projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
x86: add %gs accessors for x86_32
[linux-2.6-block.git] / arch / x86 / kernel / entry_32.S
CommitLineData
1da177e4 1/*
1da177e4
LT		 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 *
889f21ce 17 * Stack layout in 'syscall_exit':
1da177e4
LT		 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
464d1a78 32 * 24(%esp) - %fs
f95d47ca
JF		 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
1da177e4
LT		 39 *
40 * "current" is in register %ebx during any slow entries.
41 */
42
1da177e4
LT		 43#include <linux/linkage.h>
44#include <asm/thread_info.h>
55f327fa 45#include <asm/irqflags.h>
1da177e4
LT		 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>
be44d2aa 51#include <asm/percpu.h>
fe7cacc1 52#include <asm/dwarf2.h>
ab68ed98 53#include <asm/processor-flags.h>
395a59d0 54#include <asm/ftrace.h>
9b7dc567 55#include <asm/irq_vectors.h>
1da177e4 56
af0575bb
RM		 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
139ec7c4
RR		 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")
d75cd22f 72 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
139ec7c4
RR		 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
1da177e4
LT		 80#define nr_syscalls ((syscall_table_size)/4)
81
1da177e4 82#ifdef CONFIG_PREEMPT
139ec7c4 83#define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
1da177e4 84#else
139ec7c4 85#define preempt_stop(clobbers)
1da177e4
LT		 86#define resume_kernel restore_nocheck
87#endif
88
55f327fa
IM		 89.macro TRACE_IRQS_IRET
90#ifdef CONFIG_TRACE_IRQFLAGS
ab68ed98 91 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off?
55f327fa
IM		 92 jz 1f
93 TRACE_IRQS_ON
941:
95#endif
96.endm
97
4031ff38
AG		 98#ifdef CONFIG_VM86
99#define resume_userspace_sig check_userspace
100#else
101#define resume_userspace_sig resume_userspace
102#endif
103
f0d96110
TH		 104.macro 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
464d1a78 140 movl %edx, %fs
f0d96110 141.endm
1da177e4 142
f0d96110
TH		 143.macro RESTORE_INT_REGS
144 popl %ebx
145 CFI_ADJUST_CFA_OFFSET -4
146 CFI_RESTORE ebx
147 popl %ecx
148 CFI_ADJUST_CFA_OFFSET -4
149 CFI_RESTORE ecx
150 popl %edx
151 CFI_ADJUST_CFA_OFFSET -4
152 CFI_RESTORE edx
153 popl %esi
154 CFI_ADJUST_CFA_OFFSET -4
155 CFI_RESTORE esi
156 popl %edi
157 CFI_ADJUST_CFA_OFFSET -4
158 CFI_RESTORE edi
159 popl %ebp
160 CFI_ADJUST_CFA_OFFSET -4
161 CFI_RESTORE ebp
162 popl %eax
163 CFI_ADJUST_CFA_OFFSET -4
fe7cacc1 164 CFI_RESTORE eax
f0d96110 165.endm
1da177e4 166
f0d96110
TH		 167.macro RESTORE_REGS
168 RESTORE_INT_REGS
1691: popl %ds
170 CFI_ADJUST_CFA_OFFSET -4
171 /*CFI_RESTORE ds;*/
1722: popl %es
173 CFI_ADJUST_CFA_OFFSET -4
174 /*CFI_RESTORE es;*/
1753: popl %fs
176 CFI_ADJUST_CFA_OFFSET -4
177 /*CFI_RESTORE fs;*/
178.pushsection .fixup, "ax"
1794: movl $0, (%esp)
180 jmp 1b
1815: movl $0, (%esp)
182 jmp 2b
1836: movl $0, (%esp)
184 jmp 3b
185.section __ex_table, "a"
186 .align 4
187 .long 1b, 4b
188 .long 2b, 5b
189 .long 3b, 6b
f95d47ca 190.popsection
f0d96110 191.endm
1da177e4 192
f0d96110
TH		 193.macro RING0_INT_FRAME
194 CFI_STARTPROC simple
195 CFI_SIGNAL_FRAME
196 CFI_DEF_CFA esp, 3*4
197 /*CFI_OFFSET cs, -2*4;*/
fe7cacc1 198 CFI_OFFSET eip, -3*4
f0d96110 199.endm
fe7cacc1 200
f0d96110
TH		 201.macro RING0_EC_FRAME
202 CFI_STARTPROC simple
203 CFI_SIGNAL_FRAME
204 CFI_DEF_CFA esp, 4*4
205 /*CFI_OFFSET cs, -2*4;*/
fe7cacc1 206 CFI_OFFSET eip, -3*4
f0d96110 207.endm
fe7cacc1 208
f0d96110
TH		 209.macro RING0_PTREGS_FRAME
210 CFI_STARTPROC simple
211 CFI_SIGNAL_FRAME
212 CFI_DEF_CFA esp, PT_OLDESP-PT_EBX
213 /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/
214 CFI_OFFSET eip, PT_EIP-PT_OLDESP
215 /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/
216 /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/
217 CFI_OFFSET eax, PT_EAX-PT_OLDESP
218 CFI_OFFSET ebp, PT_EBP-PT_OLDESP
219 CFI_OFFSET edi, PT_EDI-PT_OLDESP
220 CFI_OFFSET esi, PT_ESI-PT_OLDESP
221 CFI_OFFSET edx, PT_EDX-PT_OLDESP
222 CFI_OFFSET ecx, PT_ECX-PT_OLDESP
eb5b7b9d 223 CFI_OFFSET ebx, PT_EBX-PT_OLDESP
f0d96110 224.endm
1da177e4
LT		 225
226ENTRY(ret_from_fork)
fe7cacc1 227 CFI_STARTPROC
1da177e4 228 pushl %eax
25d7dfda 229 CFI_ADJUST_CFA_OFFSET 4
1da177e4
LT		 230 call schedule_tail
231 GET_THREAD_INFO(%ebp)
232 popl %eax
fe7cacc1 233 CFI_ADJUST_CFA_OFFSET -4
47a5c6fa
LT		 234 pushl $0x0202 # Reset kernel eflags
235 CFI_ADJUST_CFA_OFFSET 4
236 popfl
237 CFI_ADJUST_CFA_OFFSET -4
1da177e4 238 jmp syscall_exit
fe7cacc1 239 CFI_ENDPROC
47a55cd7 240END(ret_from_fork)
1da177e4
LT		 241
242/*
243 * Return to user mode is not as complex as all this looks,
244 * but we want the default path for a system call return to
245 * go as quickly as possible which is why some of this is
246 * less clear than it otherwise should be.
247 */
248
249 # userspace resumption stub bypassing syscall exit tracing
250 ALIGN
fe7cacc1 251 RING0_PTREGS_FRAME
1da177e4 252ret_from_exception:
139ec7c4 253 preempt_stop(CLBR_ANY)
1da177e4
LT		 254ret_from_intr:
255 GET_THREAD_INFO(%ebp)
4031ff38 256check_userspace:
eb5b7b9d
JF		 257 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
258 movb PT_CS(%esp), %al
ab68ed98 259 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
78be3706
RR		 260 cmpl $USER_RPL, %eax
261 jb resume_kernel # not returning to v8086 or userspace
f95d47ca 262
1da177e4 263ENTRY(resume_userspace)
c7e872e7 264 LOCKDEP_SYS_EXIT
139ec7c4 265 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
1da177e4
LT		 266 # setting need_resched or sigpending
267 # between sampling and the iret
e32e58a9 268 TRACE_IRQS_OFF
1da177e4
LT		 269 movl TI_flags(%ebp), %ecx
270 andl $_TIF_WORK_MASK, %ecx # is there any work to be done on
271 # int/exception return?
272 jne work_pending
273 jmp restore_all
47a55cd7 274END(ret_from_exception)
1da177e4
LT		 275
276#ifdef CONFIG_PREEMPT
277ENTRY(resume_kernel)
139ec7c4 278 DISABLE_INTERRUPTS(CLBR_ANY)
1da177e4
LT		 279 cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ?
280 jnz restore_nocheck
281need_resched:
282 movl TI_flags(%ebp), %ecx # need_resched set ?
283 testb $_TIF_NEED_RESCHED, %cl
284 jz restore_all
ab68ed98 285 testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ?
1da177e4
LT		 286 jz restore_all
287 call preempt_schedule_irq
288 jmp need_resched
47a55cd7 289END(resume_kernel)
1da177e4 290#endif
fe7cacc1 291 CFI_ENDPROC
1da177e4
LT		 292
293/* SYSENTER_RETURN points to after the "sysenter" instruction in
294 the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */
295
296 # sysenter call handler stub
0aa97fb2 297ENTRY(ia32_sysenter_target)
fe7cacc1 298 CFI_STARTPROC simple
adf14236 299 CFI_SIGNAL_FRAME
fe7cacc1
JB		 300 CFI_DEF_CFA esp, 0
301 CFI_REGISTER esp, ebp
faca6227 302 movl TSS_sysenter_sp0(%esp),%esp
1da177e4 303sysenter_past_esp:
55f327fa 304 /*
d93c870b
JF		 305 * Interrupts are disabled here, but we can't trace it until
306 * enough kernel state to call TRACE_IRQS_OFF can be called - but
307 * we immediately enable interrupts at that point anyway.
55f327fa 308 */
1da177e4 309 pushl $(__USER_DS)
fe7cacc1
JB		 310 CFI_ADJUST_CFA_OFFSET 4
311 /*CFI_REL_OFFSET ss, 0*/
1da177e4 312 pushl %ebp
fe7cacc1
JB		 313 CFI_ADJUST_CFA_OFFSET 4
314 CFI_REL_OFFSET esp, 0
1da177e4 315 pushfl
d93c870b 316 orl $X86_EFLAGS_IF, (%esp)
fe7cacc1 317 CFI_ADJUST_CFA_OFFSET 4
1da177e4 318 pushl $(__USER_CS)
fe7cacc1
JB		 319 CFI_ADJUST_CFA_OFFSET 4
320 /*CFI_REL_OFFSET cs, 0*/
e6e5494c
IM		 321 /*
322 * Push current_thread_info()->sysenter_return to the stack.
323 * A tiny bit of offset fixup is necessary - 4*4 means the 4 words
324 * pushed above; +8 corresponds to copy_thread's esp0 setting.
325 */
326 pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp)
fe7cacc1
JB		 327 CFI_ADJUST_CFA_OFFSET 4
328 CFI_REL_OFFSET eip, 0
1da177e4 329
d93c870b
JF		 330 pushl %eax
331 CFI_ADJUST_CFA_OFFSET 4
332 SAVE_ALL
333 ENABLE_INTERRUPTS(CLBR_NONE)
334
1da177e4
LT		 335/*
336 * Load the potential sixth argument from user stack.
337 * Careful about security.
338 */
339 cmpl $__PAGE_OFFSET-3,%ebp
340 jae syscall_fault
3411: movl (%ebp),%ebp
d93c870b 342 movl %ebp,PT_EBP(%esp)
1da177e4
LT		 343.section __ex_table,"a"
344 .align 4
345 .long 1b,syscall_fault
346.previous
347
1da177e4
LT		 348 GET_THREAD_INFO(%ebp)
349
350 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
d4d67150 351 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
af0575bb
RM		 352 jnz sysenter_audit
353sysenter_do_call:
1da177e4
LT		 354 cmpl $(nr_syscalls), %eax
355 jae syscall_badsys
356 call *sys_call_table(,%eax,4)
eb5b7b9d 357 movl %eax,PT_EAX(%esp)
c7e872e7 358 LOCKDEP_SYS_EXIT
42c24fa2 359 DISABLE_INTERRUPTS(CLBR_ANY)
55f327fa 360 TRACE_IRQS_OFF
1da177e4
LT		 361 movl TI_flags(%ebp), %ecx
362 testw $_TIF_ALLWORK_MASK, %cx
af0575bb
RM		 363 jne sysexit_audit
364sysenter_exit:
1da177e4 365/* if something modifies registers it must also disable sysexit */
eb5b7b9d
JF		 366 movl PT_EIP(%esp), %edx
367 movl PT_OLDESP(%esp), %ecx
1da177e4 368 xorl %ebp,%ebp
55f327fa 369 TRACE_IRQS_ON
464d1a78 3701: mov PT_FS(%esp), %fs
d75cd22f 371 ENABLE_INTERRUPTS_SYSEXIT
af0575bb
RM		 372
373#ifdef CONFIG_AUDITSYSCALL
374sysenter_audit:
375 testw $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp)
376 jnz syscall_trace_entry
377 addl $4,%esp
378 CFI_ADJUST_CFA_OFFSET -4
379 /* %esi already in 8(%esp) 6th arg: 4th syscall arg */
380 /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */
381 /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */
382 movl %ebx,%ecx /* 3rd arg: 1st syscall arg */
383 movl %eax,%edx /* 2nd arg: syscall number */
384 movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */
385 call audit_syscall_entry
386 pushl %ebx
387 CFI_ADJUST_CFA_OFFSET 4
388 movl PT_EAX(%esp),%eax /* reload syscall number */
389 jmp sysenter_do_call
390
391sysexit_audit:
392 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
393 jne syscall_exit_work
394 TRACE_IRQS_ON
395 ENABLE_INTERRUPTS(CLBR_ANY)
396 movl %eax,%edx /* second arg, syscall return value */
397 cmpl $0,%eax /* is it < 0? */
398 setl %al /* 1 if so, 0 if not */
399 movzbl %al,%eax /* zero-extend that */
400 inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
401 call audit_syscall_exit
402 DISABLE_INTERRUPTS(CLBR_ANY)
403 TRACE_IRQS_OFF
404 movl TI_flags(%ebp), %ecx
405 testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx
406 jne syscall_exit_work
407 movl PT_EAX(%esp),%eax /* reload syscall return value */
408 jmp sysenter_exit
409#endif
410
fe7cacc1 411 CFI_ENDPROC
f95d47ca 412.pushsection .fixup,"ax"
464d1a78 4132: movl $0,PT_FS(%esp)
f95d47ca
JF		 414 jmp 1b
415.section __ex_table,"a"
416 .align 4
417 .long 1b,2b
418.popsection
0aa97fb2 419ENDPROC(ia32_sysenter_target)
1da177e4
LT		 420
421 # system call handler stub
422ENTRY(system_call)
fe7cacc1 423 RING0_INT_FRAME # can't unwind into user space anyway
1da177e4 424 pushl %eax # save orig_eax
fe7cacc1 425 CFI_ADJUST_CFA_OFFSET 4
1da177e4
LT		 426 SAVE_ALL
427 GET_THREAD_INFO(%ebp)
ed75e8d5 428 # system call tracing in operation / emulation
1da177e4 429 /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */
d4d67150 430 testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp)
1da177e4
LT		 431 jnz syscall_trace_entry
432 cmpl $(nr_syscalls), %eax
433 jae syscall_badsys
434syscall_call:
435 call *sys_call_table(,%eax,4)
eb5b7b9d 436 movl %eax,PT_EAX(%esp) # store the return value
1da177e4 437syscall_exit:
c7e872e7 438 LOCKDEP_SYS_EXIT
139ec7c4 439 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
1da177e4
LT		 440 # setting need_resched or sigpending
441 # between sampling and the iret
55f327fa 442 TRACE_IRQS_OFF
1da177e4
LT		 443 movl TI_flags(%ebp), %ecx
444 testw $_TIF_ALLWORK_MASK, %cx # current->work
445 jne syscall_exit_work
446
447restore_all:
eb5b7b9d
JF		 448 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
449 # Warning: PT_OLDSS(%esp) contains the wrong/random values if we
5df24082
SS		 450 # are returning to the kernel.
451 # See comments in process.c:copy_thread() for details.
eb5b7b9d
JF		 452 movb PT_OLDSS(%esp), %ah
453 movb PT_CS(%esp), %al
ab68ed98 454 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
78be3706 455 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
fe7cacc1 456 CFI_REMEMBER_STATE
1da177e4
LT		 457 je ldt_ss # returning to user-space with LDT SS
458restore_nocheck:
55f327fa
IM		 459 TRACE_IRQS_IRET
460restore_nocheck_notrace:
1da177e4 461 RESTORE_REGS
f95d47ca 462 addl $4, %esp # skip orig_eax/error_code
fe7cacc1 463 CFI_ADJUST_CFA_OFFSET -4
f7f3d791 464irq_return:
3701d863 465 INTERRUPT_RETURN
1da177e4 466.section .fixup,"ax"
90e9f536 467ENTRY(iret_exc)
a879cbbb
LT		 468 pushl $0 # no error code
469 pushl $do_iret_error
470 jmp error_code
1da177e4
LT		 471.previous
472.section __ex_table,"a"
473 .align 4
3701d863 474 .long irq_return,iret_exc
1da177e4
LT		 475.previous
476
fe7cacc1 477 CFI_RESTORE_STATE
1da177e4 478ldt_ss:
eb5b7b9d 479 larl PT_OLDSS(%esp), %eax
1da177e4
LT		 480 jnz restore_nocheck
481 testl $0x00400000, %eax # returning to 32bit stack?
482 jnz restore_nocheck # allright, normal return
d3561b7f
RR		 483
484#ifdef CONFIG_PARAVIRT
485 /*
486 * The kernel can't run on a non-flat stack if paravirt mode
487 * is active. Rather than try to fixup the high bits of
488 * ESP, bypass this code entirely. This may break DOSemu
489 * and/or Wine support in a paravirt VM, although the option
490 * is still available to implement the setting of the high
491 * 16-bits in the INTERRUPT_RETURN paravirt-op.
492 */
93b1eab3 493 cmpl $0, pv_info+PARAVIRT_enabled
d3561b7f
RR		 494 jne restore_nocheck
495#endif
496
1da177e4
LT		 497 /* If returning to userspace with 16bit stack,
498 * try to fix the higher word of ESP, as the CPU
499 * won't restore it.
500 * This is an "official" bug of all the x86-compatible
501 * CPUs, which we can try to work around to make
502 * dosemu and wine happy. */
eb5b7b9d 503 movl PT_OLDESP(%esp), %eax
be44d2aa
SS		 504 movl %esp, %edx
505 call patch_espfix_desc
506 pushl $__ESPFIX_SS
507 CFI_ADJUST_CFA_OFFSET 4
508 pushl %eax
509 CFI_ADJUST_CFA_OFFSET 4
139ec7c4 510 DISABLE_INTERRUPTS(CLBR_EAX)
55f327fa 511 TRACE_IRQS_OFF
be44d2aa
SS		 512 lss (%esp), %esp
513 CFI_ADJUST_CFA_OFFSET -8
514 jmp restore_nocheck
fe7cacc1 515 CFI_ENDPROC
47a55cd7 516ENDPROC(system_call)
1da177e4
LT		 517
518 # perform work that needs to be done immediately before resumption
519 ALIGN
fe7cacc1 520 RING0_PTREGS_FRAME # can't unwind into user space anyway
1da177e4
LT		 521work_pending:
522 testb $_TIF_NEED_RESCHED, %cl
523 jz work_notifysig
524work_resched:
525 call schedule
c7e872e7 526 LOCKDEP_SYS_EXIT
139ec7c4 527 DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt
1da177e4
LT		 528 # setting need_resched or sigpending
529 # between sampling and the iret
55f327fa 530 TRACE_IRQS_OFF
1da177e4
LT		 531 movl TI_flags(%ebp), %ecx
532 andl $_TIF_WORK_MASK, %ecx # is there any work to be done other
533 # than syscall tracing?
534 jz restore_all
535 testb $_TIF_NEED_RESCHED, %cl
536 jnz work_resched
537
538work_notifysig: # deal with pending signals and
539 # notify-resume requests
74b47a78 540#ifdef CONFIG_VM86
ab68ed98 541 testl $X86_EFLAGS_VM, PT_EFLAGS(%esp)
1da177e4
LT		 542 movl %esp, %eax
543 jne work_notifysig_v86 # returning to kernel-space or
544 # vm86-space
545 xorl %edx, %edx
546 call do_notify_resume
4031ff38 547 jmp resume_userspace_sig
1da177e4
LT		 548
549 ALIGN
550work_notifysig_v86:
551 pushl %ecx # save ti_flags for do_notify_resume
fe7cacc1 552 CFI_ADJUST_CFA_OFFSET 4
1da177e4
LT		 553 call save_v86_state # %eax contains pt_regs pointer
554 popl %ecx
fe7cacc1 555 CFI_ADJUST_CFA_OFFSET -4
1da177e4 556 movl %eax, %esp
74b47a78
JK		 557#else
558 movl %esp, %eax
559#endif
1da177e4
LT		 560 xorl %edx, %edx
561 call do_notify_resume
4031ff38 562 jmp resume_userspace_sig
47a55cd7 563END(work_pending)
1da177e4
LT		 564
565 # perform syscall exit tracing
566 ALIGN
567syscall_trace_entry:
eb5b7b9d 568 movl $-ENOSYS,PT_EAX(%esp)
1da177e4 569 movl %esp, %eax
d4d67150
RM		 570 call syscall_trace_enter
571 /* What it returned is what we'll actually use. */
1da177e4
LT		 572 cmpl $(nr_syscalls), %eax
573 jnae syscall_call
574 jmp syscall_exit
47a55cd7 575END(syscall_trace_entry)
1da177e4
LT		 576
577 # perform syscall exit tracing
578 ALIGN
579syscall_exit_work:
d4d67150 580 testb $_TIF_WORK_SYSCALL_EXIT, %cl
1da177e4 581 jz work_pending
55f327fa 582 TRACE_IRQS_ON
d4d67150 583 ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call
1da177e4
LT		 584 # schedule() instead
585 movl %esp, %eax
d4d67150 586 call syscall_trace_leave
1da177e4 587 jmp resume_userspace
47a55cd7 588END(syscall_exit_work)
fe7cacc1 589 CFI_ENDPROC
1da177e4 590
fe7cacc1 591 RING0_INT_FRAME # can't unwind into user space anyway
1da177e4 592syscall_fault:
1da177e4 593 GET_THREAD_INFO(%ebp)
eb5b7b9d 594 movl $-EFAULT,PT_EAX(%esp)
1da177e4 595 jmp resume_userspace
47a55cd7 596END(syscall_fault)
1da177e4 597
1da177e4 598syscall_badsys:
eb5b7b9d 599 movl $-ENOSYS,PT_EAX(%esp)
1da177e4 600 jmp resume_userspace
47a55cd7 601END(syscall_badsys)
fe7cacc1 602 CFI_ENDPROC
1da177e4 603
f0d96110
TH		 604.macro FIXUP_ESPFIX_STACK
605 /* since we are on a wrong stack, we cant make it a C code :( */
606 PER_CPU(gdt_page, %ebx)
607 GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah)
608 addl %esp, %eax
609 pushl $__KERNEL_DS
610 CFI_ADJUST_CFA_OFFSET 4
611 pushl %eax
612 CFI_ADJUST_CFA_OFFSET 4
613 lss (%esp), %esp
614 CFI_ADJUST_CFA_OFFSET -8
615.endm
616.macro UNWIND_ESPFIX_STACK
617 movl %ss, %eax
618 /* see if on espfix stack */
619 cmpw $__ESPFIX_SS, %ax
620 jne 27f
621 movl $__KERNEL_DS, %eax
622 movl %eax, %ds
623 movl %eax, %es
624 /* switch to normal stack */
625 FIXUP_ESPFIX_STACK
62627:
627.endm
1da177e4
LT		 628
629/*
b7c6244f
PA		 630 * Build the entry stubs and pointer table with some assembler magic.
631 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
632 * single cache line on all modern x86 implementations.
1da177e4 633 */
4687518c 634.section .init.rodata,"a"
1da177e4
LT		 635ENTRY(interrupt)
636.text
b7c6244f
PA		 637 .p2align 5
638 .p2align CONFIG_X86_L1_CACHE_SHIFT
1da177e4 639ENTRY(irq_entries_start)
fe7cacc1 640 RING0_INT_FRAME
4687518c 641vector=FIRST_EXTERNAL_VECTOR
b7c6244f
PA		 642.rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
643 .balign 32
644 .rept 7
645 .if vector < NR_VECTORS
8665596e 646 .if vector <> FIRST_EXTERNAL_VECTOR
fe7cacc1 647 CFI_ADJUST_CFA_OFFSET -4
b7c6244f
PA		 648 .endif
6491: pushl $(~vector+0x80) /* Note: always in signed byte range */
fe7cacc1 650 CFI_ADJUST_CFA_OFFSET 4
8665596e 651 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
b7c6244f
PA		 652 jmp 2f
653 .endif
654 .previous
1da177e4 655 .long 1b
b7c6244f 656 .text
1da177e4 657vector=vector+1
b7c6244f
PA		 658 .endif
659 .endr
6602: jmp common_interrupt
1da177e4 661.endr
47a55cd7
JB		 662END(irq_entries_start)
663
664.previous
665END(interrupt)
666.previous
1da177e4 667
55f327fa
IM		 668/*
669 * the CPU automatically disables interrupts when executing an IRQ vector,
670 * so IRQ-flags tracing has to follow that:
671 */
b7c6244f 672 .p2align CONFIG_X86_L1_CACHE_SHIFT
1da177e4 673common_interrupt:
b7c6244f 674 addl $-0x80,(%esp) /* Adjust vector into the [-256,-1] range */
1da177e4 675 SAVE_ALL
55f327fa 676 TRACE_IRQS_OFF
1da177e4
LT		 677 movl %esp,%eax
678 call do_IRQ
679 jmp ret_from_intr
47a55cd7 680ENDPROC(common_interrupt)
fe7cacc1 681 CFI_ENDPROC
1da177e4 682
02cf94c3 683#define BUILD_INTERRUPT3(name, nr, fn) \
1da177e4 684ENTRY(name) \
fe7cacc1 685 RING0_INT_FRAME; \
19eadf98 686 pushl $~(nr); \
fe7cacc1
JB		 687 CFI_ADJUST_CFA_OFFSET 4; \
688 SAVE_ALL; \
55f327fa 689 TRACE_IRQS_OFF \
1da177e4 690 movl %esp,%eax; \
02cf94c3 691 call fn; \
55f327fa 692 jmp ret_from_intr; \
47a55cd7
JB		 693 CFI_ENDPROC; \
694ENDPROC(name)
1da177e4 695
02cf94c3
TH		 696#define BUILD_INTERRUPT(name, nr) BUILD_INTERRUPT3(name, nr, smp_##name)
697
1da177e4
LT		 698/* The include is where all of the SMP etc. interrupts come from */
699#include "entry_arch.h"
700
1da177e4 701ENTRY(coprocessor_error)
fe7cacc1 702 RING0_INT_FRAME
1da177e4 703 pushl $0
fe7cacc1 704 CFI_ADJUST_CFA_OFFSET 4
1da177e4 705 pushl $do_coprocessor_error
fe7cacc1 706 CFI_ADJUST_CFA_OFFSET 4
1da177e4 707 jmp error_code
fe7cacc1 708 CFI_ENDPROC
47a55cd7 709END(coprocessor_error)
1da177e4
LT		 710
711ENTRY(simd_coprocessor_error)
fe7cacc1 712 RING0_INT_FRAME
1da177e4 713 pushl $0
fe7cacc1 714 CFI_ADJUST_CFA_OFFSET 4
1da177e4 715 pushl $do_simd_coprocessor_error
fe7cacc1 716 CFI_ADJUST_CFA_OFFSET 4
1da177e4 717 jmp error_code
fe7cacc1 718 CFI_ENDPROC
47a55cd7 719END(simd_coprocessor_error)
1da177e4
LT		 720
721ENTRY(device_not_available)
fe7cacc1 722 RING0_INT_FRAME
1da177e4 723 pushl $-1 # mark this as an int
fe7cacc1 724 CFI_ADJUST_CFA_OFFSET 4
7643e9b9 725 pushl $do_device_not_available
fe7cacc1 726 CFI_ADJUST_CFA_OFFSET 4
7643e9b9 727 jmp error_code
fe7cacc1 728 CFI_ENDPROC
47a55cd7 729END(device_not_available)
1da177e4 730
d3561b7f
RR		 731#ifdef CONFIG_PARAVIRT
732ENTRY(native_iret)
3701d863 733 iret
d3561b7f
RR		 734.section __ex_table,"a"
735 .align 4
3701d863 736 .long native_iret, iret_exc
d3561b7f 737.previous
47a55cd7 738END(native_iret)
d3561b7f 739
d75cd22f 740ENTRY(native_irq_enable_sysexit)
d3561b7f
RR		 741 sti
742 sysexit
d75cd22f 743END(native_irq_enable_sysexit)
d3561b7f
RR		 744#endif
745
1da177e4 746ENTRY(overflow)
fe7cacc1 747 RING0_INT_FRAME
1da177e4 748 pushl $0
fe7cacc1 749 CFI_ADJUST_CFA_OFFSET 4
1da177e4 750 pushl $do_overflow
fe7cacc1 751 CFI_ADJUST_CFA_OFFSET 4
1da177e4 752 jmp error_code
fe7cacc1 753 CFI_ENDPROC
47a55cd7 754END(overflow)
1da177e4
LT		 755
756ENTRY(bounds)
fe7cacc1 757 RING0_INT_FRAME
1da177e4 758 pushl $0
fe7cacc1 759 CFI_ADJUST_CFA_OFFSET 4
1da177e4 760 pushl $do_bounds
fe7cacc1 761 CFI_ADJUST_CFA_OFFSET 4
1da177e4 762 jmp error_code
fe7cacc1 763 CFI_ENDPROC
47a55cd7 764END(bounds)
1da177e4
LT		 765
766ENTRY(invalid_op)
fe7cacc1 767 RING0_INT_FRAME
1da177e4 768 pushl $0
fe7cacc1 769 CFI_ADJUST_CFA_OFFSET 4
1da177e4 770 pushl $do_invalid_op
fe7cacc1 771 CFI_ADJUST_CFA_OFFSET 4
1da177e4 772 jmp error_code
fe7cacc1 773 CFI_ENDPROC
47a55cd7 774END(invalid_op)
1da177e4
LT		 775
776ENTRY(coprocessor_segment_overrun)
fe7cacc1 777 RING0_INT_FRAME
1da177e4 778 pushl $0
fe7cacc1 779 CFI_ADJUST_CFA_OFFSET 4
1da177e4 780 pushl $do_coprocessor_segment_overrun
fe7cacc1 781 CFI_ADJUST_CFA_OFFSET 4
1da177e4 782 jmp error_code
fe7cacc1 783 CFI_ENDPROC
47a55cd7 784END(coprocessor_segment_overrun)
1da177e4
LT		 785
786ENTRY(invalid_TSS)
fe7cacc1 787 RING0_EC_FRAME
1da177e4 788 pushl $do_invalid_TSS
fe7cacc1 789 CFI_ADJUST_CFA_OFFSET 4
1da177e4 790 jmp error_code
fe7cacc1 791 CFI_ENDPROC
47a55cd7 792END(invalid_TSS)
1da177e4
LT		 793
794ENTRY(segment_not_present)
fe7cacc1 795 RING0_EC_FRAME
1da177e4 796 pushl $do_segment_not_present
fe7cacc1 797 CFI_ADJUST_CFA_OFFSET 4
1da177e4 798 jmp error_code
fe7cacc1 799 CFI_ENDPROC
47a55cd7 800END(segment_not_present)
1da177e4
LT		 801
802ENTRY(stack_segment)
fe7cacc1 803 RING0_EC_FRAME
1da177e4 804 pushl $do_stack_segment
fe7cacc1 805 CFI_ADJUST_CFA_OFFSET 4
1da177e4 806 jmp error_code
fe7cacc1 807 CFI_ENDPROC
47a55cd7 808END(stack_segment)
1da177e4 809
1da177e4 810ENTRY(alignment_check)
fe7cacc1 811 RING0_EC_FRAME
1da177e4 812 pushl $do_alignment_check
fe7cacc1 813 CFI_ADJUST_CFA_OFFSET 4
1da177e4 814 jmp error_code
fe7cacc1 815 CFI_ENDPROC
47a55cd7 816END(alignment_check)
1da177e4 817
d28c4393
P		 818ENTRY(divide_error)
819 RING0_INT_FRAME
820 pushl $0 # no error code
821 CFI_ADJUST_CFA_OFFSET 4
822 pushl $do_divide_error
fe7cacc1 823 CFI_ADJUST_CFA_OFFSET 4
1da177e4 824 jmp error_code
fe7cacc1 825 CFI_ENDPROC
47a55cd7 826END(divide_error)
1da177e4
LT		 827
828#ifdef CONFIG_X86_MCE
829ENTRY(machine_check)
fe7cacc1 830 RING0_INT_FRAME
1da177e4 831 pushl $0
fe7cacc1 832 CFI_ADJUST_CFA_OFFSET 4
d2f6f7ae 833 pushl machine_check_vector
fe7cacc1 834 CFI_ADJUST_CFA_OFFSET 4
1da177e4 835 jmp error_code
fe7cacc1 836 CFI_ENDPROC
47a55cd7 837END(machine_check)
1da177e4
LT		 838#endif
839
840ENTRY(spurious_interrupt_bug)
fe7cacc1 841 RING0_INT_FRAME
1da177e4 842 pushl $0
fe7cacc1 843 CFI_ADJUST_CFA_OFFSET 4
1da177e4 844 pushl $do_spurious_interrupt_bug
fe7cacc1 845 CFI_ADJUST_CFA_OFFSET 4
1da177e4 846 jmp error_code
fe7cacc1 847 CFI_ENDPROC
47a55cd7 848END(spurious_interrupt_bug)
1da177e4 849
02ba1a32
AK		 850ENTRY(kernel_thread_helper)
851 pushl $0 # fake return address for unwinder
852 CFI_STARTPROC
853 movl %edx,%eax
854 push %edx
855 CFI_ADJUST_CFA_OFFSET 4
856 call *%ebx
857 push %eax
858 CFI_ADJUST_CFA_OFFSET 4
859 call do_exit
5f5db591 860 ud2 # padding for call trace
02ba1a32
AK		 861 CFI_ENDPROC
862ENDPROC(kernel_thread_helper)
863
5ead97c8 864#ifdef CONFIG_XEN
e2a81baf
JF		 865/* Xen doesn't set %esp to be precisely what the normal sysenter
866 entrypoint expects, so fix it up before using the normal path. */
867ENTRY(xen_sysenter_target)
868 RING0_INT_FRAME
869 addl $5*4, %esp /* remove xen-provided frame */
2ddf9b7b 870 CFI_ADJUST_CFA_OFFSET -5*4
e2a81baf 871 jmp sysenter_past_esp
557d7d4e 872 CFI_ENDPROC
e2a81baf 873
5ead97c8
JF		 874ENTRY(xen_hypervisor_callback)
875 CFI_STARTPROC
876 pushl $0
877 CFI_ADJUST_CFA_OFFSET 4
878 SAVE_ALL
879 TRACE_IRQS_OFF
9ec2b804
JF		 880
881 /* Check to see if we got the event in the critical
882 region in xen_iret_direct, after we've reenabled
883 events and checked for pending events. This simulates
884 iret instruction's behaviour where it delivers a
885 pending interrupt when enabling interrupts. */
886 movl PT_EIP(%esp),%eax
887 cmpl $xen_iret_start_crit,%eax
888 jb 1f
889 cmpl $xen_iret_end_crit,%eax
890 jae 1f
891
0f2c8769 892 jmp xen_iret_crit_fixup
e2a81baf 893
e2a81baf 894ENTRY(xen_do_upcall)
b77797fb 8951: mov %esp, %eax
5ead97c8
JF		 896 call xen_evtchn_do_upcall
897 jmp ret_from_intr
898 CFI_ENDPROC
899ENDPROC(xen_hypervisor_callback)
900
901# Hypervisor uses this for application faults while it executes.
902# We get here for two reasons:
903# 1. Fault while reloading DS, ES, FS or GS
904# 2. Fault while executing IRET
905# Category 1 we fix up by reattempting the load, and zeroing the segment
906# register if the load fails.
907# Category 2 we fix up by jumping to do_iret_error. We cannot use the
908# normal Linux return path in this case because if we use the IRET hypercall
909# to pop the stack frame we end up in an infinite loop of failsafe callbacks.
910# We distinguish between categories by maintaining a status value in EAX.
911ENTRY(xen_failsafe_callback)
912 CFI_STARTPROC
913 pushl %eax
914 CFI_ADJUST_CFA_OFFSET 4
915 movl $1,%eax
9161: mov 4(%esp),%ds
9172: mov 8(%esp),%es
9183: mov 12(%esp),%fs
9194: mov 16(%esp),%gs
920 testl %eax,%eax
921 popl %eax
922 CFI_ADJUST_CFA_OFFSET -4
923 lea 16(%esp),%esp
924 CFI_ADJUST_CFA_OFFSET -16
925 jz 5f
926 addl $16,%esp
927 jmp iret_exc # EAX != 0 => Category 2 (Bad IRET)
9285: pushl $0 # EAX == 0 => Category 1 (Bad segment)
929 CFI_ADJUST_CFA_OFFSET 4
930 SAVE_ALL
931 jmp ret_from_exception
932 CFI_ENDPROC
933
934.section .fixup,"ax"
9356: xorl %eax,%eax
936 movl %eax,4(%esp)
937 jmp 1b
9387: xorl %eax,%eax
939 movl %eax,8(%esp)
940 jmp 2b
9418: xorl %eax,%eax
942 movl %eax,12(%esp)
943 jmp 3b
9449: xorl %eax,%eax
945 movl %eax,16(%esp)
946 jmp 4b
947.previous
948.section __ex_table,"a"
949 .align 4
950 .long 1b,6b
951 .long 2b,7b
952 .long 3b,8b
953 .long 4b,9b
954.previous
955ENDPROC(xen_failsafe_callback)
956
957#endif /* CONFIG_XEN */
958
606576ce 959#ifdef CONFIG_FUNCTION_TRACER
d61f82d0
SR		 960#ifdef CONFIG_DYNAMIC_FTRACE
961
962ENTRY(mcount)
d61f82d0
SR		 963 ret
964END(mcount)
965
966ENTRY(ftrace_caller)
60a7ecf4
SR		 967 cmpl $0, function_trace_stop
968 jne ftrace_stub
969
d61f82d0
SR		 970 pushl %eax
971 pushl %ecx
972 pushl %edx
973 movl 0xc(%esp), %eax
974 movl 0x4(%ebp), %edx
395a59d0 975 subl $MCOUNT_INSN_SIZE, %eax
d61f82d0
SR		 976
977.globl ftrace_call
978ftrace_call:
979 call ftrace_stub
980
981 popl %edx
982 popl %ecx
983 popl %eax
5a45cfe1
SR		 984#ifdef CONFIG_FUNCTION_GRAPH_TRACER
985.globl ftrace_graph_call
986ftrace_graph_call:
987 jmp ftrace_stub
988#endif
d61f82d0
SR		 989
990.globl ftrace_stub
991ftrace_stub:
992 ret
993END(ftrace_caller)
994
995#else /* ! CONFIG_DYNAMIC_FTRACE */
996
16444a8a 997ENTRY(mcount)
60a7ecf4
SR		 998 cmpl $0, function_trace_stop
999 jne ftrace_stub
1000
16444a8a
ACM		 1001 cmpl $ftrace_stub, ftrace_trace_function
1002 jnz trace
fb52607a 1003#ifdef CONFIG_FUNCTION_GRAPH_TRACER
c2324b69 1004 cmpl $ftrace_stub, ftrace_graph_return
fb52607a 1005 jnz ftrace_graph_caller
e49dc19c
SR		 1006
1007 cmpl $ftrace_graph_entry_stub, ftrace_graph_entry
1008 jnz ftrace_graph_caller
caf4b323 1009#endif
16444a8a
ACM		 1010.globl ftrace_stub
1011ftrace_stub:
1012 ret
1013
1014 /* taken from glibc */
1015trace:
1016 pushl %eax
1017 pushl %ecx
1018 pushl %edx
1019 movl 0xc(%esp), %eax
1020 movl 0x4(%ebp), %edx
395a59d0 1021 subl $MCOUNT_INSN_SIZE, %eax
16444a8a 1022
d61f82d0 1023 call *ftrace_trace_function
16444a8a
ACM		 1024
1025 popl %edx
1026 popl %ecx
1027 popl %eax
16444a8a
ACM		 1028 jmp ftrace_stub
1029END(mcount)
d61f82d0 1030#endif /* CONFIG_DYNAMIC_FTRACE */
606576ce 1031#endif /* CONFIG_FUNCTION_TRACER */
16444a8a 1032
fb52607a
FW		 1033#ifdef CONFIG_FUNCTION_GRAPH_TRACER
1034ENTRY(ftrace_graph_caller)
e7d3737e
FW		 1035 cmpl $0, function_trace_stop
1036 jne ftrace_stub
1037
caf4b323
FW		 1038 pushl %eax
1039 pushl %ecx
1040 pushl %edx
1dc1c6ad 1041 movl 0xc(%esp), %edx
caf4b323 1042 lea 0x4(%ebp), %eax
bb4304c7 1043 subl $MCOUNT_INSN_SIZE, %edx
caf4b323 1044 call prepare_ftrace_return
caf4b323
FW		 1045 popl %edx
1046 popl %ecx
1047 popl %eax
e7d3737e 1048 ret
fb52607a 1049END(ftrace_graph_caller)
caf4b323
FW		 1050
1051.globl return_to_handler
1052return_to_handler:
1053 pushl $0
1054 pushl %eax
1055 pushl %ecx
1056 pushl %edx
1057 call ftrace_return_to_handler
1058 movl %eax, 0xc(%esp)
1059 popl %edx
1060 popl %ecx
1061 popl %eax
1062 ret
e7d3737e 1063#endif
16444a8a 1064
bb152f53 1065.section .rodata,"a"
541054d9 1066#include "syscall_table_32.S"
1da177e4
LT		 1067
1068syscall_table_size=(.-sys_call_table)
d211af05
AH		 1069
1070/*
1071 * Some functions should be protected against kprobes
1072 */
1073 .pushsection .kprobes.text, "ax"
1074
1075ENTRY(page_fault)
1076 RING0_EC_FRAME
1077 pushl $do_page_fault
1078 CFI_ADJUST_CFA_OFFSET 4
1079 ALIGN
1080error_code:
1081 /* the function address is in %fs's slot on the stack */
1082 pushl %es
1083 CFI_ADJUST_CFA_OFFSET 4
1084 /*CFI_REL_OFFSET es, 0*/
1085 pushl %ds
1086 CFI_ADJUST_CFA_OFFSET 4
1087 /*CFI_REL_OFFSET ds, 0*/
1088 pushl %eax
1089 CFI_ADJUST_CFA_OFFSET 4
1090 CFI_REL_OFFSET eax, 0
1091 pushl %ebp
1092 CFI_ADJUST_CFA_OFFSET 4
1093 CFI_REL_OFFSET ebp, 0
1094 pushl %edi
1095 CFI_ADJUST_CFA_OFFSET 4
1096 CFI_REL_OFFSET edi, 0
1097 pushl %esi
1098 CFI_ADJUST_CFA_OFFSET 4
1099 CFI_REL_OFFSET esi, 0
1100 pushl %edx
1101 CFI_ADJUST_CFA_OFFSET 4
1102 CFI_REL_OFFSET edx, 0
1103 pushl %ecx
1104 CFI_ADJUST_CFA_OFFSET 4
1105 CFI_REL_OFFSET ecx, 0
1106 pushl %ebx
1107 CFI_ADJUST_CFA_OFFSET 4
1108 CFI_REL_OFFSET ebx, 0
1109 cld
1110 pushl %fs
1111 CFI_ADJUST_CFA_OFFSET 4
1112 /*CFI_REL_OFFSET fs, 0*/
1113 movl $(__KERNEL_PERCPU), %ecx
1114 movl %ecx, %fs
1115 UNWIND_ESPFIX_STACK
1116 popl %ecx
1117 CFI_ADJUST_CFA_OFFSET -4
1118 /*CFI_REGISTER es, ecx*/
1119 movl PT_FS(%esp), %edi # get the function address
1120 movl PT_ORIG_EAX(%esp), %edx # get the error code
1121 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1122 mov %ecx, PT_FS(%esp)
1123 /*CFI_REL_OFFSET fs, ES*/
1124 movl $(__USER_DS), %ecx
1125 movl %ecx, %ds
1126 movl %ecx, %es
1127 TRACE_IRQS_OFF
1128 movl %esp,%eax # pt_regs pointer
1129 call *%edi
1130 jmp ret_from_exception
1131 CFI_ENDPROC
1132END(page_fault)
1133
1134/*
1135 * Debug traps and NMI can happen at the one SYSENTER instruction
1136 * that sets up the real kernel stack. Check here, since we can't
1137 * allow the wrong stack to be used.
1138 *
1139 * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have
1140 * already pushed 3 words if it hits on the sysenter instruction:
1141 * eflags, cs and eip.
1142 *
1143 * We just load the right stack, and push the three (known) values
1144 * by hand onto the new stack - while updating the return eip past
1145 * the instruction that would have done it for sysenter.
1146 */
f0d96110
TH		 1147.macro FIX_STACK offset ok label
1148 cmpw $__KERNEL_CS, 4(%esp)
1149 jne \ok
1150\label:
1151 movl TSS_sysenter_sp0 + \offset(%esp), %esp
1152 CFI_DEF_CFA esp, 0
1153 CFI_UNDEFINED eip
1154 pushfl
1155 CFI_ADJUST_CFA_OFFSET 4
1156 pushl $__KERNEL_CS
1157 CFI_ADJUST_CFA_OFFSET 4
1158 pushl $sysenter_past_esp
1159 CFI_ADJUST_CFA_OFFSET 4
d211af05 1160 CFI_REL_OFFSET eip, 0
f0d96110 1161.endm
d211af05
AH		 1162
1163ENTRY(debug)
1164 RING0_INT_FRAME
1165 cmpl $ia32_sysenter_target,(%esp)
1166 jne debug_stack_correct
f0d96110 1167 FIX_STACK 12, debug_stack_correct, debug_esp_fix_insn
d211af05
AH		 1168debug_stack_correct:
1169 pushl $-1 # mark this as an int
1170 CFI_ADJUST_CFA_OFFSET 4
1171 SAVE_ALL
1172 TRACE_IRQS_OFF
1173 xorl %edx,%edx # error code 0
1174 movl %esp,%eax # pt_regs pointer
1175 call do_debug
1176 jmp ret_from_exception
1177 CFI_ENDPROC
1178END(debug)
1179
1180/*
1181 * NMI is doubly nasty. It can happen _while_ we're handling
1182 * a debug fault, and the debug fault hasn't yet been able to
1183 * clear up the stack. So we first check whether we got an
1184 * NMI on the sysenter entry path, but after that we need to
1185 * check whether we got an NMI on the debug path where the debug
1186 * fault happened on the sysenter path.
1187 */
1188ENTRY(nmi)
1189 RING0_INT_FRAME
1190 pushl %eax
1191 CFI_ADJUST_CFA_OFFSET 4
1192 movl %ss, %eax
1193 cmpw $__ESPFIX_SS, %ax
1194 popl %eax
1195 CFI_ADJUST_CFA_OFFSET -4
1196 je nmi_espfix_stack
1197 cmpl $ia32_sysenter_target,(%esp)
1198 je nmi_stack_fixup
1199 pushl %eax
1200 CFI_ADJUST_CFA_OFFSET 4
1201 movl %esp,%eax
1202 /* Do not access memory above the end of our stack page,
1203 * it might not exist.
1204 */
1205 andl $(THREAD_SIZE-1),%eax
1206 cmpl $(THREAD_SIZE-20),%eax
1207 popl %eax
1208 CFI_ADJUST_CFA_OFFSET -4
1209 jae nmi_stack_correct
1210 cmpl $ia32_sysenter_target,12(%esp)
1211 je nmi_debug_stack_check
1212nmi_stack_correct:
1213 /* We have a RING0_INT_FRAME here */
1214 pushl %eax
1215 CFI_ADJUST_CFA_OFFSET 4
1216 SAVE_ALL
d211af05
AH		 1217 xorl %edx,%edx # zero error code
1218 movl %esp,%eax # pt_regs pointer
1219 call do_nmi
1220 jmp restore_nocheck_notrace
1221 CFI_ENDPROC
1222
1223nmi_stack_fixup:
1224 RING0_INT_FRAME
f0d96110 1225 FIX_STACK 12, nmi_stack_correct, 1
d211af05
AH		 1226 jmp nmi_stack_correct
1227
1228nmi_debug_stack_check:
1229 /* We have a RING0_INT_FRAME here */
1230 cmpw $__KERNEL_CS,16(%esp)
1231 jne nmi_stack_correct
1232 cmpl $debug,(%esp)
1233 jb nmi_stack_correct
1234 cmpl $debug_esp_fix_insn,(%esp)
1235 ja nmi_stack_correct
f0d96110 1236 FIX_STACK 24, nmi_stack_correct, 1
d211af05
AH		 1237 jmp nmi_stack_correct
1238
1239nmi_espfix_stack:
1240 /* We have a RING0_INT_FRAME here.
1241 *
1242 * create the pointer to lss back
1243 */
1244 pushl %ss
1245 CFI_ADJUST_CFA_OFFSET 4
1246 pushl %esp
1247 CFI_ADJUST_CFA_OFFSET 4
1248 addw $4, (%esp)
1249 /* copy the iret frame of 12 bytes */
1250 .rept 3
1251 pushl 16(%esp)
1252 CFI_ADJUST_CFA_OFFSET 4
1253 .endr
1254 pushl %eax
1255 CFI_ADJUST_CFA_OFFSET 4
1256 SAVE_ALL
d211af05
AH		 1257 FIXUP_ESPFIX_STACK # %eax == %esp
1258 xorl %edx,%edx # zero error code
1259 call do_nmi
1260 RESTORE_REGS
1261 lss 12+4(%esp), %esp # back to espfix stack
1262 CFI_ADJUST_CFA_OFFSET -24
1263 jmp irq_return
1264 CFI_ENDPROC
1265END(nmi)
1266
1267ENTRY(int3)
1268 RING0_INT_FRAME
1269 pushl $-1 # mark this as an int
1270 CFI_ADJUST_CFA_OFFSET 4
1271 SAVE_ALL
1272 TRACE_IRQS_OFF
1273 xorl %edx,%edx # zero error code
1274 movl %esp,%eax # pt_regs pointer
1275 call do_int3
1276 jmp ret_from_exception
1277 CFI_ENDPROC
1278END(int3)
1279
1280ENTRY(general_protection)
1281 RING0_EC_FRAME
1282 pushl $do_general_protection
1283 CFI_ADJUST_CFA_OFFSET 4
1284 jmp error_code
1285 CFI_ENDPROC
1286END(general_protection)
1287
1288/*
1289 * End of kprobes section
1290 */
1291 .popsection
Linux 6.x block layer and io_uring tree(s)