2 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
12 #include <sys/ptrace.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"
23 #include "ptrace_user.h"
24 #include "registers.h"
26 #include "skas_ptrace.h"
28 #include "sysdep/stub.h"
30 int is_skas_winch(int pid
, int fd
, void *data
)
35 register_winch_irq(-1, fd
, -1, data
, 0);
39 static int ptrace_dump_regs(int pid
)
41 unsigned long regs
[MAX_REG_NR
];
44 if (ptrace(PTRACE_GETREGS
, pid
, 0, regs
) < 0)
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
]);
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
)
68 CATCH_EINTR(n
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
69 if ((n
< 0) || !WIFSTOPPED(status
))
72 if (((1 << WSTOPSIG(status
)) & STUB_SIG_MASK
) == 0)
75 err
= ptrace(PTRACE_CONT
, pid
, 0, 0);
77 printk(UM_KERN_ERR
"wait_stub_done : continue failed, "
78 "errno = %d\n", errno
);
83 if (((1 << WSTOPSIG(status
)) & STUB_DONE_MASK
) != 0)
87 err
= ptrace_dump_regs(pid
);
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
,
97 extern unsigned long current_stub_stack(void);
99 static void get_skas_faultinfo(int pid
, struct faultinfo
*fi
)
103 if (ptrace_faultinfo
) {
104 err
= ptrace(PTRACE_FAULTINFO
, pid
, 0, fi
);
106 printk(UM_KERN_ERR
"get_skas_faultinfo - "
107 "PTRACE_FAULTINFO failed, errno = %d\n", errno
);
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
));
118 unsigned long fpregs
[FP_SIZE
];
120 err
= get_fp_registers(pid
, fpregs
);
122 printk(UM_KERN_ERR
"save_fp_registers returned %d\n",
126 err
= ptrace(PTRACE_CONT
, pid
, 0, SIGSEGV
);
128 printk(UM_KERN_ERR
"Failed to continue stub, pid = %d, "
129 "errno = %d\n", pid
, errno
);
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
);
142 printk(UM_KERN_ERR
"put_fp_registers returned %d\n",
149 static void handle_segv(int pid
, struct uml_pt_regs
* regs
)
151 get_skas_faultinfo(pid
, ®s
->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
)
164 if ((UPT_IP(regs
) >= STUB_START
) && (UPT_IP(regs
) < STUB_END
))
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
,
175 printk(UM_KERN_ERR
"handle_trap - nullifying syscall "
176 "failed, errno = %d\n", errno
);
180 err
= ptrace(PTRACE_SYSCALL
, pid
, 0, 0);
182 printk(UM_KERN_ERR
"handle_trap - continuing to end of "
183 "syscall failed, errno = %d\n", errno
);
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
);
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",
201 handle_syscall(regs
);
204 extern int __syscall_stub_start
;
206 static int userspace_tramp(void *stack
)
211 ptrace(PTRACE_TRACEME
, 0, 0, 0);
213 signal(SIGTERM
, SIG_DFL
);
214 signal(SIGWINCH
, SIG_IGN
);
215 err
= set_interval();
217 printk(UM_KERN_ERR
"userspace_tramp - setting timer failed, "
218 "errno = %d\n", err
);
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
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
);
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",
251 if (!ptrace_faultinfo
&& (stack
!= NULL
)) {
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
);
270 kill(os_getpid(), SIGSTOP
);
274 /* Each element set once, and only accessed by a single processor anyway */
277 int userspace_pid
[NR_CPUS
];
279 int start_userspace(unsigned long stub_stack
)
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
) {
290 printk(UM_KERN_ERR
"start_userspace : mmap failed, "
291 "errno = %d\n", errno
);
295 sp
= (unsigned long) stack
+ UM_KERN_PAGE_SIZE
- sizeof(void *);
303 pid
= clone(userspace_tramp
, (void *) sp
, flags
, (void *) stub_stack
);
306 printk(UM_KERN_ERR
"start_userspace : clone failed, "
307 "errno = %d\n", errno
);
312 CATCH_EINTR(n
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
315 printk(UM_KERN_ERR
"start_userspace : wait failed, "
316 "errno = %d\n", errno
);
319 } while (WIFSTOPPED(status
) && (WSTOPSIG(status
) == SIGVTALRM
));
321 if (!WIFSTOPPED(status
) || (WSTOPSIG(status
) != SIGSTOP
)) {
323 printk(UM_KERN_ERR
"start_userspace : expected SIGSTOP, got "
324 "status = %d\n", status
);
328 if (ptrace(PTRACE_OLDSETOPTIONS
, pid
, NULL
,
329 (void *) PTRACE_O_TRACESYSGOOD
) < 0) {
331 printk(UM_KERN_ERR
"start_userspace : PTRACE_OLDSETOPTIONS "
332 "failed, errno = %d\n", errno
);
336 if (munmap(stack
, UM_KERN_PAGE_SIZE
) < 0) {
338 printk(UM_KERN_ERR
"start_userspace : munmap failed, "
339 "errno = %d\n", errno
);
346 os_kill_ptraced_process(pid
, 1);
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
;
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
))
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
);
388 CATCH_EINTR(err
= waitpid(pid
, &status
, WUNTRACED
| __WALL
));
390 printk(UM_KERN_ERR
"userspace - wait failed, "
391 "errno = %d\n", errno
);
396 if (ptrace(PTRACE_GETREGS
, pid
, 0, regs
->gp
)) {
397 printk(UM_KERN_ERR
"userspace - PTRACE_GETREGS failed, "
398 "errno = %d\n", errno
);
402 UPT_SYSCALL_NR(regs
) = -1; /* Assume: It's not a syscall */
404 if (WIFSTOPPED(status
)) {
405 int sig
= WSTOPSIG(status
);
408 if (PTRACE_FULL_FAULTINFO
||
410 get_skas_faultinfo(pid
,
412 (*sig_info
[SIGSEGV
])(SIGSEGV
, regs
);
414 else handle_segv(pid
, regs
);
417 handle_trap(pid
, regs
, local_using_sysemu
);
420 relay_signal(SIGTRAP
, regs
);
427 (*sig_info
[sig
])(sig
, regs
);
429 nsecs
= timer
.it_value
.tv_sec
*
431 timer
.it_value
.tv_usec
*
441 (*sig_info
[sig
])(sig
, regs
);
445 printk(UM_KERN_ERR
"userspace - child stopped "
446 "with signal %d\n", sig
);
449 pid
= userspace_pid
[0];
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
-
470 #ifdef __SIGNAL_FRAMESIZE
471 thread_regs
[REGS_SP_INDEX
] -= __SIGNAL_FRAMESIZE
;
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
};
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
),
494 .timer
= ((struct itimerval
)
496 .it_interval
= tv
}) });
498 err
= ptrace_setregs(pid
, thread_regs
);
501 printk(UM_KERN_ERR
"copy_context_skas0 : PTRACE_SETREGS "
502 "failed, pid = %d, errno = %d\n", pid
, -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);
516 printk(UM_KERN_ERR
"Failed to continue new process, pid = %d, "
517 "errno = %d\n", pid
, errno
);
525 printk(UM_KERN_ERR
"copy_context_skas0 - stub-parent reports "
531 * Wait, until child has finished too: read child's result from
532 * child's stack and check it.
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
;
542 if (ptrace(PTRACE_OLDSETOPTIONS
, pid
, NULL
,
543 (void *)PTRACE_O_TRACESYSGOOD
) < 0) {
545 printk(UM_KERN_ERR
"copy_context_skas0 : PTRACE_OLDSETOPTIONS "
546 "failed, errno = %d\n", errno
);
553 os_kill_ptraced_process(pid
, 1);
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
,
565 struct proc_mm_op mmop
;
567 unsigned long long code_offset
;
568 int code_fd
= phys_mapping(to_phys((void *) &__syscall_stub_start
),
571 mmop
= ((struct proc_mm_op
) { .op
= MM_MMAP
,
575 .len
= UM_KERN_PAGE_SIZE
,
577 .flags
= MAP_FIXED
| MAP_PRIVATE
,
579 .offset
= code_offset
581 CATCH_EINTR(n
= write(fd
, &mmop
, sizeof(mmop
)));
582 if (n
!= sizeof(mmop
)) {
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
);
593 unsigned long long map_offset
;
594 int map_fd
= phys_mapping(to_phys((void *)stack
), &map_offset
);
595 mmop
= ((struct proc_mm_op
)
600 .len
= UM_KERN_PAGE_SIZE
,
601 .prot
= PROT_READ
| PROT_WRITE
,
602 .flags
= MAP_FIXED
| MAP_SHARED
,
606 CATCH_EINTR(n
= write(fd
, &mmop
, sizeof(mmop
)));
607 if (n
!= sizeof(mmop
)) {
609 printk(UM_KERN_ERR
"map_stub_pages : /proc/mm map for "
610 "data failed, err = %d\n", n
);
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
-
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)
636 static jmp_buf initial_jmpbuf
;
638 /* XXX Make these percpu */
639 static void (*cb_proc
)(void *arg
);
641 static jmp_buf *cb_back
;
643 int start_idle_thread(void *stack
, jmp_buf *switch_buf
)
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
);
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 *);
665 case INIT_JMP_CALLBACK
:
667 longjmp(*cb_back
, 1);
672 case INIT_JMP_REBOOT
:
676 printk(UM_KERN_ERR
"Bad sigsetjmp return in "
677 "start_idle_thread - %d\n", n
);
680 longjmp(*switch_buf
, 1);
683 void initial_thread_cb_skas(void (*proc
)(void *), void *arg
)
692 if (UML_SETJMP(&here
) == 0)
693 UML_LONGJMP(&initial_jmpbuf
, INIT_JMP_CALLBACK
);
704 UML_LONGJMP(&initial_jmpbuf
, INIT_JMP_HALT
);
707 void reboot_skas(void)
710 UML_LONGJMP(&initial_jmpbuf
, INIT_JMP_REBOOT
);
713 void __switch_mm(struct mm_id
*mm_idp
)
717 /* FIXME: need cpu pid in __switch_mm */
719 err
= ptrace(PTRACE_SWITCH_MM
, userspace_pid
[0], 0,
722 printk(UM_KERN_ERR
"__switch_mm - PTRACE_SWITCH_MM "
723 "failed, errno = %d\n", errno
);
727 else userspace_pid
[0] = mm_idp
->u
.pid
;