perfcounters, x86: fix sw counters on non-PMC CPUs
[deliverable/linux.git] / arch / x86 / kernel / entry_32.S
1 /*
2 *
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 */
5
6 /*
7 * entry.S contains the system-call and fault low-level handling routines.
8 * This also contains the timer-interrupt handler, as well as all interrupts
9 * and faults that can result in a task-switch.
10 *
11 * NOTE: This code handles signal-recognition, which happens every time
12 * after a timer-interrupt and after each system call.
13 *
14 * I changed all the .align's to 4 (16 byte alignment), as that's faster
15 * on a 486.
16 *
17 * Stack layout in 'syscall_exit':
18 * ptrace needs to have all regs on the stack.
19 * if the order here is changed, it needs to be
20 * updated in fork.c:copy_process, signal.c:do_signal,
21 * ptrace.c and ptrace.h
22 *
23 * 0(%esp) - %ebx
24 * 4(%esp) - %ecx
25 * 8(%esp) - %edx
26 * C(%esp) - %esi
27 * 10(%esp) - %edi
28 * 14(%esp) - %ebp
29 * 18(%esp) - %eax
30 * 1C(%esp) - %ds
31 * 20(%esp) - %es
32 * 24(%esp) - %fs
33 * 28(%esp) - orig_eax
34 * 2C(%esp) - %eip
35 * 30(%esp) - %cs
36 * 34(%esp) - %eflags
37 * 38(%esp) - %oldesp
38 * 3C(%esp) - %oldss
39 *
40 * "current" is in register %ebx during any slow entries.
41 */
42
43 #include <linux/linkage.h>
44 #include <asm/thread_info.h>
45 #include <asm/irqflags.h>
46 #include <asm/errno.h>
47 #include <asm/segment.h>
48 #include <asm/smp.h>
49 #include <asm/page.h>
50 #include <asm/desc.h>
51 #include <asm/percpu.h>
52 #include <asm/dwarf2.h>
53 #include <asm/processor-flags.h>
54 #include <asm/ftrace.h>
55 #include <asm/irq_vectors.h>
56
57 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
58 #include <linux/elf-em.h>
59 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
60 #define __AUDIT_ARCH_LE 0x40000000
61
62 #ifndef CONFIG_AUDITSYSCALL
63 #define sysenter_audit syscall_trace_entry
64 #define sysexit_audit syscall_exit_work
65 #endif
66
67 /*
68 * We use macros for low-level operations which need to be overridden
69 * for paravirtualization. The following will never clobber any registers:
70 * INTERRUPT_RETURN (aka. "iret")
71 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
72 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
73 *
74 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
75 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
76 * Allowing a register to be clobbered can shrink the paravirt replacement
77 * enough to patch inline, increasing performance.
78 */
79
80 #define nr_syscalls ((syscall_table_size)/4)
81
82 #ifdef CONFIG_PREEMPT
83 #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
84 #else
85 #define preempt_stop(clobbers)
86 #define resume_kernel restore_nocheck
87 #endif
88
89 .macro TRACE_IRQS_IRET
90 #ifdef CONFIG_TRACE_IRQFLAGS
91 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
92 jz 1f
93 TRACE_IRQS_ON
94 1:
95 #endif
96 .endm
97
98 #ifdef CONFIG_VM86
99 #define resume_userspace_sig check_userspace
100 #else
101 #define resume_userspace_sig resume_userspace
102 #endif
103
104 #define SAVE_ALL \
105 cld; \
106 pushl %fs; \
107 CFI_ADJUST_CFA_OFFSET 4;\
108 /*CFI_REL_OFFSET fs, 0;*/\
109 pushl %es; \
110 CFI_ADJUST_CFA_OFFSET 4;\
111 /*CFI_REL_OFFSET es, 0;*/\
112 pushl %ds; \
113 CFI_ADJUST_CFA_OFFSET 4;\
114 /*CFI_REL_OFFSET ds, 0;*/\
115 pushl %eax; \
116 CFI_ADJUST_CFA_OFFSET 4;\
117 CFI_REL_OFFSET eax, 0;\
118 pushl %ebp; \
119 CFI_ADJUST_CFA_OFFSET 4;\
120 CFI_REL_OFFSET ebp, 0;\
121 pushl %edi; \
122 CFI_ADJUST_CFA_OFFSET 4;\
123 CFI_REL_OFFSET edi, 0;\
124 pushl %esi; \
125 CFI_ADJUST_CFA_OFFSET 4;\
126 CFI_REL_OFFSET esi, 0;\
127 pushl %edx; \
128 CFI_ADJUST_CFA_OFFSET 4;\
129 CFI_REL_OFFSET edx, 0;\
130 pushl %ecx; \
131 CFI_ADJUST_CFA_OFFSET 4;\
132 CFI_REL_OFFSET ecx, 0;\
133 pushl %ebx; \
134 CFI_ADJUST_CFA_OFFSET 4;\
135 CFI_REL_OFFSET ebx, 0;\
136 movl $(__USER_DS), %edx; \
137 movl %edx, %ds; \
138 movl %edx, %es; \
139 movl $(__KERNEL_PERCPU), %edx; \
140 movl %edx, %fs
141
142 #define RESTORE_INT_REGS \
143 popl %ebx; \
144 CFI_ADJUST_CFA_OFFSET -4;\
145 CFI_RESTORE ebx;\
146 popl %ecx; \
147 CFI_ADJUST_CFA_OFFSET -4;\
148 CFI_RESTORE ecx;\
149 popl %edx; \
150 CFI_ADJUST_CFA_OFFSET -4;\
151 CFI_RESTORE edx;\
152 popl %esi; \
153 CFI_ADJUST_CFA_OFFSET -4;\
154 CFI_RESTORE esi;\
155 popl %edi; \
156 CFI_ADJUST_CFA_OFFSET -4;\
157 CFI_RESTORE edi;\
158 popl %ebp; \
159 CFI_ADJUST_CFA_OFFSET -4;\
160 CFI_RESTORE ebp;\
161 popl %eax; \
162 CFI_ADJUST_CFA_OFFSET -4;\
163 CFI_RESTORE eax
164
165 #define RESTORE_REGS \
166 RESTORE_INT_REGS; \
167 1: popl %ds; \
168 CFI_ADJUST_CFA_OFFSET -4;\
169 /*CFI_RESTORE ds;*/\
170 2: popl %es; \
171 CFI_ADJUST_CFA_OFFSET -4;\
172 /*CFI_RESTORE es;*/\
173 3: popl %fs; \
174 CFI_ADJUST_CFA_OFFSET -4;\
175 /*CFI_RESTORE fs;*/\
176 .pushsection .fixup,"ax"; \
177 4: movl $0,(%esp); \
178 jmp 1b; \
179 5: movl $0,(%esp); \
180 jmp 2b; \
181 6: movl $0,(%esp); \
182 jmp 3b; \
183 .section __ex_table,"a";\
184 .align 4; \
185 .long 1b,4b; \
186 .long 2b,5b; \
187 .long 3b,6b; \
188 .popsection
189
190 #define RING0_INT_FRAME \
191 CFI_STARTPROC simple;\
192 CFI_SIGNAL_FRAME;\
193 CFI_DEF_CFA esp, 3*4;\
194 /*CFI_OFFSET cs, -2*4;*/\
195 CFI_OFFSET eip, -3*4
196
197 #define RING0_EC_FRAME \
198 CFI_STARTPROC simple;\
199 CFI_SIGNAL_FRAME;\
200 CFI_DEF_CFA esp, 4*4;\
201 /*CFI_OFFSET cs, -2*4;*/\
202 CFI_OFFSET eip, -3*4
203
204 #define RING0_PTREGS_FRAME \
205 CFI_STARTPROC simple;\
206 CFI_SIGNAL_FRAME;\
207 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\
208 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\
209 CFI_OFFSET eip, PT_EIP-PT_OLDESP;\
210 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\
211 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\
212 CFI_OFFSET eax, PT_EAX-PT_OLDESP;\
213 CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\
214 CFI_OFFSET edi, PT_EDI-PT_OLDESP;\
215 CFI_OFFSET esi, PT_ESI-PT_OLDESP;\
216 CFI_OFFSET edx, PT_EDX-PT_OLDESP;\
217 CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\
218 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
219
220 ENTRY(ret_from_fork)
221 CFI_STARTPROC
222 pushl %eax
223 CFI_ADJUST_CFA_OFFSET 4
224 call schedule_tail
225 GET_THREAD_INFO(%ebp)
226 popl %eax
227 CFI_ADJUST_CFA_OFFSET -4
228 pushl $0x0202 # Reset kernel eflags
229 CFI_ADJUST_CFA_OFFSET 4
230 popfl
231 CFI_ADJUST_CFA_OFFSET -4
232 jmp syscall_exit
233 CFI_ENDPROC
234 END(ret_from_fork)
235
236 /*
237 * Return to user mode is not as complex as all this looks,
238 * but we want the default path for a system call return to
239 * go as quickly as possible which is why some of this is
240 * less clear than it otherwise should be.
241 */
242
243 # userspace resumption stub bypassing syscall exit tracing
244 ALIGN
245 RING0_PTREGS_FRAME
246 ret_from_exception:
247 preempt_stop(CLBR_ANY)
248 ret_from_intr:
249 GET_THREAD_INFO(%ebp)
250 check_userspace:
251 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
252 movb PT_CS(%esp), %al
253 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
254 cmpl $USER_RPL, %eax
255 jb resume_kernel # not returning to v8086 or userspace
256
257 ENTRY(resume_userspace)
258 LOCKDEP_SYS_EXIT
259 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
260 # setting need_resched or sigpending
261 # between sampling and the iret
262 TRACE_IRQS_OFF
263 movl TI_flags(%ebp), %ecx
264 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
265 # int/exception return?
266 jne work_pending
267 jmp restore_all
268 END(ret_from_exception)
269
270 #ifdef CONFIG_PREEMPT
271 ENTRY(resume_kernel)
272 DISABLE_INTERRUPTS(CLBR_ANY)
273 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
274 jnz restore_nocheck
275 need_resched:
276 movl TI_flags(%ebp), %ecx # need_resched set ?
277 testb $_TIF_NEED_RESCHED, %cl
278 jz restore_all
279 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
280 jz restore_all
281 call preempt_schedule_irq
282 jmp need_resched
283 END(resume_kernel)
284 #endif
285 CFI_ENDPROC
286
287 /* SYSENTER_RETURN points to after the "sysenter" instruction in
288 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
289
290 # sysenter call handler stub
291 ENTRY(ia32_sysenter_target)
292 CFI_STARTPROC simple
293 CFI_SIGNAL_FRAME
294 CFI_DEF_CFA esp, 0
295 CFI_REGISTER esp, ebp
296 movl TSS_sysenter_sp0(%esp),%esp
297 sysenter_past_esp:
298 /*
299 * Interrupts are disabled here, but we can't trace it until
300 * enough kernel state to call TRACE_IRQS_OFF can be called - but
301 * we immediately enable interrupts at that point anyway.
302 */
303 pushl $(__USER_DS)
304 CFI_ADJUST_CFA_OFFSET 4
305 /*CFI_REL_OFFSET ss, 0*/
306 pushl %ebp
307 CFI_ADJUST_CFA_OFFSET 4
308 CFI_REL_OFFSET esp, 0
309 pushfl
310 orl $X86_EFLAGS_IF, (%esp)
311 CFI_ADJUST_CFA_OFFSET 4
312 pushl $(__USER_CS)
313 CFI_ADJUST_CFA_OFFSET 4
314 /*CFI_REL_OFFSET cs, 0*/
315 /*
316 * Push current_thread_info()->sysenter_return to the stack.
317 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
318 * pushed above; +8 corresponds to copy_thread's esp0 setting.
319 */
320 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
321 CFI_ADJUST_CFA_OFFSET 4
322 CFI_REL_OFFSET eip, 0
323
324 pushl %eax
325 CFI_ADJUST_CFA_OFFSET 4
326 SAVE_ALL
327 ENABLE_INTERRUPTS(CLBR_NONE)
328
329 /*
330 * Load the potential sixth argument from user stack.
331 * Careful about security.
332 */
333 cmpl $__PAGE_OFFSET-3,%ebp
334 jae syscall_fault
335 1: movl (%ebp),%ebp
336 movl %ebp,PT_EBP(%esp)
337 .section __ex_table,"a"
338 .align 4
339 .long 1b,syscall_fault
340 .previous
341
342 GET_THREAD_INFO(%ebp)
343
344 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
345 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
346 jnz sysenter_audit
347 sysenter_do_call:
348 cmpl $(nr_syscalls), %eax
349 jae syscall_badsys
350 call *sys_call_table(,%eax,4)
351 movl %eax,PT_EAX(%esp)
352 LOCKDEP_SYS_EXIT
353 DISABLE_INTERRUPTS(CLBR_ANY)
354 TRACE_IRQS_OFF
355 movl TI_flags(%ebp), %ecx
356 testw $_TIF_ALLWORK_MASK, %cx
357 jne sysexit_audit
358 sysenter_exit:
359 /* if something modifies registers it must also disable sysexit */
360 movl PT_EIP(%esp), %edx
361 movl PT_OLDESP(%esp), %ecx
362 xorl %ebp,%ebp
363 TRACE_IRQS_ON
364 1: mov PT_FS(%esp), %fs
365 ENABLE_INTERRUPTS_SYSEXIT
366
367 #ifdef CONFIG_AUDITSYSCALL
368 sysenter_audit:
369 testw $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
370 jnz syscall_trace_entry
371 addl $4,%esp
372 CFI_ADJUST_CFA_OFFSET -4
373 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
374 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
375 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
376 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
377 movl %eax,%edx /* 2nd arg: syscall number */
378 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
379 call audit_syscall_entry
380 pushl %ebx
381 CFI_ADJUST_CFA_OFFSET 4
382 movl PT_EAX(%esp),%eax /* reload syscall number */
383 jmp sysenter_do_call
384
385 sysexit_audit:
386 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
387 jne syscall_exit_work
388 TRACE_IRQS_ON
389 ENABLE_INTERRUPTS(CLBR_ANY)
390 movl %eax,%edx /* second arg, syscall return value */
391 cmpl $0,%eax /* is it < 0? */
392 setl %al /* 1 if so, 0 if not */
393 movzbl %al,%eax /* zero-extend that */
394 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
395 call audit_syscall_exit
396 DISABLE_INTERRUPTS(CLBR_ANY)
397 TRACE_IRQS_OFF
398 movl TI_flags(%ebp), %ecx
399 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
400 jne syscall_exit_work
401 movl PT_EAX(%esp),%eax /* reload syscall return value */
402 jmp sysenter_exit
403 #endif
404
405 CFI_ENDPROC
406 .pushsection .fixup,"ax"
407 2: movl $0,PT_FS(%esp)
408 jmp 1b
409 .section __ex_table,"a"
410 .align 4
411 .long 1b,2b
412 .popsection
413 ENDPROC(ia32_sysenter_target)
414
415 # system call handler stub
416 ENTRY(system_call)
417 RING0_INT_FRAME # can't unwind into user space anyway
418 pushl %eax # save orig_eax
419 CFI_ADJUST_CFA_OFFSET 4
420 SAVE_ALL
421 GET_THREAD_INFO(%ebp)
422 # system call tracing in operation / emulation
423 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
424 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
425 jnz syscall_trace_entry
426 cmpl $(nr_syscalls), %eax
427 jae syscall_badsys
428 syscall_call:
429 call *sys_call_table(,%eax,4)
430 movl %eax,PT_EAX(%esp) # store the return value
431 syscall_exit:
432 LOCKDEP_SYS_EXIT
433 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
434 # setting need_resched or sigpending
435 # between sampling and the iret
436 TRACE_IRQS_OFF
437 movl TI_flags(%ebp), %ecx
438 testw $_TIF_ALLWORK_MASK, %cx # current->work
439 jne syscall_exit_work
440
441 restore_all:
442 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
443 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
444 # are returning to the kernel.
445 # See comments in process.c:copy_thread() for details.
446 movb PT_OLDSS(%esp), %ah
447 movb PT_CS(%esp), %al
448 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
449 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
450 CFI_REMEMBER_STATE
451 je ldt_ss # returning to user-space with LDT SS
452 restore_nocheck:
453 TRACE_IRQS_IRET
454 restore_nocheck_notrace:
455 RESTORE_REGS
456 addl $4, %esp # skip orig_eax/error_code
457 CFI_ADJUST_CFA_OFFSET -4
458 irq_return:
459 INTERRUPT_RETURN
460 .section .fixup,"ax"
461 ENTRY(iret_exc)
462 pushl $0 # no error code
463 pushl $do_iret_error
464 jmp error_code
465 .previous
466 .section __ex_table,"a"
467 .align 4
468 .long irq_return,iret_exc
469 .previous
470
471 CFI_RESTORE_STATE
472 ldt_ss:
473 larl PT_OLDSS(%esp), %eax
474 jnz restore_nocheck
475 testl $0x00400000, %eax # returning to 32bit stack?
476 jnz restore_nocheck # allright, normal return
477
478 #ifdef CONFIG_PARAVIRT
479 /*
480 * The kernel can't run on a non-flat stack if paravirt mode
481 * is active. Rather than try to fixup the high bits of
482 * ESP, bypass this code entirely. This may break DOSemu
483 * and/or Wine support in a paravirt VM, although the option
484 * is still available to implement the setting of the high
485 * 16-bits in the INTERRUPT_RETURN paravirt-op.
486 */
487 cmpl $0, pv_info+PARAVIRT_enabled
488 jne restore_nocheck
489 #endif
490
491 /* If returning to userspace with 16bit stack,
492 * try to fix the higher word of ESP, as the CPU
493 * won't restore it.
494 * This is an "official" bug of all the x86-compatible
495 * CPUs, which we can try to work around to make
496 * dosemu and wine happy. */
497 movl PT_OLDESP(%esp), %eax
498 movl %esp, %edx
499 call patch_espfix_desc
500 pushl $__ESPFIX_SS
501 CFI_ADJUST_CFA_OFFSET 4
502 pushl %eax
503 CFI_ADJUST_CFA_OFFSET 4
504 DISABLE_INTERRUPTS(CLBR_EAX)
505 TRACE_IRQS_OFF
506 lss (%esp), %esp
507 CFI_ADJUST_CFA_OFFSET -8
508 jmp restore_nocheck
509 CFI_ENDPROC
510 ENDPROC(system_call)
511
512 # perform work that needs to be done immediately before resumption
513 ALIGN
514 RING0_PTREGS_FRAME # can't unwind into user space anyway
515 work_pending:
516 testb $_TIF_NEED_RESCHED, %cl
517 jz work_notifysig
518 work_resched:
519 call schedule
520 LOCKDEP_SYS_EXIT
521 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
522 # setting need_resched or sigpending
523 # between sampling and the iret
524 TRACE_IRQS_OFF
525 movl TI_flags(%ebp), %ecx
526 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
527 # than syscall tracing?
528 jz restore_all
529 testb $_TIF_NEED_RESCHED, %cl
530 jnz work_resched
531
532 work_notifysig: # deal with pending signals and
533 # notify-resume requests
534 #ifdef CONFIG_VM86
535 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
536 movl %esp, %eax
537 jne work_notifysig_v86 # returning to kernel-space or
538 # vm86-space
539 xorl %edx, %edx
540 call do_notify_resume
541 jmp resume_userspace_sig
542
543 ALIGN
544 work_notifysig_v86:
545 pushl %ecx # save ti_flags for do_notify_resume
546 CFI_ADJUST_CFA_OFFSET 4
547 call save_v86_state # %eax contains pt_regs pointer
548 popl %ecx
549 CFI_ADJUST_CFA_OFFSET -4
550 movl %eax, %esp
551 #else
552 movl %esp, %eax
553 #endif
554 xorl %edx, %edx
555 call do_notify_resume
556 jmp resume_userspace_sig
557 END(work_pending)
558
559 # perform syscall exit tracing
560 ALIGN
561 syscall_trace_entry:
562 movl $-ENOSYS,PT_EAX(%esp)
563 movl %esp, %eax
564 call syscall_trace_enter
565 /* What it returned is what we'll actually use. */
566 cmpl $(nr_syscalls), %eax
567 jnae syscall_call
568 jmp syscall_exit
569 END(syscall_trace_entry)
570
571 # perform syscall exit tracing
572 ALIGN
573 syscall_exit_work:
574 testb $_TIF_WORK_SYSCALL_EXIT, %cl
575 jz work_pending
576 TRACE_IRQS_ON
577 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
578 # schedule() instead
579 movl %esp, %eax
580 call syscall_trace_leave
581 jmp resume_userspace
582 END(syscall_exit_work)
583 CFI_ENDPROC
584
585 RING0_INT_FRAME # can't unwind into user space anyway
586 syscall_fault:
587 GET_THREAD_INFO(%ebp)
588 movl $-EFAULT,PT_EAX(%esp)
589 jmp resume_userspace
590 END(syscall_fault)
591
592 syscall_badsys:
593 movl $-ENOSYS,PT_EAX(%esp)
594 jmp resume_userspace
595 END(syscall_badsys)
596 CFI_ENDPROC
597
598 #define FIXUP_ESPFIX_STACK \
599 /* since we are on a wrong stack, we cant make it a C code :( */ \
600 PER_CPU(gdt_page, %ebx); \
601 GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \
602 addl %esp, %eax; \
603 pushl $__KERNEL_DS; \
604 CFI_ADJUST_CFA_OFFSET 4; \
605 pushl %eax; \
606 CFI_ADJUST_CFA_OFFSET 4; \
607 lss (%esp), %esp; \
608 CFI_ADJUST_CFA_OFFSET -8;
609 #define UNWIND_ESPFIX_STACK \
610 movl %ss, %eax; \
611 /* see if on espfix stack */ \
612 cmpw $__ESPFIX_SS, %ax; \
613 jne 27f; \
614 movl $__KERNEL_DS, %eax; \
615 movl %eax, %ds; \
616 movl %eax, %es; \
617 /* switch to normal stack */ \
618 FIXUP_ESPFIX_STACK; \
619 27:;
620
621 /*
622 * Build the entry stubs and pointer table with some assembler magic.
623 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
624 * single cache line on all modern x86 implementations.
625 */
626 .section .init.rodata,"a"
627 ENTRY(interrupt)
628 .text
629 .p2align 5
630 .p2align CONFIG_X86_L1_CACHE_SHIFT
631 ENTRY(irq_entries_start)
632 RING0_INT_FRAME
633 vector=FIRST_EXTERNAL_VECTOR
634 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
635 .balign 32
636 .rept 7
637 .if vector < NR_VECTORS
638 .if vector <> FIRST_EXTERNAL_VECTOR
639 CFI_ADJUST_CFA_OFFSET -4
640 .endif
641 1: pushl $(~vector+0x80) /* Note: always in signed byte range */
642 CFI_ADJUST_CFA_OFFSET 4
643 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
644 jmp 2f
645 .endif
646 .previous
647 .long 1b
648 .text
649 vector=vector+1
650 .endif
651 .endr
652 2: jmp common_interrupt
653 .endr
654 END(irq_entries_start)
655
656 .previous
657 END(interrupt)
658 .previous
659
660 /*
661 * the CPU automatically disables interrupts when executing an IRQ vector,
662 * so IRQ-flags tracing has to follow that:
663 */
664 .p2align CONFIG_X86_L1_CACHE_SHIFT
665 common_interrupt:
666 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
667 SAVE_ALL
668 TRACE_IRQS_OFF
669 movl %esp,%eax
670 call do_IRQ
671 jmp ret_from_intr
672 ENDPROC(common_interrupt)
673 CFI_ENDPROC
674
675 #define BUILD_INTERRUPT(name, nr) \
676 ENTRY(name) \
677 RING0_INT_FRAME; \
678 pushl $~(nr); \
679 CFI_ADJUST_CFA_OFFSET 4; \
680 SAVE_ALL; \
681 TRACE_IRQS_OFF \
682 movl %esp,%eax; \
683 call smp_##name; \
684 jmp ret_from_intr; \
685 CFI_ENDPROC; \
686 ENDPROC(name)
687
688 /* The include is where all of the SMP etc. interrupts come from */
689 #include "entry_arch.h"
690
691 ENTRY(coprocessor_error)
692 RING0_INT_FRAME
693 pushl $0
694 CFI_ADJUST_CFA_OFFSET 4
695 pushl $do_coprocessor_error
696 CFI_ADJUST_CFA_OFFSET 4
697 jmp error_code
698 CFI_ENDPROC
699 END(coprocessor_error)
700
701 ENTRY(simd_coprocessor_error)
702 RING0_INT_FRAME
703 pushl $0
704 CFI_ADJUST_CFA_OFFSET 4
705 pushl $do_simd_coprocessor_error
706 CFI_ADJUST_CFA_OFFSET 4
707 jmp error_code
708 CFI_ENDPROC
709 END(simd_coprocessor_error)
710
711 ENTRY(device_not_available)
712 RING0_INT_FRAME
713 pushl $-1 # mark this as an int
714 CFI_ADJUST_CFA_OFFSET 4
715 pushl $do_device_not_available
716 CFI_ADJUST_CFA_OFFSET 4
717 jmp error_code
718 CFI_ENDPROC
719 END(device_not_available)
720
721 #ifdef CONFIG_PARAVIRT
722 ENTRY(native_iret)
723 iret
724 .section __ex_table,"a"
725 .align 4
726 .long native_iret, iret_exc
727 .previous
728 END(native_iret)
729
730 ENTRY(native_irq_enable_sysexit)
731 sti
732 sysexit
733 END(native_irq_enable_sysexit)
734 #endif
735
736 ENTRY(overflow)
737 RING0_INT_FRAME
738 pushl $0
739 CFI_ADJUST_CFA_OFFSET 4
740 pushl $do_overflow
741 CFI_ADJUST_CFA_OFFSET 4
742 jmp error_code
743 CFI_ENDPROC
744 END(overflow)
745
746 ENTRY(bounds)
747 RING0_INT_FRAME
748 pushl $0
749 CFI_ADJUST_CFA_OFFSET 4
750 pushl $do_bounds
751 CFI_ADJUST_CFA_OFFSET 4
752 jmp error_code
753 CFI_ENDPROC
754 END(bounds)
755
756 ENTRY(invalid_op)
757 RING0_INT_FRAME
758 pushl $0
759 CFI_ADJUST_CFA_OFFSET 4
760 pushl $do_invalid_op
761 CFI_ADJUST_CFA_OFFSET 4
762 jmp error_code
763 CFI_ENDPROC
764 END(invalid_op)
765
766 ENTRY(coprocessor_segment_overrun)
767 RING0_INT_FRAME
768 pushl $0
769 CFI_ADJUST_CFA_OFFSET 4
770 pushl $do_coprocessor_segment_overrun
771 CFI_ADJUST_CFA_OFFSET 4
772 jmp error_code
773 CFI_ENDPROC
774 END(coprocessor_segment_overrun)
775
776 ENTRY(invalid_TSS)
777 RING0_EC_FRAME
778 pushl $do_invalid_TSS
779 CFI_ADJUST_CFA_OFFSET 4
780 jmp error_code
781 CFI_ENDPROC
782 END(invalid_TSS)
783
784 ENTRY(segment_not_present)
785 RING0_EC_FRAME
786 pushl $do_segment_not_present
787 CFI_ADJUST_CFA_OFFSET 4
788 jmp error_code
789 CFI_ENDPROC
790 END(segment_not_present)
791
792 ENTRY(stack_segment)
793 RING0_EC_FRAME
794 pushl $do_stack_segment
795 CFI_ADJUST_CFA_OFFSET 4
796 jmp error_code
797 CFI_ENDPROC
798 END(stack_segment)
799
800 ENTRY(alignment_check)
801 RING0_EC_FRAME
802 pushl $do_alignment_check
803 CFI_ADJUST_CFA_OFFSET 4
804 jmp error_code
805 CFI_ENDPROC
806 END(alignment_check)
807
808 ENTRY(divide_error)
809 RING0_INT_FRAME
810 pushl $0 # no error code
811 CFI_ADJUST_CFA_OFFSET 4
812 pushl $do_divide_error
813 CFI_ADJUST_CFA_OFFSET 4
814 jmp error_code
815 CFI_ENDPROC
816 END(divide_error)
817
818 #ifdef CONFIG_X86_MCE
819 ENTRY(machine_check)
820 RING0_INT_FRAME
821 pushl $0
822 CFI_ADJUST_CFA_OFFSET 4
823 pushl machine_check_vector
824 CFI_ADJUST_CFA_OFFSET 4
825 jmp error_code
826 CFI_ENDPROC
827 END(machine_check)
828 #endif
829
830 ENTRY(spurious_interrupt_bug)
831 RING0_INT_FRAME
832 pushl $0
833 CFI_ADJUST_CFA_OFFSET 4
834 pushl $do_spurious_interrupt_bug
835 CFI_ADJUST_CFA_OFFSET 4
836 jmp error_code
837 CFI_ENDPROC
838 END(spurious_interrupt_bug)
839
840 ENTRY(kernel_thread_helper)
841 pushl $0 # fake return address for unwinder
842 CFI_STARTPROC
843 movl %edx,%eax
844 push %edx
845 CFI_ADJUST_CFA_OFFSET 4
846 call *%ebx
847 push %eax
848 CFI_ADJUST_CFA_OFFSET 4
849 call do_exit
850 ud2 # padding for call trace
851 CFI_ENDPROC
852 ENDPROC(kernel_thread_helper)
853
854 #ifdef CONFIG_XEN
855 /* Xen doesn't set %esp to be precisely what the normal sysenter
856 entrypoint expects, so fix it up before using the normal path. */
857 ENTRY(xen_sysenter_target)
858 RING0_INT_FRAME
859 addl $5*4, %esp /* remove xen-provided frame */
860 CFI_ADJUST_CFA_OFFSET -5*4
861 jmp sysenter_past_esp
862 CFI_ENDPROC
863
864 ENTRY(xen_hypervisor_callback)
865 CFI_STARTPROC
866 pushl $0
867 CFI_ADJUST_CFA_OFFSET 4
868 SAVE_ALL
869 TRACE_IRQS_OFF
870
871 /* Check to see if we got the event in the critical
872 region in xen_iret_direct, after we've reenabled
873 events and checked for pending events. This simulates
874 iret instruction's behaviour where it delivers a
875 pending interrupt when enabling interrupts. */
876 movl PT_EIP(%esp),%eax
877 cmpl $xen_iret_start_crit,%eax
878 jb 1f
879 cmpl $xen_iret_end_crit,%eax
880 jae 1f
881
882 jmp xen_iret_crit_fixup
883
884 ENTRY(xen_do_upcall)
885 1: mov %esp, %eax
886 call xen_evtchn_do_upcall
887 jmp ret_from_intr
888 CFI_ENDPROC
889 ENDPROC(xen_hypervisor_callback)
890
891 # Hypervisor uses this for application faults while it executes.
892 # We get here for two reasons:
893 # 1. Fault while reloading DS, ES, FS or GS
894 # 2. Fault while executing IRET
895 # Category 1 we fix up by reattempting the load, and zeroing the segment
896 # register if the load fails.
897 # Category 2 we fix up by jumping to do_iret_error. We cannot use the
898 # normal Linux return path in this case because if we use the IRET hypercall
899 # to pop the stack frame we end up in an infinite loop of failsafe callbacks.
900 # We distinguish between categories by maintaining a status value in EAX.
901 ENTRY(xen_failsafe_callback)
902 CFI_STARTPROC
903 pushl %eax
904 CFI_ADJUST_CFA_OFFSET 4
905 movl $1,%eax
906 1: mov 4(%esp),%ds
907 2: mov 8(%esp),%es
908 3: mov 12(%esp),%fs
909 4: mov 16(%esp),%gs
910 testl %eax,%eax
911 popl %eax
912 CFI_ADJUST_CFA_OFFSET -4
913 lea 16(%esp),%esp
914 CFI_ADJUST_CFA_OFFSET -16
915 jz 5f
916 addl $16,%esp
917 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
918 5: pushl $0 # EAX == 0 => Category 1 (Bad segment)
919 CFI_ADJUST_CFA_OFFSET 4
920 SAVE_ALL
921 jmp ret_from_exception
922 CFI_ENDPROC
923
924 .section .fixup,"ax"
925 6: xorl %eax,%eax
926 movl %eax,4(%esp)
927 jmp 1b
928 7: xorl %eax,%eax
929 movl %eax,8(%esp)
930 jmp 2b
931 8: xorl %eax,%eax
932 movl %eax,12(%esp)
933 jmp 3b
934 9: xorl %eax,%eax
935 movl %eax,16(%esp)
936 jmp 4b
937 .previous
938 .section __ex_table,"a"
939 .align 4
940 .long 1b,6b
941 .long 2b,7b
942 .long 3b,8b
943 .long 4b,9b
944 .previous
945 ENDPROC(xen_failsafe_callback)
946
947 #endif /* CONFIG_XEN */
948
949 #ifdef CONFIG_FUNCTION_TRACER
950 #ifdef CONFIG_DYNAMIC_FTRACE
951
952 ENTRY(mcount)
953 ret
954 END(mcount)
955
956 ENTRY(ftrace_caller)
957 pushl %eax
958 pushl %ecx
959 pushl %edx
960 movl 0xc(%esp), %eax
961 movl 0x4(%ebp), %edx
962 subl $MCOUNT_INSN_SIZE, %eax
963
964 .globl ftrace_call
965 ftrace_call:
966 call ftrace_stub
967
968 popl %edx
969 popl %ecx
970 popl %eax
971
972 .globl ftrace_stub
973 ftrace_stub:
974 ret
975 END(ftrace_caller)
976
977 #else /* ! CONFIG_DYNAMIC_FTRACE */
978
979 ENTRY(mcount)
980 cmpl $ftrace_stub, ftrace_trace_function
981 jnz trace
982 .globl ftrace_stub
983 ftrace_stub:
984 ret
985
986 /* taken from glibc */
987 trace:
988 pushl %eax
989 pushl %ecx
990 pushl %edx
991 movl 0xc(%esp), %eax
992 movl 0x4(%ebp), %edx
993 subl $MCOUNT_INSN_SIZE, %eax
994
995 call *ftrace_trace_function
996
997 popl %edx
998 popl %ecx
999 popl %eax
1000
1001 jmp ftrace_stub
1002 END(mcount)
1003 #endif /* CONFIG_DYNAMIC_FTRACE */
1004 #endif /* CONFIG_FUNCTION_TRACER */
1005
1006 .section .rodata,"a"
1007 #include "syscall_table_32.S"
1008
1009 syscall_table_size=(.-sys_call_table)
1010
1011 /*
1012 * Some functions should be protected against kprobes
1013 */
1014 .pushsection .kprobes.text, "ax"
1015
1016 ENTRY(page_fault)
1017 RING0_EC_FRAME
1018 pushl $do_page_fault
1019 CFI_ADJUST_CFA_OFFSET 4
1020 ALIGN
1021 error_code:
1022 /* the function address is in %fs's slot on the stack */
1023 pushl %es
1024 CFI_ADJUST_CFA_OFFSET 4
1025 /*CFI_REL_OFFSET es, 0*/
1026 pushl %ds
1027 CFI_ADJUST_CFA_OFFSET 4
1028 /*CFI_REL_OFFSET ds, 0*/
1029 pushl %eax
1030 CFI_ADJUST_CFA_OFFSET 4
1031 CFI_REL_OFFSET eax, 0
1032 pushl %ebp
1033 CFI_ADJUST_CFA_OFFSET 4
1034 CFI_REL_OFFSET ebp, 0
1035 pushl %edi
1036 CFI_ADJUST_CFA_OFFSET 4
1037 CFI_REL_OFFSET edi, 0
1038 pushl %esi
1039 CFI_ADJUST_CFA_OFFSET 4
1040 CFI_REL_OFFSET esi, 0
1041 pushl %edx
1042 CFI_ADJUST_CFA_OFFSET 4
1043 CFI_REL_OFFSET edx, 0
1044 pushl %ecx
1045 CFI_ADJUST_CFA_OFFSET 4
1046 CFI_REL_OFFSET ecx, 0
1047 pushl %ebx
1048 CFI_ADJUST_CFA_OFFSET 4
1049 CFI_REL_OFFSET ebx, 0
1050 cld
1051 pushl %fs
1052 CFI_ADJUST_CFA_OFFSET 4
1053 /*CFI_REL_OFFSET fs, 0*/
1054 movl $(__KERNEL_PERCPU), %ecx
1055 movl %ecx, %fs
1056 UNWIND_ESPFIX_STACK
1057 popl %ecx
1058 CFI_ADJUST_CFA_OFFSET -4
1059 /*CFI_REGISTER es, ecx*/
1060 movl PT_FS(%esp), %edi # get the function address
1061 movl PT_ORIG_EAX(%esp), %edx # get the error code
1062 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1063 mov %ecx, PT_FS(%esp)
1064 /*CFI_REL_OFFSET fs, ES*/
1065 movl $(__USER_DS), %ecx
1066 movl %ecx, %ds
1067 movl %ecx, %es
1068 TRACE_IRQS_OFF
1069 movl %esp,%eax # pt_regs pointer
1070 call *%edi
1071 jmp ret_from_exception
1072 CFI_ENDPROC
1073 END(page_fault)
1074
1075 /*
1076 * Debug traps and NMI can happen at the one SYSENTER instruction
1077 * that sets up the real kernel stack. Check here, since we can't
1078 * allow the wrong stack to be used.
1079 *
1080 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1081 * already pushed 3 words if it hits on the sysenter instruction:
1082 * eflags, cs and eip.
1083 *
1084 * We just load the right stack, and push the three (known) values
1085 * by hand onto the new stack - while updating the return eip past
1086 * the instruction that would have done it for sysenter.
1087 */
1088 #define FIX_STACK(offset, ok, label) \
1089 cmpw $__KERNEL_CS,4(%esp); \
1090 jne ok; \
1091 label: \
1092 movl TSS_sysenter_sp0+offset(%esp),%esp; \
1093 CFI_DEF_CFA esp, 0; \
1094 CFI_UNDEFINED eip; \
1095 pushfl; \
1096 CFI_ADJUST_CFA_OFFSET 4; \
1097 pushl $__KERNEL_CS; \
1098 CFI_ADJUST_CFA_OFFSET 4; \
1099 pushl $sysenter_past_esp; \
1100 CFI_ADJUST_CFA_OFFSET 4; \
1101 CFI_REL_OFFSET eip, 0
1102
1103 ENTRY(debug)
1104 RING0_INT_FRAME
1105 cmpl $ia32_sysenter_target,(%esp)
1106 jne debug_stack_correct
1107 FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn)
1108 debug_stack_correct:
1109 pushl $-1 # mark this as an int
1110 CFI_ADJUST_CFA_OFFSET 4
1111 SAVE_ALL
1112 TRACE_IRQS_OFF
1113 xorl %edx,%edx # error code 0
1114 movl %esp,%eax # pt_regs pointer
1115 call do_debug
1116 jmp ret_from_exception
1117 CFI_ENDPROC
1118 END(debug)
1119
1120 /*
1121 * NMI is doubly nasty. It can happen _while_ we're handling
1122 * a debug fault, and the debug fault hasn't yet been able to
1123 * clear up the stack. So we first check whether we got an
1124 * NMI on the sysenter entry path, but after that we need to
1125 * check whether we got an NMI on the debug path where the debug
1126 * fault happened on the sysenter path.
1127 */
1128 ENTRY(nmi)
1129 RING0_INT_FRAME
1130 pushl %eax
1131 CFI_ADJUST_CFA_OFFSET 4
1132 movl %ss, %eax
1133 cmpw $__ESPFIX_SS, %ax
1134 popl %eax
1135 CFI_ADJUST_CFA_OFFSET -4
1136 je nmi_espfix_stack
1137 cmpl $ia32_sysenter_target,(%esp)
1138 je nmi_stack_fixup
1139 pushl %eax
1140 CFI_ADJUST_CFA_OFFSET 4
1141 movl %esp,%eax
1142 /* Do not access memory above the end of our stack page,
1143 * it might not exist.
1144 */
1145 andl $(THREAD_SIZE-1),%eax
1146 cmpl $(THREAD_SIZE-20),%eax
1147 popl %eax
1148 CFI_ADJUST_CFA_OFFSET -4
1149 jae nmi_stack_correct
1150 cmpl $ia32_sysenter_target,12(%esp)
1151 je nmi_debug_stack_check
1152 nmi_stack_correct:
1153 /* We have a RING0_INT_FRAME here */
1154 pushl %eax
1155 CFI_ADJUST_CFA_OFFSET 4
1156 SAVE_ALL
1157 TRACE_IRQS_OFF
1158 xorl %edx,%edx # zero error code
1159 movl %esp,%eax # pt_regs pointer
1160 call do_nmi
1161 jmp restore_nocheck_notrace
1162 CFI_ENDPROC
1163
1164 nmi_stack_fixup:
1165 RING0_INT_FRAME
1166 FIX_STACK(12,nmi_stack_correct, 1)
1167 jmp nmi_stack_correct
1168
1169 nmi_debug_stack_check:
1170 /* We have a RING0_INT_FRAME here */
1171 cmpw $__KERNEL_CS,16(%esp)
1172 jne nmi_stack_correct
1173 cmpl $debug,(%esp)
1174 jb nmi_stack_correct
1175 cmpl $debug_esp_fix_insn,(%esp)
1176 ja nmi_stack_correct
1177 FIX_STACK(24,nmi_stack_correct, 1)
1178 jmp nmi_stack_correct
1179
1180 nmi_espfix_stack:
1181 /* We have a RING0_INT_FRAME here.
1182 *
1183 * create the pointer to lss back
1184 */
1185 pushl %ss
1186 CFI_ADJUST_CFA_OFFSET 4
1187 pushl %esp
1188 CFI_ADJUST_CFA_OFFSET 4
1189 addw $4, (%esp)
1190 /* copy the iret frame of 12 bytes */
1191 .rept 3
1192 pushl 16(%esp)
1193 CFI_ADJUST_CFA_OFFSET 4
1194 .endr
1195 pushl %eax
1196 CFI_ADJUST_CFA_OFFSET 4
1197 SAVE_ALL
1198 TRACE_IRQS_OFF
1199 FIXUP_ESPFIX_STACK # %eax == %esp
1200 xorl %edx,%edx # zero error code
1201 call do_nmi
1202 RESTORE_REGS
1203 lss 12+4(%esp), %esp # back to espfix stack
1204 CFI_ADJUST_CFA_OFFSET -24
1205 jmp irq_return
1206 CFI_ENDPROC
1207 END(nmi)
1208
1209 ENTRY(int3)
1210 RING0_INT_FRAME
1211 pushl $-1 # mark this as an int
1212 CFI_ADJUST_CFA_OFFSET 4
1213 SAVE_ALL
1214 TRACE_IRQS_OFF
1215 xorl %edx,%edx # zero error code
1216 movl %esp,%eax # pt_regs pointer
1217 call do_int3
1218 jmp ret_from_exception
1219 CFI_ENDPROC
1220 END(int3)
1221
1222 ENTRY(general_protection)
1223 RING0_EC_FRAME
1224 pushl $do_general_protection
1225 CFI_ADJUST_CFA_OFFSET 4
1226 jmp error_code
1227 CFI_ENDPROC
1228 END(general_protection)
1229
1230 /*
1231 * End of kprobes section
1232 */
1233 .popsection
This page took 0.078513 seconds and 5 git commands to generate.