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>
20 #include <asm/pgtable_types.h>
21 #include <asm/percpu.h>
23 #include <asm/desc_defs.h>
25 #include <asm/special_insns.h>
26 #include <asm/fpu/types.h>
27 #include <asm/unwind_hints.h>
28 #include <asm/vmxfeatures.h>
29 #include <asm/vdso/processor.h>
31 #include <linux/personality.h>
32 #include <linux/cache.h>
33 #include <linux/threads.h>
34 #include <linux/math64.h>
35 #include <linux/err.h>
36 #include <linux/irqflags.h>
37 #include <linux/mem_encrypt.h>
40 * We handle most unaligned accesses in hardware. On the other hand
41 * unaligned DMA can be quite expensive on some Nehalem processors.
43 * Based on this we disable the IP header alignment in network drivers.
45 #define NET_IP_ALIGN 0
50 * These alignment constraints are for performance in the vSMP case,
51 * but in the task_struct case we must also meet hardware imposed
52 * alignment requirements of the FPU state:
54 #ifdef CONFIG_X86_VSMP
55 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
56 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
58 # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
59 # define ARCH_MIN_MMSTRUCT_ALIGN 0
67 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
68 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
69 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
70 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
71 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
72 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
73 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
76 * CPU type and hardware bug flags. Kept separately for each CPU.
77 * Members of this structure are referenced in head_32.S, so think twice
78 * before touching them. [mj]
82 __u8 x86; /* CPU family */
83 __u8 x86_vendor; /* CPU vendor */
87 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
90 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
91 __u32 vmx_capability[NVMXINTS];
95 /* CPUID returned core id bits: */
98 /* Max extended CPUID function supported: */
99 __u32 extended_cpuid_level;
100 /* Maximum supported CPUID level, -1=no CPUID: */
103 * Align to size of unsigned long because the x86_capability array
104 * is passed to bitops which require the alignment. Use unnamed
105 * union to enforce the array is aligned to size of unsigned long.
108 __u32 x86_capability[NCAPINTS + NBUGINTS];
109 unsigned long x86_capability_alignment;
111 char x86_vendor_id[16];
112 char x86_model_id[64];
113 /* in KB - valid for CPUS which support this call: */
114 unsigned int x86_cache_size;
115 int x86_cache_alignment; /* In bytes */
116 /* Cache QoS architectural values, valid only on the BSP: */
117 int x86_cache_max_rmid; /* max index */
118 int x86_cache_occ_scale; /* scale to bytes */
119 int x86_cache_mbm_width_offset;
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: */
137 /* Index into per_cpu list: */
140 /* Address space bits used by the cache internally */
142 unsigned initialized : 1;
143 } __randomize_layout;
146 u32 eax, ebx, ecx, edx;
149 enum cpuid_regs_idx {
156 #define X86_VENDOR_INTEL 0
157 #define X86_VENDOR_CYRIX 1
158 #define X86_VENDOR_AMD 2
159 #define X86_VENDOR_UMC 3
160 #define X86_VENDOR_CENTAUR 5
161 #define X86_VENDOR_TRANSMETA 7
162 #define X86_VENDOR_NSC 8
163 #define X86_VENDOR_HYGON 9
164 #define X86_VENDOR_ZHAOXIN 10
165 #define X86_VENDOR_NUM 11
167 #define X86_VENDOR_UNKNOWN 0xff
170 * capabilities of CPUs
172 extern struct cpuinfo_x86 boot_cpu_data;
173 extern struct cpuinfo_x86 new_cpu_data;
175 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
176 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
179 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
180 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
182 #define cpu_info boot_cpu_data
183 #define cpu_data(cpu) boot_cpu_data
186 extern const struct seq_operations cpuinfo_op;
188 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
190 extern void cpu_detect(struct cpuinfo_x86 *c);
192 static inline unsigned long long l1tf_pfn_limit(void)
194 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
197 extern void early_cpu_init(void);
198 extern void identify_boot_cpu(void);
199 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
200 extern void print_cpu_info(struct cpuinfo_x86 *);
201 void print_cpu_msr(struct cpuinfo_x86 *);
204 extern int have_cpuid_p(void);
206 static inline int have_cpuid_p(void)
211 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
212 unsigned int *ecx, unsigned int *edx)
214 /* ecx is often an input as well as an output. */
220 : "0" (*eax), "2" (*ecx)
224 #define native_cpuid_reg(reg) \
225 static inline unsigned int native_cpuid_##reg(unsigned int op) \
227 unsigned int eax = op, ebx, ecx = 0, edx; \
229 native_cpuid(&eax, &ebx, &ecx, &edx); \
235 * Native CPUID functions returning a single datum.
237 native_cpuid_reg(eax)
238 native_cpuid_reg(ebx)
239 native_cpuid_reg(ecx)
240 native_cpuid_reg(edx)
243 * Friendlier CR3 helpers.
245 static inline unsigned long read_cr3_pa(void)
247 return __read_cr3() & CR3_ADDR_MASK;
250 static inline unsigned long native_read_cr3_pa(void)
252 return __native_read_cr3() & CR3_ADDR_MASK;
255 static inline void load_cr3(pgd_t *pgdir)
257 write_cr3(__sme_pa(pgdir));
261 * Note that while the legacy 'TSS' name comes from 'Task State Segment',
262 * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
263 * unrelated to the task-switch mechanism:
266 /* This is the TSS defined by the hardware. */
268 unsigned short back_link, __blh;
270 unsigned short ss0, __ss0h;
274 * We don't use ring 1, so ss1 is a convenient scratch space in
275 * the same cacheline as sp0. We use ss1 to cache the value in
276 * MSR_IA32_SYSENTER_CS. When we context switch
277 * MSR_IA32_SYSENTER_CS, we first check if the new value being
278 * written matches ss1, and, if it's not, then we wrmsr the new
279 * value and update ss1.
281 * The only reason we context switch MSR_IA32_SYSENTER_CS is
282 * that we set it to zero in vm86 tasks to avoid corrupting the
283 * stack if we were to go through the sysenter path from vm86
286 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
288 unsigned short __ss1h;
290 unsigned short ss2, __ss2h;
302 unsigned short es, __esh;
303 unsigned short cs, __csh;
304 unsigned short ss, __ssh;
305 unsigned short ds, __dsh;
306 unsigned short fs, __fsh;
307 unsigned short gs, __gsh;
308 unsigned short ldt, __ldth;
309 unsigned short trace;
310 unsigned short io_bitmap_base;
312 } __attribute__((packed));
320 * Since Linux does not use ring 2, the 'sp2' slot is unused by
321 * hardware. entry_SYSCALL_64 uses it as scratch space to stash
322 * the user RSP value.
333 } __attribute__((packed));
339 #define IO_BITMAP_BITS 65536
340 #define IO_BITMAP_BYTES (IO_BITMAP_BITS / BITS_PER_BYTE)
341 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES / sizeof(long))
343 #define IO_BITMAP_OFFSET_VALID_MAP \
344 (offsetof(struct tss_struct, io_bitmap.bitmap) - \
345 offsetof(struct tss_struct, x86_tss))
347 #define IO_BITMAP_OFFSET_VALID_ALL \
348 (offsetof(struct tss_struct, io_bitmap.mapall) - \
349 offsetof(struct tss_struct, x86_tss))
351 #ifdef CONFIG_X86_IOPL_IOPERM
353 * sizeof(unsigned long) coming from an extra "long" at the end of the
354 * iobitmap. The limit is inclusive, i.e. the last valid byte.
356 # define __KERNEL_TSS_LIMIT \
357 (IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
358 sizeof(unsigned long) - 1)
360 # define __KERNEL_TSS_LIMIT \
361 (offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
364 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
365 #define IO_BITMAP_OFFSET_INVALID (__KERNEL_TSS_LIMIT + 1)
368 char stack[PAGE_SIZE];
371 struct entry_stack_page {
372 struct entry_stack stack;
373 } __aligned(PAGE_SIZE);
376 * All IO bitmap related data stored in the TSS:
378 struct x86_io_bitmap {
379 /* The sequence number of the last active bitmap. */
383 * Store the dirty size of the last io bitmap offender. The next
384 * one will have to do the cleanup as the switch out to a non io
385 * bitmap user will just set x86_tss.io_bitmap_base to a value
386 * outside of the TSS limit. So for sane tasks there is no need to
387 * actually touch the io_bitmap at all.
389 unsigned int prev_max;
392 * The extra 1 is there because the CPU will access an
393 * additional byte beyond the end of the IO permission
394 * bitmap. The extra byte must be all 1 bits, and must
395 * be within the limit.
397 unsigned long bitmap[IO_BITMAP_LONGS + 1];
400 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
401 * except the additional byte at the end.
403 unsigned long mapall[IO_BITMAP_LONGS + 1];
408 * The fixed hardware portion. This must not cross a page boundary
409 * at risk of violating the SDM's advice and potentially triggering
412 struct x86_hw_tss x86_tss;
414 struct x86_io_bitmap io_bitmap;
415 } __aligned(PAGE_SIZE);
417 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
419 /* Per CPU interrupt stacks */
421 char stack[IRQ_STACK_SIZE];
422 } __aligned(IRQ_STACK_SIZE);
424 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
427 struct fixed_percpu_data {
429 * GCC hardcodes the stack canary as %gs:40. Since the
430 * irq_stack is the object at %gs:0, we reserve the bottom
431 * 48 bytes of the irq stack for the canary.
433 * Once we are willing to require -mstack-protector-guard-symbol=
434 * support for x86_64 stackprotector, we can get rid of this.
437 unsigned long stack_canary;
440 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
441 DECLARE_INIT_PER_CPU(fixed_percpu_data);
443 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
445 return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
448 DECLARE_PER_CPU(void *, hardirq_stack_ptr);
449 DECLARE_PER_CPU(bool, hardirq_stack_inuse);
450 extern asmlinkage void ignore_sysret(void);
452 /* Save actual FS/GS selectors and bases to current->thread */
453 void current_save_fsgs(void);
455 #ifdef CONFIG_STACKPROTECTOR
456 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
458 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
459 DECLARE_PER_CPU(struct irq_stack *, softirq_stack_ptr);
462 extern unsigned int fpu_kernel_xstate_size;
463 extern unsigned int fpu_user_xstate_size;
467 struct thread_struct {
468 /* Cached TLS descriptors: */
469 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
475 unsigned long sysenter_cs;
479 unsigned short fsindex;
480 unsigned short gsindex;
484 unsigned long fsbase;
485 unsigned long gsbase;
488 * XXX: this could presumably be unsigned short. Alternatively,
489 * 32-bit kernels could be taught to use fsindex instead.
495 /* Save middle states of ptrace breakpoints */
496 struct perf_event *ptrace_bps[HBP_NUM];
497 /* Debug status used for traps, single steps, etc... */
498 unsigned long virtual_dr6;
499 /* Keep track of the exact dr7 value set by the user */
500 unsigned long ptrace_dr7;
503 unsigned long trap_nr;
504 unsigned long error_code;
506 /* Virtual 86 mode info */
509 /* IO permissions: */
510 struct io_bitmap *io_bitmap;
513 * IOPL. Privilege level dependent I/O permission which is
514 * emulated via the I/O bitmap to prevent user space from disabling
517 unsigned long iopl_emul;
519 unsigned int sig_on_uaccess_err:1;
521 /* Floating point and extended processor state */
524 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
529 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
530 static inline void arch_thread_struct_whitelist(unsigned long *offset,
533 *offset = offsetof(struct thread_struct, fpu.state);
534 *size = fpu_kernel_xstate_size;
538 native_load_sp0(unsigned long sp0)
540 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
543 static __always_inline void native_swapgs(void)
546 asm volatile("swapgs" ::: "memory");
550 static inline unsigned long current_top_of_stack(void)
553 * We can't read directly from tss.sp0: sp0 on x86_32 is special in
554 * and around vm86 mode and sp0 on x86_64 is special because of the
557 return this_cpu_read_stable(cpu_current_top_of_stack);
560 static inline bool on_thread_stack(void)
562 return (unsigned long)(current_top_of_stack() -
563 current_stack_pointer) < THREAD_SIZE;
566 #ifdef CONFIG_PARAVIRT_XXL
567 #include <asm/paravirt.h>
569 #define __cpuid native_cpuid
571 static inline void load_sp0(unsigned long sp0)
573 native_load_sp0(sp0);
576 #endif /* CONFIG_PARAVIRT_XXL */
578 /* Free all resources held by a thread. */
579 extern void release_thread(struct task_struct *);
581 unsigned long get_wchan(struct task_struct *p);
584 * Generic CPUID function
585 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
586 * resulting in stale register contents being returned.
588 static inline void cpuid(unsigned int op,
589 unsigned int *eax, unsigned int *ebx,
590 unsigned int *ecx, unsigned int *edx)
594 __cpuid(eax, ebx, ecx, edx);
597 /* Some CPUID calls want 'count' to be placed in ecx */
598 static inline void cpuid_count(unsigned int op, int count,
599 unsigned int *eax, unsigned int *ebx,
600 unsigned int *ecx, unsigned int *edx)
604 __cpuid(eax, ebx, ecx, edx);
608 * CPUID functions returning a single datum
610 static inline unsigned int cpuid_eax(unsigned int op)
612 unsigned int eax, ebx, ecx, edx;
614 cpuid(op, &eax, &ebx, &ecx, &edx);
619 static inline unsigned int cpuid_ebx(unsigned int op)
621 unsigned int eax, ebx, ecx, edx;
623 cpuid(op, &eax, &ebx, &ecx, &edx);
628 static inline unsigned int cpuid_ecx(unsigned int op)
630 unsigned int eax, ebx, ecx, edx;
632 cpuid(op, &eax, &ebx, &ecx, &edx);
637 static inline unsigned int cpuid_edx(unsigned int op)
639 unsigned int eax, ebx, ecx, edx;
641 cpuid(op, &eax, &ebx, &ecx, &edx);
646 extern void select_idle_routine(const struct cpuinfo_x86 *c);
647 extern void amd_e400_c1e_apic_setup(void);
649 extern unsigned long boot_option_idle_override;
651 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
654 extern void enable_sep_cpu(void);
655 extern int sysenter_setup(void);
658 /* Defined in head.S */
659 extern struct desc_ptr early_gdt_descr;
661 extern void switch_to_new_gdt(int);
662 extern void load_direct_gdt(int);
663 extern void load_fixmap_gdt(int);
664 extern void load_percpu_segment(int);
665 extern void cpu_init(void);
666 extern void cpu_init_exception_handling(void);
667 extern void cr4_init(void);
669 static inline unsigned long get_debugctlmsr(void)
671 unsigned long debugctlmsr = 0;
673 #ifndef CONFIG_X86_DEBUGCTLMSR
674 if (boot_cpu_data.x86 < 6)
677 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
682 static inline void update_debugctlmsr(unsigned long debugctlmsr)
684 #ifndef CONFIG_X86_DEBUGCTLMSR
685 if (boot_cpu_data.x86 < 6)
688 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
691 extern void set_task_blockstep(struct task_struct *task, bool on);
693 /* Boot loader type from the setup header: */
694 extern int bootloader_type;
695 extern int bootloader_version;
697 extern char ignore_fpu_irq;
699 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
700 #define ARCH_HAS_PREFETCHW
701 #define ARCH_HAS_SPINLOCK_PREFETCH
704 # define BASE_PREFETCH ""
705 # define ARCH_HAS_PREFETCH
707 # define BASE_PREFETCH "prefetcht0 %P1"
711 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
713 * It's not worth to care about 3dnow prefetches for the K6
714 * because they are microcoded there and very slow.
716 static inline void prefetch(const void *x)
718 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
720 "m" (*(const char *)x));
724 * 3dnow prefetch to get an exclusive cache line.
725 * Useful for spinlocks to avoid one state transition in the
726 * cache coherency protocol:
728 static __always_inline void prefetchw(const void *x)
730 alternative_input(BASE_PREFETCH, "prefetchw %P1",
731 X86_FEATURE_3DNOWPREFETCH,
732 "m" (*(const char *)x));
735 static inline void spin_lock_prefetch(const void *x)
740 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
741 TOP_OF_KERNEL_STACK_PADDING)
743 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
745 #define task_pt_regs(task) \
747 unsigned long __ptr = (unsigned long)task_stack_page(task); \
748 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
749 ((struct pt_regs *)__ptr) - 1; \
753 #define INIT_THREAD { \
754 .sp0 = TOP_OF_INIT_STACK, \
755 .sysenter_cs = __KERNEL_CS, \
758 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
761 #define INIT_THREAD { }
763 extern unsigned long KSTK_ESP(struct task_struct *task);
765 #endif /* CONFIG_X86_64 */
767 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
768 unsigned long new_sp);
771 * This decides where the kernel will search for a free chunk of vm
772 * space during mmap's.
774 #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3))
775 #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
777 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
779 /* Get/set a process' ability to use the timestamp counter instruction */
780 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
781 #define SET_TSC_CTL(val) set_tsc_mode((val))
783 extern int get_tsc_mode(unsigned long adr);
784 extern int set_tsc_mode(unsigned int val);
786 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
788 #ifdef CONFIG_CPU_SUP_AMD
789 extern u32 amd_get_nodes_per_socket(void);
791 static inline u32 amd_get_nodes_per_socket(void) { return 0; }
794 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
796 uint32_t base, eax, signature[3];
798 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
799 cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
801 if (!memcmp(sig, signature, 12) &&
802 (leaves == 0 || ((eax - base) >= leaves)))
809 extern unsigned long arch_align_stack(unsigned long sp);
810 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
811 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
813 void default_idle(void);
815 bool xen_set_default_idle(void);
817 #define xen_set_default_idle 0
820 void stop_this_cpu(void *dummy);
821 void microcode_check(void);
823 enum l1tf_mitigations {
825 L1TF_MITIGATION_FLUSH_NOWARN,
826 L1TF_MITIGATION_FLUSH,
827 L1TF_MITIGATION_FLUSH_NOSMT,
828 L1TF_MITIGATION_FULL,
829 L1TF_MITIGATION_FULL_FORCE
832 extern enum l1tf_mitigations l1tf_mitigation;
834 enum mds_mitigations {
837 MDS_MITIGATION_VMWERV,
840 #endif /* _ASM_X86_PROCESSOR_H */