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 unsigned int nr = (unsigned int)regs->orig_ax;
65 if (IS_ENABLED(CONFIG_IA32_EMULATION))
66 current_thread_info()->status |= TS_COMPAT;
68 * Subtlety here: if ptrace pokes something larger than 2^32-1 into
69 * orig_ax, the unsigned int return value truncates it. This may
70 * or may not be necessary, but it matches the old asm behavior.
72 return (unsigned int)syscall_enter_from_user_mode(regs, nr);
76 * Invoke a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL.
78 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs,
81 if (likely(nr < IA32_NR_syscalls)) {
82 instrumentation_begin();
83 nr = array_index_nospec(nr, IA32_NR_syscalls);
84 regs->ax = ia32_sys_call_table[nr](regs);
85 instrumentation_end();
89 /* Handles int $0x80 */
90 __visible noinstr void do_int80_syscall_32(struct pt_regs *regs)
92 unsigned int nr = syscall_32_enter(regs);
94 do_syscall_32_irqs_on(regs, nr);
95 syscall_exit_to_user_mode(regs);
98 static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
100 unsigned int nr = syscall_32_enter(regs);
103 instrumentation_begin();
104 /* Fetch EBP from where the vDSO stashed it. */
105 if (IS_ENABLED(CONFIG_X86_64)) {
107 * Micro-optimization: the pointer we're following is
108 * explicitly 32 bits, so it can't be out of range.
110 res = __get_user(*(u32 *)®s->bp,
111 (u32 __user __force *)(unsigned long)(u32)regs->sp);
113 res = get_user(*(u32 *)®s->bp,
114 (u32 __user __force *)(unsigned long)(u32)regs->sp);
116 instrumentation_end();
119 /* User code screwed up. */
121 syscall_exit_to_user_mode(regs);
125 /* Now this is just like a normal syscall. */
126 do_syscall_32_irqs_on(regs, nr);
127 syscall_exit_to_user_mode(regs);
131 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
132 __visible noinstr long do_fast_syscall_32(struct pt_regs *regs)
135 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
136 * convention. Adjust regs so it looks like we entered using int80.
138 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
139 vdso_image_32.sym_int80_landing_pad;
142 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
143 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
146 regs->ip = landing_pad;
148 /* Invoke the syscall. If it failed, keep it simple: use IRET. */
149 if (!__do_fast_syscall_32(regs))
154 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
155 * SYSRETL is available on all 64-bit CPUs, so we don't need to
156 * bother with SYSEXIT.
158 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
159 * because the ECX fixup above will ensure that this is essentially
162 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
163 regs->ip == landing_pad &&
164 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
167 * Opportunistic SYSEXIT: if possible, try to return using 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 * We don't allow syscalls at all from VM86 mode, but we still
174 * need to check VM, because we might be returning from sys_vm86.
176 return static_cpu_has(X86_FEATURE_SEP) &&
177 regs->cs == __USER_CS && regs->ss == __USER_DS &&
178 regs->ip == landing_pad &&
179 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
183 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
184 __visible noinstr long do_SYSENTER_32(struct pt_regs *regs)
186 /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
189 /* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */
190 regs->flags |= X86_EFLAGS_IF;
192 return do_fast_syscall_32(regs);
196 SYSCALL_DEFINE0(ni_syscall)
201 noinstr bool idtentry_enter_nmi(struct pt_regs *regs)
203 bool irq_state = lockdep_hardirqs_enabled();
206 lockdep_hardirqs_off(CALLER_ADDR0);
207 lockdep_hardirq_enter();
210 instrumentation_begin();
211 trace_hardirqs_off_finish();
213 instrumentation_end();
218 noinstr void idtentry_exit_nmi(struct pt_regs *regs, bool restore)
220 instrumentation_begin();
223 trace_hardirqs_on_prepare();
224 lockdep_hardirqs_on_prepare(CALLER_ADDR0);
226 instrumentation_end();
229 lockdep_hardirq_exit();
231 lockdep_hardirqs_on(CALLER_ADDR0);
236 #ifndef CONFIG_PREEMPTION
238 * Some hypercalls issued by the toolstack can take many 10s of
239 * seconds. Allow tasks running hypercalls via the privcmd driver to
240 * be voluntarily preempted even if full kernel preemption is
243 * Such preemptible hypercalls are bracketed by
244 * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
247 DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
248 EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall);
251 * In case of scheduling the flag must be cleared and restored after
252 * returning from schedule as the task might move to a different CPU.
254 static __always_inline bool get_and_clear_inhcall(void)
256 bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
258 __this_cpu_write(xen_in_preemptible_hcall, false);
262 static __always_inline void restore_inhcall(bool inhcall)
264 __this_cpu_write(xen_in_preemptible_hcall, inhcall);
267 static __always_inline bool get_and_clear_inhcall(void) { return false; }
268 static __always_inline void restore_inhcall(bool inhcall) { }
271 static void __xen_pv_evtchn_do_upcall(void)
274 inc_irq_stat(irq_hv_callback_count);
276 xen_hvm_evtchn_do_upcall();
281 __visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
283 struct pt_regs *old_regs;
285 irqentry_state_t state;
287 state = irqentry_enter(regs);
288 old_regs = set_irq_regs(regs);
290 instrumentation_begin();
291 run_on_irqstack_cond(__xen_pv_evtchn_do_upcall, NULL, regs);
292 instrumentation_begin();
294 set_irq_regs(old_regs);
296 inhcall = get_and_clear_inhcall();
297 if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
298 instrumentation_begin();
299 irqentry_exit_cond_resched();
300 instrumentation_end();
301 restore_inhcall(inhcall);
303 irqentry_exit(regs, state);
306 #endif /* CONFIG_XEN_PV */