Merge branch 'master'
[linux-2.6] / arch / um / kernel / skas / process.c
1 /* 
2  * Copyright (C) 2002- 2004 Jeff Dike (jdike@addtoit.com)
3  * Licensed under the GPL
4  */
5
6 #include <stdlib.h>
7 #include <string.h>
8 #include <unistd.h>
9 #include <errno.h>
10 #include <signal.h>
11 #include <setjmp.h>
12 #include <sched.h>
13 #include <sys/wait.h>
14 #include <sys/mman.h>
15 #include <sys/user.h>
16 #include <sys/time.h>
17 #include <asm/unistd.h>
18 #include <asm/types.h>
19 #include "user.h"
20 #include "ptrace_user.h"
21 #include "sysdep/ptrace.h"
22 #include "user_util.h"
23 #include "kern_util.h"
24 #include "skas.h"
25 #include "stub-data.h"
26 #include "mm_id.h"
27 #include "sysdep/sigcontext.h"
28 #include "sysdep/stub.h"
29 #include "os.h"
30 #include "proc_mm.h"
31 #include "skas_ptrace.h"
32 #include "chan_user.h"
33 #include "registers.h"
34 #include "mem.h"
35 #include "uml-config.h"
36 #include "process.h"
37
38 int is_skas_winch(int pid, int fd, void *data)
39 {
40         if(pid != os_getpgrp())
41                 return(0);
42
43         register_winch_irq(-1, fd, -1, data);
44         return(1);
45 }
46
47 void wait_stub_done(int pid, int sig, char * fname)
48 {
49         int n, status, err;
50
51         do {
52                 if ( sig != -1 ) {
53                         err = ptrace(PTRACE_CONT, pid, 0, sig);
54                         if(err)
55                                 panic("%s : continue failed, errno = %d\n",
56                                       fname, errno);
57                 }
58                 sig = 0;
59
60                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
61         } while((n >= 0) && WIFSTOPPED(status) &&
62                 ((WSTOPSIG(status) == SIGVTALRM) ||
63                  /* running UML inside a detached screen can cause
64                   * SIGWINCHes
65                   */
66                  (WSTOPSIG(status) == SIGWINCH)));
67
68         if((n < 0) || !WIFSTOPPED(status) ||
69            (WSTOPSIG(status) != SIGUSR1 && WSTOPSIG(status) != SIGTRAP)){
70                 unsigned long regs[HOST_FRAME_SIZE];
71                 if(ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
72                         printk("Failed to get registers from stub, "
73                                "errno = %d\n", errno);
74                 else {
75                         int i;
76
77                         printk("Stub registers -\n");
78                         for(i = 0; i < HOST_FRAME_SIZE; i++)
79                                 printk("\t%d - %lx\n", i, regs[i]);
80                 }
81                 panic("%s : failed to wait for SIGUSR1/SIGTRAP, "
82                       "pid = %d, n = %d, errno = %d, status = 0x%x\n",
83                       fname, pid, n, errno, status);
84         }
85 }
86
87 void get_skas_faultinfo(int pid, struct faultinfo * fi)
88 {
89         int err;
90
91         if(ptrace_faultinfo){
92                 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
93                 if(err)
94                         panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
95                               "errno = %d\n", errno);
96
97                 /* Special handling for i386, which has different structs */
98                 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
99                         memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
100                                sizeof(struct faultinfo) -
101                                sizeof(struct ptrace_faultinfo));
102         }
103         else {
104                 wait_stub_done(pid, SIGSEGV, "get_skas_faultinfo");
105
106                 /* faultinfo is prepared by the stub-segv-handler at start of
107                  * the stub stack page. We just have to copy it.
108                  */
109                 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
110         }
111 }
112
113 static void handle_segv(int pid, union uml_pt_regs * regs)
114 {
115         get_skas_faultinfo(pid, &regs->skas.faultinfo);
116         segv(regs->skas.faultinfo, 0, 1, NULL);
117 }
118
119 /*To use the same value of using_sysemu as the caller, ask it that value (in local_using_sysemu)*/
120 static void handle_trap(int pid, union uml_pt_regs *regs, int local_using_sysemu)
121 {
122         int err, status;
123
124         /* Mark this as a syscall */
125         UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->skas.regs);
126
127         if (!local_using_sysemu)
128         {
129                 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET, __NR_getpid);
130                 if(err < 0)
131                         panic("handle_trap - nullifying syscall failed errno = %d\n",
132                               errno);
133
134                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
135                 if(err < 0)
136                         panic("handle_trap - continuing to end of syscall failed, "
137                               "errno = %d\n", errno);
138
139                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
140                 if((err < 0) || !WIFSTOPPED(status) ||
141                    (WSTOPSIG(status) != SIGTRAP + 0x80))
142                         panic("handle_trap - failed to wait at end of syscall, "
143                               "errno = %d, status = %d\n", errno, status);
144         }
145
146         handle_syscall(regs);
147 }
148
149 extern int __syscall_stub_start;
150 int stub_code_fd = -1;
151 __u64 stub_code_offset;
152
153 static int userspace_tramp(void *stack)
154 {
155         void *addr;
156
157         ptrace(PTRACE_TRACEME, 0, 0, 0);
158
159         init_new_thread_signals(1);
160         enable_timer();
161
162         if(!proc_mm){
163                 /* This has a pte, but it can't be mapped in with the usual
164                  * tlb_flush mechanism because this is part of that mechanism
165                  */
166                 addr = mmap64((void *) UML_CONFIG_STUB_CODE, page_size(),
167                               PROT_EXEC, MAP_FIXED | MAP_PRIVATE,
168                               stub_code_fd, stub_code_offset);
169                 if(addr == MAP_FAILED){
170                         printk("mapping stub code failed, errno = %d\n",
171                                errno);
172                         exit(1);
173                 }
174
175                 if(stack != NULL){
176                         int fd;
177                         __u64 offset;
178
179                         fd = phys_mapping(to_phys(stack), &offset);
180                         addr = mmap((void *) UML_CONFIG_STUB_DATA, page_size(),
181                                     PROT_READ | PROT_WRITE,
182                                     MAP_FIXED | MAP_SHARED, fd, offset);
183                         if(addr == MAP_FAILED){
184                                 printk("mapping stub stack failed, "
185                                        "errno = %d\n", errno);
186                                 exit(1);
187                         }
188                 }
189         }
190         if(!ptrace_faultinfo){
191                 unsigned long v = UML_CONFIG_STUB_CODE +
192                                   (unsigned long) stub_segv_handler -
193                                   (unsigned long) &__syscall_stub_start;
194
195                 set_sigstack((void *) UML_CONFIG_STUB_DATA, page_size());
196                 set_handler(SIGSEGV, (void *) v, SA_ONSTACK,
197                             SIGIO, SIGWINCH, SIGALRM, SIGVTALRM,
198                             SIGUSR1, -1);
199         }
200
201         os_stop_process(os_getpid());
202         return(0);
203 }
204
205 /* Each element set once, and only accessed by a single processor anyway */
206 #undef NR_CPUS
207 #define NR_CPUS 1
208 int userspace_pid[NR_CPUS];
209
210 int start_userspace(unsigned long stub_stack)
211 {
212         void *stack;
213         unsigned long sp;
214         int pid, status, n, flags;
215
216         if ( stub_code_fd == -1 )
217                 stub_code_fd = phys_mapping(to_phys(&__syscall_stub_start),
218                                             &stub_code_offset);
219
220         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
221                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
222         if(stack == MAP_FAILED)
223                 panic("start_userspace : mmap failed, errno = %d", errno);
224         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
225
226         flags = CLONE_FILES | SIGCHLD;
227         if(proc_mm) flags |= CLONE_VM;
228         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
229         if(pid < 0)
230                 panic("start_userspace : clone failed, errno = %d", errno);
231
232         do {
233                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
234                 if(n < 0)
235                         panic("start_userspace : wait failed, errno = %d", 
236                               errno);
237         } while(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
238
239         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
240                 panic("start_userspace : expected SIGSTOP, got status = %d",
241                       status);
242
243         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, (void *)PTRACE_O_TRACESYSGOOD) < 0)
244                 panic("start_userspace : PTRACE_SETOPTIONS failed, errno=%d\n",
245                       errno);
246
247         if(munmap(stack, PAGE_SIZE) < 0)
248                 panic("start_userspace : munmap failed, errno = %d\n", errno);
249
250         return(pid);
251 }
252
253 void userspace(union uml_pt_regs *regs)
254 {
255         int err, status, op, pid = userspace_pid[0];
256         int local_using_sysemu; /*To prevent races if using_sysemu changes under us.*/
257
258         while(1){
259                 restore_registers(pid, regs);
260
261                 /* Now we set local_using_sysemu to be used for one loop */
262                 local_using_sysemu = get_using_sysemu();
263
264                 op = SELECT_PTRACE_OPERATION(local_using_sysemu, singlestepping(NULL));
265
266                 err = ptrace(op, pid, 0, 0);
267                 if(err)
268                         panic("userspace - could not resume userspace process, "
269                               "pid=%d, ptrace operation = %d, errno = %d\n",
270                               op, errno);
271
272                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED));
273                 if(err < 0)
274                         panic("userspace - waitpid failed, errno = %d\n", 
275                               errno);
276
277                 regs->skas.is_user = 1;
278                 save_registers(pid, regs);
279                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
280
281                 if(WIFSTOPPED(status)){
282                         switch(WSTOPSIG(status)){
283                         case SIGSEGV:
284                                 if(PTRACE_FULL_FAULTINFO || !ptrace_faultinfo)
285                                         user_signal(SIGSEGV, regs, pid);
286                                 else handle_segv(pid, regs);
287                                 break;
288                         case SIGTRAP + 0x80:
289                                 handle_trap(pid, regs, local_using_sysemu);
290                                 break;
291                         case SIGTRAP:
292                                 relay_signal(SIGTRAP, regs);
293                                 break;
294                         case SIGIO:
295                         case SIGVTALRM:
296                         case SIGILL:
297                         case SIGBUS:
298                         case SIGFPE:
299                         case SIGWINCH:
300                                 user_signal(WSTOPSIG(status), regs, pid);
301                                 break;
302                         default:
303                                 printk("userspace - child stopped with signal "
304                                        "%d\n", WSTOPSIG(status));
305                         }
306                         pid = userspace_pid[0];
307                         interrupt_end();
308
309                         /* Avoid -ERESTARTSYS handling in host */
310                         PT_SYSCALL_NR(regs->skas.regs) = -1;
311                 }
312         }
313 }
314 #define INIT_JMP_NEW_THREAD 0
315 #define INIT_JMP_REMOVE_SIGSTACK 1
316 #define INIT_JMP_CALLBACK 2
317 #define INIT_JMP_HALT 3
318 #define INIT_JMP_REBOOT 4
319
320
321 int copy_context_skas0(unsigned long new_stack, int pid)
322 {
323         int err;
324         unsigned long regs[MAX_REG_NR];
325         unsigned long current_stack = current_stub_stack();
326         struct stub_data *data = (struct stub_data *) current_stack;
327         struct stub_data *child_data = (struct stub_data *) new_stack;
328         __u64 new_offset;
329         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
330
331         /* prepare offset and fd of child's stack as argument for parent's
332          * and child's mmap2 calls
333          */
334         *data = ((struct stub_data) { .offset   = MMAP_OFFSET(new_offset),
335                                       .fd       = new_fd,
336                                       .timer    = ((struct itimerval)
337                                                    { { 0, 1000000 / hz() },
338                                                      { 0, 1000000 / hz() }})});
339         get_safe_registers(regs);
340
341         /* Set parent's instruction pointer to start of clone-stub */
342         regs[REGS_IP_INDEX] = UML_CONFIG_STUB_CODE +
343                                 (unsigned long) stub_clone_handler -
344                                 (unsigned long) &__syscall_stub_start;
345         regs[REGS_SP_INDEX] = UML_CONFIG_STUB_DATA + PAGE_SIZE -
346                 sizeof(void *);
347         err = ptrace_setregs(pid, regs);
348         if(err < 0)
349                 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
350                       "pid = %d, errno = %d\n", pid, errno);
351
352         /* set a well known return code for detection of child write failure */
353         child_data->err = 12345678;
354
355         /* Wait, until parent has finished its work: read child's pid from
356          * parent's stack, and check, if bad result.
357          */
358         wait_stub_done(pid, 0, "copy_context_skas0");
359
360         pid = data->err;
361         if(pid < 0)
362                 panic("copy_context_skas0 - stub-parent reports error %d\n",
363                       pid);
364
365         /* Wait, until child has finished too: read child's result from
366          * child's stack and check it.
367          */
368         wait_stub_done(pid, -1, "copy_context_skas0");
369         if (child_data->err != UML_CONFIG_STUB_DATA)
370                 panic("copy_context_skas0 - stub-child reports error %d\n",
371                       child_data->err);
372
373         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
374                    (void *)PTRACE_O_TRACESYSGOOD) < 0)
375                 panic("copy_context_skas0 : PTRACE_SETOPTIONS failed, "
376                       "errno = %d\n", errno);
377
378         return pid;
379 }
380
381 /*
382  * This is used only, if stub pages are needed, while proc_mm is
383  * availabl. Opening /proc/mm creates a new mm_context, which lacks
384  * the stub-pages. Thus, we map them using /proc/mm-fd
385  */
386 void map_stub_pages(int fd, unsigned long code,
387                     unsigned long data, unsigned long stack)
388 {
389         struct proc_mm_op mmop;
390         int n;
391
392         mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
393                                       .u         =
394                                       { .mmap    =
395                                         { .addr    = code,
396                                           .len     = PAGE_SIZE,
397                                           .prot    = PROT_EXEC,
398                                           .flags   = MAP_FIXED | MAP_PRIVATE,
399                                           .fd      = stub_code_fd,
400                                           .offset  = stub_code_offset
401         } } });
402         n = os_write_file(fd, &mmop, sizeof(mmop));
403         if(n != sizeof(mmop))
404                 panic("map_stub_pages : /proc/mm map for code failed, "
405                       "err = %d\n", -n);
406
407         if ( stack ) {
408                 __u64 map_offset;
409                 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
410                 mmop = ((struct proc_mm_op)
411                                 { .op        = MM_MMAP,
412                                   .u         =
413                                   { .mmap    =
414                                     { .addr    = data,
415                                       .len     = PAGE_SIZE,
416                                       .prot    = PROT_READ | PROT_WRITE,
417                                       .flags   = MAP_FIXED | MAP_SHARED,
418                                       .fd      = map_fd,
419                                       .offset  = map_offset
420                 } } });
421                 n = os_write_file(fd, &mmop, sizeof(mmop));
422                 if(n != sizeof(mmop))
423                         panic("map_stub_pages : /proc/mm map for data failed, "
424                               "err = %d\n", -n);
425         }
426 }
427
428 void new_thread(void *stack, void **switch_buf_ptr, void **fork_buf_ptr,
429                 void (*handler)(int))
430 {
431         unsigned long flags;
432         sigjmp_buf switch_buf, fork_buf;
433
434         *switch_buf_ptr = &switch_buf;
435         *fork_buf_ptr = &fork_buf;
436
437         /* Somewhat subtle - siglongjmp restores the signal mask before doing
438          * the longjmp.  This means that when jumping from one stack to another
439          * when the target stack has interrupts enabled, an interrupt may occur
440          * on the source stack.  This is bad when starting up a process because
441          * it's not supposed to get timer ticks until it has been scheduled.
442          * So, we disable interrupts around the sigsetjmp to ensure that
443          * they can't happen until we get back here where they are safe.
444          */
445         flags = get_signals();
446         block_signals();
447         if(sigsetjmp(fork_buf, 1) == 0)
448                 new_thread_proc(stack, handler);
449
450         remove_sigstack();
451
452         set_signals(flags);
453 }
454
455 void thread_wait(void *sw, void *fb)
456 {
457         sigjmp_buf buf, **switch_buf = sw, *fork_buf;
458
459         *switch_buf = &buf;
460         fork_buf = fb;
461         if(sigsetjmp(buf, 1) == 0)
462                 siglongjmp(*fork_buf, INIT_JMP_REMOVE_SIGSTACK);
463 }
464
465 void switch_threads(void *me, void *next)
466 {
467         sigjmp_buf my_buf, **me_ptr = me, *next_buf = next;
468         
469         *me_ptr = &my_buf;
470         if(sigsetjmp(my_buf, 1) == 0)
471                 siglongjmp(*next_buf, 1);
472 }
473
474 static sigjmp_buf initial_jmpbuf;
475
476 /* XXX Make these percpu */
477 static void (*cb_proc)(void *arg);
478 static void *cb_arg;
479 static sigjmp_buf *cb_back;
480
481 int start_idle_thread(void *stack, void *switch_buf_ptr, void **fork_buf_ptr)
482 {
483         sigjmp_buf **switch_buf = switch_buf_ptr;
484         int n;
485
486         set_handler(SIGWINCH, (__sighandler_t) sig_handler,
487                     SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGALRM,
488                     SIGVTALRM, -1);
489
490         *fork_buf_ptr = &initial_jmpbuf;
491         n = sigsetjmp(initial_jmpbuf, 1);
492         switch(n){
493         case INIT_JMP_NEW_THREAD:
494                 new_thread_proc((void *) stack, new_thread_handler);
495                 break;
496         case INIT_JMP_REMOVE_SIGSTACK:
497                 remove_sigstack();
498                 break;
499         case INIT_JMP_CALLBACK:
500                 (*cb_proc)(cb_arg);
501                 siglongjmp(*cb_back, 1);
502                 break;
503         case INIT_JMP_HALT:
504                 kmalloc_ok = 0;
505                 return(0);
506         case INIT_JMP_REBOOT:
507                 kmalloc_ok = 0;
508                 return(1);
509         default:
510                 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
511         }
512         siglongjmp(**switch_buf, 1);
513 }
514
515 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
516 {
517         sigjmp_buf here;
518
519         cb_proc = proc;
520         cb_arg = arg;
521         cb_back = &here;
522
523         block_signals();
524         if(sigsetjmp(here, 1) == 0)
525                 siglongjmp(initial_jmpbuf, INIT_JMP_CALLBACK);
526         unblock_signals();
527
528         cb_proc = NULL;
529         cb_arg = NULL;
530         cb_back = NULL;
531 }
532
533 void halt_skas(void)
534 {
535         block_signals();
536         siglongjmp(initial_jmpbuf, INIT_JMP_HALT);
537 }
538
539 void reboot_skas(void)
540 {
541         block_signals();
542         siglongjmp(initial_jmpbuf, INIT_JMP_REBOOT);
543 }
544
545 void switch_mm_skas(struct mm_id *mm_idp)
546 {
547         int err;
548
549 #warning need cpu pid in switch_mm_skas
550         if(proc_mm){
551                 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
552                              mm_idp->u.mm_fd);
553                 if(err)
554                         panic("switch_mm_skas - PTRACE_SWITCH_MM failed, "
555                               "errno = %d\n", errno);
556         }
557         else userspace_pid[0] = mm_idp->u.pid;
558 }
559
560 /*
561  * Overrides for Emacs so that we follow Linus's tabbing style.
562  * Emacs will notice this stuff at the end of the file and automatically
563  * adjust the settings for this buffer only.  This must remain at the end
564  * of the file.
565  * ---------------------------------------------------------------------------
566  * Local variables:
567  * c-file-style: "linux"
568  * End:
569  */