]> git.karo-electronics.de Git - mv-sheeva.git/blob - arch/powerpc/kernel/process.c
Merge commit 'paulus-perf/master' into next
[mv-sheeva.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 #ifdef CONFIG_PPC64
481         /*
482          * Collect processor utilization data per process
483          */
484         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
485                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
486                 long unsigned start_tb, current_tb;
487                 start_tb = old_thread->start_tb;
488                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
489                 old_thread->accum_tb += (current_tb - start_tb);
490                 new_thread->start_tb = current_tb;
491         }
492 #endif
493
494         local_irq_save(flags);
495
496         account_system_vtime(current);
497         account_process_vtime(current);
498         calculate_steal_time();
499
500         /*
501          * We can't take a PMU exception inside _switch() since there is a
502          * window where the kernel stack SLB and the kernel stack are out
503          * of sync. Hard disable here.
504          */
505         hard_irq_disable();
506         last = _switch(old_thread, new_thread);
507
508         local_irq_restore(flags);
509
510         return last;
511 }
512
513 static int instructions_to_print = 16;
514
515 static void show_instructions(struct pt_regs *regs)
516 {
517         int i;
518         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
519                         sizeof(int));
520
521         printk("Instruction dump:");
522
523         for (i = 0; i < instructions_to_print; i++) {
524                 int instr;
525
526                 if (!(i % 8))
527                         printk("\n");
528
529 #if !defined(CONFIG_BOOKE)
530                 /* If executing with the IMMU off, adjust pc rather
531                  * than print XXXXXXXX.
532                  */
533                 if (!(regs->msr & MSR_IR))
534                         pc = (unsigned long)phys_to_virt(pc);
535 #endif
536
537                 /* We use __get_user here *only* to avoid an OOPS on a
538                  * bad address because the pc *should* only be a
539                  * kernel address.
540                  */
541                 if (!__kernel_text_address(pc) ||
542                      __get_user(instr, (unsigned int __user *)pc)) {
543                         printk("XXXXXXXX ");
544                 } else {
545                         if (regs->nip == pc)
546                                 printk("<%08x> ", instr);
547                         else
548                                 printk("%08x ", instr);
549                 }
550
551                 pc += sizeof(int);
552         }
553
554         printk("\n");
555 }
556
557 static struct regbit {
558         unsigned long bit;
559         const char *name;
560 } msr_bits[] = {
561         {MSR_EE,        "EE"},
562         {MSR_PR,        "PR"},
563         {MSR_FP,        "FP"},
564         {MSR_VEC,       "VEC"},
565         {MSR_VSX,       "VSX"},
566         {MSR_ME,        "ME"},
567         {MSR_CE,        "CE"},
568         {MSR_DE,        "DE"},
569         {MSR_IR,        "IR"},
570         {MSR_DR,        "DR"},
571         {0,             NULL}
572 };
573
574 static void printbits(unsigned long val, struct regbit *bits)
575 {
576         const char *sep = "";
577
578         printk("<");
579         for (; bits->bit; ++bits)
580                 if (val & bits->bit) {
581                         printk("%s%s", sep, bits->name);
582                         sep = ",";
583                 }
584         printk(">");
585 }
586
587 #ifdef CONFIG_PPC64
588 #define REG             "%016lx"
589 #define REGS_PER_LINE   4
590 #define LAST_VOLATILE   13
591 #else
592 #define REG             "%08lx"
593 #define REGS_PER_LINE   8
594 #define LAST_VOLATILE   12
595 #endif
596
597 void show_regs(struct pt_regs * regs)
598 {
599         int i, trap;
600
601         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
602                regs->nip, regs->link, regs->ctr);
603         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
604                regs, regs->trap, print_tainted(), init_utsname()->release);
605         printk("MSR: "REG" ", regs->msr);
606         printbits(regs->msr, msr_bits);
607         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
608         trap = TRAP(regs);
609         if (trap == 0x300 || trap == 0x600)
610 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
611                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
612 #else
613                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
614 #endif
615         printk("TASK = %p[%d] '%s' THREAD: %p",
616                current, task_pid_nr(current), current->comm, task_thread_info(current));
617
618 #ifdef CONFIG_SMP
619         printk(" CPU: %d", raw_smp_processor_id());
620 #endif /* CONFIG_SMP */
621
622         for (i = 0;  i < 32;  i++) {
623                 if ((i % REGS_PER_LINE) == 0)
624                         printk("\nGPR%02d: ", i);
625                 printk(REG " ", regs->gpr[i]);
626                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
627                         break;
628         }
629         printk("\n");
630 #ifdef CONFIG_KALLSYMS
631         /*
632          * Lookup NIP late so we have the best change of getting the
633          * above info out without failing
634          */
635         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
636         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
637 #endif
638         show_stack(current, (unsigned long *) regs->gpr[1]);
639         if (!user_mode(regs))
640                 show_instructions(regs);
641 }
642
643 void exit_thread(void)
644 {
645         discard_lazy_cpu_state();
646 }
647
648 void flush_thread(void)
649 {
650         discard_lazy_cpu_state();
651
652 #ifdef CONFIG_HAVE_HW_BREAKPOINTS
653         flush_ptrace_hw_breakpoint(current);
654 #else /* CONFIG_HAVE_HW_BREAKPOINTS */
655         set_debug_reg_defaults(&current->thread);
656 #endif /* CONFIG_HAVE_HW_BREAKPOINTS */
657 }
658
659 void
660 release_thread(struct task_struct *t)
661 {
662 }
663
664 /*
665  * This gets called before we allocate a new thread and copy
666  * the current task into it.
667  */
668 void prepare_to_copy(struct task_struct *tsk)
669 {
670         flush_fp_to_thread(current);
671         flush_altivec_to_thread(current);
672         flush_vsx_to_thread(current);
673         flush_spe_to_thread(current);
674 #ifdef CONFIG_HAVE_HW_BREAKPOINT
675         flush_ptrace_hw_breakpoint(tsk);
676 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
677 }
678
679 /*
680  * Copy a thread..
681  */
682 int copy_thread(unsigned long clone_flags, unsigned long usp,
683                 unsigned long unused, struct task_struct *p,
684                 struct pt_regs *regs)
685 {
686         struct pt_regs *childregs, *kregs;
687         extern void ret_from_fork(void);
688         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
689
690         CHECK_FULL_REGS(regs);
691         /* Copy registers */
692         sp -= sizeof(struct pt_regs);
693         childregs = (struct pt_regs *) sp;
694         *childregs = *regs;
695         if ((childregs->msr & MSR_PR) == 0) {
696                 /* for kernel thread, set `current' and stackptr in new task */
697                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
698 #ifdef CONFIG_PPC32
699                 childregs->gpr[2] = (unsigned long) p;
700 #else
701                 clear_tsk_thread_flag(p, TIF_32BIT);
702 #endif
703                 p->thread.regs = NULL;  /* no user register state */
704         } else {
705                 childregs->gpr[1] = usp;
706                 p->thread.regs = childregs;
707                 if (clone_flags & CLONE_SETTLS) {
708 #ifdef CONFIG_PPC64
709                         if (!test_thread_flag(TIF_32BIT))
710                                 childregs->gpr[13] = childregs->gpr[6];
711                         else
712 #endif
713                                 childregs->gpr[2] = childregs->gpr[6];
714                 }
715         }
716         childregs->gpr[3] = 0;  /* Result from fork() */
717         sp -= STACK_FRAME_OVERHEAD;
718
719         /*
720          * The way this works is that at some point in the future
721          * some task will call _switch to switch to the new task.
722          * That will pop off the stack frame created below and start
723          * the new task running at ret_from_fork.  The new task will
724          * do some house keeping and then return from the fork or clone
725          * system call, using the stack frame created above.
726          */
727         sp -= sizeof(struct pt_regs);
728         kregs = (struct pt_regs *) sp;
729         sp -= STACK_FRAME_OVERHEAD;
730         p->thread.ksp = sp;
731         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
732                                 _ALIGN_UP(sizeof(struct thread_info), 16);
733
734 #ifdef CONFIG_PPC_STD_MMU_64
735         if (cpu_has_feature(CPU_FTR_SLB)) {
736                 unsigned long sp_vsid;
737                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
738
739                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
740                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
741                                 << SLB_VSID_SHIFT_1T;
742                 else
743                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
744                                 << SLB_VSID_SHIFT;
745                 sp_vsid |= SLB_VSID_KERNEL | llp;
746                 p->thread.ksp_vsid = sp_vsid;
747         }
748 #endif /* CONFIG_PPC_STD_MMU_64 */
749
750         /*
751          * The PPC64 ABI makes use of a TOC to contain function 
752          * pointers.  The function (ret_from_except) is actually a pointer
753          * to the TOC entry.  The first entry is a pointer to the actual
754          * function.
755          */
756 #ifdef CONFIG_PPC64
757         kregs->nip = *((unsigned long *)ret_from_fork);
758 #else
759         kregs->nip = (unsigned long)ret_from_fork;
760 #endif
761
762         return 0;
763 }
764
765 /*
766  * Set up a thread for executing a new program
767  */
768 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
769 {
770 #ifdef CONFIG_PPC64
771         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
772 #endif
773
774         set_fs(USER_DS);
775
776         /*
777          * If we exec out of a kernel thread then thread.regs will not be
778          * set.  Do it now.
779          */
780         if (!current->thread.regs) {
781                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
782                 current->thread.regs = regs - 1;
783         }
784
785         memset(regs->gpr, 0, sizeof(regs->gpr));
786         regs->ctr = 0;
787         regs->link = 0;
788         regs->xer = 0;
789         regs->ccr = 0;
790         regs->gpr[1] = sp;
791
792         /*
793          * We have just cleared all the nonvolatile GPRs, so make
794          * FULL_REGS(regs) return true.  This is necessary to allow
795          * ptrace to examine the thread immediately after exec.
796          */
797         regs->trap &= ~1UL;
798
799 #ifdef CONFIG_PPC32
800         regs->mq = 0;
801         regs->nip = start;
802         regs->msr = MSR_USER;
803 #else
804         if (!test_thread_flag(TIF_32BIT)) {
805                 unsigned long entry, toc;
806
807                 /* start is a relocated pointer to the function descriptor for
808                  * the elf _start routine.  The first entry in the function
809                  * descriptor is the entry address of _start and the second
810                  * entry is the TOC value we need to use.
811                  */
812                 __get_user(entry, (unsigned long __user *)start);
813                 __get_user(toc, (unsigned long __user *)start+1);
814
815                 /* Check whether the e_entry function descriptor entries
816                  * need to be relocated before we can use them.
817                  */
818                 if (load_addr != 0) {
819                         entry += load_addr;
820                         toc   += load_addr;
821                 }
822                 regs->nip = entry;
823                 regs->gpr[2] = toc;
824                 regs->msr = MSR_USER64;
825         } else {
826                 regs->nip = start;
827                 regs->gpr[2] = 0;
828                 regs->msr = MSR_USER32;
829         }
830 #endif
831
832         discard_lazy_cpu_state();
833 #ifdef CONFIG_VSX
834         current->thread.used_vsr = 0;
835 #endif
836         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
837         current->thread.fpscr.val = 0;
838 #ifdef CONFIG_ALTIVEC
839         memset(current->thread.vr, 0, sizeof(current->thread.vr));
840         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
841         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
842         current->thread.vrsave = 0;
843         current->thread.used_vr = 0;
844 #endif /* CONFIG_ALTIVEC */
845 #ifdef CONFIG_SPE
846         memset(current->thread.evr, 0, sizeof(current->thread.evr));
847         current->thread.acc = 0;
848         current->thread.spefscr = 0;
849         current->thread.used_spe = 0;
850 #endif /* CONFIG_SPE */
851 }
852
853 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
854                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
855
856 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
857 {
858         struct pt_regs *regs = tsk->thread.regs;
859
860         /* This is a bit hairy.  If we are an SPE enabled  processor
861          * (have embedded fp) we store the IEEE exception enable flags in
862          * fpexc_mode.  fpexc_mode is also used for setting FP exception
863          * mode (asyn, precise, disabled) for 'Classic' FP. */
864         if (val & PR_FP_EXC_SW_ENABLE) {
865 #ifdef CONFIG_SPE
866                 if (cpu_has_feature(CPU_FTR_SPE)) {
867                         tsk->thread.fpexc_mode = val &
868                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
869                         return 0;
870                 } else {
871                         return -EINVAL;
872                 }
873 #else
874                 return -EINVAL;
875 #endif
876         }
877
878         /* on a CONFIG_SPE this does not hurt us.  The bits that
879          * __pack_fe01 use do not overlap with bits used for
880          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
881          * on CONFIG_SPE implementations are reserved so writing to
882          * them does not change anything */
883         if (val > PR_FP_EXC_PRECISE)
884                 return -EINVAL;
885         tsk->thread.fpexc_mode = __pack_fe01(val);
886         if (regs != NULL && (regs->msr & MSR_FP) != 0)
887                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
888                         | tsk->thread.fpexc_mode;
889         return 0;
890 }
891
892 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
893 {
894         unsigned int val;
895
896         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
897 #ifdef CONFIG_SPE
898                 if (cpu_has_feature(CPU_FTR_SPE))
899                         val = tsk->thread.fpexc_mode;
900                 else
901                         return -EINVAL;
902 #else
903                 return -EINVAL;
904 #endif
905         else
906                 val = __unpack_fe01(tsk->thread.fpexc_mode);
907         return put_user(val, (unsigned int __user *) adr);
908 }
909
910 int set_endian(struct task_struct *tsk, unsigned int val)
911 {
912         struct pt_regs *regs = tsk->thread.regs;
913
914         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
915             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
916                 return -EINVAL;
917
918         if (regs == NULL)
919                 return -EINVAL;
920
921         if (val == PR_ENDIAN_BIG)
922                 regs->msr &= ~MSR_LE;
923         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
924                 regs->msr |= MSR_LE;
925         else
926                 return -EINVAL;
927
928         return 0;
929 }
930
931 int get_endian(struct task_struct *tsk, unsigned long adr)
932 {
933         struct pt_regs *regs = tsk->thread.regs;
934         unsigned int val;
935
936         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
937             !cpu_has_feature(CPU_FTR_REAL_LE))
938                 return -EINVAL;
939
940         if (regs == NULL)
941                 return -EINVAL;
942
943         if (regs->msr & MSR_LE) {
944                 if (cpu_has_feature(CPU_FTR_REAL_LE))
945                         val = PR_ENDIAN_LITTLE;
946                 else
947                         val = PR_ENDIAN_PPC_LITTLE;
948         } else
949                 val = PR_ENDIAN_BIG;
950
951         return put_user(val, (unsigned int __user *)adr);
952 }
953
954 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
955 {
956         tsk->thread.align_ctl = val;
957         return 0;
958 }
959
960 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
961 {
962         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
963 }
964
965 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
966
967 int sys_clone(unsigned long clone_flags, unsigned long usp,
968               int __user *parent_tidp, void __user *child_threadptr,
969               int __user *child_tidp, int p6,
970               struct pt_regs *regs)
971 {
972         CHECK_FULL_REGS(regs);
973         if (usp == 0)
974                 usp = regs->gpr[1];     /* stack pointer for child */
975 #ifdef CONFIG_PPC64
976         if (test_thread_flag(TIF_32BIT)) {
977                 parent_tidp = TRUNC_PTR(parent_tidp);
978                 child_tidp = TRUNC_PTR(child_tidp);
979         }
980 #endif
981         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
982 }
983
984 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
985              unsigned long p4, unsigned long p5, unsigned long p6,
986              struct pt_regs *regs)
987 {
988         CHECK_FULL_REGS(regs);
989         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
990 }
991
992 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
993               unsigned long p4, unsigned long p5, unsigned long p6,
994               struct pt_regs *regs)
995 {
996         CHECK_FULL_REGS(regs);
997         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
998                         regs, 0, NULL, NULL);
999 }
1000
1001 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
1002                unsigned long a3, unsigned long a4, unsigned long a5,
1003                struct pt_regs *regs)
1004 {
1005         int error;
1006         char *filename;
1007
1008         filename = getname((char __user *) a0);
1009         error = PTR_ERR(filename);
1010         if (IS_ERR(filename))
1011                 goto out;
1012         flush_fp_to_thread(current);
1013         flush_altivec_to_thread(current);
1014         flush_spe_to_thread(current);
1015         error = do_execve(filename, (char __user * __user *) a1,
1016                           (char __user * __user *) a2, regs);
1017         putname(filename);
1018 out:
1019         return error;
1020 }
1021
1022 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1023                                   unsigned long nbytes)
1024 {
1025         unsigned long stack_page;
1026         unsigned long cpu = task_cpu(p);
1027
1028         /*
1029          * Avoid crashing if the stack has overflowed and corrupted
1030          * task_cpu(p), which is in the thread_info struct.
1031          */
1032         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1033                 stack_page = (unsigned long) hardirq_ctx[cpu];
1034                 if (sp >= stack_page + sizeof(struct thread_struct)
1035                     && sp <= stack_page + THREAD_SIZE - nbytes)
1036                         return 1;
1037
1038                 stack_page = (unsigned long) softirq_ctx[cpu];
1039                 if (sp >= stack_page + sizeof(struct thread_struct)
1040                     && sp <= stack_page + THREAD_SIZE - nbytes)
1041                         return 1;
1042         }
1043         return 0;
1044 }
1045
1046 int validate_sp(unsigned long sp, struct task_struct *p,
1047                        unsigned long nbytes)
1048 {
1049         unsigned long stack_page = (unsigned long)task_stack_page(p);
1050
1051         if (sp >= stack_page + sizeof(struct thread_struct)
1052             && sp <= stack_page + THREAD_SIZE - nbytes)
1053                 return 1;
1054
1055         return valid_irq_stack(sp, p, nbytes);
1056 }
1057
1058 EXPORT_SYMBOL(validate_sp);
1059
1060 unsigned long get_wchan(struct task_struct *p)
1061 {
1062         unsigned long ip, sp;
1063         int count = 0;
1064
1065         if (!p || p == current || p->state == TASK_RUNNING)
1066                 return 0;
1067
1068         sp = p->thread.ksp;
1069         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1070                 return 0;
1071
1072         do {
1073                 sp = *(unsigned long *)sp;
1074                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1075                         return 0;
1076                 if (count > 0) {
1077                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1078                         if (!in_sched_functions(ip))
1079                                 return ip;
1080                 }
1081         } while (count++ < 16);
1082         return 0;
1083 }
1084
1085 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1086
1087 void show_stack(struct task_struct *tsk, unsigned long *stack)
1088 {
1089         unsigned long sp, ip, lr, newsp;
1090         int count = 0;
1091         int firstframe = 1;
1092 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1093         int curr_frame = current->curr_ret_stack;
1094         extern void return_to_handler(void);
1095         unsigned long rth = (unsigned long)return_to_handler;
1096         unsigned long mrth = -1;
1097 #ifdef CONFIG_PPC64
1098         extern void mod_return_to_handler(void);
1099         rth = *(unsigned long *)rth;
1100         mrth = (unsigned long)mod_return_to_handler;
1101         mrth = *(unsigned long *)mrth;
1102 #endif
1103 #endif
1104
1105         sp = (unsigned long) stack;
1106         if (tsk == NULL)
1107                 tsk = current;
1108         if (sp == 0) {
1109                 if (tsk == current)
1110                         asm("mr %0,1" : "=r" (sp));
1111                 else
1112                         sp = tsk->thread.ksp;
1113         }
1114
1115         lr = 0;
1116         printk("Call Trace:\n");
1117         do {
1118                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1119                         return;
1120
1121                 stack = (unsigned long *) sp;
1122                 newsp = stack[0];
1123                 ip = stack[STACK_FRAME_LR_SAVE];
1124                 if (!firstframe || ip != lr) {
1125                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1126 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1127                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1128                                 printk(" (%pS)",
1129                                        (void *)current->ret_stack[curr_frame].ret);
1130                                 curr_frame--;
1131                         }
1132 #endif
1133                         if (firstframe)
1134                                 printk(" (unreliable)");
1135                         printk("\n");
1136                 }
1137                 firstframe = 0;
1138
1139                 /*
1140                  * See if this is an exception frame.
1141                  * We look for the "regshere" marker in the current frame.
1142                  */
1143                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1144                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1145                         struct pt_regs *regs = (struct pt_regs *)
1146                                 (sp + STACK_FRAME_OVERHEAD);
1147                         lr = regs->link;
1148                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1149                                regs->trap, (void *)regs->nip, (void *)lr);
1150                         firstframe = 1;
1151                 }
1152
1153                 sp = newsp;
1154         } while (count++ < kstack_depth_to_print);
1155 }
1156
1157 void dump_stack(void)
1158 {
1159         show_stack(current, NULL);
1160 }
1161 EXPORT_SYMBOL(dump_stack);
1162
1163 #ifdef CONFIG_PPC64
1164 void ppc64_runlatch_on(void)
1165 {
1166         unsigned long ctrl;
1167
1168         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1169                 HMT_medium();
1170
1171                 ctrl = mfspr(SPRN_CTRLF);
1172                 ctrl |= CTRL_RUNLATCH;
1173                 mtspr(SPRN_CTRLT, ctrl);
1174
1175                 set_thread_flag(TIF_RUNLATCH);
1176         }
1177 }
1178
1179 void ppc64_runlatch_off(void)
1180 {
1181         unsigned long ctrl;
1182
1183         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1184                 HMT_medium();
1185
1186                 clear_thread_flag(TIF_RUNLATCH);
1187
1188                 ctrl = mfspr(SPRN_CTRLF);
1189                 ctrl &= ~CTRL_RUNLATCH;
1190                 mtspr(SPRN_CTRLT, ctrl);
1191         }
1192 }
1193 #endif
1194
1195 #if THREAD_SHIFT < PAGE_SHIFT
1196
1197 static struct kmem_cache *thread_info_cache;
1198
1199 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1200 {
1201         struct thread_info *ti;
1202
1203         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1204         if (unlikely(ti == NULL))
1205                 return NULL;
1206 #ifdef CONFIG_DEBUG_STACK_USAGE
1207         memset(ti, 0, THREAD_SIZE);
1208 #endif
1209         return ti;
1210 }
1211
1212 void free_thread_info(struct thread_info *ti)
1213 {
1214         kmem_cache_free(thread_info_cache, ti);
1215 }
1216
1217 void thread_info_cache_init(void)
1218 {
1219         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1220                                               THREAD_SIZE, 0, NULL);
1221         BUG_ON(thread_info_cache == NULL);
1222 }
1223
1224 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1225
1226 unsigned long arch_align_stack(unsigned long sp)
1227 {
1228         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1229                 sp -= get_random_int() & ~PAGE_MASK;
1230         return sp & ~0xf;
1231 }
1232
1233 static inline unsigned long brk_rnd(void)
1234 {
1235         unsigned long rnd = 0;
1236
1237         /* 8MB for 32bit, 1GB for 64bit */
1238         if (is_32bit_task())
1239                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1240         else
1241                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1242
1243         return rnd << PAGE_SHIFT;
1244 }
1245
1246 unsigned long arch_randomize_brk(struct mm_struct *mm)
1247 {
1248         unsigned long base = mm->brk;
1249         unsigned long ret;
1250
1251 #ifdef CONFIG_PPC_STD_MMU_64
1252         /*
1253          * If we are using 1TB segments and we are allowed to randomise
1254          * the heap, we can put it above 1TB so it is backed by a 1TB
1255          * segment. Otherwise the heap will be in the bottom 1TB
1256          * which always uses 256MB segments and this may result in a
1257          * performance penalty.
1258          */
1259         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1260                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1261 #endif
1262
1263         ret = PAGE_ALIGN(base + brk_rnd());
1264
1265         if (ret < mm->brk)
1266                 return mm->brk;
1267
1268         return ret;
1269 }
1270
1271 unsigned long randomize_et_dyn(unsigned long base)
1272 {
1273         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1274
1275         if (ret < base)
1276                 return base;
1277
1278         return ret;
1279 }