdir)
212 write_cr3(__sme_pa(pgdir));
216 * Note that while the legacy 'TSS' name comes from 'Task State Segment',
217 * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
218 * unrelated to the task-switch mechanism:
221 /* This is the TSS defined by the hardware. */
223 unsigned short back_link, __blh;
225 unsigned short ss0, __ss0h;
229 * We don't use ring 1, so ss1 is a convenient scratch space in
230 * the same cacheline as sp0. We use ss1 to cache the value in
231 * MSR_IA32_SYSENTER_CS. When we context switch
232 * MSR_IA32_SYSENTER_CS, we first check if the new value being
233 * written matches ss1, and, if it's not, then we wrmsr the new
234 * value and update ss1.
236 * The only reason we context switch MSR_IA32_SYSENTER_CS is
237 * that we set it to zero in vm86 tasks to avoid corrupting the
238 * stack if we were to go through the sysenter path from vm86
241 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
243 unsigned short __ss1h;
245 unsigned short ss2, __ss2h;
257 unsigned short es, __esh;
258 unsigned short cs, __csh;
259 unsigned short ss, __ssh;
260 unsigned short ds, __dsh;
261 unsigned short fs, __fsh;
262 unsigned short gs, __gsh;
263 unsigned short ldt, __ldth;
264 unsigned short trace;
265 unsigned short io_bitmap_base;
267 } __attribute__((packed));
275 * Since Linux does not use ring 2, the 'sp2' slot is unused by
276 * hardware. entry_SYSCALL_64 uses it as scratch space to stash
277 * the user RSP value.
288 } __attribute__((packed));
294 #define IO_BITMAP_BITS 65536
295 #define IO_BITMAP_BYTES (IO_BITMAP_BITS / BITS_PER_BYTE)
296 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES / sizeof(long))
298 #define IO_BITMAP_OFFSET_VALID_MAP \
299 (offsetof(struct tss_struct, io_bitmap.bitmap) - \
300 offsetof(struct tss_struct, x86_tss))
302 #define IO_BITMAP_OFFSET_VALID_ALL \
303 (offsetof(struct tss_struct, io_bitmap.mapall) - \
304 offsetof(struct tss_struct, x86_tss))
306 #ifdef CONFIG_X86_IOPL_IOPERM
308 * sizeof(unsigned long) coming from an extra "long" at the end of the
309 * iobitmap. The limit is inclusive, i.e. the last valid byte.
311 # define __KERNEL_TSS_LIMIT \
312 (IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
313 sizeof(unsigned long) - 1)
315 # define __KERNEL_TSS_LIMIT \
316 (offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
319 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
320 #define IO_BITMAP_OFFSET_INVALID (__KERNEL_TSS_LIMIT + 1)
323 char stack[PAGE_SIZE];
326 struct entry_stack_page {
327 struct entry_stack stack;
328 } __aligned(PAGE_SIZE);
331 * All IO bitmap related data stored in the TSS:
333 struct x86_io_bitmap {
334 /* The sequence number of the last active bitmap. */
338 * Store the dirty size of the last io bitmap offender. The next
339 * one will have to do the cleanup as the switch out to a non io
340 * bitmap user will just set x86_tss.io_bitmap_base to a value
341 * outside of the TSS limit. So for sane tasks there is no need to
342 * actually touch the io_bitmap at all.
344 unsigned int prev_max;
347 * The extra 1 is there because the CPU will access an
348 * additional byte beyond the end of the IO permission
349 * bitmap. The extra byte must be all 1 bits, and must
350 * be within the limit.
352 unsigned long bitmap[IO_BITMAP_LONGS + 1];
355 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
356 * except the additional byte at the end.
358 unsigned long mapall[IO_BITMAP_LONGS + 1];
363 * The fixed hardware portion. This must not cross a page boundary
364 * at risk of violating the SDM's advice and potentially triggering
367 struct x86_hw_tss x86_tss;
369 struct x86_io_bitmap io_bitmap;
370 } __aligned(PAGE_SIZE);
372 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
374 /* Per CPU interrupt stacks */
376 char stack[IRQ_STACK_SIZE];
377 } __aligned(IRQ_STACK_SIZE);
380 struct fixed_percpu_data {
382 * GCC hardcodes the stack canary as %gs:40. Since the
383 * irq_stack is the object at %gs:0, we reserve the bottom
384 * 48 bytes of the irq stack for the canary.
386 * Once we are willing to require -mstack-protector-guard-symbol=
387 * support for x86_64 stackprotector, we can get rid of this.
390 unsigned long stack_canary;
393 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
394 DECLARE_INIT_PER_CPU(fixed_percpu_data);
396 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
398 return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
401 extern asmlinkage void ignore_sysret(void);
403 /* Save actual FS/GS selectors and bases to current->thread */
404 void current_save_fsgs(void);
406 #ifdef CONFIG_STACKPROTECTOR
407 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
413 struct thread_struct {
414 /* Cached TLS descriptors: */
415 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
421 unsigned long sysenter_cs;
425 unsigned short fsindex;
426 unsigned short gsindex;
430 unsigned long fsbase;
431 unsigned long gsbase;
434 * XXX: this could presumably be unsigned short. Alternatively,
435 * 32-bit kernels could be taught to use fsindex instead.
441 /* Save middle states of ptrace breakpoints */
442 struct perf_event *ptrace_bps[HBP_NUM];
443 /* Debug status used for traps, single steps, etc... */
444 unsigned long virtual_dr6;
445 /* Keep track of the exact dr7 value set by the user */
446 unsigned long ptrace_dr7;
449 unsigned long trap_nr;
450 unsigned long error_code;
452 /* Virtual 86 mode info */
455 /* IO permissions: */
456 struct io_bitmap *io_bitmap;
459 * IOPL. Privilege level dependent I/O permission which is
460 * emulated via the I/O bitmap to prevent user space from disabling
463 unsigned long iopl_emul;
465 unsigned int iopl_warn:1;
466 unsigned int sig_on_uaccess_err:1;
469 * Protection Keys Register for Userspace. Loaded immediately on
470 * context switch. Store it in thread_struct to avoid a lookup in
471 * the tasks's FPU xstate buffer. This value is only valid when a
472 * task is scheduled out. For 'current' the authoritative source of
473 * PKRU is the hardware itself.
477 /* Floating point and extended processor state */
480 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
485 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
487 static inline void arch_thread_struct_whitelist(unsigned long *offset,
490 fpu_thread_struct_whitelist(offset, size);
494 native_load_sp0(unsigned long sp0)
496 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
499 static __always_inline void native_swapgs(void)
502 asm volatile("swapgs" ::: "memory");
506 static __always_inline unsigned long current_top_of_stack(void)
509 * We can't read directly from tss.sp0: sp0 on x86_32 is special in
510 * and around vm86 mode and sp0 on x86_64 is special because of the
513 return this_cpu_read_stable(pcpu_hot.top_of_stack);
516 static __always_inline bool on_thread_stack(void)
518 return (unsigned long)(current_top_of_stack() -
519 current_stack_pointer) < THREAD_SIZE;
522 #ifdef CONFIG_PARAVIRT_XXL
523 #include <asm/paravirt.h>
526 static inline void load_sp0(unsigned long sp0)
528 native_load_sp0(sp0);
531 #endif /* CONFIG_PARAVIRT_XXL */
533 unsigned long __get_wchan(struct task_struct *p);
535 extern void select_idle_routine(const struct cpuinfo_x86 *c);
536 extern void amd_e400_c1e_apic_setup(void);
538 extern unsigned long boot_option_idle_override;
540 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
543 extern void enable_sep_cpu(void);
546 /* Defined in head.S */
547 extern struct desc_ptr early_gdt_descr;
549 extern void switch_gdt_and_percpu_base(int);
550 extern void load_direct_gdt(int);
551 extern void load_fixmap_gdt(int);
552 extern void cpu_init(void);
553 extern void cpu_init_exception_handling(void);
554 extern void cr4_init(void);
556 static inline unsigned long get_debugctlmsr(void)
558 unsigned long debugctlmsr = 0;
560 #ifndef CONFIG_X86_DEBUGCTLMSR
561 if (boot_cpu_data.x86 < 6)
564 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
569 static inline void update_debugctlmsr(unsigned long debugctlmsr)
571 #ifndef CONFIG_X86_DEBUGCTLMSR
572 if (boot_cpu_data.x86 < 6)
575 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
578 extern void set_task_blockstep(struct task_struct *task, bool on);
580 /* Boot loader type from the setup header: */
581 extern int bootloader_type;
582 extern int bootloader_version;
584 extern char ignore_fpu_irq;
586 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
587 #define ARCH_HAS_PREFETCHW
590 # define BASE_PREFETCH ""
591 # define ARCH_HAS_PREFETCH
593 # define BASE_PREFETCH "prefetcht0 %P1"
597 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
599 * It's not worth to care about 3dnow prefetches for the K6
600 * because they are microcoded there and very slow.
602 static inline void prefetch(const void *x)
604 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
606 "m" (*(const char *)x));
610 * 3dnow prefetch to get an exclusive cache line.
611 * Useful for spinlocks to avoid one state transition in the
612 * cache coherency protocol:
614 static __always_inline void prefetchw(const void *x)
616 alternative_input(BASE_PREFETCH, "prefetchw %P1",
617 X86_FEATURE_3DNOWPREFETCH,
618 "m" (*(const char *)x));
621 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
622 TOP_OF_KERNEL_STACK_PADDING)
624 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
626 #define task_pt_regs(task) \
628 unsigned long __ptr = (unsigned long)task_stack_page(task); \
629 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
630 ((struct pt_regs *)__ptr) - 1; \
634 #define INIT_THREAD { \
635 .sp0 = TOP_OF_INIT_STACK, \
636 .sysenter_cs = __KERNEL_CS, \
639 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
642 extern unsigned long __end_init_task[];
644 #define INIT_THREAD { \
645 .sp = (unsigned long)&__end_init_task - sizeof(struct pt_regs), \
648 extern unsigned long KSTK_ESP(struct task_struct *task);
650 #endif /* CONFIG_X86_64 */
652 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
653 unsigned long new_sp);
656 * This decides where the kernel will search for a free chunk of vm
657 * space during mmap's.
659 #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3))
660 #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
662 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
664 /* Get/set a process' ability to use the timestamp counter instruction */
665 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
666 #define SET_TSC_CTL(val) set_tsc_mode((val))
668 extern int get_tsc_mode(unsigned long adr);
669 extern int set_tsc_mode(unsigned int val);
671 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
673 extern u16 get_llc_id(unsigned int cpu);
675 #ifdef CONFIG_CPU_SUP_AMD
676 extern u32 amd_get_nodes_per_socket(void);
677 extern u32 amd_get_highest_perf(void);
678 extern bool cpu_has_ibpb_brtype_microcode(void);
679 extern void amd_clear_divider(void);
680 extern void amd_check_microcode(void);
682 static inline u32 amd_get_nodes_per_socket(void) { return 0; }
683 static inline u32 amd_get_highest_perf(void) { return 0; }
684 static inline bool cpu_has_ibpb_brtype_microcode(void) { return false; }
685 static inline void amd_clear_divider(void) { }
686 static inline void amd_check_microcode(void) { }
689 extern unsigned long arch_align_stack(unsigned long sp);
690 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
691 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
693 void default_idle(void);
695 bool xen_set_default_idle(void);
697 #define xen_set_default_idle 0
700 void __noreturn stop_this_cpu(void *dummy);
701 void microcode_check(struct cpuinfo_x86 *prev_info);
702 void store_cpu_caps(struct cpuinfo_x86 *info);
704 enum l1tf_mitigations {
706 L1TF_MITIGATION_FLUSH_NOWARN,
707 L1TF_MITIGATION_FLUSH,
708 L1TF_MITIGATION_FLUSH_NOSMT,
709 L1TF_MITIGATION_FULL,
710 L1TF_MITIGATION_FULL_FORCE
713 extern enum l1tf_mitigations l1tf_mitigation;
715 enum mds_mitigations {
718 MDS_MITIGATION_VMWERV,
721 #ifdef CONFIG_X86_SGX
722 int arch_memory_failure(unsigned long pfn, int flags);
723 #define arch_memory_failure arch_memory_failure
725 bool arch_is_platform_page(u64 paddr);
726 #define arch_is_platform_page arch_is_platform_page
729 extern bool gds_ucode_mitigated(void);
731 #endif /* _ASM_X86_PROCESSOR_H */