xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision ce48b2100785e5ca629fb3aa8e3b50aca808f692)
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_vsx = NULL;
57 struct task_struct *last_task_used_spe = NULL;
58 #endif
59 
60 /*
61  * Make sure the floating-point register state in the
62  * the thread_struct is up to date for task tsk.
63  */
64 void flush_fp_to_thread(struct task_struct *tsk)
65 {
66 	if (tsk->thread.regs) {
67 		/*
68 		 * We need to disable preemption here because if we didn't,
69 		 * another process could get scheduled after the regs->msr
70 		 * test but before we have finished saving the FP registers
71 		 * to the thread_struct.  That process could take over the
72 		 * FPU, and then when we get scheduled again we would store
73 		 * bogus values for the remaining FP registers.
74 		 */
75 		preempt_disable();
76 		if (tsk->thread.regs->msr & MSR_FP) {
77 #ifdef CONFIG_SMP
78 			/*
79 			 * This should only ever be called for current or
80 			 * for a stopped child process.  Since we save away
81 			 * the FP register state on context switch on SMP,
82 			 * there is something wrong if a stopped child appears
83 			 * to still have its FP state in the CPU registers.
84 			 */
85 			BUG_ON(tsk != current);
86 #endif
87 			giveup_fpu(tsk);
88 		}
89 		preempt_enable();
90 	}
91 }
92 
93 void enable_kernel_fp(void)
94 {
95 	WARN_ON(preemptible());
96 
97 #ifdef CONFIG_SMP
98 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
99 		giveup_fpu(current);
100 	else
101 		giveup_fpu(NULL);	/* just enables FP for kernel */
102 #else
103 	giveup_fpu(last_task_used_math);
104 #endif /* CONFIG_SMP */
105 }
106 EXPORT_SYMBOL(enable_kernel_fp);
107 
108 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
109 {
110 #ifdef CONFIG_VSX
111 	int i;
112 	elf_fpreg_t *reg;
113 #endif
114 
115 	if (!tsk->thread.regs)
116 		return 0;
117 	flush_fp_to_thread(current);
118 
119 #ifdef CONFIG_VSX
120 	reg = (elf_fpreg_t *)fpregs;
121 	for (i = 0; i < ELF_NFPREG - 1; i++, reg++)
122 		*reg = tsk->thread.TS_FPR(i);
123 	memcpy(reg, &tsk->thread.fpscr, sizeof(elf_fpreg_t));
124 #else
125 	memcpy(fpregs, &tsk->thread.TS_FPR(0), sizeof(*fpregs));
126 #endif
127 
128 	return 1;
129 }
130 
131 #ifdef CONFIG_ALTIVEC
132 void enable_kernel_altivec(void)
133 {
134 	WARN_ON(preemptible());
135 
136 #ifdef CONFIG_SMP
137 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
138 		giveup_altivec(current);
139 	else
140 		giveup_altivec(NULL);	/* just enable AltiVec for kernel - force */
141 #else
142 	giveup_altivec(last_task_used_altivec);
143 #endif /* CONFIG_SMP */
144 }
145 EXPORT_SYMBOL(enable_kernel_altivec);
146 
147 /*
148  * Make sure the VMX/Altivec register state in the
149  * the thread_struct is up to date for task tsk.
150  */
151 void flush_altivec_to_thread(struct task_struct *tsk)
152 {
153 	if (tsk->thread.regs) {
154 		preempt_disable();
155 		if (tsk->thread.regs->msr & MSR_VEC) {
156 #ifdef CONFIG_SMP
157 			BUG_ON(tsk != current);
158 #endif
159 			giveup_altivec(tsk);
160 		}
161 		preempt_enable();
162 	}
163 }
164 
165 int dump_task_altivec(struct task_struct *tsk, elf_vrreg_t *vrregs)
166 {
167 	/* ELF_NVRREG includes the VSCR and VRSAVE which we need to save
168 	 * separately, see below */
169 	const int nregs = ELF_NVRREG - 2;
170 	elf_vrreg_t *reg;
171 	u32 *dest;
172 
173 	if (tsk == current)
174 		flush_altivec_to_thread(tsk);
175 
176 	reg = (elf_vrreg_t *)vrregs;
177 
178 	/* copy the 32 vr registers */
179 	memcpy(reg, &tsk->thread.vr[0], nregs * sizeof(*reg));
180 	reg += nregs;
181 
182 	/* copy the vscr */
183 	memcpy(reg, &tsk->thread.vscr, sizeof(*reg));
184 	reg++;
185 
186 	/* vrsave is stored in the high 32bit slot of the final 128bits */
187 	memset(reg, 0, sizeof(*reg));
188 	dest = (u32 *)reg;
189 	*dest = tsk->thread.vrsave;
190 
191 	return 1;
192 }
193 #endif /* CONFIG_ALTIVEC */
194 
195 #ifdef CONFIG_VSX
196 #if 0
197 /* not currently used, but some crazy RAID module might want to later */
198 void enable_kernel_vsx(void)
199 {
200 	WARN_ON(preemptible());
201 
202 #ifdef CONFIG_SMP
203 	if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
204 		giveup_vsx(current);
205 	else
206 		giveup_vsx(NULL);	/* just enable vsx for kernel - force */
207 #else
208 	giveup_vsx(last_task_used_vsx);
209 #endif /* CONFIG_SMP */
210 }
211 EXPORT_SYMBOL(enable_kernel_vsx);
212 #endif
213 
214 void flush_vsx_to_thread(struct task_struct *tsk)
215 {
216 	if (tsk->thread.regs) {
217 		preempt_disable();
218 		if (tsk->thread.regs->msr & MSR_VSX) {
219 #ifdef CONFIG_SMP
220 			BUG_ON(tsk != current);
221 #endif
222 			giveup_vsx(tsk);
223 		}
224 		preempt_enable();
225 	}
226 }
227 
228 /*
229  * This dumps the lower half 64bits of the first 32 VSX registers.
230  * This needs to be called with dump_task_fp and dump_task_altivec to
231  * get all the VSX state.
232  */
233 int dump_task_vsx(struct task_struct *tsk, elf_vrreg_t *vrregs)
234 {
235 	elf_vrreg_t *reg;
236 	double buf[32];
237 	int i;
238 
239 	if (tsk == current)
240 		flush_vsx_to_thread(tsk);
241 
242 	reg = (elf_vrreg_t *)vrregs;
243 
244 	for (i = 0; i < 32 ; i++)
245 		buf[i] = current->thread.fpr[i][TS_VSRLOWOFFSET];
246 	memcpy(reg, buf, sizeof(buf));
247 
248 	return 1;
249 }
250 #endif /* CONFIG_VSX */
251 
252 int dump_task_vector(struct task_struct *tsk, elf_vrregset_t *vrregs)
253 {
254 	int rc = 0;
255 	elf_vrreg_t *regs = (elf_vrreg_t *)vrregs;
256 #ifdef CONFIG_ALTIVEC
257 	rc = dump_task_altivec(tsk, regs);
258 	if (rc)
259 		return rc;
260 	regs += ELF_NVRREG;
261 #endif
262 
263 #ifdef CONFIG_VSX
264 	rc = dump_task_vsx(tsk, regs);
265 #endif
266 	return rc;
267 }
268 
269 #ifdef CONFIG_SPE
270 
271 void enable_kernel_spe(void)
272 {
273 	WARN_ON(preemptible());
274 
275 #ifdef CONFIG_SMP
276 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
277 		giveup_spe(current);
278 	else
279 		giveup_spe(NULL);	/* just enable SPE for kernel - force */
280 #else
281 	giveup_spe(last_task_used_spe);
282 #endif /* __SMP __ */
283 }
284 EXPORT_SYMBOL(enable_kernel_spe);
285 
286 void flush_spe_to_thread(struct task_struct *tsk)
287 {
288 	if (tsk->thread.regs) {
289 		preempt_disable();
290 		if (tsk->thread.regs->msr & MSR_SPE) {
291 #ifdef CONFIG_SMP
292 			BUG_ON(tsk != current);
293 #endif
294 			giveup_spe(tsk);
295 		}
296 		preempt_enable();
297 	}
298 }
299 
300 int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
301 {
302 	flush_spe_to_thread(current);
303 	/* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
304 	memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
305 	return 1;
306 }
307 #endif /* CONFIG_SPE */
308 
309 #ifndef CONFIG_SMP
310 /*
311  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
312  * and the current task has some state, discard it.
313  */
314 void discard_lazy_cpu_state(void)
315 {
316 	preempt_disable();
317 	if (last_task_used_math == current)
318 		last_task_used_math = NULL;
319 #ifdef CONFIG_ALTIVEC
320 	if (last_task_used_altivec == current)
321 		last_task_used_altivec = NULL;
322 #endif /* CONFIG_ALTIVEC */
323 #ifdef CONFIG_VSX
324 	if (last_task_used_vsx == current)
325 		last_task_used_vsx = NULL;
326 #endif /* CONFIG_VSX */
327 #ifdef CONFIG_SPE
328 	if (last_task_used_spe == current)
329 		last_task_used_spe = NULL;
330 #endif
331 	preempt_enable();
332 }
333 #endif /* CONFIG_SMP */
334 
335 static DEFINE_PER_CPU(unsigned long, current_dabr);
336 
337 int set_dabr(unsigned long dabr)
338 {
339 	__get_cpu_var(current_dabr) = dabr;
340 
341 #ifdef CONFIG_PPC_MERGE		/* XXX for now */
342 	if (ppc_md.set_dabr)
343 		return ppc_md.set_dabr(dabr);
344 #endif
345 
346 	/* XXX should we have a CPU_FTR_HAS_DABR ? */
347 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
348 	mtspr(SPRN_DABR, dabr);
349 #endif
350 	return 0;
351 }
352 
353 #ifdef CONFIG_PPC64
354 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
355 #endif
356 
357 struct task_struct *__switch_to(struct task_struct *prev,
358 	struct task_struct *new)
359 {
360 	struct thread_struct *new_thread, *old_thread;
361 	unsigned long flags;
362 	struct task_struct *last;
363 
364 #ifdef CONFIG_SMP
365 	/* avoid complexity of lazy save/restore of fpu
366 	 * by just saving it every time we switch out if
367 	 * this task used the fpu during the last quantum.
368 	 *
369 	 * If it tries to use the fpu again, it'll trap and
370 	 * reload its fp regs.  So we don't have to do a restore
371 	 * every switch, just a save.
372 	 *  -- Cort
373 	 */
374 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
375 		giveup_fpu(prev);
376 #ifdef CONFIG_ALTIVEC
377 	/*
378 	 * If the previous thread used altivec in the last quantum
379 	 * (thus changing altivec regs) then save them.
380 	 * We used to check the VRSAVE register but not all apps
381 	 * set it, so we don't rely on it now (and in fact we need
382 	 * to save & restore VSCR even if VRSAVE == 0).  -- paulus
383 	 *
384 	 * On SMP we always save/restore altivec regs just to avoid the
385 	 * complexity of changing processors.
386 	 *  -- Cort
387 	 */
388 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
389 		giveup_altivec(prev);
390 #endif /* CONFIG_ALTIVEC */
391 #ifdef CONFIG_VSX
392 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
393 		giveup_vsx(prev);
394 #endif /* CONFIG_VSX */
395 #ifdef CONFIG_SPE
396 	/*
397 	 * If the previous thread used spe in the last quantum
398 	 * (thus changing spe regs) then save them.
399 	 *
400 	 * On SMP we always save/restore spe regs just to avoid the
401 	 * complexity of changing processors.
402 	 */
403 	if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
404 		giveup_spe(prev);
405 #endif /* CONFIG_SPE */
406 
407 #else  /* CONFIG_SMP */
408 #ifdef CONFIG_ALTIVEC
409 	/* Avoid the trap.  On smp this this never happens since
410 	 * we don't set last_task_used_altivec -- Cort
411 	 */
412 	if (new->thread.regs && last_task_used_altivec == new)
413 		new->thread.regs->msr |= MSR_VEC;
414 #endif /* CONFIG_ALTIVEC */
415 #ifdef CONFIG_VSX
416 	if (new->thread.regs && last_task_used_vsx == new)
417 		new->thread.regs->msr |= MSR_VSX;
418 #endif /* CONFIG_VSX */
419 #ifdef CONFIG_SPE
420 	/* Avoid the trap.  On smp this this never happens since
421 	 * we don't set last_task_used_spe
422 	 */
423 	if (new->thread.regs && last_task_used_spe == new)
424 		new->thread.regs->msr |= MSR_SPE;
425 #endif /* CONFIG_SPE */
426 
427 #endif /* CONFIG_SMP */
428 
429 	if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
430 		set_dabr(new->thread.dabr);
431 
432 	new_thread = &new->thread;
433 	old_thread = &current->thread;
434 
435 #ifdef CONFIG_PPC64
436 	/*
437 	 * Collect processor utilization data per process
438 	 */
439 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
440 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
441 		long unsigned start_tb, current_tb;
442 		start_tb = old_thread->start_tb;
443 		cu->current_tb = current_tb = mfspr(SPRN_PURR);
444 		old_thread->accum_tb += (current_tb - start_tb);
445 		new_thread->start_tb = current_tb;
446 	}
447 #endif
448 
449 	local_irq_save(flags);
450 
451 	account_system_vtime(current);
452 	account_process_vtime(current);
453 	calculate_steal_time();
454 
455 	/*
456 	 * We can't take a PMU exception inside _switch() since there is a
457 	 * window where the kernel stack SLB and the kernel stack are out
458 	 * of sync. Hard disable here.
459 	 */
460 	hard_irq_disable();
461 	last = _switch(old_thread, new_thread);
462 
463 	local_irq_restore(flags);
464 
465 	return last;
466 }
467 
468 static int instructions_to_print = 16;
469 
470 static void show_instructions(struct pt_regs *regs)
471 {
472 	int i;
473 	unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
474 			sizeof(int));
475 
476 	printk("Instruction dump:");
477 
478 	for (i = 0; i < instructions_to_print; i++) {
479 		int instr;
480 
481 		if (!(i % 8))
482 			printk("\n");
483 
484 #if !defined(CONFIG_BOOKE)
485 		/* If executing with the IMMU off, adjust pc rather
486 		 * than print XXXXXXXX.
487 		 */
488 		if (!(regs->msr & MSR_IR))
489 			pc = (unsigned long)phys_to_virt(pc);
490 #endif
491 
492 		/* We use __get_user here *only* to avoid an OOPS on a
493 		 * bad address because the pc *should* only be a
494 		 * kernel address.
495 		 */
496 		if (!__kernel_text_address(pc) ||
497 		     __get_user(instr, (unsigned int __user *)pc)) {
498 			printk("XXXXXXXX ");
499 		} else {
500 			if (regs->nip == pc)
501 				printk("<%08x> ", instr);
502 			else
503 				printk("%08x ", instr);
504 		}
505 
506 		pc += sizeof(int);
507 	}
508 
509 	printk("\n");
510 }
511 
512 static struct regbit {
513 	unsigned long bit;
514 	const char *name;
515 } msr_bits[] = {
516 	{MSR_EE,	"EE"},
517 	{MSR_PR,	"PR"},
518 	{MSR_FP,	"FP"},
519 	{MSR_VEC,	"VEC"},
520 	{MSR_VSX,	"VSX"},
521 	{MSR_ME,	"ME"},
522 	{MSR_IR,	"IR"},
523 	{MSR_DR,	"DR"},
524 	{0,		NULL}
525 };
526 
527 static void printbits(unsigned long val, struct regbit *bits)
528 {
529 	const char *sep = "";
530 
531 	printk("<");
532 	for (; bits->bit; ++bits)
533 		if (val & bits->bit) {
534 			printk("%s%s", sep, bits->name);
535 			sep = ",";
536 		}
537 	printk(">");
538 }
539 
540 #ifdef CONFIG_PPC64
541 #define REG		"%016lx"
542 #define REGS_PER_LINE	4
543 #define LAST_VOLATILE	13
544 #else
545 #define REG		"%08lx"
546 #define REGS_PER_LINE	8
547 #define LAST_VOLATILE	12
548 #endif
549 
550 void show_regs(struct pt_regs * regs)
551 {
552 	int i, trap;
553 
554 	printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
555 	       regs->nip, regs->link, regs->ctr);
556 	printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
557 	       regs, regs->trap, print_tainted(), init_utsname()->release);
558 	printk("MSR: "REG" ", regs->msr);
559 	printbits(regs->msr, msr_bits);
560 	printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
561 	trap = TRAP(regs);
562 	if (trap == 0x300 || trap == 0x600)
563 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
564 		printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
565 #else
566 		printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
567 #endif
568 	printk("TASK = %p[%d] '%s' THREAD: %p",
569 	       current, task_pid_nr(current), current->comm, task_thread_info(current));
570 
571 #ifdef CONFIG_SMP
572 	printk(" CPU: %d", raw_smp_processor_id());
573 #endif /* CONFIG_SMP */
574 
575 	for (i = 0;  i < 32;  i++) {
576 		if ((i % REGS_PER_LINE) == 0)
577 			printk("\n" KERN_INFO "GPR%02d: ", i);
578 		printk(REG " ", regs->gpr[i]);
579 		if (i == LAST_VOLATILE && !FULL_REGS(regs))
580 			break;
581 	}
582 	printk("\n");
583 #ifdef CONFIG_KALLSYMS
584 	/*
585 	 * Lookup NIP late so we have the best change of getting the
586 	 * above info out without failing
587 	 */
588 	printk("NIP ["REG"] ", regs->nip);
589 	print_symbol("%s\n", regs->nip);
590 	printk("LR ["REG"] ", regs->link);
591 	print_symbol("%s\n", regs->link);
592 #endif
593 	show_stack(current, (unsigned long *) regs->gpr[1]);
594 	if (!user_mode(regs))
595 		show_instructions(regs);
596 }
597 
598 void exit_thread(void)
599 {
600 	discard_lazy_cpu_state();
601 }
602 
603 void flush_thread(void)
604 {
605 #ifdef CONFIG_PPC64
606 	struct thread_info *t = current_thread_info();
607 
608 	if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
609 		clear_ti_thread_flag(t, TIF_ABI_PENDING);
610 		if (test_ti_thread_flag(t, TIF_32BIT))
611 			clear_ti_thread_flag(t, TIF_32BIT);
612 		else
613 			set_ti_thread_flag(t, TIF_32BIT);
614 	}
615 #endif
616 
617 	discard_lazy_cpu_state();
618 
619 	if (current->thread.dabr) {
620 		current->thread.dabr = 0;
621 		set_dabr(0);
622 	}
623 }
624 
625 void
626 release_thread(struct task_struct *t)
627 {
628 }
629 
630 /*
631  * This gets called before we allocate a new thread and copy
632  * the current task into it.
633  */
634 void prepare_to_copy(struct task_struct *tsk)
635 {
636 	flush_fp_to_thread(current);
637 	flush_altivec_to_thread(current);
638 	flush_vsx_to_thread(current);
639 	flush_spe_to_thread(current);
640 }
641 
642 /*
643  * Copy a thread..
644  */
645 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
646 		unsigned long unused, struct task_struct *p,
647 		struct pt_regs *regs)
648 {
649 	struct pt_regs *childregs, *kregs;
650 	extern void ret_from_fork(void);
651 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
652 
653 	CHECK_FULL_REGS(regs);
654 	/* Copy registers */
655 	sp -= sizeof(struct pt_regs);
656 	childregs = (struct pt_regs *) sp;
657 	*childregs = *regs;
658 	if ((childregs->msr & MSR_PR) == 0) {
659 		/* for kernel thread, set `current' and stackptr in new task */
660 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
661 #ifdef CONFIG_PPC32
662 		childregs->gpr[2] = (unsigned long) p;
663 #else
664 		clear_tsk_thread_flag(p, TIF_32BIT);
665 #endif
666 		p->thread.regs = NULL;	/* no user register state */
667 	} else {
668 		childregs->gpr[1] = usp;
669 		p->thread.regs = childregs;
670 		if (clone_flags & CLONE_SETTLS) {
671 #ifdef CONFIG_PPC64
672 			if (!test_thread_flag(TIF_32BIT))
673 				childregs->gpr[13] = childregs->gpr[6];
674 			else
675 #endif
676 				childregs->gpr[2] = childregs->gpr[6];
677 		}
678 	}
679 	childregs->gpr[3] = 0;  /* Result from fork() */
680 	sp -= STACK_FRAME_OVERHEAD;
681 
682 	/*
683 	 * The way this works is that at some point in the future
684 	 * some task will call _switch to switch to the new task.
685 	 * That will pop off the stack frame created below and start
686 	 * the new task running at ret_from_fork.  The new task will
687 	 * do some house keeping and then return from the fork or clone
688 	 * system call, using the stack frame created above.
689 	 */
690 	sp -= sizeof(struct pt_regs);
691 	kregs = (struct pt_regs *) sp;
692 	sp -= STACK_FRAME_OVERHEAD;
693 	p->thread.ksp = sp;
694 	p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
695 				_ALIGN_UP(sizeof(struct thread_info), 16);
696 
697 #ifdef CONFIG_PPC64
698 	if (cpu_has_feature(CPU_FTR_SLB)) {
699 		unsigned long sp_vsid;
700 		unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
701 
702 		if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
703 			sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
704 				<< SLB_VSID_SHIFT_1T;
705 		else
706 			sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
707 				<< SLB_VSID_SHIFT;
708 		sp_vsid |= SLB_VSID_KERNEL | llp;
709 		p->thread.ksp_vsid = sp_vsid;
710 	}
711 
712 	/*
713 	 * The PPC64 ABI makes use of a TOC to contain function
714 	 * pointers.  The function (ret_from_except) is actually a pointer
715 	 * to the TOC entry.  The first entry is a pointer to the actual
716 	 * function.
717  	 */
718 	kregs->nip = *((unsigned long *)ret_from_fork);
719 #else
720 	kregs->nip = (unsigned long)ret_from_fork;
721 #endif
722 
723 	return 0;
724 }
725 
726 /*
727  * Set up a thread for executing a new program
728  */
729 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
730 {
731 #ifdef CONFIG_PPC64
732 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
733 #endif
734 
735 	set_fs(USER_DS);
736 
737 	/*
738 	 * If we exec out of a kernel thread then thread.regs will not be
739 	 * set.  Do it now.
740 	 */
741 	if (!current->thread.regs) {
742 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
743 		current->thread.regs = regs - 1;
744 	}
745 
746 	memset(regs->gpr, 0, sizeof(regs->gpr));
747 	regs->ctr = 0;
748 	regs->link = 0;
749 	regs->xer = 0;
750 	regs->ccr = 0;
751 	regs->gpr[1] = sp;
752 
753 	/*
754 	 * We have just cleared all the nonvolatile GPRs, so make
755 	 * FULL_REGS(regs) return true.  This is necessary to allow
756 	 * ptrace to examine the thread immediately after exec.
757 	 */
758 	regs->trap &= ~1UL;
759 
760 #ifdef CONFIG_PPC32
761 	regs->mq = 0;
762 	regs->nip = start;
763 	regs->msr = MSR_USER;
764 #else
765 	if (!test_thread_flag(TIF_32BIT)) {
766 		unsigned long entry, toc;
767 
768 		/* start is a relocated pointer to the function descriptor for
769 		 * the elf _start routine.  The first entry in the function
770 		 * descriptor is the entry address of _start and the second
771 		 * entry is the TOC value we need to use.
772 		 */
773 		__get_user(entry, (unsigned long __user *)start);
774 		__get_user(toc, (unsigned long __user *)start+1);
775 
776 		/* Check whether the e_entry function descriptor entries
777 		 * need to be relocated before we can use them.
778 		 */
779 		if (load_addr != 0) {
780 			entry += load_addr;
781 			toc   += load_addr;
782 		}
783 		regs->nip = entry;
784 		regs->gpr[2] = toc;
785 		regs->msr = MSR_USER64;
786 	} else {
787 		regs->nip = start;
788 		regs->gpr[2] = 0;
789 		regs->msr = MSR_USER32;
790 	}
791 #endif
792 
793 	discard_lazy_cpu_state();
794 #ifdef CONFIG_VSX
795 	current->thread.used_vsr = 0;
796 #endif
797 	memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
798 	current->thread.fpscr.val = 0;
799 #ifdef CONFIG_ALTIVEC
800 	memset(current->thread.vr, 0, sizeof(current->thread.vr));
801 	memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
802 	current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
803 	current->thread.vrsave = 0;
804 	current->thread.used_vr = 0;
805 #endif /* CONFIG_ALTIVEC */
806 #ifdef CONFIG_SPE
807 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
808 	current->thread.acc = 0;
809 	current->thread.spefscr = 0;
810 	current->thread.used_spe = 0;
811 #endif /* CONFIG_SPE */
812 }
813 
814 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
815 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
816 
817 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
818 {
819 	struct pt_regs *regs = tsk->thread.regs;
820 
821 	/* This is a bit hairy.  If we are an SPE enabled  processor
822 	 * (have embedded fp) we store the IEEE exception enable flags in
823 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
824 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
825 	if (val & PR_FP_EXC_SW_ENABLE) {
826 #ifdef CONFIG_SPE
827 		if (cpu_has_feature(CPU_FTR_SPE)) {
828 			tsk->thread.fpexc_mode = val &
829 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
830 			return 0;
831 		} else {
832 			return -EINVAL;
833 		}
834 #else
835 		return -EINVAL;
836 #endif
837 	}
838 
839 	/* on a CONFIG_SPE this does not hurt us.  The bits that
840 	 * __pack_fe01 use do not overlap with bits used for
841 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
842 	 * on CONFIG_SPE implementations are reserved so writing to
843 	 * them does not change anything */
844 	if (val > PR_FP_EXC_PRECISE)
845 		return -EINVAL;
846 	tsk->thread.fpexc_mode = __pack_fe01(val);
847 	if (regs != NULL && (regs->msr & MSR_FP) != 0)
848 		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
849 			| tsk->thread.fpexc_mode;
850 	return 0;
851 }
852 
853 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
854 {
855 	unsigned int val;
856 
857 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
858 #ifdef CONFIG_SPE
859 		if (cpu_has_feature(CPU_FTR_SPE))
860 			val = tsk->thread.fpexc_mode;
861 		else
862 			return -EINVAL;
863 #else
864 		return -EINVAL;
865 #endif
866 	else
867 		val = __unpack_fe01(tsk->thread.fpexc_mode);
868 	return put_user(val, (unsigned int __user *) adr);
869 }
870 
871 int set_endian(struct task_struct *tsk, unsigned int val)
872 {
873 	struct pt_regs *regs = tsk->thread.regs;
874 
875 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
876 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
877 		return -EINVAL;
878 
879 	if (regs == NULL)
880 		return -EINVAL;
881 
882 	if (val == PR_ENDIAN_BIG)
883 		regs->msr &= ~MSR_LE;
884 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
885 		regs->msr |= MSR_LE;
886 	else
887 		return -EINVAL;
888 
889 	return 0;
890 }
891 
892 int get_endian(struct task_struct *tsk, unsigned long adr)
893 {
894 	struct pt_regs *regs = tsk->thread.regs;
895 	unsigned int val;
896 
897 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
898 	    !cpu_has_feature(CPU_FTR_REAL_LE))
899 		return -EINVAL;
900 
901 	if (regs == NULL)
902 		return -EINVAL;
903 
904 	if (regs->msr & MSR_LE) {
905 		if (cpu_has_feature(CPU_FTR_REAL_LE))
906 			val = PR_ENDIAN_LITTLE;
907 		else
908 			val = PR_ENDIAN_PPC_LITTLE;
909 	} else
910 		val = PR_ENDIAN_BIG;
911 
912 	return put_user(val, (unsigned int __user *)adr);
913 }
914 
915 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
916 {
917 	tsk->thread.align_ctl = val;
918 	return 0;
919 }
920 
921 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
922 {
923 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
924 }
925 
926 #define TRUNC_PTR(x)	((typeof(x))(((unsigned long)(x)) & 0xffffffff))
927 
928 int sys_clone(unsigned long clone_flags, unsigned long usp,
929 	      int __user *parent_tidp, void __user *child_threadptr,
930 	      int __user *child_tidp, int p6,
931 	      struct pt_regs *regs)
932 {
933 	CHECK_FULL_REGS(regs);
934 	if (usp == 0)
935 		usp = regs->gpr[1];	/* stack pointer for child */
936 #ifdef CONFIG_PPC64
937 	if (test_thread_flag(TIF_32BIT)) {
938 		parent_tidp = TRUNC_PTR(parent_tidp);
939 		child_tidp = TRUNC_PTR(child_tidp);
940 	}
941 #endif
942  	return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
943 }
944 
945 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
946 	     unsigned long p4, unsigned long p5, unsigned long p6,
947 	     struct pt_regs *regs)
948 {
949 	CHECK_FULL_REGS(regs);
950 	return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
951 }
952 
953 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
954 	      unsigned long p4, unsigned long p5, unsigned long p6,
955 	      struct pt_regs *regs)
956 {
957 	CHECK_FULL_REGS(regs);
958 	return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
959 			regs, 0, NULL, NULL);
960 }
961 
962 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
963 	       unsigned long a3, unsigned long a4, unsigned long a5,
964 	       struct pt_regs *regs)
965 {
966 	int error;
967 	char *filename;
968 
969 	filename = getname((char __user *) a0);
970 	error = PTR_ERR(filename);
971 	if (IS_ERR(filename))
972 		goto out;
973 	flush_fp_to_thread(current);
974 	flush_altivec_to_thread(current);
975 	flush_spe_to_thread(current);
976 	error = do_execve(filename, (char __user * __user *) a1,
977 			  (char __user * __user *) a2, regs);
978 	putname(filename);
979 out:
980 	return error;
981 }
982 
983 #ifdef CONFIG_IRQSTACKS
984 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
985 				  unsigned long nbytes)
986 {
987 	unsigned long stack_page;
988 	unsigned long cpu = task_cpu(p);
989 
990 	/*
991 	 * Avoid crashing if the stack has overflowed and corrupted
992 	 * task_cpu(p), which is in the thread_info struct.
993 	 */
994 	if (cpu < NR_CPUS && cpu_possible(cpu)) {
995 		stack_page = (unsigned long) hardirq_ctx[cpu];
996 		if (sp >= stack_page + sizeof(struct thread_struct)
997 		    && sp <= stack_page + THREAD_SIZE - nbytes)
998 			return 1;
999 
1000 		stack_page = (unsigned long) softirq_ctx[cpu];
1001 		if (sp >= stack_page + sizeof(struct thread_struct)
1002 		    && sp <= stack_page + THREAD_SIZE - nbytes)
1003 			return 1;
1004 	}
1005 	return 0;
1006 }
1007 
1008 #else
1009 #define valid_irq_stack(sp, p, nb)	0
1010 #endif /* CONFIG_IRQSTACKS */
1011 
1012 int validate_sp(unsigned long sp, struct task_struct *p,
1013 		       unsigned long nbytes)
1014 {
1015 	unsigned long stack_page = (unsigned long)task_stack_page(p);
1016 
1017 	if (sp >= stack_page + sizeof(struct thread_struct)
1018 	    && sp <= stack_page + THREAD_SIZE - nbytes)
1019 		return 1;
1020 
1021 	return valid_irq_stack(sp, p, nbytes);
1022 }
1023 
1024 EXPORT_SYMBOL(validate_sp);
1025 
1026 unsigned long get_wchan(struct task_struct *p)
1027 {
1028 	unsigned long ip, sp;
1029 	int count = 0;
1030 
1031 	if (!p || p == current || p->state == TASK_RUNNING)
1032 		return 0;
1033 
1034 	sp = p->thread.ksp;
1035 	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1036 		return 0;
1037 
1038 	do {
1039 		sp = *(unsigned long *)sp;
1040 		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1041 			return 0;
1042 		if (count > 0) {
1043 			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1044 			if (!in_sched_functions(ip))
1045 				return ip;
1046 		}
1047 	} while (count++ < 16);
1048 	return 0;
1049 }
1050 
1051 static int kstack_depth_to_print = 64;
1052 
1053 void show_stack(struct task_struct *tsk, unsigned long *stack)
1054 {
1055 	unsigned long sp, ip, lr, newsp;
1056 	int count = 0;
1057 	int firstframe = 1;
1058 
1059 	sp = (unsigned long) stack;
1060 	if (tsk == NULL)
1061 		tsk = current;
1062 	if (sp == 0) {
1063 		if (tsk == current)
1064 			asm("mr %0,1" : "=r" (sp));
1065 		else
1066 			sp = tsk->thread.ksp;
1067 	}
1068 
1069 	lr = 0;
1070 	printk("Call Trace:\n");
1071 	do {
1072 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1073 			return;
1074 
1075 		stack = (unsigned long *) sp;
1076 		newsp = stack[0];
1077 		ip = stack[STACK_FRAME_LR_SAVE];
1078 		if (!firstframe || ip != lr) {
1079 			printk("["REG"] ["REG"] ", sp, ip);
1080 			print_symbol("%s", ip);
1081 			if (firstframe)
1082 				printk(" (unreliable)");
1083 			printk("\n");
1084 		}
1085 		firstframe = 0;
1086 
1087 		/*
1088 		 * See if this is an exception frame.
1089 		 * We look for the "regshere" marker in the current frame.
1090 		 */
1091 		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1092 		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1093 			struct pt_regs *regs = (struct pt_regs *)
1094 				(sp + STACK_FRAME_OVERHEAD);
1095 			printk("--- Exception: %lx", regs->trap);
1096 			print_symbol(" at %s\n", regs->nip);
1097 			lr = regs->link;
1098 			print_symbol("    LR = %s\n", lr);
1099 			firstframe = 1;
1100 		}
1101 
1102 		sp = newsp;
1103 	} while (count++ < kstack_depth_to_print);
1104 }
1105 
1106 void dump_stack(void)
1107 {
1108 	show_stack(current, NULL);
1109 }
1110 EXPORT_SYMBOL(dump_stack);
1111 
1112 #ifdef CONFIG_PPC64
1113 void ppc64_runlatch_on(void)
1114 {
1115 	unsigned long ctrl;
1116 
1117 	if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1118 		HMT_medium();
1119 
1120 		ctrl = mfspr(SPRN_CTRLF);
1121 		ctrl |= CTRL_RUNLATCH;
1122 		mtspr(SPRN_CTRLT, ctrl);
1123 
1124 		set_thread_flag(TIF_RUNLATCH);
1125 	}
1126 }
1127 
1128 void ppc64_runlatch_off(void)
1129 {
1130 	unsigned long ctrl;
1131 
1132 	if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1133 		HMT_medium();
1134 
1135 		clear_thread_flag(TIF_RUNLATCH);
1136 
1137 		ctrl = mfspr(SPRN_CTRLF);
1138 		ctrl &= ~CTRL_RUNLATCH;
1139 		mtspr(SPRN_CTRLT, ctrl);
1140 	}
1141 }
1142 #endif
1143 
1144 #if THREAD_SHIFT < PAGE_SHIFT
1145 
1146 static struct kmem_cache *thread_info_cache;
1147 
1148 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1149 {
1150 	struct thread_info *ti;
1151 
1152 	ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1153 	if (unlikely(ti == NULL))
1154 		return NULL;
1155 #ifdef CONFIG_DEBUG_STACK_USAGE
1156 	memset(ti, 0, THREAD_SIZE);
1157 #endif
1158 	return ti;
1159 }
1160 
1161 void free_thread_info(struct thread_info *ti)
1162 {
1163 	kmem_cache_free(thread_info_cache, ti);
1164 }
1165 
1166 void thread_info_cache_init(void)
1167 {
1168 	thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1169 					      THREAD_SIZE, 0, NULL);
1170 	BUG_ON(thread_info_cache == NULL);
1171 }
1172 
1173 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1174