Commit | Line | Data |
---|---|---|
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 | |
139ec7c4 RR |
57 | /* |
58 | * We use macros for low-level operations which need to be overridden | |
59 | * for paravirtualization. The following will never clobber any registers: | |
60 | * INTERRUPT_RETURN (aka. "iret") | |
61 | * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax") | |
d75cd22f | 62 | * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit"). |
139ec7c4 RR |
63 | * |
64 | * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must | |
65 | * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY). | |
66 | * Allowing a register to be clobbered can shrink the paravirt replacement | |
67 | * enough to patch inline, increasing performance. | |
68 | */ | |
69 | ||
1da177e4 LT |
70 | #define nr_syscalls ((syscall_table_size)/4) |
71 | ||
1da177e4 | 72 | #ifdef CONFIG_PREEMPT |
139ec7c4 | 73 | #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF |
1da177e4 | 74 | #else |
139ec7c4 | 75 | #define preempt_stop(clobbers) |
1da177e4 LT |
76 | #define resume_kernel restore_nocheck |
77 | #endif | |
78 | ||
55f327fa IM |
79 | .macro TRACE_IRQS_IRET |
80 | #ifdef CONFIG_TRACE_IRQFLAGS | |
ab68ed98 | 81 | testl $X86_EFLAGS_IF,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; \ | |
464d1a78 | 96 | pushl %fs; \ |
f95d47ca | 97 | CFI_ADJUST_CFA_OFFSET 4;\ |
464d1a78 | 98 | /*CFI_REL_OFFSET fs, 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 | 128 | movl %edx, %es; \ |
7c3576d2 | 129 | movl $(__KERNEL_PERCPU), %edx; \ |
464d1a78 | 130 | movl %edx, %fs |
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;*/\ | |
464d1a78 | 163 | 3: popl %fs; \ |
f95d47ca | 164 | CFI_ADJUST_CFA_OFFSET -4;\ |
464d1a78 | 165 | /*CFI_RESTORE fs;*/\ |
f95d47ca | 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 |
47a55cd7 | 224 | END(ret_from_fork) |
1da177e4 LT |
225 | |
226 | /* | |
227 | * Return to user mode is not as complex as all this looks, | |
228 | * but we want the default path for a system call return to | |
229 | * go as quickly as possible which is why some of this is | |
230 | * less clear than it otherwise should be. | |
231 | */ | |
232 | ||
233 | # userspace resumption stub bypassing syscall exit tracing | |
234 | ALIGN | |
fe7cacc1 | 235 | RING0_PTREGS_FRAME |
1da177e4 | 236 | ret_from_exception: |
139ec7c4 | 237 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
238 | ret_from_intr: |
239 | GET_THREAD_INFO(%ebp) | |
4031ff38 | 240 | check_userspace: |
eb5b7b9d JF |
241 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS |
242 | movb PT_CS(%esp), %al | |
ab68ed98 | 243 | andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax |
78be3706 RR |
244 | cmpl $USER_RPL, %eax |
245 | jb resume_kernel # not returning to v8086 or userspace | |
f95d47ca | 246 | |
1da177e4 | 247 | ENTRY(resume_userspace) |
c7e872e7 | 248 | LOCKDEP_SYS_EXIT |
139ec7c4 | 249 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
250 | # setting need_resched or sigpending |
251 | # between sampling and the iret | |
e32e58a9 | 252 | TRACE_IRQS_OFF |
1da177e4 LT |
253 | movl TI_flags(%ebp), %ecx |
254 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done on | |
255 | # int/exception return? | |
256 | jne work_pending | |
257 | jmp restore_all | |
47a55cd7 | 258 | END(ret_from_exception) |
1da177e4 LT |
259 | |
260 | #ifdef CONFIG_PREEMPT | |
261 | ENTRY(resume_kernel) | |
139ec7c4 | 262 | DISABLE_INTERRUPTS(CLBR_ANY) |
1da177e4 LT |
263 | cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ? |
264 | jnz restore_nocheck | |
265 | need_resched: | |
266 | movl TI_flags(%ebp), %ecx # need_resched set ? | |
267 | testb $_TIF_NEED_RESCHED, %cl | |
268 | jz restore_all | |
ab68ed98 | 269 | testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ? |
1da177e4 LT |
270 | jz restore_all |
271 | call preempt_schedule_irq | |
272 | jmp need_resched | |
47a55cd7 | 273 | END(resume_kernel) |
1da177e4 | 274 | #endif |
fe7cacc1 | 275 | CFI_ENDPROC |
1da177e4 LT |
276 | |
277 | /* SYSENTER_RETURN points to after the "sysenter" instruction in | |
278 | the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */ | |
279 | ||
280 | # sysenter call handler stub | |
0aa97fb2 | 281 | ENTRY(ia32_sysenter_target) |
fe7cacc1 | 282 | CFI_STARTPROC simple |
adf14236 | 283 | CFI_SIGNAL_FRAME |
fe7cacc1 JB |
284 | CFI_DEF_CFA esp, 0 |
285 | CFI_REGISTER esp, ebp | |
faca6227 | 286 | movl TSS_sysenter_sp0(%esp),%esp |
1da177e4 | 287 | sysenter_past_esp: |
55f327fa | 288 | /* |
d93c870b JF |
289 | * Interrupts are disabled here, but we can't trace it until |
290 | * enough kernel state to call TRACE_IRQS_OFF can be called - but | |
291 | * we immediately enable interrupts at that point anyway. | |
55f327fa | 292 | */ |
1da177e4 | 293 | pushl $(__USER_DS) |
fe7cacc1 JB |
294 | CFI_ADJUST_CFA_OFFSET 4 |
295 | /*CFI_REL_OFFSET ss, 0*/ | |
1da177e4 | 296 | pushl %ebp |
fe7cacc1 JB |
297 | CFI_ADJUST_CFA_OFFSET 4 |
298 | CFI_REL_OFFSET esp, 0 | |
1da177e4 | 299 | pushfl |
d93c870b | 300 | orl $X86_EFLAGS_IF, (%esp) |
fe7cacc1 | 301 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 302 | pushl $(__USER_CS) |
fe7cacc1 JB |
303 | CFI_ADJUST_CFA_OFFSET 4 |
304 | /*CFI_REL_OFFSET cs, 0*/ | |
e6e5494c IM |
305 | /* |
306 | * Push current_thread_info()->sysenter_return to the stack. | |
307 | * A tiny bit of offset fixup is necessary - 4*4 means the 4 words | |
308 | * pushed above; +8 corresponds to copy_thread's esp0 setting. | |
309 | */ | |
310 | pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp) | |
fe7cacc1 JB |
311 | CFI_ADJUST_CFA_OFFSET 4 |
312 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 313 | |
d93c870b JF |
314 | pushl %eax |
315 | CFI_ADJUST_CFA_OFFSET 4 | |
316 | SAVE_ALL | |
317 | ENABLE_INTERRUPTS(CLBR_NONE) | |
318 | ||
1da177e4 LT |
319 | /* |
320 | * Load the potential sixth argument from user stack. | |
321 | * Careful about security. | |
322 | */ | |
323 | cmpl $__PAGE_OFFSET-3,%ebp | |
324 | jae syscall_fault | |
325 | 1: movl (%ebp),%ebp | |
d93c870b | 326 | movl %ebp,PT_EBP(%esp) |
1da177e4 LT |
327 | .section __ex_table,"a" |
328 | .align 4 | |
329 | .long 1b,syscall_fault | |
330 | .previous | |
331 | ||
1da177e4 LT |
332 | GET_THREAD_INFO(%ebp) |
333 | ||
334 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ | |
ed75e8d5 | 335 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
336 | jnz syscall_trace_entry |
337 | cmpl $(nr_syscalls), %eax | |
338 | jae syscall_badsys | |
339 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 340 | movl %eax,PT_EAX(%esp) |
c7e872e7 | 341 | LOCKDEP_SYS_EXIT |
42c24fa2 | 342 | DISABLE_INTERRUPTS(CLBR_ANY) |
55f327fa | 343 | TRACE_IRQS_OFF |
1da177e4 LT |
344 | movl TI_flags(%ebp), %ecx |
345 | testw $_TIF_ALLWORK_MASK, %cx | |
346 | jne syscall_exit_work | |
347 | /* if something modifies registers it must also disable sysexit */ | |
eb5b7b9d JF |
348 | movl PT_EIP(%esp), %edx |
349 | movl PT_OLDESP(%esp), %ecx | |
1da177e4 | 350 | xorl %ebp,%ebp |
55f327fa | 351 | TRACE_IRQS_ON |
464d1a78 | 352 | 1: mov PT_FS(%esp), %fs |
d75cd22f | 353 | ENABLE_INTERRUPTS_SYSEXIT |
fe7cacc1 | 354 | CFI_ENDPROC |
f95d47ca | 355 | .pushsection .fixup,"ax" |
464d1a78 | 356 | 2: movl $0,PT_FS(%esp) |
f95d47ca JF |
357 | jmp 1b |
358 | .section __ex_table,"a" | |
359 | .align 4 | |
360 | .long 1b,2b | |
361 | .popsection | |
0aa97fb2 | 362 | ENDPROC(ia32_sysenter_target) |
1da177e4 LT |
363 | |
364 | # system call handler stub | |
365 | ENTRY(system_call) | |
fe7cacc1 | 366 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 367 | pushl %eax # save orig_eax |
fe7cacc1 | 368 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
369 | SAVE_ALL |
370 | GET_THREAD_INFO(%ebp) | |
ed75e8d5 | 371 | # system call tracing in operation / emulation |
1da177e4 | 372 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ |
ed75e8d5 | 373 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
374 | jnz syscall_trace_entry |
375 | cmpl $(nr_syscalls), %eax | |
376 | jae syscall_badsys | |
377 | syscall_call: | |
378 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 379 | movl %eax,PT_EAX(%esp) # store the return value |
1da177e4 | 380 | syscall_exit: |
c7e872e7 | 381 | LOCKDEP_SYS_EXIT |
139ec7c4 | 382 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
383 | # setting need_resched or sigpending |
384 | # between sampling and the iret | |
55f327fa | 385 | TRACE_IRQS_OFF |
ab68ed98 | 386 | testl $X86_EFLAGS_TF,PT_EFLAGS(%esp) # If tracing set singlestep flag on exit |
1e2e99f0 JW |
387 | jz no_singlestep |
388 | orl $_TIF_SINGLESTEP,TI_flags(%ebp) | |
389 | no_singlestep: | |
1da177e4 LT |
390 | movl TI_flags(%ebp), %ecx |
391 | testw $_TIF_ALLWORK_MASK, %cx # current->work | |
392 | jne syscall_exit_work | |
393 | ||
394 | restore_all: | |
eb5b7b9d JF |
395 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS |
396 | # Warning: PT_OLDSS(%esp) contains the wrong/random values if we | |
5df24082 SS |
397 | # are returning to the kernel. |
398 | # See comments in process.c:copy_thread() for details. | |
eb5b7b9d JF |
399 | movb PT_OLDSS(%esp), %ah |
400 | movb PT_CS(%esp), %al | |
ab68ed98 | 401 | andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax |
78be3706 | 402 | cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax |
fe7cacc1 | 403 | CFI_REMEMBER_STATE |
1da177e4 LT |
404 | je ldt_ss # returning to user-space with LDT SS |
405 | restore_nocheck: | |
55f327fa IM |
406 | TRACE_IRQS_IRET |
407 | restore_nocheck_notrace: | |
1da177e4 | 408 | RESTORE_REGS |
f95d47ca | 409 | addl $4, %esp # skip orig_eax/error_code |
fe7cacc1 | 410 | CFI_ADJUST_CFA_OFFSET -4 |
f7f3d791 | 411 | irq_return: |
3701d863 | 412 | INTERRUPT_RETURN |
1da177e4 | 413 | .section .fixup,"ax" |
90e9f536 | 414 | ENTRY(iret_exc) |
a879cbbb LT |
415 | pushl $0 # no error code |
416 | pushl $do_iret_error | |
417 | jmp error_code | |
1da177e4 LT |
418 | .previous |
419 | .section __ex_table,"a" | |
420 | .align 4 | |
3701d863 | 421 | .long irq_return,iret_exc |
1da177e4 LT |
422 | .previous |
423 | ||
fe7cacc1 | 424 | CFI_RESTORE_STATE |
1da177e4 | 425 | ldt_ss: |
eb5b7b9d | 426 | larl PT_OLDSS(%esp), %eax |
1da177e4 LT |
427 | jnz restore_nocheck |
428 | testl $0x00400000, %eax # returning to 32bit stack? | |
429 | jnz restore_nocheck # allright, normal return | |
d3561b7f RR |
430 | |
431 | #ifdef CONFIG_PARAVIRT | |
432 | /* | |
433 | * The kernel can't run on a non-flat stack if paravirt mode | |
434 | * is active. Rather than try to fixup the high bits of | |
435 | * ESP, bypass this code entirely. This may break DOSemu | |
436 | * and/or Wine support in a paravirt VM, although the option | |
437 | * is still available to implement the setting of the high | |
438 | * 16-bits in the INTERRUPT_RETURN paravirt-op. | |
439 | */ | |
93b1eab3 | 440 | cmpl $0, pv_info+PARAVIRT_enabled |
d3561b7f RR |
441 | jne restore_nocheck |
442 | #endif | |
443 | ||
1da177e4 LT |
444 | /* If returning to userspace with 16bit stack, |
445 | * try to fix the higher word of ESP, as the CPU | |
446 | * won't restore it. | |
447 | * This is an "official" bug of all the x86-compatible | |
448 | * CPUs, which we can try to work around to make | |
449 | * dosemu and wine happy. */ | |
eb5b7b9d | 450 | movl PT_OLDESP(%esp), %eax |
be44d2aa SS |
451 | movl %esp, %edx |
452 | call patch_espfix_desc | |
453 | pushl $__ESPFIX_SS | |
454 | CFI_ADJUST_CFA_OFFSET 4 | |
455 | pushl %eax | |
456 | CFI_ADJUST_CFA_OFFSET 4 | |
139ec7c4 | 457 | DISABLE_INTERRUPTS(CLBR_EAX) |
55f327fa | 458 | TRACE_IRQS_OFF |
be44d2aa SS |
459 | lss (%esp), %esp |
460 | CFI_ADJUST_CFA_OFFSET -8 | |
461 | jmp restore_nocheck | |
fe7cacc1 | 462 | CFI_ENDPROC |
47a55cd7 | 463 | ENDPROC(system_call) |
1da177e4 LT |
464 | |
465 | # perform work that needs to be done immediately before resumption | |
466 | ALIGN | |
fe7cacc1 | 467 | RING0_PTREGS_FRAME # can't unwind into user space anyway |
1da177e4 LT |
468 | work_pending: |
469 | testb $_TIF_NEED_RESCHED, %cl | |
470 | jz work_notifysig | |
471 | work_resched: | |
472 | call schedule | |
c7e872e7 | 473 | LOCKDEP_SYS_EXIT |
139ec7c4 | 474 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
475 | # setting need_resched or sigpending |
476 | # between sampling and the iret | |
55f327fa | 477 | TRACE_IRQS_OFF |
1da177e4 LT |
478 | movl TI_flags(%ebp), %ecx |
479 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done other | |
480 | # than syscall tracing? | |
481 | jz restore_all | |
482 | testb $_TIF_NEED_RESCHED, %cl | |
483 | jnz work_resched | |
484 | ||
485 | work_notifysig: # deal with pending signals and | |
486 | # notify-resume requests | |
74b47a78 | 487 | #ifdef CONFIG_VM86 |
ab68ed98 | 488 | testl $X86_EFLAGS_VM, PT_EFLAGS(%esp) |
1da177e4 LT |
489 | movl %esp, %eax |
490 | jne work_notifysig_v86 # returning to kernel-space or | |
491 | # vm86-space | |
492 | xorl %edx, %edx | |
493 | call do_notify_resume | |
4031ff38 | 494 | jmp resume_userspace_sig |
1da177e4 LT |
495 | |
496 | ALIGN | |
497 | work_notifysig_v86: | |
498 | pushl %ecx # save ti_flags for do_notify_resume | |
fe7cacc1 | 499 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
500 | call save_v86_state # %eax contains pt_regs pointer |
501 | popl %ecx | |
fe7cacc1 | 502 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 503 | movl %eax, %esp |
74b47a78 JK |
504 | #else |
505 | movl %esp, %eax | |
506 | #endif | |
1da177e4 LT |
507 | xorl %edx, %edx |
508 | call do_notify_resume | |
4031ff38 | 509 | jmp resume_userspace_sig |
47a55cd7 | 510 | END(work_pending) |
1da177e4 LT |
511 | |
512 | # perform syscall exit tracing | |
513 | ALIGN | |
514 | syscall_trace_entry: | |
eb5b7b9d | 515 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 LT |
516 | movl %esp, %eax |
517 | xorl %edx,%edx | |
518 | call do_syscall_trace | |
ed75e8d5 | 519 | cmpl $0, %eax |
640aa46e | 520 | jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU, |
ed75e8d5 | 521 | # so must skip actual syscall |
eb5b7b9d | 522 | movl PT_ORIG_EAX(%esp), %eax |
1da177e4 LT |
523 | cmpl $(nr_syscalls), %eax |
524 | jnae syscall_call | |
525 | jmp syscall_exit | |
47a55cd7 | 526 | END(syscall_trace_entry) |
1da177e4 LT |
527 | |
528 | # perform syscall exit tracing | |
529 | ALIGN | |
530 | syscall_exit_work: | |
531 | testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl | |
532 | jz work_pending | |
55f327fa | 533 | TRACE_IRQS_ON |
139ec7c4 | 534 | ENABLE_INTERRUPTS(CLBR_ANY) # could let do_syscall_trace() call |
1da177e4 LT |
535 | # schedule() instead |
536 | movl %esp, %eax | |
537 | movl $1, %edx | |
538 | call do_syscall_trace | |
539 | jmp resume_userspace | |
47a55cd7 | 540 | END(syscall_exit_work) |
fe7cacc1 | 541 | CFI_ENDPROC |
1da177e4 | 542 | |
fe7cacc1 | 543 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 544 | syscall_fault: |
1da177e4 | 545 | GET_THREAD_INFO(%ebp) |
eb5b7b9d | 546 | movl $-EFAULT,PT_EAX(%esp) |
1da177e4 | 547 | jmp resume_userspace |
47a55cd7 | 548 | END(syscall_fault) |
1da177e4 | 549 | |
1da177e4 | 550 | syscall_badsys: |
eb5b7b9d | 551 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 | 552 | jmp resume_userspace |
47a55cd7 | 553 | END(syscall_badsys) |
fe7cacc1 | 554 | CFI_ENDPROC |
1da177e4 LT |
555 | |
556 | #define FIXUP_ESPFIX_STACK \ | |
be44d2aa | 557 | /* since we are on a wrong stack, we cant make it a C code :( */ \ |
7a61d35d | 558 | PER_CPU(gdt_page, %ebx); \ |
be44d2aa SS |
559 | GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \ |
560 | addl %esp, %eax; \ | |
561 | pushl $__KERNEL_DS; \ | |
562 | CFI_ADJUST_CFA_OFFSET 4; \ | |
1da177e4 | 563 | pushl %eax; \ |
fe7cacc1 | 564 | CFI_ADJUST_CFA_OFFSET 4; \ |
be44d2aa SS |
565 | lss (%esp), %esp; \ |
566 | CFI_ADJUST_CFA_OFFSET -8; | |
567 | #define UNWIND_ESPFIX_STACK \ | |
1da177e4 | 568 | movl %ss, %eax; \ |
be44d2aa | 569 | /* see if on espfix stack */ \ |
1da177e4 | 570 | cmpw $__ESPFIX_SS, %ax; \ |
be44d2aa SS |
571 | jne 27f; \ |
572 | movl $__KERNEL_DS, %eax; \ | |
fe7cacc1 JB |
573 | movl %eax, %ds; \ |
574 | movl %eax, %es; \ | |
be44d2aa | 575 | /* switch to normal stack */ \ |
fe7cacc1 | 576 | FIXUP_ESPFIX_STACK; \ |
be44d2aa | 577 | 27:; |
1da177e4 LT |
578 | |
579 | /* | |
580 | * Build the entry stubs and pointer table with | |
581 | * some assembler magic. | |
582 | */ | |
3e7622f9 | 583 | .section .rodata,"a" |
1da177e4 LT |
584 | ENTRY(interrupt) |
585 | .text | |
586 | ||
1da177e4 | 587 | ENTRY(irq_entries_start) |
fe7cacc1 | 588 | RING0_INT_FRAME |
47a55cd7 | 589 | vector=0 |
1da177e4 LT |
590 | .rept NR_IRQS |
591 | ALIGN | |
fe7cacc1 JB |
592 | .if vector |
593 | CFI_ADJUST_CFA_OFFSET -4 | |
594 | .endif | |
19eadf98 | 595 | 1: pushl $~(vector) |
fe7cacc1 | 596 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 597 | jmp common_interrupt |
47a55cd7 | 598 | .previous |
1da177e4 | 599 | .long 1b |
47a55cd7 | 600 | .text |
1da177e4 LT |
601 | vector=vector+1 |
602 | .endr | |
47a55cd7 JB |
603 | END(irq_entries_start) |
604 | ||
605 | .previous | |
606 | END(interrupt) | |
607 | .previous | |
1da177e4 | 608 | |
55f327fa IM |
609 | /* |
610 | * the CPU automatically disables interrupts when executing an IRQ vector, | |
611 | * so IRQ-flags tracing has to follow that: | |
612 | */ | |
1da177e4 LT |
613 | ALIGN |
614 | common_interrupt: | |
615 | SAVE_ALL | |
55f327fa | 616 | TRACE_IRQS_OFF |
1da177e4 LT |
617 | movl %esp,%eax |
618 | call do_IRQ | |
619 | jmp ret_from_intr | |
47a55cd7 | 620 | ENDPROC(common_interrupt) |
fe7cacc1 | 621 | CFI_ENDPROC |
1da177e4 LT |
622 | |
623 | #define BUILD_INTERRUPT(name, nr) \ | |
624 | ENTRY(name) \ | |
fe7cacc1 | 625 | RING0_INT_FRAME; \ |
19eadf98 | 626 | pushl $~(nr); \ |
fe7cacc1 JB |
627 | CFI_ADJUST_CFA_OFFSET 4; \ |
628 | SAVE_ALL; \ | |
55f327fa | 629 | TRACE_IRQS_OFF \ |
1da177e4 | 630 | movl %esp,%eax; \ |
f76c3923 | 631 | call smp_##name; \ |
55f327fa | 632 | jmp ret_from_intr; \ |
47a55cd7 JB |
633 | CFI_ENDPROC; \ |
634 | ENDPROC(name) | |
1da177e4 LT |
635 | |
636 | /* The include is where all of the SMP etc. interrupts come from */ | |
637 | #include "entry_arch.h" | |
638 | ||
d28c4393 P |
639 | KPROBE_ENTRY(page_fault) |
640 | RING0_EC_FRAME | |
641 | pushl $do_page_fault | |
fe7cacc1 | 642 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
643 | ALIGN |
644 | error_code: | |
464d1a78 | 645 | /* the function address is in %fs's slot on the stack */ |
f95d47ca JF |
646 | pushl %es |
647 | CFI_ADJUST_CFA_OFFSET 4 | |
648 | /*CFI_REL_OFFSET es, 0*/ | |
1da177e4 | 649 | pushl %ds |
fe7cacc1 JB |
650 | CFI_ADJUST_CFA_OFFSET 4 |
651 | /*CFI_REL_OFFSET ds, 0*/ | |
1da177e4 | 652 | pushl %eax |
fe7cacc1 JB |
653 | CFI_ADJUST_CFA_OFFSET 4 |
654 | CFI_REL_OFFSET eax, 0 | |
1da177e4 | 655 | pushl %ebp |
fe7cacc1 JB |
656 | CFI_ADJUST_CFA_OFFSET 4 |
657 | CFI_REL_OFFSET ebp, 0 | |
1da177e4 | 658 | pushl %edi |
fe7cacc1 JB |
659 | CFI_ADJUST_CFA_OFFSET 4 |
660 | CFI_REL_OFFSET edi, 0 | |
1da177e4 | 661 | pushl %esi |
fe7cacc1 JB |
662 | CFI_ADJUST_CFA_OFFSET 4 |
663 | CFI_REL_OFFSET esi, 0 | |
1da177e4 | 664 | pushl %edx |
fe7cacc1 JB |
665 | CFI_ADJUST_CFA_OFFSET 4 |
666 | CFI_REL_OFFSET edx, 0 | |
1da177e4 | 667 | pushl %ecx |
fe7cacc1 JB |
668 | CFI_ADJUST_CFA_OFFSET 4 |
669 | CFI_REL_OFFSET ecx, 0 | |
1da177e4 | 670 | pushl %ebx |
fe7cacc1 JB |
671 | CFI_ADJUST_CFA_OFFSET 4 |
672 | CFI_REL_OFFSET ebx, 0 | |
1da177e4 | 673 | cld |
464d1a78 | 674 | pushl %fs |
fe7cacc1 | 675 | CFI_ADJUST_CFA_OFFSET 4 |
464d1a78 | 676 | /*CFI_REL_OFFSET fs, 0*/ |
7c3576d2 | 677 | movl $(__KERNEL_PERCPU), %ecx |
464d1a78 | 678 | movl %ecx, %fs |
1da177e4 LT |
679 | UNWIND_ESPFIX_STACK |
680 | popl %ecx | |
fe7cacc1 JB |
681 | CFI_ADJUST_CFA_OFFSET -4 |
682 | /*CFI_REGISTER es, ecx*/ | |
464d1a78 | 683 | movl PT_FS(%esp), %edi # get the function address |
eb5b7b9d | 684 | movl PT_ORIG_EAX(%esp), %edx # get the error code |
f95d47ca | 685 | movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart |
464d1a78 JF |
686 | mov %ecx, PT_FS(%esp) |
687 | /*CFI_REL_OFFSET fs, ES*/ | |
1da177e4 LT |
688 | movl $(__USER_DS), %ecx |
689 | movl %ecx, %ds | |
690 | movl %ecx, %es | |
691 | movl %esp,%eax # pt_regs pointer | |
692 | call *%edi | |
693 | jmp ret_from_exception | |
fe7cacc1 | 694 | CFI_ENDPROC |
d28c4393 | 695 | KPROBE_END(page_fault) |
1da177e4 LT |
696 | |
697 | ENTRY(coprocessor_error) | |
fe7cacc1 | 698 | RING0_INT_FRAME |
1da177e4 | 699 | pushl $0 |
fe7cacc1 | 700 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 701 | pushl $do_coprocessor_error |
fe7cacc1 | 702 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 703 | jmp error_code |
fe7cacc1 | 704 | CFI_ENDPROC |
47a55cd7 | 705 | END(coprocessor_error) |
1da177e4 LT |
706 | |
707 | ENTRY(simd_coprocessor_error) | |
fe7cacc1 | 708 | RING0_INT_FRAME |
1da177e4 | 709 | pushl $0 |
fe7cacc1 | 710 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 711 | pushl $do_simd_coprocessor_error |
fe7cacc1 | 712 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 713 | jmp error_code |
fe7cacc1 | 714 | CFI_ENDPROC |
47a55cd7 | 715 | END(simd_coprocessor_error) |
1da177e4 LT |
716 | |
717 | ENTRY(device_not_available) | |
fe7cacc1 | 718 | RING0_INT_FRAME |
1da177e4 | 719 | pushl $-1 # mark this as an int |
fe7cacc1 | 720 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 721 | SAVE_ALL |
0da5db31 | 722 | GET_CR0_INTO_EAX |
1da177e4 LT |
723 | testl $0x4, %eax # EM (math emulation bit) |
724 | jne device_not_available_emulate | |
139ec7c4 | 725 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
726 | call math_state_restore |
727 | jmp ret_from_exception | |
728 | device_not_available_emulate: | |
729 | pushl $0 # temporary storage for ORIG_EIP | |
fe7cacc1 | 730 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
731 | call math_emulate |
732 | addl $4, %esp | |
fe7cacc1 | 733 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 734 | jmp ret_from_exception |
fe7cacc1 | 735 | CFI_ENDPROC |
47a55cd7 | 736 | END(device_not_available) |
1da177e4 LT |
737 | |
738 | /* | |
739 | * Debug traps and NMI can happen at the one SYSENTER instruction | |
740 | * that sets up the real kernel stack. Check here, since we can't | |
741 | * allow the wrong stack to be used. | |
742 | * | |
faca6227 | 743 | * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have |
1da177e4 LT |
744 | * already pushed 3 words if it hits on the sysenter instruction: |
745 | * eflags, cs and eip. | |
746 | * | |
747 | * We just load the right stack, and push the three (known) values | |
748 | * by hand onto the new stack - while updating the return eip past | |
749 | * the instruction that would have done it for sysenter. | |
750 | */ | |
751 | #define FIX_STACK(offset, ok, label) \ | |
752 | cmpw $__KERNEL_CS,4(%esp); \ | |
753 | jne ok; \ | |
754 | label: \ | |
faca6227 | 755 | movl TSS_sysenter_sp0+offset(%esp),%esp; \ |
a549b86d CE |
756 | CFI_DEF_CFA esp, 0; \ |
757 | CFI_UNDEFINED eip; \ | |
1da177e4 | 758 | pushfl; \ |
a549b86d | 759 | CFI_ADJUST_CFA_OFFSET 4; \ |
1da177e4 | 760 | pushl $__KERNEL_CS; \ |
a549b86d CE |
761 | CFI_ADJUST_CFA_OFFSET 4; \ |
762 | pushl $sysenter_past_esp; \ | |
763 | CFI_ADJUST_CFA_OFFSET 4; \ | |
764 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 765 | |
3d97ae5b | 766 | KPROBE_ENTRY(debug) |
fe7cacc1 | 767 | RING0_INT_FRAME |
0aa97fb2 | 768 | cmpl $ia32_sysenter_target,(%esp) |
1da177e4 LT |
769 | jne debug_stack_correct |
770 | FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn) | |
771 | debug_stack_correct: | |
772 | pushl $-1 # mark this as an int | |
fe7cacc1 | 773 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
774 | SAVE_ALL |
775 | xorl %edx,%edx # error code 0 | |
776 | movl %esp,%eax # pt_regs pointer | |
777 | call do_debug | |
1da177e4 | 778 | jmp ret_from_exception |
fe7cacc1 | 779 | CFI_ENDPROC |
d28c4393 P |
780 | KPROBE_END(debug) |
781 | ||
1da177e4 LT |
782 | /* |
783 | * NMI is doubly nasty. It can happen _while_ we're handling | |
784 | * a debug fault, and the debug fault hasn't yet been able to | |
785 | * clear up the stack. So we first check whether we got an | |
786 | * NMI on the sysenter entry path, but after that we need to | |
787 | * check whether we got an NMI on the debug path where the debug | |
788 | * fault happened on the sysenter path. | |
789 | */ |