1 // SPDX-License-Identifier: GPL-2.0-only
3 * common.c - C code for kernel entry and exit
4 * Copyright (c) 2015 Andrew Lutomirski
6 * Based on asm and ptrace code by many authors. The code here originated
7 * in ptrace.c and signal.c.
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <linux/sched/task_stack.h>
13 #include <linux/entry-common.h>
15 #include <linux/smp.h>
16 #include <linux/errno.h>
17 #include <linux/ptrace.h>
18 #include <linux/export.h>
19 #include <linux/nospec.h>
20 #include <linux/syscalls.h>
21 #include <linux/uaccess.h>
24 #include <xen/xen-ops.h>
25 #include <xen/events.h>
29 #include <asm/traps.h>
31 #include <asm/cpufeature.h>
32 #include <asm/fpu/api.h>
33 #include <asm/nospec-branch.h>
34 #include <asm/io_bitmap.h>
35 #include <asm/syscall.h>
36 #include <asm/irq_stack.h>
39 __visible noinstr void do_syscall_64(unsigned long nr, struct pt_regs *regs)
41 nr = syscall_enter_from_user_mode(regs, nr);
43 instrumentation_begin();
44 if (likely(nr < NR_syscalls)) {
45 nr = array_index_nospec(nr, NR_syscalls);
46 regs->ax = sys_call_table[nr](regs);
47 #ifdef CONFIG_X86_X32_ABI
48 } else if (likely((nr & __X32_SYSCALL_BIT) &&
49 (nr & ~__X32_SYSCALL_BIT) < X32_NR_syscalls)) {
50 nr = array_index_nospec(nr & ~__X32_SYSCALL_BIT,
52 regs->ax = x32_sys_call_table[nr](regs);
55 instrumentation_end();
56 syscall_exit_to_user_mode(regs);
60 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
61 static __always_inline unsigned int syscall_32_enter(struct pt_regs *regs)
63 if (IS_ENABLED(CONFIG_IA32_EMULATION))
64 current_thread_info()->status |= TS_COMPAT;
66 return (unsigned int)regs->orig_ax;
70 * Invoke a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL.
72 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs,
75 if (likely(nr < IA32_NR_syscalls)) {
76 instrumentation_begin();
77 nr = array_index_nospec(nr, IA32_NR_syscalls);
78 regs->ax = ia32_sys_call_table[nr](regs);
79 instrumentation_end();
83 /* Handles int $0x80 */
84 __visible noinstr void do_int80_syscall_32(struct pt_regs *regs)
86 unsigned int nr = syscall_32_enter(regs);
89 * Subtlety here: if ptrace pokes something larger than 2^32-1 into
90 * orig_ax, the unsigned int return value truncates it. This may
91 * or may not be necessary, but it matches the old asm behavior.
93 nr = (unsigned int)syscall_enter_from_user_mode(regs, nr);
95 do_syscall_32_irqs_on(regs, nr);
96 syscall_exit_to_user_mode(regs);
99 static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
101 unsigned int nr = syscall_32_enter(regs);
105 * This cannot use syscall_enter_from_user_mode() as it has to
106 * fetch EBP before invoking any of the syscall entry work
109 syscall_enter_from_user_mode_prepare(regs);
111 instrumentation_begin();
112 /* Fetch EBP from where the vDSO stashed it. */
113 if (IS_ENABLED(CONFIG_X86_64)) {
115 * Micro-optimization: the pointer we're following is
116 * explicitly 32 bits, so it can't be out of range.
118 res = __get_user(*(u32 *)®s->bp,
119 (u32 __user __force *)(unsigned long)(u32)regs->sp);
121 res = get_user(*(u32 *)®s->bp,
122 (u32 __user __force *)(unsigned long)(u32)regs->sp);
124 instrumentation_end();
127 /* User code screwed up. */
129 syscall_exit_to_user_mode(regs);
133 /* The case truncates any ptrace induced syscall nr > 2^32 -1 */
134 nr = (unsigned int)syscall_enter_from_user_mode_work(regs, nr);
136 /* Now this is just like a normal syscall. */
137 do_syscall_32_irqs_on(regs, nr);
138 syscall_exit_to_user_mode(regs);
142 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
143 __visible noinstr long do_fast_syscall_32(struct pt_regs *regs)
146 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
147 * convention. Adjust regs so it looks like we entered using int80.
149 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
150 vdso_image_32.sym_int80_landing_pad;
153 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
154 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
157 regs->ip = landing_pad;
159 /* Invoke the syscall. If it failed, keep it simple: use IRET. */
160 if (!__do_fast_syscall_32(regs))
165 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
166 * SYSRETL is available on all 64-bit CPUs, so we don't need to
167 * bother with SYSEXIT.
169 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
170 * because the ECX fixup above will ensure that this is essentially
173 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
174 regs->ip == landing_pad &&
175 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
178 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
180 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
181 * because the ECX fixup above will ensure that this is essentially
184 * We don't allow syscalls at all from VM86 mode, but we still
185 * need to check VM, because we might be returning from sys_vm86.
187 return static_cpu_has(X86_FEATURE_SEP) &&
188 regs->cs == __USER_CS && regs->ss == __USER_DS &&
189 regs->ip == landing_pad &&
190 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
194 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
195 __visible noinstr long do_SYSENTER_32(struct pt_regs *regs)
197 /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
200 /* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */
201 regs->flags |= X86_EFLAGS_IF;
203 return do_fast_syscall_32(regs);
207 SYSCALL_DEFINE0(ni_syscall)
212 noinstr bool idtentry_enter_nmi(struct pt_regs *regs)
214 bool irq_state = lockdep_hardirqs_enabled();
217 lockdep_hardirqs_off(CALLER_ADDR0);
218 lockdep_hardirq_enter();
221 instrumentation_begin();
222 trace_hardirqs_off_finish();
224 instrumentation_end();
229 noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore)
231 instrumentation_begin();
234 trace_hardirqs_on_prepare();
235 lockdep_hardirqs_on_prepare(CALLER_ADDR0);
237 instrumentation_end();
240 lockdep_hardirq_exit();
242 lockdep_hardirqs_on(CALLER_ADDR0);
247 #ifndef CONFIG_PREEMPTION
249 * Some hypercalls issued by the toolstack can take many 10s of
250 * seconds. Allow tasks running hypercalls via the privcmd driver to
251 * be voluntarily preempted even if full kernel preemption is
254 * Such preemptible hypercalls are bracketed by
255 * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
258 DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
259 EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall);
262 * In case of scheduling the flag must be cleared and restored after
263 * returning from schedule as the task might move to a different CPU.
265 static __always_inline bool get_and_clear_inhcall(void)
267 bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
269 __this_cpu_write(xen_in_preemptible_hcall, false);
273 static __always_inline void restore_inhcall(bool inhcall)
275 __this_cpu_write(xen_in_preemptible_hcall, inhcall);
278 static __always_inline bool get_and_clear_inhcall(void) { return false; }
279 static __always_inline void restore_inhcall(bool inhcall) { }
282 static void __xen_pv_evtchn_do_upcall(void)
285 inc_irq_stat(irq_hv_callback_count);
287 xen_hvm_evtchn_do_upcall();
292 __visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
294 struct pt_regs *old_regs;
296 irqentry_state_t state;
298 state = irqentry_enter(regs);
299 old_regs = set_irq_regs(regs);
301 instrumentation_begin();
302 run_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs);
303 instrumentation_begin();
305 set_irq_regs(old_regs);
307 inhcall = get_and_clear_inhcall();
308 if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
309 instrumentation_begin();
310 irqentry_exit_cond_resched();
311 instrumentation_end();
312 restore_inhcall(inhcall);
314 irqentry_exit(regs, state);
317 #endif /* CONFIG_XEN_PV */