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 .section .entry.text, "ax"
21 #ifdef CONFIG_PARAVIRT
22 ENTRY(native_usergs_sysret32)
25 ENDPROC(native_usergs_sysret32)
29 * 32-bit SYSENTER instruction entry.
31 * SYSENTER loads ss, rsp, cs, and rip from previously programmed MSRs.
32 * IF and VM in rflags are cleared (IOW: interrupts are off).
33 * SYSENTER does not save anything on the stack,
34 * and does not save old rip (!!!) and rflags.
37 * eax system call number
46 * This is purely a fast path. For anything complicated we use the int 0x80
47 * path below. We set up a complete hardware stack frame to share code
48 * with the int 0x80 path.
50 ENTRY(entry_SYSENTER_compat)
51 /* Interrupts are off on entry. */
53 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
56 * User tracing code (ptrace or signal handlers) might assume that
57 * the saved RAX contains a 32-bit number when we're invoking a 32-bit
58 * syscall. Just in case the high bits are nonzero, zero-extend
59 * the syscall number. (This could almost certainly be deleted
60 * with no ill effects.)
64 /* Construct struct pt_regs on stack */
65 pushq $__USER32_DS /* pt_regs->ss */
66 pushq %rcx /* pt_regs->sp */
69 * Push flags. This is nasty. First, interrupts are currently
70 * off, but we need pt_regs->flags to have IF set. Second, even
71 * if TF was set when SYSENTER started, it's clear by now. We fix
72 * that later using TIF_SINGLESTEP.
74 pushfq /* pt_regs->flags (except IF = 0) */
75 orl $X86_EFLAGS_IF, (%rsp) /* Fix saved flags */
76 ASM_CLAC /* Clear AC after saving FLAGS */
78 pushq $__USER32_CS /* pt_regs->cs */
80 pushq %r8 /* pt_regs->ip = 0 (placeholder) */
81 pushq %rax /* pt_regs->orig_ax */
82 pushq %rdi /* pt_regs->di */
83 pushq %rsi /* pt_regs->si */
84 pushq %rdx /* pt_regs->dx */
85 pushq %rcx /* pt_regs->cx (will be overwritten) */
86 pushq $-ENOSYS /* pt_regs->ax */
87 pushq %r8 /* pt_regs->r8 = 0 */
88 pushq %r8 /* pt_regs->r9 = 0 */
89 pushq %r8 /* pt_regs->r10 = 0 */
90 pushq %r8 /* pt_regs->r11 = 0 */
91 pushq %rbx /* pt_regs->rbx */
92 pushq %rbp /* pt_regs->rbp */
93 pushq %r8 /* pt_regs->r12 = 0 */
94 pushq %r8 /* pt_regs->r13 = 0 */
95 pushq %r8 /* pt_regs->r14 = 0 */
96 pushq %r8 /* pt_regs->r15 = 0 */
100 * Sysenter doesn't filter flags, so we need to clear NT
101 * ourselves. To save a few cycles, we can check whether
102 * NT was set instead of doing an unconditional popfq.
103 * This needs to happen before enabling interrupts so that
104 * we don't get preempted with NT set.
106 * NB.: sysenter_fix_flags is a label with the code under it moved
107 * out-of-line as an optimization: NT is unlikely to be set in the
108 * majority of the cases and instead of polluting the I$ unnecessarily,
109 * we're keeping that code behind a branch which will predict as
110 * not-taken and therefore its instructions won't be fetched.
112 testl $X86_EFLAGS_NT, EFLAGS(%rsp)
113 jnz sysenter_fix_flags
114 sysenter_flags_fixed:
117 * User mode is traced as though IRQs are on, and SYSENTER
123 call do_fast_syscall_32
126 jmp sysret32_from_system_call
129 pushq $X86_EFLAGS_FIXED
131 jmp sysenter_flags_fixed
132 ENDPROC(entry_SYSENTER_compat)
135 * 32-bit SYSCALL instruction entry.
137 * 32-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
138 * then loads new ss, cs, and rip from previously programmed MSRs.
139 * rflags gets masked by a value from another MSR (so CLD and CLAC
140 * are not needed). SYSCALL does not save anything on the stack
141 * and does not change rsp.
143 * Note: rflags saving+masking-with-MSR happens only in Long mode
144 * (in legacy 32-bit mode, IF, RF and VM bits are cleared and that's it).
145 * Don't get confused: rflags saving+masking depends on Long Mode Active bit
146 * (EFER.LMA=1), NOT on bitness of userspace where SYSCALL executes
147 * or target CS descriptor's L bit (SYSCALL does not read segment descriptors).
150 * eax system call number
153 * ebp arg2 (note: not saved in the stack frame, should not be touched)
160 ENTRY(entry_SYSCALL_compat)
161 /* Interrupts are off on entry. */
164 /* Stash user ESP and switch to the kernel stack. */
166 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
168 /* Zero-extending 32-bit regs, do not remove */
171 /* Construct struct pt_regs on stack */
172 pushq $__USER32_DS /* pt_regs->ss */
173 pushq %r8 /* pt_regs->sp */
174 pushq %r11 /* pt_regs->flags */
175 pushq $__USER32_CS /* pt_regs->cs */
176 pushq %rcx /* pt_regs->ip */
177 pushq %rax /* pt_regs->orig_ax */
178 pushq %rdi /* pt_regs->di */
179 pushq %rsi /* pt_regs->si */
180 pushq %rdx /* pt_regs->dx */
181 pushq %rcx /* pt_regs->cx (will be overwritten) */
182 pushq $-ENOSYS /* pt_regs->ax */
184 pushq %r8 /* pt_regs->r8 = 0 */
185 pushq %r8 /* pt_regs->r9 = 0 */
186 pushq %r8 /* pt_regs->r10 = 0 */
187 pushq %r8 /* pt_regs->r11 = 0 */
188 pushq %rbx /* pt_regs->rbx */
189 pushq %rbp /* pt_regs->rbp */
190 pushq %r8 /* pt_regs->r12 = 0 */
191 pushq %r8 /* pt_regs->r13 = 0 */
192 pushq %r8 /* pt_regs->r14 = 0 */
193 pushq %r8 /* pt_regs->r15 = 0 */
196 * User mode is traced as though IRQs are on, and SYSENTER
202 call do_fast_syscall_32
206 /* Opportunistic SYSRET */
207 sysret32_from_system_call:
208 TRACE_IRQS_ON /* User mode traces as IRQs on. */
209 movq RBX(%rsp), %rbx /* pt_regs->rbx */
210 movq RBP(%rsp), %rbp /* pt_regs->rbp */
211 movq EFLAGS(%rsp), %r11 /* pt_regs->flags (in r11) */
212 movq RIP(%rsp), %rcx /* pt_regs->ip (in rcx) */
213 addq $RAX, %rsp /* Skip r8-r15 */
214 popq %rax /* pt_regs->rax */
215 popq %rdx /* Skip pt_regs->cx */
216 popq %rdx /* pt_regs->dx */
217 popq %rsi /* pt_regs->si */
218 popq %rdi /* pt_regs->di */
221 * USERGS_SYSRET32 does:
222 * GSBASE = user's GS base
228 * ECX will not match pt_regs->cx, but we're returning to a vDSO
229 * trampoline that will fix up RCX, so this is okay.
231 * R12-R15 are callee-saved, so they contain whatever was in them
232 * when the system call started, which is already known to user
233 * code. We zero R8-R10 to avoid info leaks.
238 movq RSP-ORIG_RAX(%rsp), %rsp
240 END(entry_SYSCALL_compat)
243 * Emulated IA32 system calls via int 0x80.
246 * eax system call number
252 * ebp arg6 (note: not saved in the stack frame, should not be touched)
255 * Uses the same stack frame as the x86-64 version.
256 * All registers except eax must be saved (but ptrace may violate that).
257 * Arguments are zero extended. For system calls that want sign extension and
258 * take long arguments a wrapper is needed. Most calls can just be called
260 * Assumes it is only called from user space and entered with interrupts off.
263 ENTRY(entry_INT80_compat)
265 * Interrupts are off on entry.
267 PARAVIRT_ADJUST_EXCEPTION_FRAME
271 * User tracing code (ptrace or signal handlers) might assume that
272 * the saved RAX contains a 32-bit number when we're invoking a 32-bit
273 * syscall. Just in case the high bits are nonzero, zero-extend
274 * the syscall number. (This could almost certainly be deleted
275 * with no ill effects.)
279 /* Construct struct pt_regs on stack (iret frame is already on stack) */
280 pushq %rax /* pt_regs->orig_ax */
281 pushq %rdi /* pt_regs->di */
282 pushq %rsi /* pt_regs->si */
283 pushq %rdx /* pt_regs->dx */
284 pushq %rcx /* pt_regs->cx */
285 pushq $-ENOSYS /* pt_regs->ax */
287 pushq %r8 /* pt_regs->r8 = 0 */
288 pushq %r8 /* pt_regs->r9 = 0 */
289 pushq %r8 /* pt_regs->r10 = 0 */
290 pushq %r8 /* pt_regs->r11 = 0 */
291 pushq %rbx /* pt_regs->rbx */
292 pushq %rbp /* pt_regs->rbp */
293 pushq %r12 /* pt_regs->r12 */
294 pushq %r13 /* pt_regs->r13 */
295 pushq %r14 /* pt_regs->r14 */
296 pushq %r15 /* pt_regs->r15 */
300 * User mode is traced as though IRQs are on, and the interrupt
301 * gate turned them off.
306 call do_syscall_32_irqs_off
309 /* Go back to user mode. */
312 jmp restore_regs_and_iret
313 END(entry_INT80_compat)
318 * The 32-bit clone ABI is: clone(..., int tls_val, int *child_tidptr).
319 * The 64-bit clone ABI is: clone(..., int *child_tidptr, int tls_val).
321 * The native 64-bit kernel's sys_clone() implements the latter,
322 * so we need to swap arguments here before calling it: