vserver 2.0 rc7
[linux-2.6.git] / arch / ppc64 / kernel / process.c
1 /*
2  *  linux/arch/ppc64/kernel/process.c
3  *
4  *  Derived from "arch/i386/kernel/process.c"
5  *    Copyright (C) 1995  Linus Torvalds
6  *
7  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
8  *  Paul Mackerras (paulus@cs.anu.edu.au)
9  *
10  *  PowerPC version 
11  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
12  *
13  *  This program is free software; you can redistribute it and/or
14  *  modify it under the terms of the GNU General Public License
15  *  as published by the Free Software Foundation; either version
16  *  2 of the License, or (at your option) any later version.
17  */
18
19 #include <linux/config.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/mm.h>
25 #include <linux/smp.h>
26 #include <linux/smp_lock.h>
27 #include <linux/stddef.h>
28 #include <linux/unistd.h>
29 #include <linux/slab.h>
30 #include <linux/user.h>
31 #include <linux/elf.h>
32 #include <linux/init.h>
33 #include <linux/init_task.h>
34 #include <linux/prctl.h>
35 #include <linux/ptrace.h>
36 #include <linux/kallsyms.h>
37 #include <linux/interrupt.h>
38 #include <linux/utsname.h>
39
40 #include <asm/pgtable.h>
41 #include <asm/uaccess.h>
42 #include <asm/system.h>
43 #include <asm/io.h>
44 #include <asm/processor.h>
45 #include <asm/mmu.h>
46 #include <asm/mmu_context.h>
47 #include <asm/prom.h>
48 #include <asm/ppcdebug.h>
49 #include <asm/machdep.h>
50 #include <asm/iSeries/HvCallHpt.h>
51 #include <asm/cputable.h>
52 #include <asm/sections.h>
53 #include <asm/tlbflush.h>
54 #include <asm/time.h>
55
56 #ifndef CONFIG_SMP
57 struct task_struct *last_task_used_math = NULL;
58 struct task_struct *last_task_used_altivec = NULL;
59 #endif
60
61 struct mm_struct ioremap_mm = {
62         .pgd            = ioremap_dir,
63         .mm_users       = ATOMIC_INIT(2),
64         .mm_count       = ATOMIC_INIT(1),
65         .cpu_vm_mask    = CPU_MASK_ALL,
66         .page_table_lock = SPIN_LOCK_UNLOCKED,
67 };
68
69 /*
70  * Make sure the floating-point register state in the
71  * the thread_struct is up to date for task tsk.
72  */
73 void flush_fp_to_thread(struct task_struct *tsk)
74 {
75         if (tsk->thread.regs) {
76                 /*
77                  * We need to disable preemption here because if we didn't,
78                  * another process could get scheduled after the regs->msr
79                  * test but before we have finished saving the FP registers
80                  * to the thread_struct.  That process could take over the
81                  * FPU, and then when we get scheduled again we would store
82                  * bogus values for the remaining FP registers.
83                  */
84                 preempt_disable();
85                 if (tsk->thread.regs->msr & MSR_FP) {
86 #ifdef CONFIG_SMP
87                         /*
88                          * This should only ever be called for current or
89                          * for a stopped child process.  Since we save away
90                          * the FP register state on context switch on SMP,
91                          * there is something wrong if a stopped child appears
92                          * to still have its FP state in the CPU registers.
93                          */
94                         BUG_ON(tsk != current);
95 #endif
96                         giveup_fpu(current);
97                 }
98                 preempt_enable();
99         }
100 }
101
102 void enable_kernel_fp(void)
103 {
104         WARN_ON(preemptible());
105
106 #ifdef CONFIG_SMP
107         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
108                 giveup_fpu(current);
109         else
110                 giveup_fpu(NULL);       /* just enables FP for kernel */
111 #else
112         giveup_fpu(last_task_used_math);
113 #endif /* CONFIG_SMP */
114 }
115 EXPORT_SYMBOL(enable_kernel_fp);
116
117 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
118 {
119         if (!tsk->thread.regs)
120                 return 0;
121         flush_fp_to_thread(current);
122
123         memcpy(fpregs, &tsk->thread.fpr[0], sizeof(*fpregs));
124
125         return 1;
126 }
127
128 #ifdef CONFIG_ALTIVEC
129
130 void enable_kernel_altivec(void)
131 {
132         WARN_ON(preemptible());
133
134 #ifdef CONFIG_SMP
135         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
136                 giveup_altivec(current);
137         else
138                 giveup_altivec(NULL);   /* just enables FP for kernel */
139 #else
140         giveup_altivec(last_task_used_altivec);
141 #endif /* CONFIG_SMP */
142 }
143 EXPORT_SYMBOL(enable_kernel_altivec);
144
145 /*
146  * Make sure the VMX/Altivec register state in the
147  * the thread_struct is up to date for task tsk.
148  */
149 void flush_altivec_to_thread(struct task_struct *tsk)
150 {
151         if (tsk->thread.regs) {
152                 preempt_disable();
153                 if (tsk->thread.regs->msr & MSR_VEC) {
154 #ifdef CONFIG_SMP
155                         BUG_ON(tsk != current);
156 #endif
157                         giveup_altivec(current);
158                 }
159                 preempt_enable();
160         }
161 }
162
163 int dump_task_altivec(struct pt_regs *regs, elf_vrregset_t *vrregs)
164 {
165         flush_altivec_to_thread(current);
166         memcpy(vrregs, &current->thread.vr[0], sizeof(*vrregs));
167         return 1;
168 }
169
170 #endif /* CONFIG_ALTIVEC */
171
172 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
173
174 struct task_struct *__switch_to(struct task_struct *prev,
175                                 struct task_struct *new)
176 {
177         struct thread_struct *new_thread, *old_thread;
178         unsigned long flags;
179         struct task_struct *last;
180
181 #ifdef CONFIG_SMP
182         /* avoid complexity of lazy save/restore of fpu
183          * by just saving it every time we switch out if
184          * this task used the fpu during the last quantum.
185          * 
186          * If it tries to use the fpu again, it'll trap and
187          * reload its fp regs.  So we don't have to do a restore
188          * every switch, just a save.
189          *  -- Cort
190          */
191         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
192                 giveup_fpu(prev);
193 #ifdef CONFIG_ALTIVEC
194         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
195                 giveup_altivec(prev);
196 #endif /* CONFIG_ALTIVEC */
197 #endif /* CONFIG_SMP */
198
199 #if defined(CONFIG_ALTIVEC) && !defined(CONFIG_SMP)
200         /* Avoid the trap.  On smp this this never happens since
201          * we don't set last_task_used_altivec -- Cort
202          */
203         if (new->thread.regs && last_task_used_altivec == new)
204                 new->thread.regs->msr |= MSR_VEC;
205 #endif /* CONFIG_ALTIVEC */
206
207         flush_tlb_pending();
208
209         new_thread = &new->thread;
210         old_thread = &current->thread;
211
212 /* Collect purr utilization data per process and per processor wise */
213 /* purr is nothing but processor time base                          */
214
215 #if defined(CONFIG_PPC_PSERIES)
216         if (cur_cpu_spec->firmware_features & FW_FEATURE_SPLPAR) {
217                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
218                 long unsigned start_tb, current_tb;
219                 start_tb = old_thread->start_tb;
220                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
221                 old_thread->accum_tb += (current_tb - start_tb);
222                 new_thread->start_tb = current_tb;
223         }
224 #endif
225
226
227         local_irq_save(flags);
228         last = _switch(old_thread, new_thread);
229
230         local_irq_restore(flags);
231
232         return last;
233 }
234
235 static int instructions_to_print = 16;
236
237 static void show_instructions(struct pt_regs *regs)
238 {
239         int i;
240         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
241                         sizeof(int));
242
243         printk("Instruction dump:");
244
245         for (i = 0; i < instructions_to_print; i++) {
246                 int instr;
247
248                 if (!(i % 8))
249                         printk("\n");
250
251                 if (((REGION_ID(pc) != KERNEL_REGION_ID) &&
252                      (REGION_ID(pc) != VMALLOC_REGION_ID)) ||
253                      __get_user(instr, (unsigned int *)pc)) {
254                         printk("XXXXXXXX ");
255                 } else {
256                         if (regs->nip == pc)
257                                 printk("<%08x> ", instr);
258                         else
259                                 printk("%08x ", instr);
260                 }
261
262                 pc += sizeof(int);
263         }
264
265         printk("\n");
266 }
267
268 void show_regs(struct pt_regs * regs)
269 {
270         int i;
271         unsigned long trap;
272
273         printk("NIP: %016lX XER: %08X LR: %016lX CTR: %016lX\n",
274                regs->nip, (unsigned int)regs->xer, regs->link, regs->ctr);
275         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
276                regs, regs->trap, print_tainted(), system_utsname.release);
277         printk("MSR: %016lx EE: %01x PR: %01x FP: %01x ME: %01x "
278                "IR/DR: %01x%01x CR: %08X\n",
279                regs->msr, regs->msr&MSR_EE ? 1 : 0, regs->msr&MSR_PR ? 1 : 0,
280                regs->msr & MSR_FP ? 1 : 0,regs->msr&MSR_ME ? 1 : 0,
281                regs->msr&MSR_IR ? 1 : 0,
282                regs->msr&MSR_DR ? 1 : 0,
283                (unsigned int)regs->ccr);
284         trap = TRAP(regs);
285         printk("DAR: %016lx DSISR: %016lx\n", regs->dar, regs->dsisr);
286         printk("TASK: %p[%d] '%s' THREAD: %p",
287                current, current->pid, current->comm, current->thread_info);
288
289 #ifdef CONFIG_SMP
290         printk(" CPU: %d", smp_processor_id());
291 #endif /* CONFIG_SMP */
292
293         for (i = 0; i < 32; i++) {
294                 if ((i % 4) == 0) {
295                         printk("\n" KERN_INFO "GPR%02d: ", i);
296                 }
297
298                 printk("%016lX ", regs->gpr[i]);
299                 if (i == 13 && !FULL_REGS(regs))
300                         break;
301         }
302         printk("\n");
303         /*
304          * Lookup NIP late so we have the best change of getting the
305          * above info out without failing
306          */
307         printk("NIP [%016lx] ", regs->nip);
308         print_symbol("%s\n", regs->nip);
309         printk("LR [%016lx] ", regs->link);
310         print_symbol("%s\n", regs->link);
311         show_stack(current, (unsigned long *)regs->gpr[1]);
312         if (!user_mode(regs))
313                 show_instructions(regs);
314 }
315
316 void exit_thread(void)
317 {
318 #ifndef CONFIG_SMP
319         if (last_task_used_math == current)
320                 last_task_used_math = NULL;
321 #ifdef CONFIG_ALTIVEC
322         if (last_task_used_altivec == current)
323                 last_task_used_altivec = NULL;
324 #endif /* CONFIG_ALTIVEC */
325 #endif /* CONFIG_SMP */
326 }
327
328 void flush_thread(void)
329 {
330         struct thread_info *t = current_thread_info();
331
332         if (t->flags & _TIF_ABI_PENDING)
333                 t->flags ^= (_TIF_ABI_PENDING | _TIF_32BIT);
334
335 #ifndef CONFIG_SMP
336         if (last_task_used_math == current)
337                 last_task_used_math = NULL;
338 #ifdef CONFIG_ALTIVEC
339         if (last_task_used_altivec == current)
340                 last_task_used_altivec = NULL;
341 #endif /* CONFIG_ALTIVEC */
342 #endif /* CONFIG_SMP */
343 }
344
345 void
346 release_thread(struct task_struct *t)
347 {
348 }
349
350
351 /*
352  * This gets called before we allocate a new thread and copy
353  * the current task into it.
354  */
355 void prepare_to_copy(struct task_struct *tsk)
356 {
357         flush_fp_to_thread(current);
358         flush_altivec_to_thread(current);
359 }
360
361 /*
362  * Copy a thread..
363  */
364 int
365 copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
366             unsigned long unused, struct task_struct *p, struct pt_regs *regs)
367 {
368         struct pt_regs *childregs, *kregs;
369         extern void ret_from_fork(void);
370         unsigned long sp = (unsigned long)p->thread_info + THREAD_SIZE;
371
372         /* Copy registers */
373         sp -= sizeof(struct pt_regs);
374         childregs = (struct pt_regs *) sp;
375         *childregs = *regs;
376         if ((childregs->msr & MSR_PR) == 0) {
377                 /* for kernel thread, set stackptr in new task */
378                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
379                 p->thread.regs = NULL;  /* no user register state */
380                 clear_ti_thread_flag(p->thread_info, TIF_32BIT);
381         } else {
382                 childregs->gpr[1] = usp;
383                 p->thread.regs = childregs;
384                 if (clone_flags & CLONE_SETTLS) {
385                         if (test_thread_flag(TIF_32BIT))
386                                 childregs->gpr[2] = childregs->gpr[6];
387                         else
388                                 childregs->gpr[13] = childregs->gpr[6];
389                 }
390         }
391         childregs->gpr[3] = 0;  /* Result from fork() */
392         sp -= STACK_FRAME_OVERHEAD;
393
394         /*
395          * The way this works is that at some point in the future
396          * some task will call _switch to switch to the new task.
397          * That will pop off the stack frame created below and start
398          * the new task running at ret_from_fork.  The new task will
399          * do some house keeping and then return from the fork or clone
400          * system call, using the stack frame created above.
401          */
402         sp -= sizeof(struct pt_regs);
403         kregs = (struct pt_regs *) sp;
404         sp -= STACK_FRAME_OVERHEAD;
405         p->thread.ksp = sp;
406         if (cpu_has_feature(CPU_FTR_SLB)) {
407                 unsigned long sp_vsid = get_kernel_vsid(sp);
408
409                 sp_vsid <<= SLB_VSID_SHIFT;
410                 sp_vsid |= SLB_VSID_KERNEL;
411                 if (cpu_has_feature(CPU_FTR_16M_PAGE))
412                         sp_vsid |= SLB_VSID_L;
413
414                 p->thread.ksp_vsid = sp_vsid;
415         }
416
417         /*
418          * The PPC64 ABI makes use of a TOC to contain function 
419          * pointers.  The function (ret_from_except) is actually a pointer
420          * to the TOC entry.  The first entry is a pointer to the actual
421          * function.
422          */
423         kregs->nip = *((unsigned long *)ret_from_fork);
424
425         return 0;
426 }
427
428 /*
429  * Set up a thread for executing a new program
430  */
431 void start_thread(struct pt_regs *regs, unsigned long fdptr, unsigned long sp)
432 {
433         unsigned long entry, toc, load_addr = regs->gpr[2];
434
435         /* fdptr is a relocated pointer to the function descriptor for
436          * the elf _start routine.  The first entry in the function
437          * descriptor is the entry address of _start and the second
438          * entry is the TOC value we need to use.
439          */
440         set_fs(USER_DS);
441         __get_user(entry, (unsigned long __user *)fdptr);
442         __get_user(toc, (unsigned long __user *)fdptr+1);
443
444         /* Check whether the e_entry function descriptor entries
445          * need to be relocated before we can use them.
446          */
447         if (load_addr != 0) {
448                 entry += load_addr;
449                 toc   += load_addr;
450         }
451
452         /*
453          * If we exec out of a kernel thread then thread.regs will not be
454          * set. Do it now.
455          */
456         if (!current->thread.regs) {
457                 unsigned long childregs = (unsigned long)current->thread_info +
458                                                 THREAD_SIZE;
459                 childregs -= sizeof(struct pt_regs);
460                 current->thread.regs = (struct pt_regs *)childregs;
461         }
462
463         regs->nip = entry;
464         regs->gpr[1] = sp;
465         regs->gpr[2] = toc;
466         regs->msr = MSR_USER64;
467 #ifndef CONFIG_SMP
468         if (last_task_used_math == current)
469                 last_task_used_math = 0;
470 #endif /* CONFIG_SMP */
471         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
472         current->thread.fpscr = 0;
473 #ifdef CONFIG_ALTIVEC
474 #ifndef CONFIG_SMP
475         if (last_task_used_altivec == current)
476                 last_task_used_altivec = 0;
477 #endif /* CONFIG_SMP */
478         memset(current->thread.vr, 0, sizeof(current->thread.vr));
479         current->thread.vscr.u[0] = 0;
480         current->thread.vscr.u[1] = 0;
481         current->thread.vscr.u[2] = 0;
482         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
483         current->thread.vrsave = 0;
484         current->thread.used_vr = 0;
485 #endif /* CONFIG_ALTIVEC */
486 }
487 EXPORT_SYMBOL(start_thread);
488
489 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
490 {
491         struct pt_regs *regs = tsk->thread.regs;
492
493         if (val > PR_FP_EXC_PRECISE)
494                 return -EINVAL;
495         tsk->thread.fpexc_mode = __pack_fe01(val);
496         if (regs != NULL && (regs->msr & MSR_FP) != 0)
497                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
498                         | tsk->thread.fpexc_mode;
499         return 0;
500 }
501
502 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
503 {
504         unsigned int val;
505
506         val = __unpack_fe01(tsk->thread.fpexc_mode);
507         return put_user(val, (unsigned int __user *) adr);
508 }
509
510 int sys_clone(unsigned long clone_flags, unsigned long p2, unsigned long p3,
511               unsigned long p4, unsigned long p5, unsigned long p6,
512               struct pt_regs *regs)
513 {
514         unsigned long parent_tidptr = 0;
515         unsigned long child_tidptr = 0;
516
517         if (p2 == 0)
518                 p2 = regs->gpr[1];      /* stack pointer for child */
519
520         if (clone_flags & (CLONE_PARENT_SETTID | CLONE_CHILD_SETTID |
521                            CLONE_CHILD_CLEARTID)) {
522                 parent_tidptr = p3;
523                 child_tidptr = p5;
524                 if (test_thread_flag(TIF_32BIT)) {
525                         parent_tidptr &= 0xffffffff;
526                         child_tidptr &= 0xffffffff;
527                 }
528         }
529
530         return do_fork(clone_flags, p2, regs, 0,
531                     (int __user *)parent_tidptr, (int __user *)child_tidptr);
532 }
533
534 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
535              unsigned long p4, unsigned long p5, unsigned long p6,
536              struct pt_regs *regs)
537 {
538         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
539 }
540
541 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
542               unsigned long p4, unsigned long p5, unsigned long p6,
543               struct pt_regs *regs)
544 {
545         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1], regs, 0,
546                     NULL, NULL);
547 }
548
549 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
550                unsigned long a3, unsigned long a4, unsigned long a5,
551                struct pt_regs *regs)
552 {
553         int error;
554         char * filename;
555         
556         filename = getname((char __user *) a0);
557         error = PTR_ERR(filename);
558         if (IS_ERR(filename))
559                 goto out;
560         flush_fp_to_thread(current);
561         flush_altivec_to_thread(current);
562         error = do_execve(filename, (char __user * __user *) a1,
563                                     (char __user * __user *) a2, regs);
564   
565         if (error == 0) {
566                 task_lock(current);
567                 current->ptrace &= ~PT_DTRACE;
568                 task_unlock(current);
569         }
570         putname(filename);
571
572 out:
573         return error;
574 }
575
576 static int kstack_depth_to_print = 64;
577
578 static int validate_sp(unsigned long sp, struct task_struct *p,
579                        unsigned long nbytes)
580 {
581         unsigned long stack_page = (unsigned long)p->thread_info;
582
583         if (sp >= stack_page + sizeof(struct thread_struct)
584             && sp <= stack_page + THREAD_SIZE - nbytes)
585                 return 1;
586
587 #ifdef CONFIG_IRQSTACKS
588         stack_page = (unsigned long) hardirq_ctx[task_cpu(p)];
589         if (sp >= stack_page + sizeof(struct thread_struct)
590             && sp <= stack_page + THREAD_SIZE - nbytes)
591                 return 1;
592
593         stack_page = (unsigned long) softirq_ctx[task_cpu(p)];
594         if (sp >= stack_page + sizeof(struct thread_struct)
595             && sp <= stack_page + THREAD_SIZE - nbytes)
596                 return 1;
597 #endif
598
599         return 0;
600 }
601
602 unsigned long get_wchan(struct task_struct *p)
603 {
604         unsigned long ip, sp;
605         int count = 0;
606
607         if (!p || p == current || p->state == TASK_RUNNING)
608                 return 0;
609
610         sp = p->thread.ksp;
611         if (!validate_sp(sp, p, 112))
612                 return 0;
613
614         do {
615                 sp = *(unsigned long *)sp;
616                 if (!validate_sp(sp, p, 112))
617                         return 0;
618                 if (count > 0) {
619                         ip = *(unsigned long *)(sp + 16);
620                         if (!in_sched_functions(ip))
621                                 return ip;
622                 }
623         } while (count++ < 16);
624         return 0;
625 }
626 EXPORT_SYMBOL(get_wchan);
627
628 void show_stack(struct task_struct *p, unsigned long *_sp)
629 {
630         unsigned long ip, newsp, lr;
631         int count = 0;
632         unsigned long sp = (unsigned long)_sp;
633         int firstframe = 1;
634
635         if (sp == 0) {
636                 if (p) {
637                         sp = p->thread.ksp;
638                 } else {
639                         sp = __get_SP();
640                         p = current;
641                 }
642         }
643
644         lr = 0;
645         printk("Call Trace:\n");
646         do {
647                 if (!validate_sp(sp, p, 112))
648                         return;
649
650                 _sp = (unsigned long *) sp;
651                 newsp = _sp[0];
652                 ip = _sp[2];
653                 if (!firstframe || ip != lr) {
654                         printk("[%016lx] [%016lx] ", sp, ip);
655                         print_symbol("%s", ip);
656                         if (firstframe)
657                                 printk(" (unreliable)");
658                         printk("\n");
659                 }
660                 firstframe = 0;
661
662                 /*
663                  * See if this is an exception frame.
664                  * We look for the "regshere" marker in the current frame.
665                  */
666                 if (validate_sp(sp, p, sizeof(struct pt_regs) + 400)
667                     && _sp[12] == 0x7265677368657265ul) {
668                         struct pt_regs *regs = (struct pt_regs *)
669                                 (sp + STACK_FRAME_OVERHEAD);
670                         printk("--- Exception: %lx", regs->trap);
671                         print_symbol(" at %s\n", regs->nip);
672                         lr = regs->link;
673                         print_symbol("    LR = %s\n", lr);
674                         firstframe = 1;
675                 }
676
677                 sp = newsp;
678         } while (count++ < kstack_depth_to_print);
679 }
680
681 void dump_stack(void)
682 {
683         show_stack(current, (unsigned long *)__get_SP());
684 }
685 EXPORT_SYMBOL(dump_stack);