| blob | b5c061f8f358489fee828e48b836d18c3bd6da79 |
1 /*
2 * linux/arch/x86_64/entry.S
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
6 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
7 */
9 /*
10 * entry.S contains the system-call and fault low-level handling routines.
11 *
12 * NOTE: This code handles signal-recognition, which happens every time
13 * after an interrupt and after each system call.
14 *
15 * Normal syscalls and interrupts don't save a full stack frame, this is
16 * only done for syscall tracing, signals or fork/exec et.al.
17 *
18 * A note on terminology:
19 * - top of stack: Architecture defined interrupt frame from SS to RIP
20 * at the top of the kernel process stack.
21 * - partial stack frame: partially saved registers upto R11.
22 * - full stack frame: Like partial stack frame, but all register saved.
23 *
24 * Some macro usage:
25 * - CFI macros are used to generate dwarf2 unwind information for better
26 * backtraces. They don't change any code.
27 * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
28 * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
29 * There are unfortunately lots of special cases where some registers
30 * not touched. The macro is a big mess that should be cleaned up.
31 * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
32 * Gives a full stack frame.
33 * - ENTRY/END Define functions in the symbol table.
34 * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
35 * frame that is otherwise undefined after a SYSCALL
36 * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
37 * - errorentry/paranoidentry/zeroentry - Define exception entry points.
38 */
40 #include <linux/linkage.h>
41 #include <asm/segment.h>
42 #include <asm/cache.h>
43 #include <asm/errno.h>
44 #include <asm/dwarf2.h>
45 #include <asm/calling.h>
46 #include <asm/asm-offsets.h>
47 #include <asm/msr.h>
48 #include <asm/unistd.h>
49 #include <asm/thread_info.h>
50 #include <asm/hw_irq.h>
51 #include <asm/page_types.h>
52 #include <asm/irqflags.h>
53 #include <asm/paravirt.h>
54 #include <asm/ftrace.h>
55 #include <asm/percpu.h>
57 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
58 #include <linux/elf-em.h>
59 #define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
60 #define __AUDIT_ARCH_64BIT 0x80000000
61 #define __AUDIT_ARCH_LE 0x40000000
63 .code64
64 #ifdef CONFIG_FUNCTION_TRACER
65 #ifdef CONFIG_DYNAMIC_FTRACE
66 ENTRY(mcount)
67 retq
68 END(mcount)
70 ENTRY(ftrace_caller)
71 cmpl $0, function_trace_stop
72 jne ftrace_stub
74 MCOUNT_SAVE_FRAME
76 movq 0x38(%rsp), %rdi
77 movq 8(%rbp), %rsi
78 subq $MCOUNT_INSN_SIZE, %rdi
80 GLOBAL(ftrace_call)
81 call ftrace_stub
83 MCOUNT_RESTORE_FRAME
85 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
86 GLOBAL(ftrace_graph_call)
87 jmp ftrace_stub
88 #endif
90 GLOBAL(ftrace_stub)
91 retq
92 END(ftrace_caller)
94 #else /* ! CONFIG_DYNAMIC_FTRACE */
95 ENTRY(mcount)
96 cmpl $0, function_trace_stop
97 jne ftrace_stub
99 cmpq $ftrace_stub, ftrace_trace_function
100 jnz trace
102 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
103 cmpq $ftrace_stub, ftrace_graph_return
104 jnz ftrace_graph_caller
106 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
107 jnz ftrace_graph_caller
108 #endif
110 GLOBAL(ftrace_stub)
111 retq
113 trace:
114 MCOUNT_SAVE_FRAME
116 movq 0x38(%rsp), %rdi
117 movq 8(%rbp), %rsi
118 subq $MCOUNT_INSN_SIZE, %rdi
120 call *ftrace_trace_function
122 MCOUNT_RESTORE_FRAME
124 jmp ftrace_stub
125 END(mcount)
126 #endif /* CONFIG_DYNAMIC_FTRACE */
127 #endif /* CONFIG_FUNCTION_TRACER */
129 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
130 ENTRY(ftrace_graph_caller)
131 cmpl $0, function_trace_stop
132 jne ftrace_stub
134 MCOUNT_SAVE_FRAME
136 leaq 8(%rbp), %rdi
137 movq 0x38(%rsp), %rsi
138 movq (%rbp), %rdx
139 subq $MCOUNT_INSN_SIZE, %rsi
141 call prepare_ftrace_return
143 MCOUNT_RESTORE_FRAME
145 retq
146 END(ftrace_graph_caller)
148 GLOBAL(return_to_handler)
149 subq $24, %rsp
151 /* Save the return values */
152 movq %rax, (%rsp)
153 movq %rdx, 8(%rsp)
154 movq %rbp, %rdi
156 call ftrace_return_to_handler
158 movq %rax, 16(%rsp)
159 movq 8(%rsp), %rdx
160 movq (%rsp), %rax
161 addq $16, %rsp
162 retq
163 #endif
166 #ifndef CONFIG_PREEMPT
167 #define retint_kernel retint_restore_args
168 #endif
170 #ifdef CONFIG_PARAVIRT
171 ENTRY(native_usergs_sysret64)
172 swapgs
173 sysretq
174 ENDPROC(native_usergs_sysret64)
175 #endif /* CONFIG_PARAVIRT */
178 .macro TRACE_IRQS_IRETQ offset=ARGOFFSET
179 #ifdef CONFIG_TRACE_IRQFLAGS
180 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
181 jnc 1f
182 TRACE_IRQS_ON
183 1:
184 #endif
185 .endm
187 /*
188 * C code is not supposed to know about undefined top of stack. Every time
189 * a C function with an pt_regs argument is called from the SYSCALL based
190 * fast path FIXUP_TOP_OF_STACK is needed.
191 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
192 * manipulation.
193 */
195 /* %rsp:at FRAMEEND */
196 .macro FIXUP_TOP_OF_STACK tmp offset=0
197 movq PER_CPU_VAR(old_rsp),\tmp
198 movq \tmp,RSP+\offset(%rsp)
199 movq $__USER_DS,SS+\offset(%rsp)
200 movq $__USER_CS,CS+\offset(%rsp)
201 movq $-1,RCX+\offset(%rsp)
202 movq R11+\offset(%rsp),\tmp /* get eflags */
203 movq \tmp,EFLAGS+\offset(%rsp)
204 .endm
206 .macro RESTORE_TOP_OF_STACK tmp offset=0
207 movq RSP+\offset(%rsp),\tmp
208 movq \tmp,PER_CPU_VAR(old_rsp)
209 movq EFLAGS+\offset(%rsp),\tmp
210 movq \tmp,R11+\offset(%rsp)
211 .endm
213 .macro FAKE_STACK_FRAME child_rip
214 /* push in order ss, rsp, eflags, cs, rip */
215 xorl %eax, %eax
216 pushq $__KERNEL_DS /* ss */
217 CFI_ADJUST_CFA_OFFSET 8
218 /*CFI_REL_OFFSET ss,0*/
219 pushq %rax /* rsp */
220 CFI_ADJUST_CFA_OFFSET 8
221 CFI_REL_OFFSET rsp,0
222 pushq $X86_EFLAGS_IF /* eflags - interrupts on */
223 CFI_ADJUST_CFA_OFFSET 8
224 /*CFI_REL_OFFSET rflags,0*/
225 pushq $__KERNEL_CS /* cs */
226 CFI_ADJUST_CFA_OFFSET 8
227 /*CFI_REL_OFFSET cs,0*/
228 pushq \child_rip /* rip */
229 CFI_ADJUST_CFA_OFFSET 8
230 CFI_REL_OFFSET rip,0
231 pushq %rax /* orig rax */
232 CFI_ADJUST_CFA_OFFSET 8
233 .endm
235 .macro UNFAKE_STACK_FRAME
236 addq $8*6, %rsp
237 CFI_ADJUST_CFA_OFFSET -(6*8)
238 .endm
240 /*
241 * initial frame state for interrupts (and exceptions without error code)
242 */
243 .macro EMPTY_FRAME start=1 offset=0
244 .if \start
245 CFI_STARTPROC simple
246 CFI_SIGNAL_FRAME
247 CFI_DEF_CFA rsp,8+\offset
248 .else
249 CFI_DEF_CFA_OFFSET 8+\offset
250 .endif
251 .endm
253 /*
254 * initial frame state for interrupts (and exceptions without error code)
255 */
256 .macro INTR_FRAME start=1 offset=0
257 EMPTY_FRAME \start, SS+8+\offset-RIP
258 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
259 CFI_REL_OFFSET rsp, RSP+\offset-RIP
260 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
261 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
262 CFI_REL_OFFSET rip, RIP+\offset-RIP
263 .endm
265 /*
266 * initial frame state for exceptions with error code (and interrupts
267 * with vector already pushed)
268 */
269 .macro XCPT_FRAME start=1 offset=0
270 INTR_FRAME \start, RIP+\offset-ORIG_RAX
271 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
272 .endm
274 /*
275 * frame that enables calling into C.
276 */
277 .macro PARTIAL_FRAME start=1 offset=0
278 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
279 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
280 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
281 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
282 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
283 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
284 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
285 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
286 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
287 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
288 .endm
290 /*
291 * frame that enables passing a complete pt_regs to a C function.
292 */
293 .macro DEFAULT_FRAME start=1 offset=0
294 PARTIAL_FRAME \start, R11+\offset-R15
295 CFI_REL_OFFSET rbx, RBX+\offset
296 CFI_REL_OFFSET rbp, RBP+\offset
297 CFI_REL_OFFSET r12, R12+\offset
298 CFI_REL_OFFSET r13, R13+\offset
299 CFI_REL_OFFSET r14, R14+\offset
300 CFI_REL_OFFSET r15, R15+\offset
301 .endm
303 /* save partial stack frame */
304 ENTRY(save_args)
305 XCPT_FRAME
306 cld
307 movq_cfi rdi, RDI+16-ARGOFFSET
308 movq_cfi rsi, RSI+16-ARGOFFSET
309 movq_cfi rdx, RDX+16-ARGOFFSET
310 movq_cfi rcx, RCX+16-ARGOFFSET
311 movq_cfi rax, RAX+16-ARGOFFSET
312 movq_cfi r8, R8+16-ARGOFFSET
313 movq_cfi r9, R9+16-ARGOFFSET
314 movq_cfi r10, R10+16-ARGOFFSET
315 movq_cfi r11, R11+16-ARGOFFSET
317 leaq -ARGOFFSET+16(%rsp),%rdi /* arg1 for handler */
318 movq_cfi rbp, 8 /* push %rbp */
319 leaq 8(%rsp), %rbp /* mov %rsp, %ebp */
320 testl $3, CS(%rdi)
321 je 1f
322 SWAPGS
323 /*
324 * irq_count is used to check if a CPU is already on an interrupt stack
325 * or not. While this is essentially redundant with preempt_count it is
326 * a little cheaper to use a separate counter in the PDA (short of
327 * moving irq_enter into assembly, which would be too much work)
328 */
329 1: incl PER_CPU_VAR(irq_count)
330 jne 2f
331 popq_cfi %rax /* move return address... */
332 mov PER_CPU_VAR(irq_stack_ptr),%rsp
333 EMPTY_FRAME 0
334 pushq_cfi %rbp /* backlink for unwinder */
335 pushq_cfi %rax /* ... to the new stack */
336 /*
337 * We entered an interrupt context - irqs are off:
338 */
339 2: TRACE_IRQS_OFF
340 ret
341 CFI_ENDPROC
342 END(save_args)
344 ENTRY(save_rest)
345 PARTIAL_FRAME 1 REST_SKIP+8
346 movq 5*8+16(%rsp), %r11 /* save return address */
347 movq_cfi rbx, RBX+16
348 movq_cfi rbp, RBP+16
349 movq_cfi r12, R12+16
350 movq_cfi r13, R13+16
351 movq_cfi r14, R14+16
352 movq_cfi r15, R15+16
353 movq %r11, 8(%rsp) /* return address */
354 FIXUP_TOP_OF_STACK %r11, 16
355 ret
356 CFI_ENDPROC
357 END(save_rest)
359 /* save complete stack frame */
360 .pushsection .kprobes.text, "ax"
361 ENTRY(save_paranoid)
362 XCPT_FRAME 1 RDI+8
363 cld
364 movq_cfi rdi, RDI+8
365 movq_cfi rsi, RSI+8
366 movq_cfi rdx, RDX+8
367 movq_cfi rcx, RCX+8
368 movq_cfi rax, RAX+8
369 movq_cfi r8, R8+8
370 movq_cfi r9, R9+8
371 movq_cfi r10, R10+8
372 movq_cfi r11, R11+8
373 movq_cfi rbx, RBX+8
374 movq_cfi rbp, RBP+8
375 movq_cfi r12, R12+8
376 movq_cfi r13, R13+8
377 movq_cfi r14, R14+8
378 movq_cfi r15, R15+8
379 movl $1,%ebx
380 movl $MSR_GS_BASE,%ecx
381 rdmsr
382 testl %edx,%edx
383 js 1f /* negative -> in kernel */
384 SWAPGS
385 xorl %ebx,%ebx
386 1: ret
387 CFI_ENDPROC
388 END(save_paranoid)
389 .popsection
391 /*
392 * A newly forked process directly context switches into this address.
393 *
394 * rdi: prev task we switched from
395 */
396 ENTRY(ret_from_fork)
397 DEFAULT_FRAME
399 LOCK ; btr $TIF_FORK,TI_flags(%r8)
401 push kernel_eflags(%rip)
402 CFI_ADJUST_CFA_OFFSET 8
403 popf # reset kernel eflags
404 CFI_ADJUST_CFA_OFFSET -8
406 call schedule_tail # rdi: 'prev' task parameter
408 GET_THREAD_INFO(%rcx)
410 RESTORE_REST
412 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
413 je int_ret_from_sys_call
415 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
416 jnz int_ret_from_sys_call
418 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
419 jmp ret_from_sys_call # go to the SYSRET fastpath
421 CFI_ENDPROC
422 END(ret_from_fork)
424 /*
425 * System call entry. Upto 6 arguments in registers are supported.
426 *
427 * SYSCALL does not save anything on the stack and does not change the
428 * stack pointer.
429 */
431 /*
432 * Register setup:
433 * rax system call number
434 * rdi arg0
435 * rcx return address for syscall/sysret, C arg3
436 * rsi arg1
437 * rdx arg2
438 * r10 arg3 (--> moved to rcx for C)
439 * r8 arg4
440 * r9 arg5
441 * r11 eflags for syscall/sysret, temporary for C
442 * r12-r15,rbp,rbx saved by C code, not touched.
443 *
444 * Interrupts are off on entry.
445 * Only called from user space.
446 *
447 * XXX if we had a free scratch register we could save the RSP into the stack frame
448 * and report it properly in ps. Unfortunately we haven't.
449 *
450 * When user can change the frames always force IRET. That is because
451 * it deals with uncanonical addresses better. SYSRET has trouble
452 * with them due to bugs in both AMD and Intel CPUs.
453 */
455 ENTRY(system_call)
456 CFI_STARTPROC simple
457 CFI_SIGNAL_FRAME
458 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
459 CFI_REGISTER rip,rcx
460 /*CFI_REGISTER rflags,r11*/
461 SWAPGS_UNSAFE_STACK
462 /*
463 * A hypervisor implementation might want to use a label
464 * after the swapgs, so that it can do the swapgs
465 * for the guest and jump here on syscall.
466 */
467 ENTRY(system_call_after_swapgs)
469 movq %rsp,PER_CPU_VAR(old_rsp)
470 movq PER_CPU_VAR(kernel_stack),%rsp
471 /*
472 * No need to follow this irqs off/on section - it's straight
473 * and short:
474 */
475 ENABLE_INTERRUPTS(CLBR_NONE)
476 SAVE_ARGS 8,1
477 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
478 movq %rcx,RIP-ARGOFFSET(%rsp)
479 CFI_REL_OFFSET rip,RIP-ARGOFFSET
480 GET_THREAD_INFO(%rcx)
481 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags(%rcx)
482 jnz tracesys
483 system_call_fastpath:
484 cmpq $__NR_syscall_max,%rax
485 ja badsys
486 movq %r10,%rcx
487 call *sys_call_table(,%rax,8) # XXX: rip relative
488 movq %rax,RAX-ARGOFFSET(%rsp)
489 /*
490 * Syscall return path ending with SYSRET (fast path)
491 * Has incomplete stack frame and undefined top of stack.
492 */
493 ret_from_sys_call:
494 movl $_TIF_ALLWORK_MASK,%edi
495 /* edi: flagmask */
496 sysret_check:
497 LOCKDEP_SYS_EXIT
498 GET_THREAD_INFO(%rcx)
499 DISABLE_INTERRUPTS(CLBR_NONE)
500 TRACE_IRQS_OFF
501 movl TI_flags(%rcx),%edx
502 andl %edi,%edx
503 jnz sysret_careful
504 CFI_REMEMBER_STATE
505 /*
506 * sysretq will re-enable interrupts:
507 */
508 TRACE_IRQS_ON
509 movq RIP-ARGOFFSET(%rsp),%rcx
510 CFI_REGISTER rip,rcx
511 RESTORE_ARGS 0,-ARG_SKIP,1
512 /*CFI_REGISTER rflags,r11*/
513 movq PER_CPU_VAR(old_rsp), %rsp
514 USERGS_SYSRET64
516 CFI_RESTORE_STATE
517 /* Handle reschedules */
518 /* edx: work, edi: workmask */
519 sysret_careful:
520 bt $TIF_NEED_RESCHED,%edx
521 jnc sysret_signal
522 TRACE_IRQS_ON
523 ENABLE_INTERRUPTS(CLBR_NONE)
524 pushq %rdi
525 CFI_ADJUST_CFA_OFFSET 8
526 call schedule
527 popq %rdi
528 CFI_ADJUST_CFA_OFFSET -8
529 jmp sysret_check
531 /* Handle a signal */
532 sysret_signal:
533 TRACE_IRQS_ON
534 ENABLE_INTERRUPTS(CLBR_NONE)
535 #ifdef CONFIG_AUDITSYSCALL
536 bt $TIF_SYSCALL_AUDIT,%edx
537 jc sysret_audit
538 #endif
539 /*
540 * We have a signal, or exit tracing or single-step.
541 * These all wind up with the iret return path anyway,
542 * so just join that path right now.
543 */
544 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
545 jmp int_check_syscall_exit_work
547 badsys:
548 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
549 jmp ret_from_sys_call
551 #ifdef CONFIG_AUDITSYSCALL
552 /*
553 * Fast path for syscall audit without full syscall trace.
554 * We just call audit_syscall_entry() directly, and then
555 * jump back to the normal fast path.
556 */
557 auditsys:
558 movq %r10,%r9 /* 6th arg: 4th syscall arg */
559 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
560 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
561 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
562 movq %rax,%rsi /* 2nd arg: syscall number */
563 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
564 call audit_syscall_entry
565 LOAD_ARGS 0 /* reload call-clobbered registers */
566 jmp system_call_fastpath
568 /*
569 * Return fast path for syscall audit. Call audit_syscall_exit()
570 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
571 * masked off.
572 */
573 sysret_audit:
574 movq %rax,%rsi /* second arg, syscall return value */
575 cmpq $0,%rax /* is it < 0? */
576 setl %al /* 1 if so, 0 if not */
577 movzbl %al,%edi /* zero-extend that into %edi */
578 inc %edi /* first arg, 0->1(AUDITSC_SUCCESS), 1->2(AUDITSC_FAILURE) */
579 call audit_syscall_exit
580 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
581 jmp sysret_check
582 #endif /* CONFIG_AUDITSYSCALL */
584 /* Do syscall tracing */
585 tracesys:
586 #ifdef CONFIG_AUDITSYSCALL
587 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags(%rcx)
588 jz auditsys
589 #endif
590 SAVE_REST
591 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
592 FIXUP_TOP_OF_STACK %rdi
593 movq %rsp,%rdi
594 call syscall_trace_enter
595 /*
596 * Reload arg registers from stack in case ptrace changed them.
597 * We don't reload %rax because syscall_trace_enter() returned
598 * the value it wants us to use in the table lookup.
599 */
600 LOAD_ARGS ARGOFFSET, 1
601 RESTORE_REST
602 cmpq $__NR_syscall_max,%rax
603 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
604 movq %r10,%rcx /* fixup for C */
605 call *sys_call_table(,%rax,8)
606 movq %rax,RAX-ARGOFFSET(%rsp)
607 /* Use IRET because user could have changed frame */
609 /*
610 * Syscall return path ending with IRET.
611 * Has correct top of stack, but partial stack frame.
612 */
613 GLOBAL(int_ret_from_sys_call)
614 DISABLE_INTERRUPTS(CLBR_NONE)
615 TRACE_IRQS_OFF
616 testl $3,CS-ARGOFFSET(%rsp)
617 je retint_restore_args
618 movl $_TIF_ALLWORK_MASK,%edi
619 /* edi: mask to check */
620 GLOBAL(int_with_check)
621 LOCKDEP_SYS_EXIT_IRQ
622 GET_THREAD_INFO(%rcx)
623 movl TI_flags(%rcx),%edx
624 andl %edi,%edx
625 jnz int_careful
626 andl $~TS_COMPAT,TI_status(%rcx)
627 jmp retint_swapgs
629 /* Either reschedule or signal or syscall exit tracking needed. */
630 /* First do a reschedule test. */
631 /* edx: work, edi: workmask */
632 int_careful:
633 bt $TIF_NEED_RESCHED,%edx
634 jnc int_very_careful
635 TRACE_IRQS_ON
636 ENABLE_INTERRUPTS(CLBR_NONE)
637 pushq %rdi
638 CFI_ADJUST_CFA_OFFSET 8
639 call schedule
640 popq %rdi
641 CFI_ADJUST_CFA_OFFSET -8
642 DISABLE_INTERRUPTS(CLBR_NONE)
643 TRACE_IRQS_OFF
644 jmp int_with_check
646 /* handle signals and tracing -- both require a full stack frame */
647 int_very_careful:
648 TRACE_IRQS_ON
649 ENABLE_INTERRUPTS(CLBR_NONE)
650 int_check_syscall_exit_work:
651 SAVE_REST
652 /* Check for syscall exit trace */
653 testl $_TIF_WORK_SYSCALL_EXIT,%edx
654 jz int_signal
655 pushq %rdi
656 CFI_ADJUST_CFA_OFFSET 8
657 leaq 8(%rsp),%rdi # &ptregs -> arg1
658 call syscall_trace_leave
659 popq %rdi
660 CFI_ADJUST_CFA_OFFSET -8
661 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
662 jmp int_restore_rest
664 int_signal:
665 testl $_TIF_DO_NOTIFY_MASK,%edx
666 jz 1f
667 movq %rsp,%rdi # &ptregs -> arg1
668 xorl %esi,%esi # oldset -> arg2
669 call do_notify_resume
670 1: movl $_TIF_WORK_MASK,%edi
671 int_restore_rest:
672 RESTORE_REST
673 DISABLE_INTERRUPTS(CLBR_NONE)
674 TRACE_IRQS_OFF
675 jmp int_with_check
676 CFI_ENDPROC
677 END(system_call)
679 /*
680 * Certain special system calls that need to save a complete full stack frame.
681 */
682 .macro PTREGSCALL label,func,arg
683 ENTRY(\label)
684 PARTIAL_FRAME 1 8 /* offset 8: return address */
685 subq $REST_SKIP, %rsp
686 CFI_ADJUST_CFA_OFFSET REST_SKIP
687 call save_rest
688 DEFAULT_FRAME 0 8 /* offset 8: return address */
689 leaq 8(%rsp), \arg /* pt_regs pointer */
690 call \func
691 jmp ptregscall_common
692 CFI_ENDPROC
693 END(\label)
694 .endm
696 PTREGSCALL stub_clone, sys_clone, %r8
697 PTREGSCALL stub_fork, sys_fork, %rdi
698 PTREGSCALL stub_vfork, sys_vfork, %rdi
699 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
700 PTREGSCALL stub_iopl, sys_iopl, %rsi
702 ENTRY(ptregscall_common)
703 DEFAULT_FRAME 1 8 /* offset 8: return address */
704 RESTORE_TOP_OF_STACK %r11, 8
705 movq_cfi_restore R15+8, r15
706 movq_cfi_restore R14+8, r14
707 movq_cfi_restore R13+8, r13
708 movq_cfi_restore R12+8, r12
709 movq_cfi_restore RBP+8, rbp
710 movq_cfi_restore RBX+8, rbx
711 ret $REST_SKIP /* pop extended registers */
712 CFI_ENDPROC
713 END(ptregscall_common)
715 ENTRY(stub_execve)
716 CFI_STARTPROC
717 popq %r11
718 CFI_ADJUST_CFA_OFFSET -8
719 CFI_REGISTER rip, r11
720 SAVE_REST
721 FIXUP_TOP_OF_STACK %r11
722 movq %rsp, %rcx
723 call sys_execve
724 RESTORE_TOP_OF_STACK %r11
725 movq %rax,RAX(%rsp)
726 RESTORE_REST
727 jmp int_ret_from_sys_call
728 CFI_ENDPROC
729 END(stub_execve)
731 /*
732 * sigreturn is special because it needs to restore all registers on return.
733 * This cannot be done with SYSRET, so use the IRET return path instead.
734 */
735 ENTRY(stub_rt_sigreturn)
736 CFI_STARTPROC
737 addq $8, %rsp
738 CFI_ADJUST_CFA_OFFSET -8
739 SAVE_REST
740 movq %rsp,%rdi
741 FIXUP_TOP_OF_STACK %r11
742 call sys_rt_sigreturn
743 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
744 RESTORE_REST
745 jmp int_ret_from_sys_call
746 CFI_ENDPROC
747 END(stub_rt_sigreturn)
749 /*
750 * Build the entry stubs and pointer table with some assembler magic.
751 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
752 * single cache line on all modern x86 implementations.
753 */
754 .section .init.rodata,"a"
755 ENTRY(interrupt)
756 .text
757 .p2align 5
758 .p2align CONFIG_X86_L1_CACHE_SHIFT
759 ENTRY(irq_entries_start)
760 INTR_FRAME
761 vector=FIRST_EXTERNAL_VECTOR
762 .rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
763 .balign 32
764 .rept 7
765 .if vector < NR_VECTORS
766 .if vector <> FIRST_EXTERNAL_VECTOR
767 CFI_ADJUST_CFA_OFFSET -8
768 .endif
769 1: pushq $(~vector+0x80) /* Note: always in signed byte range */
770 CFI_ADJUST_CFA_OFFSET 8
771 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
772 jmp 2f
773 .endif
774 .previous
775 .quad 1b
776 .text
777 vector=vector+1
778 .endif
779 .endr
780 2: jmp common_interrupt
781 .endr
782 CFI_ENDPROC
783 END(irq_entries_start)
785 .previous
786 END(interrupt)
787 .previous
789 /*
790 * Interrupt entry/exit.
791 *
792 * Interrupt entry points save only callee clobbered registers in fast path.
793 *
794 * Entry runs with interrupts off.
795 */
797 /* 0(%rsp): ~(interrupt number) */
798 .macro interrupt func
799 subq $10*8, %rsp
800 CFI_ADJUST_CFA_OFFSET 10*8
801 call save_args
802 PARTIAL_FRAME 0
803 call \func
804 .endm
806 /*
807 * The interrupt stubs push (~vector+0x80) onto the stack and
808 * then jump to common_interrupt.
809 */
810 .p2align CONFIG_X86_L1_CACHE_SHIFT
811 common_interrupt:
812 XCPT_FRAME
813 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
814 interrupt do_IRQ
815 /* 0(%rsp): old_rsp-ARGOFFSET */
816 ret_from_intr:
817 DISABLE_INTERRUPTS(CLBR_NONE)
818 TRACE_IRQS_OFF
819 decl PER_CPU_VAR(irq_count)
820 leaveq
821 CFI_DEF_CFA_REGISTER rsp
822 CFI_ADJUST_CFA_OFFSET -8
823 exit_intr:
824 GET_THREAD_INFO(%rcx)
825 testl $3,CS-ARGOFFSET(%rsp)
826 je retint_kernel
828 /* Interrupt came from user space */
829 /*
830 * Has a correct top of stack, but a partial stack frame
831 * %rcx: thread info. Interrupts off.
832 */
833 retint_with_reschedule:
834 movl $_TIF_WORK_MASK,%edi
835 retint_check:
836 LOCKDEP_SYS_EXIT_IRQ
837 movl TI_flags(%rcx),%edx
838 andl %edi,%edx
839 CFI_REMEMBER_STATE
840 jnz retint_careful
842 retint_swapgs: /* return to user-space */
843 /*
844 * The iretq could re-enable interrupts:
845 */
846 DISABLE_INTERRUPTS(CLBR_ANY)
847 TRACE_IRQS_IRETQ
848 SWAPGS
849 jmp restore_args
851 retint_restore_args: /* return to kernel space */
852 DISABLE_INTERRUPTS(CLBR_ANY)
853 /*
854 * The iretq could re-enable interrupts:
855 */
856 TRACE_IRQS_IRETQ
857 restore_args:
858 RESTORE_ARGS 0,8,0
860 irq_return:
861 INTERRUPT_RETURN
863 .section __ex_table, "a"
864 .quad irq_return, bad_iret
865 .previous
867 #ifdef CONFIG_PARAVIRT
868 ENTRY(native_iret)
869 iretq
871 .section __ex_table,"a"
872 .quad native_iret, bad_iret
873 .previous
874 #endif
876 .section .fixup,"ax"
877 bad_iret:
878 /*
879 * The iret traps when the %cs or %ss being restored is bogus.
880 * We've lost the original trap vector and error code.
881 * #GPF is the most likely one to get for an invalid selector.
882 * So pretend we completed the iret and took the #GPF in user mode.
883 *
884 * We are now running with the kernel GS after exception recovery.
885 * But error_entry expects us to have user GS to match the user %cs,
886 * so swap back.
887 */
888 pushq $0
890 SWAPGS
891 jmp general_protection
893 .previous
895 /* edi: workmask, edx: work */
896 retint_careful:
897 CFI_RESTORE_STATE
898 bt $TIF_NEED_RESCHED,%edx
899 jnc retint_signal
900 TRACE_IRQS_ON
901 ENABLE_INTERRUPTS(CLBR_NONE)
902 pushq %rdi
903 CFI_ADJUST_CFA_OFFSET 8
904 call schedule
905 popq %rdi
906 CFI_ADJUST_CFA_OFFSET -8
907 GET_THREAD_INFO(%rcx)
908 DISABLE_INTERRUPTS(CLBR_NONE)
909 TRACE_IRQS_OFF
910 jmp retint_check
912 retint_signal:
913 testl $_TIF_DO_NOTIFY_MASK,%edx
914 jz retint_swapgs
915 TRACE_IRQS_ON
916 ENABLE_INTERRUPTS(CLBR_NONE)
917 SAVE_REST
918 movq $-1,ORIG_RAX(%rsp)
919 xorl %esi,%esi # oldset
920 movq %rsp,%rdi # &pt_regs
921 call do_notify_resume
922 RESTORE_REST
923 DISABLE_INTERRUPTS(CLBR_NONE)
924 TRACE_IRQS_OFF
925 GET_THREAD_INFO(%rcx)
926 jmp retint_with_reschedule
928 #ifdef CONFIG_PREEMPT
929 /* Returning to kernel space. Check if we need preemption */
930 /* rcx: threadinfo. interrupts off. */
931 ENTRY(retint_kernel)
932 cmpl $0,TI_preempt_count(%rcx)
933 jnz retint_restore_args
934 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
935 jnc retint_restore_args
936 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
937 jnc retint_restore_args
938 call preempt_schedule_irq
939 jmp exit_intr
940 #endif
942 CFI_ENDPROC
943 END(common_interrupt)
945 /*
946 * APIC interrupts.
947 */
948 .macro apicinterrupt num sym do_sym
949 ENTRY(\sym)
950 INTR_FRAME
951 pushq $~(\num)
952 CFI_ADJUST_CFA_OFFSET 8
953 interrupt \do_sym
954 jmp ret_from_intr
955 CFI_ENDPROC
956 END(\sym)
957 .endm
959 #ifdef CONFIG_SMP
960 apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
961 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
962 apicinterrupt REBOOT_VECTOR \
963 reboot_interrupt smp_reboot_interrupt
964 #endif
966 #ifdef CONFIG_X86_UV
967 apicinterrupt UV_BAU_MESSAGE \
968 uv_bau_message_intr1 uv_bau_message_interrupt
969 #endif
970 apicinterrupt LOCAL_TIMER_VECTOR \
971 apic_timer_interrupt smp_apic_timer_interrupt
972 apicinterrupt GENERIC_INTERRUPT_VECTOR \
973 generic_interrupt smp_generic_interrupt
975 #ifdef CONFIG_SMP
976 apicinterrupt INVALIDATE_TLB_VECTOR_START+0 \
977 invalidate_interrupt0 smp_invalidate_interrupt
978 apicinterrupt INVALIDATE_TLB_VECTOR_START+1 \
979 invalidate_interrupt1 smp_invalidate_interrupt
980 apicinterrupt INVALIDATE_TLB_VECTOR_START+2 \
981 invalidate_interrupt2 smp_invalidate_interrupt
982 apicinterrupt INVALIDATE_TLB_VECTOR_START+3 \
983 invalidate_interrupt3 smp_invalidate_interrupt
984 apicinterrupt INVALIDATE_TLB_VECTOR_START+4 \
985 invalidate_interrupt4 smp_invalidate_interrupt
986 apicinterrupt INVALIDATE_TLB_VECTOR_START+5 \
987 invalidate_interrupt5 smp_invalidate_interrupt
988 apicinterrupt INVALIDATE_TLB_VECTOR_START+6 \
989 invalidate_interrupt6 smp_invalidate_interrupt
990 apicinterrupt INVALIDATE_TLB_VECTOR_START+7 \
991 invalidate_interrupt7 smp_invalidate_interrupt
992 #endif
994 apicinterrupt THRESHOLD_APIC_VECTOR \
995 threshold_interrupt smp_threshold_interrupt
996 apicinterrupt THERMAL_APIC_VECTOR \
997 thermal_interrupt smp_thermal_interrupt
999 #ifdef CONFIG_X86_MCE
1000 apicinterrupt MCE_SELF_VECTOR \
1001 mce_self_interrupt smp_mce_self_interrupt
1002 #endif
1004 #ifdef CONFIG_SMP
1005 apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1006 call_function_single_interrupt smp_call_function_single_interrupt
1007 apicinterrupt CALL_FUNCTION_VECTOR \
1008 call_function_interrupt smp_call_function_interrupt
1009 apicinterrupt RESCHEDULE_VECTOR \
1010 reschedule_interrupt smp_reschedule_interrupt
1011 #endif
1013 apicinterrupt ERROR_APIC_VECTOR \
1014 error_interrupt smp_error_interrupt
1015 apicinterrupt SPURIOUS_APIC_VECTOR \
1016 spurious_interrupt smp_spurious_interrupt
1018 #ifdef CONFIG_PERF_EVENTS
1019 apicinterrupt LOCAL_PENDING_VECTOR \
1020 perf_pending_interrupt smp_perf_pending_interrupt
1021 #endif
1023 /*
1024 * Exception entry points.
1025 */
1026 .macro zeroentry sym do_sym
1027 ENTRY(\sym)
1028 INTR_FRAME
1029 PARAVIRT_ADJUST_EXCEPTION_FRAME
1030 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1031 subq $15*8,%rsp
1032 CFI_ADJUST_CFA_OFFSET 15*8
1033 call error_entry
1034 DEFAULT_FRAME 0
1035 movq %rsp,%rdi /* pt_regs pointer */
1036 xorl %esi,%esi /* no error code */
1037 call \do_sym
1038 jmp error_exit /* %ebx: no swapgs flag */
1039 CFI_ENDPROC
1040 END(\sym)
1041 .endm
1043 .macro paranoidzeroentry sym do_sym
1044 ENTRY(\sym)
1045 INTR_FRAME
1046 PARAVIRT_ADJUST_EXCEPTION_FRAME
1047 pushq $-1 /* ORIG_RAX: no syscall to restart */
1048 CFI_ADJUST_CFA_OFFSET 8
1049 subq $15*8, %rsp
1050 call save_paranoid
1051 TRACE_IRQS_OFF
1052 movq %rsp,%rdi /* pt_regs pointer */
1053 xorl %esi,%esi /* no error code */
1054 call \do_sym
1055 jmp paranoid_exit /* %ebx: no swapgs flag */
1056 CFI_ENDPROC
1057 END(\sym)
1058 .endm
1060 .macro paranoidzeroentry_ist sym do_sym ist
1061 ENTRY(\sym)
1062 INTR_FRAME
1063 PARAVIRT_ADJUST_EXCEPTION_FRAME
1064 pushq $-1 /* ORIG_RAX: no syscall to restart */
1065 CFI_ADJUST_CFA_OFFSET 8
1066 subq $15*8, %rsp
1067 call save_paranoid
1068 TRACE_IRQS_OFF
1069 movq %rsp,%rdi /* pt_regs pointer */
1070 xorl %esi,%esi /* no error code */
1071 PER_CPU(init_tss, %rbp)
1072 subq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1073 call \do_sym
1074 addq $EXCEPTION_STKSZ, TSS_ist + (\ist - 1) * 8(%rbp)
1075 jmp paranoid_exit /* %ebx: no swapgs flag */
1076 CFI_ENDPROC
1077 END(\sym)
1078 .endm
1080 .macro errorentry sym do_sym
1081 ENTRY(\sym)
1082 XCPT_FRAME
1083 PARAVIRT_ADJUST_EXCEPTION_FRAME
1084 subq $15*8,%rsp
1085 CFI_ADJUST_CFA_OFFSET 15*8
1086 call error_entry
1087 DEFAULT_FRAME 0
1088 movq %rsp,%rdi /* pt_regs pointer */
1089 movq ORIG_RAX(%rsp),%rsi /* get error code */
1090 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1091 call \do_sym
1092 jmp error_exit /* %ebx: no swapgs flag */
1093 CFI_ENDPROC
1094 END(\sym)
1095 .endm
1097 /* error code is on the stack already */
1098 .macro paranoiderrorentry sym do_sym
1099 ENTRY(\sym)
1100 XCPT_FRAME
1101 PARAVIRT_ADJUST_EXCEPTION_FRAME
1102 subq $15*8,%rsp
1103 CFI_ADJUST_CFA_OFFSET 15*8
1104 call save_paranoid
1105 DEFAULT_FRAME 0
1106 TRACE_IRQS_OFF
1107 movq %rsp,%rdi /* pt_regs pointer */
1108 movq ORIG_RAX(%rsp),%rsi /* get error code */
1109 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1110 call \do_sym
1111 jmp paranoid_exit /* %ebx: no swapgs flag */
1112 CFI_ENDPROC
1113 END(\sym)
1114 .endm
1116 zeroentry divide_error do_divide_error
1117 zeroentry overflow do_overflow
1118 zeroentry bounds do_bounds
1119 zeroentry invalid_op do_invalid_op
1120 zeroentry device_not_available do_device_not_available
1121 paranoiderrorentry double_fault do_double_fault
1122 zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1123 errorentry invalid_TSS do_invalid_TSS
1124 errorentry segment_not_present do_segment_not_present
1125 zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1126 zeroentry coprocessor_error do_coprocessor_error
1127 errorentry alignment_check do_alignment_check
1128 zeroentry simd_coprocessor_error do_simd_coprocessor_error
1130 /* Reload gs selector with exception handling */
1131 /* edi: new selector */
1132 ENTRY(native_load_gs_index)
1133 CFI_STARTPROC
1134 pushf
1135 CFI_ADJUST_CFA_OFFSET 8
1136 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1137 SWAPGS
1138 gs_change:
1139 movl %edi,%gs
1140 2: mfence /* workaround */
1141 SWAPGS
1142 popf
1143 CFI_ADJUST_CFA_OFFSET -8
1144 ret
1145 CFI_ENDPROC
1146 END(native_load_gs_index)
1148 .section __ex_table,"a"
1149 .align 8
1150 .quad gs_change,bad_gs
1151 .previous
1152 .section .fixup,"ax"
1153 /* running with kernelgs */
1154 bad_gs:
1155 SWAPGS /* switch back to user gs */
1156 xorl %eax,%eax
1157 movl %eax,%gs
1158 jmp 2b
1159 .previous
1161 /*
1162 * Create a kernel thread.
1163 *
1164 * C extern interface:
1165 * extern long kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
1166 *
1167 * asm input arguments:
1168 * rdi: fn, rsi: arg, rdx: flags
1169 */
1170 ENTRY(kernel_thread)
1171 CFI_STARTPROC
1172 FAKE_STACK_FRAME $child_rip
1173 SAVE_ALL
1175 # rdi: flags, rsi: usp, rdx: will be &pt_regs
1176 movq %rdx,%rdi
1177 orq kernel_thread_flags(%rip),%rdi
1178 movq $-1, %rsi
1179 movq %rsp, %rdx
1181 xorl %r8d,%r8d
1182 xorl %r9d,%r9d
1184 # clone now
1185 call do_fork
1186 movq %rax,RAX(%rsp)
1187 xorl %edi,%edi
1189 /*
1190 * It isn't worth to check for reschedule here,
1191 * so internally to the x86_64 port you can rely on kernel_thread()
1192 * not to reschedule the child before returning, this avoids the need
1193 * of hacks for example to fork off the per-CPU idle tasks.
1194 * [Hopefully no generic code relies on the reschedule -AK]
1195 */
1196 RESTORE_ALL
1197 UNFAKE_STACK_FRAME
1198 ret
1199 CFI_ENDPROC
1200 END(kernel_thread)
1202 ENTRY(child_rip)
1203 pushq $0 # fake return address
1204 CFI_STARTPROC
1205 /*
1206 * Here we are in the child and the registers are set as they were
1207 * at kernel_thread() invocation in the parent.
1208 */
1209 movq %rdi, %rax
1210 movq %rsi, %rdi
1211 call *%rax
1212 # exit
1213 mov %eax, %edi
1214 call do_exit
1215 ud2 # padding for call trace
1216 CFI_ENDPROC
1217 END(child_rip)
1219 /*
1220 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1221 *
1222 * C extern interface:
1223 * extern long execve(char *name, char **argv, char **envp)
1224 *
1225 * asm input arguments:
1226 * rdi: name, rsi: argv, rdx: envp
1227 *
1228 * We want to fallback into:
1229 * extern long sys_execve(char *name, char **argv,char **envp, struct pt_regs *regs)
1230 *
1231 * do_sys_execve asm fallback arguments:
1232 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1233 */
1234 ENTRY(kernel_execve)
1235 CFI_STARTPROC
1236 FAKE_STACK_FRAME $0
1237 SAVE_ALL
1238 movq %rsp,%rcx
1239 call sys_execve
1240 movq %rax, RAX(%rsp)
1241 RESTORE_REST
1242 testq %rax,%rax
1243 je int_ret_from_sys_call
1244 RESTORE_ARGS
1245 UNFAKE_STACK_FRAME
1246 ret
1247 CFI_ENDPROC
1248 END(kernel_execve)
1250 /* Call softirq on interrupt stack. Interrupts are off. */
1251 ENTRY(call_softirq)
1252 CFI_STARTPROC
1253 push %rbp
1254 CFI_ADJUST_CFA_OFFSET 8
1255 CFI_REL_OFFSET rbp,0
1256 mov %rsp,%rbp
1257 CFI_DEF_CFA_REGISTER rbp
1258 incl PER_CPU_VAR(irq_count)
1259 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1260 push %rbp # backlink for old unwinder
1261 call __do_softirq
1262 leaveq
1263 CFI_DEF_CFA_REGISTER rsp
1264 CFI_ADJUST_CFA_OFFSET -8
1265 decl PER_CPU_VAR(irq_count)
1266 ret
1267 CFI_ENDPROC
1268 END(call_softirq)
1270 #ifdef CONFIG_XEN
1271 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1273 /*
1274 * A note on the "critical region" in our callback handler.
1275 * We want to avoid stacking callback handlers due to events occurring
1276 * during handling of the last event. To do this, we keep events disabled
1277 * until we've done all processing. HOWEVER, we must enable events before
1278 * popping the stack frame (can't be done atomically) and so it would still
1279 * be possible to get enough handler activations to overflow the stack.
1280 * Although unlikely, bugs of that kind are hard to track down, so we'd
1281 * like to avoid the possibility.
1282 * So, on entry to the handler we detect whether we interrupted an
1283 * existing activation in its critical region -- if so, we pop the current
1284 * activation and restart the handler using the previous one.
1285 */
1286 ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1287 CFI_STARTPROC
1288 /*
1289 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1290 * see the correct pointer to the pt_regs
1291 */
1292 movq %rdi, %rsp # we don't return, adjust the stack frame
1293 CFI_ENDPROC
1294 DEFAULT_FRAME
1295 11: incl PER_CPU_VAR(irq_count)
1296 movq %rsp,%rbp
1297 CFI_DEF_CFA_REGISTER rbp
1298 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1299 pushq %rbp # backlink for old unwinder
1300 call xen_evtchn_do_upcall
1301 popq %rsp
1302 CFI_DEF_CFA_REGISTER rsp
1303 decl PER_CPU_VAR(irq_count)
1304 jmp error_exit
1305 CFI_ENDPROC
1306 END(do_hypervisor_callback)
1308 /*
1309 * Hypervisor uses this for application faults while it executes.
1310 * We get here for two reasons:
1311 * 1. Fault while reloading DS, ES, FS or GS
1312 * 2. Fault while executing IRET
1313 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1314 * registers that could be reloaded and zeroed the others.
1315 * Category 2 we fix up by killing the current process. We cannot use the
1316 * normal Linux return path in this case because if we use the IRET hypercall
1317 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1318 * We distinguish between categories by comparing each saved segment register
1319 * with its current contents: any discrepancy means we in category 1.
1320 */
1321 ENTRY(xen_failsafe_callback)
1322 INTR_FRAME 1 (6*8)
1323 /*CFI_REL_OFFSET gs,GS*/
1324 /*CFI_REL_OFFSET fs,FS*/
1325 /*CFI_REL_OFFSET es,ES*/
1326 /*CFI_REL_OFFSET ds,DS*/
1327 CFI_REL_OFFSET r11,8
1328 CFI_REL_OFFSET rcx,0
1329 movw %ds,%cx
1330 cmpw %cx,0x10(%rsp)
1331 CFI_REMEMBER_STATE
1332 jne 1f
1333 movw %es,%cx
1334 cmpw %cx,0x18(%rsp)
1335 jne 1f
1336 movw %fs,%cx
1337 cmpw %cx,0x20(%rsp)
1338 jne 1f
1339 movw %gs,%cx
1340 cmpw %cx,0x28(%rsp)
1341 jne 1f
1342 /* All segments match their saved values => Category 2 (Bad IRET). */
1343 movq (%rsp),%rcx
1344 CFI_RESTORE rcx
1345 movq 8(%rsp),%r11
1346 CFI_RESTORE r11
1347 addq $0x30,%rsp
1348 CFI_ADJUST_CFA_OFFSET -0x30
1349 pushq_cfi $0 /* RIP */
1350 pushq_cfi %r11
1351 pushq_cfi %rcx
1352 jmp general_protection
1353 CFI_RESTORE_STATE
1354 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1355 movq (%rsp),%rcx
1356 CFI_RESTORE rcx
1357 movq 8(%rsp),%r11
1358 CFI_RESTORE r11
1359 addq $0x30,%rsp
1360 CFI_ADJUST_CFA_OFFSET -0x30
1361 pushq_cfi $0
1362 SAVE_ALL
1363 jmp error_exit
1364 CFI_ENDPROC
1365 END(xen_failsafe_callback)
1367 #endif /* CONFIG_XEN */
1369 /*
1370 * Some functions should be protected against kprobes
1371 */
1372 .pushsection .kprobes.text, "ax"
1374 paranoidzeroentry_ist debug do_debug DEBUG_STACK
1375 paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1376 paranoiderrorentry stack_segment do_stack_segment
1377 #ifdef CONFIG_XEN
1378 zeroentry xen_debug do_debug
1379 zeroentry xen_int3 do_int3
1380 errorentry xen_stack_segment do_stack_segment
1381 #endif
1382 errorentry general_protection do_general_protection
1383 errorentry page_fault do_page_fault
1384 #ifdef CONFIG_X86_MCE
1385 paranoidzeroentry machine_check *machine_check_vector(%rip)
1386 #endif
1388 /*
1389 * "Paranoid" exit path from exception stack.
1390 * Paranoid because this is used by NMIs and cannot take
1391 * any kernel state for granted.
1392 * We don't do kernel preemption checks here, because only
1393 * NMI should be common and it does not enable IRQs and
1394 * cannot get reschedule ticks.
1395 *
1396 * "trace" is 0 for the NMI handler only, because irq-tracing
1397 * is fundamentally NMI-unsafe. (we cannot change the soft and
1398 * hard flags at once, atomically)
1399 */
1401 /* ebx: no swapgs flag */
1402 ENTRY(paranoid_exit)
1403 INTR_FRAME
1404 DISABLE_INTERRUPTS(CLBR_NONE)
1405 TRACE_IRQS_OFF
1406 testl %ebx,%ebx /* swapgs needed? */
1407 jnz paranoid_restore
1408 testl $3,CS(%rsp)
1409 jnz paranoid_userspace
1410 paranoid_swapgs:
1411 TRACE_IRQS_IRETQ 0
1412 SWAPGS_UNSAFE_STACK
1413 RESTORE_ALL 8
1414 jmp irq_return
1415 paranoid_restore:
1416 TRACE_IRQS_IRETQ 0
1417 RESTORE_ALL 8
1418 jmp irq_return
1419 paranoid_userspace:
1420 GET_THREAD_INFO(%rcx)
1421 movl TI_flags(%rcx),%ebx
1422 andl $_TIF_WORK_MASK,%ebx
1423 jz paranoid_swapgs
1424 movq %rsp,%rdi /* &pt_regs */
1425 call sync_regs
1426 movq %rax,%rsp /* switch stack for scheduling */
1427 testl $_TIF_NEED_RESCHED,%ebx
1428 jnz paranoid_schedule
1429 movl %ebx,%edx /* arg3: thread flags */
1430 TRACE_IRQS_ON
1431 ENABLE_INTERRUPTS(CLBR_NONE)
1432 xorl %esi,%esi /* arg2: oldset */
1433 movq %rsp,%rdi /* arg1: &pt_regs */
1434 call do_notify_resume
1435 DISABLE_INTERRUPTS(CLBR_NONE)
1436 TRACE_IRQS_OFF
1437 jmp paranoid_userspace
1438 paranoid_schedule:
1439 TRACE_IRQS_ON
1440 ENABLE_INTERRUPTS(CLBR_ANY)
1441 call schedule
1442 DISABLE_INTERRUPTS(CLBR_ANY)
1443 TRACE_IRQS_OFF
1444 jmp paranoid_userspace
1445 CFI_ENDPROC
1446 END(paranoid_exit)
1448 /*
1449 * Exception entry point. This expects an error code/orig_rax on the stack.
1450 * returns in "no swapgs flag" in %ebx.
1451 */
1452 ENTRY(error_entry)
1453 XCPT_FRAME
1454 CFI_ADJUST_CFA_OFFSET 15*8
1455 /* oldrax contains error code */
1456 cld
1457 movq_cfi rdi, RDI+8
1458 movq_cfi rsi, RSI+8
1459 movq_cfi rdx, RDX+8
1460 movq_cfi rcx, RCX+8
1461 movq_cfi rax, RAX+8
1462 movq_cfi r8, R8+8
1463 movq_cfi r9, R9+8
1464 movq_cfi r10, R10+8
1465 movq_cfi r11, R11+8
1466 movq_cfi rbx, RBX+8
1467 movq_cfi rbp, RBP+8
1468 movq_cfi r12, R12+8
1469 movq_cfi r13, R13+8
1470 movq_cfi r14, R14+8
1471 movq_cfi r15, R15+8
1472 xorl %ebx,%ebx
1473 testl $3,CS+8(%rsp)
1474 je error_kernelspace
1475 error_swapgs:
1476 SWAPGS
1477 error_sti:
1478 TRACE_IRQS_OFF
1479 ret
1480 CFI_ENDPROC
1482 /*
1483 * There are two places in the kernel that can potentially fault with
1484 * usergs. Handle them here. The exception handlers after iret run with
1485 * kernel gs again, so don't set the user space flag. B stepping K8s
1486 * sometimes report an truncated RIP for IRET exceptions returning to
1487 * compat mode. Check for these here too.
1488 */
1489 error_kernelspace:
1490 incl %ebx
1491 leaq irq_return(%rip),%rcx
1492 cmpq %rcx,RIP+8(%rsp)
1493 je error_swapgs
1494 movl %ecx,%ecx /* zero extend */
1495 cmpq %rcx,RIP+8(%rsp)
1496 je error_swapgs
1497 cmpq $gs_change,RIP+8(%rsp)
1498 je error_swapgs
1499 jmp error_sti
1500 END(error_entry)
1503 /* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1504 ENTRY(error_exit)
1505 DEFAULT_FRAME
1506 movl %ebx,%eax
1507 RESTORE_REST
1508 DISABLE_INTERRUPTS(CLBR_NONE)
1509 TRACE_IRQS_OFF
1510 GET_THREAD_INFO(%rcx)
1511 testl %eax,%eax
1512 jne retint_kernel
1513 LOCKDEP_SYS_EXIT_IRQ
1514 movl TI_flags(%rcx),%edx
1515 movl $_TIF_WORK_MASK,%edi
1516 andl %edi,%edx
1517 jnz retint_careful
1518 jmp retint_swapgs
1519 CFI_ENDPROC
1520 END(error_exit)
1523 /* runs on exception stack */
1524 ENTRY(nmi)
1525 INTR_FRAME
1526 PARAVIRT_ADJUST_EXCEPTION_FRAME
1527 pushq_cfi $-1
1528 subq $15*8, %rsp
1529 CFI_ADJUST_CFA_OFFSET 15*8
1530 call save_paranoid
1531 DEFAULT_FRAME 0
1532 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1533 movq %rsp,%rdi
1534 movq $-1,%rsi
1535 call do_nmi
1536 #ifdef CONFIG_TRACE_IRQFLAGS
1537 /* paranoidexit; without TRACE_IRQS_OFF */
1538 /* ebx: no swapgs flag */
1539 DISABLE_INTERRUPTS(CLBR_NONE)
1540 testl %ebx,%ebx /* swapgs needed? */
1541 jnz nmi_restore
1542 testl $3,CS(%rsp)
1543 jnz nmi_userspace
1544 nmi_swapgs:
1545 SWAPGS_UNSAFE_STACK
1546 nmi_restore:
1547 RESTORE_ALL 8
1548 jmp irq_return
1549 nmi_userspace:
1550 GET_THREAD_INFO(%rcx)
1551 movl TI_flags(%rcx),%ebx
1552 andl $_TIF_WORK_MASK,%ebx
1553 jz nmi_swapgs
1554 movq %rsp,%rdi /* &pt_regs */
1555 call sync_regs
1556 movq %rax,%rsp /* switch stack for scheduling */
1557 testl $_TIF_NEED_RESCHED,%ebx
1558 jnz nmi_schedule
1559 movl %ebx,%edx /* arg3: thread flags */
1560 ENABLE_INTERRUPTS(CLBR_NONE)
1561 xorl %esi,%esi /* arg2: oldset */
1562 movq %rsp,%rdi /* arg1: &pt_regs */
1563 call do_notify_resume
1564 DISABLE_INTERRUPTS(CLBR_NONE)
1565 jmp nmi_userspace
1566 nmi_schedule:
1567 ENABLE_INTERRUPTS(CLBR_ANY)
1568 call schedule
1569 DISABLE_INTERRUPTS(CLBR_ANY)
1570 jmp nmi_userspace
1571 CFI_ENDPROC
1572 #else
1573 jmp paranoid_exit
1574 CFI_ENDPROC
1575 #endif
1576 END(nmi)
1578 ENTRY(ignore_sysret)
1579 CFI_STARTPROC
1580 mov $-ENOSYS,%eax
1581 sysret
1582 CFI_ENDPROC
1583 END(ignore_sysret)
1585 /*
1586 * End of kprobes section
1587 */
1588 .popsection