Print this page
    
manifest
    
      
        | Split | 
	Close | 
      
      | Expand all | 
      | Collapse all | 
    
    
          --- old/usr/src/uts/intel/os/archdep.c
          +++ new/usr/src/uts/intel/os/archdep.c
   1    1  /*
   2    2   * CDDL HEADER START
   3    3   *
   4    4   * The contents of this file are subject to the terms of the
   5    5   * Common Development and Distribution License (the "License").
   6    6   * You may not use this file except in compliance with the License.
   7    7   *
   8    8   * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9    9   * or http://www.opensolaris.org/os/licensing.
  10   10   * See the License for the specific language governing permissions
  11   11   * and limitations under the License.
  12   12   *
  13   13   * When distributing Covered Code, include this CDDL HEADER in each
  14   14   * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15   15   * If applicable, add the following below this CDDL HEADER, with the
  16   16   * fields enclosed by brackets "[]" replaced with your own identifying
  17   17   * information: Portions Copyright [yyyy] [name of copyright owner]
  18   18   *
  19   19   * CDDL HEADER END
  
    | 
      ↓ open down ↓ | 
    19 lines elided | 
    
      ↑ open up ↑ | 
  
  20   20   */
  21   21  /*
  22   22   * Copyright (c) 1992, 2010, Oracle and/or its affiliates. All rights reserved.
  23   23   */
  24   24  
  25   25  /*      Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
  26   26  /*        All Rights Reserved   */
  27   27  /*
  28   28   * Copyright (c) 2018, Joyent, Inc.
  29   29   * Copyright 2012 Nexenta Systems, Inc.  All rights reserved.
  30      - * Copyright 2022 Oxide Computer Company
       30 + * Copyright 2023 Oxide Computer Company
  31   31   */
  32   32  
  33   33  #include <sys/param.h>
  34   34  #include <sys/types.h>
  35   35  #include <sys/vmparam.h>
  36   36  #include <sys/systm.h>
  37   37  #include <sys/signal.h>
  38   38  #include <sys/stack.h>
  39   39  #include <sys/regset.h>
  40   40  #include <sys/privregs.h>
  41   41  #include <sys/frame.h>
  42   42  #include <sys/proc.h>
  43   43  #include <sys/psw.h>
  44   44  #include <sys/siginfo.h>
  45   45  #include <sys/cpuvar.h>
  46   46  #include <sys/asm_linkage.h>
  47   47  #include <sys/kmem.h>
  48   48  #include <sys/errno.h>
  49   49  #include <sys/bootconf.h>
  50   50  #include <sys/archsystm.h>
  51   51  #include <sys/debug.h>
  52   52  #include <sys/elf.h>
  53   53  #include <sys/spl.h>
  54   54  #include <sys/time.h>
  55   55  #include <sys/atomic.h>
  56   56  #include <sys/sysmacros.h>
  57   57  #include <sys/cmn_err.h>
  58   58  #include <sys/modctl.h>
  59   59  #include <sys/kobj.h>
  60   60  #include <sys/panic.h>
  61   61  #include <sys/reboot.h>
  62   62  #include <sys/time.h>
  63   63  #include <sys/fp.h>
  64   64  #include <sys/x86_archext.h>
  65   65  #include <sys/auxv.h>
  66   66  #include <sys/auxv_386.h>
  67   67  #include <sys/dtrace.h>
  68   68  #include <sys/brand.h>
  69   69  #include <sys/machbrand.h>
  70   70  #include <sys/cmn_err.h>
  71   71  
  72   72  /*
  73   73   * Map an fnsave-formatted save area into an fxsave-formatted save area.
  74   74   *
  75   75   * Most fields are the same width, content and semantics.  However
  76   76   * the tag word is compressed.
  77   77   */
  78   78  static void
  79   79  fnsave_to_fxsave(const struct fnsave_state *fn, struct fxsave_state *fx)
  80   80  {
  81   81          uint_t i, tagbits;
  82   82  
  83   83          fx->fx_fcw = fn->f_fcw;
  84   84          fx->fx_fsw = fn->f_fsw;
  85   85  
  86   86          /*
  87   87           * copy element by element (because of holes)
  88   88           */
  89   89          for (i = 0; i < 8; i++)
  90   90                  bcopy(&fn->f_st[i].fpr_16[0], &fx->fx_st[i].fpr_16[0],
  91   91                      sizeof (fn->f_st[0].fpr_16)); /* 80-bit x87-style floats */
  92   92  
  93   93          /*
  94   94           * synthesize compressed tag bits
  95   95           */
  96   96          fx->fx_fctw = 0;
  97   97          for (tagbits = fn->f_ftw, i = 0; i < 8; i++, tagbits >>= 2)
  98   98                  if ((tagbits & 3) != 3)
  99   99                          fx->fx_fctw |= (1 << i);
 100  100  
 101  101          fx->fx_fop = fn->f_fop;
 102  102  
 103  103          fx->fx_rip = (uint64_t)fn->f_eip;
 104  104          fx->fx_rdp = (uint64_t)fn->f_dp;
 105  105  }
 106  106  
 107  107  /*
 108  108   * Map from an fxsave-format save area to an fnsave-format save area.
 109  109   */
 110  110  static void
 111  111  fxsave_to_fnsave(const struct fxsave_state *fx, struct fnsave_state *fn)
 112  112  {
 113  113          uint_t i, top, tagbits;
 114  114  
 115  115          fn->f_fcw = fx->fx_fcw;
 116  116          fn->__f_ign0 = 0;
 117  117          fn->f_fsw = fx->fx_fsw;
 118  118          fn->__f_ign1 = 0;
 119  119  
 120  120          top = (fx->fx_fsw & FPS_TOP) >> 11;
 121  121  
 122  122          /*
 123  123           * copy element by element (because of holes)
 124  124           */
 125  125          for (i = 0; i < 8; i++)
 126  126                  bcopy(&fx->fx_st[i].fpr_16[0], &fn->f_st[i].fpr_16[0],
 127  127                      sizeof (fn->f_st[0].fpr_16)); /* 80-bit x87-style floats */
 128  128  
 129  129          /*
 130  130           * synthesize uncompressed tag bits
 131  131           */
 132  132          fn->f_ftw = 0;
 133  133          for (tagbits = fx->fx_fctw, i = 0; i < 8; i++, tagbits >>= 1) {
 134  134                  uint_t ibit, expo;
 135  135                  const uint16_t *fpp;
 136  136                  static const uint16_t zero[5] = { 0, 0, 0, 0, 0 };
 137  137  
 138  138                  if ((tagbits & 1) == 0) {
 139  139                          fn->f_ftw |= 3 << (i << 1);     /* empty */
 140  140                          continue;
 141  141                  }
 142  142  
 143  143                  /*
 144  144                   * (tags refer to *physical* registers)
 145  145                   */
 146  146                  fpp = &fx->fx_st[(i - top + 8) & 7].fpr_16[0];
 147  147                  ibit = fpp[3] >> 15;
 148  148                  expo = fpp[4] & 0x7fff;
 149  149  
 150  150                  if (ibit && expo != 0 && expo != 0x7fff)
 151  151                          continue;                       /* valid fp number */
 152  152  
 153  153                  if (bcmp(fpp, &zero, sizeof (zero)))
 154  154                          fn->f_ftw |= 2 << (i << 1);     /* NaN */
 155  155                  else
 156  156                          fn->f_ftw |= 1 << (i << 1);     /* fp zero */
 157  157          }
 158  158  
 159  159          fn->f_fop = fx->fx_fop;
 160  160  
 161  161          fn->__f_ign2 = 0;
 162  162          fn->f_eip = (uint32_t)fx->fx_rip;
 163  163          fn->f_cs = U32CS_SEL;
 164  164          fn->f_dp = (uint32_t)fx->fx_rdp;
 165  165          fn->f_ds = UDS_SEL;
 166  166          fn->__f_ign3 = 0;
 167  167  }
 168  168  
 169  169  /*
 170  170   * Map from an fpregset_t into an fxsave-format save area
 171  171   */
 172  172  static void
 173  173  fpregset_to_fxsave(const fpregset_t *fp, struct fxsave_state *fx)
 174  174  {
 175  175          bcopy(fp, fx, sizeof (*fx));
 176  176          /*
 177  177           * avoid useless #gp exceptions - mask reserved bits
 178  178           */
 179  179          fx->fx_mxcsr &= sse_mxcsr_mask;
 180  180  }
 181  181  
 182  182  /*
 183  183   * Map from an fxsave-format save area into a fpregset_t
 184  184   */
 185  185  static void
 186  186  fxsave_to_fpregset(const struct fxsave_state *fx, fpregset_t *fp)
 187  187  {
 188  188          bcopy(fx, fp, sizeof (*fx));
 189  189  }
 190  190  
 191  191  #if defined(_SYSCALL32_IMPL)
 192  192  static void
 193  193  fpregset32_to_fxsave(const fpregset32_t *fp, struct fxsave_state *fx)
 194  194  {
 195  195          const struct fpchip32_state *fc = &fp->fp_reg_set.fpchip_state;
 196  196  
 197  197          fnsave_to_fxsave((const struct fnsave_state *)fc, fx);
 198  198          /*
 199  199           * avoid useless #gp exceptions - mask reserved bits
 200  200           */
 201  201          fx->fx_mxcsr = sse_mxcsr_mask & fc->mxcsr;
 202  202          bcopy(&fc->xmm[0], &fx->fx_xmm[0], sizeof (fc->xmm));
 203  203  }
 204  204  
 205  205  static void
 206  206  fxsave_to_fpregset32(const struct fxsave_state *fx, fpregset32_t *fp)
 207  207  {
 208  208          struct fpchip32_state *fc = &fp->fp_reg_set.fpchip_state;
 209  209  
 210  210          fxsave_to_fnsave(fx, (struct fnsave_state *)fc);
 211  211          fc->mxcsr = fx->fx_mxcsr;
 212  212          bcopy(&fx->fx_xmm[0], &fc->xmm[0], sizeof (fc->xmm));
 213  213  }
 214  214  
 215  215  static void
 216  216  fpregset_nto32(const fpregset_t *src, fpregset32_t *dst)
 217  217  {
 218  218          fxsave_to_fpregset32((struct fxsave_state *)src, dst);
 219  219          dst->fp_reg_set.fpchip_state.status =
 220  220              src->fp_reg_set.fpchip_state.status;
 221  221          dst->fp_reg_set.fpchip_state.xstatus =
 222  222              src->fp_reg_set.fpchip_state.xstatus;
 223  223  }
 224  224  
 225  225  static void
 226  226  fpregset_32ton(const fpregset32_t *src, fpregset_t *dst)
 227  227  {
 228  228          fpregset32_to_fxsave(src, (struct fxsave_state *)dst);
 229  229          dst->fp_reg_set.fpchip_state.status =
 230  230              src->fp_reg_set.fpchip_state.status;
 231  231          dst->fp_reg_set.fpchip_state.xstatus =
  
    | 
      ↓ open down ↓ | 
    191 lines elided | 
    
      ↑ open up ↑ | 
  
 232  232              src->fp_reg_set.fpchip_state.xstatus;
 233  233  }
 234  234  #endif
 235  235  
 236  236  /*
 237  237   * Set floating-point registers from a native fpregset_t.
 238  238   */
 239  239  void
 240  240  setfpregs(klwp_t *lwp, fpregset_t *fp)
 241  241  {
 242      -        struct fpu_ctx *fpu = &lwp->lwp_pcb.pcb_fpu;
 243      -
 244      -        if (fpu->fpu_flags & FPU_EN) {
 245      -                if (!(fpu->fpu_flags & FPU_VALID)) {
 246      -                        /*
 247      -                         * FPU context is still active, release the
 248      -                         * ownership.
 249      -                         */
 250      -                        fp_free(fpu);
 251      -                }
 252      -        }
 253      -        /*
 254      -         * Else: if we are trying to change the FPU state of a thread which
 255      -         * hasn't yet initialized floating point, store the state in
 256      -         * the pcb and indicate that the state is valid.  When the
 257      -         * thread enables floating point, it will use this state instead
 258      -         * of the default state.
 259      -         */
 260      -
 261      -        switch (fp_save_mech) {
 262      -        case FP_FXSAVE:
 263      -                fpregset_to_fxsave(fp, fpu->fpu_regs.kfpu_u.kfpu_fx);
 264      -                fpu->fpu_regs.kfpu_xstatus =
 265      -                    fp->fp_reg_set.fpchip_state.xstatus;
 266      -                break;
 267      -
 268      -        case FP_XSAVE:
 269      -                fpregset_to_fxsave(fp,
 270      -                    &fpu->fpu_regs.kfpu_u.kfpu_xs->xs_fxsave);
 271      -                fpu->fpu_regs.kfpu_xstatus =
 272      -                    fp->fp_reg_set.fpchip_state.xstatus;
 273      -                fpu->fpu_regs.kfpu_u.kfpu_xs->xs_header.xsh_xstate_bv |=
 274      -                    (XFEATURE_LEGACY_FP | XFEATURE_SSE);
 275      -                break;
 276      -        default:
 277      -                panic("Invalid fp_save_mech");
 278      -                /*NOTREACHED*/
 279      -        }
 280      -
 281      -        fpu->fpu_regs.kfpu_status = fp->fp_reg_set.fpchip_state.status;
 282      -        fpu->fpu_flags |= FPU_VALID;
 283      -        PCB_SET_UPDATE_FPU(&lwp->lwp_pcb);
      242 +        fpu_set_fpregset(lwp, fp);
 284  243  }
 285  244  
 286  245  /*
 287  246   * Get floating-point registers into a native fpregset_t.
 288  247   */
 289  248  void
 290  249  getfpregs(klwp_t *lwp, fpregset_t *fp)
 291  250  {
 292      -        struct fpu_ctx *fpu = &lwp->lwp_pcb.pcb_fpu;
 293      -
 294      -        kpreempt_disable();
 295      -        if (fpu->fpu_flags & FPU_EN) {
 296      -                /*
 297      -                 * If we have FPU hw and the thread's pcb doesn't have
 298      -                 * a valid FPU state then get the state from the hw.
 299      -                 */
 300      -                if (fpu_exists && ttolwp(curthread) == lwp &&
 301      -                    !(fpu->fpu_flags & FPU_VALID))
 302      -                        fp_save(fpu); /* get the current FPU state */
 303      -        }
 304      -
 305      -        /*
 306      -         * There are 3 possible cases we have to be aware of here:
 307      -         *
 308      -         * 1. FPU is enabled.  FPU state is stored in the current LWP.
 309      -         *
 310      -         * 2. FPU is not enabled, and there have been no intervening /proc
 311      -         *    modifications.  Return initial FPU state.
 312      -         *
 313      -         * 3. FPU is not enabled, but a /proc consumer has modified FPU state.
 314      -         *    FPU state is stored in the current LWP.
 315      -         */
 316      -        if ((fpu->fpu_flags & FPU_EN) || (fpu->fpu_flags & FPU_VALID)) {
 317      -                /*
 318      -                 * Cases 1 and 3.
 319      -                 */
 320      -                switch (fp_save_mech) {
 321      -                case FP_FXSAVE:
 322      -                        fxsave_to_fpregset(fpu->fpu_regs.kfpu_u.kfpu_fx, fp);
 323      -                        fp->fp_reg_set.fpchip_state.xstatus =
 324      -                            fpu->fpu_regs.kfpu_xstatus;
 325      -                        break;
 326      -                case FP_XSAVE:
 327      -                        fxsave_to_fpregset(
 328      -                            &fpu->fpu_regs.kfpu_u.kfpu_xs->xs_fxsave, fp);
 329      -                        fp->fp_reg_set.fpchip_state.xstatus =
 330      -                            fpu->fpu_regs.kfpu_xstatus;
 331      -                        break;
 332      -                default:
 333      -                        panic("Invalid fp_save_mech");
 334      -                        /*NOTREACHED*/
 335      -                }
 336      -                fp->fp_reg_set.fpchip_state.status = fpu->fpu_regs.kfpu_status;
 337      -        } else {
 338      -                /*
 339      -                 * Case 2.
 340      -                 */
 341      -                switch (fp_save_mech) {
 342      -                case FP_FXSAVE:
 343      -                case FP_XSAVE:
 344      -                        /*
 345      -                         * For now, we don't have any AVX specific field in ABI.
 346      -                         * If we add any in the future, we need to initial them
 347      -                         * as well.
 348      -                         */
 349      -                        fxsave_to_fpregset(&sse_initial, fp);
 350      -                        fp->fp_reg_set.fpchip_state.xstatus =
 351      -                            fpu->fpu_regs.kfpu_xstatus;
 352      -                        break;
 353      -                default:
 354      -                        panic("Invalid fp_save_mech");
 355      -                        /*NOTREACHED*/
 356      -                }
 357      -                fp->fp_reg_set.fpchip_state.status = fpu->fpu_regs.kfpu_status;
 358      -        }
 359      -        kpreempt_enable();
      251 +        bzero(fp, sizeof (*fp));
      252 +        fpu_get_fpregset(lwp, fp);
 360  253  }
 361  254  
 362  255  #if defined(_SYSCALL32_IMPL)
 363  256  
 364  257  /*
 365  258   * Set floating-point registers from an fpregset32_t.
 366  259   */
 367  260  void
 368  261  setfpregs32(klwp_t *lwp, fpregset32_t *fp)
 369  262  {
 370  263          fpregset_t fpregs;
 371  264  
 372  265          fpregset_32ton(fp, &fpregs);
 373  266          setfpregs(lwp, &fpregs);
 374  267  }
 375  268  
 376  269  /*
 377  270   * Get floating-point registers into an fpregset32_t.
 378  271   */
 379  272  void
 380  273  getfpregs32(klwp_t *lwp, fpregset32_t *fp)
 381  274  {
 382  275          fpregset_t fpregs;
 383  276  
 384  277          getfpregs(lwp, &fpregs);
 385  278          fpregset_nto32(&fpregs, fp);
 386  279  }
 387  280  
 388  281  #endif  /* _SYSCALL32_IMPL */
 389  282  
 390  283  /*
 391  284   * Return the general registers
 392  285   */
 393  286  void
 394  287  getgregs(klwp_t *lwp, gregset_t grp)
 395  288  {
 396  289          struct regs *rp = lwptoregs(lwp);
 397  290          struct pcb *pcb = &lwp->lwp_pcb;
 398  291          int thisthread = lwptot(lwp) == curthread;
 399  292  
 400  293          grp[REG_RDI] = rp->r_rdi;
 401  294          grp[REG_RSI] = rp->r_rsi;
 402  295          grp[REG_RDX] = rp->r_rdx;
 403  296          grp[REG_RCX] = rp->r_rcx;
 404  297          grp[REG_R8] = rp->r_r8;
 405  298          grp[REG_R9] = rp->r_r9;
 406  299          grp[REG_RAX] = rp->r_rax;
 407  300          grp[REG_RBX] = rp->r_rbx;
 408  301          grp[REG_RBP] = rp->r_rbp;
 409  302          grp[REG_R10] = rp->r_r10;
 410  303          grp[REG_R11] = rp->r_r11;
 411  304          grp[REG_R12] = rp->r_r12;
 412  305          grp[REG_R13] = rp->r_r13;
 413  306          grp[REG_R14] = rp->r_r14;
 414  307          grp[REG_R15] = rp->r_r15;
 415  308          grp[REG_FSBASE] = pcb->pcb_fsbase;
 416  309          grp[REG_GSBASE] = pcb->pcb_gsbase;
 417  310          if (thisthread)
 418  311                  kpreempt_disable();
 419  312          if (PCB_NEED_UPDATE_SEGS(pcb)) {
 420  313                  grp[REG_DS] = pcb->pcb_ds;
 421  314                  grp[REG_ES] = pcb->pcb_es;
 422  315                  grp[REG_FS] = pcb->pcb_fs;
 423  316                  grp[REG_GS] = pcb->pcb_gs;
 424  317          } else {
 425  318                  grp[REG_DS] = rp->r_ds;
 426  319                  grp[REG_ES] = rp->r_es;
 427  320                  grp[REG_FS] = rp->r_fs;
 428  321                  grp[REG_GS] = rp->r_gs;
 429  322          }
 430  323          if (thisthread)
 431  324                  kpreempt_enable();
 432  325          grp[REG_TRAPNO] = rp->r_trapno;
 433  326          grp[REG_ERR] = rp->r_err;
 434  327          grp[REG_RIP] = rp->r_rip;
 435  328          grp[REG_CS] = rp->r_cs;
 436  329          grp[REG_SS] = rp->r_ss;
 437  330          grp[REG_RFL] = rp->r_rfl;
 438  331          grp[REG_RSP] = rp->r_rsp;
 439  332  }
 440  333  
 441  334  #if defined(_SYSCALL32_IMPL)
 442  335  
 443  336  void
 444  337  getgregs32(klwp_t *lwp, gregset32_t grp)
 445  338  {
 446  339          struct regs *rp = lwptoregs(lwp);
 447  340          struct pcb *pcb = &lwp->lwp_pcb;
 448  341          int thisthread = lwptot(lwp) == curthread;
 449  342  
 450  343          if (thisthread)
 451  344                  kpreempt_disable();
 452  345          if (PCB_NEED_UPDATE_SEGS(pcb)) {
 453  346                  grp[GS] = (uint16_t)pcb->pcb_gs;
 454  347                  grp[FS] = (uint16_t)pcb->pcb_fs;
 455  348                  grp[DS] = (uint16_t)pcb->pcb_ds;
 456  349                  grp[ES] = (uint16_t)pcb->pcb_es;
 457  350          } else {
 458  351                  grp[GS] = (uint16_t)rp->r_gs;
 459  352                  grp[FS] = (uint16_t)rp->r_fs;
 460  353                  grp[DS] = (uint16_t)rp->r_ds;
 461  354                  grp[ES] = (uint16_t)rp->r_es;
 462  355          }
 463  356          if (thisthread)
 464  357                  kpreempt_enable();
 465  358          grp[EDI] = (greg32_t)rp->r_rdi;
 466  359          grp[ESI] = (greg32_t)rp->r_rsi;
 467  360          grp[EBP] = (greg32_t)rp->r_rbp;
 468  361          grp[ESP] = 0;
 469  362          grp[EBX] = (greg32_t)rp->r_rbx;
 470  363          grp[EDX] = (greg32_t)rp->r_rdx;
 471  364          grp[ECX] = (greg32_t)rp->r_rcx;
 472  365          grp[EAX] = (greg32_t)rp->r_rax;
 473  366          grp[TRAPNO] = (greg32_t)rp->r_trapno;
 474  367          grp[ERR] = (greg32_t)rp->r_err;
 475  368          grp[EIP] = (greg32_t)rp->r_rip;
 476  369          grp[CS] = (uint16_t)rp->r_cs;
 477  370          grp[EFL] = (greg32_t)rp->r_rfl;
 478  371          grp[UESP] = (greg32_t)rp->r_rsp;
 479  372          grp[SS] = (uint16_t)rp->r_ss;
 480  373  }
 481  374  
 482  375  void
 483  376  ucontext_32ton(const ucontext32_t *src, ucontext_t *dst)
 484  377  {
 485  378          mcontext_t *dmc = &dst->uc_mcontext;
 486  379          const mcontext32_t *smc = &src->uc_mcontext;
 487  380  
 488  381          bzero(dst, sizeof (*dst));
 489  382          dst->uc_flags = src->uc_flags;
 490  383          dst->uc_link = (ucontext_t *)(uintptr_t)src->uc_link;
 491  384  
 492  385          bcopy(&src->uc_sigmask, &dst->uc_sigmask, sizeof (dst->uc_sigmask));
 493  386  
 494  387          dst->uc_stack.ss_sp = (void *)(uintptr_t)src->uc_stack.ss_sp;
 495  388          dst->uc_stack.ss_size = (size_t)src->uc_stack.ss_size;
 496  389          dst->uc_stack.ss_flags = src->uc_stack.ss_flags;
 497  390  
 498  391          dmc->gregs[REG_GS] = (greg_t)(uint32_t)smc->gregs[GS];
 499  392          dmc->gregs[REG_FS] = (greg_t)(uint32_t)smc->gregs[FS];
 500  393          dmc->gregs[REG_ES] = (greg_t)(uint32_t)smc->gregs[ES];
 501  394          dmc->gregs[REG_DS] = (greg_t)(uint32_t)smc->gregs[DS];
 502  395          dmc->gregs[REG_RDI] = (greg_t)(uint32_t)smc->gregs[EDI];
 503  396          dmc->gregs[REG_RSI] = (greg_t)(uint32_t)smc->gregs[ESI];
 504  397          dmc->gregs[REG_RBP] = (greg_t)(uint32_t)smc->gregs[EBP];
 505  398          dmc->gregs[REG_RBX] = (greg_t)(uint32_t)smc->gregs[EBX];
 506  399          dmc->gregs[REG_RDX] = (greg_t)(uint32_t)smc->gregs[EDX];
 507  400          dmc->gregs[REG_RCX] = (greg_t)(uint32_t)smc->gregs[ECX];
 508  401          dmc->gregs[REG_RAX] = (greg_t)(uint32_t)smc->gregs[EAX];
 509  402          dmc->gregs[REG_TRAPNO] = (greg_t)(uint32_t)smc->gregs[TRAPNO];
 510  403          dmc->gregs[REG_ERR] = (greg_t)(uint32_t)smc->gregs[ERR];
 511  404          dmc->gregs[REG_RIP] = (greg_t)(uint32_t)smc->gregs[EIP];
 512  405          dmc->gregs[REG_CS] = (greg_t)(uint32_t)smc->gregs[CS];
 513  406          dmc->gregs[REG_RFL] = (greg_t)(uint32_t)smc->gregs[EFL];
 514  407          dmc->gregs[REG_RSP] = (greg_t)(uint32_t)smc->gregs[UESP];
 515  408          dmc->gregs[REG_SS] = (greg_t)(uint32_t)smc->gregs[SS];
 516  409  
 517  410          /*
 518  411           * A valid fpregs is only copied in if uc.uc_flags has UC_FPU set
 519  412           * otherwise there is no guarantee that anything in fpregs is valid.
 520  413           */
  
    | 
      ↓ open down ↓ | 
    151 lines elided | 
    
      ↑ open up ↑ | 
  
 521  414          if (src->uc_flags & UC_FPU)
 522  415                  fpregset_32ton(&src->uc_mcontext.fpregs,
 523  416                      &dst->uc_mcontext.fpregs);
 524  417  
 525  418          /*
 526  419           * Copy the brand-private data:
 527  420           */
 528  421          dst->uc_brand_data[0] = (void *)(uintptr_t)src->uc_brand_data[0];
 529  422          dst->uc_brand_data[1] = (void *)(uintptr_t)src->uc_brand_data[1];
 530  423          dst->uc_brand_data[2] = (void *)(uintptr_t)src->uc_brand_data[2];
      424 +
      425 +        if (src->uc_flags & UC_XSAVE) {
      426 +                dst->uc_xsave = (long)(uint32_t)src->uc_xsave;
      427 +        } else {
      428 +                dst->uc_xsave = 0;
      429 +        }
 531  430  }
 532  431  
 533  432  #endif  /* _SYSCALL32_IMPL */
 534  433  
 535  434  /*
 536  435   * Return the user-level PC.
 537  436   * If in a system call, return the address of the syscall trap.
 538  437   */
 539  438  greg_t
 540      -getuserpc()
      439 +getuserpc(void)
 541  440  {
 542  441          greg_t upc = lwptoregs(ttolwp(curthread))->r_pc;
 543  442          uint32_t insn;
 544  443  
 545  444          if (curthread->t_sysnum == 0)
 546  445                  return (upc);
 547  446  
 548  447          /*
 549  448           * We might've gotten here from sysenter (0xf 0x34),
 550  449           * syscall (0xf 0x5) or lcall (0x9a 0 0 0 0 0x27 0).
 551  450           *
 552  451           * Go peek at the binary to figure it out..
 553  452           */
 554  453          if (fuword32((void *)(upc - 2), &insn) != -1 &&
 555  454              (insn & 0xffff) == 0x340f || (insn & 0xffff) == 0x050f)
 556  455                  return (upc - 2);
 557  456          return (upc - 7);
 558  457  }
 559  458  
 560  459  /*
 561  460   * Protect segment registers from non-user privilege levels and GDT selectors
 562  461   * other than USER_CS, USER_DS and lwp FS and GS values.  If the segment
 563  462   * selector is non-null and not USER_CS/USER_DS, we make sure that the
 564  463   * TI bit is set to point into the LDT and that the RPL is set to 3.
 565  464   *
 566  465   * Since struct regs stores each 16-bit segment register as a 32-bit greg_t, we
 567  466   * also explicitly zero the top 16 bits since they may be coming from the
 568  467   * user's address space via setcontext(2) or /proc.
 569  468   *
 570  469   * Note about null selector. When running on the hypervisor if we allow a
 571  470   * process to set its %cs to null selector with RPL of 0 the hypervisor will
 572  471   * crash the domain. If running on bare metal we would get a #gp fault and
 573  472   * be able to kill the process and continue on. Therefore we make sure to
 574  473   * force RPL to SEL_UPL even for null selector when setting %cs.
 575  474   */
 576  475  
 577  476  #if defined(IS_CS) || defined(IS_NOT_CS)
 578  477  #error  "IS_CS and IS_NOT_CS already defined"
 579  478  #endif
 580  479  
 581  480  #define IS_CS           1
 582  481  #define IS_NOT_CS       0
 583  482  
 584  483  /*ARGSUSED*/
 585  484  greg_t
 586  485  fix_segreg(greg_t sr, int iscs, model_t datamodel)
 587  486  {
 588  487          kthread_t *t = curthread;
 589  488  
 590  489          switch (sr &= 0xffff) {
 591  490  
 592  491          case 0:
 593  492                  if (iscs == IS_CS)
 594  493                          return (0 | SEL_UPL);
 595  494                  else
 596  495                          return (0);
 597  496  
 598  497          /*
 599  498           * If lwp attempts to switch data model then force their
 600  499           * code selector to be null selector.
 601  500           */
 602  501          case U32CS_SEL:
 603  502                  if (datamodel == DATAMODEL_NATIVE)
 604  503                          return (0 | SEL_UPL);
 605  504                  else
 606  505                          return (sr);
 607  506  
 608  507          case UCS_SEL:
 609  508                  if (datamodel == DATAMODEL_ILP32)
 610  509                          return (0 | SEL_UPL);
 611  510          /*FALLTHROUGH*/
 612  511          case UDS_SEL:
 613  512          case LWPFS_SEL:
 614  513          case LWPGS_SEL:
 615  514          case SEL_UPL:
 616  515                  return (sr);
 617  516          default:
 618  517                  break;
 619  518          }
 620  519  
 621  520          /*
 622  521           * Allow this process's brand to do any necessary segment register
 623  522           * manipulation.
 624  523           */
 625  524          if (PROC_IS_BRANDED(t->t_procp) && BRMOP(t->t_procp)->b_fixsegreg) {
 626  525                  greg_t bsr = BRMOP(t->t_procp)->b_fixsegreg(sr, datamodel);
 627  526  
 628  527                  if (bsr == 0 && iscs == IS_CS)
 629  528                          return (0 | SEL_UPL);
 630  529                  else
 631  530                          return (bsr);
 632  531          }
 633  532  
 634  533          /*
 635  534           * Force it into the LDT in ring 3 for 32-bit processes, which by
 636  535           * default do not have an LDT, so that any attempt to use an invalid
 637  536           * selector will reference the (non-existant) LDT, and cause a #gp
 638  537           * fault for the process.
 639  538           *
 640  539           * 64-bit processes get the null gdt selector since they
 641  540           * are not allowed to have a private LDT.
 642  541           */
 643  542          if (datamodel == DATAMODEL_ILP32) {
 644  543                  return (sr | SEL_TI_LDT | SEL_UPL);
 645  544          } else {
 646  545                  if (iscs == IS_CS)
 647  546                          return (0 | SEL_UPL);
 648  547                  else
 649  548                          return (0);
 650  549          }
 651  550  
 652  551  }
 653  552  
 654  553  /*
 655  554   * Set general registers.
 656  555   */
 657  556  void
 658  557  setgregs(klwp_t *lwp, gregset_t grp)
 659  558  {
 660  559          struct regs *rp = lwptoregs(lwp);
 661  560          model_t datamodel = lwp_getdatamodel(lwp);
 662  561  
 663  562          struct pcb *pcb = &lwp->lwp_pcb;
 664  563          int thisthread = lwptot(lwp) == curthread;
 665  564  
 666  565          if (datamodel == DATAMODEL_NATIVE) {
 667  566                  if (thisthread)
 668  567                          (void) save_syscall_args();     /* copy the args */
 669  568  
 670  569                  rp->r_rdi = grp[REG_RDI];
 671  570                  rp->r_rsi = grp[REG_RSI];
 672  571                  rp->r_rdx = grp[REG_RDX];
 673  572                  rp->r_rcx = grp[REG_RCX];
 674  573                  rp->r_r8 = grp[REG_R8];
 675  574                  rp->r_r9 = grp[REG_R9];
 676  575                  rp->r_rax = grp[REG_RAX];
 677  576                  rp->r_rbx = grp[REG_RBX];
 678  577                  rp->r_rbp = grp[REG_RBP];
 679  578                  rp->r_r10 = grp[REG_R10];
 680  579                  rp->r_r11 = grp[REG_R11];
 681  580                  rp->r_r12 = grp[REG_R12];
 682  581                  rp->r_r13 = grp[REG_R13];
 683  582                  rp->r_r14 = grp[REG_R14];
 684  583                  rp->r_r15 = grp[REG_R15];
 685  584                  rp->r_trapno = grp[REG_TRAPNO];
 686  585                  rp->r_err = grp[REG_ERR];
 687  586                  rp->r_rip = grp[REG_RIP];
 688  587                  /*
 689  588                   * Setting %cs or %ss to anything else is quietly but
 690  589                   * quite definitely forbidden!
 691  590                   */
 692  591                  rp->r_cs = UCS_SEL;
 693  592                  rp->r_ss = UDS_SEL;
 694  593                  rp->r_rsp = grp[REG_RSP];
 695  594  
 696  595                  if (thisthread)
 697  596                          kpreempt_disable();
 698  597  
 699  598                  pcb->pcb_ds = UDS_SEL;
 700  599                  pcb->pcb_es = UDS_SEL;
 701  600  
 702  601                  /*
 703  602                   * 64-bit processes -are- allowed to set their fsbase/gsbase
 704  603                   * values directly, but only if they're using the segment
 705  604                   * selectors that allow that semantic.
 706  605                   *
 707  606                   * (32-bit processes must use lwp_set_private().)
 708  607                   */
 709  608                  pcb->pcb_fsbase = grp[REG_FSBASE];
 710  609                  pcb->pcb_gsbase = grp[REG_GSBASE];
 711  610                  pcb->pcb_fs = fix_segreg(grp[REG_FS], IS_NOT_CS, datamodel);
 712  611                  pcb->pcb_gs = fix_segreg(grp[REG_GS], IS_NOT_CS, datamodel);
 713  612  
 714  613                  /*
 715  614                   * Ensure that we go out via update_sregs
 716  615                   */
 717  616                  PCB_SET_UPDATE_SEGS(pcb);
 718  617                  lwptot(lwp)->t_post_sys = 1;
 719  618                  if (thisthread)
 720  619                          kpreempt_enable();
 721  620  #if defined(_SYSCALL32_IMPL)
 722  621          } else {
 723  622                  rp->r_rdi = (uint32_t)grp[REG_RDI];
 724  623                  rp->r_rsi = (uint32_t)grp[REG_RSI];
 725  624                  rp->r_rdx = (uint32_t)grp[REG_RDX];
 726  625                  rp->r_rcx = (uint32_t)grp[REG_RCX];
 727  626                  rp->r_rax = (uint32_t)grp[REG_RAX];
 728  627                  rp->r_rbx = (uint32_t)grp[REG_RBX];
 729  628                  rp->r_rbp = (uint32_t)grp[REG_RBP];
 730  629                  rp->r_trapno = (uint32_t)grp[REG_TRAPNO];
 731  630                  rp->r_err = (uint32_t)grp[REG_ERR];
 732  631                  rp->r_rip = (uint32_t)grp[REG_RIP];
 733  632  
 734  633                  rp->r_cs = fix_segreg(grp[REG_CS], IS_CS, datamodel);
 735  634                  rp->r_ss = fix_segreg(grp[REG_DS], IS_NOT_CS, datamodel);
 736  635  
 737  636                  rp->r_rsp = (uint32_t)grp[REG_RSP];
 738  637  
 739  638                  if (thisthread)
 740  639                          kpreempt_disable();
 741  640  
 742  641                  pcb->pcb_ds = fix_segreg(grp[REG_DS], IS_NOT_CS, datamodel);
 743  642                  pcb->pcb_es = fix_segreg(grp[REG_ES], IS_NOT_CS, datamodel);
 744  643  
 745  644                  /*
 746  645                   * (See fsbase/gsbase commentary above)
 747  646                   */
 748  647                  pcb->pcb_fs = fix_segreg(grp[REG_FS], IS_NOT_CS, datamodel);
 749  648                  pcb->pcb_gs = fix_segreg(grp[REG_GS], IS_NOT_CS, datamodel);
 750  649  
 751  650                  /*
 752  651                   * Ensure that we go out via update_sregs
 753  652                   */
 754  653                  PCB_SET_UPDATE_SEGS(pcb);
 755  654                  lwptot(lwp)->t_post_sys = 1;
 756  655                  if (thisthread)
 757  656                          kpreempt_enable();
 758  657  #endif
 759  658          }
 760  659  
 761  660          /*
 762  661           * Only certain bits of the flags register can be modified.
 763  662           */
 764  663          rp->r_rfl = (rp->r_rfl & ~PSL_USERMASK) |
 765  664              (grp[REG_RFL] & PSL_USERMASK);
 766  665  }
 767  666  
 768  667  /*
 769  668   * Determine whether eip is likely to have an interrupt frame
 770  669   * on the stack.  We do this by comparing the address to the
 771  670   * range of addresses spanned by several well-known routines.
 772  671   */
 773  672  extern void _interrupt();
 774  673  extern void _allsyscalls();
 775  674  extern void _cmntrap();
 776  675  extern void fakesoftint();
 777  676  
 778  677  extern size_t _interrupt_size;
 779  678  extern size_t _allsyscalls_size;
 780  679  extern size_t _cmntrap_size;
 781  680  extern size_t _fakesoftint_size;
 782  681  
 783  682  /*
 784  683   * Get a pc-only stacktrace.  Used for kmem_alloc() buffer ownership tracking.
 785  684   * Returns MIN(current stack depth, pcstack_limit).
 786  685   */
 787  686  int
 788  687  getpcstack(pc_t *pcstack, int pcstack_limit)
 789  688  {
 790  689          struct frame *fp = (struct frame *)getfp();
 791  690          struct frame *nextfp, *minfp, *stacktop;
 792  691          int depth = 0;
 793  692          int on_intr;
 794  693          uintptr_t pc;
 795  694  
 796  695          if ((on_intr = CPU_ON_INTR(CPU)) != 0)
 797  696                  stacktop = (struct frame *)(CPU->cpu_intr_stack + SA(MINFRAME));
 798  697          else
 799  698                  stacktop = (struct frame *)curthread->t_stk;
 800  699          minfp = fp;
 801  700  
 802  701          pc = ((struct regs *)fp)->r_pc;
 803  702  
 804  703          while (depth < pcstack_limit) {
 805  704                  nextfp = (struct frame *)fp->fr_savfp;
 806  705                  pc = fp->fr_savpc;
 807  706                  if (nextfp <= minfp || nextfp >= stacktop) {
 808  707                          if (on_intr) {
 809  708                                  /*
 810  709                                   * Hop from interrupt stack to thread stack.
 811  710                                   */
 812  711                                  stacktop = (struct frame *)curthread->t_stk;
 813  712                                  minfp = (struct frame *)curthread->t_stkbase;
 814  713                                  on_intr = 0;
 815  714                                  continue;
 816  715                          }
 817  716                          break;
 818  717                  }
 819  718                  pcstack[depth++] = (pc_t)pc;
 820  719                  fp = nextfp;
 821  720                  minfp = fp;
 822  721          }
 823  722          return (depth);
 824  723  }
 825  724  
 826  725  /*
 827  726   * The following ELF header fields are defined as processor-specific
 828  727   * in the V8 ABI:
 829  728   *
 830  729   *      e_ident[EI_DATA]        encoding of the processor-specific
 831  730   *                              data in the object file
 832  731   *      e_machine               processor identification
 833  732   *      e_flags                 processor-specific flags associated
 834  733   *                              with the file
 835  734   */
 836  735  
 837  736  /*
 838  737   * The value of at_flags reflects a platform's cpu module support.
 839  738   * at_flags is used to check for allowing a binary to execute and
 840  739   * is passed as the value of the AT_FLAGS auxiliary vector.
 841  740   */
 842  741  int at_flags = 0;
 843  742  
 844  743  /*
 845  744   * Check the processor-specific fields of an ELF header.
 846  745   *
 847  746   * returns 1 if the fields are valid, 0 otherwise
 848  747   */
 849  748  /*ARGSUSED2*/
 850  749  int
 851  750  elfheadcheck(
 852  751          unsigned char e_data,
 853  752          Elf32_Half e_machine,
 854  753          Elf32_Word e_flags)
 855  754  {
 856  755          if (e_data != ELFDATA2LSB)
 857  756                  return (0);
 858  757          if (e_machine == EM_AMD64)
 859  758                  return (1);
 860  759          return (e_machine == EM_386);
 861  760  }
 862  761  
 863  762  uint_t auxv_hwcap_include = 0;  /* patch to enable unrecognized features */
 864  763  uint_t auxv_hwcap_include_2 = 0;        /* second word */
 865  764  uint_t auxv_hwcap_exclude = 0;  /* patch for broken cpus, debugging */
 866  765  uint_t auxv_hwcap_exclude_2 = 0;        /* second word */
 867  766  #if defined(_SYSCALL32_IMPL)
 868  767  uint_t auxv_hwcap32_include = 0;        /* ditto for 32-bit apps */
 869  768  uint_t auxv_hwcap32_include_2 = 0;      /* ditto for 32-bit apps */
 870  769  uint_t auxv_hwcap32_exclude = 0;        /* ditto for 32-bit apps */
 871  770  uint_t auxv_hwcap32_exclude_2 = 0;      /* ditto for 32-bit apps */
 872  771  #endif
 873  772  
 874  773  /*
 875  774   * Gather information about the processor and place it into auxv_hwcap
 876  775   * so that it can be exported to the linker via the aux vector.
 877  776   *
 878  777   * We use this seemingly complicated mechanism so that we can ensure
 879  778   * that /etc/system can be used to override what the system can or
 880  779   * cannot discover for itself. Due to a lack of use, this has not
 881  780   * been extended to the 3rd word.
 882  781   */
 883  782  void
 884  783  bind_hwcap(void)
 885  784  {
 886  785          uint_t cpu_hwcap_flags[3];
 887  786          cpuid_execpass(NULL, CPUID_PASS_RESOLVE, cpu_hwcap_flags);
 888  787  
 889  788          auxv_hwcap = (auxv_hwcap_include | cpu_hwcap_flags[0]) &
 890  789              ~auxv_hwcap_exclude;
 891  790          auxv_hwcap_2 = (auxv_hwcap_include_2 | cpu_hwcap_flags[1]) &
 892  791              ~auxv_hwcap_exclude_2;
 893  792          auxv_hwcap_3 = cpu_hwcap_flags[2];
 894  793  
 895  794          /*
 896  795           * On AMD processors, sysenter just doesn't work at all
 897  796           * when the kernel is in long mode.  On IA-32e processors
 898  797           * it does, but there's no real point in all the alternate
 899  798           * mechanism when syscall works on both.
 900  799           *
 901  800           * Besides, the kernel's sysenter handler is expecting a
 902  801           * 32-bit lwp ...
 903  802           */
 904  803          auxv_hwcap &= ~AV_386_SEP;
 905  804  
 906  805          if (auxv_hwcap_include || auxv_hwcap_exclude || auxv_hwcap_include_2 ||
 907  806              auxv_hwcap_exclude_2) {
 908  807                  /*
 909  808                   * The below assignment is regrettably required to get lint
 910  809                   * to accept the validity of our format string.  The format
 911  810                   * string is in fact valid, but whatever intelligence in lint
 912  811                   * understands the cmn_err()-specific %b appears to have an
 913  812                   * off-by-one error:  it (mistakenly) complains about bit
 914  813                   * number 32 (even though this is explicitly permitted).
 915  814                   * Normally, one would will away such warnings with a "LINTED"
 916  815                   * directive, but for reasons unclear and unknown, lint
 917  816                   * refuses to be assuaged in this case.  Fortunately, lint
 918  817                   * doesn't pretend to have solved the Halting Problem --
 919  818                   * and as soon as the format string is programmatic, it
 920  819                   * knows enough to shut up.
 921  820                   */
 922  821                  char *fmt = "?user ABI extensions: %b\n";
 923  822                  cmn_err(CE_CONT, fmt, auxv_hwcap, FMT_AV_386);
 924  823                  fmt = "?user ABI extensions (word 2): %b\n";
 925  824                  cmn_err(CE_CONT, fmt, auxv_hwcap_2, FMT_AV_386_2);
 926  825                  fmt = "?user ABI extensions (word 2): %b\n";
 927  826                  cmn_err(CE_CONT, fmt, auxv_hwcap_3, FMT_AV_386_3);
 928  827          }
 929  828  
 930  829  #if defined(_SYSCALL32_IMPL)
 931  830          auxv_hwcap32 = (auxv_hwcap32_include | cpu_hwcap_flags[0]) &
 932  831              ~auxv_hwcap32_exclude;
 933  832          auxv_hwcap32_2 = (auxv_hwcap32_include_2 | cpu_hwcap_flags[1]) &
 934  833              ~auxv_hwcap32_exclude_2;
 935  834          auxv_hwcap32_3 = auxv_hwcap_3;
 936  835  
 937  836          /*
 938  837           * If this is an amd64 architecture machine from Intel, then
 939  838           * syscall -doesn't- work in compatibility mode, only sysenter does.
 940  839           *
 941  840           * Sigh.
 942  841           */
 943  842          if (!cpuid_syscall32_insn(NULL))
 944  843                  auxv_hwcap32 &= ~AV_386_AMD_SYSC;
 945  844  
 946  845          /*
 947  846           * 32-bit processes can -always- use the lahf/sahf instructions
 948  847           */
 949  848          auxv_hwcap32 |= AV_386_AHF;
 950  849  
 951  850          /*
 952  851           * 32-bit processes can -never- use fsgsbase instructions.
 953  852           */
 954  853          auxv_hwcap32_2 &= ~AV_386_2_FSGSBASE;
 955  854  
 956  855          if (auxv_hwcap32_include || auxv_hwcap32_exclude ||
 957  856              auxv_hwcap32_include_2 || auxv_hwcap32_exclude_2) {
 958  857                  /*
 959  858                   * See the block comment in the cmn_err() of auxv_hwcap, above.
 960  859                   */
 961  860                  char *fmt = "?32-bit user ABI extensions: %b\n";
 962  861                  cmn_err(CE_CONT, fmt, auxv_hwcap32, FMT_AV_386);
 963  862                  fmt = "?32-bit user ABI extensions (word 2): %b\n";
 964  863                  cmn_err(CE_CONT, fmt, auxv_hwcap32_2, FMT_AV_386_2);
 965  864                  fmt = "?32-bit user ABI extensions (word 3): %b\n";
 966  865                  cmn_err(CE_CONT, fmt, auxv_hwcap32_3, FMT_AV_386_3);
 967  866          }
 968  867  #endif
 969  868  }
 970  869  
 971  870  /*
 972  871   *      sync_icache() - this is called
 973  872   *      in proc/fs/prusrio.c. x86 has an unified cache and therefore
 974  873   *      this is a nop.
 975  874   */
 976  875  /* ARGSUSED */
 977  876  void
 978  877  sync_icache(caddr_t addr, uint_t len)
 979  878  {
 980  879          /* Do nothing for now */
 981  880  }
 982  881  
 983  882  /*ARGSUSED*/
 984  883  void
 985  884  sync_data_memory(caddr_t va, size_t len)
 986  885  {
 987  886          /* Not implemented for this platform */
 988  887  }
 989  888  
 990  889  int
 991  890  __ipltospl(int ipl)
 992  891  {
 993  892          return (ipltospl(ipl));
 994  893  }
 995  894  
 996  895  /*
 997  896   * The panic code invokes panic_saveregs() to record the contents of a
 998  897   * regs structure into the specified panic_data structure for debuggers.
 999  898   */
1000  899  void
1001  900  panic_saveregs(panic_data_t *pdp, struct regs *rp)
1002  901  {
1003  902          panic_nv_t *pnv = PANICNVGET(pdp);
1004  903  
1005  904          struct cregs    creg;
1006  905  
1007  906          getcregs(&creg);
1008  907  
1009  908          PANICNVADD(pnv, "rdi", rp->r_rdi);
1010  909          PANICNVADD(pnv, "rsi", rp->r_rsi);
1011  910          PANICNVADD(pnv, "rdx", rp->r_rdx);
1012  911          PANICNVADD(pnv, "rcx", rp->r_rcx);
1013  912          PANICNVADD(pnv, "r8", rp->r_r8);
1014  913          PANICNVADD(pnv, "r9", rp->r_r9);
1015  914          PANICNVADD(pnv, "rax", rp->r_rax);
1016  915          PANICNVADD(pnv, "rbx", rp->r_rbx);
1017  916          PANICNVADD(pnv, "rbp", rp->r_rbp);
1018  917          PANICNVADD(pnv, "r10", rp->r_r10);
1019  918          PANICNVADD(pnv, "r11", rp->r_r11);
1020  919          PANICNVADD(pnv, "r12", rp->r_r12);
1021  920          PANICNVADD(pnv, "r13", rp->r_r13);
1022  921          PANICNVADD(pnv, "r14", rp->r_r14);
1023  922          PANICNVADD(pnv, "r15", rp->r_r15);
1024  923          PANICNVADD(pnv, "fsbase", rdmsr(MSR_AMD_FSBASE));
1025  924          PANICNVADD(pnv, "gsbase", rdmsr(MSR_AMD_GSBASE));
1026  925          PANICNVADD(pnv, "ds", rp->r_ds);
1027  926          PANICNVADD(pnv, "es", rp->r_es);
1028  927          PANICNVADD(pnv, "fs", rp->r_fs);
1029  928          PANICNVADD(pnv, "gs", rp->r_gs);
1030  929          PANICNVADD(pnv, "trapno", rp->r_trapno);
1031  930          PANICNVADD(pnv, "err", rp->r_err);
1032  931          PANICNVADD(pnv, "rip", rp->r_rip);
1033  932          PANICNVADD(pnv, "cs", rp->r_cs);
1034  933          PANICNVADD(pnv, "rflags", rp->r_rfl);
1035  934          PANICNVADD(pnv, "rsp", rp->r_rsp);
1036  935          PANICNVADD(pnv, "ss", rp->r_ss);
1037  936          PANICNVADD(pnv, "gdt_hi", (uint64_t)(creg.cr_gdt._l[3]));
1038  937          PANICNVADD(pnv, "gdt_lo", (uint64_t)(creg.cr_gdt._l[0]));
1039  938          PANICNVADD(pnv, "idt_hi", (uint64_t)(creg.cr_idt._l[3]));
1040  939          PANICNVADD(pnv, "idt_lo", (uint64_t)(creg.cr_idt._l[0]));
1041  940  
1042  941          PANICNVADD(pnv, "ldt", creg.cr_ldt);
1043  942          PANICNVADD(pnv, "task", creg.cr_task);
1044  943          PANICNVADD(pnv, "cr0", creg.cr_cr0);
1045  944          PANICNVADD(pnv, "cr2", creg.cr_cr2);
1046  945          PANICNVADD(pnv, "cr3", creg.cr_cr3);
1047  946          if (creg.cr_cr4)
1048  947                  PANICNVADD(pnv, "cr4", creg.cr_cr4);
1049  948  
1050  949          PANICNVSET(pdp, pnv);
1051  950  }
1052  951  
1053  952  #define TR_ARG_MAX 6    /* Max args to print, same as SPARC */
1054  953  
1055  954  
1056  955  /*
1057  956   * Print a stack backtrace using the specified frame pointer.  We delay two
1058  957   * seconds before continuing, unless this is the panic traceback.
1059  958   * If we are in the process of panicking, we also attempt to write the
1060  959   * stack backtrace to a staticly assigned buffer, to allow the panic
1061  960   * code to find it and write it in to uncompressed pages within the
1062  961   * system crash dump.
1063  962   * Note that the frame for the starting stack pointer value is omitted because
1064  963   * the corresponding %eip is not known.
1065  964   */
1066  965  
1067  966  extern char *dump_stack_scratch;
1068  967  
1069  968  
1070  969  void
1071  970  traceback(caddr_t fpreg)
1072  971  {
1073  972          struct frame    *fp = (struct frame *)fpreg;
1074  973          struct frame    *nextfp;
1075  974          uintptr_t       pc, nextpc;
1076  975          ulong_t         off;
1077  976          char            args[TR_ARG_MAX * 2 + 16], *sym;
1078  977          uint_t    offset = 0;
1079  978          uint_t    next_offset = 0;
1080  979          char        stack_buffer[1024];
1081  980  
1082  981          if (!panicstr)
1083  982                  printf("traceback: %%fp = %p\n", (void *)fp);
1084  983  
1085  984          if (panicstr && !dump_stack_scratch) {
1086  985                  printf("Warning - stack not written to the dump buffer\n");
1087  986          }
1088  987  
1089  988          fp = (struct frame *)plat_traceback(fpreg);
1090  989          if ((uintptr_t)fp < KERNELBASE)
1091  990                  goto out;
1092  991  
1093  992          pc = fp->fr_savpc;
1094  993          fp = (struct frame *)fp->fr_savfp;
1095  994  
1096  995          while ((uintptr_t)fp >= KERNELBASE) {
1097  996                  /*
1098  997                   * XX64 Until port is complete tolerate 8-byte aligned
1099  998                   * frame pointers but flag with a warning so they can
1100  999                   * be fixed.
1101 1000                   */
1102 1001                  if (((uintptr_t)fp & (STACK_ALIGN - 1)) != 0) {
1103 1002                          if (((uintptr_t)fp & (8 - 1)) == 0) {
1104 1003                                  printf("  >> warning! 8-byte"
1105 1004                                      " aligned %%fp = %p\n", (void *)fp);
1106 1005                          } else {
1107 1006                                  printf(
1108 1007                                      "  >> mis-aligned %%fp = %p\n", (void *)fp);
1109 1008                                  break;
1110 1009                          }
1111 1010                  }
1112 1011  
1113 1012                  args[0] = '\0';
1114 1013                  nextpc = (uintptr_t)fp->fr_savpc;
1115 1014                  nextfp = (struct frame *)fp->fr_savfp;
1116 1015                  if ((sym = kobj_getsymname(pc, &off)) != NULL) {
1117 1016                          printf("%016lx %s:%s+%lx (%s)\n", (uintptr_t)fp,
1118 1017                              mod_containing_pc((caddr_t)pc), sym, off, args);
1119 1018                          (void) snprintf(stack_buffer, sizeof (stack_buffer),
1120 1019                              "%s:%s+%lx (%s) | ",
1121 1020                              mod_containing_pc((caddr_t)pc), sym, off, args);
1122 1021                  } else {
1123 1022                          printf("%016lx %lx (%s)\n",
1124 1023                              (uintptr_t)fp, pc, args);
1125 1024                          (void) snprintf(stack_buffer, sizeof (stack_buffer),
1126 1025                              "%lx (%s) | ", pc, args);
1127 1026                  }
1128 1027  
1129 1028                  if (panicstr && dump_stack_scratch) {
1130 1029                          next_offset = offset + strlen(stack_buffer);
1131 1030                          if (next_offset < STACK_BUF_SIZE) {
1132 1031                                  bcopy(stack_buffer, dump_stack_scratch + offset,
1133 1032                                      strlen(stack_buffer));
1134 1033                                  offset = next_offset;
1135 1034                          } else {
1136 1035                                  /*
1137 1036                                   * In attempting to save the panic stack
1138 1037                                   * to the dumpbuf we have overflowed that area.
1139 1038                                   * Print a warning and continue to printf the
1140 1039                                   * stack to the msgbuf
1141 1040                                   */
1142 1041                                  printf("Warning: stack in the dump buffer"
1143 1042                                      " may be incomplete\n");
1144 1043                                  offset = next_offset;
1145 1044                          }
1146 1045                  }
1147 1046  
1148 1047                  pc = nextpc;
1149 1048                  fp = nextfp;
1150 1049          }
1151 1050  out:
1152 1051          if (!panicstr) {
1153 1052                  printf("end of traceback\n");
1154 1053                  DELAY(2 * MICROSEC);
1155 1054          } else if (dump_stack_scratch) {
1156 1055                  dump_stack_scratch[offset] = '\0';
1157 1056          }
1158 1057  }
1159 1058  
1160 1059  
1161 1060  /*
1162 1061   * Generate a stack backtrace from a saved register set.
1163 1062   */
1164 1063  void
1165 1064  traceregs(struct regs *rp)
1166 1065  {
1167 1066          traceback((caddr_t)rp->r_fp);
1168 1067  }
1169 1068  
1170 1069  void
1171 1070  exec_set_sp(size_t stksize)
1172 1071  {
1173 1072          klwp_t *lwp = ttolwp(curthread);
1174 1073  
1175 1074          lwptoregs(lwp)->r_sp = (uintptr_t)curproc->p_usrstack - stksize;
1176 1075  }
1177 1076  
1178 1077  hrtime_t
1179 1078  gethrtime_waitfree(void)
1180 1079  {
1181 1080          return (dtrace_gethrtime());
1182 1081  }
1183 1082  
1184 1083  hrtime_t
1185 1084  gethrtime(void)
1186 1085  {
1187 1086          return (gethrtimef());
1188 1087  }
1189 1088  
1190 1089  hrtime_t
1191 1090  gethrtime_unscaled(void)
1192 1091  {
1193 1092          return (gethrtimeunscaledf());
1194 1093  }
1195 1094  
1196 1095  void
1197 1096  scalehrtime(hrtime_t *hrt)
1198 1097  {
1199 1098          scalehrtimef(hrt);
1200 1099  }
1201 1100  
1202 1101  uint64_t
1203 1102  unscalehrtime(hrtime_t nsecs)
1204 1103  {
1205 1104          return (unscalehrtimef(nsecs));
1206 1105  }
1207 1106  
1208 1107  void
1209 1108  gethrestime(timespec_t *tp)
1210 1109  {
1211 1110          gethrestimef(tp);
1212 1111  }
1213 1112  
1214 1113  /*
1215 1114   * Part of the implementation of hres_tick(); this routine is
1216 1115   * easier in C than assembler .. called with the hres_lock held.
1217 1116   *
1218 1117   * XX64 Many of these timekeeping variables need to be extern'ed in a header
1219 1118   */
1220 1119  
1221 1120  #include <sys/time.h>
1222 1121  #include <sys/machlock.h>
1223 1122  
1224 1123  extern int one_sec;
1225 1124  extern int max_hres_adj;
1226 1125  
1227 1126  void
1228 1127  __adj_hrestime(void)
1229 1128  {
1230 1129          long long adj;
1231 1130  
1232 1131          if (hrestime_adj == 0)
1233 1132                  adj = 0;
1234 1133          else if (hrestime_adj > 0) {
1235 1134                  if (hrestime_adj < max_hres_adj)
1236 1135                          adj = hrestime_adj;
1237 1136                  else
1238 1137                          adj = max_hres_adj;
1239 1138          } else {
1240 1139                  if (hrestime_adj < -max_hres_adj)
1241 1140                          adj = -max_hres_adj;
1242 1141                  else
1243 1142                          adj = hrestime_adj;
1244 1143          }
1245 1144  
1246 1145          timedelta -= adj;
1247 1146          hrestime_adj = timedelta;
1248 1147          hrestime.tv_nsec += adj;
1249 1148  
1250 1149          while (hrestime.tv_nsec >= NANOSEC) {
1251 1150                  one_sec++;
1252 1151                  hrestime.tv_sec++;
1253 1152                  hrestime.tv_nsec -= NANOSEC;
1254 1153          }
1255 1154  }
1256 1155  
1257 1156  /*
1258 1157   * Wrapper functions to maintain backwards compability
1259 1158   */
1260 1159  int
1261 1160  xcopyin(const void *uaddr, void *kaddr, size_t count)
1262 1161  {
1263 1162          return (xcopyin_nta(uaddr, kaddr, count, UIO_COPY_CACHED));
1264 1163  }
1265 1164  
1266 1165  int
1267 1166  xcopyout(const void *kaddr, void *uaddr, size_t count)
1268 1167  {
1269 1168          return (xcopyout_nta(kaddr, uaddr, count, UIO_COPY_CACHED));
1270 1169  }
  
    | 
      ↓ open down ↓ | 
    720 lines elided | 
    
      ↑ open up ↑ | 
  
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX