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> | |
7 | * | |
8 | * $Id$ | |
9 | */ | |
10 | ||
11 | /* | |
12 | * entry.S contains the system-call and fault low-level handling routines. | |
13 | * | |
14 | * NOTE: This code handles signal-recognition, which happens every time | |
15 | * after an interrupt and after each system call. | |
16 | * | |
17 | * Normal syscalls and interrupts don't save a full stack frame, this is | |
18 | * only done for syscall tracing, signals or fork/exec et.al. | |
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. | |
23 | * - partial stack frame: partially saved registers upto R11. | |
24 | * - full stack frame: Like partial stack frame, but all register saved. | |
25 | * | |
26 | * TODO: | |
27 | * - schedule it carefully for the final hardware. | |
28 | */ | |
29 | ||
30 | #define ASSEMBLY 1 | |
31 | #include <linux/config.h> | |
32 | #include <linux/linkage.h> | |
33 | #include <asm/segment.h> | |
34 | #include <asm/smp.h> | |
35 | #include <asm/cache.h> | |
36 | #include <asm/errno.h> | |
37 | #include <asm/dwarf2.h> | |
38 | #include <asm/calling.h> | |
39 | #include <asm/offset.h> | |
40 | #include <asm/msr.h> | |
41 | #include <asm/unistd.h> | |
42 | #include <asm/thread_info.h> | |
43 | #include <asm/hw_irq.h> | |
44 | ||
45 | .code64 | |
46 | ||
dc37db4d | 47 | #ifndef CONFIG_PREEMPT |
1da177e4 LT |
48 | #define retint_kernel retint_restore_args |
49 | #endif | |
50 | ||
51 | /* | |
52 | * C code is not supposed to know about undefined top of stack. Every time | |
53 | * a C function with an pt_regs argument is called from the SYSCALL based | |
54 | * fast path FIXUP_TOP_OF_STACK is needed. | |
55 | * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs | |
56 | * manipulation. | |
57 | */ | |
58 | ||
59 | /* %rsp:at FRAMEEND */ | |
60 | .macro FIXUP_TOP_OF_STACK tmp | |
61 | movq %gs:pda_oldrsp,\tmp | |
62 | movq \tmp,RSP(%rsp) | |
63 | movq $__USER_DS,SS(%rsp) | |
64 | movq $__USER_CS,CS(%rsp) | |
65 | movq $-1,RCX(%rsp) | |
66 | movq R11(%rsp),\tmp /* get eflags */ | |
67 | movq \tmp,EFLAGS(%rsp) | |
68 | .endm | |
69 | ||
70 | .macro RESTORE_TOP_OF_STACK tmp,offset=0 | |
71 | movq RSP-\offset(%rsp),\tmp | |
72 | movq \tmp,%gs:pda_oldrsp | |
73 | movq EFLAGS-\offset(%rsp),\tmp | |
74 | movq \tmp,R11-\offset(%rsp) | |
75 | .endm | |
76 | ||
77 | .macro FAKE_STACK_FRAME child_rip | |
78 | /* push in order ss, rsp, eflags, cs, rip */ | |
3829ee6b | 79 | xorl %eax, %eax |
1da177e4 LT |
80 | pushq %rax /* ss */ |
81 | CFI_ADJUST_CFA_OFFSET 8 | |
82 | pushq %rax /* rsp */ | |
83 | CFI_ADJUST_CFA_OFFSET 8 | |
84 | CFI_OFFSET rip,0 | |
85 | pushq $(1<<9) /* eflags - interrupts on */ | |
86 | CFI_ADJUST_CFA_OFFSET 8 | |
87 | pushq $__KERNEL_CS /* cs */ | |
88 | CFI_ADJUST_CFA_OFFSET 8 | |
89 | pushq \child_rip /* rip */ | |
90 | CFI_ADJUST_CFA_OFFSET 8 | |
91 | CFI_OFFSET rip,0 | |
92 | pushq %rax /* orig rax */ | |
93 | CFI_ADJUST_CFA_OFFSET 8 | |
94 | .endm | |
95 | ||
96 | .macro UNFAKE_STACK_FRAME | |
97 | addq $8*6, %rsp | |
98 | CFI_ADJUST_CFA_OFFSET -(6*8) | |
99 | .endm | |
100 | ||
101 | .macro CFI_DEFAULT_STACK | |
102 | CFI_ADJUST_CFA_OFFSET (SS) | |
103 | CFI_OFFSET r15,R15-SS | |
104 | CFI_OFFSET r14,R14-SS | |
105 | CFI_OFFSET r13,R13-SS | |
106 | CFI_OFFSET r12,R12-SS | |
107 | CFI_OFFSET rbp,RBP-SS | |
108 | CFI_OFFSET rbx,RBX-SS | |
109 | CFI_OFFSET r11,R11-SS | |
110 | CFI_OFFSET r10,R10-SS | |
111 | CFI_OFFSET r9,R9-SS | |
112 | CFI_OFFSET r8,R8-SS | |
113 | CFI_OFFSET rax,RAX-SS | |
114 | CFI_OFFSET rcx,RCX-SS | |
115 | CFI_OFFSET rdx,RDX-SS | |
116 | CFI_OFFSET rsi,RSI-SS | |
117 | CFI_OFFSET rdi,RDI-SS | |
118 | CFI_OFFSET rsp,RSP-SS | |
119 | CFI_OFFSET rip,RIP-SS | |
120 | .endm | |
121 | /* | |
122 | * A newly forked process directly context switches into this. | |
123 | */ | |
124 | /* rdi: prev */ | |
125 | ENTRY(ret_from_fork) | |
126 | CFI_STARTPROC | |
127 | CFI_DEFAULT_STACK | |
128 | call schedule_tail | |
129 | GET_THREAD_INFO(%rcx) | |
130 | testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT),threadinfo_flags(%rcx) | |
131 | jnz rff_trace | |
132 | rff_action: | |
133 | RESTORE_REST | |
134 | testl $3,CS-ARGOFFSET(%rsp) # from kernel_thread? | |
135 | je int_ret_from_sys_call | |
136 | testl $_TIF_IA32,threadinfo_flags(%rcx) | |
137 | jnz int_ret_from_sys_call | |
138 | RESTORE_TOP_OF_STACK %rdi,ARGOFFSET | |
139 | jmp ret_from_sys_call | |
140 | rff_trace: | |
141 | movq %rsp,%rdi | |
142 | call syscall_trace_leave | |
143 | GET_THREAD_INFO(%rcx) | |
144 | jmp rff_action | |
145 | CFI_ENDPROC | |
146 | ||
147 | /* | |
148 | * System call entry. Upto 6 arguments in registers are supported. | |
149 | * | |
150 | * SYSCALL does not save anything on the stack and does not change the | |
151 | * stack pointer. | |
152 | */ | |
153 | ||
154 | /* | |
155 | * Register setup: | |
156 | * rax system call number | |
157 | * rdi arg0 | |
158 | * rcx return address for syscall/sysret, C arg3 | |
159 | * rsi arg1 | |
160 | * rdx arg2 | |
161 | * r10 arg3 (--> moved to rcx for C) | |
162 | * r8 arg4 | |
163 | * r9 arg5 | |
164 | * r11 eflags for syscall/sysret, temporary for C | |
165 | * r12-r15,rbp,rbx saved by C code, not touched. | |
166 | * | |
167 | * Interrupts are off on entry. | |
168 | * Only called from user space. | |
169 | * | |
170 | * XXX if we had a free scratch register we could save the RSP into the stack frame | |
171 | * and report it properly in ps. Unfortunately we haven't. | |
172 | */ | |
173 | ||
174 | ENTRY(system_call) | |
175 | CFI_STARTPROC | |
176 | swapgs | |
177 | movq %rsp,%gs:pda_oldrsp | |
178 | movq %gs:pda_kernelstack,%rsp | |
179 | sti | |
180 | SAVE_ARGS 8,1 | |
181 | movq %rax,ORIG_RAX-ARGOFFSET(%rsp) | |
182 | movq %rcx,RIP-ARGOFFSET(%rsp) | |
183 | GET_THREAD_INFO(%rcx) | |
184 | testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SECCOMP),threadinfo_flags(%rcx) | |
185 | jnz tracesys | |
186 | cmpq $__NR_syscall_max,%rax | |
187 | ja badsys | |
188 | movq %r10,%rcx | |
189 | call *sys_call_table(,%rax,8) # XXX: rip relative | |
190 | movq %rax,RAX-ARGOFFSET(%rsp) | |
191 | /* | |
192 | * Syscall return path ending with SYSRET (fast path) | |
193 | * Has incomplete stack frame and undefined top of stack. | |
194 | */ | |
195 | .globl ret_from_sys_call | |
196 | ret_from_sys_call: | |
11b854b2 | 197 | movl $_TIF_ALLWORK_MASK,%edi |
1da177e4 LT |
198 | /* edi: flagmask */ |
199 | sysret_check: | |
200 | GET_THREAD_INFO(%rcx) | |
201 | cli | |
202 | movl threadinfo_flags(%rcx),%edx | |
203 | andl %edi,%edx | |
204 | jnz sysret_careful | |
205 | movq RIP-ARGOFFSET(%rsp),%rcx | |
206 | RESTORE_ARGS 0,-ARG_SKIP,1 | |
207 | movq %gs:pda_oldrsp,%rsp | |
208 | swapgs | |
209 | sysretq | |
210 | ||
211 | /* Handle reschedules */ | |
212 | /* edx: work, edi: workmask */ | |
213 | sysret_careful: | |
214 | bt $TIF_NEED_RESCHED,%edx | |
215 | jnc sysret_signal | |
216 | sti | |
217 | pushq %rdi | |
218 | call schedule | |
219 | popq %rdi | |
220 | jmp sysret_check | |
221 | ||
222 | /* Handle a signal */ | |
223 | sysret_signal: | |
224 | sti | |
10ffdbb8 AK |
225 | testl $(_TIF_SIGPENDING|_TIF_NOTIFY_RESUME|_TIF_SINGLESTEP),%edx |
226 | jz 1f | |
227 | ||
228 | /* Really a signal */ | |
229 | /* edx: work flags (arg3) */ | |
1da177e4 LT |
230 | leaq do_notify_resume(%rip),%rax |
231 | leaq -ARGOFFSET(%rsp),%rdi # &pt_regs -> arg1 | |
232 | xorl %esi,%esi # oldset -> arg2 | |
233 | call ptregscall_common | |
10ffdbb8 | 234 | 1: movl $_TIF_NEED_RESCHED,%edi |
1da177e4 LT |
235 | jmp sysret_check |
236 | ||
237 | /* Do syscall tracing */ | |
238 | tracesys: | |
239 | SAVE_REST | |
240 | movq $-ENOSYS,RAX(%rsp) | |
241 | FIXUP_TOP_OF_STACK %rdi | |
242 | movq %rsp,%rdi | |
243 | call syscall_trace_enter | |
244 | LOAD_ARGS ARGOFFSET /* reload args from stack in case ptrace changed it */ | |
245 | RESTORE_REST | |
246 | cmpq $__NR_syscall_max,%rax | |
247 | ja 1f | |
248 | movq %r10,%rcx /* fixup for C */ | |
249 | call *sys_call_table(,%rax,8) | |
250 | movq %rax,RAX-ARGOFFSET(%rsp) | |
251 | 1: SAVE_REST | |
252 | movq %rsp,%rdi | |
253 | call syscall_trace_leave | |
254 | RESTORE_TOP_OF_STACK %rbx | |
255 | RESTORE_REST | |
256 | jmp ret_from_sys_call | |
257 | ||
258 | badsys: | |
259 | movq $-ENOSYS,RAX-ARGOFFSET(%rsp) | |
260 | jmp ret_from_sys_call | |
261 | ||
262 | /* | |
263 | * Syscall return path ending with IRET. | |
264 | * Has correct top of stack, but partial stack frame. | |
265 | */ | |
266 | ENTRY(int_ret_from_sys_call) | |
267 | cli | |
268 | testl $3,CS-ARGOFFSET(%rsp) | |
269 | je retint_restore_args | |
270 | movl $_TIF_ALLWORK_MASK,%edi | |
271 | /* edi: mask to check */ | |
272 | int_with_check: | |
273 | GET_THREAD_INFO(%rcx) | |
274 | movl threadinfo_flags(%rcx),%edx | |
275 | andl %edi,%edx | |
276 | jnz int_careful | |
277 | jmp retint_swapgs | |
278 | ||
279 | /* Either reschedule or signal or syscall exit tracking needed. */ | |
280 | /* First do a reschedule test. */ | |
281 | /* edx: work, edi: workmask */ | |
282 | int_careful: | |
283 | bt $TIF_NEED_RESCHED,%edx | |
284 | jnc int_very_careful | |
285 | sti | |
286 | pushq %rdi | |
287 | call schedule | |
288 | popq %rdi | |
cdd219cd | 289 | cli |
1da177e4 LT |
290 | jmp int_with_check |
291 | ||
292 | /* handle signals and tracing -- both require a full stack frame */ | |
293 | int_very_careful: | |
294 | sti | |
295 | SAVE_REST | |
296 | /* Check for syscall exit trace */ | |
297 | testl $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edx | |
298 | jz int_signal | |
299 | pushq %rdi | |
300 | leaq 8(%rsp),%rdi # &ptregs -> arg1 | |
301 | call syscall_trace_leave | |
302 | popq %rdi | |
36c1104e | 303 | andl $~(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP),%edi |
be9e6870 | 304 | cli |
1da177e4 LT |
305 | jmp int_restore_rest |
306 | ||
307 | int_signal: | |
308 | testl $(_TIF_NOTIFY_RESUME|_TIF_SIGPENDING|_TIF_SINGLESTEP),%edx | |
309 | jz 1f | |
310 | movq %rsp,%rdi # &ptregs -> arg1 | |
311 | xorl %esi,%esi # oldset -> arg2 | |
312 | call do_notify_resume | |
313 | 1: movl $_TIF_NEED_RESCHED,%edi | |
314 | int_restore_rest: | |
315 | RESTORE_REST | |
be9e6870 | 316 | cli |
1da177e4 LT |
317 | jmp int_with_check |
318 | CFI_ENDPROC | |
319 | ||
320 | /* | |
321 | * Certain special system calls that need to save a complete full stack frame. | |
322 | */ | |
323 | ||
324 | .macro PTREGSCALL label,func,arg | |
325 | .globl \label | |
326 | \label: | |
327 | leaq \func(%rip),%rax | |
328 | leaq -ARGOFFSET+8(%rsp),\arg /* 8 for return address */ | |
329 | jmp ptregscall_common | |
330 | .endm | |
331 | ||
332 | PTREGSCALL stub_clone, sys_clone, %r8 | |
333 | PTREGSCALL stub_fork, sys_fork, %rdi | |
334 | PTREGSCALL stub_vfork, sys_vfork, %rdi | |
335 | PTREGSCALL stub_rt_sigsuspend, sys_rt_sigsuspend, %rdx | |
336 | PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx | |
337 | PTREGSCALL stub_iopl, sys_iopl, %rsi | |
338 | ||
339 | ENTRY(ptregscall_common) | |
340 | CFI_STARTPROC | |
341 | popq %r11 | |
342 | CFI_ADJUST_CFA_OFFSET -8 | |
343 | SAVE_REST | |
344 | movq %r11, %r15 | |
345 | FIXUP_TOP_OF_STACK %r11 | |
346 | call *%rax | |
347 | RESTORE_TOP_OF_STACK %r11 | |
348 | movq %r15, %r11 | |
349 | RESTORE_REST | |
350 | pushq %r11 | |
351 | CFI_ADJUST_CFA_OFFSET 8 | |
352 | ret | |
353 | CFI_ENDPROC | |
354 | ||
355 | ENTRY(stub_execve) | |
356 | CFI_STARTPROC | |
357 | popq %r11 | |
358 | CFI_ADJUST_CFA_OFFSET -8 | |
359 | SAVE_REST | |
360 | movq %r11, %r15 | |
361 | FIXUP_TOP_OF_STACK %r11 | |
362 | call sys_execve | |
363 | GET_THREAD_INFO(%rcx) | |
364 | bt $TIF_IA32,threadinfo_flags(%rcx) | |
365 | jc exec_32bit | |
366 | RESTORE_TOP_OF_STACK %r11 | |
367 | movq %r15, %r11 | |
368 | RESTORE_REST | |
369 | push %r11 | |
370 | ret | |
371 | ||
372 | exec_32bit: | |
373 | CFI_ADJUST_CFA_OFFSET REST_SKIP | |
374 | movq %rax,RAX(%rsp) | |
375 | RESTORE_REST | |
376 | jmp int_ret_from_sys_call | |
377 | CFI_ENDPROC | |
378 | ||
379 | /* | |
380 | * sigreturn is special because it needs to restore all registers on return. | |
381 | * This cannot be done with SYSRET, so use the IRET return path instead. | |
382 | */ | |
383 | ENTRY(stub_rt_sigreturn) | |
384 | CFI_STARTPROC | |
385 | addq $8, %rsp | |
386 | SAVE_REST | |
387 | movq %rsp,%rdi | |
388 | FIXUP_TOP_OF_STACK %r11 | |
389 | call sys_rt_sigreturn | |
390 | movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer | |
391 | RESTORE_REST | |
392 | jmp int_ret_from_sys_call | |
393 | CFI_ENDPROC | |
394 | ||
395 | /* | |
396 | * Interrupt entry/exit. | |
397 | * | |
398 | * Interrupt entry points save only callee clobbered registers in fast path. | |
399 | * | |
400 | * Entry runs with interrupts off. | |
401 | */ | |
402 | ||
403 | /* 0(%rsp): interrupt number */ | |
404 | .macro interrupt func | |
405 | CFI_STARTPROC simple | |
406 | CFI_DEF_CFA rsp,(SS-RDI) | |
407 | CFI_REL_OFFSET rsp,(RSP-ORIG_RAX) | |
408 | CFI_REL_OFFSET rip,(RIP-ORIG_RAX) | |
409 | cld | |
410 | #ifdef CONFIG_DEBUG_INFO | |
411 | SAVE_ALL | |
412 | movq %rsp,%rdi | |
413 | /* | |
414 | * Setup a stack frame pointer. This allows gdb to trace | |
415 | * back to the original stack. | |
416 | */ | |
417 | movq %rsp,%rbp | |
418 | CFI_DEF_CFA_REGISTER rbp | |
419 | #else | |
420 | SAVE_ARGS | |
421 | leaq -ARGOFFSET(%rsp),%rdi # arg1 for handler | |
422 | #endif | |
423 | testl $3,CS(%rdi) | |
424 | je 1f | |
425 | swapgs | |
3829ee6b | 426 | 1: incl %gs:pda_irqcount # RED-PEN should check preempt count |
1da177e4 LT |
427 | movq %gs:pda_irqstackptr,%rax |
428 | cmoveq %rax,%rsp | |
429 | pushq %rdi # save old stack | |
430 | call \func | |
431 | .endm | |
432 | ||
433 | ENTRY(common_interrupt) | |
434 | interrupt do_IRQ | |
435 | /* 0(%rsp): oldrsp-ARGOFFSET */ | |
436 | ret_from_intr: | |
437 | popq %rdi | |
438 | cli | |
3829ee6b | 439 | decl %gs:pda_irqcount |
1da177e4 LT |
440 | #ifdef CONFIG_DEBUG_INFO |
441 | movq RBP(%rdi),%rbp | |
442 | #endif | |
443 | leaq ARGOFFSET(%rdi),%rsp | |
444 | exit_intr: | |
445 | GET_THREAD_INFO(%rcx) | |
446 | testl $3,CS-ARGOFFSET(%rsp) | |
447 | je retint_kernel | |
448 | ||
449 | /* Interrupt came from user space */ | |
450 | /* | |
451 | * Has a correct top of stack, but a partial stack frame | |
452 | * %rcx: thread info. Interrupts off. | |
453 | */ | |
454 | retint_with_reschedule: | |
455 | movl $_TIF_WORK_MASK,%edi | |
456 | retint_check: | |
457 | movl threadinfo_flags(%rcx),%edx | |
458 | andl %edi,%edx | |
459 | jnz retint_careful | |
460 | retint_swapgs: | |
1da177e4 LT |
461 | swapgs |
462 | retint_restore_args: | |
463 | cli | |
464 | RESTORE_ARGS 0,8,0 | |
465 | iret_label: | |
466 | iretq | |
467 | ||
468 | .section __ex_table,"a" | |
469 | .quad iret_label,bad_iret | |
470 | .previous | |
471 | .section .fixup,"ax" | |
472 | /* force a signal here? this matches i386 behaviour */ | |
473 | /* running with kernel gs */ | |
474 | bad_iret: | |
475 | movq $-9999,%rdi /* better code? */ | |
476 | jmp do_exit | |
477 | .previous | |
478 | ||
479 | /* edi: workmask, edx: work */ | |
480 | retint_careful: | |
481 | bt $TIF_NEED_RESCHED,%edx | |
482 | jnc retint_signal | |
483 | sti | |
484 | pushq %rdi | |
485 | call schedule | |
486 | popq %rdi | |
487 | GET_THREAD_INFO(%rcx) | |
488 | cli | |
489 | jmp retint_check | |
490 | ||
491 | retint_signal: | |
10ffdbb8 AK |
492 | testl $(_TIF_SIGPENDING|_TIF_NOTIFY_RESUME|_TIF_SINGLESTEP),%edx |
493 | jz retint_swapgs | |
1da177e4 LT |
494 | sti |
495 | SAVE_REST | |
496 | movq $-1,ORIG_RAX(%rsp) | |
3829ee6b | 497 | xorl %esi,%esi # oldset |
1da177e4 LT |
498 | movq %rsp,%rdi # &pt_regs |
499 | call do_notify_resume | |
500 | RESTORE_REST | |
501 | cli | |
10ffdbb8 | 502 | movl $_TIF_NEED_RESCHED,%edi |
be9e6870 | 503 | GET_THREAD_INFO(%rcx) |
1da177e4 LT |
504 | jmp retint_check |
505 | ||
506 | #ifdef CONFIG_PREEMPT | |
507 | /* Returning to kernel space. Check if we need preemption */ | |
508 | /* rcx: threadinfo. interrupts off. */ | |
509 | .p2align | |
510 | retint_kernel: | |
511 | cmpl $0,threadinfo_preempt_count(%rcx) | |
512 | jnz retint_restore_args | |
513 | bt $TIF_NEED_RESCHED,threadinfo_flags(%rcx) | |
514 | jnc retint_restore_args | |
515 | bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */ | |
516 | jnc retint_restore_args | |
517 | call preempt_schedule_irq | |
518 | jmp exit_intr | |
519 | #endif | |
520 | CFI_ENDPROC | |
521 | ||
522 | /* | |
523 | * APIC interrupts. | |
524 | */ | |
525 | .macro apicinterrupt num,func | |
526 | pushq $\num-256 | |
527 | interrupt \func | |
528 | jmp ret_from_intr | |
529 | CFI_ENDPROC | |
530 | .endm | |
531 | ||
532 | ENTRY(thermal_interrupt) | |
533 | apicinterrupt THERMAL_APIC_VECTOR,smp_thermal_interrupt | |
534 | ||
535 | #ifdef CONFIG_SMP | |
536 | ENTRY(reschedule_interrupt) | |
537 | apicinterrupt RESCHEDULE_VECTOR,smp_reschedule_interrupt | |
538 | ||
539 | ENTRY(invalidate_interrupt) | |
540 | apicinterrupt INVALIDATE_TLB_VECTOR,smp_invalidate_interrupt | |
541 | ||
542 | ENTRY(call_function_interrupt) | |
543 | apicinterrupt CALL_FUNCTION_VECTOR,smp_call_function_interrupt | |
544 | #endif | |
545 | ||
546 | #ifdef CONFIG_X86_LOCAL_APIC | |
547 | ENTRY(apic_timer_interrupt) | |
548 | apicinterrupt LOCAL_TIMER_VECTOR,smp_apic_timer_interrupt | |
549 | ||
550 | ENTRY(error_interrupt) | |
551 | apicinterrupt ERROR_APIC_VECTOR,smp_error_interrupt | |
552 | ||
553 | ENTRY(spurious_interrupt) | |
554 | apicinterrupt SPURIOUS_APIC_VECTOR,smp_spurious_interrupt | |
555 | #endif | |
556 | ||
557 | /* | |
558 | * Exception entry points. | |
559 | */ | |
560 | .macro zeroentry sym | |
561 | pushq $0 /* push error code/oldrax */ | |
562 | pushq %rax /* push real oldrax to the rdi slot */ | |
563 | leaq \sym(%rip),%rax | |
564 | jmp error_entry | |
565 | .endm | |
566 | ||
567 | .macro errorentry sym | |
568 | pushq %rax | |
569 | leaq \sym(%rip),%rax | |
570 | jmp error_entry | |
571 | .endm | |
572 | ||
573 | /* error code is on the stack already */ | |
574 | /* handle NMI like exceptions that can happen everywhere */ | |
575 | .macro paranoidentry sym | |
576 | SAVE_ALL | |
577 | cld | |
578 | movl $1,%ebx | |
579 | movl $MSR_GS_BASE,%ecx | |
580 | rdmsr | |
581 | testl %edx,%edx | |
582 | js 1f | |
583 | swapgs | |
584 | xorl %ebx,%ebx | |
585 | 1: movq %rsp,%rdi | |
586 | movq ORIG_RAX(%rsp),%rsi | |
587 | movq $-1,ORIG_RAX(%rsp) | |
588 | call \sym | |
6fefb0d1 | 589 | cli |
1da177e4 LT |
590 | .endm |
591 | ||
592 | /* | |
593 | * Exception entry point. This expects an error code/orig_rax on the stack | |
594 | * and the exception handler in %rax. | |
595 | */ | |
596 | ENTRY(error_entry) | |
597 | CFI_STARTPROC simple | |
598 | CFI_DEF_CFA rsp,(SS-RDI) | |
599 | CFI_REL_OFFSET rsp,(RSP-RDI) | |
600 | CFI_REL_OFFSET rip,(RIP-RDI) | |
601 | /* rdi slot contains rax, oldrax contains error code */ | |
602 | cld | |
603 | subq $14*8,%rsp | |
604 | CFI_ADJUST_CFA_OFFSET (14*8) | |
605 | movq %rsi,13*8(%rsp) | |
606 | CFI_REL_OFFSET rsi,RSI | |
607 | movq 14*8(%rsp),%rsi /* load rax from rdi slot */ | |
608 | movq %rdx,12*8(%rsp) | |
609 | CFI_REL_OFFSET rdx,RDX | |
610 | movq %rcx,11*8(%rsp) | |
611 | CFI_REL_OFFSET rcx,RCX | |
612 | movq %rsi,10*8(%rsp) /* store rax */ | |
613 | CFI_REL_OFFSET rax,RAX | |
614 | movq %r8, 9*8(%rsp) | |
615 | CFI_REL_OFFSET r8,R8 | |
616 | movq %r9, 8*8(%rsp) | |
617 | CFI_REL_OFFSET r9,R9 | |
618 | movq %r10,7*8(%rsp) | |
619 | CFI_REL_OFFSET r10,R10 | |
620 | movq %r11,6*8(%rsp) | |
621 | CFI_REL_OFFSET r11,R11 | |
622 | movq %rbx,5*8(%rsp) | |
623 | CFI_REL_OFFSET rbx,RBX | |
624 | movq %rbp,4*8(%rsp) | |
625 | CFI_REL_OFFSET rbp,RBP | |
626 | movq %r12,3*8(%rsp) | |
627 | CFI_REL_OFFSET r12,R12 | |
628 | movq %r13,2*8(%rsp) | |
629 | CFI_REL_OFFSET r13,R13 | |
630 | movq %r14,1*8(%rsp) | |
631 | CFI_REL_OFFSET r14,R14 | |
632 | movq %r15,(%rsp) | |
633 | CFI_REL_OFFSET r15,R15 | |
634 | xorl %ebx,%ebx | |
635 | testl $3,CS(%rsp) | |
636 | je error_kernelspace | |
637 | error_swapgs: | |
638 | swapgs | |
639 | error_sti: | |
640 | movq %rdi,RDI(%rsp) | |
641 | movq %rsp,%rdi | |
642 | movq ORIG_RAX(%rsp),%rsi /* get error code */ | |
643 | movq $-1,ORIG_RAX(%rsp) | |
644 | call *%rax | |
645 | /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */ | |
646 | error_exit: | |
647 | movl %ebx,%eax | |
648 | RESTORE_REST | |
649 | cli | |
650 | GET_THREAD_INFO(%rcx) | |
651 | testl %eax,%eax | |
652 | jne retint_kernel | |
653 | movl threadinfo_flags(%rcx),%edx | |
654 | movl $_TIF_WORK_MASK,%edi | |
655 | andl %edi,%edx | |
656 | jnz retint_careful | |
657 | swapgs | |
658 | RESTORE_ARGS 0,8,0 | |
659 | iretq | |
660 | CFI_ENDPROC | |
661 | ||
662 | error_kernelspace: | |
663 | incl %ebx | |
664 | /* There are two places in the kernel that can potentially fault with | |
665 | usergs. Handle them here. The exception handlers after | |
666 | iret run with kernel gs again, so don't set the user space flag. | |
667 | B stepping K8s sometimes report an truncated RIP for IRET | |
668 | exceptions returning to compat mode. Check for these here too. */ | |
669 | leaq iret_label(%rip),%rbp | |
670 | cmpq %rbp,RIP(%rsp) | |
671 | je error_swapgs | |
672 | movl %ebp,%ebp /* zero extend */ | |
673 | cmpq %rbp,RIP(%rsp) | |
674 | je error_swapgs | |
675 | cmpq $gs_change,RIP(%rsp) | |
676 | je error_swapgs | |
677 | jmp error_sti | |
678 | ||
679 | /* Reload gs selector with exception handling */ | |
680 | /* edi: new selector */ | |
681 | ENTRY(load_gs_index) | |
682 | pushf | |
683 | cli | |
684 | swapgs | |
685 | gs_change: | |
686 | movl %edi,%gs | |
687 | 2: mfence /* workaround */ | |
688 | swapgs | |
689 | popf | |
690 | ret | |
691 | ||
692 | .section __ex_table,"a" | |
693 | .align 8 | |
694 | .quad gs_change,bad_gs | |
695 | .previous | |
696 | .section .fixup,"ax" | |
697 | /* running with kernelgs */ | |
698 | bad_gs: | |
699 | swapgs /* switch back to user gs */ | |
700 | xorl %eax,%eax | |
701 | movl %eax,%gs | |
702 | jmp 2b | |
703 | .previous | |
704 | ||
705 | /* | |
706 | * Create a kernel thread. | |
707 | * | |
708 | * C extern interface: | |
709 | * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags) | |
710 | * | |
711 | * asm input arguments: | |
712 | * rdi: fn, rsi: arg, rdx: flags | |
713 | */ | |
714 | ENTRY(kernel_thread) | |
715 | CFI_STARTPROC | |
716 | FAKE_STACK_FRAME $child_rip | |
717 | SAVE_ALL | |
718 | ||
719 | # rdi: flags, rsi: usp, rdx: will be &pt_regs | |
720 | movq %rdx,%rdi | |
721 | orq kernel_thread_flags(%rip),%rdi | |
722 | movq $-1, %rsi | |
723 | movq %rsp, %rdx | |
724 | ||
725 | xorl %r8d,%r8d | |
726 | xorl %r9d,%r9d | |
727 | ||
728 | # clone now | |
729 | call do_fork | |
730 | movq %rax,RAX(%rsp) | |
731 | xorl %edi,%edi | |
732 | ||
733 | /* | |
734 | * It isn't worth to check for reschedule here, | |
735 | * so internally to the x86_64 port you can rely on kernel_thread() | |
736 | * not to reschedule the child before returning, this avoids the need | |
737 | * of hacks for example to fork off the per-CPU idle tasks. | |
738 | * [Hopefully no generic code relies on the reschedule -AK] | |
739 | */ | |
740 | RESTORE_ALL | |
741 | UNFAKE_STACK_FRAME | |
742 | ret | |
743 | CFI_ENDPROC | |
744 | ||
745 | ||
746 | child_rip: | |
747 | /* | |
748 | * Here we are in the child and the registers are set as they were | |
749 | * at kernel_thread() invocation in the parent. | |
750 | */ | |
751 | movq %rdi, %rax | |
752 | movq %rsi, %rdi | |
753 | call *%rax | |
754 | # exit | |
3829ee6b | 755 | xorl %edi, %edi |
1da177e4 LT |
756 | call do_exit |
757 | ||
758 | /* | |
759 | * execve(). This function needs to use IRET, not SYSRET, to set up all state properly. | |
760 | * | |
761 | * C extern interface: | |
762 | * extern long execve(char *name, char **argv, char **envp) | |
763 | * | |
764 | * asm input arguments: | |
765 | * rdi: name, rsi: argv, rdx: envp | |
766 | * | |
767 | * We want to fallback into: | |
768 | * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs regs) | |
769 | * | |
770 | * do_sys_execve asm fallback arguments: | |
771 | * rdi: name, rsi: argv, rdx: envp, fake frame on the stack | |
772 | */ | |
773 | ENTRY(execve) | |
774 | CFI_STARTPROC | |
775 | FAKE_STACK_FRAME $0 | |
776 | SAVE_ALL | |
777 | call sys_execve | |
778 | movq %rax, RAX(%rsp) | |
779 | RESTORE_REST | |
780 | testq %rax,%rax | |
781 | je int_ret_from_sys_call | |
782 | RESTORE_ARGS | |
783 | UNFAKE_STACK_FRAME | |
784 | ret | |
785 | CFI_ENDPROC | |
786 | ||
787 | ENTRY(page_fault) | |
788 | errorentry do_page_fault | |
789 | ||
790 | ENTRY(coprocessor_error) | |
791 | zeroentry do_coprocessor_error | |
792 | ||
793 | ENTRY(simd_coprocessor_error) | |
794 | zeroentry do_simd_coprocessor_error | |
795 | ||
796 | ENTRY(device_not_available) | |
797 | zeroentry math_state_restore | |
798 | ||
799 | /* runs on exception stack */ | |
800 | ENTRY(debug) | |
801 | CFI_STARTPROC | |
802 | pushq $0 | |
803 | CFI_ADJUST_CFA_OFFSET 8 | |
804 | paranoidentry do_debug | |
1da177e4 LT |
805 | jmp paranoid_exit |
806 | CFI_ENDPROC | |
807 | ||
808 | /* runs on exception stack */ | |
809 | ENTRY(nmi) | |
810 | CFI_STARTPROC | |
811 | pushq $-1 | |
812 | CFI_ADJUST_CFA_OFFSET 8 | |
813 | paranoidentry do_nmi | |
6fefb0d1 AK |
814 | /* |
815 | * "Paranoid" exit path from exception stack. | |
816 | * Paranoid because this is used by NMIs and cannot take | |
817 | * any kernel state for granted. | |
818 | * We don't do kernel preemption checks here, because only | |
819 | * NMI should be common and it does not enable IRQs and | |
820 | * cannot get reschedule ticks. | |
821 | */ | |
1da177e4 LT |
822 | /* ebx: no swapgs flag */ |
823 | paranoid_exit: | |
824 | testl %ebx,%ebx /* swapgs needed? */ | |
825 | jnz paranoid_restore | |
6fefb0d1 AK |
826 | testl $3,CS(%rsp) |
827 | jnz paranoid_userspace | |
1da177e4 | 828 | paranoid_swapgs: |
1da177e4 LT |
829 | swapgs |
830 | paranoid_restore: | |
831 | RESTORE_ALL 8 | |
832 | iretq | |
833 | paranoid_userspace: | |
1da177e4 | 834 | GET_THREAD_INFO(%rcx) |
6fefb0d1 AK |
835 | movl threadinfo_flags(%rcx),%ebx |
836 | andl $_TIF_WORK_MASK,%ebx | |
11b854b2 | 837 | jz paranoid_swapgs |
6fefb0d1 AK |
838 | movq %rsp,%rdi /* &pt_regs */ |
839 | call sync_regs | |
840 | movq %rax,%rsp /* switch stack for scheduling */ | |
841 | testl $_TIF_NEED_RESCHED,%ebx | |
842 | jnz paranoid_schedule | |
843 | movl %ebx,%edx /* arg3: thread flags */ | |
1da177e4 | 844 | sti |
6fefb0d1 AK |
845 | xorl %esi,%esi /* arg2: oldset */ |
846 | movq %rsp,%rdi /* arg1: &pt_regs */ | |
1da177e4 | 847 | call do_notify_resume |
6fefb0d1 AK |
848 | cli |
849 | jmp paranoid_userspace | |
850 | paranoid_schedule: | |
11b854b2 AK |
851 | sti |
852 | call schedule | |
6fefb0d1 AK |
853 | cli |
854 | jmp paranoid_userspace | |
1da177e4 | 855 | CFI_ENDPROC |
6fefb0d1 | 856 | |
1da177e4 LT |
857 | ENTRY(int3) |
858 | zeroentry do_int3 | |
859 | ||
860 | ENTRY(overflow) | |
861 | zeroentry do_overflow | |
862 | ||
863 | ENTRY(bounds) | |
864 | zeroentry do_bounds | |
865 | ||
866 | ENTRY(invalid_op) | |
867 | zeroentry do_invalid_op | |
868 | ||
869 | ENTRY(coprocessor_segment_overrun) | |
870 | zeroentry do_coprocessor_segment_overrun | |
871 | ||
872 | ENTRY(reserved) | |
873 | zeroentry do_reserved | |
874 | ||
875 | /* runs on exception stack */ | |
876 | ENTRY(double_fault) | |
877 | CFI_STARTPROC | |
878 | paranoidentry do_double_fault | |
1da177e4 LT |
879 | jmp paranoid_exit |
880 | CFI_ENDPROC | |
881 | ||
882 | ENTRY(invalid_TSS) | |
883 | errorentry do_invalid_TSS | |
884 | ||
885 | ENTRY(segment_not_present) | |
886 | errorentry do_segment_not_present | |
887 | ||
888 | /* runs on exception stack */ | |
889 | ENTRY(stack_segment) | |
890 | CFI_STARTPROC | |
891 | paranoidentry do_stack_segment | |
1da177e4 LT |
892 | jmp paranoid_exit |
893 | CFI_ENDPROC | |
894 | ||
895 | ENTRY(general_protection) | |
896 | errorentry do_general_protection | |
897 | ||
898 | ENTRY(alignment_check) | |
899 | errorentry do_alignment_check | |
900 | ||
901 | ENTRY(divide_error) | |
902 | zeroentry do_divide_error | |
903 | ||
904 | ENTRY(spurious_interrupt_bug) | |
905 | zeroentry do_spurious_interrupt_bug | |
906 | ||
907 | #ifdef CONFIG_X86_MCE | |
908 | /* runs on exception stack */ | |
909 | ENTRY(machine_check) | |
910 | CFI_STARTPROC | |
911 | pushq $0 | |
912 | CFI_ADJUST_CFA_OFFSET 8 | |
913 | paranoidentry do_machine_check | |
914 | jmp paranoid_exit | |
915 | CFI_ENDPROC | |
916 | #endif | |
917 | ||
918 | ENTRY(call_debug) | |
919 | zeroentry do_call_debug | |
920 |