Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
[linux-2.6.git] / arch / powerpc / kernel / process.c
1 /*
2  *  Derived from "arch/i386/kernel/process.c"
3  *    Copyright (C) 1995  Linus Torvalds
4  *
5  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
6  *  Paul Mackerras (paulus@cs.anu.edu.au)
7  *
8  *  PowerPC version
9  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
10  *
11  *  This program is free software; you can redistribute it and/or
12  *  modify it under the terms of the GNU General Public License
13  *  as published by the Free Software Foundation; either version
14  *  2 of the License, or (at your option) any later version.
15  */
16
17 #include <linux/errno.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/unistd.h>
24 #include <linux/ptrace.h>
25 #include <linux/slab.h>
26 #include <linux/user.h>
27 #include <linux/elf.h>
28 #include <linux/init.h>
29 #include <linux/prctl.h>
30 #include <linux/init_task.h>
31 #include <linux/module.h>
32 #include <linux/kallsyms.h>
33 #include <linux/mqueue.h>
34 #include <linux/hardirq.h>
35 #include <linux/utsname.h>
36 #include <linux/ftrace.h>
37 #include <linux/kernel_stat.h>
38 #include <linux/personality.h>
39 #include <linux/random.h>
40 #include <linux/hw_breakpoint.h>
41
42 #include <asm/pgtable.h>
43 #include <asm/uaccess.h>
44 #include <asm/system.h>
45 #include <asm/io.h>
46 #include <asm/processor.h>
47 #include <asm/mmu.h>
48 #include <asm/prom.h>
49 #include <asm/machdep.h>
50 #include <asm/time.h>
51 #include <asm/syscalls.h>
52 #ifdef CONFIG_PPC64
53 #include <asm/firmware.h>
54 #endif
55 #include <linux/kprobes.h>
56 #include <linux/kdebug.h>
57
58 extern unsigned long _get_SP(void);
59
60 #ifndef CONFIG_SMP
61 struct task_struct *last_task_used_math = NULL;
62 struct task_struct *last_task_used_altivec = NULL;
63 struct task_struct *last_task_used_vsx = NULL;
64 struct task_struct *last_task_used_spe = NULL;
65 #endif
66
67 /*
68  * Make sure the floating-point register state in the
69  * the thread_struct is up to date for task tsk.
70  */
71 void flush_fp_to_thread(struct task_struct *tsk)
72 {
73         if (tsk->thread.regs) {
74                 /*
75                  * We need to disable preemption here because if we didn't,
76                  * another process could get scheduled after the regs->msr
77                  * test but before we have finished saving the FP registers
78                  * to the thread_struct.  That process could take over the
79                  * FPU, and then when we get scheduled again we would store
80                  * bogus values for the remaining FP registers.
81                  */
82                 preempt_disable();
83                 if (tsk->thread.regs->msr & MSR_FP) {
84 #ifdef CONFIG_SMP
85                         /*
86                          * This should only ever be called for current or
87                          * for a stopped child process.  Since we save away
88                          * the FP register state on context switch on SMP,
89                          * there is something wrong if a stopped child appears
90                          * to still have its FP state in the CPU registers.
91                          */
92                         BUG_ON(tsk != current);
93 #endif
94                         giveup_fpu(tsk);
95                 }
96                 preempt_enable();
97         }
98 }
99
100 void enable_kernel_fp(void)
101 {
102         WARN_ON(preemptible());
103
104 #ifdef CONFIG_SMP
105         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
106                 giveup_fpu(current);
107         else
108                 giveup_fpu(NULL);       /* just enables FP for kernel */
109 #else
110         giveup_fpu(last_task_used_math);
111 #endif /* CONFIG_SMP */
112 }
113 EXPORT_SYMBOL(enable_kernel_fp);
114
115 #ifdef CONFIG_ALTIVEC
116 void enable_kernel_altivec(void)
117 {
118         WARN_ON(preemptible());
119
120 #ifdef CONFIG_SMP
121         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
122                 giveup_altivec(current);
123         else
124                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
125 #else
126         giveup_altivec(last_task_used_altivec);
127 #endif /* CONFIG_SMP */
128 }
129 EXPORT_SYMBOL(enable_kernel_altivec);
130
131 /*
132  * Make sure the VMX/Altivec register state in the
133  * the thread_struct is up to date for task tsk.
134  */
135 void flush_altivec_to_thread(struct task_struct *tsk)
136 {
137         if (tsk->thread.regs) {
138                 preempt_disable();
139                 if (tsk->thread.regs->msr & MSR_VEC) {
140 #ifdef CONFIG_SMP
141                         BUG_ON(tsk != current);
142 #endif
143                         giveup_altivec(tsk);
144                 }
145                 preempt_enable();
146         }
147 }
148 #endif /* CONFIG_ALTIVEC */
149
150 #ifdef CONFIG_VSX
151 #if 0
152 /* not currently used, but some crazy RAID module might want to later */
153 void enable_kernel_vsx(void)
154 {
155         WARN_ON(preemptible());
156
157 #ifdef CONFIG_SMP
158         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
159                 giveup_vsx(current);
160         else
161                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
162 #else
163         giveup_vsx(last_task_used_vsx);
164 #endif /* CONFIG_SMP */
165 }
166 EXPORT_SYMBOL(enable_kernel_vsx);
167 #endif
168
169 void giveup_vsx(struct task_struct *tsk)
170 {
171         giveup_fpu(tsk);
172         giveup_altivec(tsk);
173         __giveup_vsx(tsk);
174 }
175
176 void flush_vsx_to_thread(struct task_struct *tsk)
177 {
178         if (tsk->thread.regs) {
179                 preempt_disable();
180                 if (tsk->thread.regs->msr & MSR_VSX) {
181 #ifdef CONFIG_SMP
182                         BUG_ON(tsk != current);
183 #endif
184                         giveup_vsx(tsk);
185                 }
186                 preempt_enable();
187         }
188 }
189 #endif /* CONFIG_VSX */
190
191 #ifdef CONFIG_SPE
192
193 void enable_kernel_spe(void)
194 {
195         WARN_ON(preemptible());
196
197 #ifdef CONFIG_SMP
198         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
199                 giveup_spe(current);
200         else
201                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
202 #else
203         giveup_spe(last_task_used_spe);
204 #endif /* __SMP __ */
205 }
206 EXPORT_SYMBOL(enable_kernel_spe);
207
208 void flush_spe_to_thread(struct task_struct *tsk)
209 {
210         if (tsk->thread.regs) {
211                 preempt_disable();
212                 if (tsk->thread.regs->msr & MSR_SPE) {
213 #ifdef CONFIG_SMP
214                         BUG_ON(tsk != current);
215 #endif
216                         giveup_spe(tsk);
217                 }
218                 preempt_enable();
219         }
220 }
221 #endif /* CONFIG_SPE */
222
223 #ifndef CONFIG_SMP
224 /*
225  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
226  * and the current task has some state, discard it.
227  */
228 void discard_lazy_cpu_state(void)
229 {
230         preempt_disable();
231         if (last_task_used_math == current)
232                 last_task_used_math = NULL;
233 #ifdef CONFIG_ALTIVEC
234         if (last_task_used_altivec == current)
235                 last_task_used_altivec = NULL;
236 #endif /* CONFIG_ALTIVEC */
237 #ifdef CONFIG_VSX
238         if (last_task_used_vsx == current)
239                 last_task_used_vsx = NULL;
240 #endif /* CONFIG_VSX */
241 #ifdef CONFIG_SPE
242         if (last_task_used_spe == current)
243                 last_task_used_spe = NULL;
244 #endif
245         preempt_enable();
246 }
247 #endif /* CONFIG_SMP */
248
249 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
250 void do_send_trap(struct pt_regs *regs, unsigned long address,
251                   unsigned long error_code, int signal_code, int breakpt)
252 {
253         siginfo_t info;
254
255         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
256                         11, SIGSEGV) == NOTIFY_STOP)
257                 return;
258
259         /* Deliver the signal to userspace */
260         info.si_signo = SIGTRAP;
261         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
262         info.si_code = signal_code;
263         info.si_addr = (void __user *)address;
264         force_sig_info(SIGTRAP, &info, current);
265 }
266 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
267 void do_dabr(struct pt_regs *regs, unsigned long address,
268                     unsigned long error_code)
269 {
270         siginfo_t info;
271
272         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
273                         11, SIGSEGV) == NOTIFY_STOP)
274                 return;
275
276         if (debugger_dabr_match(regs))
277                 return;
278
279         /* Clear the DABR */
280         set_dabr(0);
281
282         /* Deliver the signal to userspace */
283         info.si_signo = SIGTRAP;
284         info.si_errno = 0;
285         info.si_code = TRAP_HWBKPT;
286         info.si_addr = (void __user *)address;
287         force_sig_info(SIGTRAP, &info, current);
288 }
289 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
290
291 static DEFINE_PER_CPU(unsigned long, current_dabr);
292
293 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
294 /*
295  * Set the debug registers back to their default "safe" values.
296  */
297 static void set_debug_reg_defaults(struct thread_struct *thread)
298 {
299         thread->iac1 = thread->iac2 = 0;
300 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
301         thread->iac3 = thread->iac4 = 0;
302 #endif
303         thread->dac1 = thread->dac2 = 0;
304 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
305         thread->dvc1 = thread->dvc2 = 0;
306 #endif
307         thread->dbcr0 = 0;
308 #ifdef CONFIG_BOOKE
309         /*
310          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
311          */
312         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
313                         DBCR1_IAC3US | DBCR1_IAC4US;
314         /*
315          * Force Data Address Compare User/Supervisor bits to be User-only
316          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
317          */
318         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
319 #else
320         thread->dbcr1 = 0;
321 #endif
322 }
323
324 static void prime_debug_regs(struct thread_struct *thread)
325 {
326         mtspr(SPRN_IAC1, thread->iac1);
327         mtspr(SPRN_IAC2, thread->iac2);
328 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
329         mtspr(SPRN_IAC3, thread->iac3);
330         mtspr(SPRN_IAC4, thread->iac4);
331 #endif
332         mtspr(SPRN_DAC1, thread->dac1);
333         mtspr(SPRN_DAC2, thread->dac2);
334 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
335         mtspr(SPRN_DVC1, thread->dvc1);
336         mtspr(SPRN_DVC2, thread->dvc2);
337 #endif
338         mtspr(SPRN_DBCR0, thread->dbcr0);
339         mtspr(SPRN_DBCR1, thread->dbcr1);
340 #ifdef CONFIG_BOOKE
341         mtspr(SPRN_DBCR2, thread->dbcr2);
342 #endif
343 }
344 /*
345  * Unless neither the old or new thread are making use of the
346  * debug registers, set the debug registers from the values
347  * stored in the new thread.
348  */
349 static void switch_booke_debug_regs(struct thread_struct *new_thread)
350 {
351         if ((current->thread.dbcr0 & DBCR0_IDM)
352                 || (new_thread->dbcr0 & DBCR0_IDM))
353                         prime_debug_regs(new_thread);
354 }
355 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
356 static void set_debug_reg_defaults(struct thread_struct *thread)
357 {
358         if (thread->dabr) {
359                 thread->dabr = 0;
360                 set_dabr(0);
361         }
362 }
363 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
364
365 int set_dabr(unsigned long dabr)
366 {
367         __get_cpu_var(current_dabr) = dabr;
368
369         if (ppc_md.set_dabr)
370                 return ppc_md.set_dabr(dabr);
371
372         /* XXX should we have a CPU_FTR_HAS_DABR ? */
373 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
374         mtspr(SPRN_DAC1, dabr);
375 #ifdef CONFIG_PPC_47x
376         isync();
377 #endif
378 #elif defined(CONFIG_PPC_BOOK3S)
379         mtspr(SPRN_DABR, dabr);
380 #endif
381
382
383         return 0;
384 }
385
386 #ifdef CONFIG_PPC64
387 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
388 #endif
389
390 struct task_struct *__switch_to(struct task_struct *prev,
391         struct task_struct *new)
392 {
393         struct thread_struct *new_thread, *old_thread;
394         unsigned long flags;
395         struct task_struct *last;
396
397 #ifdef CONFIG_SMP
398         /* avoid complexity of lazy save/restore of fpu
399          * by just saving it every time we switch out if
400          * this task used the fpu during the last quantum.
401          *
402          * If it tries to use the fpu again, it'll trap and
403          * reload its fp regs.  So we don't have to do a restore
404          * every switch, just a save.
405          *  -- Cort
406          */
407         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
408                 giveup_fpu(prev);
409 #ifdef CONFIG_ALTIVEC
410         /*
411          * If the previous thread used altivec in the last quantum
412          * (thus changing altivec regs) then save them.
413          * We used to check the VRSAVE register but not all apps
414          * set it, so we don't rely on it now (and in fact we need
415          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
416          *
417          * On SMP we always save/restore altivec regs just to avoid the
418          * complexity of changing processors.
419          *  -- Cort
420          */
421         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
422                 giveup_altivec(prev);
423 #endif /* CONFIG_ALTIVEC */
424 #ifdef CONFIG_VSX
425         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
426                 /* VMX and FPU registers are already save here */
427                 __giveup_vsx(prev);
428 #endif /* CONFIG_VSX */
429 #ifdef CONFIG_SPE
430         /*
431          * If the previous thread used spe in the last quantum
432          * (thus changing spe regs) then save them.
433          *
434          * On SMP we always save/restore spe regs just to avoid the
435          * complexity of changing processors.
436          */
437         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
438                 giveup_spe(prev);
439 #endif /* CONFIG_SPE */
440
441 #else  /* CONFIG_SMP */
442 #ifdef CONFIG_ALTIVEC
443         /* Avoid the trap.  On smp this this never happens since
444          * we don't set last_task_used_altivec -- Cort
445          */
446         if (new->thread.regs && last_task_used_altivec == new)
447                 new->thread.regs->msr |= MSR_VEC;
448 #endif /* CONFIG_ALTIVEC */
449 #ifdef CONFIG_VSX
450         if (new->thread.regs && last_task_used_vsx == new)
451                 new->thread.regs->msr |= MSR_VSX;
452 #endif /* CONFIG_VSX */
453 #ifdef CONFIG_SPE
454         /* Avoid the trap.  On smp this this never happens since
455          * we don't set last_task_used_spe
456          */
457         if (new->thread.regs && last_task_used_spe == new)
458                 new->thread.regs->msr |= MSR_SPE;
459 #endif /* CONFIG_SPE */
460
461 #endif /* CONFIG_SMP */
462
463 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
464         switch_booke_debug_regs(&new->thread);
465 #else
466 /*
467  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
468  * schedule DABR
469  */
470 #ifndef CONFIG_HAVE_HW_BREAKPOINT
471         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
472                 set_dabr(new->thread.dabr);
473 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
474 #endif
475
476
477         new_thread = &new->thread;
478         old_thread = &current->thread;
479
480 #if defined(CONFIG_PPC_BOOK3E_64)
481         /* XXX Current Book3E code doesn't deal with kernel side DBCR0,
482          * we always hold the user values, so we set it now.
483          *
484          * However, we ensure the kernel MSR:DE is appropriately cleared too
485          * to avoid spurrious single step exceptions in the kernel.
486          *
487          * This will have to change to merge with the ppc32 code at some point,
488          * but I don't like much what ppc32 is doing today so there's some
489          * thinking needed there
490          */
491         if ((new_thread->dbcr0 | old_thread->dbcr0) & DBCR0_IDM) {
492                 u32 dbcr0;
493
494                 mtmsr(mfmsr() & ~MSR_DE);
495                 isync();
496                 dbcr0 = mfspr(SPRN_DBCR0);
497                 dbcr0 = (dbcr0 & DBCR0_EDM) | new_thread->dbcr0;
498                 mtspr(SPRN_DBCR0, dbcr0);
499         }
500 #endif /* CONFIG_PPC64_BOOK3E */
501
502 #ifdef CONFIG_PPC64
503         /*
504          * Collect processor utilization data per process
505          */
506         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
507                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
508                 long unsigned start_tb, current_tb;
509                 start_tb = old_thread->start_tb;
510                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
511                 old_thread->accum_tb += (current_tb - start_tb);
512                 new_thread->start_tb = current_tb;
513         }
514 #endif
515
516         local_irq_save(flags);
517
518         account_system_vtime(current);
519         account_process_vtime(current);
520
521         /*
522          * We can't take a PMU exception inside _switch() since there is a
523          * window where the kernel stack SLB and the kernel stack are out
524          * of sync. Hard disable here.
525          */
526         hard_irq_disable();
527         last = _switch(old_thread, new_thread);
528
529         local_irq_restore(flags);
530
531         return last;
532 }
533
534 static int instructions_to_print = 16;
535
536 static void show_instructions(struct pt_regs *regs)
537 {
538         int i;
539         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
540                         sizeof(int));
541
542         printk("Instruction dump:");
543
544         for (i = 0; i < instructions_to_print; i++) {
545                 int instr;
546
547                 if (!(i % 8))
548                         printk("\n");
549
550 #if !defined(CONFIG_BOOKE)
551                 /* If executing with the IMMU off, adjust pc rather
552                  * than print XXXXXXXX.
553                  */
554                 if (!(regs->msr & MSR_IR))
555                         pc = (unsigned long)phys_to_virt(pc);
556 #endif
557
558                 /* We use __get_user here *only* to avoid an OOPS on a
559                  * bad address because the pc *should* only be a
560                  * kernel address.
561                  */
562                 if (!__kernel_text_address(pc) ||
563                      __get_user(instr, (unsigned int __user *)pc)) {
564                         printk("XXXXXXXX ");
565                 } else {
566                         if (regs->nip == pc)
567                                 printk("<%08x> ", instr);
568                         else
569                                 printk("%08x ", instr);
570                 }
571
572                 pc += sizeof(int);
573         }
574
575         printk("\n");
576 }
577
578 static struct regbit {
579         unsigned long bit;
580         const char *name;
581 } msr_bits[] = {
582         {MSR_EE,        "EE"},
583         {MSR_PR,        "PR"},
584         {MSR_FP,        "FP"},
585         {MSR_VEC,       "VEC"},
586         {MSR_VSX,       "VSX"},
587         {MSR_ME,        "ME"},
588         {MSR_CE,        "CE"},
589         {MSR_DE,        "DE"},
590         {MSR_IR,        "IR"},
591         {MSR_DR,        "DR"},
592         {0,             NULL}
593 };
594
595 static void printbits(unsigned long val, struct regbit *bits)
596 {
597         const char *sep = "";
598
599         printk("<");
600         for (; bits->bit; ++bits)
601                 if (val & bits->bit) {
602                         printk("%s%s", sep, bits->name);
603                         sep = ",";
604                 }
605         printk(">");
606 }
607
608 #ifdef CONFIG_PPC64
609 #define REG             "%016lx"
610 #define REGS_PER_LINE   4
611 #define LAST_VOLATILE   13
612 #else
613 #define REG             "%08lx"
614 #define REGS_PER_LINE   8
615 #define LAST_VOLATILE   12
616 #endif
617
618 void show_regs(struct pt_regs * regs)
619 {
620         int i, trap;
621
622         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
623                regs->nip, regs->link, regs->ctr);
624         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
625                regs, regs->trap, print_tainted(), init_utsname()->release);
626         printk("MSR: "REG" ", regs->msr);
627         printbits(regs->msr, msr_bits);
628         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
629         trap = TRAP(regs);
630         if (trap == 0x300 || trap == 0x600)
631 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
632                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
633 #else
634                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
635 #endif
636         printk("TASK = %p[%d] '%s' THREAD: %p",
637                current, task_pid_nr(current), current->comm, task_thread_info(current));
638
639 #ifdef CONFIG_SMP
640         printk(" CPU: %d", raw_smp_processor_id());
641 #endif /* CONFIG_SMP */
642
643         for (i = 0;  i < 32;  i++) {
644                 if ((i % REGS_PER_LINE) == 0)
645                         printk("\nGPR%02d: ", i);
646                 printk(REG " ", regs->gpr[i]);
647                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
648                         break;
649         }
650         printk("\n");
651 #ifdef CONFIG_KALLSYMS
652         /*
653          * Lookup NIP late so we have the best change of getting the
654          * above info out without failing
655          */
656         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
657         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
658 #endif
659         show_stack(current, (unsigned long *) regs->gpr[1]);
660         if (!user_mode(regs))
661                 show_instructions(regs);
662 }
663
664 void exit_thread(void)
665 {
666         discard_lazy_cpu_state();
667 }
668
669 void flush_thread(void)
670 {
671         discard_lazy_cpu_state();
672
673 #ifdef CONFIG_HAVE_HW_BREAKPOINTS
674         flush_ptrace_hw_breakpoint(current);
675 #else /* CONFIG_HAVE_HW_BREAKPOINTS */
676         set_debug_reg_defaults(&current->thread);
677 #endif /* CONFIG_HAVE_HW_BREAKPOINTS */
678 }
679
680 void
681 release_thread(struct task_struct *t)
682 {
683 }
684
685 /*
686  * This gets called before we allocate a new thread and copy
687  * the current task into it.
688  */
689 void prepare_to_copy(struct task_struct *tsk)
690 {
691         flush_fp_to_thread(current);
692         flush_altivec_to_thread(current);
693         flush_vsx_to_thread(current);
694         flush_spe_to_thread(current);
695 #ifdef CONFIG_HAVE_HW_BREAKPOINT
696         flush_ptrace_hw_breakpoint(tsk);
697 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
698 }
699
700 /*
701  * Copy a thread..
702  */
703 int copy_thread(unsigned long clone_flags, unsigned long usp,
704                 unsigned long unused, struct task_struct *p,
705                 struct pt_regs *regs)
706 {
707         struct pt_regs *childregs, *kregs;
708         extern void ret_from_fork(void);
709         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
710
711         CHECK_FULL_REGS(regs);
712         /* Copy registers */
713         sp -= sizeof(struct pt_regs);
714         childregs = (struct pt_regs *) sp;
715         *childregs = *regs;
716         if ((childregs->msr & MSR_PR) == 0) {
717                 /* for kernel thread, set `current' and stackptr in new task */
718                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
719 #ifdef CONFIG_PPC32
720                 childregs->gpr[2] = (unsigned long) p;
721 #else
722                 clear_tsk_thread_flag(p, TIF_32BIT);
723 #endif
724                 p->thread.regs = NULL;  /* no user register state */
725         } else {
726                 childregs->gpr[1] = usp;
727                 p->thread.regs = childregs;
728                 if (clone_flags & CLONE_SETTLS) {
729 #ifdef CONFIG_PPC64
730                         if (!is_32bit_task())
731                                 childregs->gpr[13] = childregs->gpr[6];
732                         else
733 #endif
734                                 childregs->gpr[2] = childregs->gpr[6];
735                 }
736         }
737         childregs->gpr[3] = 0;  /* Result from fork() */
738         sp -= STACK_FRAME_OVERHEAD;
739
740         /*
741          * The way this works is that at some point in the future
742          * some task will call _switch to switch to the new task.
743          * That will pop off the stack frame created below and start
744          * the new task running at ret_from_fork.  The new task will
745          * do some house keeping and then return from the fork or clone
746          * system call, using the stack frame created above.
747          */
748         sp -= sizeof(struct pt_regs);
749         kregs = (struct pt_regs *) sp;
750         sp -= STACK_FRAME_OVERHEAD;
751         p->thread.ksp = sp;
752         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
753                                 _ALIGN_UP(sizeof(struct thread_info), 16);
754
755 #ifdef CONFIG_PPC_STD_MMU_64
756         if (cpu_has_feature(CPU_FTR_SLB)) {
757                 unsigned long sp_vsid;
758                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
759
760                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
761                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
762                                 << SLB_VSID_SHIFT_1T;
763                 else
764                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
765                                 << SLB_VSID_SHIFT;
766                 sp_vsid |= SLB_VSID_KERNEL | llp;
767                 p->thread.ksp_vsid = sp_vsid;
768         }
769 #endif /* CONFIG_PPC_STD_MMU_64 */
770
771         /*
772          * The PPC64 ABI makes use of a TOC to contain function 
773          * pointers.  The function (ret_from_except) is actually a pointer
774          * to the TOC entry.  The first entry is a pointer to the actual
775          * function.
776          */
777 #ifdef CONFIG_PPC64
778         kregs->nip = *((unsigned long *)ret_from_fork);
779 #else
780         kregs->nip = (unsigned long)ret_from_fork;
781 #endif
782
783         return 0;
784 }
785
786 /*
787  * Set up a thread for executing a new program
788  */
789 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
790 {
791 #ifdef CONFIG_PPC64
792         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
793 #endif
794
795         set_fs(USER_DS);
796
797         /*
798          * If we exec out of a kernel thread then thread.regs will not be
799          * set.  Do it now.
800          */
801         if (!current->thread.regs) {
802                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
803                 current->thread.regs = regs - 1;
804         }
805
806         memset(regs->gpr, 0, sizeof(regs->gpr));
807         regs->ctr = 0;
808         regs->link = 0;
809         regs->xer = 0;
810         regs->ccr = 0;
811         regs->gpr[1] = sp;
812
813         /*
814          * We have just cleared all the nonvolatile GPRs, so make
815          * FULL_REGS(regs) return true.  This is necessary to allow
816          * ptrace to examine the thread immediately after exec.
817          */
818         regs->trap &= ~1UL;
819
820 #ifdef CONFIG_PPC32
821         regs->mq = 0;
822         regs->nip = start;
823         regs->msr = MSR_USER;
824 #else
825         if (!is_32bit_task()) {
826                 unsigned long entry, toc;
827
828                 /* start is a relocated pointer to the function descriptor for
829                  * the elf _start routine.  The first entry in the function
830                  * descriptor is the entry address of _start and the second
831                  * entry is the TOC value we need to use.
832                  */
833                 __get_user(entry, (unsigned long __user *)start);
834                 __get_user(toc, (unsigned long __user *)start+1);
835
836                 /* Check whether the e_entry function descriptor entries
837                  * need to be relocated before we can use them.
838                  */
839                 if (load_addr != 0) {
840                         entry += load_addr;
841                         toc   += load_addr;
842                 }
843                 regs->nip = entry;
844                 regs->gpr[2] = toc;
845                 regs->msr = MSR_USER64;
846         } else {
847                 regs->nip = start;
848                 regs->gpr[2] = 0;
849                 regs->msr = MSR_USER32;
850         }
851 #endif
852
853         discard_lazy_cpu_state();
854 #ifdef CONFIG_VSX
855         current->thread.used_vsr = 0;
856 #endif
857         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
858         current->thread.fpscr.val = 0;
859 #ifdef CONFIG_ALTIVEC
860         memset(current->thread.vr, 0, sizeof(current->thread.vr));
861         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
862         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
863         current->thread.vrsave = 0;
864         current->thread.used_vr = 0;
865 #endif /* CONFIG_ALTIVEC */
866 #ifdef CONFIG_SPE
867         memset(current->thread.evr, 0, sizeof(current->thread.evr));
868         current->thread.acc = 0;
869         current->thread.spefscr = 0;
870         current->thread.used_spe = 0;
871 #endif /* CONFIG_SPE */
872 }
873
874 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
875                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
876
877 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
878 {
879         struct pt_regs *regs = tsk->thread.regs;
880
881         /* This is a bit hairy.  If we are an SPE enabled  processor
882          * (have embedded fp) we store the IEEE exception enable flags in
883          * fpexc_mode.  fpexc_mode is also used for setting FP exception
884          * mode (asyn, precise, disabled) for 'Classic' FP. */
885         if (val & PR_FP_EXC_SW_ENABLE) {
886 #ifdef CONFIG_SPE
887                 if (cpu_has_feature(CPU_FTR_SPE)) {
888                         tsk->thread.fpexc_mode = val &
889                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
890                         return 0;
891                 } else {
892                         return -EINVAL;
893                 }
894 #else
895                 return -EINVAL;
896 #endif
897         }
898
899         /* on a CONFIG_SPE this does not hurt us.  The bits that
900          * __pack_fe01 use do not overlap with bits used for
901          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
902          * on CONFIG_SPE implementations are reserved so writing to
903          * them does not change anything */
904         if (val > PR_FP_EXC_PRECISE)
905                 return -EINVAL;
906         tsk->thread.fpexc_mode = __pack_fe01(val);
907         if (regs != NULL && (regs->msr & MSR_FP) != 0)
908                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
909                         | tsk->thread.fpexc_mode;
910         return 0;
911 }
912
913 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
914 {
915         unsigned int val;
916
917         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
918 #ifdef CONFIG_SPE
919                 if (cpu_has_feature(CPU_FTR_SPE))
920                         val = tsk->thread.fpexc_mode;
921                 else
922                         return -EINVAL;
923 #else
924                 return -EINVAL;
925 #endif
926         else
927                 val = __unpack_fe01(tsk->thread.fpexc_mode);
928         return put_user(val, (unsigned int __user *) adr);
929 }
930
931 int set_endian(struct task_struct *tsk, unsigned int val)
932 {
933         struct pt_regs *regs = tsk->thread.regs;
934
935         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
936             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
937                 return -EINVAL;
938
939         if (regs == NULL)
940                 return -EINVAL;
941
942         if (val == PR_ENDIAN_BIG)
943                 regs->msr &= ~MSR_LE;
944         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
945                 regs->msr |= MSR_LE;
946         else
947                 return -EINVAL;
948
949         return 0;
950 }
951
952 int get_endian(struct task_struct *tsk, unsigned long adr)
953 {
954         struct pt_regs *regs = tsk->thread.regs;
955         unsigned int val;
956
957         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
958             !cpu_has_feature(CPU_FTR_REAL_LE))
959                 return -EINVAL;
960
961         if (regs == NULL)
962                 return -EINVAL;
963
964         if (regs->msr & MSR_LE) {
965                 if (cpu_has_feature(CPU_FTR_REAL_LE))
966                         val = PR_ENDIAN_LITTLE;
967                 else
968                         val = PR_ENDIAN_PPC_LITTLE;
969         } else
970                 val = PR_ENDIAN_BIG;
971
972         return put_user(val, (unsigned int __user *)adr);
973 }
974
975 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
976 {
977         tsk->thread.align_ctl = val;
978         return 0;
979 }
980
981 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
982 {
983         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
984 }
985
986 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
987
988 int sys_clone(unsigned long clone_flags, unsigned long usp,
989               int __user *parent_tidp, void __user *child_threadptr,
990               int __user *child_tidp, int p6,
991               struct pt_regs *regs)
992 {
993         CHECK_FULL_REGS(regs);
994         if (usp == 0)
995                 usp = regs->gpr[1];     /* stack pointer for child */
996 #ifdef CONFIG_PPC64
997         if (is_32bit_task()) {
998                 parent_tidp = TRUNC_PTR(parent_tidp);
999                 child_tidp = TRUNC_PTR(child_tidp);
1000         }
1001 #endif
1002         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
1003 }
1004
1005 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
1006              unsigned long p4, unsigned long p5, unsigned long p6,
1007              struct pt_regs *regs)
1008 {
1009         CHECK_FULL_REGS(regs);
1010         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
1011 }
1012
1013 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
1014               unsigned long p4, unsigned long p5, unsigned long p6,
1015               struct pt_regs *regs)
1016 {
1017         CHECK_FULL_REGS(regs);
1018         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
1019                         regs, 0, NULL, NULL);
1020 }
1021
1022 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
1023                unsigned long a3, unsigned long a4, unsigned long a5,
1024                struct pt_regs *regs)
1025 {
1026         int error;
1027         char *filename;
1028
1029         filename = getname((const char __user *) a0);
1030         error = PTR_ERR(filename);
1031         if (IS_ERR(filename))
1032                 goto out;
1033         flush_fp_to_thread(current);
1034         flush_altivec_to_thread(current);
1035         flush_spe_to_thread(current);
1036         error = do_execve(filename,
1037                           (const char __user *const __user *) a1,
1038                           (const char __user *const __user *) a2, regs);
1039         putname(filename);
1040 out:
1041         return error;
1042 }
1043
1044 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1045                                   unsigned long nbytes)
1046 {
1047         unsigned long stack_page;
1048         unsigned long cpu = task_cpu(p);
1049
1050         /*
1051          * Avoid crashing if the stack has overflowed and corrupted
1052          * task_cpu(p), which is in the thread_info struct.
1053          */
1054         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1055                 stack_page = (unsigned long) hardirq_ctx[cpu];
1056                 if (sp >= stack_page + sizeof(struct thread_struct)
1057                     && sp <= stack_page + THREAD_SIZE - nbytes)
1058                         return 1;
1059
1060                 stack_page = (unsigned long) softirq_ctx[cpu];
1061                 if (sp >= stack_page + sizeof(struct thread_struct)
1062                     && sp <= stack_page + THREAD_SIZE - nbytes)
1063                         return 1;
1064         }
1065         return 0;
1066 }
1067
1068 int validate_sp(unsigned long sp, struct task_struct *p,
1069                        unsigned long nbytes)
1070 {
1071         unsigned long stack_page = (unsigned long)task_stack_page(p);
1072
1073         if (sp >= stack_page + sizeof(struct thread_struct)
1074             && sp <= stack_page + THREAD_SIZE - nbytes)
1075                 return 1;
1076
1077         return valid_irq_stack(sp, p, nbytes);
1078 }
1079
1080 EXPORT_SYMBOL(validate_sp);
1081
1082 unsigned long get_wchan(struct task_struct *p)
1083 {
1084         unsigned long ip, sp;
1085         int count = 0;
1086
1087         if (!p || p == current || p->state == TASK_RUNNING)
1088                 return 0;
1089
1090         sp = p->thread.ksp;
1091         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1092                 return 0;
1093
1094         do {
1095                 sp = *(unsigned long *)sp;
1096                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1097                         return 0;
1098                 if (count > 0) {
1099                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1100                         if (!in_sched_functions(ip))
1101                                 return ip;
1102                 }
1103         } while (count++ < 16);
1104         return 0;
1105 }
1106
1107 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1108
1109 void show_stack(struct task_struct *tsk, unsigned long *stack)
1110 {
1111         unsigned long sp, ip, lr, newsp;
1112         int count = 0;
1113         int firstframe = 1;
1114 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1115         int curr_frame = current->curr_ret_stack;
1116         extern void return_to_handler(void);
1117         unsigned long rth = (unsigned long)return_to_handler;
1118         unsigned long mrth = -1;
1119 #ifdef CONFIG_PPC64
1120         extern void mod_return_to_handler(void);
1121         rth = *(unsigned long *)rth;
1122         mrth = (unsigned long)mod_return_to_handler;
1123         mrth = *(unsigned long *)mrth;
1124 #endif
1125 #endif
1126
1127         sp = (unsigned long) stack;
1128         if (tsk == NULL)
1129                 tsk = current;
1130         if (sp == 0) {
1131                 if (tsk == current)
1132                         asm("mr %0,1" : "=r" (sp));
1133                 else
1134                         sp = tsk->thread.ksp;
1135         }
1136
1137         lr = 0;
1138         printk("Call Trace:\n");
1139         do {
1140                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1141                         return;
1142
1143                 stack = (unsigned long *) sp;
1144                 newsp = stack[0];
1145                 ip = stack[STACK_FRAME_LR_SAVE];
1146                 if (!firstframe || ip != lr) {
1147                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1148 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1149                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1150                                 printk(" (%pS)",
1151                                        (void *)current->ret_stack[curr_frame].ret);
1152                                 curr_frame--;
1153                         }
1154 #endif
1155                         if (firstframe)
1156                                 printk(" (unreliable)");
1157                         printk("\n");
1158                 }
1159                 firstframe = 0;
1160
1161                 /*
1162                  * See if this is an exception frame.
1163                  * We look for the "regshere" marker in the current frame.
1164                  */
1165                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1166                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1167                         struct pt_regs *regs = (struct pt_regs *)
1168                                 (sp + STACK_FRAME_OVERHEAD);
1169                         lr = regs->link;
1170                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1171                                regs->trap, (void *)regs->nip, (void *)lr);
1172                         firstframe = 1;
1173                 }
1174
1175                 sp = newsp;
1176         } while (count++ < kstack_depth_to_print);
1177 }
1178
1179 void dump_stack(void)
1180 {
1181         show_stack(current, NULL);
1182 }
1183 EXPORT_SYMBOL(dump_stack);
1184
1185 #ifdef CONFIG_PPC64
1186 void ppc64_runlatch_on(void)
1187 {
1188         unsigned long ctrl;
1189
1190         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1191                 HMT_medium();
1192
1193                 ctrl = mfspr(SPRN_CTRLF);
1194                 ctrl |= CTRL_RUNLATCH;
1195                 mtspr(SPRN_CTRLT, ctrl);
1196
1197                 set_thread_flag(TIF_RUNLATCH);
1198         }
1199 }
1200
1201 void __ppc64_runlatch_off(void)
1202 {
1203         unsigned long ctrl;
1204
1205         HMT_medium();
1206
1207         clear_thread_flag(TIF_RUNLATCH);
1208
1209         ctrl = mfspr(SPRN_CTRLF);
1210         ctrl &= ~CTRL_RUNLATCH;
1211         mtspr(SPRN_CTRLT, ctrl);
1212 }
1213 #endif
1214
1215 #if THREAD_SHIFT < PAGE_SHIFT
1216
1217 static struct kmem_cache *thread_info_cache;
1218
1219 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1220 {
1221         struct thread_info *ti;
1222
1223         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1224         if (unlikely(ti == NULL))
1225                 return NULL;
1226 #ifdef CONFIG_DEBUG_STACK_USAGE
1227         memset(ti, 0, THREAD_SIZE);
1228 #endif
1229         return ti;
1230 }
1231
1232 void free_thread_info(struct thread_info *ti)
1233 {
1234         kmem_cache_free(thread_info_cache, ti);
1235 }
1236
1237 void thread_info_cache_init(void)
1238 {
1239         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1240                                               THREAD_SIZE, 0, NULL);
1241         BUG_ON(thread_info_cache == NULL);
1242 }
1243
1244 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1245
1246 unsigned long arch_align_stack(unsigned long sp)
1247 {
1248         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1249                 sp -= get_random_int() & ~PAGE_MASK;
1250         return sp & ~0xf;
1251 }
1252
1253 static inline unsigned long brk_rnd(void)
1254 {
1255         unsigned long rnd = 0;
1256
1257         /* 8MB for 32bit, 1GB for 64bit */
1258         if (is_32bit_task())
1259                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1260         else
1261                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1262
1263         return rnd << PAGE_SHIFT;
1264 }
1265
1266 unsigned long arch_randomize_brk(struct mm_struct *mm)
1267 {
1268         unsigned long base = mm->brk;
1269         unsigned long ret;
1270
1271 #ifdef CONFIG_PPC_STD_MMU_64
1272         /*
1273          * If we are using 1TB segments and we are allowed to randomise
1274          * the heap, we can put it above 1TB so it is backed by a 1TB
1275          * segment. Otherwise the heap will be in the bottom 1TB
1276          * which always uses 256MB segments and this may result in a
1277          * performance penalty.
1278          */
1279         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1280                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1281 #endif
1282
1283         ret = PAGE_ALIGN(base + brk_rnd());
1284
1285         if (ret < mm->brk)
1286                 return mm->brk;
1287
1288         return ret;
1289 }
1290
1291 unsigned long randomize_et_dyn(unsigned long base)
1292 {
1293         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1294
1295         if (ret < base)
1296                 return base;
1297
1298         return ret;
1299 }