xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision f42b3800)
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 
37 #include <asm/pgtable.h>
38 #include <asm/uaccess.h>
39 #include <asm/system.h>
40 #include <asm/io.h>
41 #include <asm/processor.h>
42 #include <asm/mmu.h>
43 #include <asm/prom.h>
44 #include <asm/machdep.h>
45 #include <asm/time.h>
46 #include <asm/syscalls.h>
47 #ifdef CONFIG_PPC64
48 #include <asm/firmware.h>
49 #endif
50 
51 extern unsigned long _get_SP(void);
52 
53 #ifndef CONFIG_SMP
54 struct task_struct *last_task_used_math = NULL;
55 struct task_struct *last_task_used_altivec = NULL;
56 struct task_struct *last_task_used_spe = NULL;
57 #endif
58 
59 /*
60  * Make sure the floating-point register state in the
61  * the thread_struct is up to date for task tsk.
62  */
63 void flush_fp_to_thread(struct task_struct *tsk)
64 {
65 	if (tsk->thread.regs) {
66 		/*
67 		 * We need to disable preemption here because if we didn't,
68 		 * another process could get scheduled after the regs->msr
69 		 * test but before we have finished saving the FP registers
70 		 * to the thread_struct.  That process could take over the
71 		 * FPU, and then when we get scheduled again we would store
72 		 * bogus values for the remaining FP registers.
73 		 */
74 		preempt_disable();
75 		if (tsk->thread.regs->msr & MSR_FP) {
76 #ifdef CONFIG_SMP
77 			/*
78 			 * This should only ever be called for current or
79 			 * for a stopped child process.  Since we save away
80 			 * the FP register state on context switch on SMP,
81 			 * there is something wrong if a stopped child appears
82 			 * to still have its FP state in the CPU registers.
83 			 */
84 			BUG_ON(tsk != current);
85 #endif
86 			giveup_fpu(tsk);
87 		}
88 		preempt_enable();
89 	}
90 }
91 
92 void enable_kernel_fp(void)
93 {
94 	WARN_ON(preemptible());
95 
96 #ifdef CONFIG_SMP
97 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
98 		giveup_fpu(current);
99 	else
100 		giveup_fpu(NULL);	/* just enables FP for kernel */
101 #else
102 	giveup_fpu(last_task_used_math);
103 #endif /* CONFIG_SMP */
104 }
105 EXPORT_SYMBOL(enable_kernel_fp);
106 
107 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
108 {
109 	if (!tsk->thread.regs)
110 		return 0;
111 	flush_fp_to_thread(current);
112 
113 	memcpy(fpregs, &tsk->thread.fpr[0], sizeof(*fpregs));
114 
115 	return 1;
116 }
117 
118 #ifdef CONFIG_ALTIVEC
119 void enable_kernel_altivec(void)
120 {
121 	WARN_ON(preemptible());
122 
123 #ifdef CONFIG_SMP
124 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
125 		giveup_altivec(current);
126 	else
127 		giveup_altivec(NULL);	/* just enable AltiVec for kernel - force */
128 #else
129 	giveup_altivec(last_task_used_altivec);
130 #endif /* CONFIG_SMP */
131 }
132 EXPORT_SYMBOL(enable_kernel_altivec);
133 
134 /*
135  * Make sure the VMX/Altivec register state in the
136  * the thread_struct is up to date for task tsk.
137  */
138 void flush_altivec_to_thread(struct task_struct *tsk)
139 {
140 	if (tsk->thread.regs) {
141 		preempt_disable();
142 		if (tsk->thread.regs->msr & MSR_VEC) {
143 #ifdef CONFIG_SMP
144 			BUG_ON(tsk != current);
145 #endif
146 			giveup_altivec(tsk);
147 		}
148 		preempt_enable();
149 	}
150 }
151 
152 int dump_task_altivec(struct task_struct *tsk, elf_vrregset_t *vrregs)
153 {
154 	/* ELF_NVRREG includes the VSCR and VRSAVE which we need to save
155 	 * separately, see below */
156 	const int nregs = ELF_NVRREG - 2;
157 	elf_vrreg_t *reg;
158 	u32 *dest;
159 
160 	if (tsk == current)
161 		flush_altivec_to_thread(tsk);
162 
163 	reg = (elf_vrreg_t *)vrregs;
164 
165 	/* copy the 32 vr registers */
166 	memcpy(reg, &tsk->thread.vr[0], nregs * sizeof(*reg));
167 	reg += nregs;
168 
169 	/* copy the vscr */
170 	memcpy(reg, &tsk->thread.vscr, sizeof(*reg));
171 	reg++;
172 
173 	/* vrsave is stored in the high 32bit slot of the final 128bits */
174 	memset(reg, 0, sizeof(*reg));
175 	dest = (u32 *)reg;
176 	*dest = tsk->thread.vrsave;
177 
178 	return 1;
179 }
180 #endif /* CONFIG_ALTIVEC */
181 
182 #ifdef CONFIG_SPE
183 
184 void enable_kernel_spe(void)
185 {
186 	WARN_ON(preemptible());
187 
188 #ifdef CONFIG_SMP
189 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
190 		giveup_spe(current);
191 	else
192 		giveup_spe(NULL);	/* just enable SPE for kernel - force */
193 #else
194 	giveup_spe(last_task_used_spe);
195 #endif /* __SMP __ */
196 }
197 EXPORT_SYMBOL(enable_kernel_spe);
198 
199 void flush_spe_to_thread(struct task_struct *tsk)
200 {
201 	if (tsk->thread.regs) {
202 		preempt_disable();
203 		if (tsk->thread.regs->msr & MSR_SPE) {
204 #ifdef CONFIG_SMP
205 			BUG_ON(tsk != current);
206 #endif
207 			giveup_spe(tsk);
208 		}
209 		preempt_enable();
210 	}
211 }
212 
213 int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
214 {
215 	flush_spe_to_thread(current);
216 	/* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
217 	memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
218 	return 1;
219 }
220 #endif /* CONFIG_SPE */
221 
222 #ifndef CONFIG_SMP
223 /*
224  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
225  * and the current task has some state, discard it.
226  */
227 void discard_lazy_cpu_state(void)
228 {
229 	preempt_disable();
230 	if (last_task_used_math == current)
231 		last_task_used_math = NULL;
232 #ifdef CONFIG_ALTIVEC
233 	if (last_task_used_altivec == current)
234 		last_task_used_altivec = NULL;
235 #endif /* CONFIG_ALTIVEC */
236 #ifdef CONFIG_SPE
237 	if (last_task_used_spe == current)
238 		last_task_used_spe = NULL;
239 #endif
240 	preempt_enable();
241 }
242 #endif /* CONFIG_SMP */
243 
244 static DEFINE_PER_CPU(unsigned long, current_dabr);
245 
246 int set_dabr(unsigned long dabr)
247 {
248 	__get_cpu_var(current_dabr) = dabr;
249 
250 #ifdef CONFIG_PPC_MERGE		/* XXX for now */
251 	if (ppc_md.set_dabr)
252 		return ppc_md.set_dabr(dabr);
253 #endif
254 
255 	/* XXX should we have a CPU_FTR_HAS_DABR ? */
256 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
257 	mtspr(SPRN_DABR, dabr);
258 #endif
259 	return 0;
260 }
261 
262 #ifdef CONFIG_PPC64
263 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
264 #endif
265 
266 struct task_struct *__switch_to(struct task_struct *prev,
267 	struct task_struct *new)
268 {
269 	struct thread_struct *new_thread, *old_thread;
270 	unsigned long flags;
271 	struct task_struct *last;
272 
273 #ifdef CONFIG_SMP
274 	/* avoid complexity of lazy save/restore of fpu
275 	 * by just saving it every time we switch out if
276 	 * this task used the fpu during the last quantum.
277 	 *
278 	 * If it tries to use the fpu again, it'll trap and
279 	 * reload its fp regs.  So we don't have to do a restore
280 	 * every switch, just a save.
281 	 *  -- Cort
282 	 */
283 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
284 		giveup_fpu(prev);
285 #ifdef CONFIG_ALTIVEC
286 	/*
287 	 * If the previous thread used altivec in the last quantum
288 	 * (thus changing altivec regs) then save them.
289 	 * We used to check the VRSAVE register but not all apps
290 	 * set it, so we don't rely on it now (and in fact we need
291 	 * to save & restore VSCR even if VRSAVE == 0).  -- paulus
292 	 *
293 	 * On SMP we always save/restore altivec regs just to avoid the
294 	 * complexity of changing processors.
295 	 *  -- Cort
296 	 */
297 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
298 		giveup_altivec(prev);
299 #endif /* CONFIG_ALTIVEC */
300 #ifdef CONFIG_SPE
301 	/*
302 	 * If the previous thread used spe in the last quantum
303 	 * (thus changing spe regs) then save them.
304 	 *
305 	 * On SMP we always save/restore spe regs just to avoid the
306 	 * complexity of changing processors.
307 	 */
308 	if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
309 		giveup_spe(prev);
310 #endif /* CONFIG_SPE */
311 
312 #else  /* CONFIG_SMP */
313 #ifdef CONFIG_ALTIVEC
314 	/* Avoid the trap.  On smp this this never happens since
315 	 * we don't set last_task_used_altivec -- Cort
316 	 */
317 	if (new->thread.regs && last_task_used_altivec == new)
318 		new->thread.regs->msr |= MSR_VEC;
319 #endif /* CONFIG_ALTIVEC */
320 #ifdef CONFIG_SPE
321 	/* Avoid the trap.  On smp this this never happens since
322 	 * we don't set last_task_used_spe
323 	 */
324 	if (new->thread.regs && last_task_used_spe == new)
325 		new->thread.regs->msr |= MSR_SPE;
326 #endif /* CONFIG_SPE */
327 
328 #endif /* CONFIG_SMP */
329 
330 	if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
331 		set_dabr(new->thread.dabr);
332 
333 	new_thread = &new->thread;
334 	old_thread = &current->thread;
335 
336 #ifdef CONFIG_PPC64
337 	/*
338 	 * Collect processor utilization data per process
339 	 */
340 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
341 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
342 		long unsigned start_tb, current_tb;
343 		start_tb = old_thread->start_tb;
344 		cu->current_tb = current_tb = mfspr(SPRN_PURR);
345 		old_thread->accum_tb += (current_tb - start_tb);
346 		new_thread->start_tb = current_tb;
347 	}
348 #endif
349 
350 	local_irq_save(flags);
351 
352 	account_system_vtime(current);
353 	account_process_vtime(current);
354 	calculate_steal_time();
355 
356 	/*
357 	 * We can't take a PMU exception inside _switch() since there is a
358 	 * window where the kernel stack SLB and the kernel stack are out
359 	 * of sync. Hard disable here.
360 	 */
361 	hard_irq_disable();
362 	last = _switch(old_thread, new_thread);
363 
364 	local_irq_restore(flags);
365 
366 	return last;
367 }
368 
369 static int instructions_to_print = 16;
370 
371 static void show_instructions(struct pt_regs *regs)
372 {
373 	int i;
374 	unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
375 			sizeof(int));
376 
377 	printk("Instruction dump:");
378 
379 	for (i = 0; i < instructions_to_print; i++) {
380 		int instr;
381 
382 		if (!(i % 8))
383 			printk("\n");
384 
385 #if !defined(CONFIG_BOOKE)
386 		/* If executing with the IMMU off, adjust pc rather
387 		 * than print XXXXXXXX.
388 		 */
389 		if (!(regs->msr & MSR_IR))
390 			pc = (unsigned long)phys_to_virt(pc);
391 #endif
392 
393 		/* We use __get_user here *only* to avoid an OOPS on a
394 		 * bad address because the pc *should* only be a
395 		 * kernel address.
396 		 */
397 		if (!__kernel_text_address(pc) ||
398 		     __get_user(instr, (unsigned int __user *)pc)) {
399 			printk("XXXXXXXX ");
400 		} else {
401 			if (regs->nip == pc)
402 				printk("<%08x> ", instr);
403 			else
404 				printk("%08x ", instr);
405 		}
406 
407 		pc += sizeof(int);
408 	}
409 
410 	printk("\n");
411 }
412 
413 static struct regbit {
414 	unsigned long bit;
415 	const char *name;
416 } msr_bits[] = {
417 	{MSR_EE,	"EE"},
418 	{MSR_PR,	"PR"},
419 	{MSR_FP,	"FP"},
420 	{MSR_ME,	"ME"},
421 	{MSR_IR,	"IR"},
422 	{MSR_DR,	"DR"},
423 	{0,		NULL}
424 };
425 
426 static void printbits(unsigned long val, struct regbit *bits)
427 {
428 	const char *sep = "";
429 
430 	printk("<");
431 	for (; bits->bit; ++bits)
432 		if (val & bits->bit) {
433 			printk("%s%s", sep, bits->name);
434 			sep = ",";
435 		}
436 	printk(">");
437 }
438 
439 #ifdef CONFIG_PPC64
440 #define REG		"%016lx"
441 #define REGS_PER_LINE	4
442 #define LAST_VOLATILE	13
443 #else
444 #define REG		"%08lx"
445 #define REGS_PER_LINE	8
446 #define LAST_VOLATILE	12
447 #endif
448 
449 void show_regs(struct pt_regs * regs)
450 {
451 	int i, trap;
452 
453 	printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
454 	       regs->nip, regs->link, regs->ctr);
455 	printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
456 	       regs, regs->trap, print_tainted(), init_utsname()->release);
457 	printk("MSR: "REG" ", regs->msr);
458 	printbits(regs->msr, msr_bits);
459 	printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
460 	trap = TRAP(regs);
461 	if (trap == 0x300 || trap == 0x600)
462 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
463 		printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
464 #else
465 		printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
466 #endif
467 	printk("TASK = %p[%d] '%s' THREAD: %p",
468 	       current, task_pid_nr(current), current->comm, task_thread_info(current));
469 
470 #ifdef CONFIG_SMP
471 	printk(" CPU: %d", raw_smp_processor_id());
472 #endif /* CONFIG_SMP */
473 
474 	for (i = 0;  i < 32;  i++) {
475 		if ((i % REGS_PER_LINE) == 0)
476 			printk("\n" KERN_INFO "GPR%02d: ", i);
477 		printk(REG " ", regs->gpr[i]);
478 		if (i == LAST_VOLATILE && !FULL_REGS(regs))
479 			break;
480 	}
481 	printk("\n");
482 #ifdef CONFIG_KALLSYMS
483 	/*
484 	 * Lookup NIP late so we have the best change of getting the
485 	 * above info out without failing
486 	 */
487 	printk("NIP ["REG"] ", regs->nip);
488 	print_symbol("%s\n", regs->nip);
489 	printk("LR ["REG"] ", regs->link);
490 	print_symbol("%s\n", regs->link);
491 #endif
492 	show_stack(current, (unsigned long *) regs->gpr[1]);
493 	if (!user_mode(regs))
494 		show_instructions(regs);
495 }
496 
497 void exit_thread(void)
498 {
499 	discard_lazy_cpu_state();
500 }
501 
502 void flush_thread(void)
503 {
504 #ifdef CONFIG_PPC64
505 	struct thread_info *t = current_thread_info();
506 
507 	if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
508 		clear_ti_thread_flag(t, TIF_ABI_PENDING);
509 		if (test_ti_thread_flag(t, TIF_32BIT))
510 			clear_ti_thread_flag(t, TIF_32BIT);
511 		else
512 			set_ti_thread_flag(t, TIF_32BIT);
513 	}
514 #endif
515 
516 	discard_lazy_cpu_state();
517 
518 	if (current->thread.dabr) {
519 		current->thread.dabr = 0;
520 		set_dabr(0);
521 	}
522 }
523 
524 void
525 release_thread(struct task_struct *t)
526 {
527 }
528 
529 /*
530  * This gets called before we allocate a new thread and copy
531  * the current task into it.
532  */
533 void prepare_to_copy(struct task_struct *tsk)
534 {
535 	flush_fp_to_thread(current);
536 	flush_altivec_to_thread(current);
537 	flush_spe_to_thread(current);
538 }
539 
540 /*
541  * Copy a thread..
542  */
543 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
544 		unsigned long unused, struct task_struct *p,
545 		struct pt_regs *regs)
546 {
547 	struct pt_regs *childregs, *kregs;
548 	extern void ret_from_fork(void);
549 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
550 
551 	CHECK_FULL_REGS(regs);
552 	/* Copy registers */
553 	sp -= sizeof(struct pt_regs);
554 	childregs = (struct pt_regs *) sp;
555 	*childregs = *regs;
556 	if ((childregs->msr & MSR_PR) == 0) {
557 		/* for kernel thread, set `current' and stackptr in new task */
558 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
559 #ifdef CONFIG_PPC32
560 		childregs->gpr[2] = (unsigned long) p;
561 #else
562 		clear_tsk_thread_flag(p, TIF_32BIT);
563 #endif
564 		p->thread.regs = NULL;	/* no user register state */
565 	} else {
566 		childregs->gpr[1] = usp;
567 		p->thread.regs = childregs;
568 		if (clone_flags & CLONE_SETTLS) {
569 #ifdef CONFIG_PPC64
570 			if (!test_thread_flag(TIF_32BIT))
571 				childregs->gpr[13] = childregs->gpr[6];
572 			else
573 #endif
574 				childregs->gpr[2] = childregs->gpr[6];
575 		}
576 	}
577 	childregs->gpr[3] = 0;  /* Result from fork() */
578 	sp -= STACK_FRAME_OVERHEAD;
579 
580 	/*
581 	 * The way this works is that at some point in the future
582 	 * some task will call _switch to switch to the new task.
583 	 * That will pop off the stack frame created below and start
584 	 * the new task running at ret_from_fork.  The new task will
585 	 * do some house keeping and then return from the fork or clone
586 	 * system call, using the stack frame created above.
587 	 */
588 	sp -= sizeof(struct pt_regs);
589 	kregs = (struct pt_regs *) sp;
590 	sp -= STACK_FRAME_OVERHEAD;
591 	p->thread.ksp = sp;
592 
593 #ifdef CONFIG_PPC64
594 	if (cpu_has_feature(CPU_FTR_SLB)) {
595 		unsigned long sp_vsid;
596 		unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
597 
598 		if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
599 			sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
600 				<< SLB_VSID_SHIFT_1T;
601 		else
602 			sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
603 				<< SLB_VSID_SHIFT;
604 		sp_vsid |= SLB_VSID_KERNEL | llp;
605 		p->thread.ksp_vsid = sp_vsid;
606 	}
607 
608 	/*
609 	 * The PPC64 ABI makes use of a TOC to contain function
610 	 * pointers.  The function (ret_from_except) is actually a pointer
611 	 * to the TOC entry.  The first entry is a pointer to the actual
612 	 * function.
613  	 */
614 	kregs->nip = *((unsigned long *)ret_from_fork);
615 #else
616 	kregs->nip = (unsigned long)ret_from_fork;
617 #endif
618 
619 	return 0;
620 }
621 
622 /*
623  * Set up a thread for executing a new program
624  */
625 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
626 {
627 #ifdef CONFIG_PPC64
628 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
629 #endif
630 
631 	set_fs(USER_DS);
632 
633 	/*
634 	 * If we exec out of a kernel thread then thread.regs will not be
635 	 * set.  Do it now.
636 	 */
637 	if (!current->thread.regs) {
638 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
639 		current->thread.regs = regs - 1;
640 	}
641 
642 	memset(regs->gpr, 0, sizeof(regs->gpr));
643 	regs->ctr = 0;
644 	regs->link = 0;
645 	regs->xer = 0;
646 	regs->ccr = 0;
647 	regs->gpr[1] = sp;
648 
649 	/*
650 	 * We have just cleared all the nonvolatile GPRs, so make
651 	 * FULL_REGS(regs) return true.  This is necessary to allow
652 	 * ptrace to examine the thread immediately after exec.
653 	 */
654 	regs->trap &= ~1UL;
655 
656 #ifdef CONFIG_PPC32
657 	regs->mq = 0;
658 	regs->nip = start;
659 	regs->msr = MSR_USER;
660 #else
661 	if (!test_thread_flag(TIF_32BIT)) {
662 		unsigned long entry, toc;
663 
664 		/* start is a relocated pointer to the function descriptor for
665 		 * the elf _start routine.  The first entry in the function
666 		 * descriptor is the entry address of _start and the second
667 		 * entry is the TOC value we need to use.
668 		 */
669 		__get_user(entry, (unsigned long __user *)start);
670 		__get_user(toc, (unsigned long __user *)start+1);
671 
672 		/* Check whether the e_entry function descriptor entries
673 		 * need to be relocated before we can use them.
674 		 */
675 		if (load_addr != 0) {
676 			entry += load_addr;
677 			toc   += load_addr;
678 		}
679 		regs->nip = entry;
680 		regs->gpr[2] = toc;
681 		regs->msr = MSR_USER64;
682 	} else {
683 		regs->nip = start;
684 		regs->gpr[2] = 0;
685 		regs->msr = MSR_USER32;
686 	}
687 #endif
688 
689 	discard_lazy_cpu_state();
690 	memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
691 	current->thread.fpscr.val = 0;
692 #ifdef CONFIG_ALTIVEC
693 	memset(current->thread.vr, 0, sizeof(current->thread.vr));
694 	memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
695 	current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
696 	current->thread.vrsave = 0;
697 	current->thread.used_vr = 0;
698 #endif /* CONFIG_ALTIVEC */
699 #ifdef CONFIG_SPE
700 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
701 	current->thread.acc = 0;
702 	current->thread.spefscr = 0;
703 	current->thread.used_spe = 0;
704 #endif /* CONFIG_SPE */
705 }
706 
707 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
708 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
709 
710 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
711 {
712 	struct pt_regs *regs = tsk->thread.regs;
713 
714 	/* This is a bit hairy.  If we are an SPE enabled  processor
715 	 * (have embedded fp) we store the IEEE exception enable flags in
716 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
717 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
718 	if (val & PR_FP_EXC_SW_ENABLE) {
719 #ifdef CONFIG_SPE
720 		if (cpu_has_feature(CPU_FTR_SPE)) {
721 			tsk->thread.fpexc_mode = val &
722 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
723 			return 0;
724 		} else {
725 			return -EINVAL;
726 		}
727 #else
728 		return -EINVAL;
729 #endif
730 	}
731 
732 	/* on a CONFIG_SPE this does not hurt us.  The bits that
733 	 * __pack_fe01 use do not overlap with bits used for
734 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
735 	 * on CONFIG_SPE implementations are reserved so writing to
736 	 * them does not change anything */
737 	if (val > PR_FP_EXC_PRECISE)
738 		return -EINVAL;
739 	tsk->thread.fpexc_mode = __pack_fe01(val);
740 	if (regs != NULL && (regs->msr & MSR_FP) != 0)
741 		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
742 			| tsk->thread.fpexc_mode;
743 	return 0;
744 }
745 
746 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
747 {
748 	unsigned int val;
749 
750 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
751 #ifdef CONFIG_SPE
752 		if (cpu_has_feature(CPU_FTR_SPE))
753 			val = tsk->thread.fpexc_mode;
754 		else
755 			return -EINVAL;
756 #else
757 		return -EINVAL;
758 #endif
759 	else
760 		val = __unpack_fe01(tsk->thread.fpexc_mode);
761 	return put_user(val, (unsigned int __user *) adr);
762 }
763 
764 int set_endian(struct task_struct *tsk, unsigned int val)
765 {
766 	struct pt_regs *regs = tsk->thread.regs;
767 
768 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
769 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
770 		return -EINVAL;
771 
772 	if (regs == NULL)
773 		return -EINVAL;
774 
775 	if (val == PR_ENDIAN_BIG)
776 		regs->msr &= ~MSR_LE;
777 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
778 		regs->msr |= MSR_LE;
779 	else
780 		return -EINVAL;
781 
782 	return 0;
783 }
784 
785 int get_endian(struct task_struct *tsk, unsigned long adr)
786 {
787 	struct pt_regs *regs = tsk->thread.regs;
788 	unsigned int val;
789 
790 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
791 	    !cpu_has_feature(CPU_FTR_REAL_LE))
792 		return -EINVAL;
793 
794 	if (regs == NULL)
795 		return -EINVAL;
796 
797 	if (regs->msr & MSR_LE) {
798 		if (cpu_has_feature(CPU_FTR_REAL_LE))
799 			val = PR_ENDIAN_LITTLE;
800 		else
801 			val = PR_ENDIAN_PPC_LITTLE;
802 	} else
803 		val = PR_ENDIAN_BIG;
804 
805 	return put_user(val, (unsigned int __user *)adr);
806 }
807 
808 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
809 {
810 	tsk->thread.align_ctl = val;
811 	return 0;
812 }
813 
814 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
815 {
816 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
817 }
818 
819 #define TRUNC_PTR(x)	((typeof(x))(((unsigned long)(x)) & 0xffffffff))
820 
821 int sys_clone(unsigned long clone_flags, unsigned long usp,
822 	      int __user *parent_tidp, void __user *child_threadptr,
823 	      int __user *child_tidp, int p6,
824 	      struct pt_regs *regs)
825 {
826 	CHECK_FULL_REGS(regs);
827 	if (usp == 0)
828 		usp = regs->gpr[1];	/* stack pointer for child */
829 #ifdef CONFIG_PPC64
830 	if (test_thread_flag(TIF_32BIT)) {
831 		parent_tidp = TRUNC_PTR(parent_tidp);
832 		child_tidp = TRUNC_PTR(child_tidp);
833 	}
834 #endif
835  	return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
836 }
837 
838 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
839 	     unsigned long p4, unsigned long p5, unsigned long p6,
840 	     struct pt_regs *regs)
841 {
842 	CHECK_FULL_REGS(regs);
843 	return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
844 }
845 
846 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
847 	      unsigned long p4, unsigned long p5, unsigned long p6,
848 	      struct pt_regs *regs)
849 {
850 	CHECK_FULL_REGS(regs);
851 	return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
852 			regs, 0, NULL, NULL);
853 }
854 
855 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
856 	       unsigned long a3, unsigned long a4, unsigned long a5,
857 	       struct pt_regs *regs)
858 {
859 	int error;
860 	char *filename;
861 
862 	filename = getname((char __user *) a0);
863 	error = PTR_ERR(filename);
864 	if (IS_ERR(filename))
865 		goto out;
866 	flush_fp_to_thread(current);
867 	flush_altivec_to_thread(current);
868 	flush_spe_to_thread(current);
869 	error = do_execve(filename, (char __user * __user *) a1,
870 			  (char __user * __user *) a2, regs);
871 	if (error == 0) {
872 		task_lock(current);
873 		current->ptrace &= ~PT_DTRACE;
874 		task_unlock(current);
875 	}
876 	putname(filename);
877 out:
878 	return error;
879 }
880 
881 #ifdef CONFIG_IRQSTACKS
882 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
883 				  unsigned long nbytes)
884 {
885 	unsigned long stack_page;
886 	unsigned long cpu = task_cpu(p);
887 
888 	/*
889 	 * Avoid crashing if the stack has overflowed and corrupted
890 	 * task_cpu(p), which is in the thread_info struct.
891 	 */
892 	if (cpu < NR_CPUS && cpu_possible(cpu)) {
893 		stack_page = (unsigned long) hardirq_ctx[cpu];
894 		if (sp >= stack_page + sizeof(struct thread_struct)
895 		    && sp <= stack_page + THREAD_SIZE - nbytes)
896 			return 1;
897 
898 		stack_page = (unsigned long) softirq_ctx[cpu];
899 		if (sp >= stack_page + sizeof(struct thread_struct)
900 		    && sp <= stack_page + THREAD_SIZE - nbytes)
901 			return 1;
902 	}
903 	return 0;
904 }
905 
906 #else
907 #define valid_irq_stack(sp, p, nb)	0
908 #endif /* CONFIG_IRQSTACKS */
909 
910 int validate_sp(unsigned long sp, struct task_struct *p,
911 		       unsigned long nbytes)
912 {
913 	unsigned long stack_page = (unsigned long)task_stack_page(p);
914 
915 	if (sp >= stack_page + sizeof(struct thread_struct)
916 	    && sp <= stack_page + THREAD_SIZE - nbytes)
917 		return 1;
918 
919 	return valid_irq_stack(sp, p, nbytes);
920 }
921 
922 #ifdef CONFIG_PPC64
923 #define MIN_STACK_FRAME	112	/* same as STACK_FRAME_OVERHEAD, in fact */
924 #define FRAME_LR_SAVE	2
925 #define INT_FRAME_SIZE	(sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD + 288)
926 #define REGS_MARKER	0x7265677368657265ul
927 #define FRAME_MARKER	12
928 #else
929 #define MIN_STACK_FRAME	16
930 #define FRAME_LR_SAVE	1
931 #define INT_FRAME_SIZE	(sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD)
932 #define REGS_MARKER	0x72656773ul
933 #define FRAME_MARKER	2
934 #endif
935 
936 EXPORT_SYMBOL(validate_sp);
937 
938 unsigned long get_wchan(struct task_struct *p)
939 {
940 	unsigned long ip, sp;
941 	int count = 0;
942 
943 	if (!p || p == current || p->state == TASK_RUNNING)
944 		return 0;
945 
946 	sp = p->thread.ksp;
947 	if (!validate_sp(sp, p, MIN_STACK_FRAME))
948 		return 0;
949 
950 	do {
951 		sp = *(unsigned long *)sp;
952 		if (!validate_sp(sp, p, MIN_STACK_FRAME))
953 			return 0;
954 		if (count > 0) {
955 			ip = ((unsigned long *)sp)[FRAME_LR_SAVE];
956 			if (!in_sched_functions(ip))
957 				return ip;
958 		}
959 	} while (count++ < 16);
960 	return 0;
961 }
962 
963 static int kstack_depth_to_print = 64;
964 
965 void show_stack(struct task_struct *tsk, unsigned long *stack)
966 {
967 	unsigned long sp, ip, lr, newsp;
968 	int count = 0;
969 	int firstframe = 1;
970 
971 	sp = (unsigned long) stack;
972 	if (tsk == NULL)
973 		tsk = current;
974 	if (sp == 0) {
975 		if (tsk == current)
976 			asm("mr %0,1" : "=r" (sp));
977 		else
978 			sp = tsk->thread.ksp;
979 	}
980 
981 	lr = 0;
982 	printk("Call Trace:\n");
983 	do {
984 		if (!validate_sp(sp, tsk, MIN_STACK_FRAME))
985 			return;
986 
987 		stack = (unsigned long *) sp;
988 		newsp = stack[0];
989 		ip = stack[FRAME_LR_SAVE];
990 		if (!firstframe || ip != lr) {
991 			printk("["REG"] ["REG"] ", sp, ip);
992 			print_symbol("%s", ip);
993 			if (firstframe)
994 				printk(" (unreliable)");
995 			printk("\n");
996 		}
997 		firstframe = 0;
998 
999 		/*
1000 		 * See if this is an exception frame.
1001 		 * We look for the "regshere" marker in the current frame.
1002 		 */
1003 		if (validate_sp(sp, tsk, INT_FRAME_SIZE)
1004 		    && stack[FRAME_MARKER] == REGS_MARKER) {
1005 			struct pt_regs *regs = (struct pt_regs *)
1006 				(sp + STACK_FRAME_OVERHEAD);
1007 			printk("--- Exception: %lx", regs->trap);
1008 			print_symbol(" at %s\n", regs->nip);
1009 			lr = regs->link;
1010 			print_symbol("    LR = %s\n", lr);
1011 			firstframe = 1;
1012 		}
1013 
1014 		sp = newsp;
1015 	} while (count++ < kstack_depth_to_print);
1016 }
1017 
1018 void dump_stack(void)
1019 {
1020 	show_stack(current, NULL);
1021 }
1022 EXPORT_SYMBOL(dump_stack);
1023 
1024 #ifdef CONFIG_PPC64
1025 void ppc64_runlatch_on(void)
1026 {
1027 	unsigned long ctrl;
1028 
1029 	if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1030 		HMT_medium();
1031 
1032 		ctrl = mfspr(SPRN_CTRLF);
1033 		ctrl |= CTRL_RUNLATCH;
1034 		mtspr(SPRN_CTRLT, ctrl);
1035 
1036 		set_thread_flag(TIF_RUNLATCH);
1037 	}
1038 }
1039 
1040 void ppc64_runlatch_off(void)
1041 {
1042 	unsigned long ctrl;
1043 
1044 	if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1045 		HMT_medium();
1046 
1047 		clear_thread_flag(TIF_RUNLATCH);
1048 
1049 		ctrl = mfspr(SPRN_CTRLF);
1050 		ctrl &= ~CTRL_RUNLATCH;
1051 		mtspr(SPRN_CTRLT, ctrl);
1052 	}
1053 }
1054 #endif
1055