]>
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> |
1da177e4 LT |
54 | #include "irq_vectors.h" |
55 | ||
139ec7c4 RR |
56 | /* |
57 | * We use macros for low-level operations which need to be overridden | |
58 | * for paravirtualization. The following will never clobber any registers: | |
59 | * INTERRUPT_RETURN (aka. "iret") | |
60 | * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax") | |
6abcd98f | 61 | * ENABLE_INTERRUPTS_SYSCALL_RET (aka "sti; sysexit"). |
139ec7c4 RR |
62 | * |
63 | * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must | |
64 | * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY). | |
65 | * Allowing a register to be clobbered can shrink the paravirt replacement | |
66 | * enough to patch inline, increasing performance. | |
67 | */ | |
68 | ||
1da177e4 LT |
69 | #define nr_syscalls ((syscall_table_size)/4) |
70 | ||
1da177e4 | 71 | #ifdef CONFIG_PREEMPT |
139ec7c4 | 72 | #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF |
1da177e4 | 73 | #else |
139ec7c4 | 74 | #define preempt_stop(clobbers) |
1da177e4 LT |
75 | #define resume_kernel restore_nocheck |
76 | #endif | |
77 | ||
55f327fa IM |
78 | .macro TRACE_IRQS_IRET |
79 | #ifdef CONFIG_TRACE_IRQFLAGS | |
ab68ed98 | 80 | testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off? |
55f327fa IM |
81 | jz 1f |
82 | TRACE_IRQS_ON | |
83 | 1: | |
84 | #endif | |
85 | .endm | |
86 | ||
4031ff38 AG |
87 | #ifdef CONFIG_VM86 |
88 | #define resume_userspace_sig check_userspace | |
89 | #else | |
90 | #define resume_userspace_sig resume_userspace | |
91 | #endif | |
92 | ||
1da177e4 LT |
93 | #define SAVE_ALL \ |
94 | cld; \ | |
464d1a78 | 95 | pushl %fs; \ |
f95d47ca | 96 | CFI_ADJUST_CFA_OFFSET 4;\ |
464d1a78 | 97 | /*CFI_REL_OFFSET fs, 0;*/\ |
1da177e4 | 98 | pushl %es; \ |
fe7cacc1 JB |
99 | CFI_ADJUST_CFA_OFFSET 4;\ |
100 | /*CFI_REL_OFFSET es, 0;*/\ | |
1da177e4 | 101 | pushl %ds; \ |
fe7cacc1 JB |
102 | CFI_ADJUST_CFA_OFFSET 4;\ |
103 | /*CFI_REL_OFFSET ds, 0;*/\ | |
1da177e4 | 104 | pushl %eax; \ |
fe7cacc1 JB |
105 | CFI_ADJUST_CFA_OFFSET 4;\ |
106 | CFI_REL_OFFSET eax, 0;\ | |
1da177e4 | 107 | pushl %ebp; \ |
fe7cacc1 JB |
108 | CFI_ADJUST_CFA_OFFSET 4;\ |
109 | CFI_REL_OFFSET ebp, 0;\ | |
1da177e4 | 110 | pushl %edi; \ |
fe7cacc1 JB |
111 | CFI_ADJUST_CFA_OFFSET 4;\ |
112 | CFI_REL_OFFSET edi, 0;\ | |
1da177e4 | 113 | pushl %esi; \ |
fe7cacc1 JB |
114 | CFI_ADJUST_CFA_OFFSET 4;\ |
115 | CFI_REL_OFFSET esi, 0;\ | |
1da177e4 | 116 | pushl %edx; \ |
fe7cacc1 JB |
117 | CFI_ADJUST_CFA_OFFSET 4;\ |
118 | CFI_REL_OFFSET edx, 0;\ | |
1da177e4 | 119 | pushl %ecx; \ |
fe7cacc1 JB |
120 | CFI_ADJUST_CFA_OFFSET 4;\ |
121 | CFI_REL_OFFSET ecx, 0;\ | |
1da177e4 | 122 | pushl %ebx; \ |
fe7cacc1 JB |
123 | CFI_ADJUST_CFA_OFFSET 4;\ |
124 | CFI_REL_OFFSET ebx, 0;\ | |
1da177e4 LT |
125 | movl $(__USER_DS), %edx; \ |
126 | movl %edx, %ds; \ | |
f95d47ca | 127 | movl %edx, %es; \ |
7c3576d2 | 128 | movl $(__KERNEL_PERCPU), %edx; \ |
464d1a78 | 129 | movl %edx, %fs |
1da177e4 LT |
130 | |
131 | #define RESTORE_INT_REGS \ | |
132 | popl %ebx; \ | |
fe7cacc1 JB |
133 | CFI_ADJUST_CFA_OFFSET -4;\ |
134 | CFI_RESTORE ebx;\ | |
1da177e4 | 135 | popl %ecx; \ |
fe7cacc1 JB |
136 | CFI_ADJUST_CFA_OFFSET -4;\ |
137 | CFI_RESTORE ecx;\ | |
1da177e4 | 138 | popl %edx; \ |
fe7cacc1 JB |
139 | CFI_ADJUST_CFA_OFFSET -4;\ |
140 | CFI_RESTORE edx;\ | |
1da177e4 | 141 | popl %esi; \ |
fe7cacc1 JB |
142 | CFI_ADJUST_CFA_OFFSET -4;\ |
143 | CFI_RESTORE esi;\ | |
1da177e4 | 144 | popl %edi; \ |
fe7cacc1 JB |
145 | CFI_ADJUST_CFA_OFFSET -4;\ |
146 | CFI_RESTORE edi;\ | |
1da177e4 | 147 | popl %ebp; \ |
fe7cacc1 JB |
148 | CFI_ADJUST_CFA_OFFSET -4;\ |
149 | CFI_RESTORE ebp;\ | |
150 | popl %eax; \ | |
151 | CFI_ADJUST_CFA_OFFSET -4;\ | |
152 | CFI_RESTORE eax | |
1da177e4 LT |
153 | |
154 | #define RESTORE_REGS \ | |
155 | RESTORE_INT_REGS; \ | |
156 | 1: popl %ds; \ | |
fe7cacc1 JB |
157 | CFI_ADJUST_CFA_OFFSET -4;\ |
158 | /*CFI_RESTORE ds;*/\ | |
1da177e4 | 159 | 2: popl %es; \ |
fe7cacc1 JB |
160 | CFI_ADJUST_CFA_OFFSET -4;\ |
161 | /*CFI_RESTORE es;*/\ | |
464d1a78 | 162 | 3: popl %fs; \ |
f95d47ca | 163 | CFI_ADJUST_CFA_OFFSET -4;\ |
464d1a78 | 164 | /*CFI_RESTORE fs;*/\ |
f95d47ca | 165 | .pushsection .fixup,"ax"; \ |
1da177e4 | 166 | 4: movl $0,(%esp); \ |
f95d47ca JF |
167 | jmp 1b; \ |
168 | 5: movl $0,(%esp); \ | |
1da177e4 | 169 | jmp 2b; \ |
f95d47ca JF |
170 | 6: movl $0,(%esp); \ |
171 | jmp 3b; \ | |
1da177e4 LT |
172 | .section __ex_table,"a";\ |
173 | .align 4; \ | |
f95d47ca JF |
174 | .long 1b,4b; \ |
175 | .long 2b,5b; \ | |
176 | .long 3b,6b; \ | |
177 | .popsection | |
1da177e4 | 178 | |
fe7cacc1 JB |
179 | #define RING0_INT_FRAME \ |
180 | CFI_STARTPROC simple;\ | |
adf14236 | 181 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
182 | CFI_DEF_CFA esp, 3*4;\ |
183 | /*CFI_OFFSET cs, -2*4;*/\ | |
184 | CFI_OFFSET eip, -3*4 | |
185 | ||
186 | #define RING0_EC_FRAME \ | |
187 | CFI_STARTPROC simple;\ | |
adf14236 | 188 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
189 | CFI_DEF_CFA esp, 4*4;\ |
190 | /*CFI_OFFSET cs, -2*4;*/\ | |
191 | CFI_OFFSET eip, -3*4 | |
192 | ||
193 | #define RING0_PTREGS_FRAME \ | |
194 | CFI_STARTPROC simple;\ | |
adf14236 | 195 | CFI_SIGNAL_FRAME;\ |
eb5b7b9d JF |
196 | CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\ |
197 | /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\ | |
198 | CFI_OFFSET eip, PT_EIP-PT_OLDESP;\ | |
199 | /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\ | |
200 | /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\ | |
201 | CFI_OFFSET eax, PT_EAX-PT_OLDESP;\ | |
202 | CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\ | |
203 | CFI_OFFSET edi, PT_EDI-PT_OLDESP;\ | |
204 | CFI_OFFSET esi, PT_ESI-PT_OLDESP;\ | |
205 | CFI_OFFSET edx, PT_EDX-PT_OLDESP;\ | |
206 | CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\ | |
207 | CFI_OFFSET ebx, PT_EBX-PT_OLDESP | |
1da177e4 LT |
208 | |
209 | ENTRY(ret_from_fork) | |
fe7cacc1 | 210 | CFI_STARTPROC |
1da177e4 | 211 | pushl %eax |
25d7dfda | 212 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
213 | call schedule_tail |
214 | GET_THREAD_INFO(%ebp) | |
215 | popl %eax | |
fe7cacc1 | 216 | CFI_ADJUST_CFA_OFFSET -4 |
47a5c6fa LT |
217 | pushl $0x0202 # Reset kernel eflags |
218 | CFI_ADJUST_CFA_OFFSET 4 | |
219 | popfl | |
220 | CFI_ADJUST_CFA_OFFSET -4 | |
1da177e4 | 221 | jmp syscall_exit |
fe7cacc1 | 222 | CFI_ENDPROC |
47a55cd7 | 223 | END(ret_from_fork) |
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 | 235 | ret_from_exception: |
139ec7c4 | 236 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
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 | |
ab68ed98 | 242 | andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax |
78be3706 RR |
243 | cmpl $USER_RPL, %eax |
244 | jb resume_kernel # not returning to v8086 or userspace | |
f95d47ca | 245 | |
1da177e4 | 246 | ENTRY(resume_userspace) |
c7e872e7 | 247 | LOCKDEP_SYS_EXIT |
139ec7c4 | 248 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
249 | # setting need_resched or sigpending |
250 | # between sampling and the iret | |
251 | movl TI_flags(%ebp), %ecx | |
252 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done on | |
253 | # int/exception return? | |
254 | jne work_pending | |
255 | jmp restore_all | |
47a55cd7 | 256 | END(ret_from_exception) |
1da177e4 LT |
257 | |
258 | #ifdef CONFIG_PREEMPT | |
259 | ENTRY(resume_kernel) | |
139ec7c4 | 260 | DISABLE_INTERRUPTS(CLBR_ANY) |
1da177e4 LT |
261 | cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ? |
262 | jnz restore_nocheck | |
263 | need_resched: | |
264 | movl TI_flags(%ebp), %ecx # need_resched set ? | |
265 | testb $_TIF_NEED_RESCHED, %cl | |
266 | jz restore_all | |
ab68ed98 | 267 | testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ? |
1da177e4 LT |
268 | jz restore_all |
269 | call preempt_schedule_irq | |
270 | jmp need_resched | |
47a55cd7 | 271 | END(resume_kernel) |
1da177e4 | 272 | #endif |
fe7cacc1 | 273 | CFI_ENDPROC |
1da177e4 LT |
274 | |
275 | /* SYSENTER_RETURN points to after the "sysenter" instruction in | |
276 | the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */ | |
277 | ||
278 | # sysenter call handler stub | |
0aa97fb2 | 279 | ENTRY(ia32_sysenter_target) |
fe7cacc1 | 280 | CFI_STARTPROC simple |
adf14236 | 281 | CFI_SIGNAL_FRAME |
fe7cacc1 JB |
282 | CFI_DEF_CFA esp, 0 |
283 | CFI_REGISTER esp, ebp | |
faca6227 | 284 | movl TSS_sysenter_sp0(%esp),%esp |
1da177e4 | 285 | sysenter_past_esp: |
55f327fa | 286 | /* |
d93c870b JF |
287 | * Interrupts are disabled here, but we can't trace it until |
288 | * enough kernel state to call TRACE_IRQS_OFF can be called - but | |
289 | * we immediately enable interrupts at that point anyway. | |
55f327fa | 290 | */ |
1da177e4 | 291 | pushl $(__USER_DS) |
fe7cacc1 JB |
292 | CFI_ADJUST_CFA_OFFSET 4 |
293 | /*CFI_REL_OFFSET ss, 0*/ | |
1da177e4 | 294 | pushl %ebp |
fe7cacc1 JB |
295 | CFI_ADJUST_CFA_OFFSET 4 |
296 | CFI_REL_OFFSET esp, 0 | |
1da177e4 | 297 | pushfl |
d93c870b | 298 | orl $X86_EFLAGS_IF, (%esp) |
fe7cacc1 | 299 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 300 | pushl $(__USER_CS) |
fe7cacc1 JB |
301 | CFI_ADJUST_CFA_OFFSET 4 |
302 | /*CFI_REL_OFFSET cs, 0*/ | |
e6e5494c IM |
303 | /* |
304 | * Push current_thread_info()->sysenter_return to the stack. | |
305 | * A tiny bit of offset fixup is necessary - 4*4 means the 4 words | |
306 | * pushed above; +8 corresponds to copy_thread's esp0 setting. | |
307 | */ | |
308 | pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp) | |
fe7cacc1 JB |
309 | CFI_ADJUST_CFA_OFFSET 4 |
310 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 311 | |
d93c870b JF |
312 | pushl %eax |
313 | CFI_ADJUST_CFA_OFFSET 4 | |
314 | SAVE_ALL | |
315 | ENABLE_INTERRUPTS(CLBR_NONE) | |
316 | ||
1da177e4 LT |
317 | /* |
318 | * Load the potential sixth argument from user stack. | |
319 | * Careful about security. | |
320 | */ | |
321 | cmpl $__PAGE_OFFSET-3,%ebp | |
322 | jae syscall_fault | |
323 | 1: movl (%ebp),%ebp | |
d93c870b | 324 | movl %ebp,PT_EBP(%esp) |
1da177e4 LT |
325 | .section __ex_table,"a" |
326 | .align 4 | |
327 | .long 1b,syscall_fault | |
328 | .previous | |
329 | ||
1da177e4 LT |
330 | GET_THREAD_INFO(%ebp) |
331 | ||
332 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ | |
ed75e8d5 | 333 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
334 | jnz syscall_trace_entry |
335 | cmpl $(nr_syscalls), %eax | |
336 | jae syscall_badsys | |
337 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 338 | movl %eax,PT_EAX(%esp) |
c7e872e7 | 339 | LOCKDEP_SYS_EXIT |
42c24fa2 | 340 | DISABLE_INTERRUPTS(CLBR_ANY) |
55f327fa | 341 | TRACE_IRQS_OFF |
1da177e4 LT |
342 | movl TI_flags(%ebp), %ecx |
343 | testw $_TIF_ALLWORK_MASK, %cx | |
344 | jne syscall_exit_work | |
345 | /* if something modifies registers it must also disable sysexit */ | |
eb5b7b9d JF |
346 | movl PT_EIP(%esp), %edx |
347 | movl PT_OLDESP(%esp), %ecx | |
1da177e4 | 348 | xorl %ebp,%ebp |
55f327fa | 349 | TRACE_IRQS_ON |
464d1a78 | 350 | 1: mov PT_FS(%esp), %fs |
6abcd98f | 351 | ENABLE_INTERRUPTS_SYSCALL_RET |
fe7cacc1 | 352 | CFI_ENDPROC |
f95d47ca | 353 | .pushsection .fixup,"ax" |
464d1a78 | 354 | 2: movl $0,PT_FS(%esp) |
f95d47ca JF |
355 | jmp 1b |
356 | .section __ex_table,"a" | |
357 | .align 4 | |
358 | .long 1b,2b | |
359 | .popsection | |
0aa97fb2 | 360 | ENDPROC(ia32_sysenter_target) |
1da177e4 LT |
361 | |
362 | # system call handler stub | |
363 | ENTRY(system_call) | |
fe7cacc1 | 364 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 365 | pushl %eax # save orig_eax |
fe7cacc1 | 366 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
367 | SAVE_ALL |
368 | GET_THREAD_INFO(%ebp) | |
ed75e8d5 | 369 | # system call tracing in operation / emulation |
1da177e4 | 370 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ |
ed75e8d5 | 371 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
372 | jnz syscall_trace_entry |
373 | cmpl $(nr_syscalls), %eax | |
374 | jae syscall_badsys | |
375 | syscall_call: | |
376 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 377 | movl %eax,PT_EAX(%esp) # store the return value |
1da177e4 | 378 | syscall_exit: |
c7e872e7 | 379 | LOCKDEP_SYS_EXIT |
139ec7c4 | 380 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
381 | # setting need_resched or sigpending |
382 | # between sampling and the iret | |
55f327fa | 383 | TRACE_IRQS_OFF |
ab68ed98 | 384 | testl $X86_EFLAGS_TF,PT_EFLAGS(%esp) # If tracing set singlestep flag on exit |
1e2e99f0 JW |
385 | jz no_singlestep |
386 | orl $_TIF_SINGLESTEP,TI_flags(%ebp) | |
387 | no_singlestep: | |
1da177e4 LT |
388 | movl TI_flags(%ebp), %ecx |
389 | testw $_TIF_ALLWORK_MASK, %cx # current->work | |
390 | jne syscall_exit_work | |
391 | ||
392 | restore_all: | |
eb5b7b9d JF |
393 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS |
394 | # Warning: PT_OLDSS(%esp) contains the wrong/random values if we | |
5df24082 SS |
395 | # are returning to the kernel. |
396 | # See comments in process.c:copy_thread() for details. | |
eb5b7b9d JF |
397 | movb PT_OLDSS(%esp), %ah |
398 | movb PT_CS(%esp), %al | |
ab68ed98 | 399 | andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax |
78be3706 | 400 | cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax |
fe7cacc1 | 401 | CFI_REMEMBER_STATE |
1da177e4 LT |
402 | je ldt_ss # returning to user-space with LDT SS |
403 | restore_nocheck: | |
55f327fa IM |
404 | TRACE_IRQS_IRET |
405 | restore_nocheck_notrace: | |
1da177e4 | 406 | RESTORE_REGS |
f95d47ca | 407 | addl $4, %esp # skip orig_eax/error_code |
fe7cacc1 | 408 | CFI_ADJUST_CFA_OFFSET -4 |
f7f3d791 | 409 | irq_return: |
3701d863 | 410 | INTERRUPT_RETURN |
1da177e4 | 411 | .section .fixup,"ax" |
90e9f536 | 412 | ENTRY(iret_exc) |
a879cbbb LT |
413 | pushl $0 # no error code |
414 | pushl $do_iret_error | |
415 | jmp error_code | |
1da177e4 LT |
416 | .previous |
417 | .section __ex_table,"a" | |
418 | .align 4 | |
3701d863 | 419 | .long irq_return,iret_exc |
1da177e4 LT |
420 | .previous |
421 | ||
fe7cacc1 | 422 | CFI_RESTORE_STATE |
1da177e4 | 423 | ldt_ss: |
eb5b7b9d | 424 | larl PT_OLDSS(%esp), %eax |
1da177e4 LT |
425 | jnz restore_nocheck |
426 | testl $0x00400000, %eax # returning to 32bit stack? | |
427 | jnz restore_nocheck # allright, normal return | |
d3561b7f RR |
428 | |
429 | #ifdef CONFIG_PARAVIRT | |
430 | /* | |
431 | * The kernel can't run on a non-flat stack if paravirt mode | |
432 | * is active. Rather than try to fixup the high bits of | |
433 | * ESP, bypass this code entirely. This may break DOSemu | |
434 | * and/or Wine support in a paravirt VM, although the option | |
435 | * is still available to implement the setting of the high | |
436 | * 16-bits in the INTERRUPT_RETURN paravirt-op. | |
437 | */ | |
93b1eab3 | 438 | cmpl $0, pv_info+PARAVIRT_enabled |
d3561b7f RR |
439 | jne restore_nocheck |
440 | #endif | |
441 | ||
1da177e4 LT |
442 | /* If returning to userspace with 16bit stack, |
443 | * try to fix the higher word of ESP, as the CPU | |
444 | * won't restore it. | |
445 | * This is an "official" bug of all the x86-compatible | |
446 | * CPUs, which we can try to work around to make | |
447 | * dosemu and wine happy. */ | |
eb5b7b9d | 448 | movl PT_OLDESP(%esp), %eax |
be44d2aa SS |
449 | movl %esp, %edx |
450 | call patch_espfix_desc | |
451 | pushl $__ESPFIX_SS | |
452 | CFI_ADJUST_CFA_OFFSET 4 | |
453 | pushl %eax | |
454 | CFI_ADJUST_CFA_OFFSET 4 | |
139ec7c4 | 455 | DISABLE_INTERRUPTS(CLBR_EAX) |
55f327fa | 456 | TRACE_IRQS_OFF |
be44d2aa SS |
457 | lss (%esp), %esp |
458 | CFI_ADJUST_CFA_OFFSET -8 | |
459 | jmp restore_nocheck | |
fe7cacc1 | 460 | CFI_ENDPROC |
47a55cd7 | 461 | ENDPROC(system_call) |
1da177e4 LT |
462 | |
463 | # perform work that needs to be done immediately before resumption | |
464 | ALIGN | |
fe7cacc1 | 465 | RING0_PTREGS_FRAME # can't unwind into user space anyway |
1da177e4 LT |
466 | work_pending: |
467 | testb $_TIF_NEED_RESCHED, %cl | |
468 | jz work_notifysig | |
469 | work_resched: | |
470 | call schedule | |
c7e872e7 | 471 | LOCKDEP_SYS_EXIT |
139ec7c4 | 472 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
473 | # setting need_resched or sigpending |
474 | # between sampling and the iret | |
55f327fa | 475 | TRACE_IRQS_OFF |
1da177e4 LT |
476 | movl TI_flags(%ebp), %ecx |
477 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done other | |
478 | # than syscall tracing? | |
479 | jz restore_all | |
480 | testb $_TIF_NEED_RESCHED, %cl | |
481 | jnz work_resched | |
482 | ||
483 | work_notifysig: # deal with pending signals and | |
484 | # notify-resume requests | |
74b47a78 | 485 | #ifdef CONFIG_VM86 |
ab68ed98 | 486 | testl $X86_EFLAGS_VM, PT_EFLAGS(%esp) |
1da177e4 LT |
487 | movl %esp, %eax |
488 | jne work_notifysig_v86 # returning to kernel-space or | |
489 | # vm86-space | |
490 | xorl %edx, %edx | |
491 | call do_notify_resume | |
4031ff38 | 492 | jmp resume_userspace_sig |
1da177e4 LT |
493 | |
494 | ALIGN | |
495 | work_notifysig_v86: | |
496 | pushl %ecx # save ti_flags for do_notify_resume | |
fe7cacc1 | 497 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
498 | call save_v86_state # %eax contains pt_regs pointer |
499 | popl %ecx | |
fe7cacc1 | 500 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 501 | movl %eax, %esp |
74b47a78 JK |
502 | #else |
503 | movl %esp, %eax | |
504 | #endif | |
1da177e4 LT |
505 | xorl %edx, %edx |
506 | call do_notify_resume | |
4031ff38 | 507 | jmp resume_userspace_sig |
47a55cd7 | 508 | END(work_pending) |
1da177e4 LT |
509 | |
510 | # perform syscall exit tracing | |
511 | ALIGN | |
512 | syscall_trace_entry: | |
eb5b7b9d | 513 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 LT |
514 | movl %esp, %eax |
515 | xorl %edx,%edx | |
516 | call do_syscall_trace | |
ed75e8d5 | 517 | cmpl $0, %eax |
640aa46e | 518 | jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU, |
ed75e8d5 | 519 | # so must skip actual syscall |
eb5b7b9d | 520 | movl PT_ORIG_EAX(%esp), %eax |
1da177e4 LT |
521 | cmpl $(nr_syscalls), %eax |
522 | jnae syscall_call | |
523 | jmp syscall_exit | |
47a55cd7 | 524 | END(syscall_trace_entry) |
1da177e4 LT |
525 | |
526 | # perform syscall exit tracing | |
527 | ALIGN | |
528 | syscall_exit_work: | |
529 | testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl | |
530 | jz work_pending | |
55f327fa | 531 | TRACE_IRQS_ON |
139ec7c4 | 532 | ENABLE_INTERRUPTS(CLBR_ANY) # could let do_syscall_trace() call |
1da177e4 LT |
533 | # schedule() instead |
534 | movl %esp, %eax | |
535 | movl $1, %edx | |
536 | call do_syscall_trace | |
537 | jmp resume_userspace | |
47a55cd7 | 538 | END(syscall_exit_work) |
fe7cacc1 | 539 | CFI_ENDPROC |
1da177e4 | 540 | |
fe7cacc1 | 541 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 542 | syscall_fault: |
1da177e4 | 543 | GET_THREAD_INFO(%ebp) |
eb5b7b9d | 544 | movl $-EFAULT,PT_EAX(%esp) |
1da177e4 | 545 | jmp resume_userspace |
47a55cd7 | 546 | END(syscall_fault) |
1da177e4 | 547 | |
1da177e4 | 548 | syscall_badsys: |
eb5b7b9d | 549 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 | 550 | jmp resume_userspace |
47a55cd7 | 551 | END(syscall_badsys) |
fe7cacc1 | 552 | CFI_ENDPROC |
1da177e4 LT |
553 | |
554 | #define FIXUP_ESPFIX_STACK \ | |
be44d2aa | 555 | /* since we are on a wrong stack, we cant make it a C code :( */ \ |
7a61d35d | 556 | PER_CPU(gdt_page, %ebx); \ |
be44d2aa SS |
557 | GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \ |
558 | addl %esp, %eax; \ | |
559 | pushl $__KERNEL_DS; \ | |
560 | CFI_ADJUST_CFA_OFFSET 4; \ | |
1da177e4 | 561 | pushl %eax; \ |
fe7cacc1 | 562 | CFI_ADJUST_CFA_OFFSET 4; \ |
be44d2aa SS |
563 | lss (%esp), %esp; \ |
564 | CFI_ADJUST_CFA_OFFSET -8; | |
565 | #define UNWIND_ESPFIX_STACK \ | |
1da177e4 | 566 | movl %ss, %eax; \ |
be44d2aa | 567 | /* see if on espfix stack */ \ |
1da177e4 | 568 | cmpw $__ESPFIX_SS, %ax; \ |
be44d2aa SS |
569 | jne 27f; \ |
570 | movl $__KERNEL_DS, %eax; \ | |
fe7cacc1 JB |
571 | movl %eax, %ds; \ |
572 | movl %eax, %es; \ | |
be44d2aa | 573 | /* switch to normal stack */ \ |
fe7cacc1 | 574 | FIXUP_ESPFIX_STACK; \ |
be44d2aa | 575 | 27:; |
1da177e4 LT |
576 | |
577 | /* | |
578 | * Build the entry stubs and pointer table with | |
579 | * some assembler magic. | |
580 | */ | |
3e7622f9 | 581 | .section .rodata,"a" |
1da177e4 LT |
582 | ENTRY(interrupt) |
583 | .text | |
584 | ||
1da177e4 | 585 | ENTRY(irq_entries_start) |
fe7cacc1 | 586 | RING0_INT_FRAME |
47a55cd7 | 587 | vector=0 |
1da177e4 LT |
588 | .rept NR_IRQS |
589 | ALIGN | |
fe7cacc1 JB |
590 | .if vector |
591 | CFI_ADJUST_CFA_OFFSET -4 | |
592 | .endif | |
19eadf98 | 593 | 1: pushl $~(vector) |
fe7cacc1 | 594 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 595 | jmp common_interrupt |
47a55cd7 | 596 | .previous |
1da177e4 | 597 | .long 1b |
47a55cd7 | 598 | .text |
1da177e4 LT |
599 | vector=vector+1 |
600 | .endr | |
47a55cd7 JB |
601 | END(irq_entries_start) |
602 | ||
603 | .previous | |
604 | END(interrupt) | |
605 | .previous | |
1da177e4 | 606 | |
55f327fa IM |
607 | /* |
608 | * the CPU automatically disables interrupts when executing an IRQ vector, | |
609 | * so IRQ-flags tracing has to follow that: | |
610 | */ | |
1da177e4 LT |
611 | ALIGN |
612 | common_interrupt: | |
613 | SAVE_ALL | |
55f327fa | 614 | TRACE_IRQS_OFF |
1da177e4 LT |
615 | movl %esp,%eax |
616 | call do_IRQ | |
617 | jmp ret_from_intr | |
47a55cd7 | 618 | ENDPROC(common_interrupt) |
fe7cacc1 | 619 | CFI_ENDPROC |
1da177e4 LT |
620 | |
621 | #define BUILD_INTERRUPT(name, nr) \ | |
622 | ENTRY(name) \ | |
fe7cacc1 | 623 | RING0_INT_FRAME; \ |
19eadf98 | 624 | pushl $~(nr); \ |
fe7cacc1 JB |
625 | CFI_ADJUST_CFA_OFFSET 4; \ |
626 | SAVE_ALL; \ | |
55f327fa | 627 | TRACE_IRQS_OFF \ |
1da177e4 | 628 | movl %esp,%eax; \ |
f76c3923 | 629 | call smp_##name; \ |
55f327fa | 630 | jmp ret_from_intr; \ |
47a55cd7 JB |
631 | CFI_ENDPROC; \ |
632 | ENDPROC(name) | |
1da177e4 LT |
633 | |
634 | /* The include is where all of the SMP etc. interrupts come from */ | |
635 | #include "entry_arch.h" | |
636 | ||
d28c4393 P |
637 | KPROBE_ENTRY(page_fault) |
638 | RING0_EC_FRAME | |
639 | pushl $do_page_fault | |
fe7cacc1 | 640 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
641 | ALIGN |
642 | error_code: | |
464d1a78 | 643 | /* the function address is in %fs's slot on the stack */ |
f95d47ca JF |
644 | pushl %es |
645 | CFI_ADJUST_CFA_OFFSET 4 | |
646 | /*CFI_REL_OFFSET es, 0*/ | |
1da177e4 | 647 | pushl %ds |
fe7cacc1 JB |
648 | CFI_ADJUST_CFA_OFFSET 4 |
649 | /*CFI_REL_OFFSET ds, 0*/ | |
1da177e4 | 650 | pushl %eax |
fe7cacc1 JB |
651 | CFI_ADJUST_CFA_OFFSET 4 |
652 | CFI_REL_OFFSET eax, 0 | |
1da177e4 | 653 | pushl %ebp |
fe7cacc1 JB |
654 | CFI_ADJUST_CFA_OFFSET 4 |
655 | CFI_REL_OFFSET ebp, 0 | |
1da177e4 | 656 | pushl %edi |
fe7cacc1 JB |
657 | CFI_ADJUST_CFA_OFFSET 4 |
658 | CFI_REL_OFFSET edi, 0 | |
1da177e4 | 659 | pushl %esi |
fe7cacc1 JB |
660 | CFI_ADJUST_CFA_OFFSET 4 |
661 | CFI_REL_OFFSET esi, 0 | |
1da177e4 | 662 | pushl %edx |
fe7cacc1 JB |
663 | CFI_ADJUST_CFA_OFFSET 4 |
664 | CFI_REL_OFFSET edx, 0 | |
1da177e4 | 665 | pushl %ecx |
fe7cacc1 JB |
666 | CFI_ADJUST_CFA_OFFSET 4 |
667 | CFI_REL_OFFSET ecx, 0 | |
1da177e4 | 668 | pushl %ebx |
fe7cacc1 JB |
669 | CFI_ADJUST_CFA_OFFSET 4 |
670 | CFI_REL_OFFSET ebx, 0 | |
1da177e4 | 671 | cld |
464d1a78 | 672 | pushl %fs |
fe7cacc1 | 673 | CFI_ADJUST_CFA_OFFSET 4 |
464d1a78 | 674 | /*CFI_REL_OFFSET fs, 0*/ |
7c3576d2 | 675 | movl $(__KERNEL_PERCPU), %ecx |
464d1a78 | 676 | movl %ecx, %fs |
1da177e4 LT |
677 | UNWIND_ESPFIX_STACK |
678 | popl %ecx | |
fe7cacc1 JB |
679 | CFI_ADJUST_CFA_OFFSET -4 |
680 | /*CFI_REGISTER es, ecx*/ | |
464d1a78 | 681 | movl PT_FS(%esp), %edi # get the function address |
eb5b7b9d | 682 | movl PT_ORIG_EAX(%esp), %edx # get the error code |
f95d47ca | 683 | movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart |
464d1a78 JF |
684 | mov %ecx, PT_FS(%esp) |
685 | /*CFI_REL_OFFSET fs, ES*/ | |
1da177e4 LT |
686 | movl $(__USER_DS), %ecx |
687 | movl %ecx, %ds | |
688 | movl %ecx, %es | |
689 | movl %esp,%eax # pt_regs pointer | |
690 | call *%edi | |
691 | jmp ret_from_exception | |
fe7cacc1 | 692 | CFI_ENDPROC |
d28c4393 | 693 | KPROBE_END(page_fault) |
1da177e4 LT |
694 | |
695 | ENTRY(coprocessor_error) | |
fe7cacc1 | 696 | RING0_INT_FRAME |
1da177e4 | 697 | pushl $0 |
fe7cacc1 | 698 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 699 | pushl $do_coprocessor_error |
fe7cacc1 | 700 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 701 | jmp error_code |
fe7cacc1 | 702 | CFI_ENDPROC |
47a55cd7 | 703 | END(coprocessor_error) |
1da177e4 LT |
704 | |
705 | ENTRY(simd_coprocessor_error) | |
fe7cacc1 | 706 | RING0_INT_FRAME |
1da177e4 | 707 | pushl $0 |
fe7cacc1 | 708 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 709 | pushl $do_simd_coprocessor_error |
fe7cacc1 | 710 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 711 | jmp error_code |
fe7cacc1 | 712 | CFI_ENDPROC |
47a55cd7 | 713 | END(simd_coprocessor_error) |
1da177e4 LT |
714 | |
715 | ENTRY(device_not_available) | |
fe7cacc1 | 716 | RING0_INT_FRAME |
1da177e4 | 717 | pushl $-1 # mark this as an int |
fe7cacc1 | 718 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 719 | SAVE_ALL |
0da5db31 | 720 | GET_CR0_INTO_EAX |
1da177e4 LT |
721 | testl $0x4, %eax # EM (math emulation bit) |
722 | jne device_not_available_emulate | |
139ec7c4 | 723 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
724 | call math_state_restore |
725 | jmp ret_from_exception | |
726 | device_not_available_emulate: | |
727 | pushl $0 # temporary storage for ORIG_EIP | |
fe7cacc1 | 728 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
729 | call math_emulate |
730 | addl $4, %esp | |
fe7cacc1 | 731 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 732 | jmp ret_from_exception |
fe7cacc1 | 733 | CFI_ENDPROC |
47a55cd7 | 734 | END(device_not_available) |
1da177e4 LT |
735 | |
736 | /* | |
737 | * Debug traps and NMI can happen at the one SYSENTER instruction | |
738 | * that sets up the real kernel stack. Check here, since we can't | |
739 | * allow the wrong stack to be used. | |
740 | * | |
faca6227 | 741 | * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have |
1da177e4 LT |
742 | * already pushed 3 words if it hits on the sysenter instruction: |
743 | * eflags, cs and eip. | |
744 | * | |
745 | * We just load the right stack, and push the three (known) values | |
746 | * by hand onto the new stack - while updating the return eip past | |
747 | * the instruction that would have done it for sysenter. | |
748 | */ | |
749 | #define FIX_STACK(offset, ok, label) \ | |
750 | cmpw $__KERNEL_CS,4(%esp); \ | |
751 | jne ok; \ | |
752 | label: \ | |
faca6227 | 753 | movl TSS_sysenter_sp0+offset(%esp),%esp; \ |
a549b86d CE |
754 | CFI_DEF_CFA esp, 0; \ |
755 | CFI_UNDEFINED eip; \ | |
1da177e4 | 756 | pushfl; \ |
a549b86d | 757 | CFI_ADJUST_CFA_OFFSET 4; \ |
1da177e4 | 758 | pushl $__KERNEL_CS; \ |
a549b86d CE |
759 | CFI_ADJUST_CFA_OFFSET 4; \ |
760 | pushl $sysenter_past_esp; \ | |
761 | CFI_ADJUST_CFA_OFFSET 4; \ | |
762 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 763 | |
3d97ae5b | 764 | KPROBE_ENTRY(debug) |
fe7cacc1 | 765 | RING0_INT_FRAME |
0aa97fb2 | 766 | cmpl $ia32_sysenter_target,(%esp) |
1da177e4 LT |
767 | jne debug_stack_correct |
768 | FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn) | |
769 | debug_stack_correct: | |
770 | pushl $-1 # mark this as an int | |
fe7cacc1 | 771 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
772 | SAVE_ALL |
773 | xorl %edx,%edx # error code 0 | |
774 | movl %esp,%eax # pt_regs pointer | |
775 | call do_debug | |
1da177e4 | 776 | jmp ret_from_exception |
fe7cacc1 | 777 | CFI_ENDPROC |
d28c4393 P |
778 | KPROBE_END(debug) |
779 | ||
1da177e4 LT |
780 | /* |
781 | * NMI is doubly nasty. It can happen _while_ we're handling | |
782 | * a debug fault, and the debug fault hasn't yet been able to | |
783 | * clear up the stack. So we first check whether we got an | |
784 | * NMI on the sysenter entry path, but after that we need to | |
785 | * check whether we got an NMI on the debug path where the debug | |
786 | * fault happened on the sysenter path. | |
787 | */ | |