arch/x86/entry/common.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * common.c - C code for kernel entry and exit
   4  * Copyright (c) 2015 Andrew Lutomirski
   5  *
   6  * Based on asm and ptrace code by many authors.  The code here originated
   7  * in ptrace.c and signal.c.
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/sched/task_stack.h>
  13 #include <linux/entry-common.h>
  14 #include <linux/mm.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
  23 #ifdef CONFIG_XEN_PV
  24 #include <xen/xen-ops.h>
  25 #include <xen/events.h>
  26 #endif
  27
  28 #include <asm/desc.h>
  29 #include <asm/traps.h>
  30 #include <asm/vdso.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>
  37
  38 #ifdef CONFIG_X86_64
  39 __visible noinstr void do_syscall_64(unsigned long nr, struct pt_regs *regs)
  40 {
  41         nr = syscall_enter_from_user_mode(regs, nr);
  42
  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,
  51                                         X32_NR_syscalls);
  52                 regs->ax = x32_sys_call_table[nr](regs);
  53 #endif
  54         }
  55         instrumentation_end();
  56         syscall_exit_to_user_mode(regs);
  57 }
  58 #endif
  59
  60 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
  61 static __always_inline unsigned int syscall_32_enter(struct pt_regs *regs)
  62 {
  63         if (IS_ENABLED(CONFIG_IA32_EMULATION))
  64                 current_thread_info()->status |= TS_COMPAT;
  65
  66         return (unsigned int)regs->orig_ax;
  67 }
  68
  69 /*
  70  * Invoke a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.
  71  */
  72 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs,
  73                                                   unsigned int nr)
  74 {
  75         if (likely(nr < IA32_NR_syscalls)) {
  76                 nr = array_index_nospec(nr, IA32_NR_syscalls);
  77                 regs->ax = ia32_sys_call_table[nr](regs);
  78         }
  79 }
  80
  81 /* Handles int $0x80 */
  82 __visible noinstr void do_int80_syscall_32(struct pt_regs *regs)
  83 {
  84         unsigned int nr = syscall_32_enter(regs);
  85
  86         /*
  87          * Subtlety here: if ptrace pokes something larger than 2^32-1 into
  88          * orig_ax, the unsigned int return value truncates it.  This may
  89          * or may not be necessary, but it matches the old asm behavior.
  90          */
  91         nr = (unsigned int)syscall_enter_from_user_mode(regs, nr);
  92         instrumentation_begin();
  93
  94         do_syscall_32_irqs_on(regs, nr);
  95
  96         instrumentation_end();
  97         syscall_exit_to_user_mode(regs);
  98 }
  99
 100 static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
 101 {
 102         unsigned int nr = syscall_32_enter(regs);
 103         int res;
 104
 105         /*
 106          * This cannot use syscall_enter_from_user_mode() as it has to
 107          * fetch EBP before invoking any of the syscall entry work
 108          * functions.
 109          */
 110         syscall_enter_from_user_mode_prepare(regs);
 111
 112         instrumentation_begin();
 113         /* Fetch EBP from where the vDSO stashed it. */
 114         if (IS_ENABLED(CONFIG_X86_64)) {
 115                 /*
 116                  * Micro-optimization: the pointer we're following is
 117                  * explicitly 32 bits, so it can't be out of range.
 118                  */
 119                 res = __get_user(*(u32 *)&regs->bp,
 120                          (u32 __user __force *)(unsigned long)(u32)regs->sp);
 121         } else {
 122                 res = get_user(*(u32 *)&regs->bp,
 123                        (u32 __user __force *)(unsigned long)(u32)regs->sp);
 124         }
 125
 126         if (res) {
 127                 /* User code screwed up. */
 128                 regs->ax = -EFAULT;
 129
 130                 instrumentation_end();
 131                 syscall_exit_to_user_mode(regs);
 132                 return false;
 133         }
 134
 135         /* The case truncates any ptrace induced syscall nr > 2^32 -1 */
 136         nr = (unsigned int)syscall_enter_from_user_mode_work(regs, nr);
 137
 138         /* Now this is just like a normal syscall. */
 139         do_syscall_32_irqs_on(regs, nr);
 140
 141         instrumentation_end();
 142         syscall_exit_to_user_mode(regs);
 143         return true;
 144 }
 145
 146 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
 147 __visible noinstr long do_fast_syscall_32(struct pt_regs *regs)
 148 {
 149         /*
 150          * Called using the internal vDSO SYSENTER/SYSCALL32 calling
 151          * convention.  Adjust regs so it looks like we entered using int80.
 152          */
 153         unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
 154                                         vdso_image_32.sym_int80_landing_pad;
 155
 156         /*
 157          * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
 158          * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
 159          * Fix it up.
 160          */
 161         regs->ip = landing_pad;
 162
 163         /* Invoke the syscall. If it failed, keep it simple: use IRET. */
 164         if (!__do_fast_syscall_32(regs))
 165                 return 0;
 166
 167 #ifdef CONFIG_X86_64
 168         /*
 169          * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
 170          * SYSRETL is available on all 64-bit CPUs, so we don't need to
 171          * bother with SYSEXIT.
 172          *
 173          * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
 174          * because the ECX fixup above will ensure that this is essentially
 175          * never the case.
 176          */
 177         return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
 178                 regs->ip == landing_pad &&
 179                 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
 180 #else
 181         /*
 182          * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
 183          *
 184          * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
 185          * because the ECX fixup above will ensure that this is essentially
 186          * never the case.
 187          *
 188          * We don't allow syscalls at all from VM86 mode, but we still
 189          * need to check VM, because we might be returning from sys_vm86.
 190          */
 191         return static_cpu_has(X86_FEATURE_SEP) &&
 192                 regs->cs == __USER_CS && regs->ss == __USER_DS &&
 193                 regs->ip == landing_pad &&
 194                 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
 195 #endif
 196 }
 197
 198 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
 199 __visible noinstr long do_SYSENTER_32(struct pt_regs *regs)
 200 {
 201         /* SYSENTER loses RSP, but the vDSO saved it in RBP. */
 202         regs->sp = regs->bp;
 203
 204         /* SYSENTER clobbers EFLAGS.IF.  Assume it was set in usermode. */
 205         regs->flags |= X86_EFLAGS_IF;
 206
 207         return do_fast_syscall_32(regs);
 208 }
 209 #endif
 210
 211 SYSCALL_DEFINE0(ni_syscall)
 212 {
 213         return -ENOSYS;
 214 }
 215
 216 #ifdef CONFIG_XEN_PV
 217 #ifndef CONFIG_PREEMPTION
 218 /*
 219  * Some hypercalls issued by the toolstack can take many 10s of
 220  * seconds. Allow tasks running hypercalls via the privcmd driver to
 221  * be voluntarily preempted even if full kernel preemption is
 222  * disabled.
 223  *
 224  * Such preemptible hypercalls are bracketed by
 225  * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
 226  * calls.
 227  */
 228 DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
 229 EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall);
 230
 231 /*
 232  * In case of scheduling the flag must be cleared and restored after
 233  * returning from schedule as the task might move to a different CPU.
 234  */
 235 static __always_inline bool get_and_clear_inhcall(void)
 236 {
 237         bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
 238
 239         __this_cpu_write(xen_in_preemptible_hcall, false);
 240         return inhcall;
 241 }
 242
 243 static __always_inline void restore_inhcall(bool inhcall)
 244 {
 245         __this_cpu_write(xen_in_preemptible_hcall, inhcall);
 246 }
 247 #else
 248 static __always_inline bool get_and_clear_inhcall(void) { return false; }
 249 static __always_inline void restore_inhcall(bool inhcall) { }
 250 #endif
 251
 252 static void __xen_pv_evtchn_do_upcall(void)
 253 {
 254         irq_enter_rcu();
 255         inc_irq_stat(irq_hv_callback_count);
 256
 257         xen_hvm_evtchn_do_upcall();
 258
 259         irq_exit_rcu();
 260 }
 261
 262 __visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
 263 {
 264         struct pt_regs *old_regs;
 265         bool inhcall;
 266         irqentry_state_t state;
 267
 268         state = irqentry_enter(regs);
 269         old_regs = set_irq_regs(regs);
 270
 271         instrumentation_begin();
 272         run_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs);
 273         instrumentation_begin();
 274
 275         set_irq_regs(old_regs);
 276
 277         inhcall = get_and_clear_inhcall();
 278         if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
 279                 instrumentation_begin();
 280                 irqentry_exit_cond_resched();
 281                 instrumentation_end();
 282                 restore_inhcall(inhcall);
 283         } else {
 284                 irqentry_exit(regs, state);
 285         }
 286 }
 287 #endif /* CONFIG_XEN_PV */