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 add_random_kstack_offset();
42 nr = syscall_enter_from_user_mode(regs, nr);
44 instrumentation_begin();
45 if (likely(nr < NR_syscalls)) {
46 nr = array_index_nospec(nr, NR_syscalls);
47 regs->ax = sys_call_table[nr](regs);
48 #ifdef CONFIG_X86_X32_ABI
49 } else if (likely((nr & __X32_SYSCALL_BIT) &&
50 (nr & ~__X32_SYSCALL_BIT) < X32_NR_syscalls)) {
51 nr = array_index_nospec(nr & ~__X32_SYSCALL_BIT,
53 regs->ax = x32_sys_call_table[nr](regs);
56 instrumentation_end();
57 syscall_exit_to_user_mode(regs);
61 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
62 static __always_inline unsigned int syscall_32_enter(struct pt_regs *regs)
64 if (IS_ENABLED(CONFIG_IA32_EMULATION))
65 current_thread_info()->status |= TS_COMPAT;
67 return (unsigned int)regs->orig_ax;
71 * Invoke a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL.
73 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs,
76 if (likely(nr < IA32_NR_syscalls)) {
77 nr = array_index_nospec(nr, IA32_NR_syscalls);
78 regs->ax = ia32_sys_call_table[nr](regs);
82 /* Handles int $0x80 */
83 __visible noinstr void do_int80_syscall_32(struct pt_regs *regs)
85 unsigned int nr = syscall_32_enter(regs);
87 add_random_kstack_offset();
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);
94 instrumentation_begin();
96 do_syscall_32_irqs_on(regs, nr);
98 instrumentation_end();
99 syscall_exit_to_user_mode(regs);
102 static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
104 unsigned int nr = syscall_32_enter(regs);
107 add_random_kstack_offset();
109 * This cannot use syscall_enter_from_user_mode() as it has to
110 * fetch EBP before invoking any of the syscall entry work
113 syscall_enter_from_user_mode_prepare(regs);
115 instrumentation_begin();
116 /* Fetch EBP from where the vDSO stashed it. */
117 if (IS_ENABLED(CONFIG_X86_64)) {
119 * Micro-optimization: the pointer we're following is
120 * explicitly 32 bits, so it can't be out of range.
122 res = __get_user(*(u32 *)®s->bp,
123 (u32 __user __force *)(unsigned long)(u32)regs->sp);
125 res = get_user(*(u32 *)®s->bp,
126 (u32 __user __force *)(unsigned long)(u32)regs->sp);
130 /* User code screwed up. */
133 instrumentation_end();
135 irqentry_exit_to_user_mode(regs);
139 /* The case truncates any ptrace induced syscall nr > 2^32 -1 */
140 nr = (unsigned int)syscall_enter_from_user_mode_work(regs, nr);
142 /* Now this is just like a normal syscall. */
143 do_syscall_32_irqs_on(regs, nr);
145 instrumentation_end();
146 syscall_exit_to_user_mode(regs);
150 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
151 __visible noinstr long do_fast_syscall_32(struct pt_regs *regs)
154 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
155 * convention. Adjust regs so it looks like we entered using int80.
157 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
158 vdso_image_32.sym_int80_landing_pad;
161 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
162 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
165 regs->ip = landing_pad;
167 /* Invoke the syscall. If it failed, keep it simple: use IRET. */
168 if (!__do_fast_syscall_32(regs))
173 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
174 * SYSRETL is available on all 64-bit CPUs, so we don't need to
175 * bother with SYSEXIT.
177 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
178 * because the ECX fixup above will ensure that this is essentially
181 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
182 regs->ip == landing_pad &&
183 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
186 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
188 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
189 * because the ECX fixup above will ensure that this is essentially
192 * We don't allow syscalls at all from VM86 mode, but we still
193 * need to check VM, because we might be returning from sys_vm86.
195 return static_cpu_has(X86_FEATURE_SEP) &&
196 regs->cs == __USER_CS && regs->ss == __USER_DS &&
197 regs->ip == landing_pad &&
198 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
202 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
203 __visible noinstr long do_SYSENTER_32(struct pt_regs *regs)
205 /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
208 /* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */
209 regs->flags |= X86_EFLAGS_IF;
211 return do_fast_syscall_32(regs);
215 SYSCALL_DEFINE0(ni_syscall)
221 #ifndef CONFIG_PREEMPTION
223 * Some hypercalls issued by the toolstack can take many 10s of
224 * seconds. Allow tasks running hypercalls via the privcmd driver to
225 * be voluntarily preempted even if full kernel preemption is
228 * Such preemptible hypercalls are bracketed by
229 * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
232 DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
233 EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall);
236 * In case of scheduling the flag must be cleared and restored after
237 * returning from schedule as the task might move to a different CPU.
239 static __always_inline bool get_and_clear_inhcall(void)
241 bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
243 __this_cpu_write(xen_in_preemptible_hcall, false);
247 static __always_inline void restore_inhcall(bool inhcall)
249 __this_cpu_write(xen_in_preemptible_hcall, inhcall);
252 static __always_inline bool get_and_clear_inhcall(void) { return false; }
253 static __always_inline void restore_inhcall(bool inhcall) { }
256 static void __xen_pv_evtchn_do_upcall(struct pt_regs *regs)
258 struct pt_regs *old_regs = set_irq_regs(regs);
260 inc_irq_stat(irq_hv_callback_count);
262 xen_hvm_evtchn_do_upcall();
264 set_irq_regs(old_regs);
267 __visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
269 irqentry_state_t state = irqentry_enter(regs);
272 run_sysvec_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs);
274 inhcall = get_and_clear_inhcall();
275 if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
276 instrumentation_begin();
277 irqentry_exit_cond_resched();
278 instrumentation_end();
279 restore_inhcall(inhcall);
281 irqentry_exit(regs, state);
284 #endif /* CONFIG_XEN_PV */