1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PROCESSOR_H
3 #define _ASM_X86_PROCESSOR_H
4
5 #include <asm/processor-flags.h>
6
7 /* Forward declaration, a strange C thing */
8 struct task_struct;
9 struct mm_struct;
10 struct vm86;
11
12 #include <asm/math_emu.h>
13 #include <asm/segment.h>
14 #include <asm/types.h>
15 #include <uapi/asm/sigcontext.h>
16 #include <asm/current.h>
17 #include <asm/cpufeatures.h>
18 #include <asm/page.h>
19 #include <asm/pgtable_types.h>
20 #include <asm/percpu.h>
21 #include <asm/msr.h>
22 #include <asm/desc_defs.h>
23 #include <asm/nops.h>
24 #include <asm/special_insns.h>
25 #include <asm/fpu/types.h>
26 #include <asm/unwind_hints.h>
27
28 #include <linux/personality.h>
29 #include <linux/cache.h>
30 #include <linux/threads.h>
31 #include <linux/math64.h>
32 #include <linux/err.h>
33 #include <linux/irqflags.h>
34 #include <linux/mem_encrypt.h>
35
36 /*
37 * We handle most unaligned accesses in hardware. On the other hand
38 * unaligned DMA can be quite expensive on some Nehalem processors.
39 *
40 * Based on this we disable the IP header alignment in network drivers.
41 */
42 #define NET_IP_ALIGN 0
43
44 #define HBP_NUM 4
45 /*
46 * Default implementation of macro that returns current
47 * instruction pointer ("program counter").
48 */
current_text_addr(void)49 static inline void *current_text_addr(void)
50 {
51 void *pc;
52
53 asm volatile("mov $1f, %0; 1:":"=r" (pc));
54
55 return pc;
56 }
57
58 /*
59 * These alignment constraints are for performance in the vSMP case,
60 * but in the task_struct case we must also meet hardware imposed
61 * alignment requirements of the FPU state:
62 */
63 #ifdef CONFIG_X86_VSMP
64 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
65 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
66 #else
67 # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
68 # define ARCH_MIN_MMSTRUCT_ALIGN 0
69 #endif
70
71 enum tlb_infos {
72 ENTRIES,
73 NR_INFO
74 };
75
76 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
77 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
78 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
79 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
80 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
81 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
82 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
83
84 /*
85 * CPU type and hardware bug flags. Kept separately for each CPU.
86 * Members of this structure are referenced in head_32.S, so think twice
87 * before touching them. [mj]
88 */
89
90 struct cpuinfo_x86 {
91 __u8 x86; /* CPU family */
92 __u8 x86_vendor; /* CPU vendor */
93 __u8 x86_model;
94 __u8 x86_stepping;
95 #ifdef CONFIG_X86_64
96 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
97 int x86_tlbsize;
98 #endif
99 __u8 x86_virt_bits;
100 __u8 x86_phys_bits;
101 /* CPUID returned core id bits: */
102 __u8 x86_coreid_bits;
103 __u8 cu_id;
104 /* Max extended CPUID function supported: */
105 __u32 extended_cpuid_level;
106 /* Maximum supported CPUID level, -1=no CPUID: */
107 int cpuid_level;
108 __u32 x86_capability[NCAPINTS + NBUGINTS];
109 char x86_vendor_id[16];
110 char x86_model_id[64];
111 /* in KB - valid for CPUS which support this call: */
112 unsigned int x86_cache_size;
113 int x86_cache_alignment; /* In bytes */
114 /* Cache QoS architectural values: */
115 int x86_cache_max_rmid; /* max index */
116 int x86_cache_occ_scale; /* scale to bytes */
117 int x86_power;
118 unsigned long loops_per_jiffy;
119 /* cpuid returned max cores value: */
120 u16 x86_max_cores;
121 u16 apicid;
122 u16 initial_apicid;
123 u16 x86_clflush_size;
124 /* number of cores as seen by the OS: */
125 u16 booted_cores;
126 /* Physical processor id: */
127 u16 phys_proc_id;
128 /* Logical processor id: */
129 u16 logical_proc_id;
130 /* Core id: */
131 u16 cpu_core_id;
132 /* Index into per_cpu list: */
133 u16 cpu_index;
134 u32 microcode;
135 /* Address space bits used by the cache internally */
136 u8 x86_cache_bits;
137 unsigned initialized : 1;
138 } __randomize_layout;
139
140 struct cpuid_regs {
141 u32 eax, ebx, ecx, edx;
142 };
143
144 enum cpuid_regs_idx {
145 CPUID_EAX = 0,
146 CPUID_EBX,
147 CPUID_ECX,
148 CPUID_EDX,
149 };
150
151 #define X86_VENDOR_INTEL 0
152 #define X86_VENDOR_CYRIX 1
153 #define X86_VENDOR_AMD 2
154 #define X86_VENDOR_UMC 3
155 #define X86_VENDOR_CENTAUR 5
156 #define X86_VENDOR_TRANSMETA 7
157 #define X86_VENDOR_NSC 8
158 #define X86_VENDOR_NUM 9
159
160 #define X86_VENDOR_UNKNOWN 0xff
161
162 /*
163 * capabilities of CPUs
164 */
165 extern struct cpuinfo_x86 boot_cpu_data;
166 extern struct cpuinfo_x86 new_cpu_data;
167
168 extern struct x86_hw_tss doublefault_tss;
169 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
170 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
171
172 #ifdef CONFIG_SMP
173 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
174 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
175 #else
176 #define cpu_info boot_cpu_data
177 #define cpu_data(cpu) boot_cpu_data
178 #endif
179
180 extern const struct seq_operations cpuinfo_op;
181
182 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
183
184 extern void cpu_detect(struct cpuinfo_x86 *c);
185
l1tf_pfn_limit(void)186 static inline unsigned long long l1tf_pfn_limit(void)
187 {
188 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
189 }
190
191 extern void early_cpu_init(void);
192 extern void identify_boot_cpu(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 *);
196
197 #ifdef CONFIG_X86_32
198 extern int have_cpuid_p(void);
199 #else
have_cpuid_p(void)200 static inline int have_cpuid_p(void)
201 {
202 return 1;
203 }
204 #endif
native_cpuid(unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)205 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
206 unsigned int *ecx, unsigned int *edx)
207 {
208 /* ecx is often an input as well as an output. */
209 asm volatile("cpuid"
210 : "=a" (*eax),
211 "=b" (*ebx),
212 "=c" (*ecx),
213 "=d" (*edx)
214 : "0" (*eax), "2" (*ecx)
215 : "memory");
216 }
217
218 #define native_cpuid_reg(reg) \
219 static inline unsigned int native_cpuid_##reg(unsigned int op) \
220 { \
221 unsigned int eax = op, ebx, ecx = 0, edx; \
222 \
223 native_cpuid(&eax, &ebx, &ecx, &edx); \
224 \
225 return reg; \
226 }
227
228 /*
229 * Native CPUID functions returning a single datum.
230 */
231 native_cpuid_reg(eax)
native_cpuid_reg(ebx)232 native_cpuid_reg(ebx)
233 native_cpuid_reg(ecx)
234 native_cpuid_reg(edx)
235
236 /*
237 * Friendlier CR3 helpers.
238 */
239 static inline unsigned long read_cr3_pa(void)
240 {
241 return __read_cr3() & CR3_ADDR_MASK;
242 }
243
native_read_cr3_pa(void)244 static inline unsigned long native_read_cr3_pa(void)
245 {
246 return __native_read_cr3() & CR3_ADDR_MASK;
247 }
248
load_cr3(pgd_t * pgdir)249 static inline void load_cr3(pgd_t *pgdir)
250 {
251 write_cr3(__sme_pa(pgdir));
252 }
253
254 /*
255 * Note that while the legacy 'TSS' name comes from 'Task State Segment',
256 * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
257 * unrelated to the task-switch mechanism:
258 */
259 #ifdef CONFIG_X86_32
260 /* This is the TSS defined by the hardware. */
261 struct x86_hw_tss {
262 unsigned short back_link, __blh;
263 unsigned long sp0;
264 unsigned short ss0, __ss0h;
265 unsigned long sp1;
266
267 /*
268 * We don't use ring 1, so ss1 is a convenient scratch space in
269 * the same cacheline as sp0. We use ss1 to cache the value in
270 * MSR_IA32_SYSENTER_CS. When we context switch
271 * MSR_IA32_SYSENTER_CS, we first check if the new value being
272 * written matches ss1, and, if it's not, then we wrmsr the new
273 * value and update ss1.
274 *
275 * The only reason we context switch MSR_IA32_SYSENTER_CS is
276 * that we set it to zero in vm86 tasks to avoid corrupting the
277 * stack if we were to go through the sysenter path from vm86
278 * mode.
279 */
280 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
281
282 unsigned short __ss1h;
283 unsigned long sp2;
284 unsigned short ss2, __ss2h;
285 unsigned long __cr3;
286 unsigned long ip;
287 unsigned long flags;
288 unsigned long ax;
289 unsigned long cx;
290 unsigned long dx;
291 unsigned long bx;
292 unsigned long sp;
293 unsigned long bp;
294 unsigned long si;
295 unsigned long di;
296 unsigned short es, __esh;
297 unsigned short cs, __csh;
298 unsigned short ss, __ssh;
299 unsigned short ds, __dsh;
300 unsigned short fs, __fsh;
301 unsigned short gs, __gsh;
302 unsigned short ldt, __ldth;
303 unsigned short trace;
304 unsigned short io_bitmap_base;
305
306 } __attribute__((packed));
307 #else
308 struct x86_hw_tss {
309 u32 reserved1;
310 u64 sp0;
311
312 /*
313 * We store cpu_current_top_of_stack in sp1 so it's always accessible.
314 * Linux does not use ring 1, so sp1 is not otherwise needed.
315 */
316 u64 sp1;
317
318 u64 sp2;
319 u64 reserved2;
320 u64 ist[7];
321 u32 reserved3;
322 u32 reserved4;
323 u16 reserved5;
324 u16 io_bitmap_base;
325
326 } __attribute__((packed));
327 #endif
328
329 /*
330 * IO-bitmap sizes:
331 */
332 #define IO_BITMAP_BITS 65536
333 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
334 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
335 #define IO_BITMAP_OFFSET (offsetof(struct tss_struct, io_bitmap) - offsetof(struct tss_struct, x86_tss))
336 #define INVALID_IO_BITMAP_OFFSET 0x8000
337
338 struct entry_stack {
339 unsigned long words[64];
340 };
341
342 struct entry_stack_page {
343 struct entry_stack stack;
344 } __aligned(PAGE_SIZE);
345
346 struct tss_struct {
347 /*
348 * The fixed hardware portion. This must not cross a page boundary
349 * at risk of violating the SDM's advice and potentially triggering
350 * errata.
351 */
352 struct x86_hw_tss x86_tss;
353
354 /*
355 * The extra 1 is there because the CPU will access an
356 * additional byte beyond the end of the IO permission
357 * bitmap. The extra byte must be all 1 bits, and must
358 * be within the limit.
359 */
360 unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
361 } __aligned(PAGE_SIZE);
362
363 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
364
365 /*
366 * sizeof(unsigned long) coming from an extra "long" at the end
367 * of the iobitmap.
368 *
369 * -1? seg base+limit should be pointing to the address of the
370 * last valid byte
371 */
372 #define __KERNEL_TSS_LIMIT \
373 (IO_BITMAP_OFFSET + IO_BITMAP_BYTES + sizeof(unsigned long) - 1)
374
375 #ifdef CONFIG_X86_32
376 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
377 #else
378 /* The RO copy can't be accessed with this_cpu_xyz(), so use the RW copy. */
379 #define cpu_current_top_of_stack cpu_tss_rw.x86_tss.sp1
380 #endif
381
382 /*
383 * Save the original ist values for checking stack pointers during debugging
384 */
385 struct orig_ist {
386 unsigned long ist[7];
387 };
388
389 #ifdef CONFIG_X86_64
390 DECLARE_PER_CPU(struct orig_ist, orig_ist);
391
392 union irq_stack_union {
393 char irq_stack[IRQ_STACK_SIZE];
394 /*
395 * GCC hardcodes the stack canary as %gs:40. Since the
396 * irq_stack is the object at %gs:0, we reserve the bottom
397 * 48 bytes of the irq stack for the canary.
398 */
399 struct {
400 char gs_base[40];
401 unsigned long stack_canary;
402 };
403 };
404
405 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
406 DECLARE_INIT_PER_CPU(irq_stack_union);
407
cpu_kernelmode_gs_base(int cpu)408 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
409 {
410 return (unsigned long)per_cpu(irq_stack_union.gs_base, cpu);
411 }
412
413 DECLARE_PER_CPU(char *, irq_stack_ptr);
414 DECLARE_PER_CPU(unsigned int, irq_count);
415 extern asmlinkage void ignore_sysret(void);
416
417 #if IS_ENABLED(CONFIG_KVM)
418 /* Save actual FS/GS selectors and bases to current->thread */
419 void save_fsgs_for_kvm(void);
420 #endif
421 #else /* X86_64 */
422 #ifdef CONFIG_STACKPROTECTOR
423 /*
424 * Make sure stack canary segment base is cached-aligned:
425 * "For Intel Atom processors, avoid non zero segment base address
426 * that is not aligned to cache line boundary at all cost."
427 * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
428 */
429 struct stack_canary {
430 char __pad[20]; /* canary at %gs:20 */
431 unsigned long canary;
432 };
433 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
434 #endif
435 /*
436 * per-CPU IRQ handling stacks
437 */
438 struct irq_stack {
439 u32 stack[THREAD_SIZE/sizeof(u32)];
440 } __aligned(THREAD_SIZE);
441
442 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
443 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
444 #endif /* X86_64 */
445
446 extern unsigned int fpu_kernel_xstate_size;
447 extern unsigned int fpu_user_xstate_size;
448
449 struct perf_event;
450
451 typedef struct {
452 unsigned long seg;
453 } mm_segment_t;
454
455 struct thread_struct {
456 /* Cached TLS descriptors: */
457 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
458 #ifdef CONFIG_X86_32
459 unsigned long sp0;
460 #endif
461 unsigned long sp;
462 #ifdef CONFIG_X86_32
463 unsigned long sysenter_cs;
464 #else
465 unsigned short es;
466 unsigned short ds;
467 unsigned short fsindex;
468 unsigned short gsindex;
469 #endif
470
471 #ifdef CONFIG_X86_64
472 unsigned long fsbase;
473 unsigned long gsbase;
474 #else
475 /*
476 * XXX: this could presumably be unsigned short. Alternatively,
477 * 32-bit kernels could be taught to use fsindex instead.
478 */
479 unsigned long fs;
480 unsigned long gs;
481 #endif
482
483 /* Save middle states of ptrace breakpoints */
484 struct perf_event *ptrace_bps[HBP_NUM];
485 /* Debug status used for traps, single steps, etc... */
486 unsigned long debugreg6;
487 /* Keep track of the exact dr7 value set by the user */
488 unsigned long ptrace_dr7;
489 /* Fault info: */
490 unsigned long cr2;
491 unsigned long trap_nr;
492 unsigned long error_code;
493 #ifdef CONFIG_VM86
494 /* Virtual 86 mode info */
495 struct vm86 *vm86;
496 #endif
497 /* IO permissions: */
498 unsigned long *io_bitmap_ptr;
499 unsigned long iopl;
500 /* Max allowed port in the bitmap, in bytes: */
501 unsigned io_bitmap_max;
502
503 mm_segment_t addr_limit;
504
505 unsigned int sig_on_uaccess_err:1;
506 unsigned int uaccess_err:1; /* uaccess failed */
507
508 /* Floating point and extended processor state */
509 struct fpu fpu;
510 /*
511 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
512 * the end.
513 */
514 };
515
516 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
arch_thread_struct_whitelist(unsigned long * offset,unsigned long * size)517 static inline void arch_thread_struct_whitelist(unsigned long *offset,
518 unsigned long *size)
519 {
520 *offset = offsetof(struct thread_struct, fpu.state);
521 *size = fpu_kernel_xstate_size;
522 }
523
524 /*
525 * Thread-synchronous status.
526 *
527 * This is different from the flags in that nobody else
528 * ever touches our thread-synchronous status, so we don't
529 * have to worry about atomic accesses.
530 */
531 #define TS_COMPAT 0x0002 /* 32bit syscall active (64BIT)*/
532
533 /*
534 * Set IOPL bits in EFLAGS from given mask
535 */
native_set_iopl_mask(unsigned mask)536 static inline void native_set_iopl_mask(unsigned mask)
537 {
538 #ifdef CONFIG_X86_32
539 unsigned int reg;
540
541 asm volatile ("pushfl;"
542 "popl %0;"
543 "andl %1, %0;"
544 "orl %2, %0;"
545 "pushl %0;"
546 "popfl"
547 : "=&r" (reg)
548 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
549 #endif
550 }
551
552 static inline void
native_load_sp0(unsigned long sp0)553 native_load_sp0(unsigned long sp0)
554 {
555 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
556 }
557
native_swapgs(void)558 static inline void native_swapgs(void)
559 {
560 #ifdef CONFIG_X86_64
561 asm volatile("swapgs" ::: "memory");
562 #endif
563 }
564
current_top_of_stack(void)565 static inline unsigned long current_top_of_stack(void)
566 {
567 /*
568 * We can't read directly from tss.sp0: sp0 on x86_32 is special in
569 * and around vm86 mode and sp0 on x86_64 is special because of the
570 * entry trampoline.
571 */
572 return this_cpu_read_stable(cpu_current_top_of_stack);
573 }
574
on_thread_stack(void)575 static inline bool on_thread_stack(void)
576 {
577 return (unsigned long)(current_top_of_stack() -
578 current_stack_pointer) < THREAD_SIZE;
579 }
580
581 #ifdef CONFIG_PARAVIRT
582 #include <asm/paravirt.h>
583 #else
584 #define __cpuid native_cpuid
585
load_sp0(unsigned long sp0)586 static inline void load_sp0(unsigned long sp0)
587 {
588 native_load_sp0(sp0);
589 }
590
591 #define set_iopl_mask native_set_iopl_mask
592 #endif /* CONFIG_PARAVIRT */
593
594 /* Free all resources held by a thread. */
595 extern void release_thread(struct task_struct *);
596
597 unsigned long get_wchan(struct task_struct *p);
598
599 /*
600 * Generic CPUID function
601 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
602 * resulting in stale register contents being returned.
603 */
cpuid(unsigned int op,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)604 static inline void cpuid(unsigned int op,
605 unsigned int *eax, unsigned int *ebx,
606 unsigned int *ecx, unsigned int *edx)
607 {
608 *eax = op;
609 *ecx = 0;
610 __cpuid(eax, ebx, ecx, edx);
611 }
612
613 /* Some CPUID calls want 'count' to be placed in ecx */
cpuid_count(unsigned int op,int count,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)614 static inline void cpuid_count(unsigned int op, int count,
615 unsigned int *eax, unsigned int *ebx,
616 unsigned int *ecx, unsigned int *edx)
617 {
618 *eax = op;
619 *ecx = count;
620 __cpuid(eax, ebx, ecx, edx);
621 }
622
623 /*
624 * CPUID functions returning a single datum
625 */
cpuid_eax(unsigned int op)626 static inline unsigned int cpuid_eax(unsigned int op)
627 {
628 unsigned int eax, ebx, ecx, edx;
629
630 cpuid(op, &eax, &ebx, &ecx, &edx);
631
632 return eax;
633 }
634
cpuid_ebx(unsigned int op)635 static inline unsigned int cpuid_ebx(unsigned int op)
636 {
637 unsigned int eax, ebx, ecx, edx;
638
639 cpuid(op, &eax, &ebx, &ecx, &edx);
640
641 return ebx;
642 }
643
cpuid_ecx(unsigned int op)644 static inline unsigned int cpuid_ecx(unsigned int op)
645 {
646 unsigned int eax, ebx, ecx, edx;
647
648 cpuid(op, &eax, &ebx, &ecx, &edx);
649
650 return ecx;
651 }
652
cpuid_edx(unsigned int op)653 static inline unsigned int cpuid_edx(unsigned int op)
654 {
655 unsigned int eax, ebx, ecx, edx;
656
657 cpuid(op, &eax, &ebx, &ecx, &edx);
658
659 return edx;
660 }
661
662 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
rep_nop(void)663 static __always_inline void rep_nop(void)
664 {
665 asm volatile("rep; nop" ::: "memory");
666 }
667
cpu_relax(void)668 static __always_inline void cpu_relax(void)
669 {
670 rep_nop();
671 }
672
673 /*
674 * This function forces the icache and prefetched instruction stream to
675 * catch up with reality in two very specific cases:
676 *
677 * a) Text was modified using one virtual address and is about to be executed
678 * from the same physical page at a different virtual address.
679 *
680 * b) Text was modified on a different CPU, may subsequently be
681 * executed on this CPU, and you want to make sure the new version
682 * gets executed. This generally means you're calling this in a IPI.
683 *
684 * If you're calling this for a different reason, you're probably doing
685 * it wrong.
686 */
sync_core(void)687 static inline void sync_core(void)
688 {
689 /*
690 * There are quite a few ways to do this. IRET-to-self is nice
691 * because it works on every CPU, at any CPL (so it's compatible
692 * with paravirtualization), and it never exits to a hypervisor.
693 * The only down sides are that it's a bit slow (it seems to be
694 * a bit more than 2x slower than the fastest options) and that
695 * it unmasks NMIs. The "push %cs" is needed because, in
696 * paravirtual environments, __KERNEL_CS may not be a valid CS
697 * value when we do IRET directly.
698 *
699 * In case NMI unmasking or performance ever becomes a problem,
700 * the next best option appears to be MOV-to-CR2 and an
701 * unconditional jump. That sequence also works on all CPUs,
702 * but it will fault at CPL3 (i.e. Xen PV).
703 *
704 * CPUID is the conventional way, but it's nasty: it doesn't
705 * exist on some 486-like CPUs, and it usually exits to a
706 * hypervisor.
707 *
708 * Like all of Linux's memory ordering operations, this is a
709 * compiler barrier as well.
710 */
711 #ifdef CONFIG_X86_32
712 asm volatile (
713 "pushfl\n\t"
714 "pushl %%cs\n\t"
715 "pushl $1f\n\t"
716 "iret\n\t"
717 "1:"
718 : ASM_CALL_CONSTRAINT : : "memory");
719 #else
720 unsigned int tmp;
721
722 asm volatile (
723 UNWIND_HINT_SAVE
724 "mov %%ss, %0\n\t"
725 "pushq %q0\n\t"
726 "pushq %%rsp\n\t"
727 "addq $8, (%%rsp)\n\t"
728 "pushfq\n\t"
729 "mov %%cs, %0\n\t"
730 "pushq %q0\n\t"
731 "pushq $1f\n\t"
732 "iretq\n\t"
733 UNWIND_HINT_RESTORE
734 "1:"
735 : "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
736 #endif
737 }
738
739 extern void select_idle_routine(const struct cpuinfo_x86 *c);
740 extern void amd_e400_c1e_apic_setup(void);
741
742 extern unsigned long boot_option_idle_override;
743
744 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
745 IDLE_POLL};
746
747 extern void enable_sep_cpu(void);
748 extern int sysenter_setup(void);
749
750 void early_trap_pf_init(void);
751
752 /* Defined in head.S */
753 extern struct desc_ptr early_gdt_descr;
754
755 extern void switch_to_new_gdt(int);
756 extern void load_direct_gdt(int);
757 extern void load_fixmap_gdt(int);
758 extern void load_percpu_segment(int);
759 extern void cpu_init(void);
760
get_debugctlmsr(void)761 static inline unsigned long get_debugctlmsr(void)
762 {
763 unsigned long debugctlmsr = 0;
764
765 #ifndef CONFIG_X86_DEBUGCTLMSR
766 if (boot_cpu_data.x86 < 6)
767 return 0;
768 #endif
769 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
770
771 return debugctlmsr;
772 }
773
update_debugctlmsr(unsigned long debugctlmsr)774 static inline void update_debugctlmsr(unsigned long debugctlmsr)
775 {
776 #ifndef CONFIG_X86_DEBUGCTLMSR
777 if (boot_cpu_data.x86 < 6)
778 return;
779 #endif
780 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
781 }
782
783 extern void set_task_blockstep(struct task_struct *task, bool on);
784
785 /* Boot loader type from the setup header: */
786 extern int bootloader_type;
787 extern int bootloader_version;
788
789 extern char ignore_fpu_irq;
790
791 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
792 #define ARCH_HAS_PREFETCHW
793 #define ARCH_HAS_SPINLOCK_PREFETCH
794
795 #ifdef CONFIG_X86_32
796 # define BASE_PREFETCH ""
797 # define ARCH_HAS_PREFETCH
798 #else
799 # define BASE_PREFETCH "prefetcht0 %P1"
800 #endif
801
802 /*
803 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
804 *
805 * It's not worth to care about 3dnow prefetches for the K6
806 * because they are microcoded there and very slow.
807 */
prefetch(const void * x)808 static inline void prefetch(const void *x)
809 {
810 alternative_input(BASE_PREFETCH, "prefetchnta %P1",
811 X86_FEATURE_XMM,
812 "m" (*(const char *)x));
813 }
814
815 /*
816 * 3dnow prefetch to get an exclusive cache line.
817 * Useful for spinlocks to avoid one state transition in the
818 * cache coherency protocol:
819 */
prefetchw(const void * x)820 static inline void prefetchw(const void *x)
821 {
822 alternative_input(BASE_PREFETCH, "prefetchw %P1",
823 X86_FEATURE_3DNOWPREFETCH,
824 "m" (*(const char *)x));
825 }
826
spin_lock_prefetch(const void * x)827 static inline void spin_lock_prefetch(const void *x)
828 {
829 prefetchw(x);
830 }
831
832 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
833 TOP_OF_KERNEL_STACK_PADDING)
834
835 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
836
837 #define task_pt_regs(task) \
838 ({ \
839 unsigned long __ptr = (unsigned long)task_stack_page(task); \
840 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
841 ((struct pt_regs *)__ptr) - 1; \
842 })
843
844 #ifdef CONFIG_X86_32
845 /*
846 * User space process size: 3GB (default).
847 */
848 #define IA32_PAGE_OFFSET PAGE_OFFSET
849 #define TASK_SIZE PAGE_OFFSET
850 #define TASK_SIZE_LOW TASK_SIZE
851 #define TASK_SIZE_MAX TASK_SIZE
852 #define DEFAULT_MAP_WINDOW TASK_SIZE
853 #define STACK_TOP TASK_SIZE
854 #define STACK_TOP_MAX STACK_TOP
855
856 #define INIT_THREAD { \
857 .sp0 = TOP_OF_INIT_STACK, \
858 .sysenter_cs = __KERNEL_CS, \
859 .io_bitmap_ptr = NULL, \
860 .addr_limit = KERNEL_DS, \
861 }
862
863 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
864
865 #else
866 /*
867 * User space process size. This is the first address outside the user range.
868 * There are a few constraints that determine this:
869 *
870 * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
871 * address, then that syscall will enter the kernel with a
872 * non-canonical return address, and SYSRET will explode dangerously.
873 * We avoid this particular problem by preventing anything executable
874 * from being mapped at the maximum canonical address.
875 *
876 * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
877 * CPUs malfunction if they execute code from the highest canonical page.
878 * They'll speculate right off the end of the canonical space, and
879 * bad things happen. This is worked around in the same way as the
880 * Intel problem.
881 *
882 * With page table isolation enabled, we map the LDT in ... [stay tuned]
883 */
884 #define TASK_SIZE_MAX ((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
885
886 #define DEFAULT_MAP_WINDOW ((1UL << 47) - PAGE_SIZE)
887
888 /* This decides where the kernel will search for a free chunk of vm
889 * space during mmap's.
890 */
891 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
892 0xc0000000 : 0xFFFFe000)
893
894 #define TASK_SIZE_LOW (test_thread_flag(TIF_ADDR32) ? \
895 IA32_PAGE_OFFSET : DEFAULT_MAP_WINDOW)
896 #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \
897 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
898 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
899 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
900
901 #define STACK_TOP TASK_SIZE_LOW
902 #define STACK_TOP_MAX TASK_SIZE_MAX
903
904 #define INIT_THREAD { \
905 .addr_limit = KERNEL_DS, \
906 }
907
908 extern unsigned long KSTK_ESP(struct task_struct *task);
909
910 #endif /* CONFIG_X86_64 */
911
912 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
913 unsigned long new_sp);
914
915 /*
916 * This decides where the kernel will search for a free chunk of vm
917 * space during mmap's.
918 */
919 #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3))
920 #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
921
922 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
923
924 /* Get/set a process' ability to use the timestamp counter instruction */
925 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
926 #define SET_TSC_CTL(val) set_tsc_mode((val))
927
928 extern int get_tsc_mode(unsigned long adr);
929 extern int set_tsc_mode(unsigned int val);
930
931 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
932
933 /* Register/unregister a process' MPX related resource */
934 #define MPX_ENABLE_MANAGEMENT() mpx_enable_management()
935 #define MPX_DISABLE_MANAGEMENT() mpx_disable_management()
936
937 #ifdef CONFIG_X86_INTEL_MPX
938 extern int mpx_enable_management(void);
939 extern int mpx_disable_management(void);
940 #else
mpx_enable_management(void)941 static inline int mpx_enable_management(void)
942 {
943 return -EINVAL;
944 }
mpx_disable_management(void)945 static inline int mpx_disable_management(void)
946 {
947 return -EINVAL;
948 }
949 #endif /* CONFIG_X86_INTEL_MPX */
950
951 #ifdef CONFIG_CPU_SUP_AMD
952 extern u16 amd_get_nb_id(int cpu);
953 extern u32 amd_get_nodes_per_socket(void);
954 #else
amd_get_nb_id(int cpu)955 static inline u16 amd_get_nb_id(int cpu) { return 0; }
amd_get_nodes_per_socket(void)956 static inline u32 amd_get_nodes_per_socket(void) { return 0; }
957 #endif
958
hypervisor_cpuid_base(const char * sig,uint32_t leaves)959 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
960 {
961 uint32_t base, eax, signature[3];
962
963 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
964 cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
965
966 if (!memcmp(sig, signature, 12) &&
967 (leaves == 0 || ((eax - base) >= leaves)))
968 return base;
969 }
970
971 return 0;
972 }
973
974 extern unsigned long arch_align_stack(unsigned long sp);
975 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
976 extern void free_kernel_image_pages(void *begin, void *end);
977
978 void default_idle(void);
979 #ifdef CONFIG_XEN
980 bool xen_set_default_idle(void);
981 #else
982 #define xen_set_default_idle 0
983 #endif
984
985 void stop_this_cpu(void *dummy);
986 void df_debug(struct pt_regs *regs, long error_code);
987 void microcode_check(void);
988
989 enum l1tf_mitigations {
990 L1TF_MITIGATION_OFF,
991 L1TF_MITIGATION_FLUSH_NOWARN,
992 L1TF_MITIGATION_FLUSH,
993 L1TF_MITIGATION_FLUSH_NOSMT,
994 L1TF_MITIGATION_FULL,
995 L1TF_MITIGATION_FULL_FORCE
996 };
997
998 extern enum l1tf_mitigations l1tf_mitigation;
999
1000 #endif /* _ASM_X86_PROCESSOR_H */
1001