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
26 # define sysretl_audit ia32_ret_from_sys_call
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(ia32_sysenter_target)
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 */
144 xorl %edx, %edx /* Do not leak kernel information */
148 movl EFLAGS(%rsp), %r11d /* User eflags */
152 * SYSRETL works even on Intel CPUs. Use it in preference to SYSEXIT,
153 * since it avoids a dicey window with interrupts enabled.
158 * USERGS_SYSRET32 does:
159 * gsbase = user's gs base
165 * The prologue set RIP(%rsp) to VDSO32_SYSENTER_RETURN, which does:
171 * Therefore, we invoke SYSRETL with EDX and R8-R10 zeroed to
172 * avoid info leaks. R11 ends up with VDSO32_SYSENTER_RETURN's
173 * address (already known to user code), and R12-R15 are
174 * callee-saved and therefore don't contain any interesting
179 #ifdef CONFIG_AUDITSYSCALL
180 .macro auditsys_entry_common
181 movl %esi, %r8d /* 5th arg: 4th syscall arg */
182 movl %ecx, %r9d /* swap with edx */
183 movl %edx, %ecx /* 4th arg: 3rd syscall arg */
184 movl %r9d, %edx /* 3rd arg: 2nd syscall arg */
185 movl %ebx, %esi /* 2nd arg: 1st syscall arg */
186 movl %eax, %edi /* 1st arg: syscall number */
187 call __audit_syscall_entry
188 movl ORIG_RAX(%rsp), %eax /* reload syscall number */
189 movl %ebx, %edi /* reload 1st syscall arg */
190 movl RCX(%rsp), %esi /* reload 2nd syscall arg */
191 movl RDX(%rsp), %edx /* reload 3rd syscall arg */
192 movl RSI(%rsp), %ecx /* reload 4th syscall arg */
193 movl RDI(%rsp), %r8d /* reload 5th syscall arg */
196 .macro auditsys_exit exit
197 testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
198 jnz ia32_ret_from_sys_call
200 ENABLE_INTERRUPTS(CLBR_NONE)
201 movl %eax, %esi /* second arg, syscall return value */
202 cmpl $-MAX_ERRNO, %eax /* is it an error ? */
204 movslq %eax, %rsi /* if error sign extend to 64 bits */
205 1: setbe %al /* 1 if error, 0 if not */
206 movzbl %al, %edi /* zero-extend that into %edi */
207 call __audit_syscall_exit
208 movq RAX(%rsp), %rax /* reload syscall return value */
209 movl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT), %edi
210 DISABLE_INTERRUPTS(CLBR_NONE)
212 testl %edi, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
214 xorl %eax, %eax /* Do not leak kernel information */
223 auditsys_entry_common
224 movl %ebp, %r9d /* reload 6th syscall arg */
225 jmp sysenter_dispatch
228 auditsys_exit sysexit_from_sys_call
232 pushq $(X86_EFLAGS_IF|X86_EFLAGS_FIXED)
234 jmp sysenter_flags_fixed
237 #ifdef CONFIG_AUDITSYSCALL
238 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT), ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
242 xorl %eax, %eax /* Do not leak kernel information */
247 movq %rsp, %rdi /* &pt_regs -> arg1 */
248 call syscall_trace_enter
250 /* Reload arg registers from stack. (see sysenter_tracesys) */
255 movl %eax, %eax /* zero extension */
259 ENDPROC(ia32_sysenter_target)
262 * 32-bit SYSCALL instruction entry.
264 * 32-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
265 * then loads new ss, cs, and rip from previously programmed MSRs.
266 * rflags gets masked by a value from another MSR (so CLD and CLAC
267 * are not needed). SYSCALL does not save anything on the stack
268 * and does not change rsp.
270 * Note: rflags saving+masking-with-MSR happens only in Long mode
271 * (in legacy 32-bit mode, IF, RF and VM bits are cleared and that's it).
272 * Don't get confused: rflags saving+masking depends on Long Mode Active bit
273 * (EFER.LMA=1), NOT on bitness of userspace where SYSCALL executes
274 * or target CS descriptor's L bit (SYSCALL does not read segment descriptors).
277 * eax system call number
280 * ebp arg2 (note: not saved in the stack frame, should not be touched)
287 * This is purely a fast path. For anything complicated we use the int 0x80
288 * path below. We set up a complete hardware stack frame to share code
289 * with the int 0x80 path.
291 ENTRY(ia32_cstar_target)
293 * Interrupts are off on entry.
294 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
295 * it is too small to ever cause noticeable irq latency.
299 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
300 ENABLE_INTERRUPTS(CLBR_NONE)
302 /* Zero-extending 32-bit regs, do not remove */
305 /* Construct struct pt_regs on stack */
306 pushq $__USER32_DS /* pt_regs->ss */
307 pushq %r8 /* pt_regs->sp */
308 pushq %r11 /* pt_regs->flags */
309 pushq $__USER32_CS /* pt_regs->cs */
310 pushq %rcx /* pt_regs->ip */
311 pushq %rax /* pt_regs->orig_ax */
312 pushq %rdi /* pt_regs->di */
313 pushq %rsi /* pt_regs->si */
314 pushq %rdx /* pt_regs->dx */
315 pushq %rbp /* pt_regs->cx */
317 pushq $-ENOSYS /* pt_regs->ax */
318 sub $(10*8), %rsp /* pt_regs->r8-11, bp, bx, r12-15 not saved */
321 * No need to do an access_ok check here because r8 has been
322 * 32-bit zero extended:
326 _ASM_EXTABLE(1b, ia32_badarg)
328 orl $TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
329 testl $_TIF_WORK_SYSCALL_ENTRY, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
333 /* 32-bit syscall -> 64-bit C ABI argument conversion */
334 movl %edi, %r8d /* arg5 */
335 movl %ebp, %r9d /* arg6 */
336 xchg %ecx, %esi /* rsi:arg2, rcx:arg4 */
337 movl %ebx, %edi /* arg1 */
338 movl %edx, %edx /* arg3 (zero extension) */
341 cmpq $(IA32_NR_syscalls-1), %rax
344 call *ia32_sys_call_table(, %rax, 8)
347 DISABLE_INTERRUPTS(CLBR_NONE)
349 testl $_TIF_ALLWORK_MASK, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
352 sysretl_from_sys_call:
353 andl $~TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
357 movl EFLAGS(%rsp), %r11d
364 * 64-bit->32-bit SYSRET restores eip from ecx,
365 * eflags from r11 (but RF and VM bits are forced to 0),
366 * cs and ss are loaded from MSRs.
367 * (Note: 32-bit->32-bit SYSRET is different: since r11
368 * does not exist, it merely sets eflags.IF=1).
370 * NB: On AMD CPUs with the X86_BUG_SYSRET_SS_ATTRS bug, the ss
371 * descriptor is not reinitialized. This means that we must
372 * avoid SYSRET with SS == NULL, which could happen if we schedule,
373 * exit the kernel, and re-enter using an interrupt vector. (All
374 * interrupt entries on x86_64 set SS to NULL.) We prevent that
375 * from happening by reloading SS in __switch_to.
379 #ifdef CONFIG_AUDITSYSCALL
381 auditsys_entry_common
382 movl %ebp, %r9d /* reload 6th syscall arg */
386 auditsys_exit sysretl_from_sys_call
390 #ifdef CONFIG_AUDITSYSCALL
391 testl $(_TIF_WORK_SYSCALL_ENTRY & ~_TIF_SYSCALL_AUDIT), ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
395 xorl %eax, %eax /* Do not leak kernel information */
400 movq %rsp, %rdi /* &pt_regs -> arg1 */
401 call syscall_trace_enter
403 /* Reload arg registers from stack. (see sysenter_tracesys) */
408 movl %eax, %eax /* zero extension */
412 END(ia32_cstar_target)
419 ia32_ret_from_sys_call:
420 xorl %eax, %eax /* Do not leak kernel information */
425 jmp int_ret_from_sys_call
428 * Emulated IA32 system calls via int 0x80.
431 * eax system call number
437 * ebp arg6 (note: not saved in the stack frame, should not be touched)
440 * Uses the same stack frame as the x86-64 version.
441 * All registers except eax must be saved (but ptrace may violate that).
442 * Arguments are zero extended. For system calls that want sign extension and
443 * take long arguments a wrapper is needed. Most calls can just be called
445 * Assumes it is only called from user space and entered with interrupts off.
450 * Interrupts are off on entry.
451 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
452 * it is too small to ever cause noticeable irq latency.
454 PARAVIRT_ADJUST_EXCEPTION_FRAME
456 ENABLE_INTERRUPTS(CLBR_NONE)
458 /* Zero-extending 32-bit regs, do not remove */
461 /* Construct struct pt_regs on stack (iret frame is already on stack) */
462 pushq %rax /* pt_regs->orig_ax */
463 pushq %rdi /* pt_regs->di */
464 pushq %rsi /* pt_regs->si */
465 pushq %rdx /* pt_regs->dx */
466 pushq %rcx /* pt_regs->cx */
467 pushq $-ENOSYS /* pt_regs->ax */
468 pushq $0 /* pt_regs->r8 */
469 pushq $0 /* pt_regs->r9 */
470 pushq $0 /* pt_regs->r10 */
471 pushq $0 /* pt_regs->r11 */
473 sub $(6*8), %rsp /* pt_regs->bp, bx, r12-15 not saved */
475 orl $TS_COMPAT, ASM_THREAD_INFO(TI_status, %rsp, SIZEOF_PTREGS)
476 testl $_TIF_WORK_SYSCALL_ENTRY, ASM_THREAD_INFO(TI_flags, %rsp, SIZEOF_PTREGS)
480 /* 32-bit syscall -> 64-bit C ABI argument conversion */
481 movl %edi, %r8d /* arg5 */
482 movl %ebp, %r9d /* arg6 */
483 xchg %ecx, %esi /* rsi:arg2, rcx:arg4 */
484 movl %ebx, %edi /* arg1 */
485 movl %edx, %edx /* arg3 (zero extension) */
486 cmpq $(IA32_NR_syscalls-1), %rax
489 call *ia32_sys_call_table(, %rax, 8) /* RIP relative */
494 jmp int_ret_from_sys_call
498 movq %rsp, %rdi /* &pt_regs -> arg1 */
499 call syscall_trace_enter
501 * Reload arg registers from stack in case ptrace changed them.
502 * Don't reload %eax because syscall_trace_enter() returned
503 * the %rax value we should see. But do truncate it to 32 bits.
504 * If it's -1 to make us punt the syscall, then (u32)-1 is still
505 * an appropriately invalid value.
511 movl %eax, %eax /* zero extension */
516 .macro PTREGSCALL label, func
519 leaq \func(%rip), %rax
520 jmp ia32_ptregs_common
523 PTREGSCALL stub32_rt_sigreturn, sys32_rt_sigreturn
524 PTREGSCALL stub32_sigreturn, sys32_sigreturn
525 PTREGSCALL stub32_fork, sys_fork
526 PTREGSCALL stub32_vfork, sys_vfork
530 leaq sys_clone(%rip), %rax
532 * The 32-bit clone ABI is: clone(..., int tls_val, int *child_tidptr).
533 * The 64-bit clone ABI is: clone(..., int *child_tidptr, int tls_val).
535 * The native 64-bit kernel's sys_clone() implements the latter,
536 * so we need to swap arguments here before calling it:
539 jmp ia32_ptregs_common
547 END(ia32_ptregs_common)