Print this page
10924 Need mitigation of L1TF (CVE-2018-3646)
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Peter Tribble <peter.tribble@gmail.com>


  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25 
  26 /*
  27  * Copyright (c) 2009, Intel Corporation
  28  * All rights reserved.
  29  */
  30 
  31 /*       Copyright (c) 1990, 1991 UNIX System Laboratories, Inc.        */
  32 /*       Copyright (c) 1984, 1986, 1987, 1988, 1989, 1990 AT&T              */
  33 /*         All Rights Reserved                                          */
  34 
  35 /*       Copyright (c) 1987, 1988 Microsoft Corporation                 */
  36 /*         All Rights Reserved                                          */
  37 
  38 /*
  39  * Copyright 2016 Joyent, Inc.
  40  */
  41 
  42 #include <sys/errno.h>
  43 #include <sys/asm_linkage.h>
  44 
  45 #if defined(__lint)
  46 #include <sys/types.h>
  47 #include <sys/systm.h>
  48 #else   /* __lint */
  49 #include "assym.h"
  50 #endif  /* __lint */
  51 
  52 #define KCOPY_MIN_SIZE  128     /* Must be >= 16 bytes */
  53 #define XCOPY_MIN_SIZE  128     /* Must be >= 16 bytes */
  54 /*
  55  * Non-temopral access (NTA) alignment requirement
  56  */
  57 #define NTA_ALIGN_SIZE  4       /* Must be at least 4-byte aligned */
  58 #define NTA_ALIGN_MASK  _CONST(NTA_ALIGN_SIZE-1)
  59 #define COUNT_ALIGN_SIZE        16      /* Must be at least 16-byte aligned */


 849         movzbq -0x1(%rdi), %rcx
 850         mov    %r8d, -0x7(%rsi)
 851         mov    %r10w, -0x3(%rsi)
 852         mov    %cl, -0x1(%rsi)
 853         ret   
 854 
 855         /*
 856          * For large sizes rep smovq is fastest.
 857          * Transition point determined experimentally as measured on
 858          * Intel Xeon processors (incl. Nehalem and previous generations) and
 859          * AMD Opteron. The transition value is patched at boot time to avoid
 860          * memory reference hit.
 861          */
 862         .globl bcopy_patch_start
 863 bcopy_patch_start:
 864         cmpq    $BCOPY_NHM_REP, %rdx
 865         .globl bcopy_patch_end
 866 bcopy_patch_end:
 867 
 868         .p2align 4
 869         .globl bcopy_ck_size
 870 bcopy_ck_size:
 871         cmpq    $BCOPY_DFLT_REP, %rdx
 872         jae     L(use_rep)
 873 
 874         /*
 875          * Align to a 8-byte boundary. Avoids penalties from unaligned stores
 876          * as well as from stores spanning cachelines.
 877          */
 878         test    $0x7, %rsi
 879         jz      L(aligned_loop)
 880         test    $0x1, %rsi
 881         jz      2f
 882         movzbq  (%rdi), %r8
 883         dec     %rdx
 884         inc     %rdi
 885         mov     %r8b, (%rsi)
 886         inc     %rsi
 887 2:
 888         test    $0x2, %rsi
 889         jz      4f
 890         movzwq  (%rdi), %r8


 939         movslq  (%r10,%rdx,4), %rcx
 940         leaq    (%rcx,%r10,1), %r10
 941         jmpq    *%r10
 942 
 943         /*
 944          * Use rep smovq. Clear remainder via unrolled code
 945          */
 946         .p2align 4
 947 L(use_rep):
 948         xchgq   %rdi, %rsi              /* %rsi = source, %rdi = destination */
 949         movq    %rdx, %rcx              /* %rcx = count */
 950         shrq    $3, %rcx                /* 8-byte word count */
 951         rep
 952           smovq
 953 
 954         xchgq   %rsi, %rdi              /* %rdi = src, %rsi = destination */
 955         andq    $7, %rdx                /* remainder */
 956         jnz     L(do_remainder)
 957         ret
 958 #undef  L

 959 
 960 #ifdef DEBUG
 961         /*
 962          * Setup frame on the run-time stack. The end of the input argument
 963          * area must be aligned on a 16 byte boundary. The stack pointer %rsp,
 964          * always points to the end of the latest allocated stack frame.
 965          * panic(const char *format, ...) is a varargs function. When a
 966          * function taking variable arguments is called, %rax must be set
 967          * to eight times the number of floating point parameters passed
 968          * to the function in SSE registers.
 969          */
 970 call_panic:
 971         pushq   %rbp                    /* align stack properly */
 972         movq    %rsp, %rbp
 973         xorl    %eax, %eax              /* no variable arguments */
 974         call    panic                   /* %rdi = format string */
 975 #endif
 976         SET_SIZE(bcopy_altentry)
 977         SET_SIZE(bcopy)
 978 




  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25 
  26 /*
  27  * Copyright (c) 2009, Intel Corporation
  28  * All rights reserved.
  29  */
  30 
  31 /*       Copyright (c) 1990, 1991 UNIX System Laboratories, Inc.        */
  32 /*       Copyright (c) 1984, 1986, 1987, 1988, 1989, 1990 AT&T              */
  33 /*         All Rights Reserved                                          */
  34 
  35 /*       Copyright (c) 1987, 1988 Microsoft Corporation                 */
  36 /*         All Rights Reserved                                          */
  37 
  38 /*
  39  * Copyright (c) 2018 Joyent, Inc.
  40  */
  41 
  42 #include <sys/errno.h>
  43 #include <sys/asm_linkage.h>
  44 
  45 #if defined(__lint)
  46 #include <sys/types.h>
  47 #include <sys/systm.h>
  48 #else   /* __lint */
  49 #include "assym.h"
  50 #endif  /* __lint */
  51 
  52 #define KCOPY_MIN_SIZE  128     /* Must be >= 16 bytes */
  53 #define XCOPY_MIN_SIZE  128     /* Must be >= 16 bytes */
  54 /*
  55  * Non-temopral access (NTA) alignment requirement
  56  */
  57 #define NTA_ALIGN_SIZE  4       /* Must be at least 4-byte aligned */
  58 #define NTA_ALIGN_MASK  _CONST(NTA_ALIGN_SIZE-1)
  59 #define COUNT_ALIGN_SIZE        16      /* Must be at least 16-byte aligned */


 849         movzbq -0x1(%rdi), %rcx
 850         mov    %r8d, -0x7(%rsi)
 851         mov    %r10w, -0x3(%rsi)
 852         mov    %cl, -0x1(%rsi)
 853         ret   
 854 
 855         /*
 856          * For large sizes rep smovq is fastest.
 857          * Transition point determined experimentally as measured on
 858          * Intel Xeon processors (incl. Nehalem and previous generations) and
 859          * AMD Opteron. The transition value is patched at boot time to avoid
 860          * memory reference hit.
 861          */
 862         .globl bcopy_patch_start
 863 bcopy_patch_start:
 864         cmpq    $BCOPY_NHM_REP, %rdx
 865         .globl bcopy_patch_end
 866 bcopy_patch_end:
 867 
 868         .p2align 4
 869         ALTENTRY(bcopy_ck_size)
 870 
 871         cmpq    $BCOPY_DFLT_REP, %rdx
 872         jae     L(use_rep)
 873 
 874         /*
 875          * Align to a 8-byte boundary. Avoids penalties from unaligned stores
 876          * as well as from stores spanning cachelines.
 877          */
 878         test    $0x7, %rsi
 879         jz      L(aligned_loop)
 880         test    $0x1, %rsi
 881         jz      2f
 882         movzbq  (%rdi), %r8
 883         dec     %rdx
 884         inc     %rdi
 885         mov     %r8b, (%rsi)
 886         inc     %rsi
 887 2:
 888         test    $0x2, %rsi
 889         jz      4f
 890         movzwq  (%rdi), %r8


 939         movslq  (%r10,%rdx,4), %rcx
 940         leaq    (%rcx,%r10,1), %r10
 941         jmpq    *%r10
 942 
 943         /*
 944          * Use rep smovq. Clear remainder via unrolled code
 945          */
 946         .p2align 4
 947 L(use_rep):
 948         xchgq   %rdi, %rsi              /* %rsi = source, %rdi = destination */
 949         movq    %rdx, %rcx              /* %rcx = count */
 950         shrq    $3, %rcx                /* 8-byte word count */
 951         rep
 952           smovq
 953 
 954         xchgq   %rsi, %rdi              /* %rdi = src, %rsi = destination */
 955         andq    $7, %rdx                /* remainder */
 956         jnz     L(do_remainder)
 957         ret
 958 #undef  L
 959         SET_SIZE(bcopy_ck_size)
 960 
 961 #ifdef DEBUG
 962         /*
 963          * Setup frame on the run-time stack. The end of the input argument
 964          * area must be aligned on a 16 byte boundary. The stack pointer %rsp,
 965          * always points to the end of the latest allocated stack frame.
 966          * panic(const char *format, ...) is a varargs function. When a
 967          * function taking variable arguments is called, %rax must be set
 968          * to eight times the number of floating point parameters passed
 969          * to the function in SSE registers.
 970          */
 971 call_panic:
 972         pushq   %rbp                    /* align stack properly */
 973         movq    %rsp, %rbp
 974         xorl    %eax, %eax              /* no variable arguments */
 975         call    panic                   /* %rdi = format string */
 976 #endif
 977         SET_SIZE(bcopy_altentry)
 978         SET_SIZE(bcopy)
 979