Loading...
Note: File does not exist in v5.9.
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 */
8
9/*
10 * entry.S contains the system-call and fault low-level handling routines.
11 *
12 * Some of this is documented in Documentation/x86/entry_64.txt
13 *
14 * NOTE: This code handles signal-recognition, which happens every time
15 * after an interrupt and after each system call.
16 *
17 * Normal syscalls and interrupts don't save a full stack frame, this is
18 * only done for syscall tracing, signals or fork/exec et.al.
19 *
20 * A note on terminology:
21 * - top of stack: Architecture defined interrupt frame from SS to RIP
22 * at the top of the kernel process stack.
23 * - partial stack frame: partially saved registers up to R11.
24 * - full stack frame: Like partial stack frame, but all register saved.
25 *
26 * Some macro usage:
27 * - CFI macros are used to generate dwarf2 unwind information for better
28 * backtraces. They don't change any code.
29 * - SAVE_ALL/RESTORE_ALL - Save/restore all registers
30 * - SAVE_ARGS/RESTORE_ARGS - Save/restore registers that C functions modify.
31 * There are unfortunately lots of special cases where some registers
32 * not touched. The macro is a big mess that should be cleaned up.
33 * - SAVE_REST/RESTORE_REST - Handle the registers not saved by SAVE_ARGS.
34 * Gives a full stack frame.
35 * - ENTRY/END Define functions in the symbol table.
36 * - FIXUP_TOP_OF_STACK/RESTORE_TOP_OF_STACK - Fix up the hardware stack
37 * frame that is otherwise undefined after a SYSCALL
38 * - TRACE_IRQ_* - Trace hard interrupt state for lock debugging.
39 * - errorentry/paranoidentry/zeroentry - Define exception entry points.
40 */
41
42#include <linux/linkage.h>
43#include <asm/segment.h>
44#include <asm/cache.h>
45#include <asm/errno.h>
46#include <asm/dwarf2.h>
47#include <asm/calling.h>
48#include <asm/asm-offsets.h>
49#include <asm/msr.h>
50#include <asm/unistd.h>
51#include <asm/thread_info.h>
52#include <asm/hw_irq.h>
53#include <asm/page_types.h>
54#include <asm/irqflags.h>
55#include <asm/paravirt.h>
56#include <asm/ftrace.h>
57#include <asm/percpu.h>
58#include <asm/asm.h>
59#include <linux/err.h>
60
61/* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
62#include <linux/elf-em.h>
63#define AUDIT_ARCH_X86_64 (EM_X86_64|__AUDIT_ARCH_64BIT|__AUDIT_ARCH_LE)
64#define __AUDIT_ARCH_64BIT 0x80000000
65#define __AUDIT_ARCH_LE 0x40000000
66
67 .code64
68 .section .entry.text, "ax"
69
70#ifdef CONFIG_FUNCTION_TRACER
71#ifdef CONFIG_DYNAMIC_FTRACE
72ENTRY(mcount)
73 retq
74END(mcount)
75
76ENTRY(ftrace_caller)
77 cmpl $0, function_trace_stop
78 jne ftrace_stub
79
80 MCOUNT_SAVE_FRAME
81
82 movq 0x38(%rsp), %rdi
83 movq 8(%rbp), %rsi
84 subq $MCOUNT_INSN_SIZE, %rdi
85
86GLOBAL(ftrace_call)
87 call ftrace_stub
88
89 MCOUNT_RESTORE_FRAME
90
91#ifdef CONFIG_FUNCTION_GRAPH_TRACER
92GLOBAL(ftrace_graph_call)
93 jmp ftrace_stub
94#endif
95
96GLOBAL(ftrace_stub)
97 retq
98END(ftrace_caller)
99
100#else /* ! CONFIG_DYNAMIC_FTRACE */
101ENTRY(mcount)
102 cmpl $0, function_trace_stop
103 jne ftrace_stub
104
105 cmpq $ftrace_stub, ftrace_trace_function
106 jnz trace
107
108#ifdef CONFIG_FUNCTION_GRAPH_TRACER
109 cmpq $ftrace_stub, ftrace_graph_return
110 jnz ftrace_graph_caller
111
112 cmpq $ftrace_graph_entry_stub, ftrace_graph_entry
113 jnz ftrace_graph_caller
114#endif
115
116GLOBAL(ftrace_stub)
117 retq
118
119trace:
120 MCOUNT_SAVE_FRAME
121
122 movq 0x38(%rsp), %rdi
123 movq 8(%rbp), %rsi
124 subq $MCOUNT_INSN_SIZE, %rdi
125
126 call *ftrace_trace_function
127
128 MCOUNT_RESTORE_FRAME
129
130 jmp ftrace_stub
131END(mcount)
132#endif /* CONFIG_DYNAMIC_FTRACE */
133#endif /* CONFIG_FUNCTION_TRACER */
134
135#ifdef CONFIG_FUNCTION_GRAPH_TRACER
136ENTRY(ftrace_graph_caller)
137 cmpl $0, function_trace_stop
138 jne ftrace_stub
139
140 MCOUNT_SAVE_FRAME
141
142 leaq 8(%rbp), %rdi
143 movq 0x38(%rsp), %rsi
144 movq (%rbp), %rdx
145 subq $MCOUNT_INSN_SIZE, %rsi
146
147 call prepare_ftrace_return
148
149 MCOUNT_RESTORE_FRAME
150
151 retq
152END(ftrace_graph_caller)
153
154GLOBAL(return_to_handler)
155 subq $24, %rsp
156
157 /* Save the return values */
158 movq %rax, (%rsp)
159 movq %rdx, 8(%rsp)
160 movq %rbp, %rdi
161
162 call ftrace_return_to_handler
163
164 movq %rax, %rdi
165 movq 8(%rsp), %rdx
166 movq (%rsp), %rax
167 addq $24, %rsp
168 jmp *%rdi
169#endif
170
171
172#ifndef CONFIG_PREEMPT
173#define retint_kernel retint_restore_args
174#endif
175
176#ifdef CONFIG_PARAVIRT
177ENTRY(native_usergs_sysret64)
178 swapgs
179 sysretq
180ENDPROC(native_usergs_sysret64)
181#endif /* CONFIG_PARAVIRT */
182
183
184.macro TRACE_IRQS_IRETQ offset=ARGOFFSET
185#ifdef CONFIG_TRACE_IRQFLAGS
186 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
187 jnc 1f
188 TRACE_IRQS_ON
1891:
190#endif
191.endm
192
193/*
194 * When dynamic function tracer is enabled it will add a breakpoint
195 * to all locations that it is about to modify, sync CPUs, update
196 * all the code, sync CPUs, then remove the breakpoints. In this time
197 * if lockdep is enabled, it might jump back into the debug handler
198 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
199 *
200 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
201 * make sure the stack pointer does not get reset back to the top
202 * of the debug stack, and instead just reuses the current stack.
203 */
204#if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
205
206.macro TRACE_IRQS_OFF_DEBUG
207 call debug_stack_set_zero
208 TRACE_IRQS_OFF
209 call debug_stack_reset
210.endm
211
212.macro TRACE_IRQS_ON_DEBUG
213 call debug_stack_set_zero
214 TRACE_IRQS_ON
215 call debug_stack_reset
216.endm
217
218.macro TRACE_IRQS_IRETQ_DEBUG offset=ARGOFFSET
219 bt $9,EFLAGS-\offset(%rsp) /* interrupts off? */
220 jnc 1f
221 TRACE_IRQS_ON_DEBUG
2221:
223.endm
224
225#else
226# define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
227# define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
228# define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
229#endif
230
231/*
232 * C code is not supposed to know about undefined top of stack. Every time
233 * a C function with an pt_regs argument is called from the SYSCALL based
234 * fast path FIXUP_TOP_OF_STACK is needed.
235 * RESTORE_TOP_OF_STACK syncs the syscall state after any possible ptregs
236 * manipulation.
237 */
238
239 /* %rsp:at FRAMEEND */
240 .macro FIXUP_TOP_OF_STACK tmp offset=0
241 movq PER_CPU_VAR(old_rsp),\tmp
242 movq \tmp,RSP+\offset(%rsp)
243 movq $__USER_DS,SS+\offset(%rsp)
244 movq $__USER_CS,CS+\offset(%rsp)
245 movq $-1,RCX+\offset(%rsp)
246 movq R11+\offset(%rsp),\tmp /* get eflags */
247 movq \tmp,EFLAGS+\offset(%rsp)
248 .endm
249
250 .macro RESTORE_TOP_OF_STACK tmp offset=0
251 movq RSP+\offset(%rsp),\tmp
252 movq \tmp,PER_CPU_VAR(old_rsp)
253 movq EFLAGS+\offset(%rsp),\tmp
254 movq \tmp,R11+\offset(%rsp)
255 .endm
256
257 .macro FAKE_STACK_FRAME child_rip
258 /* push in order ss, rsp, eflags, cs, rip */
259 xorl %eax, %eax
260 pushq_cfi $__KERNEL_DS /* ss */
261 /*CFI_REL_OFFSET ss,0*/
262 pushq_cfi %rax /* rsp */
263 CFI_REL_OFFSET rsp,0
264 pushq_cfi $(X86_EFLAGS_IF|X86_EFLAGS_BIT1) /* eflags - interrupts on */
265 /*CFI_REL_OFFSET rflags,0*/
266 pushq_cfi $__KERNEL_CS /* cs */
267 /*CFI_REL_OFFSET cs,0*/
268 pushq_cfi \child_rip /* rip */
269 CFI_REL_OFFSET rip,0
270 pushq_cfi %rax /* orig rax */
271 .endm
272
273 .macro UNFAKE_STACK_FRAME
274 addq $8*6, %rsp
275 CFI_ADJUST_CFA_OFFSET -(6*8)
276 .endm
277
278/*
279 * initial frame state for interrupts (and exceptions without error code)
280 */
281 .macro EMPTY_FRAME start=1 offset=0
282 .if \start
283 CFI_STARTPROC simple
284 CFI_SIGNAL_FRAME
285 CFI_DEF_CFA rsp,8+\offset
286 .else
287 CFI_DEF_CFA_OFFSET 8+\offset
288 .endif
289 .endm
290
291/*
292 * initial frame state for interrupts (and exceptions without error code)
293 */
294 .macro INTR_FRAME start=1 offset=0
295 EMPTY_FRAME \start, SS+8+\offset-RIP
296 /*CFI_REL_OFFSET ss, SS+\offset-RIP*/
297 CFI_REL_OFFSET rsp, RSP+\offset-RIP
298 /*CFI_REL_OFFSET rflags, EFLAGS+\offset-RIP*/
299 /*CFI_REL_OFFSET cs, CS+\offset-RIP*/
300 CFI_REL_OFFSET rip, RIP+\offset-RIP
301 .endm
302
303/*
304 * initial frame state for exceptions with error code (and interrupts
305 * with vector already pushed)
306 */
307 .macro XCPT_FRAME start=1 offset=0
308 INTR_FRAME \start, RIP+\offset-ORIG_RAX
309 /*CFI_REL_OFFSET orig_rax, ORIG_RAX-ORIG_RAX*/
310 .endm
311
312/*
313 * frame that enables calling into C.
314 */
315 .macro PARTIAL_FRAME start=1 offset=0
316 XCPT_FRAME \start, ORIG_RAX+\offset-ARGOFFSET
317 CFI_REL_OFFSET rdi, RDI+\offset-ARGOFFSET
318 CFI_REL_OFFSET rsi, RSI+\offset-ARGOFFSET
319 CFI_REL_OFFSET rdx, RDX+\offset-ARGOFFSET
320 CFI_REL_OFFSET rcx, RCX+\offset-ARGOFFSET
321 CFI_REL_OFFSET rax, RAX+\offset-ARGOFFSET
322 CFI_REL_OFFSET r8, R8+\offset-ARGOFFSET
323 CFI_REL_OFFSET r9, R9+\offset-ARGOFFSET
324 CFI_REL_OFFSET r10, R10+\offset-ARGOFFSET
325 CFI_REL_OFFSET r11, R11+\offset-ARGOFFSET
326 .endm
327
328/*
329 * frame that enables passing a complete pt_regs to a C function.
330 */
331 .macro DEFAULT_FRAME start=1 offset=0
332 PARTIAL_FRAME \start, R11+\offset-R15
333 CFI_REL_OFFSET rbx, RBX+\offset
334 CFI_REL_OFFSET rbp, RBP+\offset
335 CFI_REL_OFFSET r12, R12+\offset
336 CFI_REL_OFFSET r13, R13+\offset
337 CFI_REL_OFFSET r14, R14+\offset
338 CFI_REL_OFFSET r15, R15+\offset
339 .endm
340
341/* save partial stack frame */
342 .macro SAVE_ARGS_IRQ
343 cld
344 /* start from rbp in pt_regs and jump over */
345 movq_cfi rdi, RDI-RBP
346 movq_cfi rsi, RSI-RBP
347 movq_cfi rdx, RDX-RBP
348 movq_cfi rcx, RCX-RBP
349 movq_cfi rax, RAX-RBP
350 movq_cfi r8, R8-RBP
351 movq_cfi r9, R9-RBP
352 movq_cfi r10, R10-RBP
353 movq_cfi r11, R11-RBP
354
355 /* Save rbp so that we can unwind from get_irq_regs() */
356 movq_cfi rbp, 0
357
358 /* Save previous stack value */
359 movq %rsp, %rsi
360
361 leaq -RBP(%rsp),%rdi /* arg1 for handler */
362 testl $3, CS-RBP(%rsi)
363 je 1f
364 SWAPGS
365 /*
366 * irq_count is used to check if a CPU is already on an interrupt stack
367 * or not. While this is essentially redundant with preempt_count it is
368 * a little cheaper to use a separate counter in the PDA (short of
369 * moving irq_enter into assembly, which would be too much work)
370 */
3711: incl PER_CPU_VAR(irq_count)
372 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
373 CFI_DEF_CFA_REGISTER rsi
374
375 /* Store previous stack value */
376 pushq %rsi
377 CFI_ESCAPE 0x0f /* DW_CFA_def_cfa_expression */, 6, \
378 0x77 /* DW_OP_breg7 */, 0, \
379 0x06 /* DW_OP_deref */, \
380 0x08 /* DW_OP_const1u */, SS+8-RBP, \
381 0x22 /* DW_OP_plus */
382 /* We entered an interrupt context - irqs are off: */
383 TRACE_IRQS_OFF
384 .endm
385
386ENTRY(save_rest)
387 PARTIAL_FRAME 1 REST_SKIP+8
388 movq 5*8+16(%rsp), %r11 /* save return address */
389 movq_cfi rbx, RBX+16
390 movq_cfi rbp, RBP+16
391 movq_cfi r12, R12+16
392 movq_cfi r13, R13+16
393 movq_cfi r14, R14+16
394 movq_cfi r15, R15+16
395 movq %r11, 8(%rsp) /* return address */
396 FIXUP_TOP_OF_STACK %r11, 16
397 ret
398 CFI_ENDPROC
399END(save_rest)
400
401/* save complete stack frame */
402 .pushsection .kprobes.text, "ax"
403ENTRY(save_paranoid)
404 XCPT_FRAME 1 RDI+8
405 cld
406 movq_cfi rdi, RDI+8
407 movq_cfi rsi, RSI+8
408 movq_cfi rdx, RDX+8
409 movq_cfi rcx, RCX+8
410 movq_cfi rax, RAX+8
411 movq_cfi r8, R8+8
412 movq_cfi r9, R9+8
413 movq_cfi r10, R10+8
414 movq_cfi r11, R11+8
415 movq_cfi rbx, RBX+8
416 movq_cfi rbp, RBP+8
417 movq_cfi r12, R12+8
418 movq_cfi r13, R13+8
419 movq_cfi r14, R14+8
420 movq_cfi r15, R15+8
421 movl $1,%ebx
422 movl $MSR_GS_BASE,%ecx
423 rdmsr
424 testl %edx,%edx
425 js 1f /* negative -> in kernel */
426 SWAPGS
427 xorl %ebx,%ebx
4281: ret
429 CFI_ENDPROC
430END(save_paranoid)
431 .popsection
432
433/*
434 * A newly forked process directly context switches into this address.
435 *
436 * rdi: prev task we switched from
437 */
438ENTRY(ret_from_fork)
439 DEFAULT_FRAME
440
441 LOCK ; btr $TIF_FORK,TI_flags(%r8)
442
443 pushq_cfi kernel_eflags(%rip)
444 popfq_cfi # reset kernel eflags
445
446 call schedule_tail # rdi: 'prev' task parameter
447
448 GET_THREAD_INFO(%rcx)
449
450 RESTORE_REST
451
452 testl $3, CS-ARGOFFSET(%rsp) # from kernel_thread?
453 jz retint_restore_args
454
455 testl $_TIF_IA32, TI_flags(%rcx) # 32-bit compat task needs IRET
456 jnz int_ret_from_sys_call
457
458 RESTORE_TOP_OF_STACK %rdi, -ARGOFFSET
459 jmp ret_from_sys_call # go to the SYSRET fastpath
460
461 CFI_ENDPROC
462END(ret_from_fork)
463
464/*
465 * System call entry. Up to 6 arguments in registers are supported.
466 *
467 * SYSCALL does not save anything on the stack and does not change the
468 * stack pointer.
469 */
470
471/*
472 * Register setup:
473 * rax system call number
474 * rdi arg0
475 * rcx return address for syscall/sysret, C arg3
476 * rsi arg1
477 * rdx arg2
478 * r10 arg3 (--> moved to rcx for C)
479 * r8 arg4
480 * r9 arg5
481 * r11 eflags for syscall/sysret, temporary for C
482 * r12-r15,rbp,rbx saved by C code, not touched.
483 *
484 * Interrupts are off on entry.
485 * Only called from user space.
486 *
487 * XXX if we had a free scratch register we could save the RSP into the stack frame
488 * and report it properly in ps. Unfortunately we haven't.
489 *
490 * When user can change the frames always force IRET. That is because
491 * it deals with uncanonical addresses better. SYSRET has trouble
492 * with them due to bugs in both AMD and Intel CPUs.
493 */
494
495ENTRY(system_call)
496 CFI_STARTPROC simple
497 CFI_SIGNAL_FRAME
498 CFI_DEF_CFA rsp,KERNEL_STACK_OFFSET
499 CFI_REGISTER rip,rcx
500 /*CFI_REGISTER rflags,r11*/
501 SWAPGS_UNSAFE_STACK
502 /*
503 * A hypervisor implementation might want to use a label
504 * after the swapgs, so that it can do the swapgs
505 * for the guest and jump here on syscall.
506 */
507GLOBAL(system_call_after_swapgs)
508
509 movq %rsp,PER_CPU_VAR(old_rsp)
510 movq PER_CPU_VAR(kernel_stack),%rsp
511 /*
512 * No need to follow this irqs off/on section - it's straight
513 * and short:
514 */
515 ENABLE_INTERRUPTS(CLBR_NONE)
516 SAVE_ARGS 8,0
517 movq %rax,ORIG_RAX-ARGOFFSET(%rsp)
518 movq %rcx,RIP-ARGOFFSET(%rsp)
519 CFI_REL_OFFSET rip,RIP-ARGOFFSET
520 testl $_TIF_WORK_SYSCALL_ENTRY,TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
521 jnz tracesys
522system_call_fastpath:
523#if __SYSCALL_MASK == ~0
524 cmpq $__NR_syscall_max,%rax
525#else
526 andl $__SYSCALL_MASK,%eax
527 cmpl $__NR_syscall_max,%eax
528#endif
529 ja badsys
530 movq %r10,%rcx
531 call *sys_call_table(,%rax,8) # XXX: rip relative
532 movq %rax,RAX-ARGOFFSET(%rsp)
533/*
534 * Syscall return path ending with SYSRET (fast path)
535 * Has incomplete stack frame and undefined top of stack.
536 */
537ret_from_sys_call:
538 movl $_TIF_ALLWORK_MASK,%edi
539 /* edi: flagmask */
540sysret_check:
541 LOCKDEP_SYS_EXIT
542 DISABLE_INTERRUPTS(CLBR_NONE)
543 TRACE_IRQS_OFF
544 movl TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET),%edx
545 andl %edi,%edx
546 jnz sysret_careful
547 CFI_REMEMBER_STATE
548 /*
549 * sysretq will re-enable interrupts:
550 */
551 TRACE_IRQS_ON
552 movq RIP-ARGOFFSET(%rsp),%rcx
553 CFI_REGISTER rip,rcx
554 RESTORE_ARGS 1,-ARG_SKIP,0
555 /*CFI_REGISTER rflags,r11*/
556 movq PER_CPU_VAR(old_rsp), %rsp
557 USERGS_SYSRET64
558
559 CFI_RESTORE_STATE
560 /* Handle reschedules */
561 /* edx: work, edi: workmask */
562sysret_careful:
563 bt $TIF_NEED_RESCHED,%edx
564 jnc sysret_signal
565 TRACE_IRQS_ON
566 ENABLE_INTERRUPTS(CLBR_NONE)
567 pushq_cfi %rdi
568 call schedule
569 popq_cfi %rdi
570 jmp sysret_check
571
572 /* Handle a signal */
573sysret_signal:
574 TRACE_IRQS_ON
575 ENABLE_INTERRUPTS(CLBR_NONE)
576#ifdef CONFIG_AUDITSYSCALL
577 bt $TIF_SYSCALL_AUDIT,%edx
578 jc sysret_audit
579#endif
580 /*
581 * We have a signal, or exit tracing or single-step.
582 * These all wind up with the iret return path anyway,
583 * so just join that path right now.
584 */
585 FIXUP_TOP_OF_STACK %r11, -ARGOFFSET
586 jmp int_check_syscall_exit_work
587
588badsys:
589 movq $-ENOSYS,RAX-ARGOFFSET(%rsp)
590 jmp ret_from_sys_call
591
592#ifdef CONFIG_AUDITSYSCALL
593 /*
594 * Fast path for syscall audit without full syscall trace.
595 * We just call __audit_syscall_entry() directly, and then
596 * jump back to the normal fast path.
597 */
598auditsys:
599 movq %r10,%r9 /* 6th arg: 4th syscall arg */
600 movq %rdx,%r8 /* 5th arg: 3rd syscall arg */
601 movq %rsi,%rcx /* 4th arg: 2nd syscall arg */
602 movq %rdi,%rdx /* 3rd arg: 1st syscall arg */
603 movq %rax,%rsi /* 2nd arg: syscall number */
604 movl $AUDIT_ARCH_X86_64,%edi /* 1st arg: audit arch */
605 call __audit_syscall_entry
606 LOAD_ARGS 0 /* reload call-clobbered registers */
607 jmp system_call_fastpath
608
609 /*
610 * Return fast path for syscall audit. Call __audit_syscall_exit()
611 * directly and then jump back to the fast path with TIF_SYSCALL_AUDIT
612 * masked off.
613 */
614sysret_audit:
615 movq RAX-ARGOFFSET(%rsp),%rsi /* second arg, syscall return value */
616 cmpq $-MAX_ERRNO,%rsi /* is it < -MAX_ERRNO? */
617 setbe %al /* 1 if so, 0 if not */
618 movzbl %al,%edi /* zero-extend that into %edi */
619 call __audit_syscall_exit
620 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),%edi
621 jmp sysret_check
622#endif /* CONFIG_AUDITSYSCALL */
623
624 /* Do syscall tracing */
625tracesys:
626#ifdef CONFIG_AUDITSYSCALL
627 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT),TI_flags+THREAD_INFO(%rsp,RIP-ARGOFFSET)
628 jz auditsys
629#endif
630 SAVE_REST
631 movq $-ENOSYS,RAX(%rsp) /* ptrace can change this for a bad syscall */
632 FIXUP_TOP_OF_STACK %rdi
633 movq %rsp,%rdi
634 call syscall_trace_enter
635 /*
636 * Reload arg registers from stack in case ptrace changed them.
637 * We don't reload %rax because syscall_trace_enter() returned
638 * the value it wants us to use in the table lookup.
639 */
640 LOAD_ARGS ARGOFFSET, 1
641 RESTORE_REST
642#if __SYSCALL_MASK == ~0
643 cmpq $__NR_syscall_max,%rax
644#else
645 andl $__SYSCALL_MASK,%eax
646 cmpl $__NR_syscall_max,%eax
647#endif
648 ja int_ret_from_sys_call /* RAX(%rsp) set to -ENOSYS above */
649 movq %r10,%rcx /* fixup for C */
650 call *sys_call_table(,%rax,8)
651 movq %rax,RAX-ARGOFFSET(%rsp)
652 /* Use IRET because user could have changed frame */
653
654/*
655 * Syscall return path ending with IRET.
656 * Has correct top of stack, but partial stack frame.
657 */
658GLOBAL(int_ret_from_sys_call)
659 DISABLE_INTERRUPTS(CLBR_NONE)
660 TRACE_IRQS_OFF
661 movl $_TIF_ALLWORK_MASK,%edi
662 /* edi: mask to check */
663GLOBAL(int_with_check)
664 LOCKDEP_SYS_EXIT_IRQ
665 GET_THREAD_INFO(%rcx)
666 movl TI_flags(%rcx),%edx
667 andl %edi,%edx
668 jnz int_careful
669 andl $~TS_COMPAT,TI_status(%rcx)
670 jmp retint_swapgs
671
672 /* Either reschedule or signal or syscall exit tracking needed. */
673 /* First do a reschedule test. */
674 /* edx: work, edi: workmask */
675int_careful:
676 bt $TIF_NEED_RESCHED,%edx
677 jnc int_very_careful
678 TRACE_IRQS_ON
679 ENABLE_INTERRUPTS(CLBR_NONE)
680 pushq_cfi %rdi
681 call schedule
682 popq_cfi %rdi
683 DISABLE_INTERRUPTS(CLBR_NONE)
684 TRACE_IRQS_OFF
685 jmp int_with_check
686
687 /* handle signals and tracing -- both require a full stack frame */
688int_very_careful:
689 TRACE_IRQS_ON
690 ENABLE_INTERRUPTS(CLBR_NONE)
691int_check_syscall_exit_work:
692 SAVE_REST
693 /* Check for syscall exit trace */
694 testl $_TIF_WORK_SYSCALL_EXIT,%edx
695 jz int_signal
696 pushq_cfi %rdi
697 leaq 8(%rsp),%rdi # &ptregs -> arg1
698 call syscall_trace_leave
699 popq_cfi %rdi
700 andl $~(_TIF_WORK_SYSCALL_EXIT|_TIF_SYSCALL_EMU),%edi
701 jmp int_restore_rest
702
703int_signal:
704 testl $_TIF_DO_NOTIFY_MASK,%edx
705 jz 1f
706 movq %rsp,%rdi # &ptregs -> arg1
707 xorl %esi,%esi # oldset -> arg2
708 call do_notify_resume
7091: movl $_TIF_WORK_MASK,%edi
710int_restore_rest:
711 RESTORE_REST
712 DISABLE_INTERRUPTS(CLBR_NONE)
713 TRACE_IRQS_OFF
714 jmp int_with_check
715 CFI_ENDPROC
716END(system_call)
717
718/*
719 * Certain special system calls that need to save a complete full stack frame.
720 */
721 .macro PTREGSCALL label,func,arg
722ENTRY(\label)
723 PARTIAL_FRAME 1 8 /* offset 8: return address */
724 subq $REST_SKIP, %rsp
725 CFI_ADJUST_CFA_OFFSET REST_SKIP
726 call save_rest
727 DEFAULT_FRAME 0 8 /* offset 8: return address */
728 leaq 8(%rsp), \arg /* pt_regs pointer */
729 call \func
730 jmp ptregscall_common
731 CFI_ENDPROC
732END(\label)
733 .endm
734
735 PTREGSCALL stub_clone, sys_clone, %r8
736 PTREGSCALL stub_fork, sys_fork, %rdi
737 PTREGSCALL stub_vfork, sys_vfork, %rdi
738 PTREGSCALL stub_sigaltstack, sys_sigaltstack, %rdx
739 PTREGSCALL stub_iopl, sys_iopl, %rsi
740
741ENTRY(ptregscall_common)
742 DEFAULT_FRAME 1 8 /* offset 8: return address */
743 RESTORE_TOP_OF_STACK %r11, 8
744 movq_cfi_restore R15+8, r15
745 movq_cfi_restore R14+8, r14
746 movq_cfi_restore R13+8, r13
747 movq_cfi_restore R12+8, r12
748 movq_cfi_restore RBP+8, rbp
749 movq_cfi_restore RBX+8, rbx
750 ret $REST_SKIP /* pop extended registers */
751 CFI_ENDPROC
752END(ptregscall_common)
753
754ENTRY(stub_execve)
755 CFI_STARTPROC
756 addq $8, %rsp
757 PARTIAL_FRAME 0
758 SAVE_REST
759 FIXUP_TOP_OF_STACK %r11
760 movq %rsp, %rcx
761 call sys_execve
762 RESTORE_TOP_OF_STACK %r11
763 movq %rax,RAX(%rsp)
764 RESTORE_REST
765 jmp int_ret_from_sys_call
766 CFI_ENDPROC
767END(stub_execve)
768
769/*
770 * sigreturn is special because it needs to restore all registers on return.
771 * This cannot be done with SYSRET, so use the IRET return path instead.
772 */
773ENTRY(stub_rt_sigreturn)
774 CFI_STARTPROC
775 addq $8, %rsp
776 PARTIAL_FRAME 0
777 SAVE_REST
778 movq %rsp,%rdi
779 FIXUP_TOP_OF_STACK %r11
780 call sys_rt_sigreturn
781 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
782 RESTORE_REST
783 jmp int_ret_from_sys_call
784 CFI_ENDPROC
785END(stub_rt_sigreturn)
786
787#ifdef CONFIG_X86_X32_ABI
788 PTREGSCALL stub_x32_sigaltstack, sys32_sigaltstack, %rdx
789
790ENTRY(stub_x32_rt_sigreturn)
791 CFI_STARTPROC
792 addq $8, %rsp
793 PARTIAL_FRAME 0
794 SAVE_REST
795 movq %rsp,%rdi
796 FIXUP_TOP_OF_STACK %r11
797 call sys32_x32_rt_sigreturn
798 movq %rax,RAX(%rsp) # fixme, this could be done at the higher layer
799 RESTORE_REST
800 jmp int_ret_from_sys_call
801 CFI_ENDPROC
802END(stub_x32_rt_sigreturn)
803
804ENTRY(stub_x32_execve)
805 CFI_STARTPROC
806 addq $8, %rsp
807 PARTIAL_FRAME 0
808 SAVE_REST
809 FIXUP_TOP_OF_STACK %r11
810 movq %rsp, %rcx
811 call sys32_execve
812 RESTORE_TOP_OF_STACK %r11
813 movq %rax,RAX(%rsp)
814 RESTORE_REST
815 jmp int_ret_from_sys_call
816 CFI_ENDPROC
817END(stub_x32_execve)
818
819#endif
820
821/*
822 * Build the entry stubs and pointer table with some assembler magic.
823 * We pack 7 stubs into a single 32-byte chunk, which will fit in a
824 * single cache line on all modern x86 implementations.
825 */
826 .section .init.rodata,"a"
827ENTRY(interrupt)
828 .section .entry.text
829 .p2align 5
830 .p2align CONFIG_X86_L1_CACHE_SHIFT
831ENTRY(irq_entries_start)
832 INTR_FRAME
833vector=FIRST_EXTERNAL_VECTOR
834.rept (NR_VECTORS-FIRST_EXTERNAL_VECTOR+6)/7
835 .balign 32
836 .rept 7
837 .if vector < NR_VECTORS
838 .if vector <> FIRST_EXTERNAL_VECTOR
839 CFI_ADJUST_CFA_OFFSET -8
840 .endif
8411: pushq_cfi $(~vector+0x80) /* Note: always in signed byte range */
842 .if ((vector-FIRST_EXTERNAL_VECTOR)%7) <> 6
843 jmp 2f
844 .endif
845 .previous
846 .quad 1b
847 .section .entry.text
848vector=vector+1
849 .endif
850 .endr
8512: jmp common_interrupt
852.endr
853 CFI_ENDPROC
854END(irq_entries_start)
855
856.previous
857END(interrupt)
858.previous
859
860/*
861 * Interrupt entry/exit.
862 *
863 * Interrupt entry points save only callee clobbered registers in fast path.
864 *
865 * Entry runs with interrupts off.
866 */
867
868/* 0(%rsp): ~(interrupt number) */
869 .macro interrupt func
870 /* reserve pt_regs for scratch regs and rbp */
871 subq $ORIG_RAX-RBP, %rsp
872 CFI_ADJUST_CFA_OFFSET ORIG_RAX-RBP
873 SAVE_ARGS_IRQ
874 call \func
875 .endm
876
877/*
878 * Interrupt entry/exit should be protected against kprobes
879 */
880 .pushsection .kprobes.text, "ax"
881 /*
882 * The interrupt stubs push (~vector+0x80) onto the stack and
883 * then jump to common_interrupt.
884 */
885 .p2align CONFIG_X86_L1_CACHE_SHIFT
886common_interrupt:
887 XCPT_FRAME
888 addq $-0x80,(%rsp) /* Adjust vector to [-256,-1] range */
889 interrupt do_IRQ
890 /* 0(%rsp): old_rsp-ARGOFFSET */
891ret_from_intr:
892 DISABLE_INTERRUPTS(CLBR_NONE)
893 TRACE_IRQS_OFF
894 decl PER_CPU_VAR(irq_count)
895
896 /* Restore saved previous stack */
897 popq %rsi
898 CFI_DEF_CFA rsi,SS+8-RBP /* reg/off reset after def_cfa_expr */
899 leaq ARGOFFSET-RBP(%rsi), %rsp
900 CFI_DEF_CFA_REGISTER rsp
901 CFI_ADJUST_CFA_OFFSET RBP-ARGOFFSET
902
903exit_intr:
904 GET_THREAD_INFO(%rcx)
905 testl $3,CS-ARGOFFSET(%rsp)
906 je retint_kernel
907
908 /* Interrupt came from user space */
909 /*
910 * Has a correct top of stack, but a partial stack frame
911 * %rcx: thread info. Interrupts off.
912 */
913retint_with_reschedule:
914 movl $_TIF_WORK_MASK,%edi
915retint_check:
916 LOCKDEP_SYS_EXIT_IRQ
917 movl TI_flags(%rcx),%edx
918 andl %edi,%edx
919 CFI_REMEMBER_STATE
920 jnz retint_careful
921
922retint_swapgs: /* return to user-space */
923 /*
924 * The iretq could re-enable interrupts:
925 */
926 DISABLE_INTERRUPTS(CLBR_ANY)
927 TRACE_IRQS_IRETQ
928 SWAPGS
929 jmp restore_args
930
931retint_restore_args: /* return to kernel space */
932 DISABLE_INTERRUPTS(CLBR_ANY)
933 /*
934 * The iretq could re-enable interrupts:
935 */
936 TRACE_IRQS_IRETQ
937restore_args:
938 RESTORE_ARGS 1,8,1
939
940irq_return:
941 INTERRUPT_RETURN
942 _ASM_EXTABLE(irq_return, bad_iret)
943
944#ifdef CONFIG_PARAVIRT
945ENTRY(native_iret)
946 iretq
947 _ASM_EXTABLE(native_iret, bad_iret)
948#endif
949
950 .section .fixup,"ax"
951bad_iret:
952 /*
953 * The iret traps when the %cs or %ss being restored is bogus.
954 * We've lost the original trap vector and error code.
955 * #GPF is the most likely one to get for an invalid selector.
956 * So pretend we completed the iret and took the #GPF in user mode.
957 *
958 * We are now running with the kernel GS after exception recovery.
959 * But error_entry expects us to have user GS to match the user %cs,
960 * so swap back.
961 */
962 pushq $0
963
964 SWAPGS
965 jmp general_protection
966
967 .previous
968
969 /* edi: workmask, edx: work */
970retint_careful:
971 CFI_RESTORE_STATE
972 bt $TIF_NEED_RESCHED,%edx
973 jnc retint_signal
974 TRACE_IRQS_ON
975 ENABLE_INTERRUPTS(CLBR_NONE)
976 pushq_cfi %rdi
977 call schedule
978 popq_cfi %rdi
979 GET_THREAD_INFO(%rcx)
980 DISABLE_INTERRUPTS(CLBR_NONE)
981 TRACE_IRQS_OFF
982 jmp retint_check
983
984retint_signal:
985 testl $_TIF_DO_NOTIFY_MASK,%edx
986 jz retint_swapgs
987 TRACE_IRQS_ON
988 ENABLE_INTERRUPTS(CLBR_NONE)
989 SAVE_REST
990 movq $-1,ORIG_RAX(%rsp)
991 xorl %esi,%esi # oldset
992 movq %rsp,%rdi # &pt_regs
993 call do_notify_resume
994 RESTORE_REST
995 DISABLE_INTERRUPTS(CLBR_NONE)
996 TRACE_IRQS_OFF
997 GET_THREAD_INFO(%rcx)
998 jmp retint_with_reschedule
999
1000#ifdef CONFIG_PREEMPT
1001 /* Returning to kernel space. Check if we need preemption */
1002 /* rcx: threadinfo. interrupts off. */
1003ENTRY(retint_kernel)
1004 cmpl $0,TI_preempt_count(%rcx)
1005 jnz retint_restore_args
1006 bt $TIF_NEED_RESCHED,TI_flags(%rcx)
1007 jnc retint_restore_args
1008 bt $9,EFLAGS-ARGOFFSET(%rsp) /* interrupts off? */
1009 jnc retint_restore_args
1010 call preempt_schedule_irq
1011 jmp exit_intr
1012#endif
1013
1014 CFI_ENDPROC
1015END(common_interrupt)
1016/*
1017 * End of kprobes section
1018 */
1019 .popsection
1020
1021/*
1022 * APIC interrupts.
1023 */
1024.macro apicinterrupt num sym do_sym
1025ENTRY(\sym)
1026 INTR_FRAME
1027 pushq_cfi $~(\num)
1028.Lcommon_\sym:
1029 interrupt \do_sym
1030 jmp ret_from_intr
1031 CFI_ENDPROC
1032END(\sym)
1033.endm
1034
1035#ifdef CONFIG_SMP
1036apicinterrupt IRQ_MOVE_CLEANUP_VECTOR \
1037 irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
1038apicinterrupt REBOOT_VECTOR \
1039 reboot_interrupt smp_reboot_interrupt
1040#endif
1041
1042#ifdef CONFIG_X86_UV
1043apicinterrupt UV_BAU_MESSAGE \
1044 uv_bau_message_intr1 uv_bau_message_interrupt
1045#endif
1046apicinterrupt LOCAL_TIMER_VECTOR \
1047 apic_timer_interrupt smp_apic_timer_interrupt
1048apicinterrupt X86_PLATFORM_IPI_VECTOR \
1049 x86_platform_ipi smp_x86_platform_ipi
1050
1051#ifdef CONFIG_SMP
1052 ALIGN
1053 INTR_FRAME
1054.irp idx,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, \
1055 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
1056.if NUM_INVALIDATE_TLB_VECTORS > \idx
1057ENTRY(invalidate_interrupt\idx)
1058 pushq_cfi $~(INVALIDATE_TLB_VECTOR_START+\idx)
1059 jmp .Lcommon_invalidate_interrupt0
1060 CFI_ADJUST_CFA_OFFSET -8
1061END(invalidate_interrupt\idx)
1062.endif
1063.endr
1064 CFI_ENDPROC
1065apicinterrupt INVALIDATE_TLB_VECTOR_START, \
1066 invalidate_interrupt0, smp_invalidate_interrupt
1067#endif
1068
1069apicinterrupt THRESHOLD_APIC_VECTOR \
1070 threshold_interrupt smp_threshold_interrupt
1071apicinterrupt THERMAL_APIC_VECTOR \
1072 thermal_interrupt smp_thermal_interrupt
1073
1074#ifdef CONFIG_SMP
1075apicinterrupt CALL_FUNCTION_SINGLE_VECTOR \
1076 call_function_single_interrupt smp_call_function_single_interrupt
1077apicinterrupt CALL_FUNCTION_VECTOR \
1078 call_function_interrupt smp_call_function_interrupt
1079apicinterrupt RESCHEDULE_VECTOR \
1080 reschedule_interrupt smp_reschedule_interrupt
1081#endif
1082
1083apicinterrupt ERROR_APIC_VECTOR \
1084 error_interrupt smp_error_interrupt
1085apicinterrupt SPURIOUS_APIC_VECTOR \
1086 spurious_interrupt smp_spurious_interrupt
1087
1088#ifdef CONFIG_IRQ_WORK
1089apicinterrupt IRQ_WORK_VECTOR \
1090 irq_work_interrupt smp_irq_work_interrupt
1091#endif
1092
1093/*
1094 * Exception entry points.
1095 */
1096.macro zeroentry sym do_sym
1097ENTRY(\sym)
1098 INTR_FRAME
1099 PARAVIRT_ADJUST_EXCEPTION_FRAME
1100 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1101 subq $ORIG_RAX-R15, %rsp
1102 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1103 call error_entry
1104 DEFAULT_FRAME 0
1105 movq %rsp,%rdi /* pt_regs pointer */
1106 xorl %esi,%esi /* no error code */
1107 call \do_sym
1108 jmp error_exit /* %ebx: no swapgs flag */
1109 CFI_ENDPROC
1110END(\sym)
1111.endm
1112
1113.macro paranoidzeroentry sym do_sym
1114ENTRY(\sym)
1115 INTR_FRAME
1116 PARAVIRT_ADJUST_EXCEPTION_FRAME
1117 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1118 subq $ORIG_RAX-R15, %rsp
1119 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1120 call save_paranoid
1121 TRACE_IRQS_OFF
1122 movq %rsp,%rdi /* pt_regs pointer */
1123 xorl %esi,%esi /* no error code */
1124 call \do_sym
1125 jmp paranoid_exit /* %ebx: no swapgs flag */
1126 CFI_ENDPROC
1127END(\sym)
1128.endm
1129
1130#define INIT_TSS_IST(x) PER_CPU_VAR(init_tss) + (TSS_ist + ((x) - 1) * 8)
1131.macro paranoidzeroentry_ist sym do_sym ist
1132ENTRY(\sym)
1133 INTR_FRAME
1134 PARAVIRT_ADJUST_EXCEPTION_FRAME
1135 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1136 subq $ORIG_RAX-R15, %rsp
1137 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1138 call save_paranoid
1139 TRACE_IRQS_OFF_DEBUG
1140 movq %rsp,%rdi /* pt_regs pointer */
1141 xorl %esi,%esi /* no error code */
1142 subq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1143 call \do_sym
1144 addq $EXCEPTION_STKSZ, INIT_TSS_IST(\ist)
1145 jmp paranoid_exit /* %ebx: no swapgs flag */
1146 CFI_ENDPROC
1147END(\sym)
1148.endm
1149
1150.macro errorentry sym do_sym
1151ENTRY(\sym)
1152 XCPT_FRAME
1153 PARAVIRT_ADJUST_EXCEPTION_FRAME
1154 subq $ORIG_RAX-R15, %rsp
1155 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1156 call error_entry
1157 DEFAULT_FRAME 0
1158 movq %rsp,%rdi /* pt_regs pointer */
1159 movq ORIG_RAX(%rsp),%rsi /* get error code */
1160 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1161 call \do_sym
1162 jmp error_exit /* %ebx: no swapgs flag */
1163 CFI_ENDPROC
1164END(\sym)
1165.endm
1166
1167 /* error code is on the stack already */
1168.macro paranoiderrorentry sym do_sym
1169ENTRY(\sym)
1170 XCPT_FRAME
1171 PARAVIRT_ADJUST_EXCEPTION_FRAME
1172 subq $ORIG_RAX-R15, %rsp
1173 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1174 call save_paranoid
1175 DEFAULT_FRAME 0
1176 TRACE_IRQS_OFF
1177 movq %rsp,%rdi /* pt_regs pointer */
1178 movq ORIG_RAX(%rsp),%rsi /* get error code */
1179 movq $-1,ORIG_RAX(%rsp) /* no syscall to restart */
1180 call \do_sym
1181 jmp paranoid_exit /* %ebx: no swapgs flag */
1182 CFI_ENDPROC
1183END(\sym)
1184.endm
1185
1186zeroentry divide_error do_divide_error
1187zeroentry overflow do_overflow
1188zeroentry bounds do_bounds
1189zeroentry invalid_op do_invalid_op
1190zeroentry device_not_available do_device_not_available
1191paranoiderrorentry double_fault do_double_fault
1192zeroentry coprocessor_segment_overrun do_coprocessor_segment_overrun
1193errorentry invalid_TSS do_invalid_TSS
1194errorentry segment_not_present do_segment_not_present
1195zeroentry spurious_interrupt_bug do_spurious_interrupt_bug
1196zeroentry coprocessor_error do_coprocessor_error
1197errorentry alignment_check do_alignment_check
1198zeroentry simd_coprocessor_error do_simd_coprocessor_error
1199
1200
1201 /* Reload gs selector with exception handling */
1202 /* edi: new selector */
1203ENTRY(native_load_gs_index)
1204 CFI_STARTPROC
1205 pushfq_cfi
1206 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
1207 SWAPGS
1208gs_change:
1209 movl %edi,%gs
12102: mfence /* workaround */
1211 SWAPGS
1212 popfq_cfi
1213 ret
1214 CFI_ENDPROC
1215END(native_load_gs_index)
1216
1217 _ASM_EXTABLE(gs_change,bad_gs)
1218 .section .fixup,"ax"
1219 /* running with kernelgs */
1220bad_gs:
1221 SWAPGS /* switch back to user gs */
1222 xorl %eax,%eax
1223 movl %eax,%gs
1224 jmp 2b
1225 .previous
1226
1227ENTRY(kernel_thread_helper)
1228 pushq $0 # fake return address
1229 CFI_STARTPROC
1230 /*
1231 * Here we are in the child and the registers are set as they were
1232 * at kernel_thread() invocation in the parent.
1233 */
1234 call *%rsi
1235 # exit
1236 mov %eax, %edi
1237 call do_exit
1238 ud2 # padding for call trace
1239 CFI_ENDPROC
1240END(kernel_thread_helper)
1241
1242/*
1243 * execve(). This function needs to use IRET, not SYSRET, to set up all state properly.
1244 *
1245 * C extern interface:
1246 * extern long execve(const char *name, char **argv, char **envp)
1247 *
1248 * asm input arguments:
1249 * rdi: name, rsi: argv, rdx: envp
1250 *
1251 * We want to fallback into:
1252 * extern long sys_execve(const char *name, char **argv,char **envp, struct pt_regs *regs)
1253 *
1254 * do_sys_execve asm fallback arguments:
1255 * rdi: name, rsi: argv, rdx: envp, rcx: fake frame on the stack
1256 */
1257ENTRY(kernel_execve)
1258 CFI_STARTPROC
1259 FAKE_STACK_FRAME $0
1260 SAVE_ALL
1261 movq %rsp,%rcx
1262 call sys_execve
1263 movq %rax, RAX(%rsp)
1264 RESTORE_REST
1265 testq %rax,%rax
1266 je int_ret_from_sys_call
1267 RESTORE_ARGS
1268 UNFAKE_STACK_FRAME
1269 ret
1270 CFI_ENDPROC
1271END(kernel_execve)
1272
1273/* Call softirq on interrupt stack. Interrupts are off. */
1274ENTRY(call_softirq)
1275 CFI_STARTPROC
1276 pushq_cfi %rbp
1277 CFI_REL_OFFSET rbp,0
1278 mov %rsp,%rbp
1279 CFI_DEF_CFA_REGISTER rbp
1280 incl PER_CPU_VAR(irq_count)
1281 cmove PER_CPU_VAR(irq_stack_ptr),%rsp
1282 push %rbp # backlink for old unwinder
1283 call __do_softirq
1284 leaveq
1285 CFI_RESTORE rbp
1286 CFI_DEF_CFA_REGISTER rsp
1287 CFI_ADJUST_CFA_OFFSET -8
1288 decl PER_CPU_VAR(irq_count)
1289 ret
1290 CFI_ENDPROC
1291END(call_softirq)
1292
1293#ifdef CONFIG_XEN
1294zeroentry xen_hypervisor_callback xen_do_hypervisor_callback
1295
1296/*
1297 * A note on the "critical region" in our callback handler.
1298 * We want to avoid stacking callback handlers due to events occurring
1299 * during handling of the last event. To do this, we keep events disabled
1300 * until we've done all processing. HOWEVER, we must enable events before
1301 * popping the stack frame (can't be done atomically) and so it would still
1302 * be possible to get enough handler activations to overflow the stack.
1303 * Although unlikely, bugs of that kind are hard to track down, so we'd
1304 * like to avoid the possibility.
1305 * So, on entry to the handler we detect whether we interrupted an
1306 * existing activation in its critical region -- if so, we pop the current
1307 * activation and restart the handler using the previous one.
1308 */
1309ENTRY(xen_do_hypervisor_callback) # do_hypervisor_callback(struct *pt_regs)
1310 CFI_STARTPROC
1311/*
1312 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
1313 * see the correct pointer to the pt_regs
1314 */
1315 movq %rdi, %rsp # we don't return, adjust the stack frame
1316 CFI_ENDPROC
1317 DEFAULT_FRAME
131811: incl PER_CPU_VAR(irq_count)
1319 movq %rsp,%rbp
1320 CFI_DEF_CFA_REGISTER rbp
1321 cmovzq PER_CPU_VAR(irq_stack_ptr),%rsp
1322 pushq %rbp # backlink for old unwinder
1323 call xen_evtchn_do_upcall
1324 popq %rsp
1325 CFI_DEF_CFA_REGISTER rsp
1326 decl PER_CPU_VAR(irq_count)
1327 jmp error_exit
1328 CFI_ENDPROC
1329END(xen_do_hypervisor_callback)
1330
1331/*
1332 * Hypervisor uses this for application faults while it executes.
1333 * We get here for two reasons:
1334 * 1. Fault while reloading DS, ES, FS or GS
1335 * 2. Fault while executing IRET
1336 * Category 1 we do not need to fix up as Xen has already reloaded all segment
1337 * registers that could be reloaded and zeroed the others.
1338 * Category 2 we fix up by killing the current process. We cannot use the
1339 * normal Linux return path in this case because if we use the IRET hypercall
1340 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1341 * We distinguish between categories by comparing each saved segment register
1342 * with its current contents: any discrepancy means we in category 1.
1343 */
1344ENTRY(xen_failsafe_callback)
1345 INTR_FRAME 1 (6*8)
1346 /*CFI_REL_OFFSET gs,GS*/
1347 /*CFI_REL_OFFSET fs,FS*/
1348 /*CFI_REL_OFFSET es,ES*/
1349 /*CFI_REL_OFFSET ds,DS*/
1350 CFI_REL_OFFSET r11,8
1351 CFI_REL_OFFSET rcx,0
1352 movw %ds,%cx
1353 cmpw %cx,0x10(%rsp)
1354 CFI_REMEMBER_STATE
1355 jne 1f
1356 movw %es,%cx
1357 cmpw %cx,0x18(%rsp)
1358 jne 1f
1359 movw %fs,%cx
1360 cmpw %cx,0x20(%rsp)
1361 jne 1f
1362 movw %gs,%cx
1363 cmpw %cx,0x28(%rsp)
1364 jne 1f
1365 /* All segments match their saved values => Category 2 (Bad IRET). */
1366 movq (%rsp),%rcx
1367 CFI_RESTORE rcx
1368 movq 8(%rsp),%r11
1369 CFI_RESTORE r11
1370 addq $0x30,%rsp
1371 CFI_ADJUST_CFA_OFFSET -0x30
1372 pushq_cfi $0 /* RIP */
1373 pushq_cfi %r11
1374 pushq_cfi %rcx
1375 jmp general_protection
1376 CFI_RESTORE_STATE
13771: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
1378 movq (%rsp),%rcx
1379 CFI_RESTORE rcx
1380 movq 8(%rsp),%r11
1381 CFI_RESTORE r11
1382 addq $0x30,%rsp
1383 CFI_ADJUST_CFA_OFFSET -0x30
1384 pushq_cfi $0
1385 SAVE_ALL
1386 jmp error_exit
1387 CFI_ENDPROC
1388END(xen_failsafe_callback)
1389
1390apicinterrupt XEN_HVM_EVTCHN_CALLBACK \
1391 xen_hvm_callback_vector xen_evtchn_do_upcall
1392
1393#endif /* CONFIG_XEN */
1394
1395/*
1396 * Some functions should be protected against kprobes
1397 */
1398 .pushsection .kprobes.text, "ax"
1399
1400paranoidzeroentry_ist debug do_debug DEBUG_STACK
1401paranoidzeroentry_ist int3 do_int3 DEBUG_STACK
1402paranoiderrorentry stack_segment do_stack_segment
1403#ifdef CONFIG_XEN
1404zeroentry xen_debug do_debug
1405zeroentry xen_int3 do_int3
1406errorentry xen_stack_segment do_stack_segment
1407#endif
1408errorentry general_protection do_general_protection
1409errorentry page_fault do_page_fault
1410#ifdef CONFIG_KVM_GUEST
1411errorentry async_page_fault do_async_page_fault
1412#endif
1413#ifdef CONFIG_X86_MCE
1414paranoidzeroentry machine_check *machine_check_vector(%rip)
1415#endif
1416
1417 /*
1418 * "Paranoid" exit path from exception stack.
1419 * Paranoid because this is used by NMIs and cannot take
1420 * any kernel state for granted.
1421 * We don't do kernel preemption checks here, because only
1422 * NMI should be common and it does not enable IRQs and
1423 * cannot get reschedule ticks.
1424 *
1425 * "trace" is 0 for the NMI handler only, because irq-tracing
1426 * is fundamentally NMI-unsafe. (we cannot change the soft and
1427 * hard flags at once, atomically)
1428 */
1429
1430 /* ebx: no swapgs flag */
1431ENTRY(paranoid_exit)
1432 DEFAULT_FRAME
1433 DISABLE_INTERRUPTS(CLBR_NONE)
1434 TRACE_IRQS_OFF_DEBUG
1435 testl %ebx,%ebx /* swapgs needed? */
1436 jnz paranoid_restore
1437 testl $3,CS(%rsp)
1438 jnz paranoid_userspace
1439paranoid_swapgs:
1440 TRACE_IRQS_IRETQ 0
1441 SWAPGS_UNSAFE_STACK
1442 RESTORE_ALL 8
1443 jmp irq_return
1444paranoid_restore:
1445 TRACE_IRQS_IRETQ_DEBUG 0
1446 RESTORE_ALL 8
1447 jmp irq_return
1448paranoid_userspace:
1449 GET_THREAD_INFO(%rcx)
1450 movl TI_flags(%rcx),%ebx
1451 andl $_TIF_WORK_MASK,%ebx
1452 jz paranoid_swapgs
1453 movq %rsp,%rdi /* &pt_regs */
1454 call sync_regs
1455 movq %rax,%rsp /* switch stack for scheduling */
1456 testl $_TIF_NEED_RESCHED,%ebx
1457 jnz paranoid_schedule
1458 movl %ebx,%edx /* arg3: thread flags */
1459 TRACE_IRQS_ON
1460 ENABLE_INTERRUPTS(CLBR_NONE)
1461 xorl %esi,%esi /* arg2: oldset */
1462 movq %rsp,%rdi /* arg1: &pt_regs */
1463 call do_notify_resume
1464 DISABLE_INTERRUPTS(CLBR_NONE)
1465 TRACE_IRQS_OFF
1466 jmp paranoid_userspace
1467paranoid_schedule:
1468 TRACE_IRQS_ON
1469 ENABLE_INTERRUPTS(CLBR_ANY)
1470 call schedule
1471 DISABLE_INTERRUPTS(CLBR_ANY)
1472 TRACE_IRQS_OFF
1473 jmp paranoid_userspace
1474 CFI_ENDPROC
1475END(paranoid_exit)
1476
1477/*
1478 * Exception entry point. This expects an error code/orig_rax on the stack.
1479 * returns in "no swapgs flag" in %ebx.
1480 */
1481ENTRY(error_entry)
1482 XCPT_FRAME
1483 CFI_ADJUST_CFA_OFFSET 15*8
1484 /* oldrax contains error code */
1485 cld
1486 movq_cfi rdi, RDI+8
1487 movq_cfi rsi, RSI+8
1488 movq_cfi rdx, RDX+8
1489 movq_cfi rcx, RCX+8
1490 movq_cfi rax, RAX+8
1491 movq_cfi r8, R8+8
1492 movq_cfi r9, R9+8
1493 movq_cfi r10, R10+8
1494 movq_cfi r11, R11+8
1495 movq_cfi rbx, RBX+8
1496 movq_cfi rbp, RBP+8
1497 movq_cfi r12, R12+8
1498 movq_cfi r13, R13+8
1499 movq_cfi r14, R14+8
1500 movq_cfi r15, R15+8
1501 xorl %ebx,%ebx
1502 testl $3,CS+8(%rsp)
1503 je error_kernelspace
1504error_swapgs:
1505 SWAPGS
1506error_sti:
1507 TRACE_IRQS_OFF
1508 ret
1509
1510/*
1511 * There are two places in the kernel that can potentially fault with
1512 * usergs. Handle them here. The exception handlers after iret run with
1513 * kernel gs again, so don't set the user space flag. B stepping K8s
1514 * sometimes report an truncated RIP for IRET exceptions returning to
1515 * compat mode. Check for these here too.
1516 */
1517error_kernelspace:
1518 incl %ebx
1519 leaq irq_return(%rip),%rcx
1520 cmpq %rcx,RIP+8(%rsp)
1521 je error_swapgs
1522 movl %ecx,%eax /* zero extend */
1523 cmpq %rax,RIP+8(%rsp)
1524 je bstep_iret
1525 cmpq $gs_change,RIP+8(%rsp)
1526 je error_swapgs
1527 jmp error_sti
1528
1529bstep_iret:
1530 /* Fix truncated RIP */
1531 movq %rcx,RIP+8(%rsp)
1532 jmp error_swapgs
1533 CFI_ENDPROC
1534END(error_entry)
1535
1536
1537/* ebx: no swapgs flag (1: don't need swapgs, 0: need it) */
1538ENTRY(error_exit)
1539 DEFAULT_FRAME
1540 movl %ebx,%eax
1541 RESTORE_REST
1542 DISABLE_INTERRUPTS(CLBR_NONE)
1543 TRACE_IRQS_OFF
1544 GET_THREAD_INFO(%rcx)
1545 testl %eax,%eax
1546 jne retint_kernel
1547 LOCKDEP_SYS_EXIT_IRQ
1548 movl TI_flags(%rcx),%edx
1549 movl $_TIF_WORK_MASK,%edi
1550 andl %edi,%edx
1551 jnz retint_careful
1552 jmp retint_swapgs
1553 CFI_ENDPROC
1554END(error_exit)
1555
1556/*
1557 * Test if a given stack is an NMI stack or not.
1558 */
1559 .macro test_in_nmi reg stack nmi_ret normal_ret
1560 cmpq %\reg, \stack
1561 ja \normal_ret
1562 subq $EXCEPTION_STKSZ, %\reg
1563 cmpq %\reg, \stack
1564 jb \normal_ret
1565 jmp \nmi_ret
1566 .endm
1567
1568 /* runs on exception stack */
1569ENTRY(nmi)
1570 INTR_FRAME
1571 PARAVIRT_ADJUST_EXCEPTION_FRAME
1572 /*
1573 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1574 * the iretq it performs will take us out of NMI context.
1575 * This means that we can have nested NMIs where the next
1576 * NMI is using the top of the stack of the previous NMI. We
1577 * can't let it execute because the nested NMI will corrupt the
1578 * stack of the previous NMI. NMI handlers are not re-entrant
1579 * anyway.
1580 *
1581 * To handle this case we do the following:
1582 * Check the a special location on the stack that contains
1583 * a variable that is set when NMIs are executing.
1584 * The interrupted task's stack is also checked to see if it
1585 * is an NMI stack.
1586 * If the variable is not set and the stack is not the NMI
1587 * stack then:
1588 * o Set the special variable on the stack
1589 * o Copy the interrupt frame into a "saved" location on the stack
1590 * o Copy the interrupt frame into a "copy" location on the stack
1591 * o Continue processing the NMI
1592 * If the variable is set or the previous stack is the NMI stack:
1593 * o Modify the "copy" location to jump to the repeate_nmi
1594 * o return back to the first NMI
1595 *
1596 * Now on exit of the first NMI, we first clear the stack variable
1597 * The NMI stack will tell any nested NMIs at that point that it is
1598 * nested. Then we pop the stack normally with iret, and if there was
1599 * a nested NMI that updated the copy interrupt stack frame, a
1600 * jump will be made to the repeat_nmi code that will handle the second
1601 * NMI.
1602 */
1603
1604 /* Use %rdx as out temp variable throughout */
1605 pushq_cfi %rdx
1606 CFI_REL_OFFSET rdx, 0
1607
1608 /*
1609 * If %cs was not the kernel segment, then the NMI triggered in user
1610 * space, which means it is definitely not nested.
1611 */
1612 cmpl $__KERNEL_CS, 16(%rsp)
1613 jne first_nmi
1614
1615 /*
1616 * Check the special variable on the stack to see if NMIs are
1617 * executing.
1618 */
1619 cmpl $1, -8(%rsp)
1620 je nested_nmi
1621
1622 /*
1623 * Now test if the previous stack was an NMI stack.
1624 * We need the double check. We check the NMI stack to satisfy the
1625 * race when the first NMI clears the variable before returning.
1626 * We check the variable because the first NMI could be in a
1627 * breakpoint routine using a breakpoint stack.
1628 */
1629 lea 6*8(%rsp), %rdx
1630 test_in_nmi rdx, 4*8(%rsp), nested_nmi, first_nmi
1631 CFI_REMEMBER_STATE
1632
1633nested_nmi:
1634 /*
1635 * Do nothing if we interrupted the fixup in repeat_nmi.
1636 * It's about to repeat the NMI handler, so we are fine
1637 * with ignoring this one.
1638 */
1639 movq $repeat_nmi, %rdx
1640 cmpq 8(%rsp), %rdx
1641 ja 1f
1642 movq $end_repeat_nmi, %rdx
1643 cmpq 8(%rsp), %rdx
1644 ja nested_nmi_out
1645
16461:
1647 /* Set up the interrupted NMIs stack to jump to repeat_nmi */
1648 leaq -6*8(%rsp), %rdx
1649 movq %rdx, %rsp
1650 CFI_ADJUST_CFA_OFFSET 6*8
1651 pushq_cfi $__KERNEL_DS
1652 pushq_cfi %rdx
1653 pushfq_cfi
1654 pushq_cfi $__KERNEL_CS
1655 pushq_cfi $repeat_nmi
1656
1657 /* Put stack back */
1658 addq $(11*8), %rsp
1659 CFI_ADJUST_CFA_OFFSET -11*8
1660
1661nested_nmi_out:
1662 popq_cfi %rdx
1663 CFI_RESTORE rdx
1664
1665 /* No need to check faults here */
1666 INTERRUPT_RETURN
1667
1668 CFI_RESTORE_STATE
1669first_nmi:
1670 /*
1671 * Because nested NMIs will use the pushed location that we
1672 * stored in rdx, we must keep that space available.
1673 * Here's what our stack frame will look like:
1674 * +-------------------------+
1675 * | original SS |
1676 * | original Return RSP |
1677 * | original RFLAGS |
1678 * | original CS |
1679 * | original RIP |
1680 * +-------------------------+
1681 * | temp storage for rdx |
1682 * +-------------------------+
1683 * | NMI executing variable |
1684 * +-------------------------+
1685 * | Saved SS |
1686 * | Saved Return RSP |
1687 * | Saved RFLAGS |
1688 * | Saved CS |
1689 * | Saved RIP |
1690 * +-------------------------+
1691 * | copied SS |
1692 * | copied Return RSP |
1693 * | copied RFLAGS |
1694 * | copied CS |
1695 * | copied RIP |
1696 * +-------------------------+
1697 * | pt_regs |
1698 * +-------------------------+
1699 *
1700 * The saved stack frame is used to fix up the copied stack frame
1701 * that a nested NMI may change to make the interrupted NMI iret jump
1702 * to the repeat_nmi. The original stack frame and the temp storage
1703 * is also used by nested NMIs and can not be trusted on exit.
1704 */
1705 /* Do not pop rdx, nested NMIs will corrupt that part of the stack */
1706 movq (%rsp), %rdx
1707 CFI_RESTORE rdx
1708
1709 /* Set the NMI executing variable on the stack. */
1710 pushq_cfi $1
1711
1712 /* Copy the stack frame to the Saved frame */
1713 .rept 5
1714 pushq_cfi 6*8(%rsp)
1715 .endr
1716 CFI_DEF_CFA_OFFSET SS+8-RIP
1717
1718 /* Everything up to here is safe from nested NMIs */
1719
1720 /*
1721 * If there was a nested NMI, the first NMI's iret will return
1722 * here. But NMIs are still enabled and we can take another
1723 * nested NMI. The nested NMI checks the interrupted RIP to see
1724 * if it is between repeat_nmi and end_repeat_nmi, and if so
1725 * it will just return, as we are about to repeat an NMI anyway.
1726 * This makes it safe to copy to the stack frame that a nested
1727 * NMI will update.
1728 */
1729repeat_nmi:
1730 /*
1731 * Update the stack variable to say we are still in NMI (the update
1732 * is benign for the non-repeat case, where 1 was pushed just above
1733 * to this very stack slot).
1734 */
1735 movq $1, 5*8(%rsp)
1736
1737 /* Make another copy, this one may be modified by nested NMIs */
1738 .rept 5
1739 pushq_cfi 4*8(%rsp)
1740 .endr
1741 CFI_DEF_CFA_OFFSET SS+8-RIP
1742end_repeat_nmi:
1743
1744 /*
1745 * Everything below this point can be preempted by a nested
1746 * NMI if the first NMI took an exception and reset our iret stack
1747 * so that we repeat another NMI.
1748 */
1749 pushq_cfi $-1 /* ORIG_RAX: no syscall to restart */
1750 subq $ORIG_RAX-R15, %rsp
1751 CFI_ADJUST_CFA_OFFSET ORIG_RAX-R15
1752 /*
1753 * Use save_paranoid to handle SWAPGS, but no need to use paranoid_exit
1754 * as we should not be calling schedule in NMI context.
1755 * Even with normal interrupts enabled. An NMI should not be
1756 * setting NEED_RESCHED or anything that normal interrupts and
1757 * exceptions might do.
1758 */
1759 call save_paranoid
1760 DEFAULT_FRAME 0
1761 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
1762 movq %rsp,%rdi
1763 movq $-1,%rsi
1764 call do_nmi
1765 testl %ebx,%ebx /* swapgs needed? */
1766 jnz nmi_restore
1767nmi_swapgs:
1768 SWAPGS_UNSAFE_STACK
1769nmi_restore:
1770 RESTORE_ALL 8
1771 /* Clear the NMI executing stack variable */
1772 movq $0, 10*8(%rsp)
1773 jmp irq_return
1774 CFI_ENDPROC
1775END(nmi)
1776
1777ENTRY(ignore_sysret)
1778 CFI_STARTPROC
1779 mov $-ENOSYS,%eax
1780 sysret
1781 CFI_ENDPROC
1782END(ignore_sysret)
1783
1784/*
1785 * End of kprobes section
1786 */
1787 .popsection