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>
31 #include <asm/shstk.h>
33 #include <linux/personality.h>
34 #include <linux/cache.h>
35 #include <linux/threads.h>
36 #include <linux/math64.h>
37 #include <linux/err.h>
38 #include <linux/irqflags.h>
39 #include <linux/mem_encrypt.h>
42 * We handle most unaligned accesses in hardware. On the other hand
43 * unaligned DMA can be quite expensive on some Nehalem processors.
45 * Based on this we disable the IP header alignment in network drivers.
47 #define NET_IP_ALIGN 0
52 * These alignment constraints are for performance in the vSMP case,
53 * but in the task_struct case we must also meet hardware imposed
54 * alignment requirements of the FPU state:
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 __alignof__(union fpregs_state)
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.
81 struct cpuinfo_topology {
82 // Real APIC ID read from the local APIC
84 // The initial APIC ID provided by CPUID
87 // Physical package ID
90 // Physical die ID on AMD, Relative on Intel
93 // Compute unit ID - AMD specific
96 // Core ID relative to the package
99 // Logical ID mappings
103 // Cache level topology IDs
109 __u8 x86; /* CPU family */
110 __u8 x86_vendor; /* CPU vendor */
114 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
117 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
118 __u32 vmx_capability[NVMXINTS];
122 /* CPUID returned core id bits: */
123 __u8 x86_coreid_bits;
124 /* Max extended CPUID function supported: */
125 __u32 extended_cpuid_level;
126 /* Maximum supported CPUID level, -1=no CPUID: */
129 * Align to size of unsigned long because the x86_capability array
130 * is passed to bitops which require the alignment. Use unnamed
131 * union to enforce the array is aligned to size of unsigned long.
134 __u32 x86_capability[NCAPINTS + NBUGINTS];
135 unsigned long x86_capability_alignment;
137 char x86_vendor_id[16];
138 char x86_model_id[64];
139 struct cpuinfo_topology topo;
140 /* in KB - valid for CPUS which support this call: */
141 unsigned int x86_cache_size;
142 int x86_cache_alignment; /* In bytes */
143 /* Cache QoS architectural values, valid only on the BSP: */
144 int x86_cache_max_rmid; /* max index */
145 int x86_cache_occ_scale; /* scale to bytes */
146 int x86_cache_mbm_width_offset;
148 unsigned long loops_per_jiffy;
149 /* protected processor identification number */
151 /* cpuid returned max cores value: */
153 u16 x86_clflush_size;
154 /* number of cores as seen by the OS: */
156 /* Index into per_cpu list: */
158 /* Is SMT active on this core? */
161 /* Address space bits used by the cache internally */
163 unsigned initialized : 1;
164 } __randomize_layout;
166 #define X86_VENDOR_INTEL 0
167 #define X86_VENDOR_CYRIX 1
168 #define X86_VENDOR_AMD 2
169 #define X86_VENDOR_UMC 3
170 #define X86_VENDOR_CENTAUR 5
171 #define X86_VENDOR_TRANSMETA 7
172 #define X86_VENDOR_NSC 8
173 #define X86_VENDOR_HYGON 9
174 #define X86_VENDOR_ZHAOXIN 10
175 #define X86_VENDOR_VORTEX 11
176 #define X86_VENDOR_NUM 12
178 #define X86_VENDOR_UNKNOWN 0xff
181 * capabilities of CPUs
183 extern struct cpuinfo_x86 boot_cpu_data;
184 extern struct cpuinfo_x86 new_cpu_data;
186 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
187 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
190 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
191 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
193 #define cpu_info boot_cpu_data
194 #define cpu_data(cpu) boot_cpu_data
197 extern const struct seq_operations cpuinfo_op;
199 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
201 extern void cpu_detect(struct cpuinfo_x86 *c);
203 static inline unsigned long long l1tf_pfn_limit(void)
205 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
208 extern void early_cpu_init(void);
209 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
210 extern void print_cpu_info(struct cpuinfo_x86 *);
211 void print_cpu_msr(struct cpuinfo_x86 *);
214 * Friendlier CR3 helpers.
216 static inline unsigned long read_cr3_pa(void)
218 return __read_cr3() & CR3_ADDR_MASK;
221 static inline unsigned long native_read_cr3_pa(void)
223 return __native_read_cr3() & CR3_ADDR_MASK;
226 static inline void load_cr3(pgd_t *pgdir)
228 write_cr3(__sme_pa(pgdir));
232 * Note that while the legacy 'TSS' name comes from 'Task State Segment',
233 * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
234 * unrelated to the task-switch mechanism:
237 /* This is the TSS defined by the hardware. */
239 unsigned short back_link, __blh;
241 unsigned short ss0, __ss0h;
245 * We don't use ring 1, so ss1 is a convenient scratch space in
246 * the same cacheline as sp0. We use ss1 to cache the value in
247 * MSR_IA32_SYSENTER_CS. When we context switch
248 * MSR_IA32_SYSENTER_CS, we first check if the new value being
249 * written matches ss1, and, if it's not, then we wrmsr the new
250 * value and update ss1.
252 * The only reason we context switch MSR_IA32_SYSENTER_CS is
253 * that we set it to zero in vm86 tasks to avoid corrupting the
254 * stack if we were to go through the sysenter path from vm86
257 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
259 unsigned short __ss1h;
261 unsigned short ss2, __ss2h;
273 unsigned short es, __esh;
274 unsigned short cs, __csh;
275 unsigned short ss, __ssh;
276 unsigned short ds, __dsh;
277 unsigned short fs, __fsh;
278 unsigned short gs, __gsh;
279 unsigned short ldt, __ldth;
280 unsigned short trace;
281 unsigned short io_bitmap_base;
283 } __attribute__((packed));
291 * Since Linux does not use ring 2, the 'sp2' slot is unused by
292 * hardware. entry_SYSCALL_64 uses it as scratch space to stash
293 * the user RSP value.
304 } __attribute__((packed));
310 #define IO_BITMAP_BITS 65536
311 #define IO_BITMAP_BYTES (IO_BITMAP_BITS / BITS_PER_BYTE)
312 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES / sizeof(long))
314 #define IO_BITMAP_OFFSET_VALID_MAP \
315 (offsetof(struct tss_struct, io_bitmap.bitmap) - \
316 offsetof(struct tss_struct, x86_tss))
318 #define IO_BITMAP_OFFSET_VALID_ALL \
319 (offsetof(struct tss_struct, io_bitmap.mapall) - \
320 offsetof(struct tss_struct, x86_tss))
322 #ifdef CONFIG_X86_IOPL_IOPERM
324 * sizeof(unsigned long) coming from an extra "long" at the end of the
325 * iobitmap. The limit is inclusive, i.e. the last valid byte.
327 # define __KERNEL_TSS_LIMIT \
328 (IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
329 sizeof(unsigned long) - 1)
331 # define __KERNEL_TSS_LIMIT \
332 (offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
335 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
336 #define IO_BITMAP_OFFSET_INVALID (__KERNEL_TSS_LIMIT + 1)
339 char stack[PAGE_SIZE];
342 struct entry_stack_page {
343 struct entry_stack stack;
344 } __aligned(PAGE_SIZE);
347 * All IO bitmap related data stored in the TSS:
349 struct x86_io_bitmap {
350 /* The sequence number of the last active bitmap. */
354 * Store the dirty size of the last io bitmap offender. The next
355 * one will have to do the cleanup as the switch out to a non io
356 * bitmap user will just set x86_tss.io_bitmap_base to a value
357 * outside of the TSS limit. So for sane tasks there is no need to
358 * actually touch the io_bitmap at all.
360 unsigned int prev_max;
363 * The extra 1 is there because the CPU will access an
364 * additional byte beyond the end of the IO permission
365 * bitmap. The extra byte must be all 1 bits, and must
366 * be within the limit.
368 unsigned long bitmap[IO_BITMAP_LONGS + 1];
371 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
372 * except the additional byte at the end.
374 unsigned long mapall[IO_BITMAP_LONGS + 1];
379 * The fixed hardware portion. This must not cross a page boundary
380 * at risk of violating the SDM's advice and potentially triggering
383 struct x86_hw_tss x86_tss;
385 struct x86_io_bitmap io_bitmap;
386 } __aligned(PAGE_SIZE);
388 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
390 /* Per CPU interrupt stacks */
392 char stack[IRQ_STACK_SIZE];
393 } __aligned(IRQ_STACK_SIZE);
396 struct fixed_percpu_data {
398 * GCC hardcodes the stack canary as %gs:40. Since the
399 * irq_stack is the object at %gs:0, we reserve the bottom
400 * 48 bytes of the irq stack for the canary.
402 * Once we are willing to require -mstack-protector-guard-symbol=
403 * support for x86_64 stackprotector, we can get rid of this.
406 unsigned long stack_canary;
409 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
410 DECLARE_INIT_PER_CPU(fixed_percpu_data);
412 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
414 return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
417 extern asmlinkage void entry_SYSCALL32_ignore(void);
419 /* Save actual FS/GS selectors and bases to current->thread */
420 void current_save_fsgs(void);
422 #ifdef CONFIG_STACKPROTECTOR
423 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
429 struct thread_struct {
430 /* Cached TLS descriptors: */
431 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
437 unsigned long sysenter_cs;
441 unsigned short fsindex;
442 unsigned short gsindex;
446 unsigned long fsbase;
447 unsigned long gsbase;
450 * XXX: this could presumably be unsigned short. Alternatively,
451 * 32-bit kernels could be taught to use fsindex instead.
457 /* Save middle states of ptrace breakpoints */
458 struct perf_event *ptrace_bps[HBP_NUM];
459 /* Debug status used for traps, single steps, etc... */
460 unsigned long virtual_dr6;
461 /* Keep track of the exact dr7 value set by the user */
462 unsigned long ptrace_dr7;
465 unsigned long trap_nr;
466 unsigned long error_code;
468 /* Virtual 86 mode info */
471 /* IO permissions: */
472 struct io_bitmap *io_bitmap;
475 * IOPL. Privilege level dependent I/O permission which is
476 * emulated via the I/O bitmap to prevent user space from disabling
479 unsigned long iopl_emul;
481 unsigned int iopl_warn:1;
482 unsigned int sig_on_uaccess_err:1;
485 * Protection Keys Register for Userspace. Loaded immediately on
486 * context switch. Store it in thread_struct to avoid a lookup in
487 * the tasks's FPU xstate buffer. This value is only valid when a
488 * task is scheduled out. For 'current' the authoritative source of
489 * PKRU is the hardware itself.
493 #ifdef CONFIG_X86_USER_SHADOW_STACK
494 unsigned long features;
495 unsigned long features_locked;
497 struct thread_shstk shstk;
500 /* Floating point and extended processor state */
503 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
508 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
510 static inline void arch_thread_struct_whitelist(unsigned long *offset,
513 fpu_thread_struct_whitelist(offset, size);
517 native_load_sp0(unsigned long sp0)
519 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
522 static __always_inline void native_swapgs(void)
525 asm volatile("swapgs" ::: "memory");
529 static __always_inline unsigned long current_top_of_stack(void)
532 * We can't read directly from tss.sp0: sp0 on x86_32 is special in
533 * and around vm86 mode and sp0 on x86_64 is special because of the
536 return this_cpu_read_stable(pcpu_hot.top_of_stack);
539 static __always_inline bool on_thread_stack(void)
541 return (unsigned long)(current_top_of_stack() -
542 current_stack_pointer) < THREAD_SIZE;
545 #ifdef CONFIG_PARAVIRT_XXL
546 #include <asm/paravirt.h>
549 static inline void load_sp0(unsigned long sp0)
551 native_load_sp0(sp0);
554 #endif /* CONFIG_PARAVIRT_XXL */
556 unsigned long __get_wchan(struct task_struct *p);
558 extern void select_idle_routine(const struct cpuinfo_x86 *c);
559 extern void amd_e400_c1e_apic_setup(void);
561 extern unsigned long boot_option_idle_override;
563 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
566 extern void enable_sep_cpu(void);
569 /* Defined in head.S */
570 extern struct desc_ptr early_gdt_descr;
572 extern void switch_gdt_and_percpu_base(int);
573 extern void load_direct_gdt(int);
574 extern void load_fixmap_gdt(int);
575 extern void cpu_init(void);
576 extern void cpu_init_exception_handling(void);
577 extern void cr4_init(void);
579 static inline unsigned long get_debugctlmsr(void)
581 unsigned long debugctlmsr = 0;
583 #ifndef CONFIG_X86_DEBUGCTLMSR
584 if (boot_cpu_data.x86 < 6)
587 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
592 static inline void update_debugctlmsr(unsigned long debugctlmsr)
594 #ifndef CONFIG_X86_DEBUGCTLMSR
595 if (boot_cpu_data.x86 < 6)
598 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
601 extern void set_task_blockstep(struct task_struct *task, bool on);
603 /* Boot loader type from the setup header: */
604 extern int bootloader_type;
605 extern int bootloader_version;
607 extern char ignore_fpu_irq;
609 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
610 #define ARCH_HAS_PREFETCHW
613 # define BASE_PREFETCH ""
614 # define ARCH_HAS_PREFETCH
616 # define BASE_PREFETCH "prefetcht0 %P1"
620 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
622 * It's not worth to care about 3dnow prefetches for the K6
623 * because they are microcoded there and very slow.
625 static inline void prefetch(const void *x)
627 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
629 "m" (*(const char *)x));
633 * 3dnow prefetch to get an exclusive cache line.
634 * Useful for spinlocks to avoid one state transition in the
635 * cache coherency protocol:
637 static __always_inline void prefetchw(const void *x)
639 alternative_input(BASE_PREFETCH, "prefetchw %P1",
640 X86_FEATURE_3DNOWPREFETCH,
641 "m" (*(const char *)x));
644 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
645 TOP_OF_KERNEL_STACK_PADDING)
647 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
649 #define task_pt_regs(task) \
651 unsigned long __ptr = (unsigned long)task_stack_page(task); \
652 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
653 ((struct pt_regs *)__ptr) - 1; \
657 #define INIT_THREAD { \
658 .sp0 = TOP_OF_INIT_STACK, \
659 .sysenter_cs = __KERNEL_CS, \
662 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
665 extern unsigned long __end_init_task[];
667 #define INIT_THREAD { \
668 .sp = (unsigned long)&__end_init_task - sizeof(struct pt_regs), \
671 extern unsigned long KSTK_ESP(struct task_struct *task);
673 #endif /* CONFIG_X86_64 */
675 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
676 unsigned long new_sp);
679 * This decides where the kernel will search for a free chunk of vm
680 * space during mmap's.
682 #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3))
683 #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
685 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
687 /* Get/set a process' ability to use the timestamp counter instruction */
688 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
689 #define SET_TSC_CTL(val) set_tsc_mode((val))
691 extern int get_tsc_mode(unsigned long adr);
692 extern int set_tsc_mode(unsigned int val);
694 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
696 static inline u32 per_cpu_llc_id(unsigned int cpu)
698 return per_cpu(cpu_info.topo.llc_id, cpu);
701 static inline u32 per_cpu_l2c_id(unsigned int cpu)
703 return per_cpu(cpu_info.topo.l2c_id, cpu);
706 #ifdef CONFIG_CPU_SUP_AMD
707 extern u32 amd_get_nodes_per_socket(void);
708 extern u32 amd_get_highest_perf(void);
709 extern void amd_clear_divider(void);
710 extern void amd_check_microcode(void);
712 static inline u32 amd_get_nodes_per_socket(void) { return 0; }
713 static inline u32 amd_get_highest_perf(void) { return 0; }
714 static inline void amd_clear_divider(void) { }
715 static inline void amd_check_microcode(void) { }
718 extern unsigned long arch_align_stack(unsigned long sp);
719 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
720 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
722 void default_idle(void);
724 bool xen_set_default_idle(void);
726 #define xen_set_default_idle 0
729 void __noreturn stop_this_cpu(void *dummy);
730 void microcode_check(struct cpuinfo_x86 *prev_info);
731 void store_cpu_caps(struct cpuinfo_x86 *info);
733 enum l1tf_mitigations {
735 L1TF_MITIGATION_FLUSH_NOWARN,
736 L1TF_MITIGATION_FLUSH,
737 L1TF_MITIGATION_FLUSH_NOSMT,
738 L1TF_MITIGATION_FULL,
739 L1TF_MITIGATION_FULL_FORCE
742 extern enum l1tf_mitigations l1tf_mitigation;
744 enum mds_mitigations {
747 MDS_MITIGATION_VMWERV,
750 extern bool gds_ucode_mitigated(void);
753 * Make previous memory operations globally visible before
756 * MFENCE makes writes visible, but only affects load/store
757 * instructions. WRMSR is unfortunately not a load/store
758 * instruction and is unaffected by MFENCE. The LFENCE ensures
759 * that the WRMSR is not reordered.
761 * Most WRMSRs are full serializing instructions themselves and
762 * do not require this barrier. This is only required for the
763 * IA32_TSC_DEADLINE and X2APIC MSRs.
765 static inline void weak_wrmsr_fence(void)
767 alternative("mfence; lfence", "", ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE));
770 #endif /* _ASM_X86_PROCESSOR_H */