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 | |
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 | |
94 | 1: | |
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 | ||
1da177e4 LT |
104 | #define SAVE_ALL \ |
105 | cld; \ | |
464d1a78 | 106 | pushl %fs; \ |
f95d47ca | 107 | CFI_ADJUST_CFA_OFFSET 4;\ |
464d1a78 | 108 | /*CFI_REL_OFFSET fs, 0;*/\ |
1da177e4 | 109 | pushl %es; \ |
fe7cacc1 JB |
110 | CFI_ADJUST_CFA_OFFSET 4;\ |
111 | /*CFI_REL_OFFSET es, 0;*/\ | |
1da177e4 | 112 | pushl %ds; \ |
fe7cacc1 JB |
113 | CFI_ADJUST_CFA_OFFSET 4;\ |
114 | /*CFI_REL_OFFSET ds, 0;*/\ | |
1da177e4 | 115 | pushl %eax; \ |
fe7cacc1 JB |
116 | CFI_ADJUST_CFA_OFFSET 4;\ |
117 | CFI_REL_OFFSET eax, 0;\ | |
1da177e4 | 118 | pushl %ebp; \ |
fe7cacc1 JB |
119 | CFI_ADJUST_CFA_OFFSET 4;\ |
120 | CFI_REL_OFFSET ebp, 0;\ | |
1da177e4 | 121 | pushl %edi; \ |
fe7cacc1 JB |
122 | CFI_ADJUST_CFA_OFFSET 4;\ |
123 | CFI_REL_OFFSET edi, 0;\ | |
1da177e4 | 124 | pushl %esi; \ |
fe7cacc1 JB |
125 | CFI_ADJUST_CFA_OFFSET 4;\ |
126 | CFI_REL_OFFSET esi, 0;\ | |
1da177e4 | 127 | pushl %edx; \ |
fe7cacc1 JB |
128 | CFI_ADJUST_CFA_OFFSET 4;\ |
129 | CFI_REL_OFFSET edx, 0;\ | |
1da177e4 | 130 | pushl %ecx; \ |
fe7cacc1 JB |
131 | CFI_ADJUST_CFA_OFFSET 4;\ |
132 | CFI_REL_OFFSET ecx, 0;\ | |
1da177e4 | 133 | pushl %ebx; \ |
fe7cacc1 JB |
134 | CFI_ADJUST_CFA_OFFSET 4;\ |
135 | CFI_REL_OFFSET ebx, 0;\ | |
1da177e4 LT |
136 | movl $(__USER_DS), %edx; \ |
137 | movl %edx, %ds; \ | |
f95d47ca | 138 | movl %edx, %es; \ |
7c3576d2 | 139 | movl $(__KERNEL_PERCPU), %edx; \ |
464d1a78 | 140 | movl %edx, %fs |
1da177e4 LT |
141 | |
142 | #define RESTORE_INT_REGS \ | |
143 | popl %ebx; \ | |
fe7cacc1 JB |
144 | CFI_ADJUST_CFA_OFFSET -4;\ |
145 | CFI_RESTORE ebx;\ | |
1da177e4 | 146 | popl %ecx; \ |
fe7cacc1 JB |
147 | CFI_ADJUST_CFA_OFFSET -4;\ |
148 | CFI_RESTORE ecx;\ | |
1da177e4 | 149 | popl %edx; \ |
fe7cacc1 JB |
150 | CFI_ADJUST_CFA_OFFSET -4;\ |
151 | CFI_RESTORE edx;\ | |
1da177e4 | 152 | popl %esi; \ |
fe7cacc1 JB |
153 | CFI_ADJUST_CFA_OFFSET -4;\ |
154 | CFI_RESTORE esi;\ | |
1da177e4 | 155 | popl %edi; \ |
fe7cacc1 JB |
156 | CFI_ADJUST_CFA_OFFSET -4;\ |
157 | CFI_RESTORE edi;\ | |
1da177e4 | 158 | popl %ebp; \ |
fe7cacc1 JB |
159 | CFI_ADJUST_CFA_OFFSET -4;\ |
160 | CFI_RESTORE ebp;\ | |
161 | popl %eax; \ | |
162 | CFI_ADJUST_CFA_OFFSET -4;\ | |
163 | CFI_RESTORE eax | |
1da177e4 LT |
164 | |
165 | #define RESTORE_REGS \ | |
166 | RESTORE_INT_REGS; \ | |
167 | 1: popl %ds; \ | |
fe7cacc1 JB |
168 | CFI_ADJUST_CFA_OFFSET -4;\ |
169 | /*CFI_RESTORE ds;*/\ | |
1da177e4 | 170 | 2: popl %es; \ |
fe7cacc1 JB |
171 | CFI_ADJUST_CFA_OFFSET -4;\ |
172 | /*CFI_RESTORE es;*/\ | |
464d1a78 | 173 | 3: popl %fs; \ |
f95d47ca | 174 | CFI_ADJUST_CFA_OFFSET -4;\ |
464d1a78 | 175 | /*CFI_RESTORE fs;*/\ |
f95d47ca | 176 | .pushsection .fixup,"ax"; \ |
1da177e4 | 177 | 4: movl $0,(%esp); \ |
f95d47ca JF |
178 | jmp 1b; \ |
179 | 5: movl $0,(%esp); \ | |
1da177e4 | 180 | jmp 2b; \ |
f95d47ca JF |
181 | 6: movl $0,(%esp); \ |
182 | jmp 3b; \ | |
1da177e4 LT |
183 | .section __ex_table,"a";\ |
184 | .align 4; \ | |
f95d47ca JF |
185 | .long 1b,4b; \ |
186 | .long 2b,5b; \ | |
187 | .long 3b,6b; \ | |
188 | .popsection | |
1da177e4 | 189 | |
fe7cacc1 JB |
190 | #define RING0_INT_FRAME \ |
191 | CFI_STARTPROC simple;\ | |
adf14236 | 192 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
193 | CFI_DEF_CFA esp, 3*4;\ |
194 | /*CFI_OFFSET cs, -2*4;*/\ | |
195 | CFI_OFFSET eip, -3*4 | |
196 | ||
197 | #define RING0_EC_FRAME \ | |
198 | CFI_STARTPROC simple;\ | |
adf14236 | 199 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
200 | CFI_DEF_CFA esp, 4*4;\ |
201 | /*CFI_OFFSET cs, -2*4;*/\ | |
202 | CFI_OFFSET eip, -3*4 | |
203 | ||
204 | #define RING0_PTREGS_FRAME \ | |
205 | CFI_STARTPROC simple;\ | |
adf14236 | 206 | CFI_SIGNAL_FRAME;\ |
eb5b7b9d JF |
207 | CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\ |
208 | /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\ | |
209 | CFI_OFFSET eip, PT_EIP-PT_OLDESP;\ | |
210 | /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\ | |
211 | /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\ | |
212 | CFI_OFFSET eax, PT_EAX-PT_OLDESP;\ | |
213 | CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\ | |
214 | CFI_OFFSET edi, PT_EDI-PT_OLDESP;\ | |
215 | CFI_OFFSET esi, PT_ESI-PT_OLDESP;\ | |
216 | CFI_OFFSET edx, PT_EDX-PT_OLDESP;\ | |
217 | CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\ | |
218 | CFI_OFFSET ebx, PT_EBX-PT_OLDESP | |
1da177e4 LT |
219 | |
220 | ENTRY(ret_from_fork) | |
fe7cacc1 | 221 | CFI_STARTPROC |
1da177e4 | 222 | pushl %eax |
25d7dfda | 223 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
224 | call schedule_tail |
225 | GET_THREAD_INFO(%ebp) | |
226 | popl %eax | |
fe7cacc1 | 227 | CFI_ADJUST_CFA_OFFSET -4 |
47a5c6fa LT |
228 | pushl $0x0202 # Reset kernel eflags |
229 | CFI_ADJUST_CFA_OFFSET 4 | |
230 | popfl | |
231 | CFI_ADJUST_CFA_OFFSET -4 | |
1da177e4 | 232 | jmp syscall_exit |
fe7cacc1 | 233 | CFI_ENDPROC |
47a55cd7 | 234 | END(ret_from_fork) |
1da177e4 LT |
235 | |
236 | /* | |
237 | * Return to user mode is not as complex as all this looks, | |
238 | * but we want the default path for a system call return to | |
239 | * go as quickly as possible which is why some of this is | |
240 | * less clear than it otherwise should be. | |
241 | */ | |
242 | ||
243 | # userspace resumption stub bypassing syscall exit tracing | |
244 | ALIGN | |
fe7cacc1 | 245 | RING0_PTREGS_FRAME |
1da177e4 | 246 | ret_from_exception: |
139ec7c4 | 247 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
248 | ret_from_intr: |
249 | GET_THREAD_INFO(%ebp) | |
4031ff38 | 250 | check_userspace: |
eb5b7b9d JF |
251 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS |
252 | movb PT_CS(%esp), %al | |
ab68ed98 | 253 | andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax |
78be3706 RR |
254 | cmpl $USER_RPL, %eax |
255 | jb resume_kernel # not returning to v8086 or userspace | |
f95d47ca | 256 | |
1da177e4 | 257 | ENTRY(resume_userspace) |
c7e872e7 | 258 | LOCKDEP_SYS_EXIT |
139ec7c4 | 259 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
260 | # setting need_resched or sigpending |
261 | # between sampling and the iret | |
e32e58a9 | 262 | TRACE_IRQS_OFF |
1da177e4 LT |
263 | movl TI_flags(%ebp), %ecx |
264 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done on | |
265 | # int/exception return? | |
266 | jne work_pending | |
267 | jmp restore_all | |
47a55cd7 | 268 | END(ret_from_exception) |
1da177e4 LT |
269 | |
270 | #ifdef CONFIG_PREEMPT | |
271 | ENTRY(resume_kernel) | |
139ec7c4 | 272 | DISABLE_INTERRUPTS(CLBR_ANY) |
1da177e4 LT |
273 | cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ? |
274 | jnz restore_nocheck | |
275 | need_resched: | |
276 | movl TI_flags(%ebp), %ecx # need_resched set ? | |
277 | testb $_TIF_NEED_RESCHED, %cl | |
278 | jz restore_all | |
ab68ed98 | 279 | testl $X86_EFLAGS_IF,PT_EFLAGS(%esp) # interrupts off (exception path) ? |
1da177e4 LT |
280 | jz restore_all |
281 | call preempt_schedule_irq | |
282 | jmp need_resched | |
47a55cd7 | 283 | END(resume_kernel) |
1da177e4 | 284 | #endif |
fe7cacc1 | 285 | CFI_ENDPROC |
1da177e4 LT |
286 | |
287 | /* SYSENTER_RETURN points to after the "sysenter" instruction in | |
288 | the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */ | |
289 | ||
290 | # sysenter call handler stub | |
0aa97fb2 | 291 | ENTRY(ia32_sysenter_target) |
fe7cacc1 | 292 | CFI_STARTPROC simple |
adf14236 | 293 | CFI_SIGNAL_FRAME |
fe7cacc1 JB |
294 | CFI_DEF_CFA esp, 0 |
295 | CFI_REGISTER esp, ebp | |
faca6227 | 296 | movl TSS_sysenter_sp0(%esp),%esp |
1da177e4 | 297 | sysenter_past_esp: |
55f327fa | 298 | /* |
d93c870b JF |
299 | * Interrupts are disabled here, but we can't trace it until |
300 | * enough kernel state to call TRACE_IRQS_OFF can be called - but | |
301 | * we immediately enable interrupts at that point anyway. | |
55f327fa | 302 | */ |
1da177e4 | 303 | pushl $(__USER_DS) |
fe7cacc1 JB |
304 | CFI_ADJUST_CFA_OFFSET 4 |
305 | /*CFI_REL_OFFSET ss, 0*/ | |
1da177e4 | 306 | pushl %ebp |
fe7cacc1 JB |
307 | CFI_ADJUST_CFA_OFFSET 4 |
308 | CFI_REL_OFFSET esp, 0 | |
1da177e4 | 309 | pushfl |
d93c870b | 310 | orl $X86_EFLAGS_IF, (%esp) |
fe7cacc1 | 311 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 312 | pushl $(__USER_CS) |
fe7cacc1 JB |
313 | CFI_ADJUST_CFA_OFFSET 4 |
314 | /*CFI_REL_OFFSET cs, 0*/ | |
e6e5494c IM |
315 | /* |
316 | * Push current_thread_info()->sysenter_return to the stack. | |
317 | * A tiny bit of offset fixup is necessary - 4*4 means the 4 words | |
318 | * pushed above; +8 corresponds to copy_thread's esp0 setting. | |
319 | */ | |
320 | pushl (TI_sysenter_return-THREAD_SIZE+8+4*4)(%esp) | |
fe7cacc1 JB |
321 | CFI_ADJUST_CFA_OFFSET 4 |
322 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 323 | |
d93c870b JF |
324 | pushl %eax |
325 | CFI_ADJUST_CFA_OFFSET 4 | |
326 | SAVE_ALL | |
327 | ENABLE_INTERRUPTS(CLBR_NONE) | |
328 | ||
1da177e4 LT |
329 | /* |
330 | * Load the potential sixth argument from user stack. | |
331 | * Careful about security. | |
332 | */ | |
333 | cmpl $__PAGE_OFFSET-3,%ebp | |
334 | jae syscall_fault | |
335 | 1: movl (%ebp),%ebp | |
d93c870b | 336 | movl %ebp,PT_EBP(%esp) |
1da177e4 LT |
337 | .section __ex_table,"a" |
338 | .align 4 | |
339 | .long 1b,syscall_fault | |
340 | .previous | |
341 | ||
1da177e4 LT |
342 | GET_THREAD_INFO(%ebp) |
343 | ||
344 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ | |
d4d67150 | 345 | testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp) |
af0575bb RM |
346 | jnz sysenter_audit |
347 | sysenter_do_call: | |
1da177e4 LT |
348 | cmpl $(nr_syscalls), %eax |
349 | jae syscall_badsys | |
350 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 351 | movl %eax,PT_EAX(%esp) |
c7e872e7 | 352 | LOCKDEP_SYS_EXIT |
42c24fa2 | 353 | DISABLE_INTERRUPTS(CLBR_ANY) |
55f327fa | 354 | TRACE_IRQS_OFF |
1da177e4 LT |
355 | movl TI_flags(%ebp), %ecx |
356 | testw $_TIF_ALLWORK_MASK, %cx | |
af0575bb RM |
357 | jne sysexit_audit |
358 | sysenter_exit: | |
1da177e4 | 359 | /* if something modifies registers it must also disable sysexit */ |
eb5b7b9d JF |
360 | movl PT_EIP(%esp), %edx |
361 | movl PT_OLDESP(%esp), %ecx | |
1da177e4 | 362 | xorl %ebp,%ebp |
55f327fa | 363 | TRACE_IRQS_ON |
464d1a78 | 364 | 1: mov PT_FS(%esp), %fs |
d75cd22f | 365 | ENABLE_INTERRUPTS_SYSEXIT |
af0575bb RM |
366 | |
367 | #ifdef CONFIG_AUDITSYSCALL | |
368 | sysenter_audit: | |
369 | testw $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%ebp) | |
370 | jnz syscall_trace_entry | |
371 | addl $4,%esp | |
372 | CFI_ADJUST_CFA_OFFSET -4 | |
373 | /* %esi already in 8(%esp) 6th arg: 4th syscall arg */ | |
374 | /* %edx already in 4(%esp) 5th arg: 3rd syscall arg */ | |
375 | /* %ecx already in 0(%esp) 4th arg: 2nd syscall arg */ | |
376 | movl %ebx,%ecx /* 3rd arg: 1st syscall arg */ | |
377 | movl %eax,%edx /* 2nd arg: syscall number */ | |
378 | movl $AUDIT_ARCH_I386,%eax /* 1st arg: audit arch */ | |
379 | call audit_syscall_entry | |
380 | pushl %ebx | |
381 | CFI_ADJUST_CFA_OFFSET 4 | |
382 | movl PT_EAX(%esp),%eax /* reload syscall number */ | |
383 | jmp sysenter_do_call | |
384 | ||
385 | sysexit_audit: | |
386 | testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx | |
387 | jne syscall_exit_work | |
388 | TRACE_IRQS_ON | |
389 | ENABLE_INTERRUPTS(CLBR_ANY) | |
390 | movl %eax,%edx /* second arg, syscall return value */ | |
391 | cmpl $0,%eax /* is it < 0? */ | |
392 | setl %al /* 1 if so, 0 if not */ | |
393 | movzbl %al,%eax /* zero-extend that */ | |
394 | inc %eax /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */ | |
395 | call audit_syscall_exit | |
396 | DISABLE_INTERRUPTS(CLBR_ANY) | |
397 | TRACE_IRQS_OFF | |
398 | movl TI_flags(%ebp), %ecx | |
399 | testw $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %cx | |
400 | jne syscall_exit_work | |
401 | movl PT_EAX(%esp),%eax /* reload syscall return value */ | |
402 | jmp sysenter_exit | |
403 | #endif | |
404 | ||
fe7cacc1 | 405 | CFI_ENDPROC |
f95d47ca | 406 | .pushsection .fixup,"ax" |
464d1a78 | 407 | 2: movl $0,PT_FS(%esp) |
f95d47ca JF |
408 | jmp 1b |
409 | .section __ex_table,"a" | |
410 | .align 4 | |
411 | .long 1b,2b | |
412 | .popsection | |
0aa97fb2 | 413 | ENDPROC(ia32_sysenter_target) |
1da177e4 LT |
414 | |
415 | # system call handler stub | |
416 | ENTRY(system_call) | |
fe7cacc1 | 417 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 418 | pushl %eax # save orig_eax |
fe7cacc1 | 419 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
420 | SAVE_ALL |
421 | GET_THREAD_INFO(%ebp) | |
ed75e8d5 | 422 | # system call tracing in operation / emulation |
1da177e4 | 423 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ |
d4d67150 | 424 | testw $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%ebp) |
1da177e4 LT |
425 | jnz syscall_trace_entry |
426 | cmpl $(nr_syscalls), %eax | |
427 | jae syscall_badsys | |
428 | syscall_call: | |
429 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 430 | movl %eax,PT_EAX(%esp) # store the return value |
1da177e4 | 431 | syscall_exit: |
c7e872e7 | 432 | LOCKDEP_SYS_EXIT |
139ec7c4 | 433 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
434 | # setting need_resched or sigpending |
435 | # between sampling and the iret | |
55f327fa | 436 | TRACE_IRQS_OFF |
1da177e4 LT |
437 | movl TI_flags(%ebp), %ecx |
438 | testw $_TIF_ALLWORK_MASK, %cx # current->work | |
439 | jne syscall_exit_work | |
440 | ||
441 | restore_all: | |
eb5b7b9d JF |
442 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS |
443 | # Warning: PT_OLDSS(%esp) contains the wrong/random values if we | |
5df24082 SS |
444 | # are returning to the kernel. |
445 | # See comments in process.c:copy_thread() for details. | |
eb5b7b9d JF |
446 | movb PT_OLDSS(%esp), %ah |
447 | movb PT_CS(%esp), %al | |
ab68ed98 | 448 | andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax |
78be3706 | 449 | cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax |
fe7cacc1 | 450 | CFI_REMEMBER_STATE |
1da177e4 LT |
451 | je ldt_ss # returning to user-space with LDT SS |
452 | restore_nocheck: | |
55f327fa IM |
453 | TRACE_IRQS_IRET |
454 | restore_nocheck_notrace: | |
1da177e4 | 455 | RESTORE_REGS |
f95d47ca | 456 | addl $4, %esp # skip orig_eax/error_code |
fe7cacc1 | 457 | CFI_ADJUST_CFA_OFFSET -4 |
f7f3d791 | 458 | irq_return: |
3701d863 | 459 | INTERRUPT_RETURN |
1da177e4 | 460 | .section .fixup,"ax" |
90e9f536 | 461 | ENTRY(iret_exc) |
a879cbbb LT |
462 | pushl $0 # no error code |
463 | pushl $do_iret_error | |
464 | jmp error_code | |
1da177e4 LT |
465 | .previous |
466 | .section __ex_table,"a" | |
467 | .align 4 | |
3701d863 | 468 | .long irq_return,iret_exc |
1da177e4 LT |
469 | .previous |
470 | ||
fe7cacc1 | 471 | CFI_RESTORE_STATE |
1da177e4 | 472 | ldt_ss: |
eb5b7b9d | 473 | larl PT_OLDSS(%esp), %eax |
1da177e4 LT |
474 | jnz restore_nocheck |
475 | testl $0x00400000, %eax # returning to 32bit stack? | |
476 | jnz restore_nocheck # allright, normal return | |
d3561b7f RR |
477 | |
478 | #ifdef CONFIG_PARAVIRT | |
479 | /* | |
480 | * The kernel can't run on a non-flat stack if paravirt mode | |
481 | * is active. Rather than try to fixup the high bits of | |
482 | * ESP, bypass this code entirely. This may break DOSemu | |
483 | * and/or Wine support in a paravirt VM, although the option | |
484 | * is still available to implement the setting of the high | |
485 | * 16-bits in the INTERRUPT_RETURN paravirt-op. | |
486 | */ | |
93b1eab3 | 487 | cmpl $0, pv_info+PARAVIRT_enabled |
d3561b7f RR |
488 | jne restore_nocheck |
489 | #endif | |
490 | ||
1da177e4 LT |
491 | /* If returning to userspace with 16bit stack, |
492 | * try to fix the higher word of ESP, as the CPU | |
493 | * won't restore it. | |
494 | * This is an "official" bug of all the x86-compatible | |
495 | * CPUs, which we can try to work around to make | |
496 | * dosemu and wine happy. */ | |
eb5b7b9d | 497 | movl PT_OLDESP(%esp), %eax |
be44d2aa SS |
498 | movl %esp, %edx |
499 | call patch_espfix_desc | |
500 | pushl $__ESPFIX_SS | |
501 | CFI_ADJUST_CFA_OFFSET 4 | |
502 | pushl %eax | |
503 | CFI_ADJUST_CFA_OFFSET 4 | |
139ec7c4 | 504 | DISABLE_INTERRUPTS(CLBR_EAX) |
55f327fa | 505 | TRACE_IRQS_OFF |
be44d2aa SS |
506 | lss (%esp), %esp |
507 | CFI_ADJUST_CFA_OFFSET -8 | |
508 | jmp restore_nocheck | |
fe7cacc1 | 509 | CFI_ENDPROC |
47a55cd7 | 510 | ENDPROC(system_call) |
1da177e4 LT |
511 | |
512 | # perform work that needs to be done immediately before resumption | |
513 | ALIGN | |
fe7cacc1 | 514 | RING0_PTREGS_FRAME # can't unwind into user space anyway |
1da177e4 LT |
515 | work_pending: |
516 | testb $_TIF_NEED_RESCHED, %cl | |
517 | jz work_notifysig | |
518 | work_resched: | |
519 | call schedule | |
c7e872e7 | 520 | LOCKDEP_SYS_EXIT |
139ec7c4 | 521 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
522 | # setting need_resched or sigpending |
523 | # between sampling and the iret | |
55f327fa | 524 | TRACE_IRQS_OFF |
1da177e4 LT |
525 | movl TI_flags(%ebp), %ecx |
526 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done other | |
527 | # than syscall tracing? | |
528 | jz restore_all | |
529 | testb $_TIF_NEED_RESCHED, %cl | |
530 | jnz work_resched | |
531 | ||
532 | work_notifysig: # deal with pending signals and | |
533 | # notify-resume requests | |
74b47a78 | 534 | #ifdef CONFIG_VM86 |
ab68ed98 | 535 | testl $X86_EFLAGS_VM, PT_EFLAGS(%esp) |
1da177e4 LT |
536 | movl %esp, %eax |
537 | jne work_notifysig_v86 # returning to kernel-space or | |
538 | # vm86-space | |
539 | xorl %edx, %edx | |
540 | call do_notify_resume | |
4031ff38 | 541 | jmp resume_userspace_sig |
1da177e4 LT |
542 | |
543 | ALIGN | |
544 | work_notifysig_v86: | |
545 | pushl %ecx # save ti_flags for do_notify_resume | |
fe7cacc1 | 546 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
547 | call save_v86_state # %eax contains pt_regs pointer |
548 | popl %ecx | |
fe7cacc1 | 549 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 550 | movl %eax, %esp |
74b47a78 JK |
551 | #else |
552 | movl %esp, %eax | |
553 | #endif | |
1da177e4 LT |
554 | xorl %edx, %edx |
555 | call do_notify_resume | |
4031ff38 | 556 | jmp resume_userspace_sig |
47a55cd7 | 557 | END(work_pending) |
1da177e4 LT |
558 | |
559 | # perform syscall exit tracing | |
560 | ALIGN | |
561 | syscall_trace_entry: | |
eb5b7b9d | 562 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 | 563 | movl %esp, %eax |
d4d67150 RM |
564 | call syscall_trace_enter |
565 | /* What it returned is what we'll actually use. */ | |
1da177e4 LT |
566 | cmpl $(nr_syscalls), %eax |
567 | jnae syscall_call | |
568 | jmp syscall_exit | |
47a55cd7 | 569 | END(syscall_trace_entry) |
1da177e4 LT |
570 | |
571 | # perform syscall exit tracing | |
572 | ALIGN | |
573 | syscall_exit_work: | |
d4d67150 | 574 | testb $_TIF_WORK_SYSCALL_EXIT, %cl |
1da177e4 | 575 | jz work_pending |
55f327fa | 576 | TRACE_IRQS_ON |
d4d67150 | 577 | ENABLE_INTERRUPTS(CLBR_ANY) # could let syscall_trace_leave() call |
1da177e4 LT |
578 | # schedule() instead |
579 | movl %esp, %eax | |
d4d67150 | 580 | call syscall_trace_leave |
1da177e4 | 581 | jmp resume_userspace |
47a55cd7 | 582 | END(syscall_exit_work) |
fe7cacc1 | 583 | CFI_ENDPROC |
1da177e4 | 584 | |
fe7cacc1 | 585 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 586 | syscall_fault: |
1da177e4 | 587 | GET_THREAD_INFO(%ebp) |
eb5b7b9d | 588 | movl $-EFAULT,PT_EAX(%esp) |
1da177e4 | 589 | jmp resume_userspace |
47a55cd7 | 590 | END(syscall_fault) |
1da177e4 | 591 | |
1da177e4 | 592 | syscall_badsys: |
eb5b7b9d | 593 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 | 594 | jmp resume_userspace |
47a55cd7 | 595 | END(syscall_badsys) |
fe7cacc1 | 596 | CFI_ENDPROC |
1da177e4 LT |
597 | |
598 | #define FIXUP_ESPFIX_STACK \ | |
be44d2aa | 599 | /* since we are on a wrong stack, we cant make it a C code :( */ \ |
7a61d35d | 600 | PER_CPU(gdt_page, %ebx); \ |
be44d2aa SS |
601 | GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \ |
602 | addl %esp, %eax; \ | |
603 | pushl $__KERNEL_DS; \ | |
604 | CFI_ADJUST_CFA_OFFSET 4; \ | |
1da177e4 | 605 | pushl %eax; \ |
fe7cacc1 | 606 | CFI_ADJUST_CFA_OFFSET 4; \ |
be44d2aa SS |
607 | lss (%esp), %esp; \ |
608 | CFI_ADJUST_CFA_OFFSET -8; | |
609 | #define UNWIND_ESPFIX_STACK \ | |
1da177e4 | 610 | movl %ss, %eax; \ |
be44d2aa | 611 | /* see if on espfix stack */ \ |
1da177e4 | 612 | cmpw $__ESPFIX_SS, %ax; \ |
be44d2aa SS |
613 | jne 27f; \ |
614 | movl $__KERNEL_DS, %eax; \ | |
fe7cacc1 JB |
615 | movl %eax, %ds; \ |
616 | movl %eax, %es; \ | |
be44d2aa | 617 | /* switch to normal stack */ \ |
fe7cacc1 | 618 | FIXUP_ESPFIX_STACK; \ |
be44d2aa | 619 | 27:; |
1da177e4 LT |
620 | |
621 | /* | |
622 | * Build the entry stubs and pointer table with | |
623 | * some assembler magic. | |
624 | */ | |
3e7622f9 | 625 | .section .rodata,"a" |
1da177e4 LT |
626 | ENTRY(interrupt) |
627 | .text | |
628 | ||
1da177e4 | 629 | ENTRY(irq_entries_start) |
fe7cacc1 | 630 | RING0_INT_FRAME |
47a55cd7 | 631 | vector=0 |
497c9a19 | 632 | .rept NR_VECTORS |
1da177e4 | 633 | ALIGN |
fe7cacc1 JB |
634 | .if vector |
635 | CFI_ADJUST_CFA_OFFSET -4 | |
636 | .endif | |
19eadf98 | 637 | 1: pushl $~(vector) |
fe7cacc1 | 638 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 639 | jmp common_interrupt |
47a55cd7 | 640 | .previous |
1da177e4 | 641 | .long 1b |
47a55cd7 | 642 | .text |
1da177e4 LT |
643 | vector=vector+1 |
644 | .endr | |
47a55cd7 JB |
645 | END(irq_entries_start) |
646 | ||
647 | .previous | |
648 | END(interrupt) | |
649 | .previous | |
1da177e4 | 650 | |
55f327fa IM |
651 | /* |
652 | * the CPU automatically disables interrupts when executing an IRQ vector, | |
653 | * so IRQ-flags tracing has to follow that: | |
654 | */ | |
1da177e4 LT |
655 | ALIGN |
656 | common_interrupt: | |
657 | SAVE_ALL | |
55f327fa | 658 | TRACE_IRQS_OFF |
1da177e4 LT |
659 | movl %esp,%eax |
660 | call do_IRQ | |
661 | jmp ret_from_intr | |
47a55cd7 | 662 | ENDPROC(common_interrupt) |
fe7cacc1 | 663 | CFI_ENDPROC |
1da177e4 LT |
664 | |
665 | #define BUILD_INTERRUPT(name, nr) \ | |
666 | ENTRY(name) \ | |
fe7cacc1 | 667 | RING0_INT_FRAME; \ |
19eadf98 | 668 | pushl $~(nr); \ |
fe7cacc1 JB |
669 | CFI_ADJUST_CFA_OFFSET 4; \ |
670 | SAVE_ALL; \ | |
55f327fa | 671 | TRACE_IRQS_OFF \ |
1da177e4 | 672 | movl %esp,%eax; \ |
f76c3923 | 673 | call smp_##name; \ |
55f327fa | 674 | jmp ret_from_intr; \ |
47a55cd7 JB |
675 | CFI_ENDPROC; \ |
676 | ENDPROC(name) | |
1da177e4 LT |
677 | |
678 | /* The include is where all of the SMP etc. interrupts come from */ | |
679 | #include "entry_arch.h" | |
680 | ||
d28c4393 P |
681 | KPROBE_ENTRY(page_fault) |
682 | RING0_EC_FRAME | |
683 | pushl $do_page_fault | |
fe7cacc1 | 684 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
685 | ALIGN |
686 | error_code: | |
464d1a78 | 687 | /* the function address is in %fs's slot on the stack */ |
f95d47ca JF |
688 | pushl %es |
689 | CFI_ADJUST_CFA_OFFSET 4 | |
690 | /*CFI_REL_OFFSET es, 0*/ | |
1da177e4 | 691 | pushl %ds |
fe7cacc1 JB |
692 | CFI_ADJUST_CFA_OFFSET 4 |
693 | /*CFI_REL_OFFSET ds, 0*/ | |
1da177e4 | 694 | pushl %eax |
fe7cacc1 JB |
695 | CFI_ADJUST_CFA_OFFSET 4 |
696 | CFI_REL_OFFSET eax, 0 | |
1da177e4 | 697 | pushl %ebp |
fe7cacc1 JB |
698 | CFI_ADJUST_CFA_OFFSET 4 |
699 | CFI_REL_OFFSET ebp, 0 | |
1da177e4 | 700 | pushl %edi |
fe7cacc1 JB |
701 | CFI_ADJUST_CFA_OFFSET 4 |
702 | CFI_REL_OFFSET edi, 0 | |
1da177e4 | 703 | pushl %esi |
fe7cacc1 JB |
704 | CFI_ADJUST_CFA_OFFSET 4 |
705 | CFI_REL_OFFSET esi, 0 | |
1da177e4 | 706 | pushl %edx |
fe7cacc1 JB |
707 | CFI_ADJUST_CFA_OFFSET 4 |
708 | CFI_REL_OFFSET edx, 0 | |
1da177e4 | 709 | pushl %ecx |
fe7cacc1 JB |
710 | CFI_ADJUST_CFA_OFFSET 4 |
711 | CFI_REL_OFFSET ecx, 0 | |
1da177e4 | 712 | pushl %ebx |
fe7cacc1 JB |
713 | CFI_ADJUST_CFA_OFFSET 4 |
714 | CFI_REL_OFFSET ebx, 0 | |
1da177e4 | 715 | cld |
464d1a78 | 716 | pushl %fs |
fe7cacc1 | 717 | CFI_ADJUST_CFA_OFFSET 4 |
464d1a78 | 718 | /*CFI_REL_OFFSET fs, 0*/ |
7c3576d2 | 719 | movl $(__KERNEL_PERCPU), %ecx |
464d1a78 | 720 | movl %ecx, %fs |
1da177e4 LT |
721 | UNWIND_ESPFIX_STACK |
722 | popl %ecx | |
fe7cacc1 JB |
723 | CFI_ADJUST_CFA_OFFSET -4 |
724 | /*CFI_REGISTER es, ecx*/ | |
464d1a78 | 725 | movl PT_FS(%esp), %edi # get the function address |
eb5b7b9d | 726 | movl PT_ORIG_EAX(%esp), %edx # get the error code |
f95d47ca | 727 | movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart |
464d1a78 JF |
728 | mov %ecx, PT_FS(%esp) |
729 | /*CFI_REL_OFFSET fs, ES*/ | |
1da177e4 LT |
730 | movl $(__USER_DS), %ecx |
731 | movl %ecx, %ds | |
732 | movl %ecx, %es | |
85cea51d | 733 | TRACE_IRQS_OFF |
1da177e4 LT |
734 | movl %esp,%eax # pt_regs pointer |
735 | call *%edi | |
736 | jmp ret_from_exception | |
fe7cacc1 | 737 | CFI_ENDPROC |
d28c4393 | 738 | KPROBE_END(page_fault) |
1da177e4 LT |
739 | |
740 | ENTRY(coprocessor_error) | |
fe7cacc1 | 741 | RING0_INT_FRAME |
1da177e4 | 742 | pushl $0 |
fe7cacc1 | 743 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 744 | pushl $do_coprocessor_error |
fe7cacc1 | 745 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 746 | jmp error_code |
fe7cacc1 | 747 | CFI_ENDPROC |
47a55cd7 | 748 | END(coprocessor_error) |
1da177e4 LT |
749 | |
750 | ENTRY(simd_coprocessor_error) | |
fe7cacc1 | 751 | RING0_INT_FRAME |
1da177e4 | 752 | pushl $0 |
fe7cacc1 | 753 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 754 | pushl $do_simd_coprocessor_error |
fe7cacc1 | 755 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 756 | jmp error_code |
fe7cacc1 | 757 | CFI_ENDPROC |
47a55cd7 | 758 | END(simd_coprocessor_error) |
1da177e4 LT |
759 | |
760 | ENTRY(device_not_available) | |
fe7cacc1 | 761 | RING0_INT_FRAME |
1da177e4 | 762 | pushl $-1 # mark this as an int |
fe7cacc1 | 763 | CFI_ADJUST_CFA_OFFSET 4 |
7643e9b9 | 764 | pushl $do_device_not_available |
fe7cacc1 | 765 | CFI_ADJUST_CFA_OFFSET 4 |
7643e9b9 | 766 | jmp error_code |
fe7cacc1 | 767 | CFI_ENDPROC |
47a55cd7 | 768 | END(device_not_available) |
1da177e4 LT |
769 | |
770 | /* | |
771 | * Debug traps and NMI can happen at the one SYSENTER instruction | |
772 | * that sets up the real kernel stack. Check here, since we can't | |
773 | * allow the wrong stack to be used. | |
774 | * | |
faca6227 | 775 | * "TSS_sysenter_sp0+12" is because the NMI/debug handler will have |
1da177e4 LT |
776 | * already pushed 3 words if it hits on the sysenter instruction: |
777 | * eflags, cs and eip. | |
778 | * | |
779 | * We just load the right stack, and push the three (known) values | |
780 | * by hand onto the new stack - while updating the return eip past | |
781 | * the instruction that would have done it for sysenter. | |
782 | */ | |
783 | #define FIX_STACK(offset, ok, label) \ | |
784 | cmpw $__KERNEL_CS,4(%esp); \ | |
785 | jne ok; \ | |
786 | label: \ | |
faca6227 | 787 | movl TSS_sysenter_sp0+offset(%esp),%esp; \ |
a549b86d CE |
788 | CFI_DEF_CFA esp, 0; \ |
789 | CFI_UNDEFINED eip; \ | |
1da177e4 | 790 | pushfl; \ |
a549b86d | 791 | CFI_ADJUST_CFA_OFFSET 4; \ |
1da177e4 | 792 | pushl $__KERNEL_CS; \ |
a549b86d CE |
793 | CFI_ADJUST_CFA_OFFSET 4; \ |
794 | pushl $sysenter_past_esp; \ | |
795 | CFI_ADJUST_CFA_OFFSET 4; \ | |
796 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 797 | |
3d97ae5b | 798 | KPROBE_ENTRY(debug) |
fe7cacc1 | 799 | RING0_INT_FRAME |
0aa97fb2 | 800 | cmpl $ia32_sysenter_target,(%esp) |
1da177e4 LT |
801 | jne debug_stack_correct |
802 | FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn) | |
803 | debug_stack_correct: | |
804 | pushl $-1 # mark this as an int | |
fe7cacc1 | 805 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 806 | SAVE_ALL |
43024a8a | 807 | TRACE_IRQS_OFF |
1da177e4 LT |
808 | xorl %edx,%edx # error code 0 |
809 | movl %esp,%eax # pt_regs pointer | |
810 | call do_debug | |
1da177e4 | 811 | jmp ret_from_exception |
fe7cacc1 | 812 | CFI_ENDPROC |
d28c4393 P |
813 | KPROBE_END(debug) |
814 | ||
1da177e4 LT |
815 | /* |
816 | * NMI is doubly nasty. It can happen _while_ we're handling | |
817 | * a debug fault, and the debug fault hasn't yet been able to | |
818 | * clear up the stack. So we first check whether we got an | |
819 | * NMI on the sysenter entry path, but after that we need to | |
820 | * check whether we got an NMI on the debug path where the debug | |
821 | * fault happened on the sysenter path. | |
822 | */ |