Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | |
f95d47ca JF |
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 | |
1da177e4 LT |
40 | * |
41 | * "current" is in register %ebx during any slow entries. | |
42 | */ | |
43 | ||
1da177e4 LT |
44 | #include <linux/linkage.h> |
45 | #include <asm/thread_info.h> | |
55f327fa | 46 | #include <asm/irqflags.h> |
1da177e4 LT |
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> | |
be44d2aa | 52 | #include <asm/percpu.h> |
fe7cacc1 | 53 | #include <asm/dwarf2.h> |
1da177e4 LT |
54 | #include "irq_vectors.h" |
55 | ||
56 | #define nr_syscalls ((syscall_table_size)/4) | |
57 | ||
1da177e4 LT |
58 | CF_MASK = 0x00000001 |
59 | TF_MASK = 0x00000100 | |
60 | IF_MASK = 0x00000200 | |
61 | DF_MASK = 0x00000400 | |
62 | NT_MASK = 0x00004000 | |
63 | VM_MASK = 0x00020000 | |
64 | ||
0da5db31 RR |
65 | /* These are replaces for paravirtualization */ |
66 | #define DISABLE_INTERRUPTS cli | |
67 | #define ENABLE_INTERRUPTS sti | |
68 | #define ENABLE_INTERRUPTS_SYSEXIT sti; sysexit | |
69 | #define INTERRUPT_RETURN iret | |
70 | #define GET_CR0_INTO_EAX movl %cr0, %eax | |
71 | ||
1da177e4 | 72 | #ifdef CONFIG_PREEMPT |
0da5db31 | 73 | #define preempt_stop DISABLE_INTERRUPTS; TRACE_IRQS_OFF |
1da177e4 LT |
74 | #else |
75 | #define preempt_stop | |
76 | #define resume_kernel restore_nocheck | |
77 | #endif | |
78 | ||
55f327fa IM |
79 | .macro TRACE_IRQS_IRET |
80 | #ifdef CONFIG_TRACE_IRQFLAGS | |
eb5b7b9d | 81 | testl $IF_MASK,PT_EFLAGS(%esp) # interrupts off? |
55f327fa IM |
82 | jz 1f |
83 | TRACE_IRQS_ON | |
84 | 1: | |
85 | #endif | |
86 | .endm | |
87 | ||
4031ff38 AG |
88 | #ifdef CONFIG_VM86 |
89 | #define resume_userspace_sig check_userspace | |
90 | #else | |
91 | #define resume_userspace_sig resume_userspace | |
92 | #endif | |
93 | ||
1da177e4 LT |
94 | #define SAVE_ALL \ |
95 | cld; \ | |
f95d47ca JF |
96 | pushl %gs; \ |
97 | CFI_ADJUST_CFA_OFFSET 4;\ | |
98 | /*CFI_REL_OFFSET gs, 0;*/\ | |
1da177e4 | 99 | pushl %es; \ |
fe7cacc1 JB |
100 | CFI_ADJUST_CFA_OFFSET 4;\ |
101 | /*CFI_REL_OFFSET es, 0;*/\ | |
1da177e4 | 102 | pushl %ds; \ |
fe7cacc1 JB |
103 | CFI_ADJUST_CFA_OFFSET 4;\ |
104 | /*CFI_REL_OFFSET ds, 0;*/\ | |
1da177e4 | 105 | pushl %eax; \ |
fe7cacc1 JB |
106 | CFI_ADJUST_CFA_OFFSET 4;\ |
107 | CFI_REL_OFFSET eax, 0;\ | |
1da177e4 | 108 | pushl %ebp; \ |
fe7cacc1 JB |
109 | CFI_ADJUST_CFA_OFFSET 4;\ |
110 | CFI_REL_OFFSET ebp, 0;\ | |
1da177e4 | 111 | pushl %edi; \ |
fe7cacc1 JB |
112 | CFI_ADJUST_CFA_OFFSET 4;\ |
113 | CFI_REL_OFFSET edi, 0;\ | |
1da177e4 | 114 | pushl %esi; \ |
fe7cacc1 JB |
115 | CFI_ADJUST_CFA_OFFSET 4;\ |
116 | CFI_REL_OFFSET esi, 0;\ | |
1da177e4 | 117 | pushl %edx; \ |
fe7cacc1 JB |
118 | CFI_ADJUST_CFA_OFFSET 4;\ |
119 | CFI_REL_OFFSET edx, 0;\ | |
1da177e4 | 120 | pushl %ecx; \ |
fe7cacc1 JB |
121 | CFI_ADJUST_CFA_OFFSET 4;\ |
122 | CFI_REL_OFFSET ecx, 0;\ | |
1da177e4 | 123 | pushl %ebx; \ |
fe7cacc1 JB |
124 | CFI_ADJUST_CFA_OFFSET 4;\ |
125 | CFI_REL_OFFSET ebx, 0;\ | |
1da177e4 LT |
126 | movl $(__USER_DS), %edx; \ |
127 | movl %edx, %ds; \ | |
f95d47ca JF |
128 | movl %edx, %es; \ |
129 | movl $(__KERNEL_PDA), %edx; \ | |
130 | movl %edx, %gs | |
1da177e4 LT |
131 | |
132 | #define RESTORE_INT_REGS \ | |
133 | popl %ebx; \ | |
fe7cacc1 JB |
134 | CFI_ADJUST_CFA_OFFSET -4;\ |
135 | CFI_RESTORE ebx;\ | |
1da177e4 | 136 | popl %ecx; \ |
fe7cacc1 JB |
137 | CFI_ADJUST_CFA_OFFSET -4;\ |
138 | CFI_RESTORE ecx;\ | |
1da177e4 | 139 | popl %edx; \ |
fe7cacc1 JB |
140 | CFI_ADJUST_CFA_OFFSET -4;\ |
141 | CFI_RESTORE edx;\ | |
1da177e4 | 142 | popl %esi; \ |
fe7cacc1 JB |
143 | CFI_ADJUST_CFA_OFFSET -4;\ |
144 | CFI_RESTORE esi;\ | |
1da177e4 | 145 | popl %edi; \ |
fe7cacc1 JB |
146 | CFI_ADJUST_CFA_OFFSET -4;\ |
147 | CFI_RESTORE edi;\ | |
1da177e4 | 148 | popl %ebp; \ |
fe7cacc1 JB |
149 | CFI_ADJUST_CFA_OFFSET -4;\ |
150 | CFI_RESTORE ebp;\ | |
151 | popl %eax; \ | |
152 | CFI_ADJUST_CFA_OFFSET -4;\ | |
153 | CFI_RESTORE eax | |
1da177e4 LT |
154 | |
155 | #define RESTORE_REGS \ | |
156 | RESTORE_INT_REGS; \ | |
157 | 1: popl %ds; \ | |
fe7cacc1 JB |
158 | CFI_ADJUST_CFA_OFFSET -4;\ |
159 | /*CFI_RESTORE ds;*/\ | |
1da177e4 | 160 | 2: popl %es; \ |
fe7cacc1 JB |
161 | CFI_ADJUST_CFA_OFFSET -4;\ |
162 | /*CFI_RESTORE es;*/\ | |
f95d47ca JF |
163 | 3: popl %gs; \ |
164 | CFI_ADJUST_CFA_OFFSET -4;\ | |
165 | /*CFI_RESTORE gs;*/\ | |
166 | .pushsection .fixup,"ax"; \ | |
1da177e4 | 167 | 4: movl $0,(%esp); \ |
f95d47ca JF |
168 | jmp 1b; \ |
169 | 5: movl $0,(%esp); \ | |
1da177e4 | 170 | jmp 2b; \ |
f95d47ca JF |
171 | 6: movl $0,(%esp); \ |
172 | jmp 3b; \ | |
1da177e4 LT |
173 | .section __ex_table,"a";\ |
174 | .align 4; \ | |
f95d47ca JF |
175 | .long 1b,4b; \ |
176 | .long 2b,5b; \ | |
177 | .long 3b,6b; \ | |
178 | .popsection | |
1da177e4 | 179 | |
fe7cacc1 JB |
180 | #define RING0_INT_FRAME \ |
181 | CFI_STARTPROC simple;\ | |
adf14236 | 182 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
183 | CFI_DEF_CFA esp, 3*4;\ |
184 | /*CFI_OFFSET cs, -2*4;*/\ | |
185 | CFI_OFFSET eip, -3*4 | |
186 | ||
187 | #define RING0_EC_FRAME \ | |
188 | CFI_STARTPROC simple;\ | |
adf14236 | 189 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
190 | CFI_DEF_CFA esp, 4*4;\ |
191 | /*CFI_OFFSET cs, -2*4;*/\ | |
192 | CFI_OFFSET eip, -3*4 | |
193 | ||
194 | #define RING0_PTREGS_FRAME \ | |
195 | CFI_STARTPROC simple;\ | |
adf14236 | 196 | CFI_SIGNAL_FRAME;\ |
eb5b7b9d JF |
197 | CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\ |
198 | /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\ | |
199 | CFI_OFFSET eip, PT_EIP-PT_OLDESP;\ | |
200 | /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\ | |
201 | /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\ | |
202 | CFI_OFFSET eax, PT_EAX-PT_OLDESP;\ | |
203 | CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\ | |
204 | CFI_OFFSET edi, PT_EDI-PT_OLDESP;\ | |
205 | CFI_OFFSET esi, PT_ESI-PT_OLDESP;\ | |
206 | CFI_OFFSET edx, PT_EDX-PT_OLDESP;\ | |
207 | CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\ | |
208 | CFI_OFFSET ebx, PT_EBX-PT_OLDESP | |
1da177e4 LT |
209 | |
210 | ENTRY(ret_from_fork) | |
fe7cacc1 | 211 | CFI_STARTPROC |
1da177e4 | 212 | pushl %eax |
25d7dfda | 213 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
214 | call schedule_tail |
215 | GET_THREAD_INFO(%ebp) | |
216 | popl %eax | |
fe7cacc1 | 217 | CFI_ADJUST_CFA_OFFSET -4 |
47a5c6fa LT |
218 | pushl $0x0202 # Reset kernel eflags |
219 | CFI_ADJUST_CFA_OFFSET 4 | |
220 | popfl | |
221 | CFI_ADJUST_CFA_OFFSET -4 | |
1da177e4 | 222 | jmp syscall_exit |
fe7cacc1 | 223 | CFI_ENDPROC |
1da177e4 LT |
224 | |
225 | /* | |
226 | * Return to user mode is not as complex as all this looks, | |
227 | * but we want the default path for a system call return to | |
228 | * go as quickly as possible which is why some of this is | |
229 | * less clear than it otherwise should be. | |
230 | */ | |
231 | ||
232 | # userspace resumption stub bypassing syscall exit tracing | |
233 | ALIGN | |
fe7cacc1 | 234 | RING0_PTREGS_FRAME |
1da177e4 LT |
235 | ret_from_exception: |
236 | preempt_stop | |
237 | ret_from_intr: | |
238 | GET_THREAD_INFO(%ebp) | |
4031ff38 | 239 | check_userspace: |
eb5b7b9d JF |
240 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS |
241 | movb PT_CS(%esp), %al | |
78be3706 RR |
242 | andl $(VM_MASK | SEGMENT_RPL_MASK), %eax |
243 | cmpl $USER_RPL, %eax | |
244 | jb resume_kernel # not returning to v8086 or userspace | |
f95d47ca | 245 | |
1da177e4 | 246 | ENTRY(resume_userspace) |
0da5db31 | 247 | DISABLE_INTERRUPTS # make sure we don't miss an interrupt |
1da177e4 LT |
248 | # setting need_resched or sigpending |
249 | # between sampling and the iret | |
250 | movl TI_flags(%ebp), %ecx | |
251 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done on | |
252 | # int/exception return? | |
253 | jne work_pending | |
254 | jmp restore_all | |
255 | ||
256 | #ifdef CONFIG_PREEMPT | |
257 | ENTRY(resume_kernel) | |
0da5db31 | 258 | DISABLE_INTERRUPTS |
1da177e4 LT |
259 | cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ? |
260 | jnz restore_nocheck | |
261 | need_resched: | |
262 | movl TI_flags(%ebp), %ecx # need_resched set ? | |
263 | testb $_TIF_NEED_RESCHED, %cl | |
264 | jz restore_all | |
eb5b7b9d | 265 | testl $IF_MASK,PT_EFLAGS(%esp) # interrupts off (exception path) ? |
1da177e4 LT |
266 | jz restore_all |
267 | call preempt_schedule_irq | |
268 | jmp need_resched | |
269 | #endif | |
fe7cacc1 | 270 | CFI_ENDPROC |
1da177e4 LT |
271 | |
272 | /* SYSENTER_RETURN points to after the "sysenter" instruction in | |
273 | the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */ | |
274 | ||
275 | # sysenter call handler stub | |
276 | ENTRY(sysenter_entry) | |
fe7cacc1 | 277 | CFI_STARTPROC simple |
adf14236 | 278 | CFI_SIGNAL_FRAME |
fe7cacc1 JB |
279 | CFI_DEF_CFA esp, 0 |
280 | CFI_REGISTER esp, ebp | |
1da177e4 LT |
281 | movl TSS_sysenter_esp0(%esp),%esp |
282 | sysenter_past_esp: | |
55f327fa IM |
283 | /* |
284 | * No need to follow this irqs on/off section: the syscall | |
285 | * disabled irqs and here we enable it straight after entry: | |
286 | */ | |
0da5db31 | 287 | ENABLE_INTERRUPTS |
1da177e4 | 288 | pushl $(__USER_DS) |
fe7cacc1 JB |
289 | CFI_ADJUST_CFA_OFFSET 4 |
290 | /*CFI_REL_OFFSET ss, 0*/ | |
1da177e4 | 291 | pushl %ebp |
fe7cacc1 JB |
292 | CFI_ADJUST_CFA_OFFSET 4 |
293 | CFI_REL_OFFSET esp, 0 | |
1da177e4 | 294 | pushfl |
fe7cacc1 | 295 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 296 | pushl $(__USER_CS) |
fe7cacc1 JB |
297 | CFI_ADJUST_CFA_OFFSET 4 |
298 | /*CFI_REL_OFFSET cs, 0*/ | |
e6e5494c IM |
299 | /* |
300 | * Push current_thread_info()->sysenter_return to the stack. | |
301 | * A tiny bit of offset fixup is necessary - 4*4 means the 4 words | |
302 | * pushed above; +8 corresponds to copy_thread's esp0 setting. | |
303 | */ | |
304 | pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp) | |
fe7cacc1 JB |
305 | CFI_ADJUST_CFA_OFFSET 4 |
306 | CFI_REL_OFFSET eip, 0 | |
1da177e4 LT |
307 | |
308 | /* | |
309 | * Load the potential sixth argument from user stack. | |
310 | * Careful about security. | |
311 | */ | |
312 | cmpl $__PAGE_OFFSET-3,%ebp | |
313 | jae syscall_fault | |
314 | 1: movl (%ebp),%ebp | |
315 | .section __ex_table,"a" | |
316 | .align 4 | |
317 | .long 1b,syscall_fault | |
318 | .previous | |
319 | ||
320 | pushl %eax | |
fe7cacc1 | 321 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
322 | SAVE_ALL |
323 | GET_THREAD_INFO(%ebp) | |
324 | ||
325 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ | |
ed75e8d5 | 326 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
327 | jnz syscall_trace_entry |
328 | cmpl $(nr_syscalls), %eax | |
329 | jae syscall_badsys | |
330 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 331 | movl %eax,PT_EAX(%esp) |
0da5db31 | 332 | DISABLE_INTERRUPTS |
55f327fa | 333 | TRACE_IRQS_OFF |
1da177e4 LT |
334 | movl TI_flags(%ebp), %ecx |
335 | testw $_TIF_ALLWORK_MASK, %cx | |
336 | jne syscall_exit_work | |
337 | /* if something modifies registers it must also disable sysexit */ | |
eb5b7b9d JF |
338 | movl PT_EIP(%esp), %edx |
339 | movl PT_OLDESP(%esp), %ecx | |
1da177e4 | 340 | xorl %ebp,%ebp |
55f327fa | 341 | TRACE_IRQS_ON |
f95d47ca | 342 | 1: mov PT_GS(%esp), %gs |
0da5db31 | 343 | ENABLE_INTERRUPTS_SYSEXIT |
fe7cacc1 | 344 | CFI_ENDPROC |
f95d47ca JF |
345 | .pushsection .fixup,"ax" |
346 | 2: movl $0,PT_GS(%esp) | |
347 | jmp 1b | |
348 | .section __ex_table,"a" | |
349 | .align 4 | |
350 | .long 1b,2b | |
351 | .popsection | |
1da177e4 LT |
352 | |
353 | # system call handler stub | |
354 | ENTRY(system_call) | |
fe7cacc1 | 355 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 356 | pushl %eax # save orig_eax |
fe7cacc1 | 357 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
358 | SAVE_ALL |
359 | GET_THREAD_INFO(%ebp) | |
eb5b7b9d | 360 | testl $TF_MASK,PT_EFLAGS(%esp) |
635cf99a CE |
361 | jz no_singlestep |
362 | orl $_TIF_SINGLESTEP,TI_flags(%ebp) | |
363 | no_singlestep: | |
ed75e8d5 | 364 | # system call tracing in operation / emulation |
1da177e4 | 365 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ |
ed75e8d5 | 366 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
367 | jnz syscall_trace_entry |
368 | cmpl $(nr_syscalls), %eax | |
369 | jae syscall_badsys | |
370 | syscall_call: | |
371 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 372 | movl %eax,PT_EAX(%esp) # store the return value |
1da177e4 | 373 | syscall_exit: |
0da5db31 | 374 | DISABLE_INTERRUPTS # make sure we don't miss an interrupt |
1da177e4 LT |
375 | # setting need_resched or sigpending |
376 | # between sampling and the iret | |
55f327fa | 377 | TRACE_IRQS_OFF |
1da177e4 LT |
378 | movl TI_flags(%ebp), %ecx |
379 | testw $_TIF_ALLWORK_MASK, %cx # current->work | |
380 | jne syscall_exit_work | |
381 | ||
382 | restore_all: | |
eb5b7b9d JF |
383 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS |
384 | # Warning: PT_OLDSS(%esp) contains the wrong/random values if we | |
5df24082 SS |
385 | # are returning to the kernel. |
386 | # See comments in process.c:copy_thread() for details. | |
eb5b7b9d JF |
387 | movb PT_OLDSS(%esp), %ah |
388 | movb PT_CS(%esp), %al | |
78be3706 RR |
389 | andl $(VM_MASK | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax |
390 | cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax | |
fe7cacc1 | 391 | CFI_REMEMBER_STATE |
1da177e4 LT |
392 | je ldt_ss # returning to user-space with LDT SS |
393 | restore_nocheck: | |
55f327fa IM |
394 | TRACE_IRQS_IRET |
395 | restore_nocheck_notrace: | |
1da177e4 | 396 | RESTORE_REGS |
f95d47ca | 397 | addl $4, %esp # skip orig_eax/error_code |
fe7cacc1 | 398 | CFI_ADJUST_CFA_OFFSET -4 |
0da5db31 | 399 | 1: INTERRUPT_RETURN |
1da177e4 LT |
400 | .section .fixup,"ax" |
401 | iret_exc: | |
55f327fa | 402 | TRACE_IRQS_ON |
0da5db31 | 403 | ENABLE_INTERRUPTS |
a879cbbb LT |
404 | pushl $0 # no error code |
405 | pushl $do_iret_error | |
406 | jmp error_code | |
1da177e4 LT |
407 | .previous |
408 | .section __ex_table,"a" | |
409 | .align 4 | |
410 | .long 1b,iret_exc | |
411 | .previous | |
412 | ||
fe7cacc1 | 413 | CFI_RESTORE_STATE |
1da177e4 | 414 | ldt_ss: |
eb5b7b9d | 415 | larl PT_OLDSS(%esp), %eax |
1da177e4 LT |
416 | jnz restore_nocheck |
417 | testl $0x00400000, %eax # returning to 32bit stack? | |
418 | jnz restore_nocheck # allright, normal return | |
419 | /* If returning to userspace with 16bit stack, | |
420 | * try to fix the higher word of ESP, as the CPU | |
421 | * won't restore it. | |
422 | * This is an "official" bug of all the x86-compatible | |
423 | * CPUs, which we can try to work around to make | |
424 | * dosemu and wine happy. */ | |
eb5b7b9d | 425 | movl PT_OLDESP(%esp), %eax |
be44d2aa SS |
426 | movl %esp, %edx |
427 | call patch_espfix_desc | |
428 | pushl $__ESPFIX_SS | |
429 | CFI_ADJUST_CFA_OFFSET 4 | |
430 | pushl %eax | |
431 | CFI_ADJUST_CFA_OFFSET 4 | |
0da5db31 | 432 | DISABLE_INTERRUPTS |
55f327fa | 433 | TRACE_IRQS_OFF |
be44d2aa SS |
434 | lss (%esp), %esp |
435 | CFI_ADJUST_CFA_OFFSET -8 | |
436 | jmp restore_nocheck | |
fe7cacc1 | 437 | CFI_ENDPROC |
1da177e4 LT |
438 | |
439 | # perform work that needs to be done immediately before resumption | |
440 | ALIGN | |
fe7cacc1 | 441 | RING0_PTREGS_FRAME # can't unwind into user space anyway |
1da177e4 LT |
442 | work_pending: |
443 | testb $_TIF_NEED_RESCHED, %cl | |
444 | jz work_notifysig | |
445 | work_resched: | |
446 | call schedule | |
0da5db31 | 447 | DISABLE_INTERRUPTS # make sure we don't miss an interrupt |
1da177e4 LT |
448 | # setting need_resched or sigpending |
449 | # between sampling and the iret | |
55f327fa | 450 | TRACE_IRQS_OFF |
1da177e4 LT |
451 | movl TI_flags(%ebp), %ecx |
452 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done other | |
453 | # than syscall tracing? | |
454 | jz restore_all | |
455 | testb $_TIF_NEED_RESCHED, %cl | |
456 | jnz work_resched | |
457 | ||
458 | work_notifysig: # deal with pending signals and | |
459 | # notify-resume requests | |
eb5b7b9d | 460 | testl $VM_MASK, PT_EFLAGS(%esp) |
1da177e4 LT |
461 | movl %esp, %eax |
462 | jne work_notifysig_v86 # returning to kernel-space or | |
463 | # vm86-space | |
464 | xorl %edx, %edx | |
465 | call do_notify_resume | |
4031ff38 | 466 | jmp resume_userspace_sig |
1da177e4 LT |
467 | |
468 | ALIGN | |
469 | work_notifysig_v86: | |
64ca9004 | 470 | #ifdef CONFIG_VM86 |
1da177e4 | 471 | pushl %ecx # save ti_flags for do_notify_resume |
fe7cacc1 | 472 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
473 | call save_v86_state # %eax contains pt_regs pointer |
474 | popl %ecx | |
fe7cacc1 | 475 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 LT |
476 | movl %eax, %esp |
477 | xorl %edx, %edx | |
478 | call do_notify_resume | |
4031ff38 | 479 | jmp resume_userspace_sig |
64ca9004 | 480 | #endif |
1da177e4 LT |
481 | |
482 | # perform syscall exit tracing | |
483 | ALIGN | |
484 | syscall_trace_entry: | |
eb5b7b9d | 485 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 LT |
486 | movl %esp, %eax |
487 | xorl %edx,%edx | |
488 | call do_syscall_trace | |
ed75e8d5 | 489 | cmpl $0, %eax |
640aa46e | 490 | jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU, |
ed75e8d5 | 491 | # so must skip actual syscall |
eb5b7b9d | 492 | movl PT_ORIG_EAX(%esp), %eax |
1da177e4 LT |
493 | cmpl $(nr_syscalls), %eax |
494 | jnae syscall_call | |
495 | jmp syscall_exit | |
496 | ||
497 | # perform syscall exit tracing | |
498 | ALIGN | |
499 | syscall_exit_work: | |
500 | testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl | |
501 | jz work_pending | |
55f327fa | 502 | TRACE_IRQS_ON |
0da5db31 | 503 | ENABLE_INTERRUPTS # could let do_syscall_trace() call |
1da177e4 LT |
504 | # schedule() instead |
505 | movl %esp, %eax | |
506 | movl $1, %edx | |
507 | call do_syscall_trace | |
508 | jmp resume_userspace | |
fe7cacc1 | 509 | CFI_ENDPROC |
1da177e4 | 510 | |
fe7cacc1 | 511 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 LT |
512 | syscall_fault: |
513 | pushl %eax # save orig_eax | |
fe7cacc1 | 514 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
515 | SAVE_ALL |
516 | GET_THREAD_INFO(%ebp) | |
eb5b7b9d | 517 | movl $-EFAULT,PT_EAX(%esp) |
1da177e4 LT |
518 | jmp resume_userspace |
519 | ||
1da177e4 | 520 | syscall_badsys: |
eb5b7b9d | 521 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 | 522 | jmp resume_userspace |
fe7cacc1 | 523 | CFI_ENDPROC |
1da177e4 LT |
524 | |
525 | #define FIXUP_ESPFIX_STACK \ | |
be44d2aa SS |
526 | /* since we are on a wrong stack, we cant make it a C code :( */ \ |
527 | GET_THREAD_INFO(%ebp); \ | |
528 | movl TI_cpu(%ebp), %ebx; \ | |
529 | PER_CPU(cpu_gdt_descr, %ebx); \ | |
530 | movl GDS_address(%ebx), %ebx; \ | |
531 | GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \ | |
532 | addl %esp, %eax; \ | |
533 | pushl $__KERNEL_DS; \ | |
534 | CFI_ADJUST_CFA_OFFSET 4; \ | |
1da177e4 | 535 | pushl %eax; \ |
fe7cacc1 | 536 | CFI_ADJUST_CFA_OFFSET 4; \ |
be44d2aa SS |
537 | lss (%esp), %esp; \ |
538 | CFI_ADJUST_CFA_OFFSET -8; | |
539 | #define UNWIND_ESPFIX_STACK \ | |
1da177e4 | 540 | movl %ss, %eax; \ |
be44d2aa | 541 | /* see if on espfix stack */ \ |
1da177e4 | 542 | cmpw $__ESPFIX_SS, %ax; \ |
be44d2aa SS |
543 | jne 27f; \ |
544 | movl $__KERNEL_DS, %eax; \ | |
fe7cacc1 JB |
545 | movl %eax, %ds; \ |
546 | movl %eax, %es; \ | |
be44d2aa | 547 | /* switch to normal stack */ \ |
fe7cacc1 | 548 | FIXUP_ESPFIX_STACK; \ |
be44d2aa | 549 | 27:; |
1da177e4 LT |
550 | |
551 | /* | |
552 | * Build the entry stubs and pointer table with | |
553 | * some assembler magic. | |
554 | */ | |
555 | .data | |
556 | ENTRY(interrupt) | |
557 | .text | |
558 | ||
559 | vector=0 | |
560 | ENTRY(irq_entries_start) | |
fe7cacc1 | 561 | RING0_INT_FRAME |
1da177e4 LT |
562 | .rept NR_IRQS |
563 | ALIGN | |
fe7cacc1 JB |
564 | .if vector |
565 | CFI_ADJUST_CFA_OFFSET -4 | |
566 | .endif | |
19eadf98 | 567 | 1: pushl $~(vector) |
fe7cacc1 | 568 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
569 | jmp common_interrupt |
570 | .data | |
571 | .long 1b | |
572 | .text | |
573 | vector=vector+1 | |
574 | .endr | |
575 | ||
55f327fa IM |
576 | /* |
577 | * the CPU automatically disables interrupts when executing an IRQ vector, | |
578 | * so IRQ-flags tracing has to follow that: | |
579 | */ | |
1da177e4 LT |
580 | ALIGN |
581 | common_interrupt: | |
582 | SAVE_ALL | |
55f327fa | 583 | TRACE_IRQS_OFF |
1da177e4 LT |
584 | movl %esp,%eax |
585 | call do_IRQ | |
586 | jmp ret_from_intr | |
fe7cacc1 | 587 | CFI_ENDPROC |
1da177e4 LT |
588 | |
589 | #define BUILD_INTERRUPT(name, nr) \ | |
590 | ENTRY(name) \ | |
fe7cacc1 | 591 | RING0_INT_FRAME; \ |
19eadf98 | 592 | pushl $~(nr); \ |
fe7cacc1 JB |
593 | CFI_ADJUST_CFA_OFFSET 4; \ |
594 | SAVE_ALL; \ | |
55f327fa | 595 | TRACE_IRQS_OFF \ |
1da177e4 LT |
596 | movl %esp,%eax; \ |
597 | call smp_/**/name; \ | |
55f327fa | 598 | jmp ret_from_intr; \ |
fe7cacc1 | 599 | CFI_ENDPROC |
1da177e4 LT |
600 | |
601 | /* The include is where all of the SMP etc. interrupts come from */ | |
602 | #include "entry_arch.h" | |
603 | ||
d28c4393 P |
604 | KPROBE_ENTRY(page_fault) |
605 | RING0_EC_FRAME | |
606 | pushl $do_page_fault | |
fe7cacc1 | 607 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
608 | ALIGN |
609 | error_code: | |
f95d47ca JF |
610 | /* the function address is in %gs's slot on the stack */ |
611 | pushl %es | |
612 | CFI_ADJUST_CFA_OFFSET 4 | |
613 | /*CFI_REL_OFFSET es, 0*/ | |
1da177e4 | 614 | pushl %ds |
fe7cacc1 JB |
615 | CFI_ADJUST_CFA_OFFSET 4 |
616 | /*CFI_REL_OFFSET ds, 0*/ | |
1da177e4 | 617 | pushl %eax |
fe7cacc1 JB |
618 | CFI_ADJUST_CFA_OFFSET 4 |
619 | CFI_REL_OFFSET eax, 0 | |
1da177e4 | 620 | pushl %ebp |
fe7cacc1 JB |
621 | CFI_ADJUST_CFA_OFFSET 4 |
622 | CFI_REL_OFFSET ebp, 0 | |
1da177e4 | 623 | pushl %edi |
fe7cacc1 JB |
624 | CFI_ADJUST_CFA_OFFSET 4 |
625 | CFI_REL_OFFSET edi, 0 | |
1da177e4 | 626 | pushl %esi |
fe7cacc1 JB |
627 | CFI_ADJUST_CFA_OFFSET 4 |
628 | CFI_REL_OFFSET esi, 0 | |
1da177e4 | 629 | pushl %edx |
fe7cacc1 JB |
630 | CFI_ADJUST_CFA_OFFSET 4 |
631 | CFI_REL_OFFSET edx, 0 | |
1da177e4 | 632 | pushl %ecx |
fe7cacc1 JB |
633 | CFI_ADJUST_CFA_OFFSET 4 |
634 | CFI_REL_OFFSET ecx, 0 | |
1da177e4 | 635 | pushl %ebx |
fe7cacc1 JB |
636 | CFI_ADJUST_CFA_OFFSET 4 |
637 | CFI_REL_OFFSET ebx, 0 | |
1da177e4 | 638 | cld |
f95d47ca | 639 | pushl %gs |
fe7cacc1 | 640 | CFI_ADJUST_CFA_OFFSET 4 |
f95d47ca JF |
641 | /*CFI_REL_OFFSET gs, 0*/ |
642 | movl $(__KERNEL_PDA), %ecx | |
643 | movl %ecx, %gs | |
1da177e4 LT |
644 | UNWIND_ESPFIX_STACK |
645 | popl %ecx | |
fe7cacc1 JB |
646 | CFI_ADJUST_CFA_OFFSET -4 |
647 | /*CFI_REGISTER es, ecx*/ | |
f95d47ca | 648 | movl PT_GS(%esp), %edi # get the function address |
eb5b7b9d | 649 | movl PT_ORIG_EAX(%esp), %edx # get the error code |
f95d47ca JF |
650 | movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart |
651 | mov %ecx, PT_GS(%esp) | |
652 | /*CFI_REL_OFFSET gs, ES*/ | |
1da177e4 LT |
653 | movl $(__USER_DS), %ecx |
654 | movl %ecx, %ds | |
655 | movl %ecx, %es | |
656 | movl %esp,%eax # pt_regs pointer | |
657 | call *%edi | |
658 | jmp ret_from_exception | |
fe7cacc1 | 659 | CFI_ENDPROC |
d28c4393 | 660 | KPROBE_END(page_fault) |
1da177e4 LT |
661 | |
662 | ENTRY(coprocessor_error) | |
fe7cacc1 | 663 | RING0_INT_FRAME |
1da177e4 | 664 | pushl $0 |
fe7cacc1 | 665 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 666 | pushl $do_coprocessor_error |
fe7cacc1 | 667 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 668 | jmp error_code |
fe7cacc1 | 669 | CFI_ENDPROC |
1da177e4 LT |
670 | |
671 | ENTRY(simd_coprocessor_error) | |
fe7cacc1 | 672 | RING0_INT_FRAME |
1da177e4 | 673 | pushl $0 |
fe7cacc1 | 674 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 675 | pushl $do_simd_coprocessor_error |
fe7cacc1 | 676 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 677 | jmp error_code |
fe7cacc1 | 678 | CFI_ENDPROC |
1da177e4 LT |
679 | |
680 | ENTRY(device_not_available) | |
fe7cacc1 | 681 | RING0_INT_FRAME |
1da177e4 | 682 | pushl $-1 # mark this as an int |
fe7cacc1 | 683 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 684 | SAVE_ALL |
0da5db31 | 685 | GET_CR0_INTO_EAX |
1da177e4 LT |
686 | testl $0x4, %eax # EM (math emulation bit) |
687 | jne device_not_available_emulate | |
688 | preempt_stop | |
689 | call math_state_restore | |
690 | jmp ret_from_exception | |
691 | device_not_available_emulate: | |
692 | pushl $0 # temporary storage for ORIG_EIP | |
fe7cacc1 | 693 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
694 | call math_emulate |
695 | addl $4, %esp | |
fe7cacc1 | 696 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 697 | jmp ret_from_exception |
fe7cacc1 | 698 | CFI_ENDPROC |
1da177e4 LT |
699 | |
700 | /* | |
701 | * Debug traps and NMI can happen at the one SYSENTER instruction | |
702 | * that sets up the real kernel stack. Check here, since we can't | |
703 | * allow the wrong stack to be used. | |
704 | * | |
705 | * "TSS_sysenter_esp0+12" is because the NMI/debug handler will have | |
706 | * already pushed 3 words if it hits on the sysenter instruction: | |
707 | * eflags, cs and eip. | |
708 | * | |
709 | * We just load the right stack, and push the three (known) values | |
710 | * by hand onto the new stack - while updating the return eip past | |
711 | * the instruction that would have done it for sysenter. | |
712 | */ | |
713 | #define FIX_STACK(offset, ok, label) \ | |
714 | cmpw $__KERNEL_CS,4(%esp); \ | |
715 | jne ok; \ | |
716 | label: \ | |
717 | movl TSS_sysenter_esp0+offset(%esp),%esp; \ | |
a549b86d CE |
718 | CFI_DEF_CFA esp, 0; \ |
719 | CFI_UNDEFINED eip; \ | |
1da177e4 | 720 | pushfl; \ |
a549b86d | 721 | CFI_ADJUST_CFA_OFFSET 4; \ |
1da177e4 | 722 | pushl $__KERNEL_CS; \ |
a549b86d CE |
723 | CFI_ADJUST_CFA_OFFSET 4; \ |
724 | pushl $sysenter_past_esp; \ | |
725 | CFI_ADJUST_CFA_OFFSET 4; \ | |
726 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 727 | |
3d97ae5b | 728 | KPROBE_ENTRY(debug) |
fe7cacc1 | 729 | RING0_INT_FRAME |
1da177e4 LT |
730 | cmpl $sysenter_entry,(%esp) |
731 | jne debug_stack_correct | |
732 | FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn) | |
733 | debug_stack_correct: | |
734 | pushl $-1 # mark this as an int | |
fe7cacc1 | 735 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
736 | SAVE_ALL |
737 | xorl %edx,%edx # error code 0 | |
738 | movl %esp,%eax # pt_regs pointer | |
739 | call do_debug | |
1da177e4 | 740 | jmp ret_from_exception |
fe7cacc1 | 741 | CFI_ENDPROC |
d28c4393 P |
742 | KPROBE_END(debug) |
743 | ||
1da177e4 LT |
744 | /* |
745 | * NMI is doubly nasty. It can happen _while_ we're handling | |
746 | * a debug fault, and the debug fault hasn't yet been able to | |
747 | * clear up the stack. So we first check whether we got an | |
748 | * NMI on the sysenter entry path, but after that we need to | |
749 | * check whether we got an NMI on the debug path where the debug | |
750 | * fault happened on the sysenter path. | |
751 | */ |