Linux 2.6.39-rc2
[pohmelfs.git] / arch / um / os-Linux / skas / process.c
blobd6e0a2234b869267e7bfcdf15170716086a55154
1 /*
2 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
4 */
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"
30 int is_skas_winch(int pid, int fd, void *data)
32 if (pid != getpgrp())
33 return 0;
35 register_winch_irq(-1, fd, -1, data, 0);
36 return 1;
39 static int ptrace_dump_regs(int pid)
41 unsigned long regs[MAX_REG_NR];
42 int i;
44 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
45 return -errno;
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]);
51 return 0;
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.
58 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
60 /* Signals that the stub will finish with - anything else is an error */
61 #define STUB_DONE_MASK (1 << SIGTRAP)
63 void wait_stub_done(int pid)
65 int n, status, err;
67 while (1) {
68 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
69 if ((n < 0) || !WIFSTOPPED(status))
70 goto bad_wait;
72 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
73 break;
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();
83 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
84 return;
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();
97 extern unsigned long current_stub_stack(void);
99 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
101 int err;
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();
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));
117 else {
118 unsigned long fpregs[FP_SIZE];
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();
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();
132 wait_stub_done(pid);
135 * faultinfo is prepared by the stub-segv-handler at start of
136 * the stub stack page. We just have to copy it.
138 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
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();
149 static void handle_segv(int pid, struct uml_pt_regs * regs)
151 get_skas_faultinfo(pid, &regs->faultinfo);
152 segv(regs->faultinfo, 0, 1, NULL);
156 * To use the same value of using_sysemu as the caller, ask it that value
157 * (in local_using_sysemu
159 static void handle_trap(int pid, struct uml_pt_regs *regs,
160 int local_using_sysemu)
162 int err, status;
164 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
165 fatal_sigsegv();
167 /* Mark this as a syscall */
168 UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
170 if (!local_using_sysemu)
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();
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();
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();
201 handle_syscall(regs);
204 extern int __syscall_stub_start;
206 static int userspace_tramp(void *stack)
208 void *addr;
209 int err;
211 ptrace(PTRACE_TRACEME, 0, 0, 0);
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);
222 if (!proc_mm) {
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
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);
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);
251 if (!ptrace_faultinfo && (stack != NULL)) {
252 struct sigaction sa;
254 unsigned long v = STUB_CODE +
255 (unsigned long) stub_segv_handler -
256 (unsigned long) &__syscall_stub_start;
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);
270 kill(os_getpid(), SIGSTOP);
271 return 0;
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];
279 int start_userspace(unsigned long stub_stack)
281 void *stack;
282 unsigned long sp;
283 int pid, status, n, flags, err;
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;
295 sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
297 flags = CLONE_FILES;
298 if (proc_mm)
299 flags |= CLONE_VM;
300 else
301 flags |= SIGCHLD;
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;
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;
319 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
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;
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;
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;
343 return pid;
345 out_kill:
346 os_kill_ptraced_process(pid, 1);
347 return err;
350 void userspace(struct uml_pt_regs *regs)
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;
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();
364 while (1) {
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.
373 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
374 fatal_sigsegv();
376 /* Now we set local_using_sysemu to be used for one loop */
377 local_using_sysemu = get_using_sysemu();
379 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
380 singlestepping(NULL));
382 if (ptrace(op, pid, 0, 0)) {
383 printk(UM_KERN_ERR "userspace - ptrace continue "
384 "failed, op = %d, errno = %d\n", op, errno);
385 fatal_sigsegv();
388 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
389 if (err < 0) {
390 printk(UM_KERN_ERR "userspace - wait failed, "
391 "errno = %d\n", errno);
392 fatal_sigsegv();
395 regs->is_user = 1;
396 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
397 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
398 "errno = %d\n", errno);
399 fatal_sigsegv();
402 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
404 if (WIFSTOPPED(status)) {
405 int sig = WSTOPSIG(status);
406 switch (sig) {
407 case SIGSEGV:
408 if (PTRACE_FULL_FAULTINFO ||
409 !ptrace_faultinfo) {
410 get_skas_faultinfo(pid,
411 &regs->faultinfo);
412 (*sig_info[SIGSEGV])(SIGSEGV, regs);
414 else handle_segv(pid, regs);
415 break;
416 case SIGTRAP + 0x80:
417 handle_trap(pid, regs, local_using_sysemu);
418 break;
419 case SIGTRAP:
420 relay_signal(SIGTRAP, regs);
421 break;
422 case SIGVTALRM:
423 now = os_nsecs();
424 if (now < nsecs)
425 break;
426 block_signals();
427 (*sig_info[sig])(sig, regs);
428 unblock_signals();
429 nsecs = timer.it_value.tv_sec *
430 UM_NSEC_PER_SEC +
431 timer.it_value.tv_usec *
432 UM_NSEC_PER_USEC;
433 nsecs += os_nsecs();
434 break;
435 case SIGIO:
436 case SIGILL:
437 case SIGBUS:
438 case SIGFPE:
439 case SIGWINCH:
440 block_signals();
441 (*sig_info[sig])(sig, regs);
442 unblock_signals();
443 break;
444 default:
445 printk(UM_KERN_ERR "userspace - child stopped "
446 "with signal %d\n", sig);
447 fatal_sigsegv();
449 pid = userspace_pid[0];
450 interrupt_end();
452 /* Avoid -ERESTARTSYS handling in host */
453 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
454 PT_SYSCALL_NR(regs->gp) = -1;
459 static unsigned long thread_regs[MAX_REG_NR];
461 static int __init init_thread_regs(void)
463 get_safe_registers(thread_regs);
464 /* Set parent's instruction pointer to start of clone-stub */
465 thread_regs[REGS_IP_INDEX] = STUB_CODE +
466 (unsigned long) stub_clone_handler -
467 (unsigned long) &__syscall_stub_start;
468 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
469 sizeof(void *);
470 #ifdef __SIGNAL_FRAMESIZE
471 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
472 #endif
473 return 0;
476 __initcall(init_thread_regs);
478 int copy_context_skas0(unsigned long new_stack, int pid)
480 struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
481 int err;
482 unsigned long current_stack = current_stub_stack();
483 struct stub_data *data = (struct stub_data *) current_stack;
484 struct stub_data *child_data = (struct stub_data *) new_stack;
485 unsigned long long new_offset;
486 int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
489 * prepare offset and fd of child's stack as argument for parent's
490 * and child's mmap2 calls
492 *data = ((struct stub_data) { .offset = MMAP_OFFSET(new_offset),
493 .fd = new_fd,
494 .timer = ((struct itimerval)
495 { .it_value = tv,
496 .it_interval = tv }) });
498 err = ptrace_setregs(pid, thread_regs);
499 if (err < 0) {
500 err = -errno;
501 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
502 "failed, pid = %d, errno = %d\n", pid, -err);
503 return err;
506 /* set a well known return code for detection of child write failure */
507 child_data->err = 12345678;
510 * Wait, until parent has finished its work: read child's pid from
511 * parent's stack, and check, if bad result.
513 err = ptrace(PTRACE_CONT, pid, 0, 0);
514 if (err) {
515 err = -errno;
516 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
517 "errno = %d\n", pid, errno);
518 return err;
521 wait_stub_done(pid);
523 pid = data->err;
524 if (pid < 0) {
525 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
526 "error %d\n", -pid);
527 return pid;
531 * Wait, until child has finished too: read child's result from
532 * child's stack and check it.
534 wait_stub_done(pid);
535 if (child_data->err != STUB_DATA) {
536 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
537 "error %ld\n", child_data->err);
538 err = child_data->err;
539 goto out_kill;
542 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
543 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
544 err = -errno;
545 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
546 "failed, errno = %d\n", errno);
547 goto out_kill;
550 return pid;
552 out_kill:
553 os_kill_ptraced_process(pid, 1);
554 return err;
558 * This is used only, if stub pages are needed, while proc_mm is
559 * available. Opening /proc/mm creates a new mm_context, which lacks
560 * the stub-pages. Thus, we map them using /proc/mm-fd
562 int map_stub_pages(int fd, unsigned long code, unsigned long data,
563 unsigned long stack)
565 struct proc_mm_op mmop;
566 int n;
567 unsigned long long code_offset;
568 int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
569 &code_offset);
571 mmop = ((struct proc_mm_op) { .op = MM_MMAP,
572 .u =
573 { .mmap =
574 { .addr = code,
575 .len = UM_KERN_PAGE_SIZE,
576 .prot = PROT_EXEC,
577 .flags = MAP_FIXED | MAP_PRIVATE,
578 .fd = code_fd,
579 .offset = code_offset
580 } } });
581 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
582 if (n != sizeof(mmop)) {
583 n = errno;
584 printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
585 "offset = %llx\n", code, code_fd,
586 (unsigned long long) code_offset);
587 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
588 "failed, err = %d\n", n);
589 return -n;
592 if (stack) {
593 unsigned long long map_offset;
594 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
595 mmop = ((struct proc_mm_op)
596 { .op = MM_MMAP,
597 .u =
598 { .mmap =
599 { .addr = data,
600 .len = UM_KERN_PAGE_SIZE,
601 .prot = PROT_READ | PROT_WRITE,
602 .flags = MAP_FIXED | MAP_SHARED,
603 .fd = map_fd,
604 .offset = map_offset
605 } } });
606 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
607 if (n != sizeof(mmop)) {
608 n = errno;
609 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
610 "data failed, err = %d\n", n);
611 return -n;
615 return 0;
618 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
620 (*buf)[0].JB_IP = (unsigned long) handler;
621 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
622 sizeof(void *);
625 #define INIT_JMP_NEW_THREAD 0
626 #define INIT_JMP_CALLBACK 1
627 #define INIT_JMP_HALT 2
628 #define INIT_JMP_REBOOT 3
630 void switch_threads(jmp_buf *me, jmp_buf *you)
632 if (UML_SETJMP(me) == 0)
633 UML_LONGJMP(you, 1);
636 static jmp_buf initial_jmpbuf;
638 /* XXX Make these percpu */
639 static void (*cb_proc)(void *arg);
640 static void *cb_arg;
641 static jmp_buf *cb_back;
643 int start_idle_thread(void *stack, jmp_buf *switch_buf)
645 int n;
647 set_handler(SIGWINCH, (__sighandler_t) sig_handler,
648 SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGVTALRM, -1);
651 * Can't use UML_SETJMP or UML_LONGJMP here because they save
652 * and restore signals, with the possible side-effect of
653 * trying to handle any signals which came when they were
654 * blocked, which can't be done on this stack.
655 * Signals must be blocked when jumping back here and restored
656 * after returning to the jumper.
658 n = setjmp(initial_jmpbuf);
659 switch (n) {
660 case INIT_JMP_NEW_THREAD:
661 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
662 (*switch_buf)[0].JB_SP = (unsigned long) stack +
663 UM_THREAD_SIZE - sizeof(void *);
664 break;
665 case INIT_JMP_CALLBACK:
666 (*cb_proc)(cb_arg);
667 longjmp(*cb_back, 1);
668 break;
669 case INIT_JMP_HALT:
670 kmalloc_ok = 0;
671 return 0;
672 case INIT_JMP_REBOOT:
673 kmalloc_ok = 0;
674 return 1;
675 default:
676 printk(UM_KERN_ERR "Bad sigsetjmp return in "
677 "start_idle_thread - %d\n", n);
678 fatal_sigsegv();
680 longjmp(*switch_buf, 1);
683 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
685 jmp_buf here;
687 cb_proc = proc;
688 cb_arg = arg;
689 cb_back = &here;
691 block_signals();
692 if (UML_SETJMP(&here) == 0)
693 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
694 unblock_signals();
696 cb_proc = NULL;
697 cb_arg = NULL;
698 cb_back = NULL;
701 void halt_skas(void)
703 block_signals();
704 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
707 void reboot_skas(void)
709 block_signals();
710 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
713 void __switch_mm(struct mm_id *mm_idp)
715 int err;
717 /* FIXME: need cpu pid in __switch_mm */
718 if (proc_mm) {
719 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
720 mm_idp->u.mm_fd);
721 if (err) {
722 printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
723 "failed, errno = %d\n", errno);
724 fatal_sigsegv();
727 else userspace_pid[0] = mm_idp->u.pid;