arch/um/os-Linux/skas/process.c

   1 /*
   2  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <stdlib.h>
   7 #include <unistd.h>
   8 #include <sched.h>
   9 #include <errno.h>
  10 #include <string.h>
  11 #include <sys/mman.h>
  12 #include <sys/ptrace.h>
  13 #include <sys/wait.h>
  14 #include <asm/unistd.h>
  15 #include "as-layout.h"
  16 #include "chan_user.h"
  17 #include "kern_constants.h"
  18 #include "kern_util.h"
  19 #include "mem.h"
  20 #include "os.h"
  21 #include "process.h"
  22 #include "proc_mm.h"
  23 #include "ptrace_user.h"
  24 #include "registers.h"
  25 #include "skas.h"
  26 #include "skas_ptrace.h"
  27 #include "user.h"
  28 #include "sysdep/stub.h"
  29
  30 int is_skas_winch(int pid, int fd, void *data)
  31 {
  32         if (pid != getpgrp())
  33                 return 0;
  34
  35         register_winch_irq(-1, fd, -1, data, 0);
  36         return 1;
  37 }
  38
  39 static int ptrace_dump_regs(int pid)
  40 {
  41         unsigned long regs[MAX_REG_NR];
  42         int i;
  43
  44         if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
  45                 return -errno;
  46
  47         printk(UM_KERN_ERR "Stub registers -\n");
  48         for (i = 0; i < ARRAY_SIZE(regs); i++)
  49                 printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
  50
  51         return 0;
  52 }
  53
  54 /*
  55  * Signals that are OK to receive in the stub - we'll just continue it.
  56  * SIGWINCH will happen when UML is inside a detached screen.
  57  */
  58 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
  59
  60 /* Signals that the stub will finish with - anything else is an error */
  61 #define STUB_DONE_MASK (1 << SIGTRAP)
  62
  63 void wait_stub_done(int pid)
  64 {
  65         int n, status, err;
  66
  67         while (1) {
  68                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
  69                 if ((n < 0) || !WIFSTOPPED(status))
  70                         goto bad_wait;
  71
  72                 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
  73                         break;
  74
  75                 err = ptrace(PTRACE_CONT, pid, 0, 0);
  76                 if (err) {
  77                         printk(UM_KERN_ERR "wait_stub_done : continue failed, "
  78                                "errno = %d\n", errno);
  79                         fatal_sigsegv();
  80                 }
  81         }
  82
  83         if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
  84                 return;
  85
  86 bad_wait:
  87         err = ptrace_dump_regs(pid);
  88         if (err)
  89                 printk(UM_KERN_ERR "Failed to get registers from stub, "
  90                        "errno = %d\n", -err);
  91         printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
  92                "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
  93                status);
  94         fatal_sigsegv();
  95 }
  96
  97 extern unsigned long current_stub_stack(void);
  98
  99 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
 100 {
 101         int err;
 102
 103         if (ptrace_faultinfo) {
 104                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
 105                 if (err) {
 106                         printk(UM_KERN_ERR "get_skas_faultinfo - "
 107                                "PTRACE_FAULTINFO failed, errno = %d\n", errno);
 108                         fatal_sigsegv();
 109                 }
 110
 111                 /* Special handling for i386, which has different structs */
 112                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
 113                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
 114                                sizeof(struct faultinfo) -
 115                                sizeof(struct ptrace_faultinfo));
 116         }
 117         else {
 118                 unsigned long fpregs[FP_SIZE];
 119
 120                 err = get_fp_registers(pid, fpregs);
 121                 if (err < 0) {
 122                         printk(UM_KERN_ERR "save_fp_registers returned %d\n",
 123                                err);
 124                         fatal_sigsegv();
 125                 }
 126                 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
 127                 if (err) {
 128                         printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
 129                                "errno = %d\n", pid, errno);
 130                         fatal_sigsegv();
 131                 }
 132                 wait_stub_done(pid);
 133
 134                 /*
 135                  * faultinfo is prepared by the stub-segv-handler at start of
 136                  * the stub stack page. We just have to copy it.
 137                  */
 138                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 139
 140                 err = put_fp_registers(pid, fpregs);
 141                 if (err < 0) {
 142                         printk(UM_KERN_ERR "put_fp_registers returned %d\n",
 143                                err);
 144                         fatal_sigsegv();
 145                 }
 146         }
 147 }
 148
 149 static void handle_segv(int pid, struct uml_pt_regs * regs)
 150 {
 151         get_skas_faultinfo(pid, &regs->faultinfo);
 152         segv(regs->faultinfo, 0, 1, NULL);
 153 }
 154
 155 /*
 156  * To use the same value of using_sysemu as the caller, ask it that value
 157  * (in local_using_sysemu
 158  */
 159 static void handle_trap(int pid, struct uml_pt_regs *regs,
 160                         int local_using_sysemu)
 161 {
 162         int err, status;
 163
 164         if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
 165                 fatal_sigsegv();
 166
 167         /* Mark this as a syscall */
 168         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
 169
 170         if (!local_using_sysemu)
 171         {
 172                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
 173                              __NR_getpid);
 174                 if (err < 0) {
 175                         printk(UM_KERN_ERR "handle_trap - nullifying syscall "
 176                                "failed, errno = %d\n", errno);
 177                         fatal_sigsegv();
 178                 }
 179
 180                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
 181                 if (err < 0) {
 182                         printk(UM_KERN_ERR "handle_trap - continuing to end of "
 183                                "syscall failed, errno = %d\n", errno);
 184                         fatal_sigsegv();
 185                 }
 186
 187                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
 188                 if ((err < 0) || !WIFSTOPPED(status) ||
 189                     (WSTOPSIG(status) != SIGTRAP + 0x80)) {
 190                         err = ptrace_dump_regs(pid);
 191                         if (err)
 192                                 printk(UM_KERN_ERR "Failed to get registers "
 193                                        "from process, errno = %d\n", -err);
 194                         printk(UM_KERN_ERR "handle_trap - failed to wait at "
 195                                "end of syscall, errno = %d, status = %d\n",
 196                                errno, status);
 197                         fatal_sigsegv();
 198                 }
 199         }
 200
 201         handle_syscall(regs);
 202 }
 203
 204 extern int __syscall_stub_start;
 205
 206 static int userspace_tramp(void *stack)
 207 {
 208         void *addr;
 209         int err;
 210
 211         ptrace(PTRACE_TRACEME, 0, 0, 0);
 212
 213         signal(SIGTERM, SIG_DFL);
 214         signal(SIGWINCH, SIG_IGN);
 215         err = set_interval();
 216         if (err) {
 217                 printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
 218                        "errno = %d\n", err);
 219                 exit(1);
 220         }
 221
 222         if (!proc_mm) {
 223                 /*
 224                  * This has a pte, but it can't be mapped in with the usual
 225                  * tlb_flush mechanism because this is part of that mechanism
 226                  */
 227                 int fd;
 228                 unsigned long long offset;
 229                 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
 230                 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
 231                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
 232                 if (addr == MAP_FAILED) {
 233                         printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
 234                                "errno = %d\n", STUB_CODE, errno);
 235                         exit(1);
 236                 }
 237
 238                 if (stack != NULL) {
 239                         fd = phys_mapping(to_phys(stack), &offset);
 240                         addr = mmap((void *) STUB_DATA,
 241                                     UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
 242                                     MAP_FIXED | MAP_SHARED, fd, offset);
 243                         if (addr == MAP_FAILED) {
 244                                 printk(UM_KERN_ERR "mapping segfault stack "
 245                                        "at 0x%lx failed, errno = %d\n",
 246                                        STUB_DATA, errno);
 247                                 exit(1);
 248                         }
 249                 }
 250         }
 251         if (!ptrace_faultinfo && (stack != NULL)) {
 252                 struct sigaction sa;
 253
 254                 unsigned long v = STUB_CODE +
 255                                   (unsigned long) stub_segv_handler -
 256                                   (unsigned long) &__syscall_stub_start;
 257
 258                 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
 259                 sigemptyset(&sa.sa_mask);
 260                 sa.sa_flags = SA_ONSTACK | SA_NODEFER;
 261                 sa.sa_handler = (void *) v;
 262                 sa.sa_restorer = NULL;
 263                 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
 264                         printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
 265                                "handler failed - errno = %d\n", errno);
 266                         exit(1);
 267                 }
 268         }
 269
 270         kill(os_getpid(), SIGSTOP);
 271         return 0;
 272 }
 273
 274 /* Each element set once, and only accessed by a single processor anyway */
 275 #undef NR_CPUS
 276 #define NR_CPUS 1
 277 int userspace_pid[NR_CPUS];
 278
 279 int start_userspace(unsigned long stub_stack)
 280 {
 281         void *stack;
 282         unsigned long sp;
 283         int pid, status, n, flags, err;
 284
 285         stack = mmap(NULL, UM_KERN_PAGE_SIZE,
 286                      PROT_READ | PROT_WRITE | PROT_EXEC,
 287                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 288         if (stack == MAP_FAILED) {
 289                 err = -errno;
 290                 printk(UM_KERN_ERR "start_userspace : mmap failed, "
 291                        "errno = %d\n", errno);
 292                 return err;
 293         }
 294
 295         sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
 296
 297         flags = CLONE_FILES;
 298         if (proc_mm)
 299                 flags |= CLONE_VM;
 300         else
 301                 flags |= SIGCHLD;
 302
 303         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
 304         if (pid < 0) {
 305                 err = -errno;
 306                 printk(UM_KERN_ERR "start_userspace : clone failed, "
 307                        "errno = %d\n", errno);
 308                 return err;
 309         }
 310
 311         do {
 312                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
 313                 if (n < 0) {
 314                         err = -errno;
 315                         printk(UM_KERN_ERR "start_userspace : wait failed, "
 316                                "errno = %d\n", errno);
 317                         goto out_kill;
 318                 }
 319         } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
 320
 321         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
 322                 err = -EINVAL;
 323                 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
 324                        "status = %d\n", status);
 325                 goto out_kill;
 326         }
 327
 328         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 329                    (void *) PTRACE_O_TRACESYSGOOD) < 0) {
 330                 err = -errno;
 331                 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
 332                        "failed, errno = %d\n", errno);
 333                 goto out_kill;
 334         }
 335
 336         if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
 337                 err = -errno;
 338                 printk(UM_KERN_ERR "start_userspace : munmap failed, "
 339                        "errno = %d\n", errno);
 340                 goto out_kill;
 341         }
 342
 343         return pid;
 344
 345  out_kill:
 346         os_kill_ptraced_process(pid, 1);
 347         return err;
 348 }
 349
 350 void userspace(struct uml_pt_regs *regs)
 351 {
 352         struct itimerval timer;
 353         unsigned long long nsecs, now;
 354         int err, status, op, pid = userspace_pid[0];
 355         /* To prevent races if using_sysemu changes under us.*/
 356         int local_using_sysemu;
 357
 358         if (getitimer(ITIMER_VIRTUAL, &timer))
 359                 printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
 360         nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
 361                 timer.it_value.tv_usec * UM_NSEC_PER_USEC;
 362         nsecs += os_nsecs();
 363
 364         while (1) {
 365                 /*
 366                  * This can legitimately fail if the process loads a
 367                  * bogus value into a segment register.  It will
 368                  * segfault and PTRACE_GETREGS will read that value
 369                  * out of the process.  However, PTRACE_SETREGS will
 370                  * fail.  In this case, there is nothing to do but
 371                  * just kill the process.
 372                  */
 373                 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
 374                         fatal_sigsegv();
 375
 376                 if (put_fp_registers(pid, regs->fp))
 377                         fatal_sigsegv();
 378
 379                 /* Now we set local_using_sysemu to be used for one loop */
 380                 local_using_sysemu = get_using_sysemu();
 381
 382                 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
 383                                              singlestepping(NULL));
 384
 385                 if (ptrace(op, pid, 0, 0)) {
 386                         printk(UM_KERN_ERR "userspace - ptrace continue "
 387                                "failed, op = %d, errno = %d\n", op, errno);
 388                         fatal_sigsegv();
 389                 }
 390
 391                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
 392                 if (err < 0) {
 393                         printk(UM_KERN_ERR "userspace - wait failed, "
 394                                "errno = %d\n", errno);
 395                         fatal_sigsegv();
 396                 }
 397
 398                 regs->is_user = 1;
 399                 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
 400                         printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
 401                                "errno = %d\n", errno);
 402                         fatal_sigsegv();
 403                 }
 404
 405                 if (get_fp_registers(pid, regs->fp)) {
 406                         printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
 407                                "errno = %d\n", errno);
 408                         fatal_sigsegv();
 409                 }
 410
 411                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
 412
 413                 if (WIFSTOPPED(status)) {
 414                         int sig = WSTOPSIG(status);
 415                         switch (sig) {
 416                         case SIGSEGV:
 417                                 if (PTRACE_FULL_FAULTINFO ||
 418                                     !ptrace_faultinfo) {
 419                                         get_skas_faultinfo(pid,
 420                                                            &regs->faultinfo);
 421                                         (*sig_info[SIGSEGV])(SIGSEGV, regs);
 422                                 }
 423                                 else handle_segv(pid, regs);
 424                                 break;
 425                         case SIGTRAP + 0x80:
 426                                 handle_trap(pid, regs, local_using_sysemu);
 427                                 break;
 428                         case SIGTRAP:
 429                                 relay_signal(SIGTRAP, regs);
 430                                 break;
 431                         case SIGVTALRM:
 432                                 now = os_nsecs();
 433                                 if (now < nsecs)
 434                                         break;
 435                                 block_signals();
 436                                 (*sig_info[sig])(sig, regs);
 437                                 unblock_signals();
 438                                 nsecs = timer.it_value.tv_sec *
 439                                         UM_NSEC_PER_SEC +
 440                                         timer.it_value.tv_usec *
 441                                         UM_NSEC_PER_USEC;
 442                                 nsecs += os_nsecs();
 443                                 break;
 444                         case SIGIO:
 445                         case SIGILL:
 446                         case SIGBUS:
 447                         case SIGFPE:
 448                         case SIGWINCH:
 449                                 block_signals();
 450                                 (*sig_info[sig])(sig, regs);
 451                                 unblock_signals();
 452                                 break;
 453                         default:
 454                                 printk(UM_KERN_ERR "userspace - child stopped "
 455                                        "with signal %d\n", sig);
 456                                 fatal_sigsegv();
 457                         }
 458                         pid = userspace_pid[0];
 459                         interrupt_end();
 460
 461                         /* Avoid -ERESTARTSYS handling in host */
 462                         if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
 463                                 PT_SYSCALL_NR(regs->gp) = -1;
 464                 }
 465         }
 466 }
 467
 468 static unsigned long thread_regs[MAX_REG_NR];
 469 static unsigned long thread_fp_regs[FP_SIZE];
 470
 471 static int __init init_thread_regs(void)
 472 {
 473         get_safe_registers(thread_regs, thread_fp_regs);
 474         /* Set parent's instruction pointer to start of clone-stub */
 475         thread_regs[REGS_IP_INDEX] = STUB_CODE +
 476                                 (unsigned long) stub_clone_handler -
 477                                 (unsigned long) &__syscall_stub_start;
 478         thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
 479                 sizeof(void *);
 480 #ifdef __SIGNAL_FRAMESIZE
 481         thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
 482 #endif
 483         return 0;
 484 }
 485
 486 __initcall(init_thread_regs);
 487
 488 int copy_context_skas0(unsigned long new_stack, int pid)
 489 {
 490         struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
 491         int err;
 492         unsigned long current_stack = current_stub_stack();
 493         struct stub_data *data = (struct stub_data *) current_stack;
 494         struct stub_data *child_data = (struct stub_data *) new_stack;
 495         unsigned long long new_offset;
 496         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
 497
 498         /*
 499          * prepare offset and fd of child's stack as argument for parent's
 500          * and child's mmap2 calls
 501          */
 502         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
 503                                       .fd       = new_fd,
 504                                       .timer    = ((struct itimerval)
 505                                                    { .it_value = tv,
 506                                                      .it_interval = tv }) });
 507
 508         err = ptrace_setregs(pid, thread_regs);
 509         if (err < 0) {
 510                 err = -errno;
 511                 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
 512                        "failed, pid = %d, errno = %d\n", pid, -err);
 513                 return err;
 514         }
 515
 516         err = put_fp_registers(pid, thread_fp_regs);
 517         if (err < 0) {
 518                 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
 519                        "failed, pid = %d, err = %d\n", pid, err);
 520                 return err;
 521         }
 522
 523         /* set a well known return code for detection of child write failure */
 524         child_data->err = 12345678;
 525
 526         /*
 527          * Wait, until parent has finished its work: read child's pid from
 528          * parent's stack, and check, if bad result.
 529          */
 530         err = ptrace(PTRACE_CONT, pid, 0, 0);
 531         if (err) {
 532                 err = -errno;
 533                 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
 534                        "errno = %d\n", pid, errno);
 535                 return err;
 536         }
 537
 538         wait_stub_done(pid);
 539
 540         pid = data->err;
 541         if (pid < 0) {
 542                 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
 543                        "error %d\n", -pid);
 544                 return pid;
 545         }
 546
 547         /*
 548          * Wait, until child has finished too: read child's result from
 549          * child's stack and check it.
 550          */
 551         wait_stub_done(pid);
 552         if (child_data->err != STUB_DATA) {
 553                 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
 554                        "error %ld\n", child_data->err);
 555                 err = child_data->err;
 556                 goto out_kill;
 557         }
 558
 559         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
 560                    (void *)PTRACE_O_TRACESYSGOOD) < 0) {
 561                 err = -errno;
 562                 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
 563                        "failed, errno = %d\n", errno);
 564                 goto out_kill;
 565         }
 566
 567         return pid;
 568
 569  out_kill:
 570         os_kill_ptraced_process(pid, 1);
 571         return err;
 572 }
 573
 574 /*
 575  * This is used only, if stub pages are needed, while proc_mm is
 576  * available. Opening /proc/mm creates a new mm_context, which lacks
 577  * the stub-pages. Thus, we map them using /proc/mm-fd
 578  */
 579 int map_stub_pages(int fd, unsigned long code, unsigned long data,
 580                    unsigned long stack)
 581 {
 582         struct proc_mm_op mmop;
 583         int n;
 584         unsigned long long code_offset;
 585         int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
 586                                    &code_offset);
 587
 588         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
 589                                       .u         =
 590                                       { .mmap    =
 591                                         { .addr    = code,
 592                                           .len     = UM_KERN_PAGE_SIZE,
 593                                           .prot    = PROT_EXEC,
 594                                           .flags   = MAP_FIXED | MAP_PRIVATE,
 595                                           .fd      = code_fd,
 596                                           .offset  = code_offset
 597         } } });
 598         CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
 599         if (n != sizeof(mmop)) {
 600                 n = errno;
 601                 printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
 602                        "offset = %llx\n", code, code_fd,
 603                        (unsigned long long) code_offset);
 604                 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
 605                        "failed, err = %d\n", n);
 606                 return -n;
 607         }
 608
 609         if (stack) {
 610                 unsigned long long map_offset;
 611                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
 612                 mmop = ((struct proc_mm_op)
 613                                 { .op        = MM_MMAP,
 614                                   .u         =
 615                                   { .mmap    =
 616                                     { .addr    = data,
 617                                       .len     = UM_KERN_PAGE_SIZE,
 618                                       .prot    = PROT_READ | PROT_WRITE,
 619                                       .flags   = MAP_FIXED | MAP_SHARED,
 620                                       .fd      = map_fd,
 621                                       .offset  = map_offset
 622                 } } });
 623                 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
 624                 if (n != sizeof(mmop)) {
 625                         n = errno;
 626                         printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
 627                                "data failed, err = %d\n", n);
 628                         return -n;
 629                 }
 630         }
 631
 632         return 0;
 633 }
 634
 635 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
 636 {
 637         (*buf)[0].JB_IP = (unsigned long) handler;
 638         (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
 639                 sizeof(void *);
 640 }
 641
 642 #define INIT_JMP_NEW_THREAD 0
 643 #define INIT_JMP_CALLBACK 1
 644 #define INIT_JMP_HALT 2
 645 #define INIT_JMP_REBOOT 3
 646
 647 void switch_threads(jmp_buf *me, jmp_buf *you)
 648 {
 649         if (UML_SETJMP(me) == 0)
 650                 UML_LONGJMP(you, 1);
 651 }
 652
 653 static jmp_buf initial_jmpbuf;
 654
 655 /* XXX Make these percpu */
 656 static void (*cb_proc)(void *arg);
 657 static void *cb_arg;
 658 static jmp_buf *cb_back;
 659
 660 int start_idle_thread(void *stack, jmp_buf *switch_buf)
 661 {
 662         int n;
 663
 664         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
 665                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGVTALRM, -1);
 666
 667         /*
 668          * Can't use UML_SETJMP or UML_LONGJMP here because they save
 669          * and restore signals, with the possible side-effect of
 670          * trying to handle any signals which came when they were
 671          * blocked, which can't be done on this stack.
 672          * Signals must be blocked when jumping back here and restored
 673          * after returning to the jumper.
 674          */
 675         n = setjmp(initial_jmpbuf);
 676         switch (n) {
 677         case INIT_JMP_NEW_THREAD:
 678                 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
 679                 (*switch_buf)[0].JB_SP = (unsigned long) stack +
 680                         UM_THREAD_SIZE - sizeof(void *);
 681                 break;
 682         case INIT_JMP_CALLBACK:
 683                 (*cb_proc)(cb_arg);
 684                 longjmp(*cb_back, 1);
 685                 break;
 686         case INIT_JMP_HALT:
 687                 kmalloc_ok = 0;
 688                 return 0;
 689         case INIT_JMP_REBOOT:
 690                 kmalloc_ok = 0;
 691                 return 1;
 692         default:
 693                 printk(UM_KERN_ERR "Bad sigsetjmp return in "
 694                        "start_idle_thread - %d\n", n);
 695                 fatal_sigsegv();
 696         }
 697         longjmp(*switch_buf, 1);
 698 }
 699
 700 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
 701 {
 702         jmp_buf here;
 703
 704         cb_proc = proc;
 705         cb_arg = arg;
 706         cb_back = &here;
 707
 708         block_signals();
 709         if (UML_SETJMP(&here) == 0)
 710                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
 711         unblock_signals();
 712
 713         cb_proc = NULL;
 714         cb_arg = NULL;
 715         cb_back = NULL;
 716 }
 717
 718 void halt_skas(void)
 719 {
 720         block_signals();
 721         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
 722 }
 723
 724 void reboot_skas(void)
 725 {
 726         block_signals();
 727         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
 728 }
 729
 730 void __switch_mm(struct mm_id *mm_idp)
 731 {
 732         int err;
 733
 734         /* FIXME: need cpu pid in __switch_mm */
 735         if (proc_mm) {
 736                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
 737                              mm_idp->u.mm_fd);
 738                 if (err) {
 739                         printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
 740                                "failed, errno = %d\n", errno);
 741                         fatal_sigsegv();
 742                 }
 743         }
 744         else userspace_pid[0] = mm_idp->u.pid;
 745 }