Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/i386/entry.S | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * entry.S contains the system-call and fault low-level handling routines. | |
9 | * This also contains the timer-interrupt handler, as well as all interrupts | |
10 | * and faults that can result in a task-switch. | |
11 | * | |
12 | * NOTE: This code handles signal-recognition, which happens every time | |
13 | * after a timer-interrupt and after each system call. | |
14 | * | |
15 | * I changed all the .align's to 4 (16 byte alignment), as that's faster | |
16 | * on a 486. | |
17 | * | |
18 | * Stack layout in 'ret_from_system_call': | |
19 | * ptrace needs to have all regs on the stack. | |
20 | * if the order here is changed, it needs to be | |
21 | * updated in fork.c:copy_process, signal.c:do_signal, | |
22 | * ptrace.c and ptrace.h | |
23 | * | |
24 | * 0(%esp) - %ebx | |
25 | * 4(%esp) - %ecx | |
26 | * 8(%esp) - %edx | |
27 | * C(%esp) - %esi | |
28 | * 10(%esp) - %edi | |
29 | * 14(%esp) - %ebp | |
30 | * 18(%esp) - %eax | |
31 | * 1C(%esp) - %ds | |
32 | * 20(%esp) - %es | |
f95d47ca JF |
33 | * 24(%esp) - %gs |
34 | * 28(%esp) - orig_eax | |
35 | * 2C(%esp) - %eip | |
36 | * 30(%esp) - %cs | |
37 | * 34(%esp) - %eflags | |
38 | * 38(%esp) - %oldesp | |
39 | * 3C(%esp) - %oldss | |
1da177e4 LT |
40 | * |
41 | * "current" is in register %ebx during any slow entries. | |
42 | */ | |
43 | ||
1da177e4 LT |
44 | #include <linux/linkage.h> |
45 | #include <asm/thread_info.h> | |
55f327fa | 46 | #include <asm/irqflags.h> |
1da177e4 LT |
47 | #include <asm/errno.h> |
48 | #include <asm/segment.h> | |
49 | #include <asm/smp.h> | |
50 | #include <asm/page.h> | |
51 | #include <asm/desc.h> | |
be44d2aa | 52 | #include <asm/percpu.h> |
fe7cacc1 | 53 | #include <asm/dwarf2.h> |
1da177e4 LT |
54 | #include "irq_vectors.h" |
55 | ||
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") | |
61 | * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit"). | |
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 LT |
71 | CF_MASK = 0x00000001 |
72 | TF_MASK = 0x00000100 | |
73 | IF_MASK = 0x00000200 | |
74 | DF_MASK = 0x00000400 | |
75 | NT_MASK = 0x00004000 | |
76 | VM_MASK = 0x00020000 | |
77 | ||
78 | #ifdef CONFIG_PREEMPT | |
139ec7c4 | 79 | #define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF |
1da177e4 | 80 | #else |
139ec7c4 | 81 | #define preempt_stop(clobbers) |
1da177e4 LT |
82 | #define resume_kernel restore_nocheck |
83 | #endif | |
84 | ||
55f327fa IM |
85 | .macro TRACE_IRQS_IRET |
86 | #ifdef CONFIG_TRACE_IRQFLAGS | |
eb5b7b9d | 87 | testl $IF_MASK,PT_EFLAGS(%esp) # interrupts off? |
55f327fa IM |
88 | jz 1f |
89 | TRACE_IRQS_ON | |
90 | 1: | |
91 | #endif | |
92 | .endm | |
93 | ||
4031ff38 AG |
94 | #ifdef CONFIG_VM86 |
95 | #define resume_userspace_sig check_userspace | |
96 | #else | |
97 | #define resume_userspace_sig resume_userspace | |
98 | #endif | |
99 | ||
1da177e4 LT |
100 | #define SAVE_ALL \ |
101 | cld; \ | |
f95d47ca JF |
102 | pushl %gs; \ |
103 | CFI_ADJUST_CFA_OFFSET 4;\ | |
104 | /*CFI_REL_OFFSET gs, 0;*/\ | |
1da177e4 | 105 | pushl %es; \ |
fe7cacc1 JB |
106 | CFI_ADJUST_CFA_OFFSET 4;\ |
107 | /*CFI_REL_OFFSET es, 0;*/\ | |
1da177e4 | 108 | pushl %ds; \ |
fe7cacc1 JB |
109 | CFI_ADJUST_CFA_OFFSET 4;\ |
110 | /*CFI_REL_OFFSET ds, 0;*/\ | |
1da177e4 | 111 | pushl %eax; \ |
fe7cacc1 JB |
112 | CFI_ADJUST_CFA_OFFSET 4;\ |
113 | CFI_REL_OFFSET eax, 0;\ | |
1da177e4 | 114 | pushl %ebp; \ |
fe7cacc1 JB |
115 | CFI_ADJUST_CFA_OFFSET 4;\ |
116 | CFI_REL_OFFSET ebp, 0;\ | |
1da177e4 | 117 | pushl %edi; \ |
fe7cacc1 JB |
118 | CFI_ADJUST_CFA_OFFSET 4;\ |
119 | CFI_REL_OFFSET edi, 0;\ | |
1da177e4 | 120 | pushl %esi; \ |
fe7cacc1 JB |
121 | CFI_ADJUST_CFA_OFFSET 4;\ |
122 | CFI_REL_OFFSET esi, 0;\ | |
1da177e4 | 123 | pushl %edx; \ |
fe7cacc1 JB |
124 | CFI_ADJUST_CFA_OFFSET 4;\ |
125 | CFI_REL_OFFSET edx, 0;\ | |
1da177e4 | 126 | pushl %ecx; \ |
fe7cacc1 JB |
127 | CFI_ADJUST_CFA_OFFSET 4;\ |
128 | CFI_REL_OFFSET ecx, 0;\ | |
1da177e4 | 129 | pushl %ebx; \ |
fe7cacc1 JB |
130 | CFI_ADJUST_CFA_OFFSET 4;\ |
131 | CFI_REL_OFFSET ebx, 0;\ | |
1da177e4 LT |
132 | movl $(__USER_DS), %edx; \ |
133 | movl %edx, %ds; \ | |
f95d47ca JF |
134 | movl %edx, %es; \ |
135 | movl $(__KERNEL_PDA), %edx; \ | |
136 | movl %edx, %gs | |
1da177e4 LT |
137 | |
138 | #define RESTORE_INT_REGS \ | |
139 | popl %ebx; \ | |
fe7cacc1 JB |
140 | CFI_ADJUST_CFA_OFFSET -4;\ |
141 | CFI_RESTORE ebx;\ | |
1da177e4 | 142 | popl %ecx; \ |
fe7cacc1 JB |
143 | CFI_ADJUST_CFA_OFFSET -4;\ |
144 | CFI_RESTORE ecx;\ | |
1da177e4 | 145 | popl %edx; \ |
fe7cacc1 JB |
146 | CFI_ADJUST_CFA_OFFSET -4;\ |
147 | CFI_RESTORE edx;\ | |
1da177e4 | 148 | popl %esi; \ |
fe7cacc1 JB |
149 | CFI_ADJUST_CFA_OFFSET -4;\ |
150 | CFI_RESTORE esi;\ | |
1da177e4 | 151 | popl %edi; \ |
fe7cacc1 JB |
152 | CFI_ADJUST_CFA_OFFSET -4;\ |
153 | CFI_RESTORE edi;\ | |
1da177e4 | 154 | popl %ebp; \ |
fe7cacc1 JB |
155 | CFI_ADJUST_CFA_OFFSET -4;\ |
156 | CFI_RESTORE ebp;\ | |
157 | popl %eax; \ | |
158 | CFI_ADJUST_CFA_OFFSET -4;\ | |
159 | CFI_RESTORE eax | |
1da177e4 LT |
160 | |
161 | #define RESTORE_REGS \ | |
162 | RESTORE_INT_REGS; \ | |
163 | 1: popl %ds; \ | |
fe7cacc1 JB |
164 | CFI_ADJUST_CFA_OFFSET -4;\ |
165 | /*CFI_RESTORE ds;*/\ | |
1da177e4 | 166 | 2: popl %es; \ |
fe7cacc1 JB |
167 | CFI_ADJUST_CFA_OFFSET -4;\ |
168 | /*CFI_RESTORE es;*/\ | |
f95d47ca JF |
169 | 3: popl %gs; \ |
170 | CFI_ADJUST_CFA_OFFSET -4;\ | |
171 | /*CFI_RESTORE gs;*/\ | |
172 | .pushsection .fixup,"ax"; \ | |
1da177e4 | 173 | 4: movl $0,(%esp); \ |
f95d47ca JF |
174 | jmp 1b; \ |
175 | 5: movl $0,(%esp); \ | |
1da177e4 | 176 | jmp 2b; \ |
f95d47ca JF |
177 | 6: movl $0,(%esp); \ |
178 | jmp 3b; \ | |
1da177e4 LT |
179 | .section __ex_table,"a";\ |
180 | .align 4; \ | |
f95d47ca JF |
181 | .long 1b,4b; \ |
182 | .long 2b,5b; \ | |
183 | .long 3b,6b; \ | |
184 | .popsection | |
1da177e4 | 185 | |
fe7cacc1 JB |
186 | #define RING0_INT_FRAME \ |
187 | CFI_STARTPROC simple;\ | |
adf14236 | 188 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
189 | CFI_DEF_CFA esp, 3*4;\ |
190 | /*CFI_OFFSET cs, -2*4;*/\ | |
191 | CFI_OFFSET eip, -3*4 | |
192 | ||
193 | #define RING0_EC_FRAME \ | |
194 | CFI_STARTPROC simple;\ | |
adf14236 | 195 | CFI_SIGNAL_FRAME;\ |
fe7cacc1 JB |
196 | CFI_DEF_CFA esp, 4*4;\ |
197 | /*CFI_OFFSET cs, -2*4;*/\ | |
198 | CFI_OFFSET eip, -3*4 | |
199 | ||
200 | #define RING0_PTREGS_FRAME \ | |
201 | CFI_STARTPROC simple;\ | |
adf14236 | 202 | CFI_SIGNAL_FRAME;\ |
eb5b7b9d JF |
203 | CFI_DEF_CFA esp, PT_OLDESP-PT_EBX;\ |
204 | /*CFI_OFFSET cs, PT_CS-PT_OLDESP;*/\ | |
205 | CFI_OFFSET eip, PT_EIP-PT_OLDESP;\ | |
206 | /*CFI_OFFSET es, PT_ES-PT_OLDESP;*/\ | |
207 | /*CFI_OFFSET ds, PT_DS-PT_OLDESP;*/\ | |
208 | CFI_OFFSET eax, PT_EAX-PT_OLDESP;\ | |
209 | CFI_OFFSET ebp, PT_EBP-PT_OLDESP;\ | |
210 | CFI_OFFSET edi, PT_EDI-PT_OLDESP;\ | |
211 | CFI_OFFSET esi, PT_ESI-PT_OLDESP;\ | |
212 | CFI_OFFSET edx, PT_EDX-PT_OLDESP;\ | |
213 | CFI_OFFSET ecx, PT_ECX-PT_OLDESP;\ | |
214 | CFI_OFFSET ebx, PT_EBX-PT_OLDESP | |
1da177e4 LT |
215 | |
216 | ENTRY(ret_from_fork) | |
fe7cacc1 | 217 | CFI_STARTPROC |
1da177e4 | 218 | pushl %eax |
25d7dfda | 219 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
220 | call schedule_tail |
221 | GET_THREAD_INFO(%ebp) | |
222 | popl %eax | |
fe7cacc1 | 223 | CFI_ADJUST_CFA_OFFSET -4 |
47a5c6fa LT |
224 | pushl $0x0202 # Reset kernel eflags |
225 | CFI_ADJUST_CFA_OFFSET 4 | |
226 | popfl | |
227 | CFI_ADJUST_CFA_OFFSET -4 | |
1da177e4 | 228 | jmp syscall_exit |
fe7cacc1 | 229 | CFI_ENDPROC |
1da177e4 LT |
230 | |
231 | /* | |
232 | * Return to user mode is not as complex as all this looks, | |
233 | * but we want the default path for a system call return to | |
234 | * go as quickly as possible which is why some of this is | |
235 | * less clear than it otherwise should be. | |
236 | */ | |
237 | ||
238 | # userspace resumption stub bypassing syscall exit tracing | |
239 | ALIGN | |
fe7cacc1 | 240 | RING0_PTREGS_FRAME |
1da177e4 | 241 | ret_from_exception: |
139ec7c4 | 242 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
243 | ret_from_intr: |
244 | GET_THREAD_INFO(%ebp) | |
4031ff38 | 245 | check_userspace: |
eb5b7b9d JF |
246 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS |
247 | movb PT_CS(%esp), %al | |
78be3706 RR |
248 | andl $(VM_MASK | SEGMENT_RPL_MASK), %eax |
249 | cmpl $USER_RPL, %eax | |
250 | jb resume_kernel # not returning to v8086 or userspace | |
f95d47ca | 251 | |
1da177e4 | 252 | ENTRY(resume_userspace) |
139ec7c4 | 253 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
254 | # setting need_resched or sigpending |
255 | # between sampling and the iret | |
256 | movl TI_flags(%ebp), %ecx | |
257 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done on | |
258 | # int/exception return? | |
259 | jne work_pending | |
260 | jmp restore_all | |
261 | ||
262 | #ifdef CONFIG_PREEMPT | |
263 | ENTRY(resume_kernel) | |
139ec7c4 | 264 | DISABLE_INTERRUPTS(CLBR_ANY) |
1da177e4 LT |
265 | cmpl $0,TI_preempt_count(%ebp) # non-zero preempt_count ? |
266 | jnz restore_nocheck | |
267 | need_resched: | |
268 | movl TI_flags(%ebp), %ecx # need_resched set ? | |
269 | testb $_TIF_NEED_RESCHED, %cl | |
270 | jz restore_all | |
eb5b7b9d | 271 | testl $IF_MASK,PT_EFLAGS(%esp) # interrupts off (exception path) ? |
1da177e4 LT |
272 | jz restore_all |
273 | call preempt_schedule_irq | |
274 | jmp need_resched | |
275 | #endif | |
fe7cacc1 | 276 | CFI_ENDPROC |
1da177e4 LT |
277 | |
278 | /* SYSENTER_RETURN points to after the "sysenter" instruction in | |
279 | the vsyscall page. See vsyscall-sysentry.S, which defines the symbol. */ | |
280 | ||
281 | # sysenter call handler stub | |
282 | ENTRY(sysenter_entry) | |
fe7cacc1 | 283 | CFI_STARTPROC simple |
adf14236 | 284 | CFI_SIGNAL_FRAME |
fe7cacc1 JB |
285 | CFI_DEF_CFA esp, 0 |
286 | CFI_REGISTER esp, ebp | |
1da177e4 LT |
287 | movl TSS_sysenter_esp0(%esp),%esp |
288 | sysenter_past_esp: | |
55f327fa IM |
289 | /* |
290 | * No need to follow this irqs on/off section: the syscall | |
291 | * disabled irqs and here we enable it straight after entry: | |
292 | */ | |
139ec7c4 | 293 | ENABLE_INTERRUPTS(CLBR_NONE) |
1da177e4 | 294 | pushl $(__USER_DS) |
fe7cacc1 JB |
295 | CFI_ADJUST_CFA_OFFSET 4 |
296 | /*CFI_REL_OFFSET ss, 0*/ | |
1da177e4 | 297 | pushl %ebp |
fe7cacc1 JB |
298 | CFI_ADJUST_CFA_OFFSET 4 |
299 | CFI_REL_OFFSET esp, 0 | |
1da177e4 | 300 | pushfl |
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 LT |
313 | |
314 | /* | |
315 | * Load the potential sixth argument from user stack. | |
316 | * Careful about security. | |
317 | */ | |
318 | cmpl $__PAGE_OFFSET-3,%ebp | |
319 | jae syscall_fault | |
320 | 1: movl (%ebp),%ebp | |
321 | .section __ex_table,"a" | |
322 | .align 4 | |
323 | .long 1b,syscall_fault | |
324 | .previous | |
325 | ||
326 | pushl %eax | |
fe7cacc1 | 327 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
328 | SAVE_ALL |
329 | GET_THREAD_INFO(%ebp) | |
330 | ||
331 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ | |
ed75e8d5 | 332 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
333 | jnz syscall_trace_entry |
334 | cmpl $(nr_syscalls), %eax | |
335 | jae syscall_badsys | |
336 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 337 | movl %eax,PT_EAX(%esp) |
139ec7c4 | 338 | DISABLE_INTERRUPTS(CLBR_ECX|CLBR_EDX) |
55f327fa | 339 | TRACE_IRQS_OFF |
1da177e4 LT |
340 | movl TI_flags(%ebp), %ecx |
341 | testw $_TIF_ALLWORK_MASK, %cx | |
342 | jne syscall_exit_work | |
343 | /* if something modifies registers it must also disable sysexit */ | |
eb5b7b9d JF |
344 | movl PT_EIP(%esp), %edx |
345 | movl PT_OLDESP(%esp), %ecx | |
1da177e4 | 346 | xorl %ebp,%ebp |
55f327fa | 347 | TRACE_IRQS_ON |
f95d47ca | 348 | 1: mov PT_GS(%esp), %gs |
0da5db31 | 349 | ENABLE_INTERRUPTS_SYSEXIT |
fe7cacc1 | 350 | CFI_ENDPROC |
f95d47ca JF |
351 | .pushsection .fixup,"ax" |
352 | 2: movl $0,PT_GS(%esp) | |
353 | jmp 1b | |
354 | .section __ex_table,"a" | |
355 | .align 4 | |
356 | .long 1b,2b | |
357 | .popsection | |
1da177e4 LT |
358 | |
359 | # system call handler stub | |
360 | ENTRY(system_call) | |
fe7cacc1 | 361 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 | 362 | pushl %eax # save orig_eax |
fe7cacc1 | 363 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
364 | SAVE_ALL |
365 | GET_THREAD_INFO(%ebp) | |
eb5b7b9d | 366 | testl $TF_MASK,PT_EFLAGS(%esp) |
635cf99a CE |
367 | jz no_singlestep |
368 | orl $_TIF_SINGLESTEP,TI_flags(%ebp) | |
369 | no_singlestep: | |
ed75e8d5 | 370 | # system call tracing in operation / emulation |
1da177e4 | 371 | /* Note, _TIF_SECCOMP is bit number 8, and so it needs testw and not testb */ |
ed75e8d5 | 372 | testw $(_TIF_SYSCALL_EMU|_TIF_SYSCALL_TRACE|_TIF_SECCOMP|_TIF_SYSCALL_AUDIT),TI_flags(%ebp) |
1da177e4 LT |
373 | jnz syscall_trace_entry |
374 | cmpl $(nr_syscalls), %eax | |
375 | jae syscall_badsys | |
376 | syscall_call: | |
377 | call *sys_call_table(,%eax,4) | |
eb5b7b9d | 378 | movl %eax,PT_EAX(%esp) # store the return value |
1da177e4 | 379 | syscall_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 |
1da177e4 LT |
384 | movl TI_flags(%ebp), %ecx |
385 | testw $_TIF_ALLWORK_MASK, %cx # current->work | |
386 | jne syscall_exit_work | |
387 | ||
388 | restore_all: | |
eb5b7b9d JF |
389 | movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS |
390 | # Warning: PT_OLDSS(%esp) contains the wrong/random values if we | |
5df24082 SS |
391 | # are returning to the kernel. |
392 | # See comments in process.c:copy_thread() for details. | |
eb5b7b9d JF |
393 | movb PT_OLDSS(%esp), %ah |
394 | movb PT_CS(%esp), %al | |
78be3706 RR |
395 | andl $(VM_MASK | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax |
396 | cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax | |
fe7cacc1 | 397 | CFI_REMEMBER_STATE |
1da177e4 LT |
398 | je ldt_ss # returning to user-space with LDT SS |
399 | restore_nocheck: | |
55f327fa IM |
400 | TRACE_IRQS_IRET |
401 | restore_nocheck_notrace: | |
1da177e4 | 402 | RESTORE_REGS |
f95d47ca | 403 | addl $4, %esp # skip orig_eax/error_code |
fe7cacc1 | 404 | CFI_ADJUST_CFA_OFFSET -4 |
0da5db31 | 405 | 1: INTERRUPT_RETURN |
1da177e4 LT |
406 | .section .fixup,"ax" |
407 | iret_exc: | |
55f327fa | 408 | TRACE_IRQS_ON |
139ec7c4 | 409 | ENABLE_INTERRUPTS(CLBR_NONE) |
a879cbbb LT |
410 | pushl $0 # no error code |
411 | pushl $do_iret_error | |
412 | jmp error_code | |
1da177e4 LT |
413 | .previous |
414 | .section __ex_table,"a" | |
415 | .align 4 | |
416 | .long 1b,iret_exc | |
417 | .previous | |
418 | ||
fe7cacc1 | 419 | CFI_RESTORE_STATE |
1da177e4 | 420 | ldt_ss: |
eb5b7b9d | 421 | larl PT_OLDSS(%esp), %eax |
1da177e4 LT |
422 | jnz restore_nocheck |
423 | testl $0x00400000, %eax # returning to 32bit stack? | |
424 | jnz restore_nocheck # allright, normal return | |
d3561b7f RR |
425 | |
426 | #ifdef CONFIG_PARAVIRT | |
427 | /* | |
428 | * The kernel can't run on a non-flat stack if paravirt mode | |
429 | * is active. Rather than try to fixup the high bits of | |
430 | * ESP, bypass this code entirely. This may break DOSemu | |
431 | * and/or Wine support in a paravirt VM, although the option | |
432 | * is still available to implement the setting of the high | |
433 | * 16-bits in the INTERRUPT_RETURN paravirt-op. | |
434 | */ | |
435 | cmpl $0, paravirt_ops+PARAVIRT_enabled | |
436 | jne restore_nocheck | |
437 | #endif | |
438 | ||
1da177e4 LT |
439 | /* If returning to userspace with 16bit stack, |
440 | * try to fix the higher word of ESP, as the CPU | |
441 | * won't restore it. | |
442 | * This is an "official" bug of all the x86-compatible | |
443 | * CPUs, which we can try to work around to make | |
444 | * dosemu and wine happy. */ | |
eb5b7b9d | 445 | movl PT_OLDESP(%esp), %eax |
be44d2aa SS |
446 | movl %esp, %edx |
447 | call patch_espfix_desc | |
448 | pushl $__ESPFIX_SS | |
449 | CFI_ADJUST_CFA_OFFSET 4 | |
450 | pushl %eax | |
451 | CFI_ADJUST_CFA_OFFSET 4 | |
139ec7c4 | 452 | DISABLE_INTERRUPTS(CLBR_EAX) |
55f327fa | 453 | TRACE_IRQS_OFF |
be44d2aa SS |
454 | lss (%esp), %esp |
455 | CFI_ADJUST_CFA_OFFSET -8 | |
456 | jmp restore_nocheck | |
fe7cacc1 | 457 | CFI_ENDPROC |
1da177e4 LT |
458 | |
459 | # perform work that needs to be done immediately before resumption | |
460 | ALIGN | |
fe7cacc1 | 461 | RING0_PTREGS_FRAME # can't unwind into user space anyway |
1da177e4 LT |
462 | work_pending: |
463 | testb $_TIF_NEED_RESCHED, %cl | |
464 | jz work_notifysig | |
465 | work_resched: | |
466 | call schedule | |
139ec7c4 | 467 | DISABLE_INTERRUPTS(CLBR_ANY) # make sure we don't miss an interrupt |
1da177e4 LT |
468 | # setting need_resched or sigpending |
469 | # between sampling and the iret | |
55f327fa | 470 | TRACE_IRQS_OFF |
1da177e4 LT |
471 | movl TI_flags(%ebp), %ecx |
472 | andl $_TIF_WORK_MASK, %ecx # is there any work to be done other | |
473 | # than syscall tracing? | |
474 | jz restore_all | |
475 | testb $_TIF_NEED_RESCHED, %cl | |
476 | jnz work_resched | |
477 | ||
478 | work_notifysig: # deal with pending signals and | |
479 | # notify-resume requests | |
74b47a78 | 480 | #ifdef CONFIG_VM86 |
eb5b7b9d | 481 | testl $VM_MASK, PT_EFLAGS(%esp) |
1da177e4 LT |
482 | movl %esp, %eax |
483 | jne work_notifysig_v86 # returning to kernel-space or | |
484 | # vm86-space | |
485 | xorl %edx, %edx | |
486 | call do_notify_resume | |
4031ff38 | 487 | jmp resume_userspace_sig |
1da177e4 LT |
488 | |
489 | ALIGN | |
490 | work_notifysig_v86: | |
491 | pushl %ecx # save ti_flags for do_notify_resume | |
fe7cacc1 | 492 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
493 | call save_v86_state # %eax contains pt_regs pointer |
494 | popl %ecx | |
fe7cacc1 | 495 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 496 | movl %eax, %esp |
74b47a78 JK |
497 | #else |
498 | movl %esp, %eax | |
499 | #endif | |
1da177e4 LT |
500 | xorl %edx, %edx |
501 | call do_notify_resume | |
4031ff38 | 502 | jmp resume_userspace_sig |
1da177e4 LT |
503 | |
504 | # perform syscall exit tracing | |
505 | ALIGN | |
506 | syscall_trace_entry: | |
eb5b7b9d | 507 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 LT |
508 | movl %esp, %eax |
509 | xorl %edx,%edx | |
510 | call do_syscall_trace | |
ed75e8d5 | 511 | cmpl $0, %eax |
640aa46e | 512 | jne resume_userspace # ret != 0 -> running under PTRACE_SYSEMU, |
ed75e8d5 | 513 | # so must skip actual syscall |
eb5b7b9d | 514 | movl PT_ORIG_EAX(%esp), %eax |
1da177e4 LT |
515 | cmpl $(nr_syscalls), %eax |
516 | jnae syscall_call | |
517 | jmp syscall_exit | |
518 | ||
519 | # perform syscall exit tracing | |
520 | ALIGN | |
521 | syscall_exit_work: | |
522 | testb $(_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SINGLESTEP), %cl | |
523 | jz work_pending | |
55f327fa | 524 | TRACE_IRQS_ON |
139ec7c4 | 525 | ENABLE_INTERRUPTS(CLBR_ANY) # could let do_syscall_trace() call |
1da177e4 LT |
526 | # schedule() instead |
527 | movl %esp, %eax | |
528 | movl $1, %edx | |
529 | call do_syscall_trace | |
530 | jmp resume_userspace | |
fe7cacc1 | 531 | CFI_ENDPROC |
1da177e4 | 532 | |
fe7cacc1 | 533 | RING0_INT_FRAME # can't unwind into user space anyway |
1da177e4 LT |
534 | syscall_fault: |
535 | pushl %eax # save orig_eax | |
fe7cacc1 | 536 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
537 | SAVE_ALL |
538 | GET_THREAD_INFO(%ebp) | |
eb5b7b9d | 539 | movl $-EFAULT,PT_EAX(%esp) |
1da177e4 LT |
540 | jmp resume_userspace |
541 | ||
1da177e4 | 542 | syscall_badsys: |
eb5b7b9d | 543 | movl $-ENOSYS,PT_EAX(%esp) |
1da177e4 | 544 | jmp resume_userspace |
fe7cacc1 | 545 | CFI_ENDPROC |
1da177e4 LT |
546 | |
547 | #define FIXUP_ESPFIX_STACK \ | |
be44d2aa | 548 | /* since we are on a wrong stack, we cant make it a C code :( */ \ |
b2938f88 | 549 | movl %gs:PDA_cpu, %ebx; \ |
be44d2aa SS |
550 | PER_CPU(cpu_gdt_descr, %ebx); \ |
551 | movl GDS_address(%ebx), %ebx; \ | |
552 | GET_DESC_BASE(GDT_ENTRY_ESPFIX_SS, %ebx, %eax, %ax, %al, %ah); \ | |
553 | addl %esp, %eax; \ | |
554 | pushl $__KERNEL_DS; \ | |
555 | CFI_ADJUST_CFA_OFFSET 4; \ | |
1da177e4 | 556 | pushl %eax; \ |
fe7cacc1 | 557 | CFI_ADJUST_CFA_OFFSET 4; \ |
be44d2aa SS |
558 | lss (%esp), %esp; \ |
559 | CFI_ADJUST_CFA_OFFSET -8; | |
560 | #define UNWIND_ESPFIX_STACK \ | |
1da177e4 | 561 | movl %ss, %eax; \ |
be44d2aa | 562 | /* see if on espfix stack */ \ |
1da177e4 | 563 | cmpw $__ESPFIX_SS, %ax; \ |
be44d2aa SS |
564 | jne 27f; \ |
565 | movl $__KERNEL_DS, %eax; \ | |
fe7cacc1 JB |
566 | movl %eax, %ds; \ |
567 | movl %eax, %es; \ | |
be44d2aa | 568 | /* switch to normal stack */ \ |
fe7cacc1 | 569 | FIXUP_ESPFIX_STACK; \ |
be44d2aa | 570 | 27:; |
1da177e4 LT |
571 | |
572 | /* | |
573 | * Build the entry stubs and pointer table with | |
574 | * some assembler magic. | |
575 | */ | |
576 | .data | |
577 | ENTRY(interrupt) | |
578 | .text | |
579 | ||
580 | vector=0 | |
581 | ENTRY(irq_entries_start) | |
fe7cacc1 | 582 | RING0_INT_FRAME |
1da177e4 LT |
583 | .rept NR_IRQS |
584 | ALIGN | |
fe7cacc1 JB |
585 | .if vector |
586 | CFI_ADJUST_CFA_OFFSET -4 | |
587 | .endif | |
19eadf98 | 588 | 1: pushl $~(vector) |
fe7cacc1 | 589 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
590 | jmp common_interrupt |
591 | .data | |
592 | .long 1b | |
593 | .text | |
594 | vector=vector+1 | |
595 | .endr | |
596 | ||
55f327fa IM |
597 | /* |
598 | * the CPU automatically disables interrupts when executing an IRQ vector, | |
599 | * so IRQ-flags tracing has to follow that: | |
600 | */ | |
1da177e4 LT |
601 | ALIGN |
602 | common_interrupt: | |
603 | SAVE_ALL | |
55f327fa | 604 | TRACE_IRQS_OFF |
1da177e4 LT |
605 | movl %esp,%eax |
606 | call do_IRQ | |
607 | jmp ret_from_intr | |
fe7cacc1 | 608 | CFI_ENDPROC |
1da177e4 LT |
609 | |
610 | #define BUILD_INTERRUPT(name, nr) \ | |
611 | ENTRY(name) \ | |
fe7cacc1 | 612 | RING0_INT_FRAME; \ |
19eadf98 | 613 | pushl $~(nr); \ |
fe7cacc1 JB |
614 | CFI_ADJUST_CFA_OFFSET 4; \ |
615 | SAVE_ALL; \ | |
55f327fa | 616 | TRACE_IRQS_OFF \ |
1da177e4 LT |
617 | movl %esp,%eax; \ |
618 | call smp_/**/name; \ | |
55f327fa | 619 | jmp ret_from_intr; \ |
fe7cacc1 | 620 | CFI_ENDPROC |
1da177e4 LT |
621 | |
622 | /* The include is where all of the SMP etc. interrupts come from */ | |
623 | #include "entry_arch.h" | |
624 | ||
d28c4393 P |
625 | KPROBE_ENTRY(page_fault) |
626 | RING0_EC_FRAME | |
627 | pushl $do_page_fault | |
fe7cacc1 | 628 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
629 | ALIGN |
630 | error_code: | |
f95d47ca JF |
631 | /* the function address is in %gs's slot on the stack */ |
632 | pushl %es | |
633 | CFI_ADJUST_CFA_OFFSET 4 | |
634 | /*CFI_REL_OFFSET es, 0*/ | |
1da177e4 | 635 | pushl %ds |
fe7cacc1 JB |
636 | CFI_ADJUST_CFA_OFFSET 4 |
637 | /*CFI_REL_OFFSET ds, 0*/ | |
1da177e4 | 638 | pushl %eax |
fe7cacc1 JB |
639 | CFI_ADJUST_CFA_OFFSET 4 |
640 | CFI_REL_OFFSET eax, 0 | |
1da177e4 | 641 | pushl %ebp |
fe7cacc1 JB |
642 | CFI_ADJUST_CFA_OFFSET 4 |
643 | CFI_REL_OFFSET ebp, 0 | |
1da177e4 | 644 | pushl %edi |
fe7cacc1 JB |
645 | CFI_ADJUST_CFA_OFFSET 4 |
646 | CFI_REL_OFFSET edi, 0 | |
1da177e4 | 647 | pushl %esi |
fe7cacc1 JB |
648 | CFI_ADJUST_CFA_OFFSET 4 |
649 | CFI_REL_OFFSET esi, 0 | |
1da177e4 | 650 | pushl %edx |
fe7cacc1 JB |
651 | CFI_ADJUST_CFA_OFFSET 4 |
652 | CFI_REL_OFFSET edx, 0 | |
1da177e4 | 653 | pushl %ecx |
fe7cacc1 JB |
654 | CFI_ADJUST_CFA_OFFSET 4 |
655 | CFI_REL_OFFSET ecx, 0 | |
1da177e4 | 656 | pushl %ebx |
fe7cacc1 JB |
657 | CFI_ADJUST_CFA_OFFSET 4 |
658 | CFI_REL_OFFSET ebx, 0 | |
1da177e4 | 659 | cld |
f95d47ca | 660 | pushl %gs |
fe7cacc1 | 661 | CFI_ADJUST_CFA_OFFSET 4 |
f95d47ca JF |
662 | /*CFI_REL_OFFSET gs, 0*/ |
663 | movl $(__KERNEL_PDA), %ecx | |
664 | movl %ecx, %gs | |
1da177e4 LT |
665 | UNWIND_ESPFIX_STACK |
666 | popl %ecx | |
fe7cacc1 JB |
667 | CFI_ADJUST_CFA_OFFSET -4 |
668 | /*CFI_REGISTER es, ecx*/ | |
f95d47ca | 669 | movl PT_GS(%esp), %edi # get the function address |
eb5b7b9d | 670 | movl PT_ORIG_EAX(%esp), %edx # get the error code |
f95d47ca JF |
671 | movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart |
672 | mov %ecx, PT_GS(%esp) | |
673 | /*CFI_REL_OFFSET gs, ES*/ | |
1da177e4 LT |
674 | movl $(__USER_DS), %ecx |
675 | movl %ecx, %ds | |
676 | movl %ecx, %es | |
677 | movl %esp,%eax # pt_regs pointer | |
678 | call *%edi | |
679 | jmp ret_from_exception | |
fe7cacc1 | 680 | CFI_ENDPROC |
d28c4393 | 681 | KPROBE_END(page_fault) |
1da177e4 LT |
682 | |
683 | ENTRY(coprocessor_error) | |
fe7cacc1 | 684 | RING0_INT_FRAME |
1da177e4 | 685 | pushl $0 |
fe7cacc1 | 686 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 687 | pushl $do_coprocessor_error |
fe7cacc1 | 688 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 689 | jmp error_code |
fe7cacc1 | 690 | CFI_ENDPROC |
1da177e4 LT |
691 | |
692 | ENTRY(simd_coprocessor_error) | |
fe7cacc1 | 693 | RING0_INT_FRAME |
1da177e4 | 694 | pushl $0 |
fe7cacc1 | 695 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 696 | pushl $do_simd_coprocessor_error |
fe7cacc1 | 697 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 698 | jmp error_code |
fe7cacc1 | 699 | CFI_ENDPROC |
1da177e4 LT |
700 | |
701 | ENTRY(device_not_available) | |
fe7cacc1 | 702 | RING0_INT_FRAME |
1da177e4 | 703 | pushl $-1 # mark this as an int |
fe7cacc1 | 704 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 | 705 | SAVE_ALL |
0da5db31 | 706 | GET_CR0_INTO_EAX |
1da177e4 LT |
707 | testl $0x4, %eax # EM (math emulation bit) |
708 | jne device_not_available_emulate | |
139ec7c4 | 709 | preempt_stop(CLBR_ANY) |
1da177e4 LT |
710 | call math_state_restore |
711 | jmp ret_from_exception | |
712 | device_not_available_emulate: | |
713 | pushl $0 # temporary storage for ORIG_EIP | |
fe7cacc1 | 714 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
715 | call math_emulate |
716 | addl $4, %esp | |
fe7cacc1 | 717 | CFI_ADJUST_CFA_OFFSET -4 |
1da177e4 | 718 | jmp ret_from_exception |
fe7cacc1 | 719 | CFI_ENDPROC |
1da177e4 LT |
720 | |
721 | /* | |
722 | * Debug traps and NMI can happen at the one SYSENTER instruction | |
723 | * that sets up the real kernel stack. Check here, since we can't | |
724 | * allow the wrong stack to be used. | |
725 | * | |
726 | * "TSS_sysenter_esp0+12" is because the NMI/debug handler will have | |
727 | * already pushed 3 words if it hits on the sysenter instruction: | |
728 | * eflags, cs and eip. | |
729 | * | |
730 | * We just load the right stack, and push the three (known) values | |
731 | * by hand onto the new stack - while updating the return eip past | |
732 | * the instruction that would have done it for sysenter. | |
733 | */ | |
734 | #define FIX_STACK(offset, ok, label) \ | |
735 | cmpw $__KERNEL_CS,4(%esp); \ | |
736 | jne ok; \ | |
737 | label: \ | |
738 | movl TSS_sysenter_esp0+offset(%esp),%esp; \ | |
a549b86d CE |
739 | CFI_DEF_CFA esp, 0; \ |
740 | CFI_UNDEFINED eip; \ | |
1da177e4 | 741 | pushfl; \ |
a549b86d | 742 | CFI_ADJUST_CFA_OFFSET 4; \ |
1da177e4 | 743 | pushl $__KERNEL_CS; \ |
a549b86d CE |
744 | CFI_ADJUST_CFA_OFFSET 4; \ |
745 | pushl $sysenter_past_esp; \ | |
746 | CFI_ADJUST_CFA_OFFSET 4; \ | |
747 | CFI_REL_OFFSET eip, 0 | |
1da177e4 | 748 | |
3d97ae5b | 749 | KPROBE_ENTRY(debug) |
fe7cacc1 | 750 | RING0_INT_FRAME |
1da177e4 LT |
751 | cmpl $sysenter_entry,(%esp) |
752 | jne debug_stack_correct | |
753 | FIX_STACK(12, debug_stack_correct, debug_esp_fix_insn) | |
754 | debug_stack_correct: | |
755 | pushl $-1 # mark this as an int | |
fe7cacc1 | 756 | CFI_ADJUST_CFA_OFFSET 4 |
1da177e4 LT |
757 | SAVE_ALL |
758 | xorl %edx,%edx # error code 0 | |
759 | movl %esp,%eax # pt_regs pointer | |
760 | call do_debug | |
1da177e4 | 761 | jmp ret_from_exception |
fe7cacc1 | 762 | CFI_ENDPROC |
d28c4393 P |
763 | KPROBE_END(debug) |
764 | ||
1da177e4 LT |
765 | /* |
766 | * NMI is doubly nasty. It can happen _while_ we're handling | |
767 | * a debug fault, and the debug fault hasn't yet been able to | |
768 | * clear up the stack. So we first check whether we got an | |
769 | * NMI on the sysenter entry path, but after that we need to | |
770 | * check whether we got an NMI on the debug path where the debug | |
771 | * fault happened on the sysenter path. | |
772 | */ |