arch/x86/entry/entry_64_compat.S

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /*
   3  * Compatibility mode system call entry point for x86-64.
   4  *
   5  * Copyright 2000-2002 Andi Kleen, SuSE Labs.
   6  */
   7 #include "calling.h"
   8 #include <asm/asm-offsets.h>
   9 #include <asm/current.h>
  10 #include <asm/errno.h>
  11 #include <asm/ia32_unistd.h>
  12 #include <asm/thread_info.h>
  13 #include <asm/segment.h>
  14 #include <asm/irqflags.h>
  15 #include <asm/asm.h>
  16 #include <asm/smap.h>
  17 #include <linux/linkage.h>
  18 #include <linux/err.h>
  19
  20         .section .entry.text, "ax"
  21
  22 /*
  23  * 32-bit SYSENTER entry.
  24  *
  25  * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
  26  * on 64-bit kernels running on Intel CPUs.
  27  *
  28  * The SYSENTER instruction, in principle, should *only* occur in the
  29  * vDSO.  In practice, a small number of Android devices were shipped
  30  * with a copy of Bionic that inlined a SYSENTER instruction.  This
  31  * never happened in any of Google's Bionic versions -- it only happened
  32  * in a narrow range of Intel-provided versions.
  33  *
  34  * SYSENTER loads SS, RSP, CS, and RIP from previously programmed MSRs.
  35  * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
  36  * SYSENTER does not save anything on the stack,
  37  * and does not save old RIP (!!!), RSP, or RFLAGS.
  38  *
  39  * Arguments:
  40  * eax  system call number
  41  * ebx  arg1
  42  * ecx  arg2
  43  * edx  arg3
  44  * esi  arg4
  45  * edi  arg5
  46  * ebp  user stack
  47  * 0(%ebp) arg6
  48  */
  49 ENTRY(entry_SYSENTER_compat)
  50         /* Interrupts are off on entry. */
  51         SWAPGS
  52
  53         /* We are about to clobber %rsp anyway, clobbering here is OK */
  54         SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp
  55
  56         movq    PER_CPU_VAR(cpu_current_top_of_stack), %rsp
  57
  58         /*
  59          * User tracing code (ptrace or signal handlers) might assume that
  60          * the saved RAX contains a 32-bit number when we're invoking a 32-bit
  61          * syscall.  Just in case the high bits are nonzero, zero-extend
  62          * the syscall number.  (This could almost certainly be deleted
  63          * with no ill effects.)
  64          */
  65         movl    %eax, %eax
  66
  67         /* Construct struct pt_regs on stack */
  68         pushq   $__USER32_DS            /* pt_regs->ss */
  69         pushq   %rbp                    /* pt_regs->sp (stashed in bp) */
  70
  71         /*
  72          * Push flags.  This is nasty.  First, interrupts are currently
  73          * off, but we need pt_regs->flags to have IF set.  Second, even
  74          * if TF was set when SYSENTER started, it's clear by now.  We fix
  75          * that later using TIF_SINGLESTEP.
  76          */
  77         pushfq                          /* pt_regs->flags (except IF = 0) */
  78         orl     $X86_EFLAGS_IF, (%rsp)  /* Fix saved flags */
  79         pushq   $__USER32_CS            /* pt_regs->cs */
  80         pushq   $0                      /* pt_regs->ip = 0 (placeholder) */
  81         pushq   %rax                    /* pt_regs->orig_ax */
  82         pushq   %rdi                    /* pt_regs->di */
  83         pushq   %rsi                    /* pt_regs->si */
  84         pushq   %rdx                    /* pt_regs->dx */
  85         pushq   %rcx                    /* pt_regs->cx */
  86         pushq   $-ENOSYS                /* pt_regs->ax */
  87         pushq   %r8                     /* pt_regs->r8 */
  88         xorl    %r8d, %r8d              /* nospec   r8 */
  89         pushq   %r9                     /* pt_regs->r9 */
  90         xorl    %r9d, %r9d              /* nospec   r9 */
  91         pushq   %r10                    /* pt_regs->r10 */
  92         xorl    %r10d, %r10d            /* nospec   r10 */
  93         pushq   %r11                    /* pt_regs->r11 */
  94         xorl    %r11d, %r11d            /* nospec   r11 */
  95         pushq   %rbx                    /* pt_regs->rbx */
  96         xorl    %ebx, %ebx              /* nospec   rbx */
  97         pushq   %rbp                    /* pt_regs->rbp (will be overwritten) */
  98         xorl    %ebp, %ebp              /* nospec   rbp */
  99         pushq   $0                      /* pt_regs->r12 = 0 */
 100         xorl    %r12d, %r12d            /* nospec   r12 */
 101         pushq   $0                      /* pt_regs->r13 = 0 */
 102         xorl    %r13d, %r13d            /* nospec   r13 */
 103         pushq   $0                      /* pt_regs->r14 = 0 */
 104         xorl    %r14d, %r14d            /* nospec   r14 */
 105         pushq   $0                      /* pt_regs->r15 = 0 */
 106         xorl    %r15d, %r15d            /* nospec   r15 */
 107         cld
 108
 109         /*
 110          * SYSENTER doesn't filter flags, so we need to clear NT and AC
 111          * ourselves.  To save a few cycles, we can check whether
 112          * either was set instead of doing an unconditional popfq.
 113          * This needs to happen before enabling interrupts so that
 114          * we don't get preempted with NT set.
 115          *
 116          * If TF is set, we will single-step all the way to here -- do_debug
 117          * will ignore all the traps.  (Yes, this is slow, but so is
 118          * single-stepping in general.  This allows us to avoid having
 119          * a more complicated code to handle the case where a user program
 120          * forces us to single-step through the SYSENTER entry code.)
 121          *
 122          * NB.: .Lsysenter_fix_flags is a label with the code under it moved
 123          * out-of-line as an optimization: NT is unlikely to be set in the
 124          * majority of the cases and instead of polluting the I$ unnecessarily,
 125          * we're keeping that code behind a branch which will predict as
 126          * not-taken and therefore its instructions won't be fetched.
 127          */
 128         testl   $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, EFLAGS(%rsp)
 129         jnz     .Lsysenter_fix_flags
 130 .Lsysenter_flags_fixed:
 131
 132         /*
 133          * User mode is traced as though IRQs are on, and SYSENTER
 134          * turned them off.
 135          */
 136         TRACE_IRQS_OFF
 137
 138         movq    %rsp, %rdi
 139         call    do_fast_syscall_32
 140         /* XEN PV guests always use IRET path */
 141         ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 142                     "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 143         jmp     sysret32_from_system_call
 144
 145 .Lsysenter_fix_flags:
 146         pushq   $X86_EFLAGS_FIXED
 147         popfq
 148         jmp     .Lsysenter_flags_fixed
 149 GLOBAL(__end_entry_SYSENTER_compat)
 150 ENDPROC(entry_SYSENTER_compat)
 151
 152 /*
 153  * 32-bit SYSCALL entry.
 154  *
 155  * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
 156  * on 64-bit kernels running on AMD CPUs.
 157  *
 158  * The SYSCALL instruction, in principle, should *only* occur in the
 159  * vDSO.  In practice, it appears that this really is the case.
 160  * As evidence:
 161  *
 162  *  - The calling convention for SYSCALL has changed several times without
 163  *    anyone noticing.
 164  *
 165  *  - Prior to the in-kernel X86_BUG_SYSRET_SS_ATTRS fixup, anything
 166  *    user task that did SYSCALL without immediately reloading SS
 167  *    would randomly crash.
 168  *
 169  *  - Most programmers do not directly target AMD CPUs, and the 32-bit
 170  *    SYSCALL instruction does not exist on Intel CPUs.  Even on AMD
 171  *    CPUs, Linux disables the SYSCALL instruction on 32-bit kernels
 172  *    because the SYSCALL instruction in legacy/native 32-bit mode (as
 173  *    opposed to compat mode) is sufficiently poorly designed as to be
 174  *    essentially unusable.
 175  *
 176  * 32-bit SYSCALL saves RIP to RCX, clears RFLAGS.RF, then saves
 177  * RFLAGS to R11, then loads new SS, CS, and RIP from previously
 178  * programmed MSRs.  RFLAGS gets masked by a value from another MSR
 179  * (so CLD and CLAC are not needed).  SYSCALL does not save anything on
 180  * the stack and does not change RSP.
 181  *
 182  * Note: RFLAGS saving+masking-with-MSR happens only in Long mode
 183  * (in legacy 32-bit mode, IF, RF and VM bits are cleared and that's it).
 184  * Don't get confused: RFLAGS saving+masking depends on Long Mode Active bit
 185  * (EFER.LMA=1), NOT on bitness of userspace where SYSCALL executes
 186  * or target CS descriptor's L bit (SYSCALL does not read segment descriptors).
 187  *
 188  * Arguments:
 189  * eax  system call number
 190  * ecx  return address
 191  * ebx  arg1
 192  * ebp  arg2    (note: not saved in the stack frame, should not be touched)
 193  * edx  arg3
 194  * esi  arg4
 195  * edi  arg5
 196  * esp  user stack
 197  * 0(%esp) arg6
 198  */
 199 ENTRY(entry_SYSCALL_compat)
 200         /* Interrupts are off on entry. */
 201         swapgs
 202
 203         /* Stash user ESP */
 204         movl    %esp, %r8d
 205
 206         /* Use %rsp as scratch reg. User ESP is stashed in r8 */
 207         SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp
 208
 209         /* Switch to the kernel stack */
 210         movq    PER_CPU_VAR(cpu_current_top_of_stack), %rsp
 211
 212         /* Construct struct pt_regs on stack */
 213         pushq   $__USER32_DS            /* pt_regs->ss */
 214         pushq   %r8                     /* pt_regs->sp */
 215         pushq   %r11                    /* pt_regs->flags */
 216         pushq   $__USER32_CS            /* pt_regs->cs */
 217         pushq   %rcx                    /* pt_regs->ip */
 218 GLOBAL(entry_SYSCALL_compat_after_hwframe)
 219         movl    %eax, %eax              /* discard orig_ax high bits */
 220         pushq   %rax                    /* pt_regs->orig_ax */
 221         pushq   %rdi                    /* pt_regs->di */
 222         pushq   %rsi                    /* pt_regs->si */
 223         xorl    %esi, %esi              /* nospec   si */
 224         pushq   %rdx                    /* pt_regs->dx */
 225         xorl    %edx, %edx              /* nospec   dx */
 226         pushq   %rbp                    /* pt_regs->cx (stashed in bp) */
 227         xorl    %ecx, %ecx              /* nospec   cx */
 228         pushq   $-ENOSYS                /* pt_regs->ax */
 229         pushq   $0                      /* pt_regs->r8  = 0 */
 230         xorl    %r8d, %r8d              /* nospec   r8 */
 231         pushq   $0                      /* pt_regs->r9  = 0 */
 232         xorl    %r9d, %r9d              /* nospec   r9 */
 233         pushq   $0                      /* pt_regs->r10 = 0 */
 234         xorl    %r10d, %r10d            /* nospec   r10 */
 235         pushq   $0                      /* pt_regs->r11 = 0 */
 236         xorl    %r11d, %r11d            /* nospec   r11 */
 237         pushq   %rbx                    /* pt_regs->rbx */
 238         xorl    %ebx, %ebx              /* nospec   rbx */
 239         pushq   %rbp                    /* pt_regs->rbp (will be overwritten) */
 240         xorl    %ebp, %ebp              /* nospec   rbp */
 241         pushq   $0                      /* pt_regs->r12 = 0 */
 242         xorl    %r12d, %r12d            /* nospec   r12 */
 243         pushq   $0                      /* pt_regs->r13 = 0 */
 244         xorl    %r13d, %r13d            /* nospec   r13 */
 245         pushq   $0                      /* pt_regs->r14 = 0 */
 246         xorl    %r14d, %r14d            /* nospec   r14 */
 247         pushq   $0                      /* pt_regs->r15 = 0 */
 248         xorl    %r15d, %r15d            /* nospec   r15 */
 249
 250         /*
 251          * User mode is traced as though IRQs are on, and SYSENTER
 252          * turned them off.
 253          */
 254         TRACE_IRQS_OFF
 255
 256         movq    %rsp, %rdi
 257         call    do_fast_syscall_32
 258         /* XEN PV guests always use IRET path */
 259         ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 260                     "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 261
 262         /* Opportunistic SYSRET */
 263 sysret32_from_system_call:
 264         TRACE_IRQS_ON                   /* User mode traces as IRQs on. */
 265         movq    RBX(%rsp), %rbx         /* pt_regs->rbx */
 266         movq    RBP(%rsp), %rbp         /* pt_regs->rbp */
 267         movq    EFLAGS(%rsp), %r11      /* pt_regs->flags (in r11) */
 268         movq    RIP(%rsp), %rcx         /* pt_regs->ip (in rcx) */
 269         addq    $RAX, %rsp              /* Skip r8-r15 */
 270         popq    %rax                    /* pt_regs->rax */
 271         popq    %rdx                    /* Skip pt_regs->cx */
 272         popq    %rdx                    /* pt_regs->dx */
 273         popq    %rsi                    /* pt_regs->si */
 274         popq    %rdi                    /* pt_regs->di */
 275
 276         /*
 277          * USERGS_SYSRET32 does:
 278          *  GSBASE = user's GS base
 279          *  EIP = ECX
 280          *  RFLAGS = R11
 281          *  CS = __USER32_CS
 282          *  SS = __USER_DS
 283          *
 284          * ECX will not match pt_regs->cx, but we're returning to a vDSO
 285          * trampoline that will fix up RCX, so this is okay.
 286          *
 287          * R12-R15 are callee-saved, so they contain whatever was in them
 288          * when the system call started, which is already known to user
 289          * code.  We zero R8-R10 to avoid info leaks.
 290          */
 291         movq    RSP-ORIG_RAX(%rsp), %rsp
 292
 293         /*
 294          * The original userspace %rsp (RSP-ORIG_RAX(%rsp)) is stored
 295          * on the process stack which is not mapped to userspace and
 296          * not readable after we SWITCH_TO_USER_CR3.  Delay the CR3
 297          * switch until after after the last reference to the process
 298          * stack.
 299          *
 300          * %r8/%r9 are zeroed before the sysret, thus safe to clobber.
 301          */
 302         SWITCH_TO_USER_CR3_NOSTACK scratch_reg=%r8 scratch_reg2=%r9
 303
 304         xorl    %r8d, %r8d
 305         xorl    %r9d, %r9d
 306         xorl    %r10d, %r10d
 307         swapgs
 308         sysretl
 309 END(entry_SYSCALL_compat)
 310
 311 /*
 312  * 32-bit legacy system call entry.
 313  *
 314  * 32-bit x86 Linux system calls traditionally used the INT $0x80
 315  * instruction.  INT $0x80 lands here.
 316  *
 317  * This entry point can be used by 32-bit and 64-bit programs to perform
 318  * 32-bit system calls.  Instances of INT $0x80 can be found inline in
 319  * various programs and libraries.  It is also used by the vDSO's
 320  * __kernel_vsyscall fallback for hardware that doesn't support a faster
 321  * entry method.  Restarted 32-bit system calls also fall back to INT
 322  * $0x80 regardless of what instruction was originally used to do the
 323  * system call.
 324  *
 325  * This is considered a slow path.  It is not used by most libc
 326  * implementations on modern hardware except during process startup.
 327  *
 328  * Arguments:
 329  * eax  system call number
 330  * ebx  arg1
 331  * ecx  arg2
 332  * edx  arg3
 333  * esi  arg4
 334  * edi  arg5
 335  * ebp  arg6
 336  */
 337 ENTRY(entry_INT80_compat)
 338         /*
 339          * Interrupts are off on entry.
 340          */
 341         ASM_CLAC                        /* Do this early to minimize exposure */
 342         SWAPGS
 343
 344         /*
 345          * User tracing code (ptrace or signal handlers) might assume that
 346          * the saved RAX contains a 32-bit number when we're invoking a 32-bit
 347          * syscall.  Just in case the high bits are nonzero, zero-extend
 348          * the syscall number.  (This could almost certainly be deleted
 349          * with no ill effects.)
 350          */
 351         movl    %eax, %eax
 352
 353         /* switch to thread stack expects orig_ax and rdi to be pushed */
 354         pushq   %rax                    /* pt_regs->orig_ax */
 355         pushq   %rdi                    /* pt_regs->di */
 356
 357         /* Need to switch before accessing the thread stack. */
 358         SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi
 359         movq    %rsp, %rdi
 360         movq    PER_CPU_VAR(cpu_current_top_of_stack), %rsp
 361
 362         pushq   6*8(%rdi)               /* regs->ss */
 363         pushq   5*8(%rdi)               /* regs->rsp */
 364         pushq   4*8(%rdi)               /* regs->eflags */
 365         pushq   3*8(%rdi)               /* regs->cs */
 366         pushq   2*8(%rdi)               /* regs->ip */
 367         pushq   1*8(%rdi)               /* regs->orig_ax */
 368
 369         pushq   (%rdi)                  /* pt_regs->di */
 370         pushq   %rsi                    /* pt_regs->si */
 371         xorl    %esi, %esi              /* nospec   si */
 372         pushq   %rdx                    /* pt_regs->dx */
 373         xorl    %edx, %edx              /* nospec   dx */
 374         pushq   %rcx                    /* pt_regs->cx */
 375         xorl    %ecx, %ecx              /* nospec   cx */
 376         pushq   $-ENOSYS                /* pt_regs->ax */
 377         pushq   $0                      /* pt_regs->r8  = 0 */
 378         xorl    %r8d, %r8d              /* nospec   r8 */
 379         pushq   $0                      /* pt_regs->r9  = 0 */
 380         xorl    %r9d, %r9d              /* nospec   r9 */
 381         pushq   $0                      /* pt_regs->r10 = 0 */
 382         xorl    %r10d, %r10d            /* nospec   r10 */
 383         pushq   $0                      /* pt_regs->r11 = 0 */
 384         xorl    %r11d, %r11d            /* nospec   r11 */
 385         pushq   %rbx                    /* pt_regs->rbx */
 386         xorl    %ebx, %ebx              /* nospec   rbx */
 387         pushq   %rbp                    /* pt_regs->rbp */
 388         xorl    %ebp, %ebp              /* nospec   rbp */
 389         pushq   %r12                    /* pt_regs->r12 */
 390         xorl    %r12d, %r12d            /* nospec   r12 */
 391         pushq   %r13                    /* pt_regs->r13 */
 392         xorl    %r13d, %r13d            /* nospec   r13 */
 393         pushq   %r14                    /* pt_regs->r14 */
 394         xorl    %r14d, %r14d            /* nospec   r14 */
 395         pushq   %r15                    /* pt_regs->r15 */
 396         xorl    %r15d, %r15d            /* nospec   r15 */
 397         cld
 398
 399         /*
 400          * User mode is traced as though IRQs are on, and the interrupt
 401          * gate turned them off.
 402          */
 403         TRACE_IRQS_OFF
 404
 405         movq    %rsp, %rdi
 406         call    do_int80_syscall_32
 407 .Lsyscall_32_done:
 408
 409         /* Go back to user mode. */
 410         TRACE_IRQS_ON
 411         jmp     swapgs_restore_regs_and_return_to_usermode
 412 END(entry_INT80_compat)