xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision 9cfc5c90)
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/prctl.h>
29 #include <linux/init_task.h>
30 #include <linux/export.h>
31 #include <linux/kallsyms.h>
32 #include <linux/mqueue.h>
33 #include <linux/hardirq.h>
34 #include <linux/utsname.h>
35 #include <linux/ftrace.h>
36 #include <linux/kernel_stat.h>
37 #include <linux/personality.h>
38 #include <linux/random.h>
39 #include <linux/hw_breakpoint.h>
40 #include <linux/uaccess.h>
41 
42 #include <asm/pgtable.h>
43 #include <asm/io.h>
44 #include <asm/processor.h>
45 #include <asm/mmu.h>
46 #include <asm/prom.h>
47 #include <asm/machdep.h>
48 #include <asm/time.h>
49 #include <asm/runlatch.h>
50 #include <asm/syscalls.h>
51 #include <asm/switch_to.h>
52 #include <asm/tm.h>
53 #include <asm/debug.h>
54 #ifdef CONFIG_PPC64
55 #include <asm/firmware.h>
56 #endif
57 #include <asm/code-patching.h>
58 #include <linux/kprobes.h>
59 #include <linux/kdebug.h>
60 
61 /* Transactional Memory debug */
62 #ifdef TM_DEBUG_SW
63 #define TM_DEBUG(x...) printk(KERN_INFO x)
64 #else
65 #define TM_DEBUG(x...) do { } while(0)
66 #endif
67 
68 extern unsigned long _get_SP(void);
69 
70 #ifndef CONFIG_SMP
71 struct task_struct *last_task_used_math = NULL;
72 struct task_struct *last_task_used_altivec = NULL;
73 struct task_struct *last_task_used_vsx = NULL;
74 struct task_struct *last_task_used_spe = NULL;
75 #endif
76 
77 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
78 void giveup_fpu_maybe_transactional(struct task_struct *tsk)
79 {
80 	/*
81 	 * If we are saving the current thread's registers, and the
82 	 * thread is in a transactional state, set the TIF_RESTORE_TM
83 	 * bit so that we know to restore the registers before
84 	 * returning to userspace.
85 	 */
86 	if (tsk == current && tsk->thread.regs &&
87 	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
88 	    !test_thread_flag(TIF_RESTORE_TM)) {
89 		tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
90 		set_thread_flag(TIF_RESTORE_TM);
91 	}
92 
93 	giveup_fpu(tsk);
94 }
95 
96 void giveup_altivec_maybe_transactional(struct task_struct *tsk)
97 {
98 	/*
99 	 * If we are saving the current thread's registers, and the
100 	 * thread is in a transactional state, set the TIF_RESTORE_TM
101 	 * bit so that we know to restore the registers before
102 	 * returning to userspace.
103 	 */
104 	if (tsk == current && tsk->thread.regs &&
105 	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
106 	    !test_thread_flag(TIF_RESTORE_TM)) {
107 		tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
108 		set_thread_flag(TIF_RESTORE_TM);
109 	}
110 
111 	giveup_altivec(tsk);
112 }
113 
114 #else
115 #define giveup_fpu_maybe_transactional(tsk)	giveup_fpu(tsk)
116 #define giveup_altivec_maybe_transactional(tsk)	giveup_altivec(tsk)
117 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
118 
119 #ifdef CONFIG_PPC_FPU
120 /*
121  * Make sure the floating-point register state in the
122  * the thread_struct is up to date for task tsk.
123  */
124 void flush_fp_to_thread(struct task_struct *tsk)
125 {
126 	if (tsk->thread.regs) {
127 		/*
128 		 * We need to disable preemption here because if we didn't,
129 		 * another process could get scheduled after the regs->msr
130 		 * test but before we have finished saving the FP registers
131 		 * to the thread_struct.  That process could take over the
132 		 * FPU, and then when we get scheduled again we would store
133 		 * bogus values for the remaining FP registers.
134 		 */
135 		preempt_disable();
136 		if (tsk->thread.regs->msr & MSR_FP) {
137 #ifdef CONFIG_SMP
138 			/*
139 			 * This should only ever be called for current or
140 			 * for a stopped child process.  Since we save away
141 			 * the FP register state on context switch on SMP,
142 			 * there is something wrong if a stopped child appears
143 			 * to still have its FP state in the CPU registers.
144 			 */
145 			BUG_ON(tsk != current);
146 #endif
147 			giveup_fpu_maybe_transactional(tsk);
148 		}
149 		preempt_enable();
150 	}
151 }
152 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
153 #endif /* CONFIG_PPC_FPU */
154 
155 void enable_kernel_fp(void)
156 {
157 	WARN_ON(preemptible());
158 
159 #ifdef CONFIG_SMP
160 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
161 		giveup_fpu_maybe_transactional(current);
162 	else
163 		giveup_fpu(NULL);	/* just enables FP for kernel */
164 #else
165 	giveup_fpu_maybe_transactional(last_task_used_math);
166 #endif /* CONFIG_SMP */
167 }
168 EXPORT_SYMBOL(enable_kernel_fp);
169 
170 #ifdef CONFIG_ALTIVEC
171 void enable_kernel_altivec(void)
172 {
173 	WARN_ON(preemptible());
174 
175 #ifdef CONFIG_SMP
176 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
177 		giveup_altivec_maybe_transactional(current);
178 	else
179 		giveup_altivec_notask();
180 #else
181 	giveup_altivec_maybe_transactional(last_task_used_altivec);
182 #endif /* CONFIG_SMP */
183 }
184 EXPORT_SYMBOL(enable_kernel_altivec);
185 
186 /*
187  * Make sure the VMX/Altivec register state in the
188  * the thread_struct is up to date for task tsk.
189  */
190 void flush_altivec_to_thread(struct task_struct *tsk)
191 {
192 	if (tsk->thread.regs) {
193 		preempt_disable();
194 		if (tsk->thread.regs->msr & MSR_VEC) {
195 #ifdef CONFIG_SMP
196 			BUG_ON(tsk != current);
197 #endif
198 			giveup_altivec_maybe_transactional(tsk);
199 		}
200 		preempt_enable();
201 	}
202 }
203 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
204 #endif /* CONFIG_ALTIVEC */
205 
206 #ifdef CONFIG_VSX
207 void enable_kernel_vsx(void)
208 {
209 	WARN_ON(preemptible());
210 
211 #ifdef CONFIG_SMP
212 	if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
213 		giveup_vsx(current);
214 	else
215 		giveup_vsx(NULL);	/* just enable vsx for kernel - force */
216 #else
217 	giveup_vsx(last_task_used_vsx);
218 #endif /* CONFIG_SMP */
219 }
220 EXPORT_SYMBOL(enable_kernel_vsx);
221 
222 void giveup_vsx(struct task_struct *tsk)
223 {
224 	giveup_fpu_maybe_transactional(tsk);
225 	giveup_altivec_maybe_transactional(tsk);
226 	__giveup_vsx(tsk);
227 }
228 EXPORT_SYMBOL(giveup_vsx);
229 
230 void flush_vsx_to_thread(struct task_struct *tsk)
231 {
232 	if (tsk->thread.regs) {
233 		preempt_disable();
234 		if (tsk->thread.regs->msr & MSR_VSX) {
235 #ifdef CONFIG_SMP
236 			BUG_ON(tsk != current);
237 #endif
238 			giveup_vsx(tsk);
239 		}
240 		preempt_enable();
241 	}
242 }
243 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
244 #endif /* CONFIG_VSX */
245 
246 #ifdef CONFIG_SPE
247 
248 void enable_kernel_spe(void)
249 {
250 	WARN_ON(preemptible());
251 
252 #ifdef CONFIG_SMP
253 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
254 		giveup_spe(current);
255 	else
256 		giveup_spe(NULL);	/* just enable SPE for kernel - force */
257 #else
258 	giveup_spe(last_task_used_spe);
259 #endif /* __SMP __ */
260 }
261 EXPORT_SYMBOL(enable_kernel_spe);
262 
263 void flush_spe_to_thread(struct task_struct *tsk)
264 {
265 	if (tsk->thread.regs) {
266 		preempt_disable();
267 		if (tsk->thread.regs->msr & MSR_SPE) {
268 #ifdef CONFIG_SMP
269 			BUG_ON(tsk != current);
270 #endif
271 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
272 			giveup_spe(tsk);
273 		}
274 		preempt_enable();
275 	}
276 }
277 #endif /* CONFIG_SPE */
278 
279 #ifndef CONFIG_SMP
280 /*
281  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
282  * and the current task has some state, discard it.
283  */
284 void discard_lazy_cpu_state(void)
285 {
286 	preempt_disable();
287 	if (last_task_used_math == current)
288 		last_task_used_math = NULL;
289 #ifdef CONFIG_ALTIVEC
290 	if (last_task_used_altivec == current)
291 		last_task_used_altivec = NULL;
292 #endif /* CONFIG_ALTIVEC */
293 #ifdef CONFIG_VSX
294 	if (last_task_used_vsx == current)
295 		last_task_used_vsx = NULL;
296 #endif /* CONFIG_VSX */
297 #ifdef CONFIG_SPE
298 	if (last_task_used_spe == current)
299 		last_task_used_spe = NULL;
300 #endif
301 	preempt_enable();
302 }
303 #endif /* CONFIG_SMP */
304 
305 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
306 void do_send_trap(struct pt_regs *regs, unsigned long address,
307 		  unsigned long error_code, int signal_code, int breakpt)
308 {
309 	siginfo_t info;
310 
311 	current->thread.trap_nr = signal_code;
312 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
313 			11, SIGSEGV) == NOTIFY_STOP)
314 		return;
315 
316 	/* Deliver the signal to userspace */
317 	info.si_signo = SIGTRAP;
318 	info.si_errno = breakpt;	/* breakpoint or watchpoint id */
319 	info.si_code = signal_code;
320 	info.si_addr = (void __user *)address;
321 	force_sig_info(SIGTRAP, &info, current);
322 }
323 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
324 void do_break (struct pt_regs *regs, unsigned long address,
325 		    unsigned long error_code)
326 {
327 	siginfo_t info;
328 
329 	current->thread.trap_nr = TRAP_HWBKPT;
330 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
331 			11, SIGSEGV) == NOTIFY_STOP)
332 		return;
333 
334 	if (debugger_break_match(regs))
335 		return;
336 
337 	/* Clear the breakpoint */
338 	hw_breakpoint_disable();
339 
340 	/* Deliver the signal to userspace */
341 	info.si_signo = SIGTRAP;
342 	info.si_errno = 0;
343 	info.si_code = TRAP_HWBKPT;
344 	info.si_addr = (void __user *)address;
345 	force_sig_info(SIGTRAP, &info, current);
346 }
347 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
348 
349 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
350 
351 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
352 /*
353  * Set the debug registers back to their default "safe" values.
354  */
355 static void set_debug_reg_defaults(struct thread_struct *thread)
356 {
357 	thread->debug.iac1 = thread->debug.iac2 = 0;
358 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
359 	thread->debug.iac3 = thread->debug.iac4 = 0;
360 #endif
361 	thread->debug.dac1 = thread->debug.dac2 = 0;
362 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
363 	thread->debug.dvc1 = thread->debug.dvc2 = 0;
364 #endif
365 	thread->debug.dbcr0 = 0;
366 #ifdef CONFIG_BOOKE
367 	/*
368 	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
369 	 */
370 	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
371 			DBCR1_IAC3US | DBCR1_IAC4US;
372 	/*
373 	 * Force Data Address Compare User/Supervisor bits to be User-only
374 	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
375 	 */
376 	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
377 #else
378 	thread->debug.dbcr1 = 0;
379 #endif
380 }
381 
382 static void prime_debug_regs(struct debug_reg *debug)
383 {
384 	/*
385 	 * We could have inherited MSR_DE from userspace, since
386 	 * it doesn't get cleared on exception entry.  Make sure
387 	 * MSR_DE is clear before we enable any debug events.
388 	 */
389 	mtmsr(mfmsr() & ~MSR_DE);
390 
391 	mtspr(SPRN_IAC1, debug->iac1);
392 	mtspr(SPRN_IAC2, debug->iac2);
393 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
394 	mtspr(SPRN_IAC3, debug->iac3);
395 	mtspr(SPRN_IAC4, debug->iac4);
396 #endif
397 	mtspr(SPRN_DAC1, debug->dac1);
398 	mtspr(SPRN_DAC2, debug->dac2);
399 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
400 	mtspr(SPRN_DVC1, debug->dvc1);
401 	mtspr(SPRN_DVC2, debug->dvc2);
402 #endif
403 	mtspr(SPRN_DBCR0, debug->dbcr0);
404 	mtspr(SPRN_DBCR1, debug->dbcr1);
405 #ifdef CONFIG_BOOKE
406 	mtspr(SPRN_DBCR2, debug->dbcr2);
407 #endif
408 }
409 /*
410  * Unless neither the old or new thread are making use of the
411  * debug registers, set the debug registers from the values
412  * stored in the new thread.
413  */
414 void switch_booke_debug_regs(struct debug_reg *new_debug)
415 {
416 	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
417 		|| (new_debug->dbcr0 & DBCR0_IDM))
418 			prime_debug_regs(new_debug);
419 }
420 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
421 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
422 #ifndef CONFIG_HAVE_HW_BREAKPOINT
423 static void set_debug_reg_defaults(struct thread_struct *thread)
424 {
425 	thread->hw_brk.address = 0;
426 	thread->hw_brk.type = 0;
427 	set_breakpoint(&thread->hw_brk);
428 }
429 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
430 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
431 
432 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
433 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
434 {
435 	mtspr(SPRN_DAC1, dabr);
436 #ifdef CONFIG_PPC_47x
437 	isync();
438 #endif
439 	return 0;
440 }
441 #elif defined(CONFIG_PPC_BOOK3S)
442 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
443 {
444 	mtspr(SPRN_DABR, dabr);
445 	if (cpu_has_feature(CPU_FTR_DABRX))
446 		mtspr(SPRN_DABRX, dabrx);
447 	return 0;
448 }
449 #else
450 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
451 {
452 	return -EINVAL;
453 }
454 #endif
455 
456 static inline int set_dabr(struct arch_hw_breakpoint *brk)
457 {
458 	unsigned long dabr, dabrx;
459 
460 	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
461 	dabrx = ((brk->type >> 3) & 0x7);
462 
463 	if (ppc_md.set_dabr)
464 		return ppc_md.set_dabr(dabr, dabrx);
465 
466 	return __set_dabr(dabr, dabrx);
467 }
468 
469 static inline int set_dawr(struct arch_hw_breakpoint *brk)
470 {
471 	unsigned long dawr, dawrx, mrd;
472 
473 	dawr = brk->address;
474 
475 	dawrx  = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
476 		                   << (63 - 58); //* read/write bits */
477 	dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
478 		                   << (63 - 59); //* translate */
479 	dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
480 		                   >> 3; //* PRIM bits */
481 	/* dawr length is stored in field MDR bits 48:53.  Matches range in
482 	   doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
483 	   0b111111=64DW.
484 	   brk->len is in bytes.
485 	   This aligns up to double word size, shifts and does the bias.
486 	*/
487 	mrd = ((brk->len + 7) >> 3) - 1;
488 	dawrx |= (mrd & 0x3f) << (63 - 53);
489 
490 	if (ppc_md.set_dawr)
491 		return ppc_md.set_dawr(dawr, dawrx);
492 	mtspr(SPRN_DAWR, dawr);
493 	mtspr(SPRN_DAWRX, dawrx);
494 	return 0;
495 }
496 
497 void __set_breakpoint(struct arch_hw_breakpoint *brk)
498 {
499 	memcpy(this_cpu_ptr(&current_brk), brk, sizeof(*brk));
500 
501 	if (cpu_has_feature(CPU_FTR_DAWR))
502 		set_dawr(brk);
503 	else
504 		set_dabr(brk);
505 }
506 
507 void set_breakpoint(struct arch_hw_breakpoint *brk)
508 {
509 	preempt_disable();
510 	__set_breakpoint(brk);
511 	preempt_enable();
512 }
513 
514 #ifdef CONFIG_PPC64
515 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
516 #endif
517 
518 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
519 			      struct arch_hw_breakpoint *b)
520 {
521 	if (a->address != b->address)
522 		return false;
523 	if (a->type != b->type)
524 		return false;
525 	if (a->len != b->len)
526 		return false;
527 	return true;
528 }
529 
530 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
531 static void tm_reclaim_thread(struct thread_struct *thr,
532 			      struct thread_info *ti, uint8_t cause)
533 {
534 	unsigned long msr_diff = 0;
535 
536 	/*
537 	 * If FP/VSX registers have been already saved to the
538 	 * thread_struct, move them to the transact_fp array.
539 	 * We clear the TIF_RESTORE_TM bit since after the reclaim
540 	 * the thread will no longer be transactional.
541 	 */
542 	if (test_ti_thread_flag(ti, TIF_RESTORE_TM)) {
543 		msr_diff = thr->ckpt_regs.msr & ~thr->regs->msr;
544 		if (msr_diff & MSR_FP)
545 			memcpy(&thr->transact_fp, &thr->fp_state,
546 			       sizeof(struct thread_fp_state));
547 		if (msr_diff & MSR_VEC)
548 			memcpy(&thr->transact_vr, &thr->vr_state,
549 			       sizeof(struct thread_vr_state));
550 		clear_ti_thread_flag(ti, TIF_RESTORE_TM);
551 		msr_diff &= MSR_FP | MSR_VEC | MSR_VSX | MSR_FE0 | MSR_FE1;
552 	}
553 
554 	tm_reclaim(thr, thr->regs->msr, cause);
555 
556 	/* Having done the reclaim, we now have the checkpointed
557 	 * FP/VSX values in the registers.  These might be valid
558 	 * even if we have previously called enable_kernel_fp() or
559 	 * flush_fp_to_thread(), so update thr->regs->msr to
560 	 * indicate their current validity.
561 	 */
562 	thr->regs->msr |= msr_diff;
563 }
564 
565 void tm_reclaim_current(uint8_t cause)
566 {
567 	tm_enable();
568 	tm_reclaim_thread(&current->thread, current_thread_info(), cause);
569 }
570 
571 static inline void tm_reclaim_task(struct task_struct *tsk)
572 {
573 	/* We have to work out if we're switching from/to a task that's in the
574 	 * middle of a transaction.
575 	 *
576 	 * In switching we need to maintain a 2nd register state as
577 	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
578 	 * checkpointed (tbegin) state in ckpt_regs and saves the transactional
579 	 * (current) FPRs into oldtask->thread.transact_fpr[].
580 	 *
581 	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
582 	 */
583 	struct thread_struct *thr = &tsk->thread;
584 
585 	if (!thr->regs)
586 		return;
587 
588 	if (!MSR_TM_ACTIVE(thr->regs->msr))
589 		goto out_and_saveregs;
590 
591 	/* Stash the original thread MSR, as giveup_fpu et al will
592 	 * modify it.  We hold onto it to see whether the task used
593 	 * FP & vector regs.  If the TIF_RESTORE_TM flag is set,
594 	 * ckpt_regs.msr is already set.
595 	 */
596 	if (!test_ti_thread_flag(task_thread_info(tsk), TIF_RESTORE_TM))
597 		thr->ckpt_regs.msr = thr->regs->msr;
598 
599 	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
600 		 "ccr=%lx, msr=%lx, trap=%lx)\n",
601 		 tsk->pid, thr->regs->nip,
602 		 thr->regs->ccr, thr->regs->msr,
603 		 thr->regs->trap);
604 
605 	tm_reclaim_thread(thr, task_thread_info(tsk), TM_CAUSE_RESCHED);
606 
607 	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
608 		 tsk->pid);
609 
610 out_and_saveregs:
611 	/* Always save the regs here, even if a transaction's not active.
612 	 * This context-switches a thread's TM info SPRs.  We do it here to
613 	 * be consistent with the restore path (in recheckpoint) which
614 	 * cannot happen later in _switch().
615 	 */
616 	tm_save_sprs(thr);
617 }
618 
619 extern void __tm_recheckpoint(struct thread_struct *thread,
620 			      unsigned long orig_msr);
621 
622 void tm_recheckpoint(struct thread_struct *thread,
623 		     unsigned long orig_msr)
624 {
625 	unsigned long flags;
626 
627 	/* We really can't be interrupted here as the TEXASR registers can't
628 	 * change and later in the trecheckpoint code, we have a userspace R1.
629 	 * So let's hard disable over this region.
630 	 */
631 	local_irq_save(flags);
632 	hard_irq_disable();
633 
634 	/* The TM SPRs are restored here, so that TEXASR.FS can be set
635 	 * before the trecheckpoint and no explosion occurs.
636 	 */
637 	tm_restore_sprs(thread);
638 
639 	__tm_recheckpoint(thread, orig_msr);
640 
641 	local_irq_restore(flags);
642 }
643 
644 static inline void tm_recheckpoint_new_task(struct task_struct *new)
645 {
646 	unsigned long msr;
647 
648 	if (!cpu_has_feature(CPU_FTR_TM))
649 		return;
650 
651 	/* Recheckpoint the registers of the thread we're about to switch to.
652 	 *
653 	 * If the task was using FP, we non-lazily reload both the original and
654 	 * the speculative FP register states.  This is because the kernel
655 	 * doesn't see if/when a TM rollback occurs, so if we take an FP
656 	 * unavoidable later, we are unable to determine which set of FP regs
657 	 * need to be restored.
658 	 */
659 	if (!new->thread.regs)
660 		return;
661 
662 	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
663 		tm_restore_sprs(&new->thread);
664 		return;
665 	}
666 	msr = new->thread.ckpt_regs.msr;
667 	/* Recheckpoint to restore original checkpointed register state. */
668 	TM_DEBUG("*** tm_recheckpoint of pid %d "
669 		 "(new->msr 0x%lx, new->origmsr 0x%lx)\n",
670 		 new->pid, new->thread.regs->msr, msr);
671 
672 	/* This loads the checkpointed FP/VEC state, if used */
673 	tm_recheckpoint(&new->thread, msr);
674 
675 	/* This loads the speculative FP/VEC state, if used */
676 	if (msr & MSR_FP) {
677 		do_load_up_transact_fpu(&new->thread);
678 		new->thread.regs->msr |=
679 			(MSR_FP | new->thread.fpexc_mode);
680 	}
681 #ifdef CONFIG_ALTIVEC
682 	if (msr & MSR_VEC) {
683 		do_load_up_transact_altivec(&new->thread);
684 		new->thread.regs->msr |= MSR_VEC;
685 	}
686 #endif
687 	/* We may as well turn on VSX too since all the state is restored now */
688 	if (msr & MSR_VSX)
689 		new->thread.regs->msr |= MSR_VSX;
690 
691 	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
692 		 "(kernel msr 0x%lx)\n",
693 		 new->pid, mfmsr());
694 }
695 
696 static inline void __switch_to_tm(struct task_struct *prev)
697 {
698 	if (cpu_has_feature(CPU_FTR_TM)) {
699 		tm_enable();
700 		tm_reclaim_task(prev);
701 	}
702 }
703 
704 /*
705  * This is called if we are on the way out to userspace and the
706  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
707  * FP and/or vector state and does so if necessary.
708  * If userspace is inside a transaction (whether active or
709  * suspended) and FP/VMX/VSX instructions have ever been enabled
710  * inside that transaction, then we have to keep them enabled
711  * and keep the FP/VMX/VSX state loaded while ever the transaction
712  * continues.  The reason is that if we didn't, and subsequently
713  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
714  * we don't know whether it's the same transaction, and thus we
715  * don't know which of the checkpointed state and the transactional
716  * state to use.
717  */
718 void restore_tm_state(struct pt_regs *regs)
719 {
720 	unsigned long msr_diff;
721 
722 	clear_thread_flag(TIF_RESTORE_TM);
723 	if (!MSR_TM_ACTIVE(regs->msr))
724 		return;
725 
726 	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
727 	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
728 	if (msr_diff & MSR_FP) {
729 		fp_enable();
730 		load_fp_state(&current->thread.fp_state);
731 		regs->msr |= current->thread.fpexc_mode;
732 	}
733 	if (msr_diff & MSR_VEC) {
734 		vec_enable();
735 		load_vr_state(&current->thread.vr_state);
736 	}
737 	regs->msr |= msr_diff;
738 }
739 
740 #else
741 #define tm_recheckpoint_new_task(new)
742 #define __switch_to_tm(prev)
743 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
744 
745 struct task_struct *__switch_to(struct task_struct *prev,
746 	struct task_struct *new)
747 {
748 	struct thread_struct *new_thread, *old_thread;
749 	struct task_struct *last;
750 #ifdef CONFIG_PPC_BOOK3S_64
751 	struct ppc64_tlb_batch *batch;
752 #endif
753 
754 	WARN_ON(!irqs_disabled());
755 
756 	/* Back up the TAR and DSCR across context switches.
757 	 * Note that the TAR is not available for use in the kernel.  (To
758 	 * provide this, the TAR should be backed up/restored on exception
759 	 * entry/exit instead, and be in pt_regs.  FIXME, this should be in
760 	 * pt_regs anyway (for debug).)
761 	 * Save the TAR and DSCR here before we do treclaim/trecheckpoint as
762 	 * these will change them.
763 	 */
764 	save_early_sprs(&prev->thread);
765 
766 	__switch_to_tm(prev);
767 
768 #ifdef CONFIG_SMP
769 	/* avoid complexity of lazy save/restore of fpu
770 	 * by just saving it every time we switch out if
771 	 * this task used the fpu during the last quantum.
772 	 *
773 	 * If it tries to use the fpu again, it'll trap and
774 	 * reload its fp regs.  So we don't have to do a restore
775 	 * every switch, just a save.
776 	 *  -- Cort
777 	 */
778 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
779 		giveup_fpu(prev);
780 #ifdef CONFIG_ALTIVEC
781 	/*
782 	 * If the previous thread used altivec in the last quantum
783 	 * (thus changing altivec regs) then save them.
784 	 * We used to check the VRSAVE register but not all apps
785 	 * set it, so we don't rely on it now (and in fact we need
786 	 * to save & restore VSCR even if VRSAVE == 0).  -- paulus
787 	 *
788 	 * On SMP we always save/restore altivec regs just to avoid the
789 	 * complexity of changing processors.
790 	 *  -- Cort
791 	 */
792 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
793 		giveup_altivec(prev);
794 #endif /* CONFIG_ALTIVEC */
795 #ifdef CONFIG_VSX
796 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
797 		/* VMX and FPU registers are already save here */
798 		__giveup_vsx(prev);
799 #endif /* CONFIG_VSX */
800 #ifdef CONFIG_SPE
801 	/*
802 	 * If the previous thread used spe in the last quantum
803 	 * (thus changing spe regs) then save them.
804 	 *
805 	 * On SMP we always save/restore spe regs just to avoid the
806 	 * complexity of changing processors.
807 	 */
808 	if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
809 		giveup_spe(prev);
810 #endif /* CONFIG_SPE */
811 
812 #else  /* CONFIG_SMP */
813 #ifdef CONFIG_ALTIVEC
814 	/* Avoid the trap.  On smp this this never happens since
815 	 * we don't set last_task_used_altivec -- Cort
816 	 */
817 	if (new->thread.regs && last_task_used_altivec == new)
818 		new->thread.regs->msr |= MSR_VEC;
819 #endif /* CONFIG_ALTIVEC */
820 #ifdef CONFIG_VSX
821 	if (new->thread.regs && last_task_used_vsx == new)
822 		new->thread.regs->msr |= MSR_VSX;
823 #endif /* CONFIG_VSX */
824 #ifdef CONFIG_SPE
825 	/* Avoid the trap.  On smp this this never happens since
826 	 * we don't set last_task_used_spe
827 	 */
828 	if (new->thread.regs && last_task_used_spe == new)
829 		new->thread.regs->msr |= MSR_SPE;
830 #endif /* CONFIG_SPE */
831 
832 #endif /* CONFIG_SMP */
833 
834 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
835 	switch_booke_debug_regs(&new->thread.debug);
836 #else
837 /*
838  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
839  * schedule DABR
840  */
841 #ifndef CONFIG_HAVE_HW_BREAKPOINT
842 	if (unlikely(!hw_brk_match(this_cpu_ptr(&current_brk), &new->thread.hw_brk)))
843 		__set_breakpoint(&new->thread.hw_brk);
844 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
845 #endif
846 
847 
848 	new_thread = &new->thread;
849 	old_thread = &current->thread;
850 
851 #ifdef CONFIG_PPC64
852 	/*
853 	 * Collect processor utilization data per process
854 	 */
855 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
856 		struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
857 		long unsigned start_tb, current_tb;
858 		start_tb = old_thread->start_tb;
859 		cu->current_tb = current_tb = mfspr(SPRN_PURR);
860 		old_thread->accum_tb += (current_tb - start_tb);
861 		new_thread->start_tb = current_tb;
862 	}
863 #endif /* CONFIG_PPC64 */
864 
865 #ifdef CONFIG_PPC_BOOK3S_64
866 	batch = this_cpu_ptr(&ppc64_tlb_batch);
867 	if (batch->active) {
868 		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
869 		if (batch->index)
870 			__flush_tlb_pending(batch);
871 		batch->active = 0;
872 	}
873 #endif /* CONFIG_PPC_BOOK3S_64 */
874 
875 	/*
876 	 * We can't take a PMU exception inside _switch() since there is a
877 	 * window where the kernel stack SLB and the kernel stack are out
878 	 * of sync. Hard disable here.
879 	 */
880 	hard_irq_disable();
881 
882 	tm_recheckpoint_new_task(new);
883 
884 	last = _switch(old_thread, new_thread);
885 
886 #ifdef CONFIG_PPC_BOOK3S_64
887 	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
888 		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
889 		batch = this_cpu_ptr(&ppc64_tlb_batch);
890 		batch->active = 1;
891 	}
892 #endif /* CONFIG_PPC_BOOK3S_64 */
893 
894 	return last;
895 }
896 
897 static int instructions_to_print = 16;
898 
899 static void show_instructions(struct pt_regs *regs)
900 {
901 	int i;
902 	unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
903 			sizeof(int));
904 
905 	printk("Instruction dump:");
906 
907 	for (i = 0; i < instructions_to_print; i++) {
908 		int instr;
909 
910 		if (!(i % 8))
911 			printk("\n");
912 
913 #if !defined(CONFIG_BOOKE)
914 		/* If executing with the IMMU off, adjust pc rather
915 		 * than print XXXXXXXX.
916 		 */
917 		if (!(regs->msr & MSR_IR))
918 			pc = (unsigned long)phys_to_virt(pc);
919 #endif
920 
921 		if (!__kernel_text_address(pc) ||
922 		     probe_kernel_address((unsigned int __user *)pc, instr)) {
923 			printk(KERN_CONT "XXXXXXXX ");
924 		} else {
925 			if (regs->nip == pc)
926 				printk(KERN_CONT "<%08x> ", instr);
927 			else
928 				printk(KERN_CONT "%08x ", instr);
929 		}
930 
931 		pc += sizeof(int);
932 	}
933 
934 	printk("\n");
935 }
936 
937 static struct regbit {
938 	unsigned long bit;
939 	const char *name;
940 } msr_bits[] = {
941 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
942 	{MSR_SF,	"SF"},
943 	{MSR_HV,	"HV"},
944 #endif
945 	{MSR_VEC,	"VEC"},
946 	{MSR_VSX,	"VSX"},
947 #ifdef CONFIG_BOOKE
948 	{MSR_CE,	"CE"},
949 #endif
950 	{MSR_EE,	"EE"},
951 	{MSR_PR,	"PR"},
952 	{MSR_FP,	"FP"},
953 	{MSR_ME,	"ME"},
954 #ifdef CONFIG_BOOKE
955 	{MSR_DE,	"DE"},
956 #else
957 	{MSR_SE,	"SE"},
958 	{MSR_BE,	"BE"},
959 #endif
960 	{MSR_IR,	"IR"},
961 	{MSR_DR,	"DR"},
962 	{MSR_PMM,	"PMM"},
963 #ifndef CONFIG_BOOKE
964 	{MSR_RI,	"RI"},
965 	{MSR_LE,	"LE"},
966 #endif
967 	{0,		NULL}
968 };
969 
970 static void printbits(unsigned long val, struct regbit *bits)
971 {
972 	const char *sep = "";
973 
974 	printk("<");
975 	for (; bits->bit; ++bits)
976 		if (val & bits->bit) {
977 			printk("%s%s", sep, bits->name);
978 			sep = ",";
979 		}
980 	printk(">");
981 }
982 
983 #ifdef CONFIG_PPC64
984 #define REG		"%016lx"
985 #define REGS_PER_LINE	4
986 #define LAST_VOLATILE	13
987 #else
988 #define REG		"%08lx"
989 #define REGS_PER_LINE	8
990 #define LAST_VOLATILE	12
991 #endif
992 
993 void show_regs(struct pt_regs * regs)
994 {
995 	int i, trap;
996 
997 	show_regs_print_info(KERN_DEFAULT);
998 
999 	printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1000 	       regs->nip, regs->link, regs->ctr);
1001 	printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
1002 	       regs, regs->trap, print_tainted(), init_utsname()->release);
1003 	printk("MSR: "REG" ", regs->msr);
1004 	printbits(regs->msr, msr_bits);
1005 	printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1006 	trap = TRAP(regs);
1007 	if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1008 		printk("CFAR: "REG" ", regs->orig_gpr3);
1009 	if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1010 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1011 		printk("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1012 #else
1013 		printk("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1014 #endif
1015 #ifdef CONFIG_PPC64
1016 	printk("SOFTE: %ld ", regs->softe);
1017 #endif
1018 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1019 	if (MSR_TM_ACTIVE(regs->msr))
1020 		printk("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1021 #endif
1022 
1023 	for (i = 0;  i < 32;  i++) {
1024 		if ((i % REGS_PER_LINE) == 0)
1025 			printk("\nGPR%02d: ", i);
1026 		printk(REG " ", regs->gpr[i]);
1027 		if (i == LAST_VOLATILE && !FULL_REGS(regs))
1028 			break;
1029 	}
1030 	printk("\n");
1031 #ifdef CONFIG_KALLSYMS
1032 	/*
1033 	 * Lookup NIP late so we have the best change of getting the
1034 	 * above info out without failing
1035 	 */
1036 	printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1037 	printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1038 #endif
1039 	show_stack(current, (unsigned long *) regs->gpr[1]);
1040 	if (!user_mode(regs))
1041 		show_instructions(regs);
1042 }
1043 
1044 void exit_thread(void)
1045 {
1046 	discard_lazy_cpu_state();
1047 }
1048 
1049 void flush_thread(void)
1050 {
1051 	discard_lazy_cpu_state();
1052 
1053 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1054 	flush_ptrace_hw_breakpoint(current);
1055 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1056 	set_debug_reg_defaults(&current->thread);
1057 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1058 }
1059 
1060 void
1061 release_thread(struct task_struct *t)
1062 {
1063 }
1064 
1065 /*
1066  * this gets called so that we can store coprocessor state into memory and
1067  * copy the current task into the new thread.
1068  */
1069 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1070 {
1071 	flush_fp_to_thread(src);
1072 	flush_altivec_to_thread(src);
1073 	flush_vsx_to_thread(src);
1074 	flush_spe_to_thread(src);
1075 	/*
1076 	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1077 	 * flush but it removes the checkpointed state from the current CPU and
1078 	 * transitions the CPU out of TM mode.  Hence we need to call
1079 	 * tm_recheckpoint_new_task() (on the same task) to restore the
1080 	 * checkpointed state back and the TM mode.
1081 	 */
1082 	__switch_to_tm(src);
1083 	tm_recheckpoint_new_task(src);
1084 
1085 	*dst = *src;
1086 
1087 	clear_task_ebb(dst);
1088 
1089 	return 0;
1090 }
1091 
1092 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1093 {
1094 #ifdef CONFIG_PPC_STD_MMU_64
1095 	unsigned long sp_vsid;
1096 	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1097 
1098 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1099 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1100 			<< SLB_VSID_SHIFT_1T;
1101 	else
1102 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1103 			<< SLB_VSID_SHIFT;
1104 	sp_vsid |= SLB_VSID_KERNEL | llp;
1105 	p->thread.ksp_vsid = sp_vsid;
1106 #endif
1107 }
1108 
1109 /*
1110  * Copy a thread..
1111  */
1112 
1113 /*
1114  * Copy architecture-specific thread state
1115  */
1116 int copy_thread(unsigned long clone_flags, unsigned long usp,
1117 		unsigned long kthread_arg, struct task_struct *p)
1118 {
1119 	struct pt_regs *childregs, *kregs;
1120 	extern void ret_from_fork(void);
1121 	extern void ret_from_kernel_thread(void);
1122 	void (*f)(void);
1123 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1124 
1125 	/* Copy registers */
1126 	sp -= sizeof(struct pt_regs);
1127 	childregs = (struct pt_regs *) sp;
1128 	if (unlikely(p->flags & PF_KTHREAD)) {
1129 		/* kernel thread */
1130 		struct thread_info *ti = (void *)task_stack_page(p);
1131 		memset(childregs, 0, sizeof(struct pt_regs));
1132 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
1133 		/* function */
1134 		if (usp)
1135 			childregs->gpr[14] = ppc_function_entry((void *)usp);
1136 #ifdef CONFIG_PPC64
1137 		clear_tsk_thread_flag(p, TIF_32BIT);
1138 		childregs->softe = 1;
1139 #endif
1140 		childregs->gpr[15] = kthread_arg;
1141 		p->thread.regs = NULL;	/* no user register state */
1142 		ti->flags |= _TIF_RESTOREALL;
1143 		f = ret_from_kernel_thread;
1144 	} else {
1145 		/* user thread */
1146 		struct pt_regs *regs = current_pt_regs();
1147 		CHECK_FULL_REGS(regs);
1148 		*childregs = *regs;
1149 		if (usp)
1150 			childregs->gpr[1] = usp;
1151 		p->thread.regs = childregs;
1152 		childregs->gpr[3] = 0;  /* Result from fork() */
1153 		if (clone_flags & CLONE_SETTLS) {
1154 #ifdef CONFIG_PPC64
1155 			if (!is_32bit_task())
1156 				childregs->gpr[13] = childregs->gpr[6];
1157 			else
1158 #endif
1159 				childregs->gpr[2] = childregs->gpr[6];
1160 		}
1161 
1162 		f = ret_from_fork;
1163 	}
1164 	sp -= STACK_FRAME_OVERHEAD;
1165 
1166 	/*
1167 	 * The way this works is that at some point in the future
1168 	 * some task will call _switch to switch to the new task.
1169 	 * That will pop off the stack frame created below and start
1170 	 * the new task running at ret_from_fork.  The new task will
1171 	 * do some house keeping and then return from the fork or clone
1172 	 * system call, using the stack frame created above.
1173 	 */
1174 	((unsigned long *)sp)[0] = 0;
1175 	sp -= sizeof(struct pt_regs);
1176 	kregs = (struct pt_regs *) sp;
1177 	sp -= STACK_FRAME_OVERHEAD;
1178 	p->thread.ksp = sp;
1179 #ifdef CONFIG_PPC32
1180 	p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1181 				_ALIGN_UP(sizeof(struct thread_info), 16);
1182 #endif
1183 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1184 	p->thread.ptrace_bps[0] = NULL;
1185 #endif
1186 
1187 	p->thread.fp_save_area = NULL;
1188 #ifdef CONFIG_ALTIVEC
1189 	p->thread.vr_save_area = NULL;
1190 #endif
1191 
1192 	setup_ksp_vsid(p, sp);
1193 
1194 #ifdef CONFIG_PPC64
1195 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1196 		p->thread.dscr_inherit = current->thread.dscr_inherit;
1197 		p->thread.dscr = current->thread.dscr;
1198 	}
1199 	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1200 		p->thread.ppr = INIT_PPR;
1201 #endif
1202 	kregs->nip = ppc_function_entry(f);
1203 	return 0;
1204 }
1205 
1206 /*
1207  * Set up a thread for executing a new program
1208  */
1209 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1210 {
1211 #ifdef CONFIG_PPC64
1212 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1213 #endif
1214 
1215 	/*
1216 	 * If we exec out of a kernel thread then thread.regs will not be
1217 	 * set.  Do it now.
1218 	 */
1219 	if (!current->thread.regs) {
1220 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1221 		current->thread.regs = regs - 1;
1222 	}
1223 
1224 	memset(regs->gpr, 0, sizeof(regs->gpr));
1225 	regs->ctr = 0;
1226 	regs->link = 0;
1227 	regs->xer = 0;
1228 	regs->ccr = 0;
1229 	regs->gpr[1] = sp;
1230 
1231 	/*
1232 	 * We have just cleared all the nonvolatile GPRs, so make
1233 	 * FULL_REGS(regs) return true.  This is necessary to allow
1234 	 * ptrace to examine the thread immediately after exec.
1235 	 */
1236 	regs->trap &= ~1UL;
1237 
1238 #ifdef CONFIG_PPC32
1239 	regs->mq = 0;
1240 	regs->nip = start;
1241 	regs->msr = MSR_USER;
1242 #else
1243 	if (!is_32bit_task()) {
1244 		unsigned long entry;
1245 
1246 		if (is_elf2_task()) {
1247 			/* Look ma, no function descriptors! */
1248 			entry = start;
1249 
1250 			/*
1251 			 * Ulrich says:
1252 			 *   The latest iteration of the ABI requires that when
1253 			 *   calling a function (at its global entry point),
1254 			 *   the caller must ensure r12 holds the entry point
1255 			 *   address (so that the function can quickly
1256 			 *   establish addressability).
1257 			 */
1258 			regs->gpr[12] = start;
1259 			/* Make sure that's restored on entry to userspace. */
1260 			set_thread_flag(TIF_RESTOREALL);
1261 		} else {
1262 			unsigned long toc;
1263 
1264 			/* start is a relocated pointer to the function
1265 			 * descriptor for the elf _start routine.  The first
1266 			 * entry in the function descriptor is the entry
1267 			 * address of _start and the second entry is the TOC
1268 			 * value we need to use.
1269 			 */
1270 			__get_user(entry, (unsigned long __user *)start);
1271 			__get_user(toc, (unsigned long __user *)start+1);
1272 
1273 			/* Check whether the e_entry function descriptor entries
1274 			 * need to be relocated before we can use them.
1275 			 */
1276 			if (load_addr != 0) {
1277 				entry += load_addr;
1278 				toc   += load_addr;
1279 			}
1280 			regs->gpr[2] = toc;
1281 		}
1282 		regs->nip = entry;
1283 		regs->msr = MSR_USER64;
1284 	} else {
1285 		regs->nip = start;
1286 		regs->gpr[2] = 0;
1287 		regs->msr = MSR_USER32;
1288 	}
1289 #endif
1290 	discard_lazy_cpu_state();
1291 #ifdef CONFIG_VSX
1292 	current->thread.used_vsr = 0;
1293 #endif
1294 	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1295 	current->thread.fp_save_area = NULL;
1296 #ifdef CONFIG_ALTIVEC
1297 	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1298 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1299 	current->thread.vr_save_area = NULL;
1300 	current->thread.vrsave = 0;
1301 	current->thread.used_vr = 0;
1302 #endif /* CONFIG_ALTIVEC */
1303 #ifdef CONFIG_SPE
1304 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1305 	current->thread.acc = 0;
1306 	current->thread.spefscr = 0;
1307 	current->thread.used_spe = 0;
1308 #endif /* CONFIG_SPE */
1309 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1310 	if (cpu_has_feature(CPU_FTR_TM))
1311 		regs->msr |= MSR_TM;
1312 	current->thread.tm_tfhar = 0;
1313 	current->thread.tm_texasr = 0;
1314 	current->thread.tm_tfiar = 0;
1315 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1316 }
1317 EXPORT_SYMBOL(start_thread);
1318 
1319 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1320 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1321 
1322 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1323 {
1324 	struct pt_regs *regs = tsk->thread.regs;
1325 
1326 	/* This is a bit hairy.  If we are an SPE enabled  processor
1327 	 * (have embedded fp) we store the IEEE exception enable flags in
1328 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1329 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1330 	if (val & PR_FP_EXC_SW_ENABLE) {
1331 #ifdef CONFIG_SPE
1332 		if (cpu_has_feature(CPU_FTR_SPE)) {
1333 			/*
1334 			 * When the sticky exception bits are set
1335 			 * directly by userspace, it must call prctl
1336 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1337 			 * in the existing prctl settings) or
1338 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1339 			 * the bits being set).  <fenv.h> functions
1340 			 * saving and restoring the whole
1341 			 * floating-point environment need to do so
1342 			 * anyway to restore the prctl settings from
1343 			 * the saved environment.
1344 			 */
1345 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1346 			tsk->thread.fpexc_mode = val &
1347 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1348 			return 0;
1349 		} else {
1350 			return -EINVAL;
1351 		}
1352 #else
1353 		return -EINVAL;
1354 #endif
1355 	}
1356 
1357 	/* on a CONFIG_SPE this does not hurt us.  The bits that
1358 	 * __pack_fe01 use do not overlap with bits used for
1359 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1360 	 * on CONFIG_SPE implementations are reserved so writing to
1361 	 * them does not change anything */
1362 	if (val > PR_FP_EXC_PRECISE)
1363 		return -EINVAL;
1364 	tsk->thread.fpexc_mode = __pack_fe01(val);
1365 	if (regs != NULL && (regs->msr & MSR_FP) != 0)
1366 		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1367 			| tsk->thread.fpexc_mode;
1368 	return 0;
1369 }
1370 
1371 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1372 {
1373 	unsigned int val;
1374 
1375 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1376 #ifdef CONFIG_SPE
1377 		if (cpu_has_feature(CPU_FTR_SPE)) {
1378 			/*
1379 			 * When the sticky exception bits are set
1380 			 * directly by userspace, it must call prctl
1381 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1382 			 * in the existing prctl settings) or
1383 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1384 			 * the bits being set).  <fenv.h> functions
1385 			 * saving and restoring the whole
1386 			 * floating-point environment need to do so
1387 			 * anyway to restore the prctl settings from
1388 			 * the saved environment.
1389 			 */
1390 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1391 			val = tsk->thread.fpexc_mode;
1392 		} else
1393 			return -EINVAL;
1394 #else
1395 		return -EINVAL;
1396 #endif
1397 	else
1398 		val = __unpack_fe01(tsk->thread.fpexc_mode);
1399 	return put_user(val, (unsigned int __user *) adr);
1400 }
1401 
1402 int set_endian(struct task_struct *tsk, unsigned int val)
1403 {
1404 	struct pt_regs *regs = tsk->thread.regs;
1405 
1406 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1407 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1408 		return -EINVAL;
1409 
1410 	if (regs == NULL)
1411 		return -EINVAL;
1412 
1413 	if (val == PR_ENDIAN_BIG)
1414 		regs->msr &= ~MSR_LE;
1415 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1416 		regs->msr |= MSR_LE;
1417 	else
1418 		return -EINVAL;
1419 
1420 	return 0;
1421 }
1422 
1423 int get_endian(struct task_struct *tsk, unsigned long adr)
1424 {
1425 	struct pt_regs *regs = tsk->thread.regs;
1426 	unsigned int val;
1427 
1428 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1429 	    !cpu_has_feature(CPU_FTR_REAL_LE))
1430 		return -EINVAL;
1431 
1432 	if (regs == NULL)
1433 		return -EINVAL;
1434 
1435 	if (regs->msr & MSR_LE) {
1436 		if (cpu_has_feature(CPU_FTR_REAL_LE))
1437 			val = PR_ENDIAN_LITTLE;
1438 		else
1439 			val = PR_ENDIAN_PPC_LITTLE;
1440 	} else
1441 		val = PR_ENDIAN_BIG;
1442 
1443 	return put_user(val, (unsigned int __user *)adr);
1444 }
1445 
1446 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1447 {
1448 	tsk->thread.align_ctl = val;
1449 	return 0;
1450 }
1451 
1452 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1453 {
1454 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1455 }
1456 
1457 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1458 				  unsigned long nbytes)
1459 {
1460 	unsigned long stack_page;
1461 	unsigned long cpu = task_cpu(p);
1462 
1463 	/*
1464 	 * Avoid crashing if the stack has overflowed and corrupted
1465 	 * task_cpu(p), which is in the thread_info struct.
1466 	 */
1467 	if (cpu < NR_CPUS && cpu_possible(cpu)) {
1468 		stack_page = (unsigned long) hardirq_ctx[cpu];
1469 		if (sp >= stack_page + sizeof(struct thread_struct)
1470 		    && sp <= stack_page + THREAD_SIZE - nbytes)
1471 			return 1;
1472 
1473 		stack_page = (unsigned long) softirq_ctx[cpu];
1474 		if (sp >= stack_page + sizeof(struct thread_struct)
1475 		    && sp <= stack_page + THREAD_SIZE - nbytes)
1476 			return 1;
1477 	}
1478 	return 0;
1479 }
1480 
1481 int validate_sp(unsigned long sp, struct task_struct *p,
1482 		       unsigned long nbytes)
1483 {
1484 	unsigned long stack_page = (unsigned long)task_stack_page(p);
1485 
1486 	if (sp >= stack_page + sizeof(struct thread_struct)
1487 	    && sp <= stack_page + THREAD_SIZE - nbytes)
1488 		return 1;
1489 
1490 	return valid_irq_stack(sp, p, nbytes);
1491 }
1492 
1493 EXPORT_SYMBOL(validate_sp);
1494 
1495 unsigned long get_wchan(struct task_struct *p)
1496 {
1497 	unsigned long ip, sp;
1498 	int count = 0;
1499 
1500 	if (!p || p == current || p->state == TASK_RUNNING)
1501 		return 0;
1502 
1503 	sp = p->thread.ksp;
1504 	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1505 		return 0;
1506 
1507 	do {
1508 		sp = *(unsigned long *)sp;
1509 		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1510 			return 0;
1511 		if (count > 0) {
1512 			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1513 			if (!in_sched_functions(ip))
1514 				return ip;
1515 		}
1516 	} while (count++ < 16);
1517 	return 0;
1518 }
1519 
1520 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1521 
1522 void show_stack(struct task_struct *tsk, unsigned long *stack)
1523 {
1524 	unsigned long sp, ip, lr, newsp;
1525 	int count = 0;
1526 	int firstframe = 1;
1527 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1528 	int curr_frame = current->curr_ret_stack;
1529 	extern void return_to_handler(void);
1530 	unsigned long rth = (unsigned long)return_to_handler;
1531 #endif
1532 
1533 	sp = (unsigned long) stack;
1534 	if (tsk == NULL)
1535 		tsk = current;
1536 	if (sp == 0) {
1537 		if (tsk == current)
1538 			sp = current_stack_pointer();
1539 		else
1540 			sp = tsk->thread.ksp;
1541 	}
1542 
1543 	lr = 0;
1544 	printk("Call Trace:\n");
1545 	do {
1546 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1547 			return;
1548 
1549 		stack = (unsigned long *) sp;
1550 		newsp = stack[0];
1551 		ip = stack[STACK_FRAME_LR_SAVE];
1552 		if (!firstframe || ip != lr) {
1553 			printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1554 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1555 			if ((ip == rth) && curr_frame >= 0) {
1556 				printk(" (%pS)",
1557 				       (void *)current->ret_stack[curr_frame].ret);
1558 				curr_frame--;
1559 			}
1560 #endif
1561 			if (firstframe)
1562 				printk(" (unreliable)");
1563 			printk("\n");
1564 		}
1565 		firstframe = 0;
1566 
1567 		/*
1568 		 * See if this is an exception frame.
1569 		 * We look for the "regshere" marker in the current frame.
1570 		 */
1571 		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1572 		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1573 			struct pt_regs *regs = (struct pt_regs *)
1574 				(sp + STACK_FRAME_OVERHEAD);
1575 			lr = regs->link;
1576 			printk("--- interrupt: %lx at %pS\n    LR = %pS\n",
1577 			       regs->trap, (void *)regs->nip, (void *)lr);
1578 			firstframe = 1;
1579 		}
1580 
1581 		sp = newsp;
1582 	} while (count++ < kstack_depth_to_print);
1583 }
1584 
1585 #ifdef CONFIG_PPC64
1586 /* Called with hard IRQs off */
1587 void notrace __ppc64_runlatch_on(void)
1588 {
1589 	struct thread_info *ti = current_thread_info();
1590 	unsigned long ctrl;
1591 
1592 	ctrl = mfspr(SPRN_CTRLF);
1593 	ctrl |= CTRL_RUNLATCH;
1594 	mtspr(SPRN_CTRLT, ctrl);
1595 
1596 	ti->local_flags |= _TLF_RUNLATCH;
1597 }
1598 
1599 /* Called with hard IRQs off */
1600 void notrace __ppc64_runlatch_off(void)
1601 {
1602 	struct thread_info *ti = current_thread_info();
1603 	unsigned long ctrl;
1604 
1605 	ti->local_flags &= ~_TLF_RUNLATCH;
1606 
1607 	ctrl = mfspr(SPRN_CTRLF);
1608 	ctrl &= ~CTRL_RUNLATCH;
1609 	mtspr(SPRN_CTRLT, ctrl);
1610 }
1611 #endif /* CONFIG_PPC64 */
1612 
1613 unsigned long arch_align_stack(unsigned long sp)
1614 {
1615 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1616 		sp -= get_random_int() & ~PAGE_MASK;
1617 	return sp & ~0xf;
1618 }
1619 
1620 static inline unsigned long brk_rnd(void)
1621 {
1622         unsigned long rnd = 0;
1623 
1624 	/* 8MB for 32bit, 1GB for 64bit */
1625 	if (is_32bit_task())
1626 		rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1627 	else
1628 		rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1629 
1630 	return rnd << PAGE_SHIFT;
1631 }
1632 
1633 unsigned long arch_randomize_brk(struct mm_struct *mm)
1634 {
1635 	unsigned long base = mm->brk;
1636 	unsigned long ret;
1637 
1638 #ifdef CONFIG_PPC_STD_MMU_64
1639 	/*
1640 	 * If we are using 1TB segments and we are allowed to randomise
1641 	 * the heap, we can put it above 1TB so it is backed by a 1TB
1642 	 * segment. Otherwise the heap will be in the bottom 1TB
1643 	 * which always uses 256MB segments and this may result in a
1644 	 * performance penalty.
1645 	 */
1646 	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1647 		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1648 #endif
1649 
1650 	ret = PAGE_ALIGN(base + brk_rnd());
1651 
1652 	if (ret < mm->brk)
1653 		return mm->brk;
1654 
1655 	return ret;
1656 }
1657 
1658