Commit | Line | Data |
---|---|---|
1965aae3 PA |
1 | #ifndef _ASM_X86_PROCESSOR_H |
2 | #define _ASM_X86_PROCESSOR_H | |
c758ecf6 | 3 | |
053de044 GOC |
4 | #include <asm/processor-flags.h> |
5 | ||
683e0253 GOC |
6 | /* Forward declaration, a strange C thing */ |
7 | struct task_struct; | |
8 | struct mm_struct; | |
9fda6a06 | 9 | struct vm86; |
683e0253 | 10 | |
2f66dcc9 GOC |
11 | #include <asm/math_emu.h> |
12 | #include <asm/segment.h> | |
2f66dcc9 | 13 | #include <asm/types.h> |
decb4c41 | 14 | #include <uapi/asm/sigcontext.h> |
2f66dcc9 | 15 | #include <asm/current.h> |
cd4d09ec | 16 | #include <asm/cpufeatures.h> |
2f66dcc9 | 17 | #include <asm/page.h> |
54321d94 | 18 | #include <asm/pgtable_types.h> |
5300db88 | 19 | #include <asm/percpu.h> |
2f66dcc9 GOC |
20 | #include <asm/msr.h> |
21 | #include <asm/desc_defs.h> | |
bd61643e | 22 | #include <asm/nops.h> |
f05e798a | 23 | #include <asm/special_insns.h> |
14b9675a | 24 | #include <asm/fpu/types.h> |
4d46a89e | 25 | |
2f66dcc9 | 26 | #include <linux/personality.h> |
5300db88 | 27 | #include <linux/cache.h> |
2f66dcc9 | 28 | #include <linux/threads.h> |
5cbc19a9 | 29 | #include <linux/math64.h> |
faa4602e | 30 | #include <linux/err.h> |
f05e798a | 31 | #include <linux/irqflags.h> |
21729f81 | 32 | #include <linux/mem_encrypt.h> |
f05e798a DH |
33 | |
34 | /* | |
35 | * We handle most unaligned accesses in hardware. On the other hand | |
36 | * unaligned DMA can be quite expensive on some Nehalem processors. | |
37 | * | |
38 | * Based on this we disable the IP header alignment in network drivers. | |
39 | */ | |
40 | #define NET_IP_ALIGN 0 | |
c72dcf83 | 41 | |
b332828c | 42 | #define HBP_NUM 4 |
0ccb8acc GOC |
43 | /* |
44 | * Default implementation of macro that returns current | |
45 | * instruction pointer ("program counter"). | |
46 | */ | |
47 | static inline void *current_text_addr(void) | |
48 | { | |
49 | void *pc; | |
4d46a89e IM |
50 | |
51 | asm volatile("mov $1f, %0; 1:":"=r" (pc)); | |
52 | ||
0ccb8acc GOC |
53 | return pc; |
54 | } | |
55 | ||
b8c1b8ea IM |
56 | /* |
57 | * These alignment constraints are for performance in the vSMP case, | |
58 | * but in the task_struct case we must also meet hardware imposed | |
59 | * alignment requirements of the FPU state: | |
60 | */ | |
dbcb4660 | 61 | #ifdef CONFIG_X86_VSMP |
4d46a89e IM |
62 | # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT) |
63 | # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT) | |
dbcb4660 | 64 | #else |
b8c1b8ea | 65 | # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state) |
4d46a89e | 66 | # define ARCH_MIN_MMSTRUCT_ALIGN 0 |
dbcb4660 GOC |
67 | #endif |
68 | ||
e0ba94f1 AS |
69 | enum tlb_infos { |
70 | ENTRIES, | |
71 | NR_INFO | |
72 | }; | |
73 | ||
74 | extern u16 __read_mostly tlb_lli_4k[NR_INFO]; | |
75 | extern u16 __read_mostly tlb_lli_2m[NR_INFO]; | |
76 | extern u16 __read_mostly tlb_lli_4m[NR_INFO]; | |
77 | extern u16 __read_mostly tlb_lld_4k[NR_INFO]; | |
78 | extern u16 __read_mostly tlb_lld_2m[NR_INFO]; | |
79 | extern u16 __read_mostly tlb_lld_4m[NR_INFO]; | |
dd360393 | 80 | extern u16 __read_mostly tlb_lld_1g[NR_INFO]; |
c4211f42 | 81 | |
5300db88 GOC |
82 | /* |
83 | * CPU type and hardware bug flags. Kept separately for each CPU. | |
04402116 | 84 | * Members of this structure are referenced in head_32.S, so think twice |
5300db88 GOC |
85 | * before touching them. [mj] |
86 | */ | |
87 | ||
88 | struct cpuinfo_x86 { | |
4d46a89e IM |
89 | __u8 x86; /* CPU family */ |
90 | __u8 x86_vendor; /* CPU vendor */ | |
91 | __u8 x86_model; | |
92 | __u8 x86_mask; | |
6415813b | 93 | #ifdef CONFIG_X86_64 |
4d46a89e | 94 | /* Number of 4K pages in DTLB/ITLB combined(in pages): */ |
b1882e68 | 95 | int x86_tlbsize; |
13c6c532 | 96 | #endif |
4d46a89e IM |
97 | __u8 x86_virt_bits; |
98 | __u8 x86_phys_bits; | |
99 | /* CPUID returned core id bits: */ | |
100 | __u8 x86_coreid_bits; | |
79a8b9aa | 101 | __u8 cu_id; |
4d46a89e IM |
102 | /* Max extended CPUID function supported: */ |
103 | __u32 extended_cpuid_level; | |
4d46a89e IM |
104 | /* Maximum supported CPUID level, -1=no CPUID: */ |
105 | int cpuid_level; | |
65fc985b | 106 | __u32 x86_capability[NCAPINTS + NBUGINTS]; |
4d46a89e IM |
107 | char x86_vendor_id[16]; |
108 | char x86_model_id[64]; | |
109 | /* in KB - valid for CPUS which support this call: */ | |
110 | int x86_cache_size; | |
111 | int x86_cache_alignment; /* In bytes */ | |
cbc82b17 PWJ |
112 | /* Cache QoS architectural values: */ |
113 | int x86_cache_max_rmid; /* max index */ | |
114 | int x86_cache_occ_scale; /* scale to bytes */ | |
4d46a89e IM |
115 | int x86_power; |
116 | unsigned long loops_per_jiffy; | |
4d46a89e IM |
117 | /* cpuid returned max cores value: */ |
118 | u16 x86_max_cores; | |
119 | u16 apicid; | |
01aaea1a | 120 | u16 initial_apicid; |
4d46a89e | 121 | u16 x86_clflush_size; |
4d46a89e IM |
122 | /* number of cores as seen by the OS: */ |
123 | u16 booted_cores; | |
124 | /* Physical processor id: */ | |
125 | u16 phys_proc_id; | |
1f12e32f TG |
126 | /* Logical processor id: */ |
127 | u16 logical_proc_id; | |
4d46a89e IM |
128 | /* Core id: */ |
129 | u16 cpu_core_id; | |
130 | /* Index into per_cpu list: */ | |
131 | u16 cpu_index; | |
506ed6b5 | 132 | u32 microcode; |
2c773dd3 | 133 | }; |
5300db88 | 134 | |
47f10a36 HC |
135 | struct cpuid_regs { |
136 | u32 eax, ebx, ecx, edx; | |
137 | }; | |
138 | ||
139 | enum cpuid_regs_idx { | |
140 | CPUID_EAX = 0, | |
141 | CPUID_EBX, | |
142 | CPUID_ECX, | |
143 | CPUID_EDX, | |
144 | }; | |
145 | ||
4d46a89e IM |
146 | #define X86_VENDOR_INTEL 0 |
147 | #define X86_VENDOR_CYRIX 1 | |
148 | #define X86_VENDOR_AMD 2 | |
149 | #define X86_VENDOR_UMC 3 | |
4d46a89e IM |
150 | #define X86_VENDOR_CENTAUR 5 |
151 | #define X86_VENDOR_TRANSMETA 7 | |
152 | #define X86_VENDOR_NSC 8 | |
153 | #define X86_VENDOR_NUM 9 | |
154 | ||
155 | #define X86_VENDOR_UNKNOWN 0xff | |
5300db88 | 156 | |
1a53905a GOC |
157 | /* |
158 | * capabilities of CPUs | |
159 | */ | |
4d46a89e IM |
160 | extern struct cpuinfo_x86 boot_cpu_data; |
161 | extern struct cpuinfo_x86 new_cpu_data; | |
162 | ||
163 | extern struct tss_struct doublefault_tss; | |
3e0c3737 YL |
164 | extern __u32 cpu_caps_cleared[NCAPINTS]; |
165 | extern __u32 cpu_caps_set[NCAPINTS]; | |
5300db88 GOC |
166 | |
167 | #ifdef CONFIG_SMP | |
2c773dd3 | 168 | DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); |
5300db88 | 169 | #define cpu_data(cpu) per_cpu(cpu_info, cpu) |
5300db88 | 170 | #else |
7b543a53 | 171 | #define cpu_info boot_cpu_data |
5300db88 | 172 | #define cpu_data(cpu) boot_cpu_data |
5300db88 GOC |
173 | #endif |
174 | ||
1c6c727d JS |
175 | extern const struct seq_operations cpuinfo_op; |
176 | ||
4d46a89e IM |
177 | #define cache_line_size() (boot_cpu_data.x86_cache_alignment) |
178 | ||
179 | extern void cpu_detect(struct cpuinfo_x86 *c); | |
1a53905a | 180 | |
f580366f | 181 | extern void early_cpu_init(void); |
1a53905a GOC |
182 | extern void identify_boot_cpu(void); |
183 | extern void identify_secondary_cpu(struct cpuinfo_x86 *); | |
5300db88 | 184 | extern void print_cpu_info(struct cpuinfo_x86 *); |
21c3fcf3 | 185 | void print_cpu_msr(struct cpuinfo_x86 *); |
5300db88 | 186 | extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c); |
47bdf337 HC |
187 | extern u32 get_scattered_cpuid_leaf(unsigned int level, |
188 | unsigned int sub_leaf, | |
189 | enum cpuid_regs_idx reg); | |
5300db88 | 190 | extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); |
04a15418 | 191 | extern void init_amd_cacheinfo(struct cpuinfo_x86 *c); |
5300db88 | 192 | |
bbb65d2d | 193 | extern void detect_extended_topology(struct cpuinfo_x86 *c); |
1a53905a | 194 | extern void detect_ht(struct cpuinfo_x86 *c); |
1a53905a | 195 | |
d288e1cf FY |
196 | #ifdef CONFIG_X86_32 |
197 | extern int have_cpuid_p(void); | |
198 | #else | |
199 | static inline int have_cpuid_p(void) | |
200 | { | |
201 | return 1; | |
202 | } | |
203 | #endif | |
c758ecf6 | 204 | static inline void native_cpuid(unsigned int *eax, unsigned int *ebx, |
4d46a89e | 205 | unsigned int *ecx, unsigned int *edx) |
c758ecf6 GOC |
206 | { |
207 | /* ecx is often an input as well as an output. */ | |
45a94d7c | 208 | asm volatile("cpuid" |
cca2e6f8 JP |
209 | : "=a" (*eax), |
210 | "=b" (*ebx), | |
211 | "=c" (*ecx), | |
212 | "=d" (*edx) | |
506ed6b5 AK |
213 | : "0" (*eax), "2" (*ecx) |
214 | : "memory"); | |
c758ecf6 GOC |
215 | } |
216 | ||
5dedade6 BP |
217 | #define native_cpuid_reg(reg) \ |
218 | static inline unsigned int native_cpuid_##reg(unsigned int op) \ | |
219 | { \ | |
220 | unsigned int eax = op, ebx, ecx = 0, edx; \ | |
221 | \ | |
222 | native_cpuid(&eax, &ebx, &ecx, &edx); \ | |
223 | \ | |
224 | return reg; \ | |
225 | } | |
226 | ||
227 | /* | |
228 | * Native CPUID functions returning a single datum. | |
229 | */ | |
230 | native_cpuid_reg(eax) | |
231 | native_cpuid_reg(ebx) | |
232 | native_cpuid_reg(ecx) | |
233 | native_cpuid_reg(edx) | |
234 | ||
6c690ee1 AL |
235 | /* |
236 | * Friendlier CR3 helpers. | |
237 | */ | |
238 | static inline unsigned long read_cr3_pa(void) | |
239 | { | |
240 | return __read_cr3() & CR3_ADDR_MASK; | |
241 | } | |
242 | ||
c72dcf83 GOC |
243 | static inline void load_cr3(pgd_t *pgdir) |
244 | { | |
21729f81 | 245 | write_cr3(__sme_pa(pgdir)); |
c72dcf83 | 246 | } |
c758ecf6 | 247 | |
ca241c75 GOC |
248 | #ifdef CONFIG_X86_32 |
249 | /* This is the TSS defined by the hardware. */ | |
250 | struct x86_hw_tss { | |
4d46a89e IM |
251 | unsigned short back_link, __blh; |
252 | unsigned long sp0; | |
253 | unsigned short ss0, __ss0h; | |
cf9328cc | 254 | unsigned long sp1; |
76e4c490 AL |
255 | |
256 | /* | |
cf9328cc AL |
257 | * We don't use ring 1, so ss1 is a convenient scratch space in |
258 | * the same cacheline as sp0. We use ss1 to cache the value in | |
259 | * MSR_IA32_SYSENTER_CS. When we context switch | |
260 | * MSR_IA32_SYSENTER_CS, we first check if the new value being | |
261 | * written matches ss1, and, if it's not, then we wrmsr the new | |
262 | * value and update ss1. | |
76e4c490 | 263 | * |
cf9328cc AL |
264 | * The only reason we context switch MSR_IA32_SYSENTER_CS is |
265 | * that we set it to zero in vm86 tasks to avoid corrupting the | |
266 | * stack if we were to go through the sysenter path from vm86 | |
267 | * mode. | |
76e4c490 | 268 | */ |
76e4c490 AL |
269 | unsigned short ss1; /* MSR_IA32_SYSENTER_CS */ |
270 | ||
271 | unsigned short __ss1h; | |
4d46a89e IM |
272 | unsigned long sp2; |
273 | unsigned short ss2, __ss2h; | |
274 | unsigned long __cr3; | |
275 | unsigned long ip; | |
276 | unsigned long flags; | |
277 | unsigned long ax; | |
278 | unsigned long cx; | |
279 | unsigned long dx; | |
280 | unsigned long bx; | |
281 | unsigned long sp; | |
282 | unsigned long bp; | |
283 | unsigned long si; | |
284 | unsigned long di; | |
285 | unsigned short es, __esh; | |
286 | unsigned short cs, __csh; | |
287 | unsigned short ss, __ssh; | |
288 | unsigned short ds, __dsh; | |
289 | unsigned short fs, __fsh; | |
290 | unsigned short gs, __gsh; | |
291 | unsigned short ldt, __ldth; | |
292 | unsigned short trace; | |
293 | unsigned short io_bitmap_base; | |
294 | ||
ca241c75 GOC |
295 | } __attribute__((packed)); |
296 | #else | |
297 | struct x86_hw_tss { | |
4d46a89e IM |
298 | u32 reserved1; |
299 | u64 sp0; | |
300 | u64 sp1; | |
301 | u64 sp2; | |
302 | u64 reserved2; | |
303 | u64 ist[7]; | |
304 | u32 reserved3; | |
305 | u32 reserved4; | |
306 | u16 reserved5; | |
307 | u16 io_bitmap_base; | |
308 | ||
d3273dea | 309 | } __attribute__((packed)); |
ca241c75 GOC |
310 | #endif |
311 | ||
312 | /* | |
4d46a89e | 313 | * IO-bitmap sizes: |
ca241c75 | 314 | */ |
4d46a89e IM |
315 | #define IO_BITMAP_BITS 65536 |
316 | #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8) | |
317 | #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long)) | |
318 | #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap) | |
319 | #define INVALID_IO_BITMAP_OFFSET 0x8000 | |
ca241c75 GOC |
320 | |
321 | struct tss_struct { | |
4d46a89e IM |
322 | /* |
323 | * The hardware state: | |
324 | */ | |
325 | struct x86_hw_tss x86_tss; | |
ca241c75 GOC |
326 | |
327 | /* | |
328 | * The extra 1 is there because the CPU will access an | |
329 | * additional byte beyond the end of the IO permission | |
330 | * bitmap. The extra byte must be all 1 bits, and must | |
331 | * be within the limit. | |
332 | */ | |
4d46a89e | 333 | unsigned long io_bitmap[IO_BITMAP_LONGS + 1]; |
4d46a89e | 334 | |
6dcc9414 | 335 | #ifdef CONFIG_X86_32 |
ca241c75 | 336 | /* |
2a41aa4f | 337 | * Space for the temporary SYSENTER stack. |
ca241c75 | 338 | */ |
2a41aa4f | 339 | unsigned long SYSENTER_stack_canary; |
d828c71f | 340 | unsigned long SYSENTER_stack[64]; |
6dcc9414 | 341 | #endif |
4d46a89e | 342 | |
84e65b0a | 343 | } ____cacheline_aligned; |
ca241c75 | 344 | |
24933b82 | 345 | DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss); |
ca241c75 | 346 | |
4f53ab14 AL |
347 | /* |
348 | * sizeof(unsigned long) coming from an extra "long" at the end | |
349 | * of the iobitmap. | |
350 | * | |
351 | * -1? seg base+limit should be pointing to the address of the | |
352 | * last valid byte | |
353 | */ | |
354 | #define __KERNEL_TSS_LIMIT \ | |
355 | (IO_BITMAP_OFFSET + IO_BITMAP_BYTES + sizeof(unsigned long) - 1) | |
356 | ||
a7fcf28d AL |
357 | #ifdef CONFIG_X86_32 |
358 | DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack); | |
359 | #endif | |
360 | ||
4d46a89e IM |
361 | /* |
362 | * Save the original ist values for checking stack pointers during debugging | |
363 | */ | |
1a53905a | 364 | struct orig_ist { |
4d46a89e | 365 | unsigned long ist[7]; |
1a53905a GOC |
366 | }; |
367 | ||
fe676203 | 368 | #ifdef CONFIG_X86_64 |
2f66dcc9 | 369 | DECLARE_PER_CPU(struct orig_ist, orig_ist); |
26f80bd6 | 370 | |
947e76cd BG |
371 | union irq_stack_union { |
372 | char irq_stack[IRQ_STACK_SIZE]; | |
373 | /* | |
374 | * GCC hardcodes the stack canary as %gs:40. Since the | |
375 | * irq_stack is the object at %gs:0, we reserve the bottom | |
376 | * 48 bytes of the irq stack for the canary. | |
377 | */ | |
378 | struct { | |
379 | char gs_base[40]; | |
380 | unsigned long stack_canary; | |
381 | }; | |
382 | }; | |
383 | ||
277d5b40 | 384 | DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible; |
2add8e23 BG |
385 | DECLARE_INIT_PER_CPU(irq_stack_union); |
386 | ||
26f80bd6 | 387 | DECLARE_PER_CPU(char *, irq_stack_ptr); |
9766cdbc | 388 | DECLARE_PER_CPU(unsigned int, irq_count); |
9766cdbc | 389 | extern asmlinkage void ignore_sysret(void); |
60a5317f TH |
390 | #else /* X86_64 */ |
391 | #ifdef CONFIG_CC_STACKPROTECTOR | |
1ea0d14e JF |
392 | /* |
393 | * Make sure stack canary segment base is cached-aligned: | |
394 | * "For Intel Atom processors, avoid non zero segment base address | |
395 | * that is not aligned to cache line boundary at all cost." | |
396 | * (Optim Ref Manual Assembly/Compiler Coding Rule 15.) | |
397 | */ | |
398 | struct stack_canary { | |
399 | char __pad[20]; /* canary at %gs:20 */ | |
400 | unsigned long canary; | |
401 | }; | |
53f82452 | 402 | DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); |
96a388de | 403 | #endif |
198d208d SR |
404 | /* |
405 | * per-CPU IRQ handling stacks | |
406 | */ | |
407 | struct irq_stack { | |
408 | u32 stack[THREAD_SIZE/sizeof(u32)]; | |
409 | } __aligned(THREAD_SIZE); | |
410 | ||
411 | DECLARE_PER_CPU(struct irq_stack *, hardirq_stack); | |
412 | DECLARE_PER_CPU(struct irq_stack *, softirq_stack); | |
60a5317f | 413 | #endif /* X86_64 */ |
c758ecf6 | 414 | |
bf15a8cf | 415 | extern unsigned int fpu_kernel_xstate_size; |
a1141e0b | 416 | extern unsigned int fpu_user_xstate_size; |
683e0253 | 417 | |
24f1e32c FW |
418 | struct perf_event; |
419 | ||
13d4ea09 AL |
420 | typedef struct { |
421 | unsigned long seg; | |
422 | } mm_segment_t; | |
423 | ||
cb38d377 | 424 | struct thread_struct { |
4d46a89e IM |
425 | /* Cached TLS descriptors: */ |
426 | struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; | |
427 | unsigned long sp0; | |
428 | unsigned long sp; | |
cb38d377 | 429 | #ifdef CONFIG_X86_32 |
4d46a89e | 430 | unsigned long sysenter_cs; |
cb38d377 | 431 | #else |
4d46a89e IM |
432 | unsigned short es; |
433 | unsigned short ds; | |
434 | unsigned short fsindex; | |
435 | unsigned short gsindex; | |
cb38d377 | 436 | #endif |
b9d989c7 AL |
437 | |
438 | u32 status; /* thread synchronous flags */ | |
439 | ||
d756f4ad | 440 | #ifdef CONFIG_X86_64 |
296f781a AL |
441 | unsigned long fsbase; |
442 | unsigned long gsbase; | |
443 | #else | |
444 | /* | |
445 | * XXX: this could presumably be unsigned short. Alternatively, | |
446 | * 32-bit kernels could be taught to use fsindex instead. | |
447 | */ | |
448 | unsigned long fs; | |
449 | unsigned long gs; | |
d756f4ad | 450 | #endif |
c5bedc68 | 451 | |
24f1e32c FW |
452 | /* Save middle states of ptrace breakpoints */ |
453 | struct perf_event *ptrace_bps[HBP_NUM]; | |
454 | /* Debug status used for traps, single steps, etc... */ | |
455 | unsigned long debugreg6; | |
326264a0 FW |
456 | /* Keep track of the exact dr7 value set by the user */ |
457 | unsigned long ptrace_dr7; | |
4d46a89e IM |
458 | /* Fault info: */ |
459 | unsigned long cr2; | |
51e7dc70 | 460 | unsigned long trap_nr; |
4d46a89e | 461 | unsigned long error_code; |
9fda6a06 | 462 | #ifdef CONFIG_VM86 |
4d46a89e | 463 | /* Virtual 86 mode info */ |
9fda6a06 | 464 | struct vm86 *vm86; |
cb38d377 | 465 | #endif |
4d46a89e IM |
466 | /* IO permissions: */ |
467 | unsigned long *io_bitmap_ptr; | |
468 | unsigned long iopl; | |
469 | /* Max allowed port in the bitmap, in bytes: */ | |
470 | unsigned io_bitmap_max; | |
0c8c0f03 | 471 | |
13d4ea09 AL |
472 | mm_segment_t addr_limit; |
473 | ||
2a53ccbc | 474 | unsigned int sig_on_uaccess_err:1; |
dfa9a942 AL |
475 | unsigned int uaccess_err:1; /* uaccess failed */ |
476 | ||
0c8c0f03 DH |
477 | /* Floating point and extended processor state */ |
478 | struct fpu fpu; | |
479 | /* | |
480 | * WARNING: 'fpu' is dynamically-sized. It *MUST* be at | |
481 | * the end. | |
482 | */ | |
cb38d377 GOC |
483 | }; |
484 | ||
b9d989c7 AL |
485 | /* |
486 | * Thread-synchronous status. | |
487 | * | |
488 | * This is different from the flags in that nobody else | |
489 | * ever touches our thread-synchronous status, so we don't | |
490 | * have to worry about atomic accesses. | |
491 | */ | |
492 | #define TS_COMPAT 0x0002 /* 32bit syscall active (64BIT)*/ | |
493 | ||
62d7d7ed GOC |
494 | /* |
495 | * Set IOPL bits in EFLAGS from given mask | |
496 | */ | |
497 | static inline void native_set_iopl_mask(unsigned mask) | |
498 | { | |
499 | #ifdef CONFIG_X86_32 | |
500 | unsigned int reg; | |
4d46a89e | 501 | |
cca2e6f8 JP |
502 | asm volatile ("pushfl;" |
503 | "popl %0;" | |
504 | "andl %1, %0;" | |
505 | "orl %2, %0;" | |
506 | "pushl %0;" | |
507 | "popfl" | |
508 | : "=&r" (reg) | |
509 | : "i" (~X86_EFLAGS_IOPL), "r" (mask)); | |
62d7d7ed GOC |
510 | #endif |
511 | } | |
512 | ||
4d46a89e IM |
513 | static inline void |
514 | native_load_sp0(struct tss_struct *tss, struct thread_struct *thread) | |
7818a1e0 GOC |
515 | { |
516 | tss->x86_tss.sp0 = thread->sp0; | |
517 | #ifdef CONFIG_X86_32 | |
4d46a89e | 518 | /* Only happens when SEP is enabled, no need to test "SEP"arately: */ |
7818a1e0 GOC |
519 | if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
520 | tss->x86_tss.ss1 = thread->sysenter_cs; | |
521 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); | |
522 | } | |
523 | #endif | |
524 | } | |
1b46cbe0 | 525 | |
e801f864 GOC |
526 | static inline void native_swapgs(void) |
527 | { | |
528 | #ifdef CONFIG_X86_64 | |
529 | asm volatile("swapgs" ::: "memory"); | |
530 | #endif | |
531 | } | |
532 | ||
a7fcf28d | 533 | static inline unsigned long current_top_of_stack(void) |
8ef46a67 | 534 | { |
a7fcf28d | 535 | #ifdef CONFIG_X86_64 |
24933b82 | 536 | return this_cpu_read_stable(cpu_tss.x86_tss.sp0); |
a7fcf28d AL |
537 | #else |
538 | /* sp0 on x86_32 is special in and around vm86 mode. */ | |
539 | return this_cpu_read_stable(cpu_current_top_of_stack); | |
540 | #endif | |
8ef46a67 AL |
541 | } |
542 | ||
7818a1e0 GOC |
543 | #ifdef CONFIG_PARAVIRT |
544 | #include <asm/paravirt.h> | |
545 | #else | |
4d46a89e | 546 | #define __cpuid native_cpuid |
1b46cbe0 | 547 | |
cca2e6f8 JP |
548 | static inline void load_sp0(struct tss_struct *tss, |
549 | struct thread_struct *thread) | |
7818a1e0 GOC |
550 | { |
551 | native_load_sp0(tss, thread); | |
552 | } | |
553 | ||
62d7d7ed | 554 | #define set_iopl_mask native_set_iopl_mask |
1b46cbe0 GOC |
555 | #endif /* CONFIG_PARAVIRT */ |
556 | ||
683e0253 GOC |
557 | /* Free all resources held by a thread. */ |
558 | extern void release_thread(struct task_struct *); | |
559 | ||
683e0253 | 560 | unsigned long get_wchan(struct task_struct *p); |
c758ecf6 GOC |
561 | |
562 | /* | |
563 | * Generic CPUID function | |
564 | * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx | |
565 | * resulting in stale register contents being returned. | |
566 | */ | |
567 | static inline void cpuid(unsigned int op, | |
568 | unsigned int *eax, unsigned int *ebx, | |
569 | unsigned int *ecx, unsigned int *edx) | |
570 | { | |
571 | *eax = op; | |
572 | *ecx = 0; | |
573 | __cpuid(eax, ebx, ecx, edx); | |
574 | } | |
575 | ||
576 | /* Some CPUID calls want 'count' to be placed in ecx */ | |
577 | static inline void cpuid_count(unsigned int op, int count, | |
578 | unsigned int *eax, unsigned int *ebx, | |
579 | unsigned int *ecx, unsigned int *edx) | |
580 | { | |
581 | *eax = op; | |
582 | *ecx = count; | |
583 | __cpuid(eax, ebx, ecx, edx); | |
584 | } | |
585 | ||
586 | /* | |
587 | * CPUID functions returning a single datum | |
588 | */ | |
589 | static inline unsigned int cpuid_eax(unsigned int op) | |
590 | { | |
591 | unsigned int eax, ebx, ecx, edx; | |
592 | ||
593 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 594 | |
c758ecf6 GOC |
595 | return eax; |
596 | } | |
4d46a89e | 597 | |
c758ecf6 GOC |
598 | static inline unsigned int cpuid_ebx(unsigned int op) |
599 | { | |
600 | unsigned int eax, ebx, ecx, edx; | |
601 | ||
602 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 603 | |
c758ecf6 GOC |
604 | return ebx; |
605 | } | |
4d46a89e | 606 | |
c758ecf6 GOC |
607 | static inline unsigned int cpuid_ecx(unsigned int op) |
608 | { | |
609 | unsigned int eax, ebx, ecx, edx; | |
610 | ||
611 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 612 | |
c758ecf6 GOC |
613 | return ecx; |
614 | } | |
4d46a89e | 615 | |
c758ecf6 GOC |
616 | static inline unsigned int cpuid_edx(unsigned int op) |
617 | { | |
618 | unsigned int eax, ebx, ecx, edx; | |
619 | ||
620 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 621 | |
c758ecf6 GOC |
622 | return edx; |
623 | } | |
624 | ||
683e0253 | 625 | /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */ |
0b101e62 | 626 | static __always_inline void rep_nop(void) |
683e0253 | 627 | { |
cca2e6f8 | 628 | asm volatile("rep; nop" ::: "memory"); |
683e0253 GOC |
629 | } |
630 | ||
0b101e62 | 631 | static __always_inline void cpu_relax(void) |
4d46a89e IM |
632 | { |
633 | rep_nop(); | |
634 | } | |
635 | ||
c198b121 AL |
636 | /* |
637 | * This function forces the icache and prefetched instruction stream to | |
638 | * catch up with reality in two very specific cases: | |
639 | * | |
640 | * a) Text was modified using one virtual address and is about to be executed | |
641 | * from the same physical page at a different virtual address. | |
642 | * | |
643 | * b) Text was modified on a different CPU, may subsequently be | |
644 | * executed on this CPU, and you want to make sure the new version | |
645 | * gets executed. This generally means you're calling this in a IPI. | |
646 | * | |
647 | * If you're calling this for a different reason, you're probably doing | |
648 | * it wrong. | |
649 | */ | |
683e0253 GOC |
650 | static inline void sync_core(void) |
651 | { | |
45c39fb0 | 652 | /* |
c198b121 AL |
653 | * There are quite a few ways to do this. IRET-to-self is nice |
654 | * because it works on every CPU, at any CPL (so it's compatible | |
655 | * with paravirtualization), and it never exits to a hypervisor. | |
656 | * The only down sides are that it's a bit slow (it seems to be | |
657 | * a bit more than 2x slower than the fastest options) and that | |
658 | * it unmasks NMIs. The "push %cs" is needed because, in | |
659 | * paravirtual environments, __KERNEL_CS may not be a valid CS | |
660 | * value when we do IRET directly. | |
661 | * | |
662 | * In case NMI unmasking or performance ever becomes a problem, | |
663 | * the next best option appears to be MOV-to-CR2 and an | |
664 | * unconditional jump. That sequence also works on all CPUs, | |
665 | * but it will fault at CPL3 (i.e. Xen PV and lguest). | |
666 | * | |
667 | * CPUID is the conventional way, but it's nasty: it doesn't | |
668 | * exist on some 486-like CPUs, and it usually exits to a | |
669 | * hypervisor. | |
670 | * | |
671 | * Like all of Linux's memory ordering operations, this is a | |
672 | * compiler barrier as well. | |
45c39fb0 | 673 | */ |
c198b121 AL |
674 | register void *__sp asm(_ASM_SP); |
675 | ||
676 | #ifdef CONFIG_X86_32 | |
677 | asm volatile ( | |
678 | "pushfl\n\t" | |
679 | "pushl %%cs\n\t" | |
680 | "pushl $1f\n\t" | |
681 | "iret\n\t" | |
682 | "1:" | |
683 | : "+r" (__sp) : : "memory"); | |
45c39fb0 | 684 | #else |
c198b121 AL |
685 | unsigned int tmp; |
686 | ||
687 | asm volatile ( | |
688 | "mov %%ss, %0\n\t" | |
689 | "pushq %q0\n\t" | |
690 | "pushq %%rsp\n\t" | |
691 | "addq $8, (%%rsp)\n\t" | |
692 | "pushfq\n\t" | |
693 | "mov %%cs, %0\n\t" | |
694 | "pushq %q0\n\t" | |
695 | "pushq $1f\n\t" | |
696 | "iretq\n\t" | |
697 | "1:" | |
698 | : "=&r" (tmp), "+r" (__sp) : : "cc", "memory"); | |
5367b688 | 699 | #endif |
683e0253 GOC |
700 | } |
701 | ||
683e0253 | 702 | extern void select_idle_routine(const struct cpuinfo_x86 *c); |
07c94a38 | 703 | extern void amd_e400_c1e_apic_setup(void); |
683e0253 | 704 | |
4d46a89e | 705 | extern unsigned long boot_option_idle_override; |
683e0253 | 706 | |
d1896049 | 707 | enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT, |
69fb3676 | 708 | IDLE_POLL}; |
d1896049 | 709 | |
1a53905a GOC |
710 | extern void enable_sep_cpu(void); |
711 | extern int sysenter_setup(void); | |
712 | ||
29c84391 | 713 | extern void early_trap_init(void); |
8170e6be | 714 | void early_trap_pf_init(void); |
29c84391 | 715 | |
1a53905a | 716 | /* Defined in head.S */ |
4d46a89e | 717 | extern struct desc_ptr early_gdt_descr; |
1a53905a GOC |
718 | |
719 | extern void cpu_set_gdt(int); | |
552be871 | 720 | extern void switch_to_new_gdt(int); |
45fc8757 | 721 | extern void load_direct_gdt(int); |
69218e47 | 722 | extern void load_fixmap_gdt(int); |
11e3a840 | 723 | extern void load_percpu_segment(int); |
1a53905a | 724 | extern void cpu_init(void); |
1a53905a | 725 | |
c2724775 MM |
726 | static inline unsigned long get_debugctlmsr(void) |
727 | { | |
ea8e61b7 | 728 | unsigned long debugctlmsr = 0; |
c2724775 MM |
729 | |
730 | #ifndef CONFIG_X86_DEBUGCTLMSR | |
731 | if (boot_cpu_data.x86 < 6) | |
732 | return 0; | |
733 | #endif | |
734 | rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); | |
735 | ||
ea8e61b7 | 736 | return debugctlmsr; |
c2724775 MM |
737 | } |
738 | ||
5b0e5084 JB |
739 | static inline void update_debugctlmsr(unsigned long debugctlmsr) |
740 | { | |
741 | #ifndef CONFIG_X86_DEBUGCTLMSR | |
742 | if (boot_cpu_data.x86 < 6) | |
743 | return; | |
744 | #endif | |
745 | wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); | |
746 | } | |
747 | ||
9bd1190a ON |
748 | extern void set_task_blockstep(struct task_struct *task, bool on); |
749 | ||
4d46a89e IM |
750 | /* Boot loader type from the setup header: */ |
751 | extern int bootloader_type; | |
5031296c | 752 | extern int bootloader_version; |
1a53905a | 753 | |
4d46a89e | 754 | extern char ignore_fpu_irq; |
683e0253 GOC |
755 | |
756 | #define HAVE_ARCH_PICK_MMAP_LAYOUT 1 | |
757 | #define ARCH_HAS_PREFETCHW | |
758 | #define ARCH_HAS_SPINLOCK_PREFETCH | |
759 | ||
ae2e15eb | 760 | #ifdef CONFIG_X86_32 |
a930dc45 | 761 | # define BASE_PREFETCH "" |
4d46a89e | 762 | # define ARCH_HAS_PREFETCH |
ae2e15eb | 763 | #else |
a930dc45 | 764 | # define BASE_PREFETCH "prefetcht0 %P1" |
ae2e15eb GOC |
765 | #endif |
766 | ||
4d46a89e IM |
767 | /* |
768 | * Prefetch instructions for Pentium III (+) and AMD Athlon (+) | |
769 | * | |
770 | * It's not worth to care about 3dnow prefetches for the K6 | |
771 | * because they are microcoded there and very slow. | |
772 | */ | |
ae2e15eb GOC |
773 | static inline void prefetch(const void *x) |
774 | { | |
a930dc45 | 775 | alternative_input(BASE_PREFETCH, "prefetchnta %P1", |
ae2e15eb | 776 | X86_FEATURE_XMM, |
a930dc45 | 777 | "m" (*(const char *)x)); |
ae2e15eb GOC |
778 | } |
779 | ||
4d46a89e IM |
780 | /* |
781 | * 3dnow prefetch to get an exclusive cache line. | |
782 | * Useful for spinlocks to avoid one state transition in the | |
783 | * cache coherency protocol: | |
784 | */ | |
ae2e15eb GOC |
785 | static inline void prefetchw(const void *x) |
786 | { | |
a930dc45 BP |
787 | alternative_input(BASE_PREFETCH, "prefetchw %P1", |
788 | X86_FEATURE_3DNOWPREFETCH, | |
789 | "m" (*(const char *)x)); | |
ae2e15eb GOC |
790 | } |
791 | ||
4d46a89e IM |
792 | static inline void spin_lock_prefetch(const void *x) |
793 | { | |
794 | prefetchw(x); | |
795 | } | |
796 | ||
d9e05cc5 AL |
797 | #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \ |
798 | TOP_OF_KERNEL_STACK_PADDING) | |
799 | ||
2f66dcc9 GOC |
800 | #ifdef CONFIG_X86_32 |
801 | /* | |
802 | * User space process size: 3GB (default). | |
803 | */ | |
8f3e474f | 804 | #define IA32_PAGE_OFFSET PAGE_OFFSET |
4d46a89e | 805 | #define TASK_SIZE PAGE_OFFSET |
d9517346 | 806 | #define TASK_SIZE_MAX TASK_SIZE |
4d46a89e IM |
807 | #define STACK_TOP TASK_SIZE |
808 | #define STACK_TOP_MAX STACK_TOP | |
809 | ||
810 | #define INIT_THREAD { \ | |
d9e05cc5 | 811 | .sp0 = TOP_OF_INIT_STACK, \ |
4d46a89e IM |
812 | .sysenter_cs = __KERNEL_CS, \ |
813 | .io_bitmap_ptr = NULL, \ | |
13d4ea09 | 814 | .addr_limit = KERNEL_DS, \ |
2f66dcc9 GOC |
815 | } |
816 | ||
2f66dcc9 | 817 | /* |
5c39403e | 818 | * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack. |
2f66dcc9 | 819 | * This is necessary to guarantee that the entire "struct pt_regs" |
b595076a | 820 | * is accessible even if the CPU haven't stored the SS/ESP registers |
2f66dcc9 GOC |
821 | * on the stack (interrupt gate does not save these registers |
822 | * when switching to the same priv ring). | |
823 | * Therefore beware: accessing the ss/esp fields of the | |
824 | * "struct pt_regs" is possible, but they may contain the | |
825 | * completely wrong values. | |
826 | */ | |
5c39403e DV |
827 | #define task_pt_regs(task) \ |
828 | ({ \ | |
829 | unsigned long __ptr = (unsigned long)task_stack_page(task); \ | |
830 | __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \ | |
831 | ((struct pt_regs *)__ptr) - 1; \ | |
2f66dcc9 GOC |
832 | }) |
833 | ||
4d46a89e | 834 | #define KSTK_ESP(task) (task_pt_regs(task)->sp) |
2f66dcc9 GOC |
835 | |
836 | #else | |
837 | /* | |
07114f0f AL |
838 | * User space process size. 47bits minus one guard page. The guard |
839 | * page is necessary on Intel CPUs: if a SYSCALL instruction is at | |
840 | * the highest possible canonical userspace address, then that | |
841 | * syscall will enter the kernel with a non-canonical return | |
842 | * address, and SYSRET will explode dangerously. We avoid this | |
843 | * particular problem by preventing anything from being mapped | |
844 | * at the maximum canonical address. | |
2f66dcc9 | 845 | */ |
d9517346 | 846 | #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE) |
2f66dcc9 GOC |
847 | |
848 | /* This decides where the kernel will search for a free chunk of vm | |
849 | * space during mmap's. | |
850 | */ | |
4d46a89e IM |
851 | #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \ |
852 | 0xc0000000 : 0xFFFFe000) | |
2f66dcc9 | 853 | |
6bd33008 | 854 | #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \ |
d9517346 | 855 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
6bd33008 | 856 | #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \ |
d9517346 | 857 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
2f66dcc9 | 858 | |
922a70d3 | 859 | #define STACK_TOP TASK_SIZE |
d9517346 | 860 | #define STACK_TOP_MAX TASK_SIZE_MAX |
922a70d3 | 861 | |
13d4ea09 AL |
862 | #define INIT_THREAD { \ |
863 | .sp0 = TOP_OF_INIT_STACK, \ | |
864 | .addr_limit = KERNEL_DS, \ | |
2f66dcc9 GOC |
865 | } |
866 | ||
4d46a89e | 867 | #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1) |
89240ba0 | 868 | extern unsigned long KSTK_ESP(struct task_struct *task); |
d046ff8b | 869 | |
2f66dcc9 GOC |
870 | #endif /* CONFIG_X86_64 */ |
871 | ||
513ad84b IM |
872 | extern void start_thread(struct pt_regs *regs, unsigned long new_ip, |
873 | unsigned long new_sp); | |
874 | ||
4d46a89e IM |
875 | /* |
876 | * This decides where the kernel will search for a free chunk of vm | |
683e0253 GOC |
877 | * space during mmap's. |
878 | */ | |
8f3e474f DS |
879 | #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3)) |
880 | #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE) | |
683e0253 | 881 | |
4d46a89e | 882 | #define KSTK_EIP(task) (task_pt_regs(task)->ip) |
683e0253 | 883 | |
529e25f6 EB |
884 | /* Get/set a process' ability to use the timestamp counter instruction */ |
885 | #define GET_TSC_CTL(adr) get_tsc_mode((adr)) | |
886 | #define SET_TSC_CTL(val) set_tsc_mode((val)) | |
887 | ||
888 | extern int get_tsc_mode(unsigned long adr); | |
889 | extern int set_tsc_mode(unsigned int val); | |
890 | ||
e9ea1e7f KH |
891 | DECLARE_PER_CPU(u64, msr_misc_features_shadow); |
892 | ||
fe3d197f | 893 | /* Register/unregister a process' MPX related resource */ |
46a6e0cf DH |
894 | #define MPX_ENABLE_MANAGEMENT() mpx_enable_management() |
895 | #define MPX_DISABLE_MANAGEMENT() mpx_disable_management() | |
fe3d197f DH |
896 | |
897 | #ifdef CONFIG_X86_INTEL_MPX | |
46a6e0cf DH |
898 | extern int mpx_enable_management(void); |
899 | extern int mpx_disable_management(void); | |
fe3d197f | 900 | #else |
46a6e0cf | 901 | static inline int mpx_enable_management(void) |
fe3d197f DH |
902 | { |
903 | return -EINVAL; | |
904 | } | |
46a6e0cf | 905 | static inline int mpx_disable_management(void) |
fe3d197f DH |
906 | { |
907 | return -EINVAL; | |
908 | } | |
909 | #endif /* CONFIG_X86_INTEL_MPX */ | |
910 | ||
bc8e80d5 | 911 | #ifdef CONFIG_CPU_SUP_AMD |
8b84c8df | 912 | extern u16 amd_get_nb_id(int cpu); |
cc2749e4 | 913 | extern u32 amd_get_nodes_per_socket(void); |
bc8e80d5 BP |
914 | #else |
915 | static inline u16 amd_get_nb_id(int cpu) { return 0; } | |
916 | static inline u32 amd_get_nodes_per_socket(void) { return 0; } | |
917 | #endif | |
6a812691 | 918 | |
96e39ac0 JW |
919 | static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves) |
920 | { | |
921 | uint32_t base, eax, signature[3]; | |
922 | ||
923 | for (base = 0x40000000; base < 0x40010000; base += 0x100) { | |
924 | cpuid(base, &eax, &signature[0], &signature[1], &signature[2]); | |
925 | ||
926 | if (!memcmp(sig, signature, 12) && | |
927 | (leaves == 0 || ((eax - base) >= leaves))) | |
928 | return base; | |
929 | } | |
930 | ||
931 | return 0; | |
932 | } | |
933 | ||
f05e798a DH |
934 | extern unsigned long arch_align_stack(unsigned long sp); |
935 | extern void free_init_pages(char *what, unsigned long begin, unsigned long end); | |
936 | ||
937 | void default_idle(void); | |
6a377ddc LB |
938 | #ifdef CONFIG_XEN |
939 | bool xen_set_default_idle(void); | |
940 | #else | |
941 | #define xen_set_default_idle 0 | |
942 | #endif | |
f05e798a DH |
943 | |
944 | void stop_this_cpu(void *dummy); | |
4d067d8e | 945 | void df_debug(struct pt_regs *regs, long error_code); |
1965aae3 | 946 | #endif /* _ASM_X86_PROCESSOR_H */ |