2 * linux/arch/i386/entry.S
4 * Copyright (C) 1991, 1992 Linus Torvalds
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.
12 * NOTE: This code handles signal-recognition, which happens every time
13 * after a timer-interrupt and after each system call.
15 * I changed all the .align's to 4 (16 byte alignment), as that's faster
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
40 * "current" is in register %ebx during any slow entries.
43 #include <linux/config.h>
44 #include <linux/linkage.h>
45 #include <asm/thread_info.h>
46 #include <asm/errno.h>
47 #include <asm/segment.h>
51 #include "irq_vectors.h"
53 #define nr_syscalls ((syscall_table_size)/4)
79 #define preempt_stop cli
82 #define resume_kernel restore_nocheck
96 movl $(__USER_DS), %edx; \
100 #define RESTORE_INT_REGS \
109 #define RESTORE_REGS \
113 .section .fixup,"ax"; \
119 .section __ex_table,"a";\
129 GET_THREAD_INFO(%ebp)
134 * Return to user mode is not as complex as all this looks,
135 * but we want the default path for a system call return to
136 * go as quickly as possible which is why some of this is
137 * less clear than it otherwise should be.
140 # userspace resumption stub bypassing syscall exit tracing
145 GET_THREAD_INFO(%ebp)
146 movl EFLAGS(%esp), %eax # mix EFLAGS and CS
148 testl $(VM_MASK | 3), %eax
150 ENTRY(resume_userspace)
151 cli # make sure we don't miss an interrupt
152 # setting need_resched or sigpending
153 # between sampling and the iret
154 movl TI_flags(%ebp), %ecx
155 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
156 # int/exception return?
160 #ifdef CONFIG_PREEMPT
163 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
166 movl TI_flags(%ebp), %ecx # need_resched set ?
167 testb $_TIF_NEED_RESCHED, %cl
169 testl $IF_MASK,EFLAGS(%esp) # interrupts off (exception path) ?
171 call preempt_schedule_irq
175 /* SYSENTER_RETURN points to after the "sysenter" instruction in
176 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
178 # sysenter call handler stub
179 ENTRY(sysenter_entry)
180 movl TSS_sysenter_esp0(%esp),%esp
187 pushl $SYSENTER_RETURN
190 * Load the potential sixth argument from user stack.
191 * Careful about security.
193 cmpl $__PAGE_OFFSET-3,%ebp
196 .section __ex_table,"a"
198 .long 1b,syscall_fault
203 GET_THREAD_INFO(%ebp)
205 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
206 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
207 jnz syscall_trace_entry
208 cmpl $(nr_syscalls), %eax
210 call *sys_call_table(,%eax,4)
213 movl TI_flags(%ebp), %ecx
214 testw $_TIF_ALLWORK_MASK, %cx
215 jne syscall_exit_work
216 /* if something modifies registers it must also disable sysexit */
218 movl OLDESP(%esp), %ecx
224 # system call handler stub
226 pushl %eax # save orig_eax
228 GET_THREAD_INFO(%ebp)
229 testl $TF_MASK,EFLAGS(%esp)
231 orl $_TIF_SINGLESTEP,TI_flags(%ebp)
233 # system call tracing in operation / emulation
234 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
235 testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
236 jnz syscall_trace_entry
237 cmpl $(nr_syscalls), %eax
240 call *sys_call_table(,%eax,4)
241 movl %eax,EAX(%esp) # store the return value
243 cli # make sure we don't miss an interrupt
244 # setting need_resched or sigpending
245 # between sampling and the iret
246 movl TI_flags(%ebp), %ecx
247 testw $_TIF_ALLWORK_MASK, %cx # current->work
248 jne syscall_exit_work
251 movl EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
252 # Warning: OLDSS(%esp) contains the wrong/random values if we
253 # are returning to the kernel.
254 # See comments in process.c:copy_thread() for details.
255 movb OLDSS(%esp), %ah
257 andl $(VM_MASK | (4 << 8) | 3), %eax
258 cmpl $((4 << 8) | 3), %eax
259 je ldt_ss # returning to user-space with LDT SS
267 pushl $0 # no error code
271 .section __ex_table,"a"
277 larl OLDSS(%esp), %eax
279 testl $0x00400000, %eax # returning to 32bit stack?
280 jnz restore_nocheck # allright, normal return
281 /* If returning to userspace with 16bit stack,
282 * try to fix the higher word of ESP, as the CPU
284 * This is an "official" bug of all the x86-compatible
285 * CPUs, which we can try to work around to make
286 * dosemu and wine happy. */
287 subl $8, %esp # reserve space for switch16 pointer
290 /* Set up the 16bit stack frame with switch32 pointer on top,
291 * and a switch16 pointer on top of the current frame. */
292 call setup_x86_bogus_stack
294 lss 20+4(%esp), %esp # switch to 16bit stack
296 .section __ex_table,"a"
301 # perform work that needs to be done immediately before resumption
304 testb $_TIF_NEED_RESCHED, %cl
308 cli # make sure we don't miss an interrupt
309 # setting need_resched or sigpending
310 # between sampling and the iret
311 movl TI_flags(%ebp), %ecx
312 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
313 # than syscall tracing?
315 testb $_TIF_NEED_RESCHED, %cl
318 work_notifysig: # deal with pending signals and
319 # notify-resume requests
320 testl $VM_MASK, EFLAGS(%esp)
322 jne work_notifysig_v86 # returning to kernel-space or
325 call do_notify_resume
331 pushl %ecx # save ti_flags for do_notify_resume
332 call save_v86_state # %eax contains pt_regs pointer
336 call do_notify_resume
340 # perform syscall exit tracing
343 movl $-ENOSYS,EAX(%esp)
346 call do_syscall_trace
348 jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU,
349 # so must skip actual syscall
350 movl ORIG_EAX(%esp), %eax
351 cmpl $(nr_syscalls), %eax
355 # perform syscall exit tracing
358 testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl
360 sti # could let do_syscall_trace() call
364 call do_syscall_trace
369 pushl %eax # save orig_eax
371 GET_THREAD_INFO(%ebp)
372 movl $-EFAULT,EAX(%esp)
377 movl $-ENOSYS,EAX(%esp)
380 #define FIXUP_ESPFIX_STACK \
382 /* switch to 32bit stack using the pointer on top of 16bit stack */ \
383 lss %ss:CPU_16BIT_STACK_SIZE-8, %esp; \
384 /* copy data from 16bit stack to 32bit stack */ \
385 call fixup_x86_bogus_stack; \
386 /* put ESP to the proper location */ \
388 #define UNWIND_ESPFIX_STACK \
391 /* see if on 16bit stack */ \
392 cmpw $__ESPFIX_SS, %ax; \
394 movl $__KERNEL_DS, %edx; \
397 /* switch to 32bit stack */ \
402 * Build the entry stubs and pointer table with
403 * some assembler magic.
410 ENTRY(irq_entries_start)
428 #define BUILD_INTERRUPT(name, nr) \
436 /* The include is where all of the SMP etc. interrupts come from */
437 #include "entry_arch.h"
440 pushl $0 # no error code
441 pushl $do_divide_error
458 movl ES(%esp), %edi # get the function address
459 movl ORIG_EAX(%esp), %edx # get the error code
460 movl %eax, ORIG_EAX(%esp)
462 movl $(__USER_DS), %ecx
465 movl %esp,%eax # pt_regs pointer
467 jmp ret_from_exception
469 ENTRY(coprocessor_error)
471 pushl $do_coprocessor_error
474 ENTRY(simd_coprocessor_error)
476 pushl $do_simd_coprocessor_error
479 ENTRY(device_not_available)
480 pushl $-1 # mark this as an int
483 testl $0x4, %eax # EM (math emulation bit)
484 jne device_not_available_emulate
486 call math_state_restore
487 jmp ret_from_exception
488 device_not_available_emulate:
489 pushl $0 # temporary storage for ORIG_EIP
492 jmp ret_from_exception
495 * Debug traps and NMI can happen at the one SYSENTER instruction
496 * that sets up the real kernel stack. Check here, since we can't
497 * allow the wrong stack to be used.
499 * "TSS_sysenter_esp0+12" is because the NMI/debug handler will have
500 * already pushed 3 words if it hits on the sysenter instruction:
501 * eflags, cs and eip.
503 * We just load the right stack, and push the three (known) values
504 * by hand onto the new stack - while updating the return eip past
505 * the instruction that would have done it for sysenter.
507 #define FIX_STACK(offset, ok, label) \
508 cmpw $__KERNEL_CS,4(%esp); \
511 movl TSS_sysenter_esp0+offset(%esp),%esp; \
513 pushl $__KERNEL_CS; \
514 pushl $sysenter_past_esp
517 cmpl $sysenter_entry,(%esp)
518 jne debug_stack_correct
519 FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn)
521 pushl $-1 # mark this as an int
523 xorl %edx,%edx # error code 0
524 movl %esp,%eax # pt_regs pointer
526 jmp ret_from_exception
529 * NMI is doubly nasty. It can happen _while_ we're handling
530 * a debug fault, and the debug fault hasn't yet been able to
531 * clear up the stack. So we first check whether we got an
532 * NMI on the sysenter entry path, but after that we need to
533 * check whether we got an NMI on the debug path where the debug
534 * fault happened on the sysenter path.
539 cmpw $__ESPFIX_SS, %ax
542 cmpl $sysenter_entry,(%esp)
546 /* Do not access memory above the end of our stack page,
547 * it might not exist.
549 andl $(THREAD_SIZE-1),%eax
550 cmpl $(THREAD_SIZE-20),%eax
552 jae nmi_stack_correct
553 cmpl $sysenter_entry,12(%esp)
554 je nmi_debug_stack_check
558 xorl %edx,%edx # zero error code
559 movl %esp,%eax # pt_regs pointer
564 FIX_STACK(12,nmi_stack_correct, 1)
565 jmp nmi_stack_correct
566 nmi_debug_stack_check:
567 cmpw $__KERNEL_CS,16(%esp)
568 jne nmi_stack_correct
571 cmpl $debug_esp_fix_insn,(%esp)
573 FIX_STACK(24,nmi_stack_correct, 1)
574 jmp nmi_stack_correct
577 /* create the pointer to lss back */
582 /* copy the iret frame of 12 bytes */
588 FIXUP_ESPFIX_STACK # %eax == %esp
589 xorl %edx,%edx # zero error code
592 lss 12+4(%esp), %esp # back to 16bit stack
594 .section __ex_table,"a"
600 pushl $-1 # mark this as an int
602 xorl %edx,%edx # zero error code
603 movl %esp,%eax # pt_regs pointer
605 jmp ret_from_exception
623 ENTRY(coprocessor_segment_overrun)
625 pushl $do_coprocessor_segment_overrun
629 pushl $do_invalid_TSS
632 ENTRY(segment_not_present)
633 pushl $do_segment_not_present
637 pushl $do_stack_segment
640 KPROBE_ENTRY(general_protection)
641 pushl $do_general_protection
645 ENTRY(alignment_check)
646 pushl $do_alignment_check
649 KPROBE_ENTRY(page_fault)
654 #ifdef CONFIG_X86_MCE
657 pushl machine_check_vector
661 ENTRY(spurious_interrupt_bug)
663 pushl $do_spurious_interrupt_bug
667 #include "syscall_table.S"
669 syscall_table_size=(.-sys_call_table)