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>
22 #include <linux/init.h>
25 #include <xen/xen-ops.h>
26 #include <xen/events.h>
31 #include <asm/traps.h>
33 #include <asm/cpufeature.h>
34 #include <asm/fpu/api.h>
35 #include <asm/nospec-branch.h>
36 #include <asm/io_bitmap.h>
37 #include <asm/syscall.h>
38 #include <asm/irq_stack.h>
42 static __always_inline bool do_syscall_x64(struct pt_regs *regs, int nr)
45 * Convert negative numbers to very high and thus out of range
46 * numbers for comparisons.
48 unsigned int unr = nr;
50 if (likely(unr < NR_syscalls)) {
51 unr = array_index_nospec(unr, NR_syscalls);
52 regs->ax = x64_sys_call(regs, unr);
58 static __always_inline bool do_syscall_x32(struct pt_regs *regs, int nr)
61 * Adjust the starting offset of the table, and convert numbers
62 * < __X32_SYSCALL_BIT to very high and thus out of range
63 * numbers for comparisons.
65 unsigned int xnr = nr - __X32_SYSCALL_BIT;
67 if (IS_ENABLED(CONFIG_X86_X32_ABI) && likely(xnr < X32_NR_syscalls)) {
68 xnr = array_index_nospec(xnr, X32_NR_syscalls);
69 regs->ax = x32_sys_call(regs, xnr);
75 /* Returns true to return using SYSRET, or false to use IRET */
76 __visible noinstr bool do_syscall_64(struct pt_regs *regs, int nr)
78 add_random_kstack_offset();
79 nr = syscall_enter_from_user_mode(regs, nr);
81 instrumentation_begin();
83 if (!do_syscall_x64(regs, nr) && !do_syscall_x32(regs, nr) && nr != -1) {
84 /* Invalid system call, but still a system call. */
85 regs->ax = __x64_sys_ni_syscall(regs);
88 instrumentation_end();
89 syscall_exit_to_user_mode(regs);
92 * Check that the register state is valid for using SYSRET to exit
93 * to userspace. Otherwise use the slower but fully capable IRET
97 /* XEN PV guests always use the IRET path */
98 if (cpu_feature_enabled(X86_FEATURE_XENPV))
101 /* SYSRET requires RCX == RIP and R11 == EFLAGS */
102 if (unlikely(regs->cx != regs->ip || regs->r11 != regs->flags))
105 /* CS and SS must match the values set in MSR_STAR */
106 if (unlikely(regs->cs != __USER_CS || regs->ss != __USER_DS))
110 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
111 * in kernel space. This essentially lets the user take over
112 * the kernel, since userspace controls RSP.
114 * TASK_SIZE_MAX covers all user-accessible addresses other than
115 * the deprecated vsyscall page.
117 if (unlikely(regs->ip >= TASK_SIZE_MAX))
121 * SYSRET cannot restore RF. It can restore TF, but unlike IRET,
122 * restoring TF results in a trap from userspace immediately after
125 if (unlikely(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)))
128 /* Use SYSRET to exit to userspace */
133 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
134 static __always_inline int syscall_32_enter(struct pt_regs *regs)
136 if (IS_ENABLED(CONFIG_IA32_EMULATION))
137 current_thread_info()->status |= TS_COMPAT;
139 return (int)regs->orig_ax;
142 #ifdef CONFIG_IA32_EMULATION
143 bool __ia32_enabled __ro_after_init = !IS_ENABLED(CONFIG_IA32_EMULATION_DEFAULT_DISABLED);
145 static int ia32_emulation_override_cmdline(char *arg)
147 return kstrtobool(arg, &__ia32_enabled);
149 early_param("ia32_emulation", ia32_emulation_override_cmdline);
153 * Invoke a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL.
155 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs, int nr)
158 * Convert negative numbers to very high and thus out of range
159 * numbers for comparisons.
161 unsigned int unr = nr;
163 if (likely(unr < IA32_NR_syscalls)) {
164 unr = array_index_nospec(unr, IA32_NR_syscalls);
165 regs->ax = ia32_sys_call(regs, unr);
166 } else if (nr != -1) {
167 regs->ax = __ia32_sys_ni_syscall(regs);
171 #ifdef CONFIG_IA32_EMULATION
172 static __always_inline bool int80_is_external(void)
174 const unsigned int offs = (0x80 / 32) * 0x10;
175 const u32 bit = BIT(0x80 % 32);
177 /* The local APIC on XENPV guests is fake */
178 if (cpu_feature_enabled(X86_FEATURE_XENPV))
182 * If vector 0x80 is set in the APIC ISR then this is an external
183 * interrupt. Either from broken hardware or injected by a VMM.
185 * Note: In guest mode this is only valid for secure guests where
186 * the secure module fully controls the vAPIC exposed to the guest.
188 return apic_read(APIC_ISR + offs) & bit;
192 * do_int80_emulation - 32-bit legacy syscall C entry from asm
194 * This entry point can be used by 32-bit and 64-bit programs to perform
195 * 32-bit system calls. Instances of INT $0x80 can be found inline in
196 * various programs and libraries. It is also used by the vDSO's
197 * __kernel_vsyscall fallback for hardware that doesn't support a faster
198 * entry method. Restarted 32-bit system calls also fall back to INT
199 * $0x80 regardless of what instruction was originally used to do the
202 * This is considered a slow path. It is not used by most libc
203 * implementations on modern hardware except during process startup.
205 * The arguments for the INT $0x80 based syscall are on stack in the
207 * eax: system call number
208 * ebx, ecx, edx, esi, edi, ebp: arg1 - arg 6
210 __visible noinstr void do_int80_emulation(struct pt_regs *regs)
214 /* Kernel does not use INT $0x80! */
215 if (unlikely(!user_mode(regs))) {
216 irqentry_enter(regs);
217 instrumentation_begin();
218 panic("Unexpected external interrupt 0x80\n");
222 * Establish kernel context for instrumentation, including for
223 * int80_is_external() below which calls into the APIC driver.
224 * Identical for soft and external interrupts.
226 enter_from_user_mode(regs);
228 instrumentation_begin();
229 add_random_kstack_offset();
231 /* Validate that this is a soft interrupt to the extent possible */
232 if (unlikely(int80_is_external()))
233 panic("Unexpected external interrupt 0x80\n");
236 * The low level idtentry code pushed -1 into regs::orig_ax
237 * and regs::ax contains the syscall number.
239 * User tracing code (ptrace or signal handlers) might assume
240 * that the regs::orig_ax contains a 32-bit number on invoking
243 * Establish the syscall convention by saving the 32bit truncated
244 * syscall number in regs::orig_ax and by invalidating regs::ax.
246 regs->orig_ax = regs->ax & GENMASK(31, 0);
249 nr = syscall_32_enter(regs);
252 nr = syscall_enter_from_user_mode_work(regs, nr);
253 do_syscall_32_irqs_on(regs, nr);
255 instrumentation_end();
256 syscall_exit_to_user_mode(regs);
258 #else /* CONFIG_IA32_EMULATION */
260 /* Handles int $0x80 on a 32bit kernel */
261 __visible noinstr void do_int80_syscall_32(struct pt_regs *regs)
263 int nr = syscall_32_enter(regs);
265 add_random_kstack_offset();
267 * Subtlety here: if ptrace pokes something larger than 2^31-1 into
268 * orig_ax, the int return value truncates it. This matches
269 * the semantics of syscall_get_nr().
271 nr = syscall_enter_from_user_mode(regs, nr);
272 instrumentation_begin();
274 do_syscall_32_irqs_on(regs, nr);
276 instrumentation_end();
277 syscall_exit_to_user_mode(regs);
279 #endif /* !CONFIG_IA32_EMULATION */
281 static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
283 int nr = syscall_32_enter(regs);
286 add_random_kstack_offset();
288 * This cannot use syscall_enter_from_user_mode() as it has to
289 * fetch EBP before invoking any of the syscall entry work
292 syscall_enter_from_user_mode_prepare(regs);
294 instrumentation_begin();
295 /* Fetch EBP from where the vDSO stashed it. */
296 if (IS_ENABLED(CONFIG_X86_64)) {
298 * Micro-optimization: the pointer we're following is
299 * explicitly 32 bits, so it can't be out of range.
301 res = __get_user(*(u32 *)®s->bp,
302 (u32 __user __force *)(unsigned long)(u32)regs->sp);
304 res = get_user(*(u32 *)®s->bp,
305 (u32 __user __force *)(unsigned long)(u32)regs->sp);
309 /* User code screwed up. */
313 instrumentation_end();
314 irqentry_exit_to_user_mode(regs);
318 nr = syscall_enter_from_user_mode_work(regs, nr);
320 /* Now this is just like a normal syscall. */
321 do_syscall_32_irqs_on(regs, nr);
323 instrumentation_end();
324 syscall_exit_to_user_mode(regs);
328 /* Returns true to return using SYSEXIT/SYSRETL, or false to use IRET */
329 __visible noinstr bool do_fast_syscall_32(struct pt_regs *regs)
332 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
333 * convention. Adjust regs so it looks like we entered using int80.
335 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
336 vdso_image_32.sym_int80_landing_pad;
339 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
340 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
343 regs->ip = landing_pad;
345 /* Invoke the syscall. If it failed, keep it simple: use IRET. */
346 if (!__do_fast_syscall_32(regs))
350 * Check that the register state is valid for using SYSRETL/SYSEXIT
351 * to exit to userspace. Otherwise use the slower but fully capable
355 /* XEN PV guests always use the IRET path */
356 if (cpu_feature_enabled(X86_FEATURE_XENPV))
359 /* EIP must point to the VDSO landing pad */
360 if (unlikely(regs->ip != landing_pad))
363 /* CS and SS must match the values set in MSR_STAR */
364 if (unlikely(regs->cs != __USER32_CS || regs->ss != __USER_DS))
367 /* If the TF, RF, or VM flags are set, use IRET */
368 if (unlikely(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)))
371 /* Use SYSRETL/SYSEXIT to exit to userspace */
375 /* Returns true to return using SYSEXIT/SYSRETL, or false to use IRET */
376 __visible noinstr bool do_SYSENTER_32(struct pt_regs *regs)
378 /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
381 /* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */
382 regs->flags |= X86_EFLAGS_IF;
384 return do_fast_syscall_32(regs);
388 SYSCALL_DEFINE0(ni_syscall)
394 #ifndef CONFIG_PREEMPTION
396 * Some hypercalls issued by the toolstack can take many 10s of
397 * seconds. Allow tasks running hypercalls via the privcmd driver to
398 * be voluntarily preempted even if full kernel preemption is
401 * Such preemptible hypercalls are bracketed by
402 * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
405 DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
406 EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall);
409 * In case of scheduling the flag must be cleared and restored after
410 * returning from schedule as the task might move to a different CPU.
412 static __always_inline bool get_and_clear_inhcall(void)
414 bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
416 __this_cpu_write(xen_in_preemptible_hcall, false);
420 static __always_inline void restore_inhcall(bool inhcall)
422 __this_cpu_write(xen_in_preemptible_hcall, inhcall);
425 static __always_inline bool get_and_clear_inhcall(void) { return false; }
426 static __always_inline void restore_inhcall(bool inhcall) { }
429 static void __xen_pv_evtchn_do_upcall(struct pt_regs *regs)
431 struct pt_regs *old_regs = set_irq_regs(regs);
433 inc_irq_stat(irq_hv_callback_count);
435 xen_evtchn_do_upcall();
437 set_irq_regs(old_regs);
440 __visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
442 irqentry_state_t state = irqentry_enter(regs);
445 instrumentation_begin();
446 run_sysvec_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs);
448 inhcall = get_and_clear_inhcall();
449 if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
450 irqentry_exit_cond_resched();
451 instrumentation_end();
452 restore_inhcall(inhcall);
454 instrumentation_end();
455 irqentry_exit(regs, state);
458 #endif /* CONFIG_XEN_PV */