[PATCH] paravirt: Patch inline replacements for paravirt intercepts
[deliverable/linux.git] / arch / i386 / kernel / entry.S
1 /*
2 * linux/arch/i386/entry.S
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 /*
8 * entry.S contains the system-call and fault low-level handling routines.
9 * This also contains the timer-interrupt handler, as well as all interrupts
10 * and faults that can result in a task-switch.
11 *
12 * NOTE: This code handles signal-recognition, which happens every time
13 * after a timer-interrupt and after each system call.
14 *
15 * I changed all the .align's to 4 (16 byte alignment), as that's faster
16 * on a 486.
17 *
18 * Stack layout in 'ret_from_system_call':
19 * ptrace needs to have all regs on the stack.
20 * if the order here is changed, it needs to be
21 * updated in fork.c:copy_process, signal.c:do_signal,
22 * ptrace.c and ptrace.h
23 *
24 * 0(%esp) - %ebx
25 * 4(%esp) - %ecx
26 * 8(%esp) - %edx
27 * C(%esp) - %esi
28 * 10(%esp) - %edi
29 * 14(%esp) - %ebp
30 * 18(%esp) - %eax
31 * 1C(%esp) - %ds
32 * 20(%esp) - %es
33 * 24(%esp) - %gs
34 * 28(%esp) - orig_eax
35 * 2C(%esp) - %eip
36 * 30(%esp) - %cs
37 * 34(%esp) - %eflags
38 * 38(%esp) - %oldesp
39 * 3C(%esp) - %oldss
40 *
41 * "current" is in register %ebx during any slow entries.
42 */
43
44 #include <linux/linkage.h>
45 #include <asm/thread_info.h>
46 #include <asm/irqflags.h>
47 #include <asm/errno.h>
48 #include <asm/segment.h>
49 #include <asm/smp.h>
50 #include <asm/page.h>
51 #include <asm/desc.h>
52 #include <asm/percpu.h>
53 #include <asm/dwarf2.h>
54 #include "irq_vectors.h"
55
56 /*
57 * We use macros for low-level operations which need to be overridden
58 * for paravirtualization. The following will never clobber any registers:
59 * INTERRUPT_RETURN (aka. "iret")
60 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
61 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
62 *
63 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
64 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
65 * Allowing a register to be clobbered can shrink the paravirt replacement
66 * enough to patch inline, increasing performance.
67 */
68
69 #define nr_syscalls ((syscall_table_size)/4)
70
71 CF_MASK = 0x00000001
72 TF_MASK = 0x00000100
73 IF_MASK = 0x00000200
74 DF_MASK = 0x00000400
75 NT_MASK = 0x00004000
76 VM_MASK = 0x00020000
77
78 #ifdef CONFIG_PREEMPT
79 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
80 #else
81 #define preempt_stop(clobbers)
82 #define resume_kernel restore_nocheck
83 #endif
84
85 .macro TRACE_IRQS_IRET
86 #ifdef CONFIG_TRACE_IRQFLAGS
87 testl $IF_MASK,PT_EFLAGS(%esp) # interrupts off?
88 jz 1f
89 TRACE_IRQS_ON
90 1:
91 #endif
92 .endm
93
94 #ifdef CONFIG_VM86
95 #define resume_userspace_sig check_userspace
96 #else
97 #define resume_userspace_sig resume_userspace
98 #endif
99
100 #define SAVE_ALL \
101 cld; \
102 pushl %gs; \
103 CFI_ADJUST_CFA_OFFSET 4;\
104 /*CFI_REL_OFFSET gs, 0;*/\
105 pushl %es; \
106 CFI_ADJUST_CFA_OFFSET 4;\
107 /*CFI_REL_OFFSET es, 0;*/\
108 pushl %ds; \
109 CFI_ADJUST_CFA_OFFSET 4;\
110 /*CFI_REL_OFFSET ds, 0;*/\
111 pushl %eax; \
112 CFI_ADJUST_CFA_OFFSET 4;\
113 CFI_REL_OFFSET eax, 0;\
114 pushl %ebp; \
115 CFI_ADJUST_CFA_OFFSET 4;\
116 CFI_REL_OFFSET ebp, 0;\
117 pushl %edi; \
118 CFI_ADJUST_CFA_OFFSET 4;\
119 CFI_REL_OFFSET edi, 0;\
120 pushl %esi; \
121 CFI_ADJUST_CFA_OFFSET 4;\
122 CFI_REL_OFFSET esi, 0;\
123 pushl %edx; \
124 CFI_ADJUST_CFA_OFFSET 4;\
125 CFI_REL_OFFSET edx, 0;\
126 pushl %ecx; \
127 CFI_ADJUST_CFA_OFFSET 4;\
128 CFI_REL_OFFSET ecx, 0;\
129 pushl %ebx; \
130 CFI_ADJUST_CFA_OFFSET 4;\
131 CFI_REL_OFFSET ebx, 0;\
132 movl $(__USER_DS), %edx; \
133 movl %edx, %ds; \
134 movl %edx, %es; \
135 movl $(__KERNEL_PDA), %edx; \
136 movl %edx, %gs
137
138 #define RESTORE_INT_REGS \
139 popl %ebx; \
140 CFI_ADJUST_CFA_OFFSET -4;\
141 CFI_RESTORE ebx;\
142 popl %ecx; \
143 CFI_ADJUST_CFA_OFFSET -4;\
144 CFI_RESTORE ecx;\
145 popl %edx; \
146 CFI_ADJUST_CFA_OFFSET -4;\
147 CFI_RESTORE edx;\
148 popl %esi; \
149 CFI_ADJUST_CFA_OFFSET -4;\
150 CFI_RESTORE esi;\
151 popl %edi; \
152 CFI_ADJUST_CFA_OFFSET -4;\
153 CFI_RESTORE edi;\
154 popl %ebp; \
155 CFI_ADJUST_CFA_OFFSET -4;\
156 CFI_RESTORE ebp;\
157 popl %eax; \
158 CFI_ADJUST_CFA_OFFSET -4;\
159 CFI_RESTORE eax
160
161 #define RESTORE_REGS \
162 RESTORE_INT_REGS; \
163 1: popl %ds; \
164 CFI_ADJUST_CFA_OFFSET -4;\
165 /*CFI_RESTORE ds;*/\
166 2: popl %es; \
167 CFI_ADJUST_CFA_OFFSET -4;\
168 /*CFI_RESTORE es;*/\
169 3: popl %gs; \
170 CFI_ADJUST_CFA_OFFSET -4;\
171 /*CFI_RESTORE gs;*/\
172 .pushsection .fixup,"ax"; \
173 4: movl $0,(%esp); \
174 jmp 1b; \
175 5: movl $0,(%esp); \
176 jmp 2b; \
177 6: movl $0,(%esp); \
178 jmp 3b; \
179 .section __ex_table,"a";\
180 .align 4; \
181 .long 1b,4b; \
182 .long 2b,5b; \
183 .long 3b,6b; \
184 .popsection
185
186 #define RING0_INT_FRAME \
187 CFI_STARTPROC simple;\
188 CFI_SIGNAL_FRAME;\
189 CFI_DEF_CFA esp, 3*4;\
190 /*CFI_OFFSET cs, -2*4;*/\
191 CFI_OFFSET eip, -3*4
192
193 #define RING0_EC_FRAME \
194 CFI_STARTPROC simple;\
195 CFI_SIGNAL_FRAME;\
196 CFI_DEF_CFA esp, 4*4;\
197 /*CFI_OFFSET cs, -2*4;*/\
198 CFI_OFFSET eip, -3*4
199
200 #define RING0_PTREGS_FRAME \
201 CFI_STARTPROC simple;\
202 CFI_SIGNAL_FRAME;\
203 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\
204 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\
205 CFI_OFFSET eip, PT_EIP-PT_OLDESP;\
206 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\
207 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\
208 CFI_OFFSET eax, PT_EAX-PT_OLDESP;\
209 CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\
210 CFI_OFFSET edi, PT_EDI-PT_OLDESP;\
211 CFI_OFFSET esi, PT_ESI-PT_OLDESP;\
212 CFI_OFFSET edx, PT_EDX-PT_OLDESP;\
213 CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\
214 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
215
216 ENTRY(ret_from_fork)
217 CFI_STARTPROC
218 pushl %eax
219 CFI_ADJUST_CFA_OFFSET 4
220 call schedule_tail
221 GET_THREAD_INFO(%ebp)
222 popl %eax
223 CFI_ADJUST_CFA_OFFSET -4
224 pushl $0x0202 # Reset kernel eflags
225 CFI_ADJUST_CFA_OFFSET 4
226 popfl
227 CFI_ADJUST_CFA_OFFSET -4
228 jmp syscall_exit
229 CFI_ENDPROC
230
231 /*
232 * Return to user mode is not as complex as all this looks,
233 * but we want the default path for a system call return to
234 * go as quickly as possible which is why some of this is
235 * less clear than it otherwise should be.
236 */
237
238 # userspace resumption stub bypassing syscall exit tracing
239 ALIGN
240 RING0_PTREGS_FRAME
241 ret_from_exception:
242 preempt_stop(CLBR_ANY)
243 ret_from_intr:
244 GET_THREAD_INFO(%ebp)
245 check_userspace:
246 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
247 movb PT_CS(%esp), %al
248 andl $(VM_MASK | SEGMENT_RPL_MASK), %eax
249 cmpl $USER_RPL, %eax
250 jb resume_kernel # not returning to v8086 or userspace
251
252 ENTRY(resume_userspace)
253 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
254 # setting need_resched or sigpending
255 # between sampling and the iret
256 movl TI_flags(%ebp), %ecx
257 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
258 # int/exception return?
259 jne work_pending
260 jmp restore_all
261
262 #ifdef CONFIG_PREEMPT
263 ENTRY(resume_kernel)
264 DISABLE_INTERRUPTS(CLBR_ANY)
265 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
266 jnz restore_nocheck
267 need_resched:
268 movl TI_flags(%ebp), %ecx # need_resched set ?
269 testb $_TIF_NEED_RESCHED, %cl
270 jz restore_all
271 testl $IF_MASK,PT_EFLAGS(%esp) # interrupts off (exception path) ?
272 jz restore_all
273 call preempt_schedule_irq
274 jmp need_resched
275 #endif
276 CFI_ENDPROC
277
278 /* SYSENTER_RETURN points to after the "sysenter" instruction in
279 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
280
281 # sysenter call handler stub
282 ENTRY(sysenter_entry)
283 CFI_STARTPROC simple
284 CFI_SIGNAL_FRAME
285 CFI_DEF_CFA esp, 0
286 CFI_REGISTER esp, ebp
287 movl TSS_sysenter_esp0(%esp),%esp
288 sysenter_past_esp:
289 /*
290 * No need to follow this irqs on/off section: the syscall
291 * disabled irqs and here we enable it straight after entry:
292 */
293 ENABLE_INTERRUPTS(CLBR_NONE)
294 pushl $(__USER_DS)
295 CFI_ADJUST_CFA_OFFSET 4
296 /*CFI_REL_OFFSET ss, 0*/
297 pushl %ebp
298 CFI_ADJUST_CFA_OFFSET 4
299 CFI_REL_OFFSET esp, 0
300 pushfl
301 CFI_ADJUST_CFA_OFFSET 4
302 pushl $(__USER_CS)
303 CFI_ADJUST_CFA_OFFSET 4
304 /*CFI_REL_OFFSET cs, 0*/
305 /*
306 * Push current_thread_info()->sysenter_return to the stack.
307 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
308 * pushed above; +8 corresponds to copy_thread's esp0 setting.
309 */
310 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
311 CFI_ADJUST_CFA_OFFSET 4
312 CFI_REL_OFFSET eip, 0
313
314 /*
315 * Load the potential sixth argument from user stack.
316 * Careful about security.
317 */
318 cmpl $__PAGE_OFFSET-3,%ebp
319 jae syscall_fault
320 1: movl (%ebp),%ebp
321 .section __ex_table,"a"
322 .align 4
323 .long 1b,syscall_fault
324 .previous
325
326 pushl %eax
327 CFI_ADJUST_CFA_OFFSET 4
328 SAVE_ALL
329 GET_THREAD_INFO(%ebp)
330
331 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
332 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
333 jnz syscall_trace_entry
334 cmpl $(nr_syscalls), %eax
335 jae syscall_badsys
336 call *sys_call_table(,%eax,4)
337 movl %eax,PT_EAX(%esp)
338 DISABLE_INTERRUPTS(CLBR_ECX|CLBR_EDX)
339 TRACE_IRQS_OFF
340 movl TI_flags(%ebp), %ecx
341 testw $_TIF_ALLWORK_MASK, %cx
342 jne syscall_exit_work
343 /* if something modifies registers it must also disable sysexit */
344 movl PT_EIP(%esp), %edx
345 movl PT_OLDESP(%esp), %ecx
346 xorl %ebp,%ebp
347 TRACE_IRQS_ON
348 1: mov PT_GS(%esp), %gs
349 ENABLE_INTERRUPTS_SYSEXIT
350 CFI_ENDPROC
351 .pushsection .fixup,"ax"
352 2: movl $0,PT_GS(%esp)
353 jmp 1b
354 .section __ex_table,"a"
355 .align 4
356 .long 1b,2b
357 .popsection
358
359 # system call handler stub
360 ENTRY(system_call)
361 RING0_INT_FRAME # can't unwind into user space anyway
362 pushl %eax # save orig_eax
363 CFI_ADJUST_CFA_OFFSET 4
364 SAVE_ALL
365 GET_THREAD_INFO(%ebp)
366 testl $TF_MASK,PT_EFLAGS(%esp)
367 jz no_singlestep
368 orl $_TIF_SINGLESTEP,TI_flags(%ebp)
369 no_singlestep:
370 # system call tracing in operation / emulation
371 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
372 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
373 jnz syscall_trace_entry
374 cmpl $(nr_syscalls), %eax
375 jae syscall_badsys
376 syscall_call:
377 call *sys_call_table(,%eax,4)
378 movl %eax,PT_EAX(%esp) # store the return value
379 syscall_exit:
380 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
381 # setting need_resched or sigpending
382 # between sampling and the iret
383 TRACE_IRQS_OFF
384 movl TI_flags(%ebp), %ecx
385 testw $_TIF_ALLWORK_MASK, %cx # current->work
386 jne syscall_exit_work
387
388 restore_all:
389 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
390 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
391 # are returning to the kernel.
392 # See comments in process.c:copy_thread() for details.
393 movb PT_OLDSS(%esp), %ah
394 movb PT_CS(%esp), %al
395 andl $(VM_MASK | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
396 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
397 CFI_REMEMBER_STATE
398 je ldt_ss # returning to user-space with LDT SS
399 restore_nocheck:
400 TRACE_IRQS_IRET
401 restore_nocheck_notrace:
402 RESTORE_REGS
403 addl $4, %esp # skip orig_eax/error_code
404 CFI_ADJUST_CFA_OFFSET -4
405 1: INTERRUPT_RETURN
406 .section .fixup,"ax"
407 iret_exc:
408 TRACE_IRQS_ON
409 ENABLE_INTERRUPTS(CLBR_NONE)
410 pushl $0 # no error code
411 pushl $do_iret_error
412 jmp error_code
413 .previous
414 .section __ex_table,"a"
415 .align 4
416 .long 1b,iret_exc
417 .previous
418
419 CFI_RESTORE_STATE
420 ldt_ss:
421 larl PT_OLDSS(%esp), %eax
422 jnz restore_nocheck
423 testl $0x00400000, %eax # returning to 32bit stack?
424 jnz restore_nocheck # allright, normal return
425
426 #ifdef CONFIG_PARAVIRT
427 /*
428 * The kernel can't run on a non-flat stack if paravirt mode
429 * is active. Rather than try to fixup the high bits of
430 * ESP, bypass this code entirely. This may break DOSemu
431 * and/or Wine support in a paravirt VM, although the option
432 * is still available to implement the setting of the high
433 * 16-bits in the INTERRUPT_RETURN paravirt-op.
434 */
435 cmpl $0, paravirt_ops+PARAVIRT_enabled
436 jne restore_nocheck
437 #endif
438
439 /* If returning to userspace with 16bit stack,
440 * try to fix the higher word of ESP, as the CPU
441 * won't restore it.
442 * This is an "official" bug of all the x86-compatible
443 * CPUs, which we can try to work around to make
444 * dosemu and wine happy. */
445 movl PT_OLDESP(%esp), %eax
446 movl %esp, %edx
447 call patch_espfix_desc
448 pushl $__ESPFIX_SS
449 CFI_ADJUST_CFA_OFFSET 4
450 pushl %eax
451 CFI_ADJUST_CFA_OFFSET 4
452 DISABLE_INTERRUPTS(CLBR_EAX)
453 TRACE_IRQS_OFF
454 lss (%esp), %esp
455 CFI_ADJUST_CFA_OFFSET -8
456 jmp restore_nocheck
457 CFI_ENDPROC
458
459 # perform work that needs to be done immediately before resumption
460 ALIGN
461 RING0_PTREGS_FRAME # can't unwind into user space anyway
462 work_pending:
463 testb $_TIF_NEED_RESCHED, %cl
464 jz work_notifysig
465 work_resched:
466 call schedule
467 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
468 # setting need_resched or sigpending
469 # between sampling and the iret
470 TRACE_IRQS_OFF
471 movl TI_flags(%ebp), %ecx
472 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
473 # than syscall tracing?
474 jz restore_all
475 testb $_TIF_NEED_RESCHED, %cl
476 jnz work_resched
477
478 work_notifysig: # deal with pending signals and
479 # notify-resume requests
480 #ifdef CONFIG_VM86
481 testl $VM_MASK, PT_EFLAGS(%esp)
482 movl %esp, %eax
483 jne work_notifysig_v86 # returning to kernel-space or
484 # vm86-space
485 xorl %edx, %edx
486 call do_notify_resume
487 jmp resume_userspace_sig
488
489 ALIGN
490 work_notifysig_v86:
491 pushl %ecx # save ti_flags for do_notify_resume
492 CFI_ADJUST_CFA_OFFSET 4
493 call save_v86_state # %eax contains pt_regs pointer
494 popl %ecx
495 CFI_ADJUST_CFA_OFFSET -4
496 movl %eax, %esp
497 #else
498 movl %esp, %eax
499 #endif
500 xorl %edx, %edx
501 call do_notify_resume
502 jmp resume_userspace_sig
503
504 # perform syscall exit tracing
505 ALIGN
506 syscall_trace_entry:
507 movl $-ENOSYS,PT_EAX(%esp)
508 movl %esp, %eax
509 xorl %edx,%edx
510 call do_syscall_trace
511 cmpl $0, %eax
512 jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU,
513 # so must skip actual syscall
514 movl PT_ORIG_EAX(%esp), %eax
515 cmpl $(nr_syscalls), %eax
516 jnae syscall_call
517 jmp syscall_exit
518
519 # perform syscall exit tracing
520 ALIGN
521 syscall_exit_work:
522 testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl
523 jz work_pending
524 TRACE_IRQS_ON
525 ENABLE_INTERRUPTS(CLBR_ANY) # could let do_syscall_trace() call
526 # schedule() instead
527 movl %esp, %eax
528 movl $1, %edx
529 call do_syscall_trace
530 jmp resume_userspace
531 CFI_ENDPROC
532
533 RING0_INT_FRAME # can't unwind into user space anyway
534 syscall_fault:
535 pushl %eax # save orig_eax
536 CFI_ADJUST_CFA_OFFSET 4
537 SAVE_ALL
538 GET_THREAD_INFO(%ebp)
539 movl $-EFAULT,PT_EAX(%esp)
540 jmp resume_userspace
541
542 syscall_badsys:
543 movl $-ENOSYS,PT_EAX(%esp)
544 jmp resume_userspace
545 CFI_ENDPROC
546
547 #define FIXUP_ESPFIX_STACK \
548 /* since we are on a wrong stack, we cant make it a C code :( */ \
549 movl %gs:PDA_cpu, %ebx; \
550 PER_CPU(cpu_gdt_descr, %ebx); \
551 movl GDS_address(%ebx), %ebx; \
552 GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \
553 addl %esp, %eax; \
554 pushl $__KERNEL_DS; \
555 CFI_ADJUST_CFA_OFFSET 4; \
556 pushl %eax; \
557 CFI_ADJUST_CFA_OFFSET 4; \
558 lss (%esp), %esp; \
559 CFI_ADJUST_CFA_OFFSET -8;
560 #define UNWIND_ESPFIX_STACK \
561 movl %ss, %eax; \
562 /* see if on espfix stack */ \
563 cmpw $__ESPFIX_SS, %ax; \
564 jne 27f; \
565 movl $__KERNEL_DS, %eax; \
566 movl %eax, %ds; \
567 movl %eax, %es; \
568 /* switch to normal stack */ \
569 FIXUP_ESPFIX_STACK; \
570 27:;
571
572 /*
573 * Build the entry stubs and pointer table with
574 * some assembler magic.
575 */
576 .data
577 ENTRY(interrupt)
578 .text
579
580 vector=0
581 ENTRY(irq_entries_start)
582 RING0_INT_FRAME
583 .rept NR_IRQS
584 ALIGN
585 .if vector
586 CFI_ADJUST_CFA_OFFSET -4
587 .endif
588 1: pushl $~(vector)
589 CFI_ADJUST_CFA_OFFSET 4
590 jmp common_interrupt
591 .data
592 .long 1b
593 .text
594 vector=vector+1
595 .endr
596
597 /*
598 * the CPU automatically disables interrupts when executing an IRQ vector,
599 * so IRQ-flags tracing has to follow that:
600 */
601 ALIGN
602 common_interrupt:
603 SAVE_ALL
604 TRACE_IRQS_OFF
605 movl %esp,%eax
606 call do_IRQ
607 jmp ret_from_intr
608 CFI_ENDPROC
609
610 #define BUILD_INTERRUPT(name, nr) \
611 ENTRY(name) \
612 RING0_INT_FRAME; \
613 pushl $~(nr); \
614 CFI_ADJUST_CFA_OFFSET 4; \
615 SAVE_ALL; \
616 TRACE_IRQS_OFF \
617 movl %esp,%eax; \
618 call smp_/**/name; \
619 jmp ret_from_intr; \
620 CFI_ENDPROC
621
622 /* The include is where all of the SMP etc. interrupts come from */
623 #include "entry_arch.h"
624
625 KPROBE_ENTRY(page_fault)
626 RING0_EC_FRAME
627 pushl $do_page_fault
628 CFI_ADJUST_CFA_OFFSET 4
629 ALIGN
630 error_code:
631 /* the function address is in %gs's slot on the stack */
632 pushl %es
633 CFI_ADJUST_CFA_OFFSET 4
634 /*CFI_REL_OFFSET es, 0*/
635 pushl %ds
636 CFI_ADJUST_CFA_OFFSET 4
637 /*CFI_REL_OFFSET ds, 0*/
638 pushl %eax
639 CFI_ADJUST_CFA_OFFSET 4
640 CFI_REL_OFFSET eax, 0
641 pushl %ebp
642 CFI_ADJUST_CFA_OFFSET 4
643 CFI_REL_OFFSET ebp, 0
644 pushl %edi
645 CFI_ADJUST_CFA_OFFSET 4
646 CFI_REL_OFFSET edi, 0
647 pushl %esi
648 CFI_ADJUST_CFA_OFFSET 4
649 CFI_REL_OFFSET esi, 0
650 pushl %edx
651 CFI_ADJUST_CFA_OFFSET 4
652 CFI_REL_OFFSET edx, 0
653 pushl %ecx
654 CFI_ADJUST_CFA_OFFSET 4
655 CFI_REL_OFFSET ecx, 0
656 pushl %ebx
657 CFI_ADJUST_CFA_OFFSET 4
658 CFI_REL_OFFSET ebx, 0
659 cld
660 pushl %gs
661 CFI_ADJUST_CFA_OFFSET 4
662 /*CFI_REL_OFFSET gs, 0*/
663 movl $(__KERNEL_PDA), %ecx
664 movl %ecx, %gs
665 UNWIND_ESPFIX_STACK
666 popl %ecx
667 CFI_ADJUST_CFA_OFFSET -4
668 /*CFI_REGISTER es, ecx*/
669 movl PT_GS(%esp), %edi # get the function address
670 movl PT_ORIG_EAX(%esp), %edx # get the error code
671 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
672 mov %ecx, PT_GS(%esp)
673 /*CFI_REL_OFFSET gs, ES*/
674 movl $(__USER_DS), %ecx
675 movl %ecx, %ds
676 movl %ecx, %es
677 movl %esp,%eax # pt_regs pointer
678 call *%edi
679 jmp ret_from_exception
680 CFI_ENDPROC
681 KPROBE_END(page_fault)
682
683 ENTRY(coprocessor_error)
684 RING0_INT_FRAME
685 pushl $0
686 CFI_ADJUST_CFA_OFFSET 4
687 pushl $do_coprocessor_error
688 CFI_ADJUST_CFA_OFFSET 4
689 jmp error_code
690 CFI_ENDPROC
691
692 ENTRY(simd_coprocessor_error)
693 RING0_INT_FRAME
694 pushl $0
695 CFI_ADJUST_CFA_OFFSET 4
696 pushl $do_simd_coprocessor_error
697 CFI_ADJUST_CFA_OFFSET 4
698 jmp error_code
699 CFI_ENDPROC
700
701 ENTRY(device_not_available)
702 RING0_INT_FRAME
703 pushl $-1 # mark this as an int
704 CFI_ADJUST_CFA_OFFSET 4
705 SAVE_ALL
706 GET_CR0_INTO_EAX
707 testl $0x4, %eax # EM (math emulation bit)
708 jne device_not_available_emulate
709 preempt_stop(CLBR_ANY)
710 call math_state_restore
711 jmp ret_from_exception
712 device_not_available_emulate:
713 pushl $0 # temporary storage for ORIG_EIP
714 CFI_ADJUST_CFA_OFFSET 4
715 call math_emulate
716 addl $4, %esp
717 CFI_ADJUST_CFA_OFFSET -4
718 jmp ret_from_exception
719 CFI_ENDPROC
720
721 /*
722 * Debug traps and NMI can happen at the one SYSENTER instruction
723 * that sets up the real kernel stack. Check here, since we can't
724 * allow the wrong stack to be used.
725 *
726 * "TSS_sysenter_esp0+12" is because the NMI/debug handler will have
727 * already pushed 3 words if it hits on the sysenter instruction:
728 * eflags, cs and eip.
729 *
730 * We just load the right stack, and push the three (known) values
731 * by hand onto the new stack - while updating the return eip past
732 * the instruction that would have done it for sysenter.
733 */
734 #define FIX_STACK(offset, ok, label) \
735 cmpw $__KERNEL_CS,4(%esp); \
736 jne ok; \
737 label: \
738 movl TSS_sysenter_esp0+offset(%esp),%esp; \
739 CFI_DEF_CFA esp, 0; \
740 CFI_UNDEFINED eip; \
741 pushfl; \
742 CFI_ADJUST_CFA_OFFSET 4; \
743 pushl $__KERNEL_CS; \
744 CFI_ADJUST_CFA_OFFSET 4; \
745 pushl $sysenter_past_esp; \
746 CFI_ADJUST_CFA_OFFSET 4; \
747 CFI_REL_OFFSET eip, 0
748
749 KPROBE_ENTRY(debug)
750 RING0_INT_FRAME
751 cmpl $sysenter_entry,(%esp)
752 jne debug_stack_correct
753 FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn)
754 debug_stack_correct:
755 pushl $-1 # mark this as an int
756 CFI_ADJUST_CFA_OFFSET 4
757 SAVE_ALL
758 xorl %edx,%edx # error code 0
759 movl %esp,%eax # pt_regs pointer
760 call do_debug
761 jmp ret_from_exception
762 CFI_ENDPROC
763 KPROBE_END(debug)
764
765 /*
766 * NMI is doubly nasty. It can happen _while_ we're handling
767 * a debug fault, and the debug fault hasn't yet been able to
768 * clear up the stack. So we first check whether we got an
769 * NMI on the sysenter entry path, but after that we need to
770 * check whether we got an NMI on the debug path where the debug
771 * fault happened on the sysenter path.
772 */
773 KPROBE_ENTRY(nmi)
774 RING0_INT_FRAME
775 pushl %eax
776 CFI_ADJUST_CFA_OFFSET 4
777 movl %ss, %eax
778 cmpw $__ESPFIX_SS, %ax
779 popl %eax
780 CFI_ADJUST_CFA_OFFSET -4
781 je nmi_espfix_stack
782 cmpl $sysenter_entry,(%esp)
783 je nmi_stack_fixup
784 pushl %eax
785 CFI_ADJUST_CFA_OFFSET 4
786 movl %esp,%eax
787 /* Do not access memory above the end of our stack page,
788 * it might not exist.
789 */
790 andl $(THREAD_SIZE-1),%eax
791 cmpl $(THREAD_SIZE-20),%eax
792 popl %eax
793 CFI_ADJUST_CFA_OFFSET -4
794 jae nmi_stack_correct
795 cmpl $sysenter_entry,12(%esp)
796 je nmi_debug_stack_check
797 nmi_stack_correct:
798 /* We have a RING0_INT_FRAME here */
799 pushl %eax
800 CFI_ADJUST_CFA_OFFSET 4
801 SAVE_ALL
802 xorl %edx,%edx # zero error code
803 movl %esp,%eax # pt_regs pointer
804 call do_nmi
805 jmp restore_nocheck_notrace
806 CFI_ENDPROC
807
808 nmi_stack_fixup:
809 RING0_INT_FRAME
810 FIX_STACK(12,nmi_stack_correct, 1)
811 jmp nmi_stack_correct
812
813 nmi_debug_stack_check:
814 /* We have a RING0_INT_FRAME here */
815 cmpw $__KERNEL_CS,16(%esp)
816 jne nmi_stack_correct
817 cmpl $debug,(%esp)
818 jb nmi_stack_correct
819 cmpl $debug_esp_fix_insn,(%esp)
820 ja nmi_stack_correct
821 FIX_STACK(24,nmi_stack_correct, 1)
822 jmp nmi_stack_correct
823
824 nmi_espfix_stack:
825 /* We have a RING0_INT_FRAME here.
826 *
827 * create the pointer to lss back
828 */
829 pushl %ss
830 CFI_ADJUST_CFA_OFFSET 4
831 pushl %esp
832 CFI_ADJUST_CFA_OFFSET 4
833 addw $4, (%esp)
834 /* copy the iret frame of 12 bytes */
835 .rept 3
836 pushl 16(%esp)
837 CFI_ADJUST_CFA_OFFSET 4
838 .endr
839 pushl %eax
840 CFI_ADJUST_CFA_OFFSET 4
841 SAVE_ALL
842 FIXUP_ESPFIX_STACK # %eax == %esp
843 xorl %edx,%edx # zero error code
844 call do_nmi
845 RESTORE_REGS
846 lss 12+4(%esp), %esp # back to espfix stack
847 CFI_ADJUST_CFA_OFFSET -24
848 1: INTERRUPT_RETURN
849 CFI_ENDPROC
850 .section __ex_table,"a"
851 .align 4
852 .long 1b,iret_exc
853 .previous
854 KPROBE_END(nmi)
855
856 #ifdef CONFIG_PARAVIRT
857 ENTRY(native_iret)
858 1: iret
859 .section __ex_table,"a"
860 .align 4
861 .long 1b,iret_exc
862 .previous
863
864 ENTRY(native_irq_enable_sysexit)
865 sti
866 sysexit
867 #endif
868
869 KPROBE_ENTRY(int3)
870 RING0_INT_FRAME
871 pushl $-1 # mark this as an int
872 CFI_ADJUST_CFA_OFFSET 4
873 SAVE_ALL
874 xorl %edx,%edx # zero error code
875 movl %esp,%eax # pt_regs pointer
876 call do_int3
877 jmp ret_from_exception
878 CFI_ENDPROC
879 KPROBE_END(int3)
880
881 ENTRY(overflow)
882 RING0_INT_FRAME
883 pushl $0
884 CFI_ADJUST_CFA_OFFSET 4
885 pushl $do_overflow
886 CFI_ADJUST_CFA_OFFSET 4
887 jmp error_code
888 CFI_ENDPROC
889
890 ENTRY(bounds)
891 RING0_INT_FRAME
892 pushl $0
893 CFI_ADJUST_CFA_OFFSET 4
894 pushl $do_bounds
895 CFI_ADJUST_CFA_OFFSET 4
896 jmp error_code
897 CFI_ENDPROC
898
899 ENTRY(invalid_op)
900 RING0_INT_FRAME
901 pushl $0
902 CFI_ADJUST_CFA_OFFSET 4
903 pushl $do_invalid_op
904 CFI_ADJUST_CFA_OFFSET 4
905 jmp error_code
906 CFI_ENDPROC
907
908 ENTRY(coprocessor_segment_overrun)
909 RING0_INT_FRAME
910 pushl $0
911 CFI_ADJUST_CFA_OFFSET 4
912 pushl $do_coprocessor_segment_overrun
913 CFI_ADJUST_CFA_OFFSET 4
914 jmp error_code
915 CFI_ENDPROC
916
917 ENTRY(invalid_TSS)
918 RING0_EC_FRAME
919 pushl $do_invalid_TSS
920 CFI_ADJUST_CFA_OFFSET 4
921 jmp error_code
922 CFI_ENDPROC
923
924 ENTRY(segment_not_present)
925 RING0_EC_FRAME
926 pushl $do_segment_not_present
927 CFI_ADJUST_CFA_OFFSET 4
928 jmp error_code
929 CFI_ENDPROC
930
931 ENTRY(stack_segment)
932 RING0_EC_FRAME
933 pushl $do_stack_segment
934 CFI_ADJUST_CFA_OFFSET 4
935 jmp error_code
936 CFI_ENDPROC
937
938 KPROBE_ENTRY(general_protection)
939 RING0_EC_FRAME
940 pushl $do_general_protection
941 CFI_ADJUST_CFA_OFFSET 4
942 jmp error_code
943 CFI_ENDPROC
944 KPROBE_END(general_protection)
945
946 ENTRY(alignment_check)
947 RING0_EC_FRAME
948 pushl $do_alignment_check
949 CFI_ADJUST_CFA_OFFSET 4
950 jmp error_code
951 CFI_ENDPROC
952
953 ENTRY(divide_error)
954 RING0_INT_FRAME
955 pushl $0 # no error code
956 CFI_ADJUST_CFA_OFFSET 4
957 pushl $do_divide_error
958 CFI_ADJUST_CFA_OFFSET 4
959 jmp error_code
960 CFI_ENDPROC
961
962 #ifdef CONFIG_X86_MCE
963 ENTRY(machine_check)
964 RING0_INT_FRAME
965 pushl $0
966 CFI_ADJUST_CFA_OFFSET 4
967 pushl machine_check_vector
968 CFI_ADJUST_CFA_OFFSET 4
969 jmp error_code
970 CFI_ENDPROC
971 #endif
972
973 ENTRY(spurious_interrupt_bug)
974 RING0_INT_FRAME
975 pushl $0
976 CFI_ADJUST_CFA_OFFSET 4
977 pushl $do_spurious_interrupt_bug
978 CFI_ADJUST_CFA_OFFSET 4
979 jmp error_code
980 CFI_ENDPROC
981
982 #ifdef CONFIG_STACK_UNWIND
983 ENTRY(arch_unwind_init_running)
984 CFI_STARTPROC
985 movl 4(%esp), %edx
986 movl (%esp), %ecx
987 leal 4(%esp), %eax
988 movl %ebx, PT_EBX(%edx)
989 xorl %ebx, %ebx
990 movl %ebx, PT_ECX(%edx)
991 movl %ebx, PT_EDX(%edx)
992 movl %esi, PT_ESI(%edx)
993 movl %edi, PT_EDI(%edx)
994 movl %ebp, PT_EBP(%edx)
995 movl %ebx, PT_EAX(%edx)
996 movl $__USER_DS, PT_DS(%edx)
997 movl $__USER_DS, PT_ES(%edx)
998 movl $0, PT_GS(%edx)
999 movl %ebx, PT_ORIG_EAX(%edx)
1000 movl %ecx, PT_EIP(%edx)
1001 movl 12(%esp), %ecx
1002 movl $__KERNEL_CS, PT_CS(%edx)
1003 movl %ebx, PT_EFLAGS(%edx)
1004 movl %eax, PT_OLDESP(%edx)
1005 movl 8(%esp), %eax
1006 movl %ecx, 8(%esp)
1007 movl PT_EBX(%edx), %ebx
1008 movl $__KERNEL_DS, PT_OLDSS(%edx)
1009 jmpl *%eax
1010 CFI_ENDPROC
1011 ENDPROC(arch_unwind_init_running)
1012 #endif
1013
1014 ENTRY(kernel_thread_helper)
1015 pushl $0 # fake return address for unwinder
1016 CFI_STARTPROC
1017 movl %edx,%eax
1018 push %edx
1019 CFI_ADJUST_CFA_OFFSET 4
1020 call *%ebx
1021 push %eax
1022 CFI_ADJUST_CFA_OFFSET 4
1023 call do_exit
1024 CFI_ENDPROC
1025 ENDPROC(kernel_thread_helper)
1026
1027 .section .rodata,"a"
1028 #include "syscall_table.S"
1029
1030 syscall_table_size=(.-sys_call_table)
This page took 0.071826 seconds and 5 git commands to generate.