Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86_64/entry.S | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs | |
6 | * Copyright (C) 2000 Pavel Machek <pavel@suse.cz> | |
1da177e4 LT |
7 | */ |
8 | ||
9 | /* | |
10 | * entry.S contains the system-call and fault low-level handling routines. | |
11 | * | |
8b4777a4 AL |
12 | * Some of this is documented in Documentation/x86/entry_64.txt |
13 | * | |
1da177e4 LT |
14 | * NOTE: This code handles signal-recognition, which happens every time |
15 | * after an interrupt and after each system call. | |
0bd7b798 AH |
16 | * |
17 | * Normal syscalls and interrupts don't save a full stack frame, this is | |
1da177e4 | 18 | * only done for syscall tracing, signals or fork/exec et.al. |
0bd7b798 AH |
19 | * |
20 | * A note on terminology: | |
21 | * - top of stack: Architecture defined interrupt frame from SS to RIP | |
22 | * at the top of the kernel process stack. | |
0d2eb44f | 23 | * - partial stack frame: partially saved registers up to R11. |
0bd7b798 | 24 | * - full stack frame: Like partial stack frame, but all register saved. |
2e91a17b AK |
25 | * |
26 | * Some macro usage: | |
27 | * - CFI macros are used to generate dwarf2 unwind information for better | |
28 | * backtraces. They don't change any code. | |
29 | * - SAVE_ALL/RESTORE_ALL - Save/restore all registers | |
30 | * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify. | |
31 | * There are unfortunately lots of special cases where some registers | |
32 | * not touched. The macro is a big mess that should be cleaned up. | |
33 | * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS. | |
34 | * Gives a full stack frame. | |
35 | * - ENTRY/END Define functions in the symbol table. | |
36 | * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack | |
37 | * frame that is otherwise undefined after a SYSCALL | |
38 | * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging. | |
39 | * - errorentry/paranoidentry/zeroentry - Define exception entry points. | |
1da177e4 LT |
40 | */ |
41 | ||
1da177e4 LT |
42 | #include <linux/linkage.h> |
43 | #include <asm/segment.h> | |
1da177e4 LT |
44 | #include <asm/cache.h> |
45 | #include <asm/errno.h> | |
46 | #include <asm/dwarf2.h> | |
47 | #include <asm/calling.h> | |
e2d5df93 | 48 | #include <asm/asm-offsets.h> |
1da177e4 LT |
49 | #include <asm/msr.h> |
50 | #include <asm/unistd.h> | |
51 | #include <asm/thread_info.h> | |
52 | #include <asm/hw_irq.h> | |
0341c14d | 53 | #include <asm/page_types.h> |
2601e64d | 54 | #include <asm/irqflags.h> |
72fe4858 | 55 | #include <asm/paravirt.h> |
395a59d0 | 56 | #include <asm/ftrace.h> |
9939ddaf | 57 | #include <asm/percpu.h> |
d7abc0fa | 58 | #include <asm/asm.h> |
91d1aa43 | 59 | #include <asm/context_tracking.h> |
63bcff2a | 60 | #include <asm/smap.h> |
d7e7528b | 61 | #include <linux/err.h> |
1da177e4 | 62 | |
86a1c34a RM |
63 | /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */ |
64 | #include <linux/elf-em.h> | |
65 | #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE) | |
66 | #define __AUDIT_ARCH_64BIT 0x80000000 | |
67 | #define __AUDIT_ARCH_LE 0x40000000 | |
68 | ||
1da177e4 | 69 | .code64 |
ea714547 JO |
70 | .section .entry.text, "ax" |
71 | ||
606576ce | 72 | #ifdef CONFIG_FUNCTION_TRACER |
d57c5d51 SR |
73 | |
74 | #ifdef CC_USING_FENTRY | |
75 | # define function_hook __fentry__ | |
76 | #else | |
77 | # define function_hook mcount | |
78 | #endif | |
79 | ||
d61f82d0 | 80 | #ifdef CONFIG_DYNAMIC_FTRACE |
d57c5d51 SR |
81 | |
82 | ENTRY(function_hook) | |
d61f82d0 | 83 | retq |
d57c5d51 | 84 | END(function_hook) |
d61f82d0 | 85 | |
08f6fba5 SR |
86 | /* skip is set if stack has been adjusted */ |
87 | .macro ftrace_caller_setup skip=0 | |
88 | MCOUNT_SAVE_FRAME \skip | |
89 | ||
90 | /* Load the ftrace_ops into the 3rd parameter */ | |
91 | leaq function_trace_op, %rdx | |
92 | ||
93 | /* Load ip into the first parameter */ | |
94 | movq RIP(%rsp), %rdi | |
95 | subq $MCOUNT_INSN_SIZE, %rdi | |
96 | /* Load the parent_ip into the second parameter */ | |
d57c5d51 SR |
97 | #ifdef CC_USING_FENTRY |
98 | movq SS+16(%rsp), %rsi | |
99 | #else | |
08f6fba5 | 100 | movq 8(%rbp), %rsi |
d57c5d51 | 101 | #endif |
08f6fba5 | 102 | .endm |
d61f82d0 SR |
103 | |
104 | ENTRY(ftrace_caller) | |
08f6fba5 | 105 | /* Check if tracing was disabled (quick check) */ |
60a7ecf4 SR |
106 | cmpl $0, function_trace_stop |
107 | jne ftrace_stub | |
d61f82d0 | 108 | |
08f6fba5 SR |
109 | ftrace_caller_setup |
110 | /* regs go into 4th parameter (but make it NULL) */ | |
111 | movq $0, %rcx | |
d61f82d0 | 112 | |
bc8b2b92 | 113 | GLOBAL(ftrace_call) |
d61f82d0 SR |
114 | call ftrace_stub |
115 | ||
d680fe44 | 116 | MCOUNT_RESTORE_FRAME |
08f6fba5 | 117 | ftrace_return: |
d61f82d0 | 118 | |
48d68b20 | 119 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER |
bc8b2b92 | 120 | GLOBAL(ftrace_graph_call) |
48d68b20 FW |
121 | jmp ftrace_stub |
122 | #endif | |
d61f82d0 | 123 | |
bc8b2b92 | 124 | GLOBAL(ftrace_stub) |
d61f82d0 SR |
125 | retq |
126 | END(ftrace_caller) | |
127 | ||
08f6fba5 SR |
128 | ENTRY(ftrace_regs_caller) |
129 | /* Save the current flags before compare (in SS location)*/ | |
130 | pushfq | |
131 | ||
132 | /* Check if tracing was disabled (quick check) */ | |
133 | cmpl $0, function_trace_stop | |
134 | jne ftrace_restore_flags | |
135 | ||
136 | /* skip=8 to skip flags saved in SS */ | |
137 | ftrace_caller_setup 8 | |
138 | ||
139 | /* Save the rest of pt_regs */ | |
140 | movq %r15, R15(%rsp) | |
141 | movq %r14, R14(%rsp) | |
142 | movq %r13, R13(%rsp) | |
143 | movq %r12, R12(%rsp) | |
144 | movq %r11, R11(%rsp) | |
145 | movq %r10, R10(%rsp) | |
146 | movq %rbp, RBP(%rsp) | |
147 | movq %rbx, RBX(%rsp) | |
148 | /* Copy saved flags */ | |
149 | movq SS(%rsp), %rcx | |
150 | movq %rcx, EFLAGS(%rsp) | |
151 | /* Kernel segments */ | |
152 | movq $__KERNEL_DS, %rcx | |
153 | movq %rcx, SS(%rsp) | |
154 | movq $__KERNEL_CS, %rcx | |
155 | movq %rcx, CS(%rsp) | |
156 | /* Stack - skipping return address */ | |
157 | leaq SS+16(%rsp), %rcx | |
158 | movq %rcx, RSP(%rsp) | |
159 | ||
160 | /* regs go into 4th parameter */ | |
161 | leaq (%rsp), %rcx | |
162 | ||
163 | GLOBAL(ftrace_regs_call) | |
164 | call ftrace_stub | |
165 | ||
166 | /* Copy flags back to SS, to restore them */ | |
167 | movq EFLAGS(%rsp), %rax | |
168 | movq %rax, SS(%rsp) | |
169 | ||
47d5a5f8 SR |
170 | /* Handlers can change the RIP */ |
171 | movq RIP(%rsp), %rax | |
172 | movq %rax, SS+8(%rsp) | |
173 | ||
08f6fba5 SR |
174 | /* restore the rest of pt_regs */ |
175 | movq R15(%rsp), %r15 | |
176 | movq R14(%rsp), %r14 | |
177 | movq R13(%rsp), %r13 | |
178 | movq R12(%rsp), %r12 | |
179 | movq R10(%rsp), %r10 | |
180 | movq RBP(%rsp), %rbp | |
181 | movq RBX(%rsp), %rbx | |
182 | ||
183 | /* skip=8 to skip flags saved in SS */ | |
184 | MCOUNT_RESTORE_FRAME 8 | |
185 | ||
186 | /* Restore flags */ | |
187 | popfq | |
188 | ||
189 | jmp ftrace_return | |
190 | ftrace_restore_flags: | |
191 | popfq | |
192 | jmp ftrace_stub | |
193 | ||
194 | END(ftrace_regs_caller) | |
195 | ||
196 | ||
d61f82d0 | 197 | #else /* ! CONFIG_DYNAMIC_FTRACE */ |
d57c5d51 SR |
198 | |
199 | ENTRY(function_hook) | |
60a7ecf4 SR |
200 | cmpl $0, function_trace_stop |
201 | jne ftrace_stub | |
202 | ||
16444a8a ACM |
203 | cmpq $ftrace_stub, ftrace_trace_function |
204 | jnz trace | |
48d68b20 FW |
205 | |
206 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER | |
207 | cmpq $ftrace_stub, ftrace_graph_return | |
208 | jnz ftrace_graph_caller | |
e49dc19c SR |
209 | |
210 | cmpq $ftrace_graph_entry_stub, ftrace_graph_entry | |
211 | jnz ftrace_graph_caller | |
48d68b20 FW |
212 | #endif |
213 | ||
bc8b2b92 | 214 | GLOBAL(ftrace_stub) |
16444a8a ACM |
215 | retq |
216 | ||
217 | trace: | |
d680fe44 | 218 | MCOUNT_SAVE_FRAME |
16444a8a | 219 | |
08f6fba5 | 220 | movq RIP(%rsp), %rdi |
d57c5d51 SR |
221 | #ifdef CC_USING_FENTRY |
222 | movq SS+16(%rsp), %rsi | |
223 | #else | |
16444a8a | 224 | movq 8(%rbp), %rsi |
d57c5d51 | 225 | #endif |
395a59d0 | 226 | subq $MCOUNT_INSN_SIZE, %rdi |
16444a8a ACM |
227 | |
228 | call *ftrace_trace_function | |
229 | ||
d680fe44 | 230 | MCOUNT_RESTORE_FRAME |
16444a8a ACM |
231 | |
232 | jmp ftrace_stub | |
d57c5d51 | 233 | END(function_hook) |
d61f82d0 | 234 | #endif /* CONFIG_DYNAMIC_FTRACE */ |
606576ce | 235 | #endif /* CONFIG_FUNCTION_TRACER */ |
16444a8a | 236 | |
48d68b20 FW |
237 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER |
238 | ENTRY(ftrace_graph_caller) | |
d680fe44 | 239 | MCOUNT_SAVE_FRAME |
48d68b20 | 240 | |
d57c5d51 SR |
241 | #ifdef CC_USING_FENTRY |
242 | leaq SS+16(%rsp), %rdi | |
243 | movq $0, %rdx /* No framepointers needed */ | |
244 | #else | |
48d68b20 | 245 | leaq 8(%rbp), %rdi |
71e308a2 | 246 | movq (%rbp), %rdx |
d57c5d51 SR |
247 | #endif |
248 | movq RIP(%rsp), %rsi | |
bb4304c7 | 249 | subq $MCOUNT_INSN_SIZE, %rsi |
48d68b20 FW |
250 | |
251 | call prepare_ftrace_return | |
252 | ||
d680fe44 CG |
253 | MCOUNT_RESTORE_FRAME |
254 | ||
48d68b20 FW |
255 | retq |
256 | END(ftrace_graph_caller) | |
257 | ||
bc8b2b92 | 258 | GLOBAL(return_to_handler) |
4818d809 | 259 | subq $24, %rsp |
48d68b20 | 260 | |
e71e99c2 | 261 | /* Save the return values */ |
16444a8a | 262 | movq %rax, (%rsp) |
e71e99c2 | 263 | movq %rdx, 8(%rsp) |
71e308a2 | 264 | movq %rbp, %rdi |
16444a8a | 265 | |
48d68b20 | 266 | call ftrace_return_to_handler |
16444a8a | 267 | |
194ec341 | 268 | movq %rax, %rdi |
e71e99c2 | 269 | movq 8(%rsp), %rdx |
16444a8a | 270 | movq (%rsp), %rax |
194ec341 SR |
271 | addq $24, %rsp |
272 | jmp *%rdi | |
48d68b20 | 273 | #endif |
16444a8a | 274 | |
16444a8a | 275 | |
dc37db4d | 276 | #ifndef CONFIG_PREEMPT |
1da177e4 | 277 | #define retint_kernel retint_restore_args |
0bd7b798 | 278 | #endif |
2601e64d | 279 | |
72fe4858 | 280 | #ifdef CONFIG_PARAVIRT |
2be29982 | 281 | ENTRY(native_usergs_sysret64) |
72fe4858 GOC |
282 | swapgs |
283 | sysretq | |
b3baaa13 | 284 | ENDPROC(native_usergs_sysret64) |
72fe4858 GOC |
285 | #endif /* CONFIG_PARAVIRT */ |
286 | ||
2601e64d IM |
287 | |
288 | .macro TRACE_IRQS_IRETQ offset=ARGOFFSET | |
289 | #ifdef CONFIG_TRACE_IRQFLAGS | |
290 | bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */ | |
291 | jnc 1f | |
292 | TRACE_IRQS_ON | |
293 | 1: | |
294 | #endif | |
295 | .endm | |
296 | ||
5963e317 SR |
297 | /* |
298 | * When dynamic function tracer is enabled it will add a breakpoint | |
299 | * to all locations that it is about to modify, sync CPUs, update | |
300 | * all the code, sync CPUs, then remove the breakpoints. In this time | |
301 | * if lockdep is enabled, it might jump back into the debug handler | |
302 | * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF). | |
303 | * | |
304 | * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to | |
305 | * make sure the stack pointer does not get reset back to the top | |
306 | * of the debug stack, and instead just reuses the current stack. | |
307 | */ | |
308 | #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS) | |
309 | ||
310 | .macro TRACE_IRQS_OFF_DEBUG | |
311 | call debug_stack_set_zero | |
312 | TRACE_IRQS_OFF | |
313 | call debug_stack_reset | |
314 | .endm | |
315 | ||
316 | .macro TRACE_IRQS_ON_DEBUG | |
317 | call debug_stack_set_zero | |
318 | TRACE_IRQS_ON | |
319 | call debug_stack_reset | |
320 | .endm | |
321 | ||
322 | .macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET | |
323 | bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */ | |
324 | jnc 1f | |
325 | TRACE_IRQS_ON_DEBUG | |
326 | 1: | |
327 | .endm | |
328 | ||
329 | #else | |
330 | # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF | |
331 | # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON | |
332 | # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ | |
333 | #endif | |
334 | ||
1da177e4 | 335 | /* |
0bd7b798 AH |
336 | * C code is not supposed to know about undefined top of stack. Every time |
337 | * a C function with an pt_regs argument is called from the SYSCALL based | |
1da177e4 LT |
338 | * fast path FIXUP_TOP_OF_STACK is needed. |
339 | * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs | |
340 | * manipulation. | |
0bd7b798 AH |
341 | */ |
342 | ||
343 | /* %rsp:at FRAMEEND */ | |
c002a1e6 | 344 | .macro FIXUP_TOP_OF_STACK tmp offset=0 |
3d1e42a7 | 345 | movq PER_CPU_VAR(old_rsp),\tmp |
c002a1e6 AH |
346 | movq \tmp,RSP+\offset(%rsp) |
347 | movq $__USER_DS,SS+\offset(%rsp) | |
348 | movq $__USER_CS,CS+\offset(%rsp) | |
349 | movq $-1,RCX+\offset(%rsp) | |
350 | movq R11+\offset(%rsp),\tmp /* get eflags */ | |
351 | movq \tmp,EFLAGS+\offset(%rsp) | |
1da177e4 LT |
352 | .endm |
353 | ||
c002a1e6 AH |
354 | .macro RESTORE_TOP_OF_STACK tmp offset=0 |
355 | movq RSP+\offset(%rsp),\tmp | |
3d1e42a7 | 356 | movq \tmp,PER_CPU_VAR(old_rsp) |
c002a1e6 AH |
357 | movq EFLAGS+\offset(%rsp),\tmp |
358 | movq \tmp,R11+\offset(%rsp) | |
1da177e4 LT |
359 | .endm |
360 | ||
361 | .macro FAKE_STACK_FRAME child_rip | |
362 | /* push in order ss, rsp, eflags, cs, rip */ | |
3829ee6b | 363 | xorl %eax, %eax |
df5d1874 | 364 | pushq_cfi $__KERNEL_DS /* ss */ |
7effaa88 | 365 | /*CFI_REL_OFFSET ss,0*/ |
df5d1874 | 366 | pushq_cfi %rax /* rsp */ |
7effaa88 | 367 | CFI_REL_OFFSET rsp,0 |
1cf8343f | 368 | pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */ |
7effaa88 | 369 | /*CFI_REL_OFFSET rflags,0*/ |
df5d1874 | 370 | pushq_cfi $__KERNEL_CS /* cs */ |
7effaa88 | 371 | /*CFI_REL_OFFSET cs,0*/ |
df5d1874 | 372 | pushq_cfi \child_rip /* rip */ |
7effaa88 | 373 | CFI_REL_OFFSET rip,0 |
df5d1874 | 374 | pushq_cfi %rax /* orig rax */ |
1da177e4 LT |
375 | .endm |
376 | ||
377 | .macro UNFAKE_STACK_FRAME | |
378 | addq $8*6, %rsp | |
379 | CFI_ADJUST_CFA_OFFSET -(6*8) | |
380 | .endm | |
381 | ||
dcd072e2 AH |
382 | /* |
383 | * initial frame state for interrupts (and exceptions without error code) | |
384 | */ | |
385 | .macro EMPTY_FRAME start=1 offset=0 | |
7effaa88 | 386 | .if \start |
dcd072e2 | 387 | CFI_STARTPROC simple |
adf14236 | 388 | CFI_SIGNAL_FRAME |
dcd072e2 | 389 | CFI_DEF_CFA rsp,8+\offset |
7effaa88 | 390 | .else |
dcd072e2 | 391 | CFI_DEF_CFA_OFFSET 8+\offset |
7effaa88 | 392 | .endif |
1da177e4 | 393 | .endm |
d99015b1 AH |
394 | |
395 | /* | |
dcd072e2 | 396 | * initial frame state for interrupts (and exceptions without error code) |
d99015b1 | 397 | */ |
dcd072e2 | 398 | .macro INTR_FRAME start=1 offset=0 |
e8a0e276 IM |
399 | EMPTY_FRAME \start, SS+8+\offset-RIP |
400 | /*CFI_REL_OFFSET ss, SS+\offset-RIP*/ | |
401 | CFI_REL_OFFSET rsp, RSP+\offset-RIP | |
402 | /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/ | |
403 | /*CFI_REL_OFFSET cs, CS+\offset-RIP*/ | |
404 | CFI_REL_OFFSET rip, RIP+\offset-RIP | |
d99015b1 AH |
405 | .endm |
406 | ||
d99015b1 AH |
407 | /* |
408 | * initial frame state for exceptions with error code (and interrupts | |
409 | * with vector already pushed) | |
410 | */ | |
dcd072e2 | 411 | .macro XCPT_FRAME start=1 offset=0 |
e8a0e276 | 412 | INTR_FRAME \start, RIP+\offset-ORIG_RAX |
dcd072e2 AH |
413 | /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/ |
414 | .endm | |
415 | ||
416 | /* | |
417 | * frame that enables calling into C. | |
418 | */ | |
419 | .macro PARTIAL_FRAME start=1 offset=0 | |
e8a0e276 IM |
420 | XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET |
421 | CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET | |
422 | CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET | |
423 | CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET | |
424 | CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET | |
425 | CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET | |
426 | CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET | |
427 | CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET | |
428 | CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET | |
429 | CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET | |
dcd072e2 AH |
430 | .endm |
431 | ||
432 | /* | |
433 | * frame that enables passing a complete pt_regs to a C function. | |
434 | */ | |
435 | .macro DEFAULT_FRAME start=1 offset=0 | |
e8a0e276 | 436 | PARTIAL_FRAME \start, R11+\offset-R15 |
dcd072e2 AH |
437 | CFI_REL_OFFSET rbx, RBX+\offset |
438 | CFI_REL_OFFSET rbp, RBP+\offset | |
439 | CFI_REL_OFFSET r12, R12+\offset | |
440 | CFI_REL_OFFSET r13, R13+\offset | |
441 | CFI_REL_OFFSET r14, R14+\offset | |
442 | CFI_REL_OFFSET r15, R15+\offset | |
443 | .endm | |
d99015b1 AH |
444 | |
445 | /* save partial stack frame */ | |
1871853f | 446 | .macro SAVE_ARGS_IRQ |
d99015b1 | 447 | cld |
1871853f | 448 | /* start from rbp in pt_regs and jump over */ |
1b2b23d8 TG |
449 | movq_cfi rdi, (RDI-RBP) |
450 | movq_cfi rsi, (RSI-RBP) | |
451 | movq_cfi rdx, (RDX-RBP) | |
452 | movq_cfi rcx, (RCX-RBP) | |
453 | movq_cfi rax, (RAX-RBP) | |
454 | movq_cfi r8, (R8-RBP) | |
455 | movq_cfi r9, (R9-RBP) | |
456 | movq_cfi r10, (R10-RBP) | |
457 | movq_cfi r11, (R11-RBP) | |
1871853f | 458 | |
a2bbe750 FW |
459 | /* Save rbp so that we can unwind from get_irq_regs() */ |
460 | movq_cfi rbp, 0 | |
461 | ||
462 | /* Save previous stack value */ | |
463 | movq %rsp, %rsi | |
3b99a3ef FW |
464 | |
465 | leaq -RBP(%rsp),%rdi /* arg1 for handler */ | |
69466466 | 466 | testl $3, CS-RBP(%rsi) |
d99015b1 AH |
467 | je 1f |
468 | SWAPGS | |
469 | /* | |
56895530 | 470 | * irq_count is used to check if a CPU is already on an interrupt stack |
d99015b1 AH |
471 | * or not. While this is essentially redundant with preempt_count it is |
472 | * a little cheaper to use a separate counter in the PDA (short of | |
473 | * moving irq_enter into assembly, which would be too much work) | |
474 | */ | |
56895530 | 475 | 1: incl PER_CPU_VAR(irq_count) |
69466466 | 476 | cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp |
eab9e613 | 477 | CFI_DEF_CFA_REGISTER rsi |
a2bbe750 | 478 | |
69466466 | 479 | /* Store previous stack value */ |
a2bbe750 | 480 | pushq %rsi |
eab9e613 JB |
481 | CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \ |
482 | 0x77 /* DW_OP_breg7 */, 0, \ | |
483 | 0x06 /* DW_OP_deref */, \ | |
484 | 0x08 /* DW_OP_const1u */, SS+8-RBP, \ | |
485 | 0x22 /* DW_OP_plus */ | |
a2bbe750 FW |
486 | /* We entered an interrupt context - irqs are off: */ |
487 | TRACE_IRQS_OFF | |
1871853f | 488 | .endm |
d99015b1 | 489 | |
c002a1e6 | 490 | ENTRY(save_rest) |
1b2b23d8 | 491 | PARTIAL_FRAME 1 (REST_SKIP+8) |
c002a1e6 AH |
492 | movq 5*8+16(%rsp), %r11 /* save return address */ |
493 | movq_cfi rbx, RBX+16 | |
494 | movq_cfi rbp, RBP+16 | |
495 | movq_cfi r12, R12+16 | |
496 | movq_cfi r13, R13+16 | |
497 | movq_cfi r14, R14+16 | |
498 | movq_cfi r15, R15+16 | |
499 | movq %r11, 8(%rsp) /* return address */ | |
500 | FIXUP_TOP_OF_STACK %r11, 16 | |
501 | ret | |
502 | CFI_ENDPROC | |
503 | END(save_rest) | |
504 | ||
e2f6bc25 | 505 | /* save complete stack frame */ |
c2810188 | 506 | .pushsection .kprobes.text, "ax" |
e2f6bc25 AH |
507 | ENTRY(save_paranoid) |
508 | XCPT_FRAME 1 RDI+8 | |
509 | cld | |
510 | movq_cfi rdi, RDI+8 | |
511 | movq_cfi rsi, RSI+8 | |
512 | movq_cfi rdx, RDX+8 | |
513 | movq_cfi rcx, RCX+8 | |
514 | movq_cfi rax, RAX+8 | |
515 | movq_cfi r8, R8+8 | |
516 | movq_cfi r9, R9+8 | |
517 | movq_cfi r10, R10+8 | |
518 | movq_cfi r11, R11+8 | |
519 | movq_cfi rbx, RBX+8 | |
520 | movq_cfi rbp, RBP+8 | |
521 | movq_cfi r12, R12+8 | |
522 | movq_cfi r13, R13+8 | |
523 | movq_cfi r14, R14+8 | |
524 | movq_cfi r15, R15+8 | |
525 | movl $1,%ebx | |
526 | movl $MSR_GS_BASE,%ecx | |
527 | rdmsr | |
528 | testl %edx,%edx | |
529 | js 1f /* negative -> in kernel */ | |
530 | SWAPGS | |
531 | xorl %ebx,%ebx | |
532 | 1: ret | |
533 | CFI_ENDPROC | |
534 | END(save_paranoid) | |
c2810188 | 535 | .popsection |
e2f6bc25 | 536 | |
1da177e4 | 537 | /* |
5b3eec0c IM |
538 | * A newly forked process directly context switches into this address. |
539 | * | |
540 | * rdi: prev task we switched from | |
0bd7b798 | 541 | */ |
1da177e4 | 542 | ENTRY(ret_from_fork) |
dcd072e2 | 543 | DEFAULT_FRAME |
5b3eec0c | 544 | |
7106a5ab BL |
545 | LOCK ; btr $TIF_FORK,TI_flags(%r8) |
546 | ||
6eebdda3 | 547 | pushq_cfi $0x0002 |
df5d1874 | 548 | popfq_cfi # reset kernel eflags |
5b3eec0c IM |
549 | |
550 | call schedule_tail # rdi: 'prev' task parameter | |
551 | ||
1da177e4 | 552 | GET_THREAD_INFO(%rcx) |
5b3eec0c | 553 | |
1da177e4 | 554 | RESTORE_REST |
5b3eec0c IM |
555 | |
556 | testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread? | |
7076aada | 557 | jz 1f |
5b3eec0c IM |
558 | |
559 | testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET | |
1da177e4 | 560 | jnz int_ret_from_sys_call |
5b3eec0c | 561 | |
c002a1e6 | 562 | RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET |
5b3eec0c IM |
563 | jmp ret_from_sys_call # go to the SYSRET fastpath |
564 | ||
7076aada | 565 | 1: |
22e2430d | 566 | subq $REST_SKIP, %rsp # leave space for volatiles |
7076aada AV |
567 | CFI_ADJUST_CFA_OFFSET REST_SKIP |
568 | movq %rbp, %rdi | |
569 | call *%rbx | |
22e2430d AV |
570 | movl $0, RAX(%rsp) |
571 | RESTORE_REST | |
572 | jmp int_ret_from_sys_call | |
1da177e4 | 573 | CFI_ENDPROC |
4b787e0b | 574 | END(ret_from_fork) |
1da177e4 LT |
575 | |
576 | /* | |
0d2eb44f | 577 | * System call entry. Up to 6 arguments in registers are supported. |
1da177e4 LT |
578 | * |
579 | * SYSCALL does not save anything on the stack and does not change the | |
63bcff2a PA |
580 | * stack pointer. However, it does mask the flags register for us, so |
581 | * CLD and CLAC are not needed. | |
1da177e4 | 582 | */ |
0bd7b798 | 583 | |
1da177e4 | 584 | /* |
0bd7b798 | 585 | * Register setup: |
1da177e4 LT |
586 | * rax system call number |
587 | * rdi arg0 | |
0bd7b798 | 588 | * rcx return address for syscall/sysret, C arg3 |
1da177e4 | 589 | * rsi arg1 |
0bd7b798 | 590 | * rdx arg2 |
1da177e4 LT |
591 | * r10 arg3 (--> moved to rcx for C) |
592 | * r8 arg4 | |
593 | * r9 arg5 | |
594 | * r11 eflags for syscall/sysret, temporary for C | |
0bd7b798 AH |
595 | * r12-r15,rbp,rbx saved by C code, not touched. |
596 | * | |
1da177e4 LT |
597 | * Interrupts are off on entry. |
598 | * Only called from user space. | |
599 | * | |
600 | * XXX if we had a free scratch register we could save the RSP into the stack frame | |
601 | * and report it properly in ps. Unfortunately we haven't. | |
7bf36bbc AK |
602 | * |
603 | * When user can change the frames always force IRET. That is because | |
604 | * it deals with uncanonical addresses better. SYSRET has trouble | |
605 | * with them due to bugs in both AMD and Intel CPUs. | |
0bd7b798 | 606 | */ |
1da177e4 LT |
607 | |
608 | ENTRY(system_call) | |
7effaa88 | 609 | CFI_STARTPROC simple |
adf14236 | 610 | CFI_SIGNAL_FRAME |
9af45651 | 611 | CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET |
7effaa88 JB |
612 | CFI_REGISTER rip,rcx |
613 | /*CFI_REGISTER rflags,r11*/ | |
72fe4858 GOC |
614 | SWAPGS_UNSAFE_STACK |
615 | /* | |
616 | * A hypervisor implementation might want to use a label | |
617 | * after the swapgs, so that it can do the swapgs | |
618 | * for the guest and jump here on syscall. | |
619 | */ | |
f6b2bc84 | 620 | GLOBAL(system_call_after_swapgs) |
72fe4858 | 621 | |
3d1e42a7 | 622 | movq %rsp,PER_CPU_VAR(old_rsp) |
9af45651 | 623 | movq PER_CPU_VAR(kernel_stack),%rsp |
2601e64d IM |
624 | /* |
625 | * No need to follow this irqs off/on section - it's straight | |
626 | * and short: | |
627 | */ | |
72fe4858 | 628 | ENABLE_INTERRUPTS(CLBR_NONE) |
cac0e0a7 | 629 | SAVE_ARGS 8,0 |
0bd7b798 | 630 | movq %rax,ORIG_RAX-ARGOFFSET(%rsp) |
7effaa88 JB |
631 | movq %rcx,RIP-ARGOFFSET(%rsp) |
632 | CFI_REL_OFFSET rip,RIP-ARGOFFSET | |
46db09d3 | 633 | testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET) |
1da177e4 | 634 | jnz tracesys |
86a1c34a | 635 | system_call_fastpath: |
fca460f9 | 636 | #if __SYSCALL_MASK == ~0 |
1da177e4 | 637 | cmpq $__NR_syscall_max,%rax |
fca460f9 PA |
638 | #else |
639 | andl $__SYSCALL_MASK,%eax | |
640 | cmpl $__NR_syscall_max,%eax | |
641 | #endif | |
1da177e4 LT |
642 | ja badsys |
643 | movq %r10,%rcx | |
644 | call *sys_call_table(,%rax,8) # XXX: rip relative | |
645 | movq %rax,RAX-ARGOFFSET(%rsp) | |
646 | /* | |
647 | * Syscall return path ending with SYSRET (fast path) | |
0bd7b798 AH |
648 | * Has incomplete stack frame and undefined top of stack. |
649 | */ | |
1da177e4 | 650 | ret_from_sys_call: |
11b854b2 | 651 | movl $_TIF_ALLWORK_MASK,%edi |
1da177e4 | 652 | /* edi: flagmask */ |
0bd7b798 | 653 | sysret_check: |
10cd706d | 654 | LOCKDEP_SYS_EXIT |
72fe4858 | 655 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 656 | TRACE_IRQS_OFF |
46db09d3 | 657 | movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx |
1da177e4 | 658 | andl %edi,%edx |
0bd7b798 | 659 | jnz sysret_careful |
bcddc015 | 660 | CFI_REMEMBER_STATE |
2601e64d IM |
661 | /* |
662 | * sysretq will re-enable interrupts: | |
663 | */ | |
664 | TRACE_IRQS_ON | |
1da177e4 | 665 | movq RIP-ARGOFFSET(%rsp),%rcx |
7effaa88 | 666 | CFI_REGISTER rip,rcx |
838feb47 | 667 | RESTORE_ARGS 1,-ARG_SKIP,0 |
7effaa88 | 668 | /*CFI_REGISTER rflags,r11*/ |
3d1e42a7 | 669 | movq PER_CPU_VAR(old_rsp), %rsp |
2be29982 | 670 | USERGS_SYSRET64 |
1da177e4 | 671 | |
bcddc015 | 672 | CFI_RESTORE_STATE |
1da177e4 | 673 | /* Handle reschedules */ |
0bd7b798 | 674 | /* edx: work, edi: workmask */ |
1da177e4 LT |
675 | sysret_careful: |
676 | bt $TIF_NEED_RESCHED,%edx | |
677 | jnc sysret_signal | |
2601e64d | 678 | TRACE_IRQS_ON |
72fe4858 | 679 | ENABLE_INTERRUPTS(CLBR_NONE) |
df5d1874 | 680 | pushq_cfi %rdi |
0430499c | 681 | SCHEDULE_USER |
df5d1874 | 682 | popq_cfi %rdi |
1da177e4 LT |
683 | jmp sysret_check |
684 | ||
0bd7b798 | 685 | /* Handle a signal */ |
1da177e4 | 686 | sysret_signal: |
2601e64d | 687 | TRACE_IRQS_ON |
72fe4858 | 688 | ENABLE_INTERRUPTS(CLBR_NONE) |
86a1c34a RM |
689 | #ifdef CONFIG_AUDITSYSCALL |
690 | bt $TIF_SYSCALL_AUDIT,%edx | |
691 | jc sysret_audit | |
692 | #endif | |
b60e714d RM |
693 | /* |
694 | * We have a signal, or exit tracing or single-step. | |
695 | * These all wind up with the iret return path anyway, | |
696 | * so just join that path right now. | |
697 | */ | |
698 | FIXUP_TOP_OF_STACK %r11, -ARGOFFSET | |
699 | jmp int_check_syscall_exit_work | |
0bd7b798 | 700 | |
7effaa88 JB |
701 | badsys: |
702 | movq $-ENOSYS,RAX-ARGOFFSET(%rsp) | |
703 | jmp ret_from_sys_call | |
704 | ||
86a1c34a RM |
705 | #ifdef CONFIG_AUDITSYSCALL |
706 | /* | |
707 | * Fast path for syscall audit without full syscall trace. | |
b05d8447 | 708 | * We just call __audit_syscall_entry() directly, and then |
86a1c34a RM |
709 | * jump back to the normal fast path. |
710 | */ | |
711 | auditsys: | |
712 | movq %r10,%r9 /* 6th arg: 4th syscall arg */ | |
713 | movq %rdx,%r8 /* 5th arg: 3rd syscall arg */ | |
714 | movq %rsi,%rcx /* 4th arg: 2nd syscall arg */ | |
715 | movq %rdi,%rdx /* 3rd arg: 1st syscall arg */ | |
716 | movq %rax,%rsi /* 2nd arg: syscall number */ | |
717 | movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */ | |
b05d8447 | 718 | call __audit_syscall_entry |
86a1c34a RM |
719 | LOAD_ARGS 0 /* reload call-clobbered registers */ |
720 | jmp system_call_fastpath | |
721 | ||
722 | /* | |
d7e7528b | 723 | * Return fast path for syscall audit. Call __audit_syscall_exit() |
86a1c34a RM |
724 | * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT |
725 | * masked off. | |
726 | */ | |
727 | sysret_audit: | |
03275591 | 728 | movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */ |
d7e7528b EP |
729 | cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */ |
730 | setbe %al /* 1 if so, 0 if not */ | |
86a1c34a | 731 | movzbl %al,%edi /* zero-extend that into %edi */ |
d7e7528b | 732 | call __audit_syscall_exit |
86a1c34a RM |
733 | movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi |
734 | jmp sysret_check | |
735 | #endif /* CONFIG_AUDITSYSCALL */ | |
736 | ||
1da177e4 | 737 | /* Do syscall tracing */ |
0bd7b798 | 738 | tracesys: |
86a1c34a | 739 | #ifdef CONFIG_AUDITSYSCALL |
46db09d3 | 740 | testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET) |
86a1c34a RM |
741 | jz auditsys |
742 | #endif | |
1da177e4 | 743 | SAVE_REST |
a31f8dd7 | 744 | movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */ |
1da177e4 LT |
745 | FIXUP_TOP_OF_STACK %rdi |
746 | movq %rsp,%rdi | |
747 | call syscall_trace_enter | |
d4d67150 RM |
748 | /* |
749 | * Reload arg registers from stack in case ptrace changed them. | |
750 | * We don't reload %rax because syscall_trace_enter() returned | |
751 | * the value it wants us to use in the table lookup. | |
752 | */ | |
753 | LOAD_ARGS ARGOFFSET, 1 | |
1da177e4 | 754 | RESTORE_REST |
fca460f9 | 755 | #if __SYSCALL_MASK == ~0 |
1da177e4 | 756 | cmpq $__NR_syscall_max,%rax |
fca460f9 PA |
757 | #else |
758 | andl $__SYSCALL_MASK,%eax | |
759 | cmpl $__NR_syscall_max,%eax | |
760 | #endif | |
a31f8dd7 | 761 | ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */ |
1da177e4 LT |
762 | movq %r10,%rcx /* fixup for C */ |
763 | call *sys_call_table(,%rax,8) | |
a31f8dd7 | 764 | movq %rax,RAX-ARGOFFSET(%rsp) |
7bf36bbc | 765 | /* Use IRET because user could have changed frame */ |
0bd7b798 AH |
766 | |
767 | /* | |
1da177e4 LT |
768 | * Syscall return path ending with IRET. |
769 | * Has correct top of stack, but partial stack frame. | |
bcddc015 | 770 | */ |
bc8b2b92 | 771 | GLOBAL(int_ret_from_sys_call) |
72fe4858 | 772 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 773 | TRACE_IRQS_OFF |
1da177e4 LT |
774 | movl $_TIF_ALLWORK_MASK,%edi |
775 | /* edi: mask to check */ | |
bc8b2b92 | 776 | GLOBAL(int_with_check) |
10cd706d | 777 | LOCKDEP_SYS_EXIT_IRQ |
1da177e4 | 778 | GET_THREAD_INFO(%rcx) |
26ccb8a7 | 779 | movl TI_flags(%rcx),%edx |
1da177e4 LT |
780 | andl %edi,%edx |
781 | jnz int_careful | |
26ccb8a7 | 782 | andl $~TS_COMPAT,TI_status(%rcx) |
1da177e4 LT |
783 | jmp retint_swapgs |
784 | ||
785 | /* Either reschedule or signal or syscall exit tracking needed. */ | |
786 | /* First do a reschedule test. */ | |
787 | /* edx: work, edi: workmask */ | |
788 | int_careful: | |
789 | bt $TIF_NEED_RESCHED,%edx | |
790 | jnc int_very_careful | |
2601e64d | 791 | TRACE_IRQS_ON |
72fe4858 | 792 | ENABLE_INTERRUPTS(CLBR_NONE) |
df5d1874 | 793 | pushq_cfi %rdi |
0430499c | 794 | SCHEDULE_USER |
df5d1874 | 795 | popq_cfi %rdi |
72fe4858 | 796 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 797 | TRACE_IRQS_OFF |
1da177e4 LT |
798 | jmp int_with_check |
799 | ||
800 | /* handle signals and tracing -- both require a full stack frame */ | |
801 | int_very_careful: | |
2601e64d | 802 | TRACE_IRQS_ON |
72fe4858 | 803 | ENABLE_INTERRUPTS(CLBR_NONE) |
b60e714d | 804 | int_check_syscall_exit_work: |
1da177e4 | 805 | SAVE_REST |
0bd7b798 | 806 | /* Check for syscall exit trace */ |
d4d67150 | 807 | testl $_TIF_WORK_SYSCALL_EXIT,%edx |
1da177e4 | 808 | jz int_signal |
df5d1874 | 809 | pushq_cfi %rdi |
0bd7b798 | 810 | leaq 8(%rsp),%rdi # &ptregs -> arg1 |
1da177e4 | 811 | call syscall_trace_leave |
df5d1874 | 812 | popq_cfi %rdi |
d4d67150 | 813 | andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi |
1da177e4 | 814 | jmp int_restore_rest |
0bd7b798 | 815 | |
1da177e4 | 816 | int_signal: |
8f4d37ec | 817 | testl $_TIF_DO_NOTIFY_MASK,%edx |
1da177e4 LT |
818 | jz 1f |
819 | movq %rsp,%rdi # &ptregs -> arg1 | |
820 | xorl %esi,%esi # oldset -> arg2 | |
821 | call do_notify_resume | |
eca91e78 | 822 | 1: movl $_TIF_WORK_MASK,%edi |
1da177e4 LT |
823 | int_restore_rest: |
824 | RESTORE_REST | |
72fe4858 | 825 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 826 | TRACE_IRQS_OFF |
1da177e4 LT |
827 | jmp int_with_check |
828 | CFI_ENDPROC | |
bcddc015 | 829 | END(system_call) |
0bd7b798 | 830 | |
1d4b4b29 AV |
831 | .macro FORK_LIKE func |
832 | ENTRY(stub_\func) | |
833 | CFI_STARTPROC | |
834 | popq %r11 /* save return address */ | |
835 | PARTIAL_FRAME 0 | |
836 | SAVE_REST | |
837 | pushq %r11 /* put it back on stack */ | |
838 | FIXUP_TOP_OF_STACK %r11, 8 | |
839 | DEFAULT_FRAME 0 8 /* offset 8: return address */ | |
840 | call sys_\func | |
841 | RESTORE_TOP_OF_STACK %r11, 8 | |
842 | ret $REST_SKIP /* pop extended registers */ | |
843 | CFI_ENDPROC | |
844 | END(stub_\func) | |
845 | .endm | |
846 | ||
b3af11af AV |
847 | .macro FIXED_FRAME label,func |
848 | ENTRY(\label) | |
849 | CFI_STARTPROC | |
850 | PARTIAL_FRAME 0 8 /* offset 8: return address */ | |
851 | FIXUP_TOP_OF_STACK %r11, 8-ARGOFFSET | |
852 | call \func | |
853 | RESTORE_TOP_OF_STACK %r11, 8-ARGOFFSET | |
854 | ret | |
855 | CFI_ENDPROC | |
856 | END(\label) | |
857 | .endm | |
858 | ||
1d4b4b29 AV |
859 | FORK_LIKE clone |
860 | FORK_LIKE fork | |
861 | FORK_LIKE vfork | |
b3af11af | 862 | FIXED_FRAME stub_iopl, sys_iopl |
1da177e4 LT |
863 | |
864 | ENTRY(ptregscall_common) | |
c002a1e6 AH |
865 | DEFAULT_FRAME 1 8 /* offset 8: return address */ |
866 | RESTORE_TOP_OF_STACK %r11, 8 | |
867 | movq_cfi_restore R15+8, r15 | |
868 | movq_cfi_restore R14+8, r14 | |
869 | movq_cfi_restore R13+8, r13 | |
870 | movq_cfi_restore R12+8, r12 | |
871 | movq_cfi_restore RBP+8, rbp | |
872 | movq_cfi_restore RBX+8, rbx | |
873 | ret $REST_SKIP /* pop extended registers */ | |
1da177e4 | 874 | CFI_ENDPROC |
4b787e0b | 875 | END(ptregscall_common) |
0bd7b798 | 876 | |
1da177e4 LT |
877 | ENTRY(stub_execve) |
878 | CFI_STARTPROC | |
e6b04b6b JB |
879 | addq $8, %rsp |
880 | PARTIAL_FRAME 0 | |
1da177e4 | 881 | SAVE_REST |
1da177e4 LT |
882 | FIXUP_TOP_OF_STACK %r11 |
883 | call sys_execve | |
1da177e4 LT |
884 | movq %rax,RAX(%rsp) |
885 | RESTORE_REST | |
886 | jmp int_ret_from_sys_call | |
887 | CFI_ENDPROC | |
4b787e0b | 888 | END(stub_execve) |
0bd7b798 | 889 | |
1da177e4 LT |
890 | /* |
891 | * sigreturn is special because it needs to restore all registers on return. | |
892 | * This cannot be done with SYSRET, so use the IRET return path instead. | |
0bd7b798 | 893 | */ |
1da177e4 LT |
894 | ENTRY(stub_rt_sigreturn) |
895 | CFI_STARTPROC | |
7effaa88 | 896 | addq $8, %rsp |
e6b04b6b | 897 | PARTIAL_FRAME 0 |
1da177e4 | 898 | SAVE_REST |
1da177e4 LT |
899 | FIXUP_TOP_OF_STACK %r11 |
900 | call sys_rt_sigreturn | |
901 | movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer | |
902 | RESTORE_REST | |
903 | jmp int_ret_from_sys_call | |
904 | CFI_ENDPROC | |
4b787e0b | 905 | END(stub_rt_sigreturn) |
1da177e4 | 906 | |
c5a37394 | 907 | #ifdef CONFIG_X86_X32_ABI |
c5a37394 PA |
908 | ENTRY(stub_x32_rt_sigreturn) |
909 | CFI_STARTPROC | |
910 | addq $8, %rsp | |
911 | PARTIAL_FRAME 0 | |
912 | SAVE_REST | |
c5a37394 PA |
913 | FIXUP_TOP_OF_STACK %r11 |
914 | call sys32_x32_rt_sigreturn | |
915 | movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer | |
916 | RESTORE_REST | |
917 | jmp int_ret_from_sys_call | |
918 | CFI_ENDPROC | |
919 | END(stub_x32_rt_sigreturn) | |
920 | ||
d1a797f3 PA |
921 | ENTRY(stub_x32_execve) |
922 | CFI_STARTPROC | |
923 | addq $8, %rsp | |
924 | PARTIAL_FRAME 0 | |
925 | SAVE_REST | |
926 | FIXUP_TOP_OF_STACK %r11 | |
6783eaa2 | 927 | call compat_sys_execve |
d1a797f3 PA |
928 | RESTORE_TOP_OF_STACK %r11 |
929 | movq %rax,RAX(%rsp) | |
930 | RESTORE_REST | |
931 | jmp int_ret_from_sys_call | |
932 | CFI_ENDPROC | |
933 | END(stub_x32_execve) | |
934 | ||
c5a37394 PA |
935 | #endif |
936 | ||
939b7871 PA |
937 | /* |
938 | * Build the entry stubs and pointer table with some assembler magic. | |
939 | * We pack 7 stubs into a single 32-byte chunk, which will fit in a | |
940 | * single cache line on all modern x86 implementations. | |
941 | */ | |
942 | .section .init.rodata,"a" | |
943 | ENTRY(interrupt) | |
ea714547 | 944 | .section .entry.text |
939b7871 PA |
945 | .p2align 5 |
946 | .p2align CONFIG_X86_L1_CACHE_SHIFT | |
947 | ENTRY(irq_entries_start) | |
948 | INTR_FRAME | |
949 | vector=FIRST_EXTERNAL_VECTOR | |
950 | .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7 | |
951 | .balign 32 | |
952 | .rept 7 | |
953 | .if vector < NR_VECTORS | |
8665596e | 954 | .if vector <> FIRST_EXTERNAL_VECTOR |
939b7871 PA |
955 | CFI_ADJUST_CFA_OFFSET -8 |
956 | .endif | |
df5d1874 | 957 | 1: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */ |
8665596e | 958 | .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6 |
939b7871 PA |
959 | jmp 2f |
960 | .endif | |
961 | .previous | |
962 | .quad 1b | |
ea714547 | 963 | .section .entry.text |
939b7871 PA |
964 | vector=vector+1 |
965 | .endif | |
966 | .endr | |
967 | 2: jmp common_interrupt | |
968 | .endr | |
969 | CFI_ENDPROC | |
970 | END(irq_entries_start) | |
971 | ||
972 | .previous | |
973 | END(interrupt) | |
974 | .previous | |
975 | ||
d99015b1 | 976 | /* |
1da177e4 LT |
977 | * Interrupt entry/exit. |
978 | * | |
979 | * Interrupt entry points save only callee clobbered registers in fast path. | |
d99015b1 AH |
980 | * |
981 | * Entry runs with interrupts off. | |
982 | */ | |
1da177e4 | 983 | |
722024db | 984 | /* 0(%rsp): ~(interrupt number) */ |
1da177e4 | 985 | .macro interrupt func |
625dbc3b FW |
986 | /* reserve pt_regs for scratch regs and rbp */ |
987 | subq $ORIG_RAX-RBP, %rsp | |
988 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP | |
1871853f | 989 | SAVE_ARGS_IRQ |
1da177e4 LT |
990 | call \func |
991 | .endm | |
992 | ||
8222d718 MH |
993 | /* |
994 | * Interrupt entry/exit should be protected against kprobes | |
995 | */ | |
996 | .pushsection .kprobes.text, "ax" | |
722024db AH |
997 | /* |
998 | * The interrupt stubs push (~vector+0x80) onto the stack and | |
999 | * then jump to common_interrupt. | |
1000 | */ | |
939b7871 PA |
1001 | .p2align CONFIG_X86_L1_CACHE_SHIFT |
1002 | common_interrupt: | |
7effaa88 | 1003 | XCPT_FRAME |
ee4eb87b | 1004 | ASM_CLAC |
722024db | 1005 | addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */ |
1da177e4 | 1006 | interrupt do_IRQ |
3d1e42a7 | 1007 | /* 0(%rsp): old_rsp-ARGOFFSET */ |
7effaa88 | 1008 | ret_from_intr: |
72fe4858 | 1009 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 1010 | TRACE_IRQS_OFF |
56895530 | 1011 | decl PER_CPU_VAR(irq_count) |
625dbc3b | 1012 | |
a2bbe750 FW |
1013 | /* Restore saved previous stack */ |
1014 | popq %rsi | |
928282e4 | 1015 | CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */ |
eab9e613 | 1016 | leaq ARGOFFSET-RBP(%rsi), %rsp |
7effaa88 | 1017 | CFI_DEF_CFA_REGISTER rsp |
eab9e613 | 1018 | CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET |
625dbc3b | 1019 | |
7effaa88 | 1020 | exit_intr: |
1da177e4 LT |
1021 | GET_THREAD_INFO(%rcx) |
1022 | testl $3,CS-ARGOFFSET(%rsp) | |
1023 | je retint_kernel | |
0bd7b798 | 1024 | |
1da177e4 LT |
1025 | /* Interrupt came from user space */ |
1026 | /* | |
1027 | * Has a correct top of stack, but a partial stack frame | |
1028 | * %rcx: thread info. Interrupts off. | |
0bd7b798 | 1029 | */ |
1da177e4 LT |
1030 | retint_with_reschedule: |
1031 | movl $_TIF_WORK_MASK,%edi | |
7effaa88 | 1032 | retint_check: |
10cd706d | 1033 | LOCKDEP_SYS_EXIT_IRQ |
26ccb8a7 | 1034 | movl TI_flags(%rcx),%edx |
1da177e4 | 1035 | andl %edi,%edx |
7effaa88 | 1036 | CFI_REMEMBER_STATE |
1da177e4 | 1037 | jnz retint_careful |
10cd706d PZ |
1038 | |
1039 | retint_swapgs: /* return to user-space */ | |
2601e64d IM |
1040 | /* |
1041 | * The iretq could re-enable interrupts: | |
1042 | */ | |
72fe4858 | 1043 | DISABLE_INTERRUPTS(CLBR_ANY) |
2601e64d | 1044 | TRACE_IRQS_IRETQ |
72fe4858 | 1045 | SWAPGS |
2601e64d IM |
1046 | jmp restore_args |
1047 | ||
10cd706d | 1048 | retint_restore_args: /* return to kernel space */ |
72fe4858 | 1049 | DISABLE_INTERRUPTS(CLBR_ANY) |
2601e64d IM |
1050 | /* |
1051 | * The iretq could re-enable interrupts: | |
1052 | */ | |
1053 | TRACE_IRQS_IRETQ | |
1054 | restore_args: | |
838feb47 | 1055 | RESTORE_ARGS 1,8,1 |
3701d863 | 1056 | |
f7f3d791 | 1057 | irq_return: |
72fe4858 | 1058 | INTERRUPT_RETURN |
d7abc0fa | 1059 | _ASM_EXTABLE(irq_return, bad_iret) |
3701d863 IM |
1060 | |
1061 | #ifdef CONFIG_PARAVIRT | |
72fe4858 | 1062 | ENTRY(native_iret) |
1da177e4 | 1063 | iretq |
d7abc0fa | 1064 | _ASM_EXTABLE(native_iret, bad_iret) |
3701d863 IM |
1065 | #endif |
1066 | ||
1da177e4 | 1067 | .section .fixup,"ax" |
1da177e4 | 1068 | bad_iret: |
3aa4b37d RM |
1069 | /* |
1070 | * The iret traps when the %cs or %ss being restored is bogus. | |
1071 | * We've lost the original trap vector and error code. | |
1072 | * #GPF is the most likely one to get for an invalid selector. | |
1073 | * So pretend we completed the iret and took the #GPF in user mode. | |
1074 | * | |
1075 | * We are now running with the kernel GS after exception recovery. | |
1076 | * But error_entry expects us to have user GS to match the user %cs, | |
1077 | * so swap back. | |
1078 | */ | |
1079 | pushq $0 | |
1080 | ||
1081 | SWAPGS | |
1082 | jmp general_protection | |
1083 | ||
72fe4858 GOC |
1084 | .previous |
1085 | ||
7effaa88 | 1086 | /* edi: workmask, edx: work */ |
1da177e4 | 1087 | retint_careful: |
7effaa88 | 1088 | CFI_RESTORE_STATE |
1da177e4 LT |
1089 | bt $TIF_NEED_RESCHED,%edx |
1090 | jnc retint_signal | |
2601e64d | 1091 | TRACE_IRQS_ON |
72fe4858 | 1092 | ENABLE_INTERRUPTS(CLBR_NONE) |
df5d1874 | 1093 | pushq_cfi %rdi |
0430499c | 1094 | SCHEDULE_USER |
df5d1874 | 1095 | popq_cfi %rdi |
1da177e4 | 1096 | GET_THREAD_INFO(%rcx) |
72fe4858 | 1097 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 1098 | TRACE_IRQS_OFF |
1da177e4 | 1099 | jmp retint_check |
0bd7b798 | 1100 | |
1da177e4 | 1101 | retint_signal: |
8f4d37ec | 1102 | testl $_TIF_DO_NOTIFY_MASK,%edx |
10ffdbb8 | 1103 | jz retint_swapgs |
2601e64d | 1104 | TRACE_IRQS_ON |
72fe4858 | 1105 | ENABLE_INTERRUPTS(CLBR_NONE) |
1da177e4 | 1106 | SAVE_REST |
0bd7b798 | 1107 | movq $-1,ORIG_RAX(%rsp) |
3829ee6b | 1108 | xorl %esi,%esi # oldset |
1da177e4 LT |
1109 | movq %rsp,%rdi # &pt_regs |
1110 | call do_notify_resume | |
1111 | RESTORE_REST | |
72fe4858 | 1112 | DISABLE_INTERRUPTS(CLBR_NONE) |
2601e64d | 1113 | TRACE_IRQS_OFF |
be9e6870 | 1114 | GET_THREAD_INFO(%rcx) |
eca91e78 | 1115 | jmp retint_with_reschedule |
1da177e4 LT |
1116 | |
1117 | #ifdef CONFIG_PREEMPT | |
1118 | /* Returning to kernel space. Check if we need preemption */ | |
1119 | /* rcx: threadinfo. interrupts off. */ | |
b06babac | 1120 | ENTRY(retint_kernel) |
26ccb8a7 | 1121 | cmpl $0,TI_preempt_count(%rcx) |
1da177e4 | 1122 | jnz retint_restore_args |
26ccb8a7 | 1123 | bt $TIF_NEED_RESCHED,TI_flags(%rcx) |
1da177e4 LT |
1124 | jnc retint_restore_args |
1125 | bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */ | |
1126 | jnc retint_restore_args | |
1127 | call preempt_schedule_irq | |
1128 | jmp exit_intr | |
0bd7b798 | 1129 | #endif |
4b787e0b | 1130 | |
1da177e4 | 1131 | CFI_ENDPROC |
4b787e0b | 1132 | END(common_interrupt) |
8222d718 MH |
1133 | /* |
1134 | * End of kprobes section | |
1135 | */ | |
1136 | .popsection | |
0bd7b798 | 1137 | |
1da177e4 LT |
1138 | /* |
1139 | * APIC interrupts. | |
0bd7b798 | 1140 | */ |
cf910e83 | 1141 | .macro apicinterrupt3 num sym do_sym |
322648d1 | 1142 | ENTRY(\sym) |
7effaa88 | 1143 | INTR_FRAME |
ee4eb87b | 1144 | ASM_CLAC |
df5d1874 | 1145 | pushq_cfi $~(\num) |
39e95433 | 1146 | .Lcommon_\sym: |
322648d1 | 1147 | interrupt \do_sym |
1da177e4 LT |
1148 | jmp ret_from_intr |
1149 | CFI_ENDPROC | |
322648d1 AH |
1150 | END(\sym) |
1151 | .endm | |
1da177e4 | 1152 | |
cf910e83 SA |
1153 | #ifdef CONFIG_TRACING |
1154 | #define trace(sym) trace_##sym | |
1155 | #define smp_trace(sym) smp_trace_##sym | |
1156 | ||
1157 | .macro trace_apicinterrupt num sym | |
1158 | apicinterrupt3 \num trace(\sym) smp_trace(\sym) | |
1159 | .endm | |
1160 | #else | |
1161 | .macro trace_apicinterrupt num sym do_sym | |
1162 | .endm | |
1163 | #endif | |
1164 | ||
1165 | .macro apicinterrupt num sym do_sym | |
1166 | apicinterrupt3 \num \sym \do_sym | |
1167 | trace_apicinterrupt \num \sym | |
1168 | .endm | |
1169 | ||
322648d1 | 1170 | #ifdef CONFIG_SMP |
cf910e83 | 1171 | apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR \ |
322648d1 | 1172 | irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt |
cf910e83 | 1173 | apicinterrupt3 REBOOT_VECTOR \ |
4ef702c1 | 1174 | reboot_interrupt smp_reboot_interrupt |
322648d1 | 1175 | #endif |
1da177e4 | 1176 | |
03b48632 | 1177 | #ifdef CONFIG_X86_UV |
cf910e83 | 1178 | apicinterrupt3 UV_BAU_MESSAGE \ |
322648d1 | 1179 | uv_bau_message_intr1 uv_bau_message_interrupt |
03b48632 | 1180 | #endif |
322648d1 AH |
1181 | apicinterrupt LOCAL_TIMER_VECTOR \ |
1182 | apic_timer_interrupt smp_apic_timer_interrupt | |
4a4de9c7 DS |
1183 | apicinterrupt X86_PLATFORM_IPI_VECTOR \ |
1184 | x86_platform_ipi smp_x86_platform_ipi | |
89b831ef | 1185 | |
d78f2664 | 1186 | #ifdef CONFIG_HAVE_KVM |
cf910e83 | 1187 | apicinterrupt3 POSTED_INTR_VECTOR \ |
d78f2664 YZ |
1188 | kvm_posted_intr_ipi smp_kvm_posted_intr_ipi |
1189 | #endif | |
1190 | ||
322648d1 | 1191 | apicinterrupt THRESHOLD_APIC_VECTOR \ |
7856f6cc | 1192 | threshold_interrupt smp_threshold_interrupt |
322648d1 AH |
1193 | apicinterrupt THERMAL_APIC_VECTOR \ |
1194 | thermal_interrupt smp_thermal_interrupt | |
1812924b | 1195 | |
322648d1 AH |
1196 | #ifdef CONFIG_SMP |
1197 | apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \ | |
1198 | call_function_single_interrupt smp_call_function_single_interrupt | |
1199 | apicinterrupt CALL_FUNCTION_VECTOR \ | |
1200 | call_function_interrupt smp_call_function_interrupt | |
1201 | apicinterrupt RESCHEDULE_VECTOR \ | |
1202 | reschedule_interrupt smp_reschedule_interrupt | |
1203 | #endif | |
1da177e4 | 1204 | |
322648d1 AH |
1205 | apicinterrupt ERROR_APIC_VECTOR \ |
1206 | error_interrupt smp_error_interrupt | |
1207 | apicinterrupt SPURIOUS_APIC_VECTOR \ | |
1208 | spurious_interrupt smp_spurious_interrupt | |
0bd7b798 | 1209 | |
e360adbe PZ |
1210 | #ifdef CONFIG_IRQ_WORK |
1211 | apicinterrupt IRQ_WORK_VECTOR \ | |
1212 | irq_work_interrupt smp_irq_work_interrupt | |
241771ef IM |
1213 | #endif |
1214 | ||
1da177e4 LT |
1215 | /* |
1216 | * Exception entry points. | |
0bd7b798 | 1217 | */ |
322648d1 AH |
1218 | .macro zeroentry sym do_sym |
1219 | ENTRY(\sym) | |
7effaa88 | 1220 | INTR_FRAME |
ee4eb87b | 1221 | ASM_CLAC |
fab58420 | 1222 | PARAVIRT_ADJUST_EXCEPTION_FRAME |
14ae22ba | 1223 | pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */ |
b1cccb1b JB |
1224 | subq $ORIG_RAX-R15, %rsp |
1225 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15 | |
d99015b1 | 1226 | call error_entry |
dcd072e2 | 1227 | DEFAULT_FRAME 0 |
d99015b1 AH |
1228 | movq %rsp,%rdi /* pt_regs pointer */ |
1229 | xorl %esi,%esi /* no error code */ | |
322648d1 | 1230 | call \do_sym |
d99015b1 | 1231 | jmp error_exit /* %ebx: no swapgs flag */ |
7effaa88 | 1232 | CFI_ENDPROC |
322648d1 AH |
1233 | END(\sym) |
1234 | .endm | |
1da177e4 | 1235 | |
322648d1 | 1236 | .macro paranoidzeroentry sym do_sym |
ddeb8f21 | 1237 | ENTRY(\sym) |
b8b1d08b | 1238 | INTR_FRAME |
ee4eb87b | 1239 | ASM_CLAC |
b8b1d08b | 1240 | PARAVIRT_ADJUST_EXCEPTION_FRAME |
b1cccb1b JB |
1241 | pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */ |
1242 | subq $ORIG_RAX-R15, %rsp | |
1243 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15 | |
b8b1d08b AH |
1244 | call save_paranoid |
1245 | TRACE_IRQS_OFF | |
1246 | movq %rsp,%rdi /* pt_regs pointer */ | |
1247 | xorl %esi,%esi /* no error code */ | |
322648d1 | 1248 | call \do_sym |
b8b1d08b AH |
1249 | jmp paranoid_exit /* %ebx: no swapgs flag */ |
1250 | CFI_ENDPROC | |
ddeb8f21 | 1251 | END(\sym) |
322648d1 | 1252 | .endm |
b8b1d08b | 1253 | |
c15a5958 | 1254 | #define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8) |
322648d1 | 1255 | .macro paranoidzeroentry_ist sym do_sym ist |
ddeb8f21 | 1256 | ENTRY(\sym) |
9f1e87ea | 1257 | INTR_FRAME |
ee4eb87b | 1258 | ASM_CLAC |
b8b1d08b | 1259 | PARAVIRT_ADJUST_EXCEPTION_FRAME |
b1cccb1b JB |
1260 | pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */ |
1261 | subq $ORIG_RAX-R15, %rsp | |
1262 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15 | |
b8b1d08b | 1263 | call save_paranoid |
5963e317 | 1264 | TRACE_IRQS_OFF_DEBUG |
b8b1d08b AH |
1265 | movq %rsp,%rdi /* pt_regs pointer */ |
1266 | xorl %esi,%esi /* no error code */ | |
c15a5958 | 1267 | subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist) |
322648d1 | 1268 | call \do_sym |
c15a5958 | 1269 | addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist) |
b8b1d08b AH |
1270 | jmp paranoid_exit /* %ebx: no swapgs flag */ |
1271 | CFI_ENDPROC | |
ddeb8f21 | 1272 | END(\sym) |
322648d1 | 1273 | .endm |
b8b1d08b | 1274 | |
ddeb8f21 | 1275 | .macro errorentry sym do_sym |
322648d1 | 1276 | ENTRY(\sym) |
7effaa88 | 1277 | XCPT_FRAME |
ee4eb87b | 1278 | ASM_CLAC |
fab58420 | 1279 | PARAVIRT_ADJUST_EXCEPTION_FRAME |
b1cccb1b JB |
1280 | subq $ORIG_RAX-R15, %rsp |
1281 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15 | |
d99015b1 | 1282 | call error_entry |
dcd072e2 | 1283 | DEFAULT_FRAME 0 |
d99015b1 AH |
1284 | movq %rsp,%rdi /* pt_regs pointer */ |
1285 | movq ORIG_RAX(%rsp),%rsi /* get error code */ | |
1286 | movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */ | |
322648d1 | 1287 | call \do_sym |
d99015b1 | 1288 | jmp error_exit /* %ebx: no swapgs flag */ |
7effaa88 | 1289 | CFI_ENDPROC |
322648d1 | 1290 | END(\sym) |
322648d1 | 1291 | .endm |
1da177e4 LT |
1292 | |
1293 | /* error code is on the stack already */ | |
ddeb8f21 | 1294 | .macro paranoiderrorentry sym do_sym |
322648d1 | 1295 | ENTRY(\sym) |
b8b1d08b | 1296 | XCPT_FRAME |
ee4eb87b | 1297 | ASM_CLAC |
b8b1d08b | 1298 | PARAVIRT_ADJUST_EXCEPTION_FRAME |
b1cccb1b JB |
1299 | subq $ORIG_RAX-R15, %rsp |
1300 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15 | |
e2f6bc25 AH |
1301 | call save_paranoid |
1302 | DEFAULT_FRAME 0 | |
7e61a793 | 1303 | TRACE_IRQS_OFF |
b8b1d08b AH |
1304 | movq %rsp,%rdi /* pt_regs pointer */ |
1305 | movq ORIG_RAX(%rsp),%rsi /* get error code */ | |
1306 | movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */ | |
322648d1 | 1307 | call \do_sym |
b8b1d08b AH |
1308 | jmp paranoid_exit /* %ebx: no swapgs flag */ |
1309 | CFI_ENDPROC | |
322648d1 | 1310 | END(\sym) |
322648d1 AH |
1311 | .endm |
1312 | ||
1313 | zeroentry divide_error do_divide_error | |
322648d1 AH |
1314 | zeroentry overflow do_overflow |
1315 | zeroentry bounds do_bounds | |
1316 | zeroentry invalid_op do_invalid_op | |
1317 | zeroentry device_not_available do_device_not_available | |
ddeb8f21 | 1318 | paranoiderrorentry double_fault do_double_fault |
322648d1 AH |
1319 | zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun |
1320 | errorentry invalid_TSS do_invalid_TSS | |
1321 | errorentry segment_not_present do_segment_not_present | |
322648d1 AH |
1322 | zeroentry spurious_interrupt_bug do_spurious_interrupt_bug |
1323 | zeroentry coprocessor_error do_coprocessor_error | |
1324 | errorentry alignment_check do_alignment_check | |
322648d1 | 1325 | zeroentry simd_coprocessor_error do_simd_coprocessor_error |
5cec93c2 | 1326 | |
2601e64d | 1327 | |
9f1e87ea CG |
1328 | /* Reload gs selector with exception handling */ |
1329 | /* edi: new selector */ | |
9f9d489a | 1330 | ENTRY(native_load_gs_index) |
7effaa88 | 1331 | CFI_STARTPROC |
df5d1874 | 1332 | pushfq_cfi |
b8aa287f | 1333 | DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI) |
9f1e87ea | 1334 | SWAPGS |
0bd7b798 | 1335 | gs_change: |
9f1e87ea | 1336 | movl %edi,%gs |
1da177e4 | 1337 | 2: mfence /* workaround */ |
72fe4858 | 1338 | SWAPGS |
df5d1874 | 1339 | popfq_cfi |
9f1e87ea | 1340 | ret |
7effaa88 | 1341 | CFI_ENDPROC |
6efdcfaf | 1342 | END(native_load_gs_index) |
0bd7b798 | 1343 | |
d7abc0fa | 1344 | _ASM_EXTABLE(gs_change,bad_gs) |
9f1e87ea | 1345 | .section .fixup,"ax" |
1da177e4 | 1346 | /* running with kernelgs */ |
0bd7b798 | 1347 | bad_gs: |
72fe4858 | 1348 | SWAPGS /* switch back to user gs */ |
1da177e4 | 1349 | xorl %eax,%eax |
9f1e87ea CG |
1350 | movl %eax,%gs |
1351 | jmp 2b | |
1352 | .previous | |
0bd7b798 | 1353 | |
2699500b | 1354 | /* Call softirq on interrupt stack. Interrupts are off. */ |
ed6b676c | 1355 | ENTRY(call_softirq) |
7effaa88 | 1356 | CFI_STARTPROC |
df5d1874 | 1357 | pushq_cfi %rbp |
2699500b AK |
1358 | CFI_REL_OFFSET rbp,0 |
1359 | mov %rsp,%rbp | |
1360 | CFI_DEF_CFA_REGISTER rbp | |
56895530 | 1361 | incl PER_CPU_VAR(irq_count) |
26f80bd6 | 1362 | cmove PER_CPU_VAR(irq_stack_ptr),%rsp |
2699500b | 1363 | push %rbp # backlink for old unwinder |
ed6b676c | 1364 | call __do_softirq |
2699500b | 1365 | leaveq |
df5d1874 | 1366 | CFI_RESTORE rbp |
7effaa88 | 1367 | CFI_DEF_CFA_REGISTER rsp |
2699500b | 1368 | CFI_ADJUST_CFA_OFFSET -8 |
56895530 | 1369 | decl PER_CPU_VAR(irq_count) |
ed6b676c | 1370 | ret |
7effaa88 | 1371 | CFI_ENDPROC |
6efdcfaf | 1372 | END(call_softirq) |
75154f40 | 1373 | |
3d75e1b8 | 1374 | #ifdef CONFIG_XEN |
322648d1 | 1375 | zeroentry xen_hypervisor_callback xen_do_hypervisor_callback |
3d75e1b8 JF |
1376 | |
1377 | /* | |
9f1e87ea CG |
1378 | * A note on the "critical region" in our callback handler. |
1379 | * We want to avoid stacking callback handlers due to events occurring | |
1380 | * during handling of the last event. To do this, we keep events disabled | |
1381 | * until we've done all processing. HOWEVER, we must enable events before | |
1382 | * popping the stack frame (can't be done atomically) and so it would still | |
1383 | * be possible to get enough handler activations to overflow the stack. | |
1384 | * Although unlikely, bugs of that kind are hard to track down, so we'd | |
1385 | * like to avoid the possibility. | |
1386 | * So, on entry to the handler we detect whether we interrupted an | |
1387 | * existing activation in its critical region -- if so, we pop the current | |
1388 | * activation and restart the handler using the previous one. | |
1389 | */ | |
3d75e1b8 JF |
1390 | ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs) |
1391 | CFI_STARTPROC | |
9f1e87ea CG |
1392 | /* |
1393 | * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will | |
1394 | * see the correct pointer to the pt_regs | |
1395 | */ | |
3d75e1b8 JF |
1396 | movq %rdi, %rsp # we don't return, adjust the stack frame |
1397 | CFI_ENDPROC | |
dcd072e2 | 1398 | DEFAULT_FRAME |
56895530 | 1399 | 11: incl PER_CPU_VAR(irq_count) |
3d75e1b8 JF |
1400 | movq %rsp,%rbp |
1401 | CFI_DEF_CFA_REGISTER rbp | |
26f80bd6 | 1402 | cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp |
3d75e1b8 JF |
1403 | pushq %rbp # backlink for old unwinder |
1404 | call xen_evtchn_do_upcall | |
1405 | popq %rsp | |
1406 | CFI_DEF_CFA_REGISTER rsp | |
56895530 | 1407 | decl PER_CPU_VAR(irq_count) |
3d75e1b8 JF |
1408 | jmp error_exit |
1409 | CFI_ENDPROC | |
371c394a | 1410 | END(xen_do_hypervisor_callback) |
3d75e1b8 JF |
1411 | |
1412 | /* | |
9f1e87ea CG |
1413 | * Hypervisor uses this for application faults while it executes. |
1414 | * We get here for two reasons: | |
1415 | * 1. Fault while reloading DS, ES, FS or GS | |
1416 | * 2. Fault while executing IRET | |
1417 | * Category 1 we do not need to fix up as Xen has already reloaded all segment | |
1418 | * registers that could be reloaded and zeroed the others. | |
1419 | * Category 2 we fix up by killing the current process. We cannot use the | |
1420 | * normal Linux return path in this case because if we use the IRET hypercall | |
1421 | * to pop the stack frame we end up in an infinite loop of failsafe callbacks. | |
1422 | * We distinguish between categories by comparing each saved segment register | |
1423 | * with its current contents: any discrepancy means we in category 1. | |
1424 | */ | |
3d75e1b8 | 1425 | ENTRY(xen_failsafe_callback) |
dcd072e2 AH |
1426 | INTR_FRAME 1 (6*8) |
1427 | /*CFI_REL_OFFSET gs,GS*/ | |
1428 | /*CFI_REL_OFFSET fs,FS*/ | |
1429 | /*CFI_REL_OFFSET es,ES*/ | |
1430 | /*CFI_REL_OFFSET ds,DS*/ | |
1431 | CFI_REL_OFFSET r11,8 | |
1432 | CFI_REL_OFFSET rcx,0 | |
3d75e1b8 JF |
1433 | movw %ds,%cx |
1434 | cmpw %cx,0x10(%rsp) | |
1435 | CFI_REMEMBER_STATE | |
1436 | jne 1f | |
1437 | movw %es,%cx | |
1438 | cmpw %cx,0x18(%rsp) | |
1439 | jne 1f | |
1440 | movw %fs,%cx | |
1441 | cmpw %cx,0x20(%rsp) | |
1442 | jne 1f | |
1443 | movw %gs,%cx | |
1444 | cmpw %cx,0x28(%rsp) | |
1445 | jne 1f | |
1446 | /* All segments match their saved values => Category 2 (Bad IRET). */ | |
1447 | movq (%rsp),%rcx | |
1448 | CFI_RESTORE rcx | |
1449 | movq 8(%rsp),%r11 | |
1450 | CFI_RESTORE r11 | |
1451 | addq $0x30,%rsp | |
1452 | CFI_ADJUST_CFA_OFFSET -0x30 | |
14ae22ba IM |
1453 | pushq_cfi $0 /* RIP */ |
1454 | pushq_cfi %r11 | |
1455 | pushq_cfi %rcx | |
4a5c3e77 | 1456 | jmp general_protection |
3d75e1b8 JF |
1457 | CFI_RESTORE_STATE |
1458 | 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */ | |
1459 | movq (%rsp),%rcx | |
1460 | CFI_RESTORE rcx | |
1461 | movq 8(%rsp),%r11 | |
1462 | CFI_RESTORE r11 | |
1463 | addq $0x30,%rsp | |
1464 | CFI_ADJUST_CFA_OFFSET -0x30 | |
a349e23d | 1465 | pushq_cfi $-1 /* orig_ax = -1 => not a system call */ |
3d75e1b8 JF |
1466 | SAVE_ALL |
1467 | jmp error_exit | |
1468 | CFI_ENDPROC | |
3d75e1b8 JF |
1469 | END(xen_failsafe_callback) |
1470 | ||
cf910e83 | 1471 | apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \ |
38e20b07 SY |
1472 | xen_hvm_callback_vector xen_evtchn_do_upcall |
1473 | ||
3d75e1b8 | 1474 | #endif /* CONFIG_XEN */ |
ddeb8f21 | 1475 | |
bc2b0331 | 1476 | #if IS_ENABLED(CONFIG_HYPERV) |
cf910e83 | 1477 | apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \ |
bc2b0331 S |
1478 | hyperv_callback_vector hyperv_vector_handler |
1479 | #endif /* CONFIG_HYPERV */ | |
1480 | ||
ddeb8f21 AH |
1481 | /* |
1482 | * Some functions should be protected against kprobes | |
1483 | */ | |
1484 | .pushsection .kprobes.text, "ax" | |
1485 | ||
1486 | paranoidzeroentry_ist debug do_debug DEBUG_STACK | |
1487 | paranoidzeroentry_ist int3 do_int3 DEBUG_STACK | |
1488 | paranoiderrorentry stack_segment do_stack_segment | |
6cac5a92 JF |
1489 | #ifdef CONFIG_XEN |
1490 | zeroentry xen_debug do_debug | |
1491 | zeroentry xen_int3 do_int3 | |
1492 | errorentry xen_stack_segment do_stack_segment | |
1493 | #endif | |
ddeb8f21 AH |
1494 | errorentry general_protection do_general_protection |
1495 | errorentry page_fault do_page_fault | |
631bc487 GN |
1496 | #ifdef CONFIG_KVM_GUEST |
1497 | errorentry async_page_fault do_async_page_fault | |
1498 | #endif | |
ddeb8f21 | 1499 | #ifdef CONFIG_X86_MCE |
5d727926 | 1500 | paranoidzeroentry machine_check *machine_check_vector(%rip) |
ddeb8f21 AH |
1501 | #endif |
1502 | ||
1503 | /* | |
9f1e87ea CG |
1504 | * "Paranoid" exit path from exception stack. |
1505 | * Paranoid because this is used by NMIs and cannot take | |
ddeb8f21 AH |
1506 | * any kernel state for granted. |
1507 | * We don't do kernel preemption checks here, because only | |
1508 | * NMI should be common and it does not enable IRQs and | |
1509 | * cannot get reschedule ticks. | |
1510 | * | |
1511 | * "trace" is 0 for the NMI handler only, because irq-tracing | |
1512 | * is fundamentally NMI-unsafe. (we cannot change the soft and | |
1513 | * hard flags at once, atomically) | |
1514 | */ | |
1515 | ||
1516 | /* ebx: no swapgs flag */ | |
1517 | ENTRY(paranoid_exit) | |
1f130a78 | 1518 | DEFAULT_FRAME |
ddeb8f21 | 1519 | DISABLE_INTERRUPTS(CLBR_NONE) |
5963e317 | 1520 | TRACE_IRQS_OFF_DEBUG |
ddeb8f21 AH |
1521 | testl %ebx,%ebx /* swapgs needed? */ |
1522 | jnz paranoid_restore | |
1523 | testl $3,CS(%rsp) | |
1524 | jnz paranoid_userspace | |
1525 | paranoid_swapgs: | |
1526 | TRACE_IRQS_IRETQ 0 | |
1527 | SWAPGS_UNSAFE_STACK | |
0300e7f1 SR |
1528 | RESTORE_ALL 8 |
1529 | jmp irq_return | |
ddeb8f21 | 1530 | paranoid_restore: |
5963e317 | 1531 | TRACE_IRQS_IRETQ_DEBUG 0 |
ddeb8f21 AH |
1532 | RESTORE_ALL 8 |
1533 | jmp irq_return | |
1534 | paranoid_userspace: | |
1535 | GET_THREAD_INFO(%rcx) | |
1536 | movl TI_flags(%rcx),%ebx | |
1537 | andl $_TIF_WORK_MASK,%ebx | |
1538 | jz paranoid_swapgs | |
1539 | movq %rsp,%rdi /* &pt_regs */ | |
1540 | call sync_regs | |
1541 | movq %rax,%rsp /* switch stack for scheduling */ | |
1542 | testl $_TIF_NEED_RESCHED,%ebx | |
1543 | jnz paranoid_schedule | |
1544 | movl %ebx,%edx /* arg3: thread flags */ | |
1545 | TRACE_IRQS_ON | |
1546 | ENABLE_INTERRUPTS(CLBR_NONE) | |
1547 | xorl %esi,%esi /* arg2: oldset */ | |
1548 | movq %rsp,%rdi /* arg1: &pt_regs */ | |
1549 | call do_notify_resume | |
1550 | DISABLE_INTERRUPTS(CLBR_NONE) | |
1551 | TRACE_IRQS_OFF | |
1552 | jmp paranoid_userspace | |
1553 | paranoid_schedule: | |
1554 | TRACE_IRQS_ON | |
1555 | ENABLE_INTERRUPTS(CLBR_ANY) | |
0430499c | 1556 | SCHEDULE_USER |
ddeb8f21 AH |
1557 | DISABLE_INTERRUPTS(CLBR_ANY) |
1558 | TRACE_IRQS_OFF | |
1559 | jmp paranoid_userspace | |
1560 | CFI_ENDPROC | |
1561 | END(paranoid_exit) | |
1562 | ||
1563 | /* | |
1564 | * Exception entry point. This expects an error code/orig_rax on the stack. | |
1565 | * returns in "no swapgs flag" in %ebx. | |
1566 | */ | |
1567 | ENTRY(error_entry) | |
1568 | XCPT_FRAME | |
1569 | CFI_ADJUST_CFA_OFFSET 15*8 | |
1570 | /* oldrax contains error code */ | |
1571 | cld | |
1572 | movq_cfi rdi, RDI+8 | |
1573 | movq_cfi rsi, RSI+8 | |
1574 | movq_cfi rdx, RDX+8 | |
1575 | movq_cfi rcx, RCX+8 | |
1576 | movq_cfi rax, RAX+8 | |
1577 | movq_cfi r8, R8+8 | |
1578 | movq_cfi r9, R9+8 | |
1579 | movq_cfi r10, R10+8 | |
1580 | movq_cfi r11, R11+8 | |
1581 | movq_cfi rbx, RBX+8 | |
1582 | movq_cfi rbp, RBP+8 | |
1583 | movq_cfi r12, R12+8 | |
1584 | movq_cfi r13, R13+8 | |
1585 | movq_cfi r14, R14+8 | |
1586 | movq_cfi r15, R15+8 | |
1587 | xorl %ebx,%ebx | |
1588 | testl $3,CS+8(%rsp) | |
1589 | je error_kernelspace | |
1590 | error_swapgs: | |
1591 | SWAPGS | |
1592 | error_sti: | |
1593 | TRACE_IRQS_OFF | |
1594 | ret | |
ddeb8f21 AH |
1595 | |
1596 | /* | |
1597 | * There are two places in the kernel that can potentially fault with | |
1598 | * usergs. Handle them here. The exception handlers after iret run with | |
1599 | * kernel gs again, so don't set the user space flag. B stepping K8s | |
1600 | * sometimes report an truncated RIP for IRET exceptions returning to | |
1601 | * compat mode. Check for these here too. | |
1602 | */ | |
1603 | error_kernelspace: | |
1604 | incl %ebx | |
1605 | leaq irq_return(%rip),%rcx | |
1606 | cmpq %rcx,RIP+8(%rsp) | |
1607 | je error_swapgs | |
ae24ffe5 BG |
1608 | movl %ecx,%eax /* zero extend */ |
1609 | cmpq %rax,RIP+8(%rsp) | |
1610 | je bstep_iret | |
ddeb8f21 | 1611 | cmpq $gs_change,RIP+8(%rsp) |
9f1e87ea | 1612 | je error_swapgs |
ddeb8f21 | 1613 | jmp error_sti |
ae24ffe5 BG |
1614 | |
1615 | bstep_iret: | |
1616 | /* Fix truncated RIP */ | |
1617 | movq %rcx,RIP+8(%rsp) | |
97829de5 | 1618 | jmp error_swapgs |
e6b04b6b | 1619 | CFI_ENDPROC |
ddeb8f21 AH |
1620 | END(error_entry) |
1621 | ||
1622 | ||
1623 | /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */ | |
1624 | ENTRY(error_exit) | |
1625 | DEFAULT_FRAME | |
1626 | movl %ebx,%eax | |
1627 | RESTORE_REST | |
1628 | DISABLE_INTERRUPTS(CLBR_NONE) | |
1629 | TRACE_IRQS_OFF | |
1630 | GET_THREAD_INFO(%rcx) | |
1631 | testl %eax,%eax | |
1632 | jne retint_kernel | |
1633 | LOCKDEP_SYS_EXIT_IRQ | |
1634 | movl TI_flags(%rcx),%edx | |
1635 | movl $_TIF_WORK_MASK,%edi | |
1636 | andl %edi,%edx | |
1637 | jnz retint_careful | |
1638 | jmp retint_swapgs | |
1639 | CFI_ENDPROC | |
1640 | END(error_exit) | |
1641 | ||
3f3c8b8c SR |
1642 | /* |
1643 | * Test if a given stack is an NMI stack or not. | |
1644 | */ | |
1645 | .macro test_in_nmi reg stack nmi_ret normal_ret | |
1646 | cmpq %\reg, \stack | |
1647 | ja \normal_ret | |
1648 | subq $EXCEPTION_STKSZ, %\reg | |
1649 | cmpq %\reg, \stack | |
1650 | jb \normal_ret | |
1651 | jmp \nmi_ret | |
1652 | .endm | |
ddeb8f21 AH |
1653 | |
1654 | /* runs on exception stack */ | |
1655 | ENTRY(nmi) | |
1656 | INTR_FRAME | |
1657 | PARAVIRT_ADJUST_EXCEPTION_FRAME | |
3f3c8b8c SR |
1658 | /* |
1659 | * We allow breakpoints in NMIs. If a breakpoint occurs, then | |
1660 | * the iretq it performs will take us out of NMI context. | |
1661 | * This means that we can have nested NMIs where the next | |
1662 | * NMI is using the top of the stack of the previous NMI. We | |
1663 | * can't let it execute because the nested NMI will corrupt the | |
1664 | * stack of the previous NMI. NMI handlers are not re-entrant | |
1665 | * anyway. | |
1666 | * | |
1667 | * To handle this case we do the following: | |
1668 | * Check the a special location on the stack that contains | |
1669 | * a variable that is set when NMIs are executing. | |
1670 | * The interrupted task's stack is also checked to see if it | |
1671 | * is an NMI stack. | |
1672 | * If the variable is not set and the stack is not the NMI | |
1673 | * stack then: | |
1674 | * o Set the special variable on the stack | |
1675 | * o Copy the interrupt frame into a "saved" location on the stack | |
1676 | * o Copy the interrupt frame into a "copy" location on the stack | |
1677 | * o Continue processing the NMI | |
1678 | * If the variable is set or the previous stack is the NMI stack: | |
1679 | * o Modify the "copy" location to jump to the repeate_nmi | |
1680 | * o return back to the first NMI | |
1681 | * | |
1682 | * Now on exit of the first NMI, we first clear the stack variable | |
1683 | * The NMI stack will tell any nested NMIs at that point that it is | |
1684 | * nested. Then we pop the stack normally with iret, and if there was | |
1685 | * a nested NMI that updated the copy interrupt stack frame, a | |
1686 | * jump will be made to the repeat_nmi code that will handle the second | |
1687 | * NMI. | |
1688 | */ | |
1689 | ||
1690 | /* Use %rdx as out temp variable throughout */ | |
1691 | pushq_cfi %rdx | |
62610913 | 1692 | CFI_REL_OFFSET rdx, 0 |
3f3c8b8c | 1693 | |
45d5a168 SR |
1694 | /* |
1695 | * If %cs was not the kernel segment, then the NMI triggered in user | |
1696 | * space, which means it is definitely not nested. | |
1697 | */ | |
a38449ef | 1698 | cmpl $__KERNEL_CS, 16(%rsp) |
45d5a168 SR |
1699 | jne first_nmi |
1700 | ||
3f3c8b8c SR |
1701 | /* |
1702 | * Check the special variable on the stack to see if NMIs are | |
1703 | * executing. | |
1704 | */ | |
a38449ef | 1705 | cmpl $1, -8(%rsp) |
3f3c8b8c SR |
1706 | je nested_nmi |
1707 | ||
1708 | /* | |
1709 | * Now test if the previous stack was an NMI stack. | |
1710 | * We need the double check. We check the NMI stack to satisfy the | |
1711 | * race when the first NMI clears the variable before returning. | |
1712 | * We check the variable because the first NMI could be in a | |
1713 | * breakpoint routine using a breakpoint stack. | |
1714 | */ | |
1715 | lea 6*8(%rsp), %rdx | |
1716 | test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi | |
62610913 | 1717 | CFI_REMEMBER_STATE |
3f3c8b8c SR |
1718 | |
1719 | nested_nmi: | |
1720 | /* | |
1721 | * Do nothing if we interrupted the fixup in repeat_nmi. | |
1722 | * It's about to repeat the NMI handler, so we are fine | |
1723 | * with ignoring this one. | |
1724 | */ | |
1725 | movq $repeat_nmi, %rdx | |
1726 | cmpq 8(%rsp), %rdx | |
1727 | ja 1f | |
1728 | movq $end_repeat_nmi, %rdx | |
1729 | cmpq 8(%rsp), %rdx | |
1730 | ja nested_nmi_out | |
1731 | ||
1732 | 1: | |
1733 | /* Set up the interrupted NMIs stack to jump to repeat_nmi */ | |
28696f43 | 1734 | leaq -1*8(%rsp), %rdx |
3f3c8b8c | 1735 | movq %rdx, %rsp |
28696f43 SQ |
1736 | CFI_ADJUST_CFA_OFFSET 1*8 |
1737 | leaq -10*8(%rsp), %rdx | |
3f3c8b8c SR |
1738 | pushq_cfi $__KERNEL_DS |
1739 | pushq_cfi %rdx | |
1740 | pushfq_cfi | |
1741 | pushq_cfi $__KERNEL_CS | |
1742 | pushq_cfi $repeat_nmi | |
1743 | ||
1744 | /* Put stack back */ | |
28696f43 SQ |
1745 | addq $(6*8), %rsp |
1746 | CFI_ADJUST_CFA_OFFSET -6*8 | |
3f3c8b8c SR |
1747 | |
1748 | nested_nmi_out: | |
1749 | popq_cfi %rdx | |
62610913 | 1750 | CFI_RESTORE rdx |
3f3c8b8c SR |
1751 | |
1752 | /* No need to check faults here */ | |
1753 | INTERRUPT_RETURN | |
1754 | ||
62610913 | 1755 | CFI_RESTORE_STATE |
3f3c8b8c SR |
1756 | first_nmi: |
1757 | /* | |
1758 | * Because nested NMIs will use the pushed location that we | |
1759 | * stored in rdx, we must keep that space available. | |
1760 | * Here's what our stack frame will look like: | |
1761 | * +-------------------------+ | |
1762 | * | original SS | | |
1763 | * | original Return RSP | | |
1764 | * | original RFLAGS | | |
1765 | * | original CS | | |
1766 | * | original RIP | | |
1767 | * +-------------------------+ | |
1768 | * | temp storage for rdx | | |
1769 | * +-------------------------+ | |
1770 | * | NMI executing variable | | |
1771 | * +-------------------------+ | |
3f3c8b8c SR |
1772 | * | copied SS | |
1773 | * | copied Return RSP | | |
1774 | * | copied RFLAGS | | |
1775 | * | copied CS | | |
1776 | * | copied RIP | | |
1777 | * +-------------------------+ | |
28696f43 SQ |
1778 | * | Saved SS | |
1779 | * | Saved Return RSP | | |
1780 | * | Saved RFLAGS | | |
1781 | * | Saved CS | | |
1782 | * | Saved RIP | | |
1783 | * +-------------------------+ | |
3f3c8b8c SR |
1784 | * | pt_regs | |
1785 | * +-------------------------+ | |
1786 | * | |
79fb4ad6 SR |
1787 | * The saved stack frame is used to fix up the copied stack frame |
1788 | * that a nested NMI may change to make the interrupted NMI iret jump | |
1789 | * to the repeat_nmi. The original stack frame and the temp storage | |
3f3c8b8c SR |
1790 | * is also used by nested NMIs and can not be trusted on exit. |
1791 | */ | |
79fb4ad6 | 1792 | /* Do not pop rdx, nested NMIs will corrupt that part of the stack */ |
62610913 JB |
1793 | movq (%rsp), %rdx |
1794 | CFI_RESTORE rdx | |
1795 | ||
3f3c8b8c SR |
1796 | /* Set the NMI executing variable on the stack. */ |
1797 | pushq_cfi $1 | |
1798 | ||
28696f43 SQ |
1799 | /* |
1800 | * Leave room for the "copied" frame | |
1801 | */ | |
1802 | subq $(5*8), %rsp | |
444723dc | 1803 | CFI_ADJUST_CFA_OFFSET 5*8 |
28696f43 | 1804 | |
3f3c8b8c SR |
1805 | /* Copy the stack frame to the Saved frame */ |
1806 | .rept 5 | |
28696f43 | 1807 | pushq_cfi 11*8(%rsp) |
3f3c8b8c | 1808 | .endr |
62610913 JB |
1809 | CFI_DEF_CFA_OFFSET SS+8-RIP |
1810 | ||
79fb4ad6 SR |
1811 | /* Everything up to here is safe from nested NMIs */ |
1812 | ||
62610913 JB |
1813 | /* |
1814 | * If there was a nested NMI, the first NMI's iret will return | |
1815 | * here. But NMIs are still enabled and we can take another | |
1816 | * nested NMI. The nested NMI checks the interrupted RIP to see | |
1817 | * if it is between repeat_nmi and end_repeat_nmi, and if so | |
1818 | * it will just return, as we are about to repeat an NMI anyway. | |
1819 | * This makes it safe to copy to the stack frame that a nested | |
1820 | * NMI will update. | |
1821 | */ | |
1822 | repeat_nmi: | |
1823 | /* | |
1824 | * Update the stack variable to say we are still in NMI (the update | |
1825 | * is benign for the non-repeat case, where 1 was pushed just above | |
1826 | * to this very stack slot). | |
1827 | */ | |
28696f43 | 1828 | movq $1, 10*8(%rsp) |
3f3c8b8c SR |
1829 | |
1830 | /* Make another copy, this one may be modified by nested NMIs */ | |
28696f43 SQ |
1831 | addq $(10*8), %rsp |
1832 | CFI_ADJUST_CFA_OFFSET -10*8 | |
3f3c8b8c | 1833 | .rept 5 |
28696f43 | 1834 | pushq_cfi -6*8(%rsp) |
3f3c8b8c | 1835 | .endr |
28696f43 | 1836 | subq $(5*8), %rsp |
62610913 JB |
1837 | CFI_DEF_CFA_OFFSET SS+8-RIP |
1838 | end_repeat_nmi: | |
3f3c8b8c SR |
1839 | |
1840 | /* | |
1841 | * Everything below this point can be preempted by a nested | |
79fb4ad6 SR |
1842 | * NMI if the first NMI took an exception and reset our iret stack |
1843 | * so that we repeat another NMI. | |
3f3c8b8c | 1844 | */ |
1fd466ef | 1845 | pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */ |
b1cccb1b JB |
1846 | subq $ORIG_RAX-R15, %rsp |
1847 | CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15 | |
1fd466ef SR |
1848 | /* |
1849 | * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit | |
1850 | * as we should not be calling schedule in NMI context. | |
1851 | * Even with normal interrupts enabled. An NMI should not be | |
1852 | * setting NEED_RESCHED or anything that normal interrupts and | |
1853 | * exceptions might do. | |
1854 | */ | |
ddeb8f21 AH |
1855 | call save_paranoid |
1856 | DEFAULT_FRAME 0 | |
7fbb98c5 SR |
1857 | |
1858 | /* | |
1859 | * Save off the CR2 register. If we take a page fault in the NMI then | |
1860 | * it could corrupt the CR2 value. If the NMI preempts a page fault | |
1861 | * handler before it was able to read the CR2 register, and then the | |
1862 | * NMI itself takes a page fault, the page fault that was preempted | |
1863 | * will read the information from the NMI page fault and not the | |
1864 | * origin fault. Save it off and restore it if it changes. | |
1865 | * Use the r12 callee-saved register. | |
1866 | */ | |
1867 | movq %cr2, %r12 | |
1868 | ||
ddeb8f21 AH |
1869 | /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */ |
1870 | movq %rsp,%rdi | |
1871 | movq $-1,%rsi | |
1872 | call do_nmi | |
7fbb98c5 SR |
1873 | |
1874 | /* Did the NMI take a page fault? Restore cr2 if it did */ | |
1875 | movq %cr2, %rcx | |
1876 | cmpq %rcx, %r12 | |
1877 | je 1f | |
1878 | movq %r12, %cr2 | |
1879 | 1: | |
1880 | ||
ddeb8f21 AH |
1881 | testl %ebx,%ebx /* swapgs needed? */ |
1882 | jnz nmi_restore | |
ddeb8f21 AH |
1883 | nmi_swapgs: |
1884 | SWAPGS_UNSAFE_STACK | |
1885 | nmi_restore: | |
444723dc JB |
1886 | /* Pop the extra iret frame at once */ |
1887 | RESTORE_ALL 6*8 | |
28696f43 | 1888 | |
3f3c8b8c | 1889 | /* Clear the NMI executing stack variable */ |
28696f43 | 1890 | movq $0, 5*8(%rsp) |
ddeb8f21 | 1891 | jmp irq_return |
9f1e87ea | 1892 | CFI_ENDPROC |
ddeb8f21 AH |
1893 | END(nmi) |
1894 | ||
1895 | ENTRY(ignore_sysret) | |
1896 | CFI_STARTPROC | |
1897 | mov $-ENOSYS,%eax | |
1898 | sysret | |
1899 | CFI_ENDPROC | |
1900 | END(ignore_sysret) | |
1901 | ||
1902 | /* | |
1903 | * End of kprobes section | |
1904 | */ | |
1905 | .popsection |