1fa697140SDominik Brodowski /* SPDX-License-Identifier: GPL-2.0 */ 2fa697140SDominik Brodowski /* 3fa697140SDominik Brodowski * syscall_wrapper.h - x86 specific wrappers to syscall definitions 4fa697140SDominik Brodowski */ 5fa697140SDominik Brodowski 6fa697140SDominik Brodowski #ifndef _ASM_X86_SYSCALL_WRAPPER_H 7fa697140SDominik Brodowski #define _ASM_X86_SYSCALL_WRAPPER_H 8fa697140SDominik Brodowski 99440c429SJiri Olsa #include <asm/ptrace.h> 106e484764SSami Tolvanen 110f78ff17SBrian Gerst extern long __x64_sys_ni_syscall(const struct pt_regs *regs); 120f78ff17SBrian Gerst extern long __ia32_sys_ni_syscall(const struct pt_regs *regs); 13cc42c045SBrian Gerst 1425c619e5SBrian Gerst /* 1525c619e5SBrian Gerst * Instead of the generic __SYSCALL_DEFINEx() definition, the x86 version takes 1625c619e5SBrian Gerst * struct pt_regs *regs as the only argument of the syscall stub(s) named as: 1725c619e5SBrian Gerst * __x64_sys_*() - 64-bit native syscall 1825c619e5SBrian Gerst * __ia32_sys_*() - 32-bit native syscall or common compat syscall 1925c619e5SBrian Gerst * __ia32_compat_sys_*() - 32-bit compat syscall 202e958a8aSMasahiro Yamada * __x64_compat_sys_*() - 64-bit X32 compat syscall 2125c619e5SBrian Gerst * 2225c619e5SBrian Gerst * The registers are decoded according to the ABI: 2325c619e5SBrian Gerst * 64-bit: RDI, RSI, RDX, R10, R8, R9 2425c619e5SBrian Gerst * 32-bit: EBX, ECX, EDX, ESI, EDI, EBP 2525c619e5SBrian Gerst * 2625c619e5SBrian Gerst * The stub then passes the decoded arguments to the __se_sys_*() wrapper to 2725c619e5SBrian Gerst * perform sign-extension (omitted for zero-argument syscalls). Finally the 2825c619e5SBrian Gerst * arguments are passed to the __do_sys_*() function which is the actual 2925c619e5SBrian Gerst * syscall. These wrappers are marked as inline so the compiler can optimize 3025c619e5SBrian Gerst * the functions where appropriate. 3125c619e5SBrian Gerst * 3225c619e5SBrian Gerst * Example assembly (slightly re-ordered for better readability): 3325c619e5SBrian Gerst * 3425c619e5SBrian Gerst * <__x64_sys_recv>: <-- syscall with 4 parameters 3525c619e5SBrian Gerst * callq <__fentry__> 3625c619e5SBrian Gerst * 3725c619e5SBrian Gerst * mov 0x70(%rdi),%rdi <-- decode regs->di 3825c619e5SBrian Gerst * mov 0x68(%rdi),%rsi <-- decode regs->si 3925c619e5SBrian Gerst * mov 0x60(%rdi),%rdx <-- decode regs->dx 4025c619e5SBrian Gerst * mov 0x38(%rdi),%rcx <-- decode regs->r10 4125c619e5SBrian Gerst * 4225c619e5SBrian Gerst * xor %r9d,%r9d <-- clear %r9 4325c619e5SBrian Gerst * xor %r8d,%r8d <-- clear %r8 4425c619e5SBrian Gerst * 4525c619e5SBrian Gerst * callq __sys_recvfrom <-- do the actual work in __sys_recvfrom() 4625c619e5SBrian Gerst * which takes 6 arguments 4725c619e5SBrian Gerst * 4825c619e5SBrian Gerst * cltq <-- extend return value to 64-bit 4925c619e5SBrian Gerst * retq <-- return 5025c619e5SBrian Gerst * 5125c619e5SBrian Gerst * This approach avoids leaking random user-provided register content down 5225c619e5SBrian Gerst * the call chain. 5325c619e5SBrian Gerst */ 5425c619e5SBrian Gerst 55ebeb8c82SDominik Brodowski /* Mapping of registers to parameters for syscalls on x86-64 and x32 */ 56ebeb8c82SDominik Brodowski #define SC_X86_64_REGS_TO_ARGS(x, ...) \ 57ebeb8c82SDominik Brodowski __MAP(x,__SC_ARGS \ 58ebeb8c82SDominik Brodowski ,,regs->di,,regs->si,,regs->dx \ 59ebeb8c82SDominik Brodowski ,,regs->r10,,regs->r8,,regs->r9) \ 60ebeb8c82SDominik Brodowski 61*de66d97aSRichard Palethorpe 62*de66d97aSRichard Palethorpe /* SYSCALL_PT_ARGS is Adapted from s390x */ 63*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARG6(m, t1, t2, t3, t4, t5, t6) \ 64*de66d97aSRichard Palethorpe SYSCALL_PT_ARG5(m, t1, t2, t3, t4, t5), m(t6, (regs->bp)) 65*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARG5(m, t1, t2, t3, t4, t5) \ 66*de66d97aSRichard Palethorpe SYSCALL_PT_ARG4(m, t1, t2, t3, t4), m(t5, (regs->di)) 67*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARG4(m, t1, t2, t3, t4) \ 68*de66d97aSRichard Palethorpe SYSCALL_PT_ARG3(m, t1, t2, t3), m(t4, (regs->si)) 69*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARG3(m, t1, t2, t3) \ 70*de66d97aSRichard Palethorpe SYSCALL_PT_ARG2(m, t1, t2), m(t3, (regs->dx)) 71*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARG2(m, t1, t2) \ 72*de66d97aSRichard Palethorpe SYSCALL_PT_ARG1(m, t1), m(t2, (regs->cx)) 73*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARG1(m, t1) m(t1, (regs->bx)) 74*de66d97aSRichard Palethorpe #define SYSCALL_PT_ARGS(x, ...) SYSCALL_PT_ARG##x(__VA_ARGS__) 75*de66d97aSRichard Palethorpe 76*de66d97aSRichard Palethorpe #define __SC_COMPAT_CAST(t, a) \ 77*de66d97aSRichard Palethorpe (__typeof(__builtin_choose_expr(__TYPE_IS_L(t), 0, 0U))) \ 78*de66d97aSRichard Palethorpe (unsigned int)a 79*de66d97aSRichard Palethorpe 80ebeb8c82SDominik Brodowski /* Mapping of registers to parameters for syscalls on i386 */ 81ebeb8c82SDominik Brodowski #define SC_IA32_REGS_TO_ARGS(x, ...) \ 82*de66d97aSRichard Palethorpe SYSCALL_PT_ARGS(x, __SC_COMPAT_CAST, \ 83*de66d97aSRichard Palethorpe __MAP(x, __SC_TYPE, __VA_ARGS__)) \ 84ebeb8c82SDominik Brodowski 85d2b5de49SBrian Gerst #define __SYS_STUB0(abi, name) \ 860f78ff17SBrian Gerst long __##abi##_##name(const struct pt_regs *regs); \ 87d2b5de49SBrian Gerst ALLOW_ERROR_INJECTION(__##abi##_##name, ERRNO); \ 880f78ff17SBrian Gerst long __##abi##_##name(const struct pt_regs *regs) \ 89d2b5de49SBrian Gerst __alias(__do_##name); 90d2b5de49SBrian Gerst 914399e0cfSBrian Gerst #define __SYS_STUBx(abi, name, ...) \ 920f78ff17SBrian Gerst long __##abi##_##name(const struct pt_regs *regs); \ 934399e0cfSBrian Gerst ALLOW_ERROR_INJECTION(__##abi##_##name, ERRNO); \ 940f78ff17SBrian Gerst long __##abi##_##name(const struct pt_regs *regs) \ 954399e0cfSBrian Gerst { \ 964399e0cfSBrian Gerst return __se_##name(__VA_ARGS__); \ 974399e0cfSBrian Gerst } 984399e0cfSBrian Gerst 996cc8d2b2SBrian Gerst #define __COND_SYSCALL(abi, name) \ 1007dfe553aSMasahiro Yamada __weak long __##abi##_##name(const struct pt_regs *__unused); \ 1010f78ff17SBrian Gerst __weak long __##abi##_##name(const struct pt_regs *__unused) \ 1026cc8d2b2SBrian Gerst { \ 1036cc8d2b2SBrian Gerst return sys_ni_syscall(); \ 1046cc8d2b2SBrian Gerst } 1056cc8d2b2SBrian Gerst 1064399e0cfSBrian Gerst #ifdef CONFIG_X86_64 107d2b5de49SBrian Gerst #define __X64_SYS_STUB0(name) \ 108d2b5de49SBrian Gerst __SYS_STUB0(x64, sys_##name) 109d2b5de49SBrian Gerst 1104399e0cfSBrian Gerst #define __X64_SYS_STUBx(x, name, ...) \ 1114399e0cfSBrian Gerst __SYS_STUBx(x64, sys##name, \ 1124399e0cfSBrian Gerst SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__)) 1136cc8d2b2SBrian Gerst 1146cc8d2b2SBrian Gerst #define __X64_COND_SYSCALL(name) \ 1156cc8d2b2SBrian Gerst __COND_SYSCALL(x64, sys_##name) 116a74d187cSBrian Gerst 1174399e0cfSBrian Gerst #else /* CONFIG_X86_64 */ 118d2b5de49SBrian Gerst #define __X64_SYS_STUB0(name) 1194399e0cfSBrian Gerst #define __X64_SYS_STUBx(x, name, ...) 1206cc8d2b2SBrian Gerst #define __X64_COND_SYSCALL(name) 1214399e0cfSBrian Gerst #endif /* CONFIG_X86_64 */ 1224399e0cfSBrian Gerst 12325c619e5SBrian Gerst #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 12425c619e5SBrian Gerst #define __IA32_SYS_STUB0(name) \ 12525c619e5SBrian Gerst __SYS_STUB0(ia32, sys_##name) 12625c619e5SBrian Gerst 12725c619e5SBrian Gerst #define __IA32_SYS_STUBx(x, name, ...) \ 12825c619e5SBrian Gerst __SYS_STUBx(ia32, sys##name, \ 12925c619e5SBrian Gerst SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__)) 13025c619e5SBrian Gerst 13125c619e5SBrian Gerst #define __IA32_COND_SYSCALL(name) \ 13225c619e5SBrian Gerst __COND_SYSCALL(ia32, sys_##name) 13325c619e5SBrian Gerst 13425c619e5SBrian Gerst #else /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ 13525c619e5SBrian Gerst #define __IA32_SYS_STUB0(name) 13625c619e5SBrian Gerst #define __IA32_SYS_STUBx(x, name, ...) 13725c619e5SBrian Gerst #define __IA32_COND_SYSCALL(name) 13825c619e5SBrian Gerst #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ 13925c619e5SBrian Gerst 140ebeb8c82SDominik Brodowski #ifdef CONFIG_IA32_EMULATION 141ebeb8c82SDominik Brodowski /* 142ebeb8c82SDominik Brodowski * For IA32 emulation, we need to handle "compat" syscalls *and* create 143e145242eSDominik Brodowski * additional wrappers (aptly named __ia32_sys_xyzzy) which decode the 144ebeb8c82SDominik Brodowski * ia32 regs in the proper order for shared or "common" syscalls. As some 145ebeb8c82SDominik Brodowski * syscalls may not be implemented, we need to expand COND_SYSCALL in 146766c6c1dSLinus Torvalds * kernel/sys_ni.c to cover this case as well. 147ebeb8c82SDominik Brodowski */ 148d2b5de49SBrian Gerst #define __IA32_COMPAT_SYS_STUB0(name) \ 149d2b5de49SBrian Gerst __SYS_STUB0(ia32, compat_sys_##name) 150cf3b83e1SAndy Lutomirski 151c76fc982SDominik Brodowski #define __IA32_COMPAT_SYS_STUBx(x, name, ...) \ 1524399e0cfSBrian Gerst __SYS_STUBx(ia32, compat_sys##name, \ 1534399e0cfSBrian Gerst SC_IA32_REGS_TO_ARGS(x, __VA_ARGS__)) 154ebeb8c82SDominik Brodowski 1556cc8d2b2SBrian Gerst #define __IA32_COMPAT_COND_SYSCALL(name) \ 1566cc8d2b2SBrian Gerst __COND_SYSCALL(ia32, compat_sys_##name) 1576cc8d2b2SBrian Gerst 158ebeb8c82SDominik Brodowski #else /* CONFIG_IA32_EMULATION */ 159d2b5de49SBrian Gerst #define __IA32_COMPAT_SYS_STUB0(name) 160c76fc982SDominik Brodowski #define __IA32_COMPAT_SYS_STUBx(x, name, ...) 1616cc8d2b2SBrian Gerst #define __IA32_COMPAT_COND_SYSCALL(name) 162ebeb8c82SDominik Brodowski #endif /* CONFIG_IA32_EMULATION */ 163ebeb8c82SDominik Brodowski 164ebeb8c82SDominik Brodowski 16583a44a4fSMasahiro Yamada #ifdef CONFIG_X86_X32_ABI 166ebeb8c82SDominik Brodowski /* 167ebeb8c82SDominik Brodowski * For the x32 ABI, we need to create a stub for compat_sys_*() which is aware 168ebeb8c82SDominik Brodowski * of the x86-64-style parameter ordering of x32 syscalls. The syscalls common 169ebeb8c82SDominik Brodowski * with x86_64 obviously do not need such care. 170ebeb8c82SDominik Brodowski */ 171d2b5de49SBrian Gerst #define __X32_COMPAT_SYS_STUB0(name) \ 1722e958a8aSMasahiro Yamada __SYS_STUB0(x64, compat_sys_##name) 173cf3b83e1SAndy Lutomirski 174c76fc982SDominik Brodowski #define __X32_COMPAT_SYS_STUBx(x, name, ...) \ 1752e958a8aSMasahiro Yamada __SYS_STUBx(x64, compat_sys##name, \ 1764399e0cfSBrian Gerst SC_X86_64_REGS_TO_ARGS(x, __VA_ARGS__)) 177ebeb8c82SDominik Brodowski 1786cc8d2b2SBrian Gerst #define __X32_COMPAT_COND_SYSCALL(name) \ 1792e958a8aSMasahiro Yamada __COND_SYSCALL(x64, compat_sys_##name) 180a74d187cSBrian Gerst 18183a44a4fSMasahiro Yamada #else /* CONFIG_X86_X32_ABI */ 182d2b5de49SBrian Gerst #define __X32_COMPAT_SYS_STUB0(name) 183c76fc982SDominik Brodowski #define __X32_COMPAT_SYS_STUBx(x, name, ...) 1846cc8d2b2SBrian Gerst #define __X32_COMPAT_COND_SYSCALL(name) 18583a44a4fSMasahiro Yamada #endif /* CONFIG_X86_X32_ABI */ 186ebeb8c82SDominik Brodowski 187ebeb8c82SDominik Brodowski 188ebeb8c82SDominik Brodowski #ifdef CONFIG_COMPAT 189ebeb8c82SDominik Brodowski /* 190ebeb8c82SDominik Brodowski * Compat means IA32_EMULATION and/or X86_X32. As they use a different 191ebeb8c82SDominik Brodowski * mapping of registers to parameters, we need to generate stubs for each 192d5a00528SDominik Brodowski * of them. 193ebeb8c82SDominik Brodowski */ 194cf3b83e1SAndy Lutomirski #define COMPAT_SYSCALL_DEFINE0(name) \ 1950f78ff17SBrian Gerst static long \ 196d2b5de49SBrian Gerst __do_compat_sys_##name(const struct pt_regs *__unused); \ 197d2b5de49SBrian Gerst __IA32_COMPAT_SYS_STUB0(name) \ 198d2b5de49SBrian Gerst __X32_COMPAT_SYS_STUB0(name) \ 1990f78ff17SBrian Gerst static long \ 200d2b5de49SBrian Gerst __do_compat_sys_##name(const struct pt_regs *__unused) 201cf3b83e1SAndy Lutomirski 202ebeb8c82SDominik Brodowski #define COMPAT_SYSCALL_DEFINEx(x, name, ...) \ 2035ac9efa3SDominik Brodowski static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \ 2045ac9efa3SDominik Brodowski static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\ 205c76fc982SDominik Brodowski __IA32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \ 206c76fc982SDominik Brodowski __X32_COMPAT_SYS_STUBx(x, name, __VA_ARGS__) \ 2075ac9efa3SDominik Brodowski static long __se_compat_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \ 208ebeb8c82SDominik Brodowski { \ 2095ac9efa3SDominik Brodowski return __do_compat_sys##name(__MAP(x,__SC_DELOUSE,__VA_ARGS__));\ 210ebeb8c82SDominik Brodowski } \ 2115ac9efa3SDominik Brodowski static inline long __do_compat_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__)) 212ebeb8c82SDominik Brodowski 213ebeb8c82SDominik Brodowski /* 214ebeb8c82SDominik Brodowski * As some compat syscalls may not be implemented, we need to expand 215766c6c1dSLinus Torvalds * COND_SYSCALL_COMPAT in kernel/sys_ni.c to cover this case as well. 216ebeb8c82SDominik Brodowski */ 217ebeb8c82SDominik Brodowski #define COND_SYSCALL_COMPAT(name) \ 2186cc8d2b2SBrian Gerst __IA32_COMPAT_COND_SYSCALL(name) \ 2196cc8d2b2SBrian Gerst __X32_COMPAT_COND_SYSCALL(name) 220ebeb8c82SDominik Brodowski 221ebeb8c82SDominik Brodowski #endif /* CONFIG_COMPAT */ 222ebeb8c82SDominik Brodowski 223fa697140SDominik Brodowski #define __SYSCALL_DEFINEx(x, name, ...) \ 224e145242eSDominik Brodowski static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)); \ 225e145242eSDominik Brodowski static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__));\ 2264399e0cfSBrian Gerst __X64_SYS_STUBx(x, name, __VA_ARGS__) \ 227c76fc982SDominik Brodowski __IA32_SYS_STUBx(x, name, __VA_ARGS__) \ 228e145242eSDominik Brodowski static long __se_sys##name(__MAP(x,__SC_LONG,__VA_ARGS__)) \ 229fa697140SDominik Brodowski { \ 230e145242eSDominik Brodowski long ret = __do_sys##name(__MAP(x,__SC_CAST,__VA_ARGS__));\ 231fa697140SDominik Brodowski __MAP(x,__SC_TEST,__VA_ARGS__); \ 232fa697140SDominik Brodowski __PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__)); \ 233fa697140SDominik Brodowski return ret; \ 234fa697140SDominik Brodowski } \ 235e145242eSDominik Brodowski static inline long __do_sys##name(__MAP(x,__SC_DECL,__VA_ARGS__)) 236fa697140SDominik Brodowski 237fa697140SDominik Brodowski /* 238d5a00528SDominik Brodowski * As the generic SYSCALL_DEFINE0() macro does not decode any parameters for 239d5a00528SDominik Brodowski * obvious reasons, and passing struct pt_regs *regs to it in %rdi does not 240d5a00528SDominik Brodowski * hurt, we only need to re-define it here to keep the naming congruent to 241766c6c1dSLinus Torvalds * SYSCALL_DEFINEx() -- which is essential for the COND_SYSCALL() macro 242766c6c1dSLinus Torvalds * to work correctly. 243d5a00528SDominik Brodowski */ 244d5a00528SDominik Brodowski #define SYSCALL_DEFINE0(sname) \ 245d5a00528SDominik Brodowski SYSCALL_METADATA(_##sname, 0); \ 2460f78ff17SBrian Gerst static long __do_sys_##sname(const struct pt_regs *__unused); \ 247d2b5de49SBrian Gerst __X64_SYS_STUB0(sname) \ 248d2b5de49SBrian Gerst __IA32_SYS_STUB0(sname) \ 2490f78ff17SBrian Gerst static long __do_sys_##sname(const struct pt_regs *__unused) 250d5a00528SDominik Brodowski 2516e484764SSami Tolvanen #define COND_SYSCALL(name) \ 2526cc8d2b2SBrian Gerst __X64_COND_SYSCALL(name) \ 2536cc8d2b2SBrian Gerst __IA32_COND_SYSCALL(name) 254d5a00528SDominik Brodowski 255d5a00528SDominik Brodowski 256d5a00528SDominik Brodowski /* 257fa697140SDominik Brodowski * For VSYSCALLS, we need to declare these three syscalls with the new 258fa697140SDominik Brodowski * pt_regs-based calling convention for in-kernel use. 259fa697140SDominik Brodowski */ 2600f78ff17SBrian Gerst long __x64_sys_getcpu(const struct pt_regs *regs); 2610f78ff17SBrian Gerst long __x64_sys_gettimeofday(const struct pt_regs *regs); 2620f78ff17SBrian Gerst long __x64_sys_time(const struct pt_regs *regs); 263fa697140SDominik Brodowski 264fa697140SDominik Brodowski #endif /* _ASM_X86_SYSCALL_WRAPPER_H */ 265