2 * Compatibility mode system call entry point for x86-64.
4 * Copyright 2000-2002 Andi Kleen, SuSE Labs.
7 #include <asm/asm-offsets.h>
8 #include <asm/current.h>
10 #include <asm/ia32_unistd.h>
11 #include <asm/thread_info.h>
12 #include <asm/segment.h>
13 #include <asm/irqflags.h>
16 #include <linux/linkage.h>
17 #include <linux/err.h>
19 /* Avoid __ASSEMBLER__'ifying <linux/audit.h> just for this. */
20 #include <linux/elf-em.h>
21 #define AUDIT_ARCH_I386 (EM_386|__AUDIT_ARCH_LE)
22 #define __AUDIT_ARCH_LE 0x40000000
24 #ifndef CONFIG_AUDITSYSCALL
25 # define sysexit_audit ia32_ret_from_sys_call_irqs_off
26 # define sysretl_audit ia32_ret_from_sys_call_irqs_off
29 .section .entry.text, "ax"
31 #ifdef CONFIG_PARAVIRT
32 ENTRY(native_usergs_sysret32)
35 ENDPROC(native_usergs_sysret32)
39 * 32-bit SYSENTER instruction entry.
41 * SYSENTER loads ss, rsp, cs, and rip from previously programmed MSRs.
42 * IF and VM in rflags are cleared (IOW: interrupts are off).
43 * SYSENTER does not save anything on the stack,
44 * and does not save old rip (!!!) and rflags.
47 * eax system call number
56 * This is purely a fast path. For anything complicated we use the int 0x80
57 * path below. We set up a complete hardware stack frame to share code
58 * with the int 0x80 path.
60 ENTRY(entry_SYSENTER_compat)
62 * Interrupts are off on entry.
63 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
64 * it is too small to ever cause noticeable irq latency.
67 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
68 ENABLE_INTERRUPTS(CLBR_NONE)
70 /* Zero-extending 32-bit regs, do not remove */
74 movl ASM_THREAD_INFO(TI_sysenter_return, %rsp, 0), %r10d
76 /* Construct struct pt_regs on stack */
77 pushq $__USER32_DS /* pt_regs->ss */
78 pushq %rbp /* pt_regs->sp */
79 pushfq /* pt_regs->flags */
80 pushq $__USER32_CS /* pt_regs->cs */
81 pushq %r10 /* pt_regs->ip = thread_info->sysenter_return */
82 pushq %rax /* pt_regs->orig_ax */
83 pushq %rdi /* pt_regs->di */
84 pushq %rsi /* pt_regs->si */
85 pushq %rdx /* pt_regs->dx */
86 pushq %rcx /* pt_regs->cx */
87 pushq $-ENOSYS /* pt_regs->ax */
89 sub $(10*8), %rsp /* pt_regs->r8-11, bp, bx, r12-15 not saved */
92 * no need to do an access_ok check here because rbp has been
93 * 32-bit zero extended
97 _ASM_EXTABLE(1b, ia32_badarg)
101 * Sysenter doesn't filter flags, so we need to clear NT
102 * ourselves. To save a few cycles, we can check whether
103 * NT was set instead of doing an unconditional popfq.
105 testl $X86_EFLAGS_NT, EFLAGS(%rsp)
106 jnz sysenter_fix_flags
107 sysenter_flags_fixed:
109 orl $TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
110 testl $_TIF_WORK_SYSCALL_ENTRY, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
111 jnz sysenter_tracesys
114 /* 32-bit syscall -> 64-bit C ABI argument conversion */
115 movl %edi, %r8d /* arg5 */
116 movl %ebp, %r9d /* arg6 */
117 xchg %ecx, %esi /* rsi:arg2, rcx:arg4 */
118 movl %ebx, %edi /* arg1 */
119 movl %edx, %edx /* arg3 (zero extension) */
121 cmpq $(IA32_NR_syscalls-1), %rax
123 call *ia32_sys_call_table(, %rax, 8)
126 DISABLE_INTERRUPTS(CLBR_NONE)
128 testl $_TIF_ALLWORK_MASK, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
130 sysexit_from_sys_call:
132 * NB: SYSEXIT is not obviously safe for 64-bit kernels -- an
133 * NMI between STI and SYSEXIT has poorly specified behavior,
134 * and and NMI followed by an IRQ with usergs is fatal. So
135 * we just pretend we're using SYSEXIT but we really use
138 * This code path is still called 'sysexit' because it pairs
139 * with 'sysenter' and it uses the SYSENTER calling convention.
141 andl $~TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
142 movl RIP(%rsp), %ecx /* User %eip */
146 xorl %edx, %edx /* Do not leak kernel information */
150 movl EFLAGS(%rsp), %r11d /* User eflags */
154 * SYSRETL works even on Intel CPUs. Use it in preference to SYSEXIT,
155 * since it avoids a dicey window with interrupts enabled.
160 * USERGS_SYSRET32 does:
161 * gsbase = user's gs base
167 * The prologue set RIP(%rsp) to VDSO32_SYSENTER_RETURN, which does:
173 * Therefore, we invoke SYSRETL with EDX and R8-R10 zeroed to
174 * avoid info leaks. R11 ends up with VDSO32_SYSENTER_RETURN's
175 * address (already known to user code), and R12-R15 are
176 * callee-saved and therefore don't contain any interesting
181 #ifdef CONFIG_AUDITSYSCALL
182 .macro auditsys_entry_common
184 * At this point, registers hold syscall args in the 32-bit syscall ABI:
185 * EAX is syscall number, the 6 args are in EBX,ECX,EDX,ESI,EDI,EBP.
187 * We want to pass them to __audit_syscall_entry(), which is a 64-bit
188 * C function with 5 parameters, so shuffle them to match what
189 * the function expects: RDI,RSI,RDX,RCX,R8.
191 movl %esi, %r8d /* arg5 (R8 ) <= 4th syscall arg (ESI) */
192 xchg %ecx, %edx /* arg4 (RCX) <= 3rd syscall arg (EDX) */
193 /* arg3 (RDX) <= 2nd syscall arg (ECX) */
194 movl %ebx, %esi /* arg2 (RSI) <= 1st syscall arg (EBX) */
195 movl %eax, %edi /* arg1 (RDI) <= syscall number (EAX) */
196 call __audit_syscall_entry
199 * We are going to jump back to the syscall dispatch code.
200 * Prepare syscall args as required by the 64-bit C ABI.
201 * Registers clobbered by __audit_syscall_entry() are
202 * loaded from pt_regs on stack:
204 movl ORIG_RAX(%rsp), %eax /* syscall number */
205 movl %ebx, %edi /* arg1 */
206 movl RCX(%rsp), %esi /* arg2 */
207 movl RDX(%rsp), %edx /* arg3 */
208 movl RSI(%rsp), %ecx /* arg4 */
209 movl RDI(%rsp), %r8d /* arg5 */
212 .macro auditsys_exit exit
214 ENABLE_INTERRUPTS(CLBR_NONE)
215 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
216 jnz ia32_ret_from_sys_call
217 movl %eax, %esi /* second arg, syscall return value */
218 cmpl $-MAX_ERRNO, %eax /* is it an error ? */
220 movslq %eax, %rsi /* if error sign extend to 64 bits */
221 1: setbe %al /* 1 if error, 0 if not */
222 movzbl %al, %edi /* zero-extend that into %edi */
223 call __audit_syscall_exit
224 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %edi
225 DISABLE_INTERRUPTS(CLBR_NONE)
227 testl %edi, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
229 xorl %eax, %eax /* Do not leak kernel information */
234 jmp int_ret_from_sys_call_irqs_off
238 auditsys_entry_common
239 movl %ebp, %r9d /* reload 6th syscall arg */
240 jmp sysenter_dispatch
243 auditsys_exit sysexit_from_sys_call
247 pushq $(X86_EFLAGS_IF|X86_EFLAGS_FIXED)
249 jmp sysenter_flags_fixed
252 #ifdef CONFIG_AUDITSYSCALL
253 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT), ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
257 xorl %eax, %eax /* Do not leak kernel information */
262 movq %rsp, %rdi /* &pt_regs -> arg1 */
263 call syscall_trace_enter
265 /* Reload arg registers from stack. (see sysenter_tracesys) */
270 movl %eax, %eax /* zero extension */
274 ENDPROC(entry_SYSENTER_compat)
277 * 32-bit SYSCALL instruction entry.
279 * 32-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
280 * then loads new ss, cs, and rip from previously programmed MSRs.
281 * rflags gets masked by a value from another MSR (so CLD and CLAC
282 * are not needed). SYSCALL does not save anything on the stack
283 * and does not change rsp.
285 * Note: rflags saving+masking-with-MSR happens only in Long mode
286 * (in legacy 32-bit mode, IF, RF and VM bits are cleared and that's it).
287 * Don't get confused: rflags saving+masking depends on Long Mode Active bit
288 * (EFER.LMA=1), NOT on bitness of userspace where SYSCALL executes
289 * or target CS descriptor's L bit (SYSCALL does not read segment descriptors).
292 * eax system call number
295 * ebp arg2 (note: not saved in the stack frame, should not be touched)
302 * This is purely a fast path. For anything complicated we use the int 0x80
303 * path below. We set up a complete hardware stack frame to share code
304 * with the int 0x80 path.
306 ENTRY(entry_SYSCALL_compat)
308 * Interrupts are off on entry.
309 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
310 * it is too small to ever cause noticeable irq latency.
314 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
315 ENABLE_INTERRUPTS(CLBR_NONE)
317 /* Zero-extending 32-bit regs, do not remove */
320 /* Construct struct pt_regs on stack */
321 pushq $__USER32_DS /* pt_regs->ss */
322 pushq %r8 /* pt_regs->sp */
323 pushq %r11 /* pt_regs->flags */
324 pushq $__USER32_CS /* pt_regs->cs */
325 pushq %rcx /* pt_regs->ip */
326 pushq %rax /* pt_regs->orig_ax */
327 pushq %rdi /* pt_regs->di */
328 pushq %rsi /* pt_regs->si */
329 pushq %rdx /* pt_regs->dx */
330 pushq %rbp /* pt_regs->cx */
332 pushq $-ENOSYS /* pt_regs->ax */
333 sub $(10*8), %rsp /* pt_regs->r8-11, bp, bx, r12-15 not saved */
336 * No need to do an access_ok check here because r8 has been
337 * 32-bit zero extended:
341 _ASM_EXTABLE(1b, ia32_badarg)
343 orl $TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
344 testl $_TIF_WORK_SYSCALL_ENTRY, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
348 /* 32-bit syscall -> 64-bit C ABI argument conversion */
349 movl %edi, %r8d /* arg5 */
350 /* r9 already loaded */ /* arg6 */
351 xchg %ecx, %esi /* rsi:arg2, rcx:arg4 */
352 movl %ebx, %edi /* arg1 */
353 movl %edx, %edx /* arg3 (zero extension) */
356 cmpq $(IA32_NR_syscalls-1), %rax
359 call *ia32_sys_call_table(, %rax, 8)
362 DISABLE_INTERRUPTS(CLBR_NONE)
364 testl $_TIF_ALLWORK_MASK, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
367 sysretl_from_sys_call:
368 andl $~TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
373 movl EFLAGS(%rsp), %r11d
381 * 64-bit->32-bit SYSRET restores eip from ecx,
382 * eflags from r11 (but RF and VM bits are forced to 0),
383 * cs and ss are loaded from MSRs.
384 * (Note: 32-bit->32-bit SYSRET is different: since r11
385 * does not exist, it merely sets eflags.IF=1).
387 * NB: On AMD CPUs with the X86_BUG_SYSRET_SS_ATTRS bug, the ss
388 * descriptor is not reinitialized. This means that we must
389 * avoid SYSRET with SS == NULL, which could happen if we schedule,
390 * exit the kernel, and re-enter using an interrupt vector. (All
391 * interrupt entries on x86_64 set SS to NULL.) We prevent that
392 * from happening by reloading SS in __switch_to.
396 #ifdef CONFIG_AUDITSYSCALL
398 movl %r9d, R9(%rsp) /* register to be clobbered by call */
399 auditsys_entry_common
400 movl R9(%rsp), %r9d /* reload 6th syscall arg */
404 auditsys_exit sysretl_from_sys_call
408 #ifdef CONFIG_AUDITSYSCALL
409 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT), ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
414 xorl %eax, %eax /* Do not leak kernel information */
419 movq %rsp, %rdi /* &pt_regs -> arg1 */
420 call syscall_trace_enter
423 /* Reload arg registers from stack. (see sysenter_tracesys) */
428 movl %eax, %eax /* zero extension */
433 END(entry_SYSCALL_compat)
437 * So far, we've entered kernel mode, set AC, turned on IRQs, and
438 * saved C regs except r8-r11. We haven't done any of the other
439 * standard entry work, though. We want to bail, but we shouldn't
440 * treat this as a syscall entry since we don't even know what the
441 * args are. Instead, treat this as a non-syscall entry, finish
442 * the entry work, and immediately exit after setting AX = -EFAULT.
444 * We're really just being polite here. Killing the task outright
445 * would be a reasonable action, too. Given that the only valid
446 * way to have gotten here is through the vDSO, and we already know
447 * that the stack pointer is bad, the task isn't going to survive
448 * for long no matter what we do.
451 ASM_CLAC /* undo STAC */
452 movq $-EFAULT, RAX(%rsp) /* return -EFAULT if possible */
454 /* Fill in the rest of pt_regs */
462 /* Turn IRQs back off. */
463 DISABLE_INTERRUPTS(CLBR_NONE)
466 /* Now finish entering normal kernel mode. */
467 #ifdef CONFIG_CONTEXT_TRACKING
468 call enter_from_user_mode
471 /* And exit again. */
474 ia32_ret_from_sys_call_irqs_off:
476 ENABLE_INTERRUPTS(CLBR_NONE)
478 ia32_ret_from_sys_call:
479 xorl %eax, %eax /* Do not leak kernel information */
484 jmp int_ret_from_sys_call
487 * Emulated IA32 system calls via int 0x80.
490 * eax system call number
496 * ebp arg6 (note: not saved in the stack frame, should not be touched)
499 * Uses the same stack frame as the x86-64 version.
500 * All registers except eax must be saved (but ptrace may violate that).
501 * Arguments are zero extended. For system calls that want sign extension and
502 * take long arguments a wrapper is needed. Most calls can just be called
504 * Assumes it is only called from user space and entered with interrupts off.
507 ENTRY(entry_INT80_compat)
509 * Interrupts are off on entry.
510 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
511 * it is too small to ever cause noticeable irq latency.
513 PARAVIRT_ADJUST_EXCEPTION_FRAME
515 ENABLE_INTERRUPTS(CLBR_NONE)
517 /* Zero-extending 32-bit regs, do not remove */
520 /* Construct struct pt_regs on stack (iret frame is already on stack) */
521 pushq %rax /* pt_regs->orig_ax */
522 pushq %rdi /* pt_regs->di */
523 pushq %rsi /* pt_regs->si */
524 pushq %rdx /* pt_regs->dx */
525 pushq %rcx /* pt_regs->cx */
526 pushq $-ENOSYS /* pt_regs->ax */
527 pushq $0 /* pt_regs->r8 */
528 pushq $0 /* pt_regs->r9 */
529 pushq $0 /* pt_regs->r10 */
530 pushq $0 /* pt_regs->r11 */
532 sub $(6*8), %rsp /* pt_regs->bp, bx, r12-15 not saved */
534 orl $TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
535 testl $_TIF_WORK_SYSCALL_ENTRY, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
539 /* 32-bit syscall -> 64-bit C ABI argument conversion */
540 movl %edi, %r8d /* arg5 */
541 movl %ebp, %r9d /* arg6 */
542 xchg %ecx, %esi /* rsi:arg2, rcx:arg4 */
543 movl %ebx, %edi /* arg1 */
544 movl %edx, %edx /* arg3 (zero extension) */
545 cmpq $(IA32_NR_syscalls-1), %rax
548 call *ia32_sys_call_table(, %rax, 8)
551 jmp int_ret_from_sys_call
555 movq %rsp, %rdi /* &pt_regs -> arg1 */
556 call syscall_trace_enter
558 * Reload arg registers from stack in case ptrace changed them.
559 * Don't reload %eax because syscall_trace_enter() returned
560 * the %rax value we should see. But do truncate it to 32 bits.
561 * If it's -1 to make us punt the syscall, then (u32)-1 is still
562 * an appropriately invalid value.
568 movl %eax, %eax /* zero extension */
571 END(entry_INT80_compat)
573 .macro PTREGSCALL label, func
576 leaq \func(%rip), %rax
577 jmp ia32_ptregs_common
580 PTREGSCALL stub32_rt_sigreturn, sys32_rt_sigreturn
581 PTREGSCALL stub32_sigreturn, sys32_sigreturn
582 PTREGSCALL stub32_fork, sys_fork
583 PTREGSCALL stub32_vfork, sys_vfork
587 leaq sys_clone(%rip), %rax
589 * The 32-bit clone ABI is: clone(..., int tls_val, int *child_tidptr).
590 * The 64-bit clone ABI is: clone(..., int *child_tidptr, int tls_val).
592 * The native 64-bit kernel's sys_clone() implements the latter,
593 * so we need to swap arguments here before calling it:
596 jmp ia32_ptregs_common
604 END(ia32_ptregs_common)