2 * Derived from "arch/i386/kernel/process.c"
3 * Copyright (C) 1995 Linus Torvalds
5 * Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
6 * Paul Mackerras (paulus@cs.anu.edu.au)
9 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
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.
17 #include <linux/errno.h>
18 #include <linux/sched.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/kernel.h>
24 #include <linux/smp.h>
25 #include <linux/stddef.h>
26 #include <linux/unistd.h>
27 #include <linux/ptrace.h>
28 #include <linux/slab.h>
29 #include <linux/user.h>
30 #include <linux/elf.h>
31 #include <linux/prctl.h>
32 #include <linux/init_task.h>
33 #include <linux/export.h>
34 #include <linux/kallsyms.h>
35 #include <linux/mqueue.h>
36 #include <linux/hardirq.h>
37 #include <linux/utsname.h>
38 #include <linux/ftrace.h>
39 #include <linux/kernel_stat.h>
40 #include <linux/personality.h>
41 #include <linux/random.h>
42 #include <linux/hw_breakpoint.h>
43 #include <linux/uaccess.h>
44 #include <linux/elf-randomize.h>
46 #include <asm/pgtable.h>
48 #include <asm/processor.h>
51 #include <asm/machdep.h>
53 #include <asm/runlatch.h>
54 #include <asm/syscalls.h>
55 #include <asm/switch_to.h>
57 #include <asm/debug.h>
59 #include <asm/firmware.h>
61 #include <asm/code-patching.h>
63 #include <asm/livepatch.h>
64 #include <asm/cpu_has_feature.h>
65 #include <asm/asm-prototypes.h>
67 #include <linux/kprobes.h>
68 #include <linux/kdebug.h>
70 /* Transactional Memory debug */
72 #define TM_DEBUG(x...) printk(KERN_INFO x)
74 #define TM_DEBUG(x...) do { } while(0)
77 extern unsigned long _get_SP(void);
79 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
80 static void check_if_tm_restore_required(struct task_struct *tsk)
83 * If we are saving the current thread's registers, and the
84 * thread is in a transactional state, set the TIF_RESTORE_TM
85 * bit so that we know to restore the registers before
86 * returning to userspace.
88 if (tsk == current && tsk->thread.regs &&
89 MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
90 !test_thread_flag(TIF_RESTORE_TM)) {
91 tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
92 set_thread_flag(TIF_RESTORE_TM);
96 static inline bool msr_tm_active(unsigned long msr)
98 return MSR_TM_ACTIVE(msr);
101 static inline bool msr_tm_active(unsigned long msr) { return false; }
102 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
103 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
105 bool strict_msr_control;
106 EXPORT_SYMBOL(strict_msr_control);
108 static int __init enable_strict_msr_control(char *str)
110 strict_msr_control = true;
111 pr_info("Enabling strict facility control\n");
115 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
117 unsigned long msr_check_and_set(unsigned long bits)
119 unsigned long oldmsr = mfmsr();
120 unsigned long newmsr;
122 newmsr = oldmsr | bits;
125 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
129 if (oldmsr != newmsr)
135 void __msr_check_and_clear(unsigned long bits)
137 unsigned long oldmsr = mfmsr();
138 unsigned long newmsr;
140 newmsr = oldmsr & ~bits;
143 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
147 if (oldmsr != newmsr)
150 EXPORT_SYMBOL(__msr_check_and_clear);
152 #ifdef CONFIG_PPC_FPU
153 void __giveup_fpu(struct task_struct *tsk)
158 msr = tsk->thread.regs->msr;
161 if (cpu_has_feature(CPU_FTR_VSX))
164 tsk->thread.regs->msr = msr;
167 void giveup_fpu(struct task_struct *tsk)
169 check_if_tm_restore_required(tsk);
171 msr_check_and_set(MSR_FP);
173 msr_check_and_clear(MSR_FP);
175 EXPORT_SYMBOL(giveup_fpu);
178 * Make sure the floating-point register state in the
179 * the thread_struct is up to date for task tsk.
181 void flush_fp_to_thread(struct task_struct *tsk)
183 if (tsk->thread.regs) {
185 * We need to disable preemption here because if we didn't,
186 * another process could get scheduled after the regs->msr
187 * test but before we have finished saving the FP registers
188 * to the thread_struct. That process could take over the
189 * FPU, and then when we get scheduled again we would store
190 * bogus values for the remaining FP registers.
193 if (tsk->thread.regs->msr & MSR_FP) {
195 * This should only ever be called for current or
196 * for a stopped child process. Since we save away
197 * the FP register state on context switch,
198 * there is something wrong if a stopped child appears
199 * to still have its FP state in the CPU registers.
201 BUG_ON(tsk != current);
207 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
209 void enable_kernel_fp(void)
211 unsigned long cpumsr;
213 WARN_ON(preemptible());
215 cpumsr = msr_check_and_set(MSR_FP);
217 if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
218 check_if_tm_restore_required(current);
220 * If a thread has already been reclaimed then the
221 * checkpointed registers are on the CPU but have definitely
222 * been saved by the reclaim code. Don't need to and *cannot*
223 * giveup as this would save to the 'live' structure not the
224 * checkpointed structure.
226 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
228 __giveup_fpu(current);
231 EXPORT_SYMBOL(enable_kernel_fp);
233 static int restore_fp(struct task_struct *tsk) {
234 if (tsk->thread.load_fp || msr_tm_active(tsk->thread.regs->msr)) {
235 load_fp_state(¤t->thread.fp_state);
236 current->thread.load_fp++;
242 static int restore_fp(struct task_struct *tsk) { return 0; }
243 #endif /* CONFIG_PPC_FPU */
245 #ifdef CONFIG_ALTIVEC
246 #define loadvec(thr) ((thr).load_vec)
248 static void __giveup_altivec(struct task_struct *tsk)
253 msr = tsk->thread.regs->msr;
256 if (cpu_has_feature(CPU_FTR_VSX))
259 tsk->thread.regs->msr = msr;
262 void giveup_altivec(struct task_struct *tsk)
264 check_if_tm_restore_required(tsk);
266 msr_check_and_set(MSR_VEC);
267 __giveup_altivec(tsk);
268 msr_check_and_clear(MSR_VEC);
270 EXPORT_SYMBOL(giveup_altivec);
272 void enable_kernel_altivec(void)
274 unsigned long cpumsr;
276 WARN_ON(preemptible());
278 cpumsr = msr_check_and_set(MSR_VEC);
280 if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
281 check_if_tm_restore_required(current);
283 * If a thread has already been reclaimed then the
284 * checkpointed registers are on the CPU but have definitely
285 * been saved by the reclaim code. Don't need to and *cannot*
286 * giveup as this would save to the 'live' structure not the
287 * checkpointed structure.
289 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
291 __giveup_altivec(current);
294 EXPORT_SYMBOL(enable_kernel_altivec);
297 * Make sure the VMX/Altivec register state in the
298 * the thread_struct is up to date for task tsk.
300 void flush_altivec_to_thread(struct task_struct *tsk)
302 if (tsk->thread.regs) {
304 if (tsk->thread.regs->msr & MSR_VEC) {
305 BUG_ON(tsk != current);
311 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
313 static int restore_altivec(struct task_struct *tsk)
315 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
316 (tsk->thread.load_vec || msr_tm_active(tsk->thread.regs->msr))) {
317 load_vr_state(&tsk->thread.vr_state);
318 tsk->thread.used_vr = 1;
319 tsk->thread.load_vec++;
326 #define loadvec(thr) 0
327 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
328 #endif /* CONFIG_ALTIVEC */
331 static void __giveup_vsx(struct task_struct *tsk)
333 if (tsk->thread.regs->msr & MSR_FP)
335 if (tsk->thread.regs->msr & MSR_VEC)
336 __giveup_altivec(tsk);
337 tsk->thread.regs->msr &= ~MSR_VSX;
340 static void giveup_vsx(struct task_struct *tsk)
342 check_if_tm_restore_required(tsk);
344 msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
346 msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
349 static void save_vsx(struct task_struct *tsk)
351 if (tsk->thread.regs->msr & MSR_FP)
353 if (tsk->thread.regs->msr & MSR_VEC)
357 void enable_kernel_vsx(void)
359 unsigned long cpumsr;
361 WARN_ON(preemptible());
363 cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
365 if (current->thread.regs && (current->thread.regs->msr & MSR_VSX)) {
366 check_if_tm_restore_required(current);
368 * If a thread has already been reclaimed then the
369 * checkpointed registers are on the CPU but have definitely
370 * been saved by the reclaim code. Don't need to and *cannot*
371 * giveup as this would save to the 'live' structure not the
372 * checkpointed structure.
374 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
376 if (current->thread.regs->msr & MSR_FP)
377 __giveup_fpu(current);
378 if (current->thread.regs->msr & MSR_VEC)
379 __giveup_altivec(current);
380 __giveup_vsx(current);
383 EXPORT_SYMBOL(enable_kernel_vsx);
385 void flush_vsx_to_thread(struct task_struct *tsk)
387 if (tsk->thread.regs) {
389 if (tsk->thread.regs->msr & MSR_VSX) {
390 BUG_ON(tsk != current);
396 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
398 static int restore_vsx(struct task_struct *tsk)
400 if (cpu_has_feature(CPU_FTR_VSX)) {
401 tsk->thread.used_vsr = 1;
408 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
409 static inline void save_vsx(struct task_struct *tsk) { }
410 #endif /* CONFIG_VSX */
413 void giveup_spe(struct task_struct *tsk)
415 check_if_tm_restore_required(tsk);
417 msr_check_and_set(MSR_SPE);
419 msr_check_and_clear(MSR_SPE);
421 EXPORT_SYMBOL(giveup_spe);
423 void enable_kernel_spe(void)
425 WARN_ON(preemptible());
427 msr_check_and_set(MSR_SPE);
429 if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
430 check_if_tm_restore_required(current);
431 __giveup_spe(current);
434 EXPORT_SYMBOL(enable_kernel_spe);
436 void flush_spe_to_thread(struct task_struct *tsk)
438 if (tsk->thread.regs) {
440 if (tsk->thread.regs->msr & MSR_SPE) {
441 BUG_ON(tsk != current);
442 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
448 #endif /* CONFIG_SPE */
450 static unsigned long msr_all_available;
452 static int __init init_msr_all_available(void)
454 #ifdef CONFIG_PPC_FPU
455 msr_all_available |= MSR_FP;
457 #ifdef CONFIG_ALTIVEC
458 if (cpu_has_feature(CPU_FTR_ALTIVEC))
459 msr_all_available |= MSR_VEC;
462 if (cpu_has_feature(CPU_FTR_VSX))
463 msr_all_available |= MSR_VSX;
466 if (cpu_has_feature(CPU_FTR_SPE))
467 msr_all_available |= MSR_SPE;
472 early_initcall(init_msr_all_available);
474 void giveup_all(struct task_struct *tsk)
476 unsigned long usermsr;
478 if (!tsk->thread.regs)
481 usermsr = tsk->thread.regs->msr;
483 if ((usermsr & msr_all_available) == 0)
486 msr_check_and_set(msr_all_available);
487 check_if_tm_restore_required(tsk);
489 #ifdef CONFIG_PPC_FPU
490 if (usermsr & MSR_FP)
493 #ifdef CONFIG_ALTIVEC
494 if (usermsr & MSR_VEC)
495 __giveup_altivec(tsk);
498 if (usermsr & MSR_VSX)
502 if (usermsr & MSR_SPE)
506 msr_check_and_clear(msr_all_available);
508 EXPORT_SYMBOL(giveup_all);
510 void restore_math(struct pt_regs *regs)
514 if (!msr_tm_active(regs->msr) &&
515 !current->thread.load_fp && !loadvec(current->thread))
519 msr_check_and_set(msr_all_available);
522 * Only reload if the bit is not set in the user MSR, the bit BEING set
523 * indicates that the registers are hot
525 if ((!(msr & MSR_FP)) && restore_fp(current))
526 msr |= MSR_FP | current->thread.fpexc_mode;
528 if ((!(msr & MSR_VEC)) && restore_altivec(current))
531 if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
532 restore_vsx(current)) {
536 msr_check_and_clear(msr_all_available);
541 void save_all(struct task_struct *tsk)
543 unsigned long usermsr;
545 if (!tsk->thread.regs)
548 usermsr = tsk->thread.regs->msr;
550 if ((usermsr & msr_all_available) == 0)
553 msr_check_and_set(msr_all_available);
556 * Saving the way the register space is in hardware, save_vsx boils
557 * down to a save_fpu() and save_altivec()
559 if (usermsr & MSR_VSX) {
562 if (usermsr & MSR_FP)
565 if (usermsr & MSR_VEC)
569 if (usermsr & MSR_SPE)
572 msr_check_and_clear(msr_all_available);
575 void flush_all_to_thread(struct task_struct *tsk)
577 if (tsk->thread.regs) {
579 BUG_ON(tsk != current);
583 if (tsk->thread.regs->msr & MSR_SPE)
584 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
590 EXPORT_SYMBOL(flush_all_to_thread);
592 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
593 void do_send_trap(struct pt_regs *regs, unsigned long address,
594 unsigned long error_code, int signal_code, int breakpt)
598 current->thread.trap_nr = signal_code;
599 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
600 11, SIGSEGV) == NOTIFY_STOP)
603 /* Deliver the signal to userspace */
604 info.si_signo = SIGTRAP;
605 info.si_errno = breakpt; /* breakpoint or watchpoint id */
606 info.si_code = signal_code;
607 info.si_addr = (void __user *)address;
608 force_sig_info(SIGTRAP, &info, current);
610 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
611 void do_break (struct pt_regs *regs, unsigned long address,
612 unsigned long error_code)
616 current->thread.trap_nr = TRAP_HWBKPT;
617 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
618 11, SIGSEGV) == NOTIFY_STOP)
621 if (debugger_break_match(regs))
624 /* Clear the breakpoint */
625 hw_breakpoint_disable();
627 /* Deliver the signal to userspace */
628 info.si_signo = SIGTRAP;
630 info.si_code = TRAP_HWBKPT;
631 info.si_addr = (void __user *)address;
632 force_sig_info(SIGTRAP, &info, current);
634 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
636 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
638 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
640 * Set the debug registers back to their default "safe" values.
642 static void set_debug_reg_defaults(struct thread_struct *thread)
644 thread->debug.iac1 = thread->debug.iac2 = 0;
645 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
646 thread->debug.iac3 = thread->debug.iac4 = 0;
648 thread->debug.dac1 = thread->debug.dac2 = 0;
649 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
650 thread->debug.dvc1 = thread->debug.dvc2 = 0;
652 thread->debug.dbcr0 = 0;
655 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
657 thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
658 DBCR1_IAC3US | DBCR1_IAC4US;
660 * Force Data Address Compare User/Supervisor bits to be User-only
661 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
663 thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
665 thread->debug.dbcr1 = 0;
669 static void prime_debug_regs(struct debug_reg *debug)
672 * We could have inherited MSR_DE from userspace, since
673 * it doesn't get cleared on exception entry. Make sure
674 * MSR_DE is clear before we enable any debug events.
676 mtmsr(mfmsr() & ~MSR_DE);
678 mtspr(SPRN_IAC1, debug->iac1);
679 mtspr(SPRN_IAC2, debug->iac2);
680 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
681 mtspr(SPRN_IAC3, debug->iac3);
682 mtspr(SPRN_IAC4, debug->iac4);
684 mtspr(SPRN_DAC1, debug->dac1);
685 mtspr(SPRN_DAC2, debug->dac2);
686 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
687 mtspr(SPRN_DVC1, debug->dvc1);
688 mtspr(SPRN_DVC2, debug->dvc2);
690 mtspr(SPRN_DBCR0, debug->dbcr0);
691 mtspr(SPRN_DBCR1, debug->dbcr1);
693 mtspr(SPRN_DBCR2, debug->dbcr2);
697 * Unless neither the old or new thread are making use of the
698 * debug registers, set the debug registers from the values
699 * stored in the new thread.
701 void switch_booke_debug_regs(struct debug_reg *new_debug)
703 if ((current->thread.debug.dbcr0 & DBCR0_IDM)
704 || (new_debug->dbcr0 & DBCR0_IDM))
705 prime_debug_regs(new_debug);
707 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
708 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
709 #ifndef CONFIG_HAVE_HW_BREAKPOINT
710 static void set_debug_reg_defaults(struct thread_struct *thread)
712 thread->hw_brk.address = 0;
713 thread->hw_brk.type = 0;
714 set_breakpoint(&thread->hw_brk);
716 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
717 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
719 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
720 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
722 mtspr(SPRN_DAC1, dabr);
723 #ifdef CONFIG_PPC_47x
728 #elif defined(CONFIG_PPC_BOOK3S)
729 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
731 mtspr(SPRN_DABR, dabr);
732 if (cpu_has_feature(CPU_FTR_DABRX))
733 mtspr(SPRN_DABRX, dabrx);
736 #elif defined(CONFIG_PPC_8xx)
737 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
739 unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
740 unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
741 unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
743 if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
745 else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
747 else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
750 mtspr(SPRN_LCTRL2, 0);
751 mtspr(SPRN_CMPE, addr);
752 mtspr(SPRN_CMPF, addr + 4);
753 mtspr(SPRN_LCTRL1, lctrl1);
754 mtspr(SPRN_LCTRL2, lctrl2);
759 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
765 static inline int set_dabr(struct arch_hw_breakpoint *brk)
767 unsigned long dabr, dabrx;
769 dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
770 dabrx = ((brk->type >> 3) & 0x7);
773 return ppc_md.set_dabr(dabr, dabrx);
775 return __set_dabr(dabr, dabrx);
778 static inline int set_dawr(struct arch_hw_breakpoint *brk)
780 unsigned long dawr, dawrx, mrd;
784 dawrx = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
785 << (63 - 58); //* read/write bits */
786 dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
787 << (63 - 59); //* translate */
788 dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
789 >> 3; //* PRIM bits */
790 /* dawr length is stored in field MDR bits 48:53. Matches range in
791 doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
793 brk->len is in bytes.
794 This aligns up to double word size, shifts and does the bias.
796 mrd = ((brk->len + 7) >> 3) - 1;
797 dawrx |= (mrd & 0x3f) << (63 - 53);
800 return ppc_md.set_dawr(dawr, dawrx);
801 mtspr(SPRN_DAWR, dawr);
802 mtspr(SPRN_DAWRX, dawrx);
806 void __set_breakpoint(struct arch_hw_breakpoint *brk)
808 memcpy(this_cpu_ptr(¤t_brk), brk, sizeof(*brk));
810 if (cpu_has_feature(CPU_FTR_DAWR))
816 void set_breakpoint(struct arch_hw_breakpoint *brk)
819 __set_breakpoint(brk);
824 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
827 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
828 struct arch_hw_breakpoint *b)
830 if (a->address != b->address)
832 if (a->type != b->type)
834 if (a->len != b->len)
839 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
841 static inline bool tm_enabled(struct task_struct *tsk)
843 return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
846 static void tm_reclaim_thread(struct thread_struct *thr,
847 struct thread_info *ti, uint8_t cause)
850 * Use the current MSR TM suspended bit to track if we have
851 * checkpointed state outstanding.
852 * On signal delivery, we'd normally reclaim the checkpointed
853 * state to obtain stack pointer (see:get_tm_stackpointer()).
854 * This will then directly return to userspace without going
855 * through __switch_to(). However, if the stack frame is bad,
856 * we need to exit this thread which calls __switch_to() which
857 * will again attempt to reclaim the already saved tm state.
858 * Hence we need to check that we've not already reclaimed
860 * We do this using the current MSR, rather tracking it in
861 * some specific thread_struct bit, as it has the additional
862 * benefit of checking for a potential TM bad thing exception.
864 if (!MSR_TM_SUSPENDED(mfmsr()))
868 * If we are in a transaction and FP is off then we can't have
869 * used FP inside that transaction. Hence the checkpointed
870 * state is the same as the live state. We need to copy the
871 * live state to the checkpointed state so that when the
872 * transaction is restored, the checkpointed state is correct
873 * and the aborted transaction sees the correct state. We use
874 * ckpt_regs.msr here as that's what tm_reclaim will use to
875 * determine if it's going to write the checkpointed state or
876 * not. So either this will write the checkpointed registers,
877 * or reclaim will. Similarly for VMX.
879 if ((thr->ckpt_regs.msr & MSR_FP) == 0)
880 memcpy(&thr->ckfp_state, &thr->fp_state,
881 sizeof(struct thread_fp_state));
882 if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
883 memcpy(&thr->ckvr_state, &thr->vr_state,
884 sizeof(struct thread_vr_state));
886 giveup_all(container_of(thr, struct task_struct, thread));
888 tm_reclaim(thr, thr->ckpt_regs.msr, cause);
891 void tm_reclaim_current(uint8_t cause)
894 tm_reclaim_thread(¤t->thread, current_thread_info(), cause);
897 static inline void tm_reclaim_task(struct task_struct *tsk)
899 /* We have to work out if we're switching from/to a task that's in the
900 * middle of a transaction.
902 * In switching we need to maintain a 2nd register state as
903 * oldtask->thread.ckpt_regs. We tm_reclaim(oldproc); this saves the
904 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
907 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
909 struct thread_struct *thr = &tsk->thread;
914 if (!MSR_TM_ACTIVE(thr->regs->msr))
915 goto out_and_saveregs;
917 TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
918 "ccr=%lx, msr=%lx, trap=%lx)\n",
919 tsk->pid, thr->regs->nip,
920 thr->regs->ccr, thr->regs->msr,
923 tm_reclaim_thread(thr, task_thread_info(tsk), TM_CAUSE_RESCHED);
925 TM_DEBUG("--- tm_reclaim on pid %d complete\n",
929 /* Always save the regs here, even if a transaction's not active.
930 * This context-switches a thread's TM info SPRs. We do it here to
931 * be consistent with the restore path (in recheckpoint) which
932 * cannot happen later in _switch().
937 extern void __tm_recheckpoint(struct thread_struct *thread,
938 unsigned long orig_msr);
940 void tm_recheckpoint(struct thread_struct *thread,
941 unsigned long orig_msr)
945 if (!(thread->regs->msr & MSR_TM))
948 /* We really can't be interrupted here as the TEXASR registers can't
949 * change and later in the trecheckpoint code, we have a userspace R1.
950 * So let's hard disable over this region.
952 local_irq_save(flags);
955 /* The TM SPRs are restored here, so that TEXASR.FS can be set
956 * before the trecheckpoint and no explosion occurs.
958 tm_restore_sprs(thread);
960 __tm_recheckpoint(thread, orig_msr);
962 local_irq_restore(flags);
965 static inline void tm_recheckpoint_new_task(struct task_struct *new)
969 if (!cpu_has_feature(CPU_FTR_TM))
972 /* Recheckpoint the registers of the thread we're about to switch to.
974 * If the task was using FP, we non-lazily reload both the original and
975 * the speculative FP register states. This is because the kernel
976 * doesn't see if/when a TM rollback occurs, so if we take an FP
977 * unavailable later, we are unable to determine which set of FP regs
978 * need to be restored.
980 if (!tm_enabled(new))
983 if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
984 tm_restore_sprs(&new->thread);
987 msr = new->thread.ckpt_regs.msr;
988 /* Recheckpoint to restore original checkpointed register state. */
989 TM_DEBUG("*** tm_recheckpoint of pid %d "
990 "(new->msr 0x%lx, new->origmsr 0x%lx)\n",
991 new->pid, new->thread.regs->msr, msr);
993 tm_recheckpoint(&new->thread, msr);
996 * The checkpointed state has been restored but the live state has
997 * not, ensure all the math functionality is turned off to trigger
998 * restore_math() to reload.
1000 new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1002 TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1003 "(kernel msr 0x%lx)\n",
1007 static inline void __switch_to_tm(struct task_struct *prev,
1008 struct task_struct *new)
1010 if (cpu_has_feature(CPU_FTR_TM)) {
1011 if (tm_enabled(prev) || tm_enabled(new))
1014 if (tm_enabled(prev)) {
1015 prev->thread.load_tm++;
1016 tm_reclaim_task(prev);
1017 if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1018 prev->thread.regs->msr &= ~MSR_TM;
1021 tm_recheckpoint_new_task(new);
1026 * This is called if we are on the way out to userspace and the
1027 * TIF_RESTORE_TM flag is set. It checks if we need to reload
1028 * FP and/or vector state and does so if necessary.
1029 * If userspace is inside a transaction (whether active or
1030 * suspended) and FP/VMX/VSX instructions have ever been enabled
1031 * inside that transaction, then we have to keep them enabled
1032 * and keep the FP/VMX/VSX state loaded while ever the transaction
1033 * continues. The reason is that if we didn't, and subsequently
1034 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1035 * we don't know whether it's the same transaction, and thus we
1036 * don't know which of the checkpointed state and the transactional
1039 void restore_tm_state(struct pt_regs *regs)
1041 unsigned long msr_diff;
1044 * This is the only moment we should clear TIF_RESTORE_TM as
1045 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1046 * again, anything else could lead to an incorrect ckpt_msr being
1047 * saved and therefore incorrect signal contexts.
1049 clear_thread_flag(TIF_RESTORE_TM);
1050 if (!MSR_TM_ACTIVE(regs->msr))
1053 msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1054 msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1056 /* Ensure that restore_math() will restore */
1057 if (msr_diff & MSR_FP)
1058 current->thread.load_fp = 1;
1059 #ifdef CONFIG_ALTIVEC
1060 if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1061 current->thread.load_vec = 1;
1065 regs->msr |= msr_diff;
1069 #define tm_recheckpoint_new_task(new)
1070 #define __switch_to_tm(prev, new)
1071 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1073 static inline void save_sprs(struct thread_struct *t)
1075 #ifdef CONFIG_ALTIVEC
1076 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1077 t->vrsave = mfspr(SPRN_VRSAVE);
1079 #ifdef CONFIG_PPC_BOOK3S_64
1080 if (cpu_has_feature(CPU_FTR_DSCR))
1081 t->dscr = mfspr(SPRN_DSCR);
1083 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1084 t->bescr = mfspr(SPRN_BESCR);
1085 t->ebbhr = mfspr(SPRN_EBBHR);
1086 t->ebbrr = mfspr(SPRN_EBBRR);
1088 t->fscr = mfspr(SPRN_FSCR);
1091 * Note that the TAR is not available for use in the kernel.
1092 * (To provide this, the TAR should be backed up/restored on
1093 * exception entry/exit instead, and be in pt_regs. FIXME,
1094 * this should be in pt_regs anyway (for debug).)
1096 t->tar = mfspr(SPRN_TAR);
1101 static inline void restore_sprs(struct thread_struct *old_thread,
1102 struct thread_struct *new_thread)
1104 #ifdef CONFIG_ALTIVEC
1105 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1106 old_thread->vrsave != new_thread->vrsave)
1107 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1109 #ifdef CONFIG_PPC_BOOK3S_64
1110 if (cpu_has_feature(CPU_FTR_DSCR)) {
1111 u64 dscr = get_paca()->dscr_default;
1112 if (new_thread->dscr_inherit)
1113 dscr = new_thread->dscr;
1115 if (old_thread->dscr != dscr)
1116 mtspr(SPRN_DSCR, dscr);
1119 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1120 if (old_thread->bescr != new_thread->bescr)
1121 mtspr(SPRN_BESCR, new_thread->bescr);
1122 if (old_thread->ebbhr != new_thread->ebbhr)
1123 mtspr(SPRN_EBBHR, new_thread->ebbhr);
1124 if (old_thread->ebbrr != new_thread->ebbrr)
1125 mtspr(SPRN_EBBRR, new_thread->ebbrr);
1127 if (old_thread->fscr != new_thread->fscr)
1128 mtspr(SPRN_FSCR, new_thread->fscr);
1130 if (old_thread->tar != new_thread->tar)
1131 mtspr(SPRN_TAR, new_thread->tar);
1136 struct task_struct *__switch_to(struct task_struct *prev,
1137 struct task_struct *new)
1139 struct thread_struct *new_thread, *old_thread;
1140 struct task_struct *last;
1141 #ifdef CONFIG_PPC_BOOK3S_64
1142 struct ppc64_tlb_batch *batch;
1145 new_thread = &new->thread;
1146 old_thread = ¤t->thread;
1148 WARN_ON(!irqs_disabled());
1152 * Collect processor utilization data per process
1154 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
1155 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
1156 long unsigned start_tb, current_tb;
1157 start_tb = old_thread->start_tb;
1158 cu->current_tb = current_tb = mfspr(SPRN_PURR);
1159 old_thread->accum_tb += (current_tb - start_tb);
1160 new_thread->start_tb = current_tb;
1162 #endif /* CONFIG_PPC64 */
1164 #ifdef CONFIG_PPC_STD_MMU_64
1165 batch = this_cpu_ptr(&ppc64_tlb_batch);
1166 if (batch->active) {
1167 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1169 __flush_tlb_pending(batch);
1172 #endif /* CONFIG_PPC_STD_MMU_64 */
1174 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1175 switch_booke_debug_regs(&new->thread.debug);
1178 * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1181 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1182 if (unlikely(!hw_brk_match(this_cpu_ptr(¤t_brk), &new->thread.hw_brk)))
1183 __set_breakpoint(&new->thread.hw_brk);
1184 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1188 * We need to save SPRs before treclaim/trecheckpoint as these will
1189 * change a number of them.
1191 save_sprs(&prev->thread);
1193 /* Save FPU, Altivec, VSX and SPE state */
1196 __switch_to_tm(prev, new);
1199 * We can't take a PMU exception inside _switch() since there is a
1200 * window where the kernel stack SLB and the kernel stack are out
1201 * of sync. Hard disable here.
1206 * Call restore_sprs() before calling _switch(). If we move it after
1207 * _switch() then we miss out on calling it for new tasks. The reason
1208 * for this is we manually create a stack frame for new tasks that
1209 * directly returns through ret_from_fork() or
1210 * ret_from_kernel_thread(). See copy_thread() for details.
1212 restore_sprs(old_thread, new_thread);
1214 last = _switch(old_thread, new_thread);
1216 #ifdef CONFIG_PPC_STD_MMU_64
1217 if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1218 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1219 batch = this_cpu_ptr(&ppc64_tlb_batch);
1223 if (current_thread_info()->task->thread.regs)
1224 restore_math(current_thread_info()->task->thread.regs);
1225 #endif /* CONFIG_PPC_STD_MMU_64 */
1230 static int instructions_to_print = 16;
1232 static void show_instructions(struct pt_regs *regs)
1235 unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
1238 printk("Instruction dump:");
1240 for (i = 0; i < instructions_to_print; i++) {
1246 #if !defined(CONFIG_BOOKE)
1247 /* If executing with the IMMU off, adjust pc rather
1248 * than print XXXXXXXX.
1250 if (!(regs->msr & MSR_IR))
1251 pc = (unsigned long)phys_to_virt(pc);
1254 if (!__kernel_text_address(pc) ||
1255 probe_kernel_address((unsigned int __user *)pc, instr)) {
1256 pr_cont("XXXXXXXX ");
1258 if (regs->nip == pc)
1259 pr_cont("<%08x> ", instr);
1261 pr_cont("%08x ", instr);
1275 static struct regbit msr_bits[] = {
1276 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1298 #ifndef CONFIG_BOOKE
1305 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1309 for (; bits->bit; ++bits)
1310 if (val & bits->bit) {
1311 pr_cont("%s%s", s, bits->name);
1316 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1317 static struct regbit msr_tm_bits[] = {
1324 static void print_tm_bits(unsigned long val)
1327 * This only prints something if at least one of the TM bit is set.
1328 * Inside the TM[], the output means:
1329 * E: Enabled (bit 32)
1330 * S: Suspended (bit 33)
1331 * T: Transactional (bit 34)
1333 if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1335 print_bits(val, msr_tm_bits, "");
1340 static void print_tm_bits(unsigned long val) {}
1343 static void print_msr_bits(unsigned long val)
1346 print_bits(val, msr_bits, ",");
1352 #define REG "%016lx"
1353 #define REGS_PER_LINE 4
1354 #define LAST_VOLATILE 13
1357 #define REGS_PER_LINE 8
1358 #define LAST_VOLATILE 12
1361 void show_regs(struct pt_regs * regs)
1365 show_regs_print_info(KERN_DEFAULT);
1367 printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1368 regs->nip, regs->link, regs->ctr);
1369 printk("REGS: %p TRAP: %04lx %s (%s)\n",
1370 regs, regs->trap, print_tainted(), init_utsname()->release);
1371 printk("MSR: "REG" ", regs->msr);
1372 print_msr_bits(regs->msr);
1373 printk(" CR: %08lx XER: %08lx\n", regs->ccr, regs->xer);
1375 if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1376 pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1377 if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1378 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1379 pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1381 pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1384 pr_cont("SOFTE: %ld ", regs->softe);
1386 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1387 if (MSR_TM_ACTIVE(regs->msr))
1388 pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1391 for (i = 0; i < 32; i++) {
1392 if ((i % REGS_PER_LINE) == 0)
1393 pr_cont("\nGPR%02d: ", i);
1394 pr_cont(REG " ", regs->gpr[i]);
1395 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1399 #ifdef CONFIG_KALLSYMS
1401 * Lookup NIP late so we have the best change of getting the
1402 * above info out without failing
1404 printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1405 printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1407 show_stack(current, (unsigned long *) regs->gpr[1]);
1408 if (!user_mode(regs))
1409 show_instructions(regs);
1412 void flush_thread(void)
1414 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1415 flush_ptrace_hw_breakpoint(current);
1416 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1417 set_debug_reg_defaults(¤t->thread);
1418 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1422 release_thread(struct task_struct *t)
1427 * this gets called so that we can store coprocessor state into memory and
1428 * copy the current task into the new thread.
1430 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1432 flush_all_to_thread(src);
1434 * Flush TM state out so we can copy it. __switch_to_tm() does this
1435 * flush but it removes the checkpointed state from the current CPU and
1436 * transitions the CPU out of TM mode. Hence we need to call
1437 * tm_recheckpoint_new_task() (on the same task) to restore the
1438 * checkpointed state back and the TM mode.
1440 * Can't pass dst because it isn't ready. Doesn't matter, passing
1441 * dst is only important for __switch_to()
1443 __switch_to_tm(src, src);
1447 clear_task_ebb(dst);
1452 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1454 #ifdef CONFIG_PPC_STD_MMU_64
1455 unsigned long sp_vsid;
1456 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1458 if (radix_enabled())
1461 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1462 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1463 << SLB_VSID_SHIFT_1T;
1465 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1467 sp_vsid |= SLB_VSID_KERNEL | llp;
1468 p->thread.ksp_vsid = sp_vsid;
1477 * Copy architecture-specific thread state
1479 int copy_thread(unsigned long clone_flags, unsigned long usp,
1480 unsigned long kthread_arg, struct task_struct *p)
1482 struct pt_regs *childregs, *kregs;
1483 extern void ret_from_fork(void);
1484 extern void ret_from_kernel_thread(void);
1486 unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1487 struct thread_info *ti = task_thread_info(p);
1489 klp_init_thread_info(ti);
1491 /* Copy registers */
1492 sp -= sizeof(struct pt_regs);
1493 childregs = (struct pt_regs *) sp;
1494 if (unlikely(p->flags & PF_KTHREAD)) {
1496 memset(childregs, 0, sizeof(struct pt_regs));
1497 childregs->gpr[1] = sp + sizeof(struct pt_regs);
1500 childregs->gpr[14] = ppc_function_entry((void *)usp);
1502 clear_tsk_thread_flag(p, TIF_32BIT);
1503 childregs->softe = 1;
1505 childregs->gpr[15] = kthread_arg;
1506 p->thread.regs = NULL; /* no user register state */
1507 ti->flags |= _TIF_RESTOREALL;
1508 f = ret_from_kernel_thread;
1511 struct pt_regs *regs = current_pt_regs();
1512 CHECK_FULL_REGS(regs);
1515 childregs->gpr[1] = usp;
1516 p->thread.regs = childregs;
1517 childregs->gpr[3] = 0; /* Result from fork() */
1518 if (clone_flags & CLONE_SETTLS) {
1520 if (!is_32bit_task())
1521 childregs->gpr[13] = childregs->gpr[6];
1524 childregs->gpr[2] = childregs->gpr[6];
1529 childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1530 sp -= STACK_FRAME_OVERHEAD;
1533 * The way this works is that at some point in the future
1534 * some task will call _switch to switch to the new task.
1535 * That will pop off the stack frame created below and start
1536 * the new task running at ret_from_fork. The new task will
1537 * do some house keeping and then return from the fork or clone
1538 * system call, using the stack frame created above.
1540 ((unsigned long *)sp)[0] = 0;
1541 sp -= sizeof(struct pt_regs);
1542 kregs = (struct pt_regs *) sp;
1543 sp -= STACK_FRAME_OVERHEAD;
1546 p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1547 _ALIGN_UP(sizeof(struct thread_info), 16);
1549 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1550 p->thread.ptrace_bps[0] = NULL;
1553 p->thread.fp_save_area = NULL;
1554 #ifdef CONFIG_ALTIVEC
1555 p->thread.vr_save_area = NULL;
1558 setup_ksp_vsid(p, sp);
1561 if (cpu_has_feature(CPU_FTR_DSCR)) {
1562 p->thread.dscr_inherit = current->thread.dscr_inherit;
1563 p->thread.dscr = mfspr(SPRN_DSCR);
1565 if (cpu_has_feature(CPU_FTR_HAS_PPR))
1566 p->thread.ppr = INIT_PPR;
1568 kregs->nip = ppc_function_entry(f);
1573 * Set up a thread for executing a new program
1575 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1578 unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1582 * If we exec out of a kernel thread then thread.regs will not be
1585 if (!current->thread.regs) {
1586 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1587 current->thread.regs = regs - 1;
1590 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1592 * Clear any transactional state, we're exec()ing. The cause is
1593 * not important as there will never be a recheckpoint so it's not
1596 if (MSR_TM_SUSPENDED(mfmsr()))
1597 tm_reclaim_current(0);
1600 memset(regs->gpr, 0, sizeof(regs->gpr));
1608 * We have just cleared all the nonvolatile GPRs, so make
1609 * FULL_REGS(regs) return true. This is necessary to allow
1610 * ptrace to examine the thread immediately after exec.
1617 regs->msr = MSR_USER;
1619 if (!is_32bit_task()) {
1620 unsigned long entry;
1622 if (is_elf2_task()) {
1623 /* Look ma, no function descriptors! */
1628 * The latest iteration of the ABI requires that when
1629 * calling a function (at its global entry point),
1630 * the caller must ensure r12 holds the entry point
1631 * address (so that the function can quickly
1632 * establish addressability).
1634 regs->gpr[12] = start;
1635 /* Make sure that's restored on entry to userspace. */
1636 set_thread_flag(TIF_RESTOREALL);
1640 /* start is a relocated pointer to the function
1641 * descriptor for the elf _start routine. The first
1642 * entry in the function descriptor is the entry
1643 * address of _start and the second entry is the TOC
1644 * value we need to use.
1646 __get_user(entry, (unsigned long __user *)start);
1647 __get_user(toc, (unsigned long __user *)start+1);
1649 /* Check whether the e_entry function descriptor entries
1650 * need to be relocated before we can use them.
1652 if (load_addr != 0) {
1659 regs->msr = MSR_USER64;
1663 regs->msr = MSR_USER32;
1667 current->thread.used_vsr = 0;
1669 memset(¤t->thread.fp_state, 0, sizeof(current->thread.fp_state));
1670 current->thread.fp_save_area = NULL;
1671 #ifdef CONFIG_ALTIVEC
1672 memset(¤t->thread.vr_state, 0, sizeof(current->thread.vr_state));
1673 current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1674 current->thread.vr_save_area = NULL;
1675 current->thread.vrsave = 0;
1676 current->thread.used_vr = 0;
1677 #endif /* CONFIG_ALTIVEC */
1679 memset(current->thread.evr, 0, sizeof(current->thread.evr));
1680 current->thread.acc = 0;
1681 current->thread.spefscr = 0;
1682 current->thread.used_spe = 0;
1683 #endif /* CONFIG_SPE */
1684 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1685 current->thread.tm_tfhar = 0;
1686 current->thread.tm_texasr = 0;
1687 current->thread.tm_tfiar = 0;
1688 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1690 EXPORT_SYMBOL(start_thread);
1692 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1693 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1695 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1697 struct pt_regs *regs = tsk->thread.regs;
1699 /* This is a bit hairy. If we are an SPE enabled processor
1700 * (have embedded fp) we store the IEEE exception enable flags in
1701 * fpexc_mode. fpexc_mode is also used for setting FP exception
1702 * mode (asyn, precise, disabled) for 'Classic' FP. */
1703 if (val & PR_FP_EXC_SW_ENABLE) {
1705 if (cpu_has_feature(CPU_FTR_SPE)) {
1707 * When the sticky exception bits are set
1708 * directly by userspace, it must call prctl
1709 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1710 * in the existing prctl settings) or
1711 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1712 * the bits being set). <fenv.h> functions
1713 * saving and restoring the whole
1714 * floating-point environment need to do so
1715 * anyway to restore the prctl settings from
1716 * the saved environment.
1718 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1719 tsk->thread.fpexc_mode = val &
1720 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1730 /* on a CONFIG_SPE this does not hurt us. The bits that
1731 * __pack_fe01 use do not overlap with bits used for
1732 * PR_FP_EXC_SW_ENABLE. Additionally, the MSR[FE0,FE1] bits
1733 * on CONFIG_SPE implementations are reserved so writing to
1734 * them does not change anything */
1735 if (val > PR_FP_EXC_PRECISE)
1737 tsk->thread.fpexc_mode = __pack_fe01(val);
1738 if (regs != NULL && (regs->msr & MSR_FP) != 0)
1739 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1740 | tsk->thread.fpexc_mode;
1744 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1748 if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1750 if (cpu_has_feature(CPU_FTR_SPE)) {
1752 * When the sticky exception bits are set
1753 * directly by userspace, it must call prctl
1754 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1755 * in the existing prctl settings) or
1756 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1757 * the bits being set). <fenv.h> functions
1758 * saving and restoring the whole
1759 * floating-point environment need to do so
1760 * anyway to restore the prctl settings from
1761 * the saved environment.
1763 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1764 val = tsk->thread.fpexc_mode;
1771 val = __unpack_fe01(tsk->thread.fpexc_mode);
1772 return put_user(val, (unsigned int __user *) adr);
1775 int set_endian(struct task_struct *tsk, unsigned int val)
1777 struct pt_regs *regs = tsk->thread.regs;
1779 if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1780 (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1786 if (val == PR_ENDIAN_BIG)
1787 regs->msr &= ~MSR_LE;
1788 else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1789 regs->msr |= MSR_LE;
1796 int get_endian(struct task_struct *tsk, unsigned long adr)
1798 struct pt_regs *regs = tsk->thread.regs;
1801 if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1802 !cpu_has_feature(CPU_FTR_REAL_LE))
1808 if (regs->msr & MSR_LE) {
1809 if (cpu_has_feature(CPU_FTR_REAL_LE))
1810 val = PR_ENDIAN_LITTLE;
1812 val = PR_ENDIAN_PPC_LITTLE;
1814 val = PR_ENDIAN_BIG;
1816 return put_user(val, (unsigned int __user *)adr);
1819 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1821 tsk->thread.align_ctl = val;
1825 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1827 return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1830 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1831 unsigned long nbytes)
1833 unsigned long stack_page;
1834 unsigned long cpu = task_cpu(p);
1837 * Avoid crashing if the stack has overflowed and corrupted
1838 * task_cpu(p), which is in the thread_info struct.
1840 if (cpu < NR_CPUS && cpu_possible(cpu)) {
1841 stack_page = (unsigned long) hardirq_ctx[cpu];
1842 if (sp >= stack_page + sizeof(struct thread_struct)
1843 && sp <= stack_page + THREAD_SIZE - nbytes)
1846 stack_page = (unsigned long) softirq_ctx[cpu];
1847 if (sp >= stack_page + sizeof(struct thread_struct)
1848 && sp <= stack_page + THREAD_SIZE - nbytes)
1854 int validate_sp(unsigned long sp, struct task_struct *p,
1855 unsigned long nbytes)
1857 unsigned long stack_page = (unsigned long)task_stack_page(p);
1859 if (sp >= stack_page + sizeof(struct thread_struct)
1860 && sp <= stack_page + THREAD_SIZE - nbytes)
1863 return valid_irq_stack(sp, p, nbytes);
1866 EXPORT_SYMBOL(validate_sp);
1868 unsigned long get_wchan(struct task_struct *p)
1870 unsigned long ip, sp;
1873 if (!p || p == current || p->state == TASK_RUNNING)
1877 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1881 sp = *(unsigned long *)sp;
1882 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1885 ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1886 if (!in_sched_functions(ip))
1889 } while (count++ < 16);
1893 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1895 void show_stack(struct task_struct *tsk, unsigned long *stack)
1897 unsigned long sp, ip, lr, newsp;
1900 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1901 int curr_frame = current->curr_ret_stack;
1902 extern void return_to_handler(void);
1903 unsigned long rth = (unsigned long)return_to_handler;
1906 sp = (unsigned long) stack;
1911 sp = current_stack_pointer();
1913 sp = tsk->thread.ksp;
1917 printk("Call Trace:\n");
1919 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1922 stack = (unsigned long *) sp;
1924 ip = stack[STACK_FRAME_LR_SAVE];
1925 if (!firstframe || ip != lr) {
1926 printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1927 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1928 if ((ip == rth) && curr_frame >= 0) {
1930 (void *)current->ret_stack[curr_frame].ret);
1935 pr_cont(" (unreliable)");
1941 * See if this is an exception frame.
1942 * We look for the "regshere" marker in the current frame.
1944 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1945 && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1946 struct pt_regs *regs = (struct pt_regs *)
1947 (sp + STACK_FRAME_OVERHEAD);
1949 printk("--- interrupt: %lx at %pS\n LR = %pS\n",
1950 regs->trap, (void *)regs->nip, (void *)lr);
1955 } while (count++ < kstack_depth_to_print);
1959 /* Called with hard IRQs off */
1960 void notrace __ppc64_runlatch_on(void)
1962 struct thread_info *ti = current_thread_info();
1965 ctrl = mfspr(SPRN_CTRLF);
1966 ctrl |= CTRL_RUNLATCH;
1967 mtspr(SPRN_CTRLT, ctrl);
1969 ti->local_flags |= _TLF_RUNLATCH;
1972 /* Called with hard IRQs off */
1973 void notrace __ppc64_runlatch_off(void)
1975 struct thread_info *ti = current_thread_info();
1978 ti->local_flags &= ~_TLF_RUNLATCH;
1980 ctrl = mfspr(SPRN_CTRLF);
1981 ctrl &= ~CTRL_RUNLATCH;
1982 mtspr(SPRN_CTRLT, ctrl);
1984 #endif /* CONFIG_PPC64 */
1986 unsigned long arch_align_stack(unsigned long sp)
1988 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1989 sp -= get_random_int() & ~PAGE_MASK;
1993 static inline unsigned long brk_rnd(void)
1995 unsigned long rnd = 0;
1997 /* 8MB for 32bit, 1GB for 64bit */
1998 if (is_32bit_task())
1999 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2001 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2003 return rnd << PAGE_SHIFT;
2006 unsigned long arch_randomize_brk(struct mm_struct *mm)
2008 unsigned long base = mm->brk;
2011 #ifdef CONFIG_PPC_STD_MMU_64
2013 * If we are using 1TB segments and we are allowed to randomise
2014 * the heap, we can put it above 1TB so it is backed by a 1TB
2015 * segment. Otherwise the heap will be in the bottom 1TB
2016 * which always uses 256MB segments and this may result in a
2017 * performance penalty. We don't need to worry about radix. For
2018 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2020 if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2021 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2024 ret = PAGE_ALIGN(base + brk_rnd());