2 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
13 #include <asm/unistd.h>
14 #include <as-layout.h>
16 #include <kern_util.h>
19 #include <ptrace_user.h>
20 #include <registers.h>
22 #include <sysdep/stub.h>
24 int is_skas_winch(int pid
, int fd
, void *data
)
26 return pid
== getpgrp();
29 static int ptrace_dump_regs(int pid
)
31 unsigned long regs
[MAX_REG_NR
];
34 if (ptrace(PTRACE_GETREGS
, pid
, 0, regs
) < 0)
37 printk(UM_KERN_ERR
"Stub registers -\n");
38 for (i
= 0; i
< ARRAY_SIZE(regs
); i
++)
39 printk(UM_KERN_ERR
"\t%d - %lx\n", i
, regs
[i
]);
45 * Signals that are OK to receive in the stub - we'll just continue it.
46 * SIGWINCH will happen when UML is inside a detached screen.
48 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
50 /* Signals that the stub will finish with - anything else is an error */
51 #define STUB_DONE_MASK (1 << SIGTRAP)
53 void wait_stub_done(int pid
)
58 CATCH_EINTR(n
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
59 if ((n
< 0) || !WIFSTOPPED(status
))
62 if (((1 << WSTOPSIG(status
)) & STUB_SIG_MASK
) == 0)
65 err
= ptrace(PTRACE_CONT
, pid
, 0, 0);
67 printk(UM_KERN_ERR
"wait_stub_done : continue failed, "
68 "errno = %d\n", errno
);
73 if (((1 << WSTOPSIG(status
)) & STUB_DONE_MASK
) != 0)
77 err
= ptrace_dump_regs(pid
);
79 printk(UM_KERN_ERR
"Failed to get registers from stub, "
80 "errno = %d\n", -err
);
81 printk(UM_KERN_ERR
"wait_stub_done : failed to wait for SIGTRAP, "
82 "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid
, n
, errno
,
87 extern unsigned long current_stub_stack(void);
89 static void get_skas_faultinfo(int pid
, struct faultinfo
*fi
)
92 unsigned long fpregs
[FP_SIZE
];
94 err
= get_fp_registers(pid
, fpregs
);
96 printk(UM_KERN_ERR
"save_fp_registers returned %d\n",
100 err
= ptrace(PTRACE_CONT
, pid
, 0, SIGSEGV
);
102 printk(UM_KERN_ERR
"Failed to continue stub, pid = %d, "
103 "errno = %d\n", pid
, errno
);
109 * faultinfo is prepared by the stub-segv-handler at start of
110 * the stub stack page. We just have to copy it.
112 memcpy(fi
, (void *)current_stub_stack(), sizeof(*fi
));
114 err
= put_fp_registers(pid
, fpregs
);
116 printk(UM_KERN_ERR
"put_fp_registers returned %d\n",
122 static void handle_segv(int pid
, struct uml_pt_regs
* regs
)
124 get_skas_faultinfo(pid
, ®s
->faultinfo
);
125 segv(regs
->faultinfo
, 0, 1, NULL
);
129 * To use the same value of using_sysemu as the caller, ask it that value
130 * (in local_using_sysemu
132 static void handle_trap(int pid
, struct uml_pt_regs
*regs
,
133 int local_using_sysemu
)
137 if ((UPT_IP(regs
) >= STUB_START
) && (UPT_IP(regs
) < STUB_END
))
140 /* Mark this as a syscall */
141 UPT_SYSCALL_NR(regs
) = PT_SYSCALL_NR(regs
->gp
);
143 if (!local_using_sysemu
)
145 err
= ptrace(PTRACE_POKEUSER
, pid
, PT_SYSCALL_NR_OFFSET
,
148 printk(UM_KERN_ERR
"handle_trap - nullifying syscall "
149 "failed, errno = %d\n", errno
);
153 err
= ptrace(PTRACE_SYSCALL
, pid
, 0, 0);
155 printk(UM_KERN_ERR
"handle_trap - continuing to end of "
156 "syscall failed, errno = %d\n", errno
);
160 CATCH_EINTR(err
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
161 if ((err
< 0) || !WIFSTOPPED(status
) ||
162 (WSTOPSIG(status
) != SIGTRAP
+ 0x80)) {
163 err
= ptrace_dump_regs(pid
);
165 printk(UM_KERN_ERR
"Failed to get registers "
166 "from process, errno = %d\n", -err
);
167 printk(UM_KERN_ERR
"handle_trap - failed to wait at "
168 "end of syscall, errno = %d, status = %d\n",
174 handle_syscall(regs
);
177 extern char __syscall_stub_start
[];
179 static int userspace_tramp(void *stack
)
183 unsigned long long offset
;
185 ptrace(PTRACE_TRACEME
, 0, 0, 0);
187 signal(SIGTERM
, SIG_DFL
);
188 signal(SIGWINCH
, SIG_IGN
);
189 err
= set_interval();
191 printk(UM_KERN_ERR
"userspace_tramp - setting timer failed, "
192 "errno = %d\n", err
);
197 * This has a pte, but it can't be mapped in with the usual
198 * tlb_flush mechanism because this is part of that mechanism
200 fd
= phys_mapping(to_phys(__syscall_stub_start
), &offset
);
201 addr
= mmap64((void *) STUB_CODE
, UM_KERN_PAGE_SIZE
,
202 PROT_EXEC
, MAP_FIXED
| MAP_PRIVATE
, fd
, offset
);
203 if (addr
== MAP_FAILED
) {
204 printk(UM_KERN_ERR
"mapping mmap stub at 0x%lx failed, "
205 "errno = %d\n", STUB_CODE
, errno
);
210 fd
= phys_mapping(to_phys(stack
), &offset
);
211 addr
= mmap((void *) STUB_DATA
,
212 UM_KERN_PAGE_SIZE
, PROT_READ
| PROT_WRITE
,
213 MAP_FIXED
| MAP_SHARED
, fd
, offset
);
214 if (addr
== MAP_FAILED
) {
215 printk(UM_KERN_ERR
"mapping segfault stack "
216 "at 0x%lx failed, errno = %d\n",
224 unsigned long v
= STUB_CODE
+
225 (unsigned long) stub_segv_handler
-
226 (unsigned long) __syscall_stub_start
;
228 set_sigstack((void *) STUB_DATA
, UM_KERN_PAGE_SIZE
);
229 sigemptyset(&sa
.sa_mask
);
230 sa
.sa_flags
= SA_ONSTACK
| SA_NODEFER
| SA_SIGINFO
;
231 sa
.sa_sigaction
= (void *) v
;
232 sa
.sa_restorer
= NULL
;
233 if (sigaction(SIGSEGV
, &sa
, NULL
) < 0) {
234 printk(UM_KERN_ERR
"userspace_tramp - setting SIGSEGV "
235 "handler failed - errno = %d\n", errno
);
240 kill(os_getpid(), SIGSTOP
);
244 /* Each element set once, and only accessed by a single processor anyway */
247 int userspace_pid
[NR_CPUS
];
249 int start_userspace(unsigned long stub_stack
)
253 int pid
, status
, n
, flags
, err
;
255 stack
= mmap(NULL
, UM_KERN_PAGE_SIZE
,
256 PROT_READ
| PROT_WRITE
| PROT_EXEC
,
257 MAP_PRIVATE
| MAP_ANONYMOUS
, -1, 0);
258 if (stack
== MAP_FAILED
) {
260 printk(UM_KERN_ERR
"start_userspace : mmap failed, "
261 "errno = %d\n", errno
);
265 sp
= (unsigned long) stack
+ UM_KERN_PAGE_SIZE
- sizeof(void *);
267 flags
= CLONE_FILES
| SIGCHLD
;
269 pid
= clone(userspace_tramp
, (void *) sp
, flags
, (void *) stub_stack
);
272 printk(UM_KERN_ERR
"start_userspace : clone failed, "
273 "errno = %d\n", errno
);
278 CATCH_EINTR(n
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
281 printk(UM_KERN_ERR
"start_userspace : wait failed, "
282 "errno = %d\n", errno
);
285 } while (WIFSTOPPED(status
) && (WSTOPSIG(status
) == SIGVTALRM
));
287 if (!WIFSTOPPED(status
) || (WSTOPSIG(status
) != SIGSTOP
)) {
289 printk(UM_KERN_ERR
"start_userspace : expected SIGSTOP, got "
290 "status = %d\n", status
);
294 if (ptrace(PTRACE_OLDSETOPTIONS
, pid
, NULL
,
295 (void *) PTRACE_O_TRACESYSGOOD
) < 0) {
297 printk(UM_KERN_ERR
"start_userspace : PTRACE_OLDSETOPTIONS "
298 "failed, errno = %d\n", errno
);
302 if (munmap(stack
, UM_KERN_PAGE_SIZE
) < 0) {
304 printk(UM_KERN_ERR
"start_userspace : munmap failed, "
305 "errno = %d\n", errno
);
312 os_kill_ptraced_process(pid
, 1);
316 void userspace(struct uml_pt_regs
*regs
)
318 struct itimerval timer
;
319 unsigned long long nsecs
, now
;
320 int err
, status
, op
, pid
= userspace_pid
[0];
321 /* To prevent races if using_sysemu changes under us.*/
322 int local_using_sysemu
;
325 /* Handle any immediate reschedules or signals */
328 if (getitimer(ITIMER_VIRTUAL
, &timer
))
329 printk(UM_KERN_ERR
"Failed to get itimer, errno = %d\n", errno
);
330 nsecs
= timer
.it_value
.tv_sec
* UM_NSEC_PER_SEC
+
331 timer
.it_value
.tv_usec
* UM_NSEC_PER_USEC
;
336 * This can legitimately fail if the process loads a
337 * bogus value into a segment register. It will
338 * segfault and PTRACE_GETREGS will read that value
339 * out of the process. However, PTRACE_SETREGS will
340 * fail. In this case, there is nothing to do but
341 * just kill the process.
343 if (ptrace(PTRACE_SETREGS
, pid
, 0, regs
->gp
))
346 if (put_fp_registers(pid
, regs
->fp
))
349 /* Now we set local_using_sysemu to be used for one loop */
350 local_using_sysemu
= get_using_sysemu();
352 op
= SELECT_PTRACE_OPERATION(local_using_sysemu
,
353 singlestepping(NULL
));
355 if (ptrace(op
, pid
, 0, 0)) {
356 printk(UM_KERN_ERR
"userspace - ptrace continue "
357 "failed, op = %d, errno = %d\n", op
, errno
);
361 CATCH_EINTR(err
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
363 printk(UM_KERN_ERR
"userspace - wait failed, "
364 "errno = %d\n", errno
);
369 if (ptrace(PTRACE_GETREGS
, pid
, 0, regs
->gp
)) {
370 printk(UM_KERN_ERR
"userspace - PTRACE_GETREGS failed, "
371 "errno = %d\n", errno
);
375 if (get_fp_registers(pid
, regs
->fp
)) {
376 printk(UM_KERN_ERR
"userspace - get_fp_registers failed, "
377 "errno = %d\n", errno
);
381 UPT_SYSCALL_NR(regs
) = -1; /* Assume: It's not a syscall */
383 if (WIFSTOPPED(status
)) {
384 int sig
= WSTOPSIG(status
);
386 ptrace(PTRACE_GETSIGINFO
, pid
, 0, (struct siginfo
*)&si
);
390 if (PTRACE_FULL_FAULTINFO
) {
391 get_skas_faultinfo(pid
,
393 (*sig_info
[SIGSEGV
])(SIGSEGV
, (struct siginfo
*)&si
,
396 else handle_segv(pid
, regs
);
399 handle_trap(pid
, regs
, local_using_sysemu
);
402 relay_signal(SIGTRAP
, (struct siginfo
*)&si
, regs
);
409 (*sig_info
[sig
])(sig
, (struct siginfo
*)&si
, regs
);
411 nsecs
= timer
.it_value
.tv_sec
*
413 timer
.it_value
.tv_usec
*
423 (*sig_info
[sig
])(sig
, (struct siginfo
*)&si
, regs
);
427 printk(UM_KERN_ERR
"userspace - child stopped "
428 "with signal %d\n", sig
);
431 pid
= userspace_pid
[0];
434 /* Avoid -ERESTARTSYS handling in host */
435 if (PT_SYSCALL_NR_OFFSET
!= PT_SYSCALL_RET_OFFSET
)
436 PT_SYSCALL_NR(regs
->gp
) = -1;
441 static unsigned long thread_regs
[MAX_REG_NR
];
442 static unsigned long thread_fp_regs
[FP_SIZE
];
444 static int __init
init_thread_regs(void)
446 get_safe_registers(thread_regs
, thread_fp_regs
);
447 /* Set parent's instruction pointer to start of clone-stub */
448 thread_regs
[REGS_IP_INDEX
] = STUB_CODE
+
449 (unsigned long) stub_clone_handler
-
450 (unsigned long) __syscall_stub_start
;
451 thread_regs
[REGS_SP_INDEX
] = STUB_DATA
+ UM_KERN_PAGE_SIZE
-
453 #ifdef __SIGNAL_FRAMESIZE
454 thread_regs
[REGS_SP_INDEX
] -= __SIGNAL_FRAMESIZE
;
459 __initcall(init_thread_regs
);
461 int copy_context_skas0(unsigned long new_stack
, int pid
)
463 struct timeval tv
= { .tv_sec
= 0, .tv_usec
= UM_USEC_PER_SEC
/ UM_HZ
};
465 unsigned long current_stack
= current_stub_stack();
466 struct stub_data
*data
= (struct stub_data
*) current_stack
;
467 struct stub_data
*child_data
= (struct stub_data
*) new_stack
;
468 unsigned long long new_offset
;
469 int new_fd
= phys_mapping(to_phys((void *)new_stack
), &new_offset
);
472 * prepare offset and fd of child's stack as argument for parent's
473 * and child's mmap2 calls
475 *data
= ((struct stub_data
) { .offset
= MMAP_OFFSET(new_offset
),
477 .timer
= ((struct itimerval
)
479 .it_interval
= tv
}) });
481 err
= ptrace_setregs(pid
, thread_regs
);
484 printk(UM_KERN_ERR
"copy_context_skas0 : PTRACE_SETREGS "
485 "failed, pid = %d, errno = %d\n", pid
, -err
);
489 err
= put_fp_registers(pid
, thread_fp_regs
);
491 printk(UM_KERN_ERR
"copy_context_skas0 : put_fp_registers "
492 "failed, pid = %d, err = %d\n", pid
, err
);
496 /* set a well known return code for detection of child write failure */
497 child_data
->err
= 12345678;
500 * Wait, until parent has finished its work: read child's pid from
501 * parent's stack, and check, if bad result.
503 err
= ptrace(PTRACE_CONT
, pid
, 0, 0);
506 printk(UM_KERN_ERR
"Failed to continue new process, pid = %d, "
507 "errno = %d\n", pid
, errno
);
515 printk(UM_KERN_ERR
"copy_context_skas0 - stub-parent reports "
521 * Wait, until child has finished too: read child's result from
522 * child's stack and check it.
525 if (child_data
->err
!= STUB_DATA
) {
526 printk(UM_KERN_ERR
"copy_context_skas0 - stub-child reports "
527 "error %ld\n", child_data
->err
);
528 err
= child_data
->err
;
532 if (ptrace(PTRACE_OLDSETOPTIONS
, pid
, NULL
,
533 (void *)PTRACE_O_TRACESYSGOOD
) < 0) {
535 printk(UM_KERN_ERR
"copy_context_skas0 : PTRACE_OLDSETOPTIONS "
536 "failed, errno = %d\n", errno
);
543 os_kill_ptraced_process(pid
, 1);
547 void new_thread(void *stack
, jmp_buf *buf
, void (*handler
)(void))
549 (*buf
)[0].JB_IP
= (unsigned long) handler
;
550 (*buf
)[0].JB_SP
= (unsigned long) stack
+ UM_THREAD_SIZE
-
554 #define INIT_JMP_NEW_THREAD 0
555 #define INIT_JMP_CALLBACK 1
556 #define INIT_JMP_HALT 2
557 #define INIT_JMP_REBOOT 3
559 void switch_threads(jmp_buf *me
, jmp_buf *you
)
561 if (UML_SETJMP(me
) == 0)
565 static jmp_buf initial_jmpbuf
;
567 /* XXX Make these percpu */
568 static void (*cb_proc
)(void *arg
);
570 static jmp_buf *cb_back
;
572 int start_idle_thread(void *stack
, jmp_buf *switch_buf
)
576 set_handler(SIGWINCH
);
579 * Can't use UML_SETJMP or UML_LONGJMP here because they save
580 * and restore signals, with the possible side-effect of
581 * trying to handle any signals which came when they were
582 * blocked, which can't be done on this stack.
583 * Signals must be blocked when jumping back here and restored
584 * after returning to the jumper.
586 n
= setjmp(initial_jmpbuf
);
588 case INIT_JMP_NEW_THREAD
:
589 (*switch_buf
)[0].JB_IP
= (unsigned long) uml_finishsetup
;
590 (*switch_buf
)[0].JB_SP
= (unsigned long) stack
+
591 UM_THREAD_SIZE
- sizeof(void *);
593 case INIT_JMP_CALLBACK
:
595 longjmp(*cb_back
, 1);
600 case INIT_JMP_REBOOT
:
604 printk(UM_KERN_ERR
"Bad sigsetjmp return in "
605 "start_idle_thread - %d\n", n
);
608 longjmp(*switch_buf
, 1);
611 void initial_thread_cb_skas(void (*proc
)(void *), void *arg
)
620 if (UML_SETJMP(&here
) == 0)
621 UML_LONGJMP(&initial_jmpbuf
, INIT_JMP_CALLBACK
);
632 UML_LONGJMP(&initial_jmpbuf
, INIT_JMP_HALT
);
635 void reboot_skas(void)
638 UML_LONGJMP(&initial_jmpbuf
, INIT_JMP_REBOOT
);
641 void __switch_mm(struct mm_id
*mm_idp
)
643 userspace_pid
[0] = mm_idp
->u
.pid
;