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>
45 #include <linux/pkeys.h>
47 #include <asm/pgtable.h>
49 #include <asm/processor.h>
52 #include <asm/machdep.h>
54 #include <asm/runlatch.h>
55 #include <asm/syscalls.h>
56 #include <asm/switch_to.h>
58 #include <asm/debug.h>
60 #include <asm/firmware.h>
61 #include <asm/hw_irq.h>
63 #include <asm/code-patching.h>
65 #include <asm/livepatch.h>
66 #include <asm/cpu_has_feature.h>
67 #include <asm/asm-prototypes.h>
69 #include <linux/kprobes.h>
70 #include <linux/kdebug.h>
72 /* Transactional Memory debug */
74 #define TM_DEBUG(x...) printk(KERN_INFO x)
76 #define TM_DEBUG(x...) do { } while(0)
79 extern unsigned long _get_SP(void);
81 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
83 * Are we running in "Suspend disabled" mode? If so we have to block any
84 * sigreturn that would get us into suspended state, and we also warn in some
85 * other paths that we should never reach with suspend disabled.
87 bool tm_suspend_disabled __ro_after_init = false;
89 static void check_if_tm_restore_required(struct task_struct *tsk)
92 * If we are saving the current thread's registers, and the
93 * thread is in a transactional state, set the TIF_RESTORE_TM
94 * bit so that we know to restore the registers before
95 * returning to userspace.
97 if (tsk == current && tsk->thread.regs &&
98 MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
99 !test_thread_flag(TIF_RESTORE_TM)) {
100 tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
101 set_thread_flag(TIF_RESTORE_TM);
105 static inline bool msr_tm_active(unsigned long msr)
107 return MSR_TM_ACTIVE(msr);
110 static bool tm_active_with_fp(struct task_struct *tsk)
112 return msr_tm_active(tsk->thread.regs->msr) &&
113 (tsk->thread.ckpt_regs.msr & MSR_FP);
116 static bool tm_active_with_altivec(struct task_struct *tsk)
118 return msr_tm_active(tsk->thread.regs->msr) &&
119 (tsk->thread.ckpt_regs.msr & MSR_VEC);
122 static inline bool msr_tm_active(unsigned long msr) { return false; }
123 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
124 static inline bool tm_active_with_fp(struct task_struct *tsk) { return false; }
125 static inline bool tm_active_with_altivec(struct task_struct *tsk) { return false; }
126 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
128 bool strict_msr_control;
129 EXPORT_SYMBOL(strict_msr_control);
131 static int __init enable_strict_msr_control(char *str)
133 strict_msr_control = true;
134 pr_info("Enabling strict facility control\n");
138 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
140 unsigned long msr_check_and_set(unsigned long bits)
142 unsigned long oldmsr = mfmsr();
143 unsigned long newmsr;
145 newmsr = oldmsr | bits;
148 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
152 if (oldmsr != newmsr)
157 EXPORT_SYMBOL_GPL(msr_check_and_set);
159 void __msr_check_and_clear(unsigned long bits)
161 unsigned long oldmsr = mfmsr();
162 unsigned long newmsr;
164 newmsr = oldmsr & ~bits;
167 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
171 if (oldmsr != newmsr)
174 EXPORT_SYMBOL(__msr_check_and_clear);
176 #ifdef CONFIG_PPC_FPU
177 static void __giveup_fpu(struct task_struct *tsk)
182 msr = tsk->thread.regs->msr;
185 if (cpu_has_feature(CPU_FTR_VSX))
188 tsk->thread.regs->msr = msr;
191 void giveup_fpu(struct task_struct *tsk)
193 check_if_tm_restore_required(tsk);
195 msr_check_and_set(MSR_FP);
197 msr_check_and_clear(MSR_FP);
199 EXPORT_SYMBOL(giveup_fpu);
202 * Make sure the floating-point register state in the
203 * the thread_struct is up to date for task tsk.
205 void flush_fp_to_thread(struct task_struct *tsk)
207 if (tsk->thread.regs) {
209 * We need to disable preemption here because if we didn't,
210 * another process could get scheduled after the regs->msr
211 * test but before we have finished saving the FP registers
212 * to the thread_struct. That process could take over the
213 * FPU, and then when we get scheduled again we would store
214 * bogus values for the remaining FP registers.
217 if (tsk->thread.regs->msr & MSR_FP) {
219 * This should only ever be called for current or
220 * for a stopped child process. Since we save away
221 * the FP register state on context switch,
222 * there is something wrong if a stopped child appears
223 * to still have its FP state in the CPU registers.
225 BUG_ON(tsk != current);
231 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
233 void enable_kernel_fp(void)
235 unsigned long cpumsr;
237 WARN_ON(preemptible());
239 cpumsr = msr_check_and_set(MSR_FP);
241 if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
242 check_if_tm_restore_required(current);
244 * If a thread has already been reclaimed then the
245 * checkpointed registers are on the CPU but have definitely
246 * been saved by the reclaim code. Don't need to and *cannot*
247 * giveup as this would save to the 'live' structure not the
248 * checkpointed structure.
250 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
252 __giveup_fpu(current);
255 EXPORT_SYMBOL(enable_kernel_fp);
257 static int restore_fp(struct task_struct *tsk)
259 if (tsk->thread.load_fp || tm_active_with_fp(tsk)) {
260 load_fp_state(¤t->thread.fp_state);
261 current->thread.load_fp++;
267 static int restore_fp(struct task_struct *tsk) { return 0; }
268 #endif /* CONFIG_PPC_FPU */
270 #ifdef CONFIG_ALTIVEC
271 #define loadvec(thr) ((thr).load_vec)
273 static void __giveup_altivec(struct task_struct *tsk)
278 msr = tsk->thread.regs->msr;
281 if (cpu_has_feature(CPU_FTR_VSX))
284 tsk->thread.regs->msr = msr;
287 void giveup_altivec(struct task_struct *tsk)
289 check_if_tm_restore_required(tsk);
291 msr_check_and_set(MSR_VEC);
292 __giveup_altivec(tsk);
293 msr_check_and_clear(MSR_VEC);
295 EXPORT_SYMBOL(giveup_altivec);
297 void enable_kernel_altivec(void)
299 unsigned long cpumsr;
301 WARN_ON(preemptible());
303 cpumsr = msr_check_and_set(MSR_VEC);
305 if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
306 check_if_tm_restore_required(current);
308 * If a thread has already been reclaimed then the
309 * checkpointed registers are on the CPU but have definitely
310 * been saved by the reclaim code. Don't need to and *cannot*
311 * giveup as this would save to the 'live' structure not the
312 * checkpointed structure.
314 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
316 __giveup_altivec(current);
319 EXPORT_SYMBOL(enable_kernel_altivec);
322 * Make sure the VMX/Altivec register state in the
323 * the thread_struct is up to date for task tsk.
325 void flush_altivec_to_thread(struct task_struct *tsk)
327 if (tsk->thread.regs) {
329 if (tsk->thread.regs->msr & MSR_VEC) {
330 BUG_ON(tsk != current);
336 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
338 static int restore_altivec(struct task_struct *tsk)
340 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
341 (tsk->thread.load_vec || tm_active_with_altivec(tsk))) {
342 load_vr_state(&tsk->thread.vr_state);
343 tsk->thread.used_vr = 1;
344 tsk->thread.load_vec++;
351 #define loadvec(thr) 0
352 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
353 #endif /* CONFIG_ALTIVEC */
356 static void __giveup_vsx(struct task_struct *tsk)
358 unsigned long msr = tsk->thread.regs->msr;
361 * We should never be ssetting MSR_VSX without also setting
364 WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
366 /* __giveup_fpu will clear MSR_VSX */
370 __giveup_altivec(tsk);
373 static void giveup_vsx(struct task_struct *tsk)
375 check_if_tm_restore_required(tsk);
377 msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
379 msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
382 void enable_kernel_vsx(void)
384 unsigned long cpumsr;
386 WARN_ON(preemptible());
388 cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
390 if (current->thread.regs &&
391 (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
392 check_if_tm_restore_required(current);
394 * If a thread has already been reclaimed then the
395 * checkpointed registers are on the CPU but have definitely
396 * been saved by the reclaim code. Don't need to and *cannot*
397 * giveup as this would save to the 'live' structure not the
398 * checkpointed structure.
400 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
402 __giveup_vsx(current);
405 EXPORT_SYMBOL(enable_kernel_vsx);
407 void flush_vsx_to_thread(struct task_struct *tsk)
409 if (tsk->thread.regs) {
411 if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
412 BUG_ON(tsk != current);
418 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
420 static int restore_vsx(struct task_struct *tsk)
422 if (cpu_has_feature(CPU_FTR_VSX)) {
423 tsk->thread.used_vsr = 1;
430 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
431 #endif /* CONFIG_VSX */
434 void giveup_spe(struct task_struct *tsk)
436 check_if_tm_restore_required(tsk);
438 msr_check_and_set(MSR_SPE);
440 msr_check_and_clear(MSR_SPE);
442 EXPORT_SYMBOL(giveup_spe);
444 void enable_kernel_spe(void)
446 WARN_ON(preemptible());
448 msr_check_and_set(MSR_SPE);
450 if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
451 check_if_tm_restore_required(current);
452 __giveup_spe(current);
455 EXPORT_SYMBOL(enable_kernel_spe);
457 void flush_spe_to_thread(struct task_struct *tsk)
459 if (tsk->thread.regs) {
461 if (tsk->thread.regs->msr & MSR_SPE) {
462 BUG_ON(tsk != current);
463 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
469 #endif /* CONFIG_SPE */
471 static unsigned long msr_all_available;
473 static int __init init_msr_all_available(void)
475 #ifdef CONFIG_PPC_FPU
476 msr_all_available |= MSR_FP;
478 #ifdef CONFIG_ALTIVEC
479 if (cpu_has_feature(CPU_FTR_ALTIVEC))
480 msr_all_available |= MSR_VEC;
483 if (cpu_has_feature(CPU_FTR_VSX))
484 msr_all_available |= MSR_VSX;
487 if (cpu_has_feature(CPU_FTR_SPE))
488 msr_all_available |= MSR_SPE;
493 early_initcall(init_msr_all_available);
495 void giveup_all(struct task_struct *tsk)
497 unsigned long usermsr;
499 if (!tsk->thread.regs)
502 usermsr = tsk->thread.regs->msr;
504 if ((usermsr & msr_all_available) == 0)
507 msr_check_and_set(msr_all_available);
508 check_if_tm_restore_required(tsk);
510 WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
512 #ifdef CONFIG_PPC_FPU
513 if (usermsr & MSR_FP)
516 #ifdef CONFIG_ALTIVEC
517 if (usermsr & MSR_VEC)
518 __giveup_altivec(tsk);
521 if (usermsr & MSR_SPE)
525 msr_check_and_clear(msr_all_available);
527 EXPORT_SYMBOL(giveup_all);
529 void restore_math(struct pt_regs *regs)
533 if (!msr_tm_active(regs->msr) &&
534 !current->thread.load_fp && !loadvec(current->thread))
538 msr_check_and_set(msr_all_available);
541 * Only reload if the bit is not set in the user MSR, the bit BEING set
542 * indicates that the registers are hot
544 if ((!(msr & MSR_FP)) && restore_fp(current))
545 msr |= MSR_FP | current->thread.fpexc_mode;
547 if ((!(msr & MSR_VEC)) && restore_altivec(current))
550 if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
551 restore_vsx(current)) {
555 msr_check_and_clear(msr_all_available);
560 static void save_all(struct task_struct *tsk)
562 unsigned long usermsr;
564 if (!tsk->thread.regs)
567 usermsr = tsk->thread.regs->msr;
569 if ((usermsr & msr_all_available) == 0)
572 msr_check_and_set(msr_all_available);
574 WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
576 if (usermsr & MSR_FP)
579 if (usermsr & MSR_VEC)
582 if (usermsr & MSR_SPE)
585 msr_check_and_clear(msr_all_available);
586 thread_pkey_regs_save(&tsk->thread);
589 void flush_all_to_thread(struct task_struct *tsk)
591 if (tsk->thread.regs) {
593 BUG_ON(tsk != current);
597 if (tsk->thread.regs->msr & MSR_SPE)
598 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
604 EXPORT_SYMBOL(flush_all_to_thread);
606 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
607 void do_send_trap(struct pt_regs *regs, unsigned long address,
608 unsigned long error_code, int breakpt)
610 current->thread.trap_nr = TRAP_HWBKPT;
611 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
612 11, SIGSEGV) == NOTIFY_STOP)
615 /* Deliver the signal to userspace */
616 force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
617 (void __user *)address);
619 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
620 void do_break (struct pt_regs *regs, unsigned long address,
621 unsigned long error_code)
623 current->thread.trap_nr = TRAP_HWBKPT;
624 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
625 11, SIGSEGV) == NOTIFY_STOP)
628 if (debugger_break_match(regs))
631 /* Clear the breakpoint */
632 hw_breakpoint_disable();
634 /* Deliver the signal to userspace */
635 force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)address, current);
637 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
639 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
641 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
643 * Set the debug registers back to their default "safe" values.
645 static void set_debug_reg_defaults(struct thread_struct *thread)
647 thread->debug.iac1 = thread->debug.iac2 = 0;
648 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
649 thread->debug.iac3 = thread->debug.iac4 = 0;
651 thread->debug.dac1 = thread->debug.dac2 = 0;
652 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
653 thread->debug.dvc1 = thread->debug.dvc2 = 0;
655 thread->debug.dbcr0 = 0;
658 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
660 thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
661 DBCR1_IAC3US | DBCR1_IAC4US;
663 * Force Data Address Compare User/Supervisor bits to be User-only
664 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
666 thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
668 thread->debug.dbcr1 = 0;
672 static void prime_debug_regs(struct debug_reg *debug)
675 * We could have inherited MSR_DE from userspace, since
676 * it doesn't get cleared on exception entry. Make sure
677 * MSR_DE is clear before we enable any debug events.
679 mtmsr(mfmsr() & ~MSR_DE);
681 mtspr(SPRN_IAC1, debug->iac1);
682 mtspr(SPRN_IAC2, debug->iac2);
683 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
684 mtspr(SPRN_IAC3, debug->iac3);
685 mtspr(SPRN_IAC4, debug->iac4);
687 mtspr(SPRN_DAC1, debug->dac1);
688 mtspr(SPRN_DAC2, debug->dac2);
689 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
690 mtspr(SPRN_DVC1, debug->dvc1);
691 mtspr(SPRN_DVC2, debug->dvc2);
693 mtspr(SPRN_DBCR0, debug->dbcr0);
694 mtspr(SPRN_DBCR1, debug->dbcr1);
696 mtspr(SPRN_DBCR2, debug->dbcr2);
700 * Unless neither the old or new thread are making use of the
701 * debug registers, set the debug registers from the values
702 * stored in the new thread.
704 void switch_booke_debug_regs(struct debug_reg *new_debug)
706 if ((current->thread.debug.dbcr0 & DBCR0_IDM)
707 || (new_debug->dbcr0 & DBCR0_IDM))
708 prime_debug_regs(new_debug);
710 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
711 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
712 #ifndef CONFIG_HAVE_HW_BREAKPOINT
713 static void set_breakpoint(struct arch_hw_breakpoint *brk)
716 __set_breakpoint(brk);
720 static void set_debug_reg_defaults(struct thread_struct *thread)
722 thread->hw_brk.address = 0;
723 thread->hw_brk.type = 0;
724 if (ppc_breakpoint_available())
725 set_breakpoint(&thread->hw_brk);
727 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
728 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
730 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
731 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
733 mtspr(SPRN_DAC1, dabr);
734 #ifdef CONFIG_PPC_47x
739 #elif defined(CONFIG_PPC_BOOK3S)
740 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
742 mtspr(SPRN_DABR, dabr);
743 if (cpu_has_feature(CPU_FTR_DABRX))
744 mtspr(SPRN_DABRX, dabrx);
747 #elif defined(CONFIG_PPC_8xx)
748 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
750 unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
751 unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
752 unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
754 if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
756 else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
758 else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
761 mtspr(SPRN_LCTRL2, 0);
762 mtspr(SPRN_CMPE, addr);
763 mtspr(SPRN_CMPF, addr + 4);
764 mtspr(SPRN_LCTRL1, lctrl1);
765 mtspr(SPRN_LCTRL2, lctrl2);
770 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
776 static inline int set_dabr(struct arch_hw_breakpoint *brk)
778 unsigned long dabr, dabrx;
780 dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
781 dabrx = ((brk->type >> 3) & 0x7);
784 return ppc_md.set_dabr(dabr, dabrx);
786 return __set_dabr(dabr, dabrx);
789 static inline int set_dawr(struct arch_hw_breakpoint *brk)
791 unsigned long dawr, dawrx, mrd;
795 dawrx = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
796 << (63 - 58); //* read/write bits */
797 dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
798 << (63 - 59); //* translate */
799 dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
800 >> 3; //* PRIM bits */
801 /* dawr length is stored in field MDR bits 48:53. Matches range in
802 doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
804 brk->len is in bytes.
805 This aligns up to double word size, shifts and does the bias.
807 mrd = ((brk->len + 7) >> 3) - 1;
808 dawrx |= (mrd & 0x3f) << (63 - 53);
811 return ppc_md.set_dawr(dawr, dawrx);
812 mtspr(SPRN_DAWR, dawr);
813 mtspr(SPRN_DAWRX, dawrx);
817 void __set_breakpoint(struct arch_hw_breakpoint *brk)
819 memcpy(this_cpu_ptr(¤t_brk), brk, sizeof(*brk));
821 if (cpu_has_feature(CPU_FTR_DAWR))
824 else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
828 // Shouldn't happen due to higher level checks
832 /* Check if we have DAWR or DABR hardware */
833 bool ppc_breakpoint_available(void)
835 if (cpu_has_feature(CPU_FTR_DAWR))
836 return true; /* POWER8 DAWR */
837 if (cpu_has_feature(CPU_FTR_ARCH_207S))
838 return false; /* POWER9 with DAWR disabled */
839 /* DABR: Everything but POWER8 and POWER9 */
842 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
844 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
845 struct arch_hw_breakpoint *b)
847 if (a->address != b->address)
849 if (a->type != b->type)
851 if (a->len != b->len)
856 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
858 static inline bool tm_enabled(struct task_struct *tsk)
860 return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
863 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
866 * Use the current MSR TM suspended bit to track if we have
867 * checkpointed state outstanding.
868 * On signal delivery, we'd normally reclaim the checkpointed
869 * state to obtain stack pointer (see:get_tm_stackpointer()).
870 * This will then directly return to userspace without going
871 * through __switch_to(). However, if the stack frame is bad,
872 * we need to exit this thread which calls __switch_to() which
873 * will again attempt to reclaim the already saved tm state.
874 * Hence we need to check that we've not already reclaimed
876 * We do this using the current MSR, rather tracking it in
877 * some specific thread_struct bit, as it has the additional
878 * benefit of checking for a potential TM bad thing exception.
880 if (!MSR_TM_SUSPENDED(mfmsr()))
883 giveup_all(container_of(thr, struct task_struct, thread));
885 tm_reclaim(thr, cause);
888 * If we are in a transaction and FP is off then we can't have
889 * used FP inside that transaction. Hence the checkpointed
890 * state is the same as the live state. We need to copy the
891 * live state to the checkpointed state so that when the
892 * transaction is restored, the checkpointed state is correct
893 * and the aborted transaction sees the correct state. We use
894 * ckpt_regs.msr here as that's what tm_reclaim will use to
895 * determine if it's going to write the checkpointed state or
896 * not. So either this will write the checkpointed registers,
897 * or reclaim will. Similarly for VMX.
899 if ((thr->ckpt_regs.msr & MSR_FP) == 0)
900 memcpy(&thr->ckfp_state, &thr->fp_state,
901 sizeof(struct thread_fp_state));
902 if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
903 memcpy(&thr->ckvr_state, &thr->vr_state,
904 sizeof(struct thread_vr_state));
907 void tm_reclaim_current(uint8_t cause)
910 tm_reclaim_thread(¤t->thread, cause);
913 static inline void tm_reclaim_task(struct task_struct *tsk)
915 /* We have to work out if we're switching from/to a task that's in the
916 * middle of a transaction.
918 * In switching we need to maintain a 2nd register state as
919 * oldtask->thread.ckpt_regs. We tm_reclaim(oldproc); this saves the
920 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
923 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
925 struct thread_struct *thr = &tsk->thread;
930 if (!MSR_TM_ACTIVE(thr->regs->msr))
931 goto out_and_saveregs;
933 WARN_ON(tm_suspend_disabled);
935 TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
936 "ccr=%lx, msr=%lx, trap=%lx)\n",
937 tsk->pid, thr->regs->nip,
938 thr->regs->ccr, thr->regs->msr,
941 tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
943 TM_DEBUG("--- tm_reclaim on pid %d complete\n",
947 /* Always save the regs here, even if a transaction's not active.
948 * This context-switches a thread's TM info SPRs. We do it here to
949 * be consistent with the restore path (in recheckpoint) which
950 * cannot happen later in _switch().
955 extern void __tm_recheckpoint(struct thread_struct *thread);
957 void tm_recheckpoint(struct thread_struct *thread)
961 if (!(thread->regs->msr & MSR_TM))
964 /* We really can't be interrupted here as the TEXASR registers can't
965 * change and later in the trecheckpoint code, we have a userspace R1.
966 * So let's hard disable over this region.
968 local_irq_save(flags);
971 /* The TM SPRs are restored here, so that TEXASR.FS can be set
972 * before the trecheckpoint and no explosion occurs.
974 tm_restore_sprs(thread);
976 __tm_recheckpoint(thread);
978 local_irq_restore(flags);
981 static inline void tm_recheckpoint_new_task(struct task_struct *new)
983 if (!cpu_has_feature(CPU_FTR_TM))
986 /* Recheckpoint the registers of the thread we're about to switch to.
988 * If the task was using FP, we non-lazily reload both the original and
989 * the speculative FP register states. This is because the kernel
990 * doesn't see if/when a TM rollback occurs, so if we take an FP
991 * unavailable later, we are unable to determine which set of FP regs
992 * need to be restored.
994 if (!tm_enabled(new))
997 if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
998 tm_restore_sprs(&new->thread);
1001 /* Recheckpoint to restore original checkpointed register state. */
1002 TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1003 new->pid, new->thread.regs->msr);
1005 tm_recheckpoint(&new->thread);
1008 * The checkpointed state has been restored but the live state has
1009 * not, ensure all the math functionality is turned off to trigger
1010 * restore_math() to reload.
1012 new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1014 TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1015 "(kernel msr 0x%lx)\n",
1019 static inline void __switch_to_tm(struct task_struct *prev,
1020 struct task_struct *new)
1022 if (cpu_has_feature(CPU_FTR_TM)) {
1023 if (tm_enabled(prev) || tm_enabled(new))
1026 if (tm_enabled(prev)) {
1027 prev->thread.load_tm++;
1028 tm_reclaim_task(prev);
1029 if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1030 prev->thread.regs->msr &= ~MSR_TM;
1033 tm_recheckpoint_new_task(new);
1038 * This is called if we are on the way out to userspace and the
1039 * TIF_RESTORE_TM flag is set. It checks if we need to reload
1040 * FP and/or vector state and does so if necessary.
1041 * If userspace is inside a transaction (whether active or
1042 * suspended) and FP/VMX/VSX instructions have ever been enabled
1043 * inside that transaction, then we have to keep them enabled
1044 * and keep the FP/VMX/VSX state loaded while ever the transaction
1045 * continues. The reason is that if we didn't, and subsequently
1046 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1047 * we don't know whether it's the same transaction, and thus we
1048 * don't know which of the checkpointed state and the transactional
1051 void restore_tm_state(struct pt_regs *regs)
1053 unsigned long msr_diff;
1056 * This is the only moment we should clear TIF_RESTORE_TM as
1057 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1058 * again, anything else could lead to an incorrect ckpt_msr being
1059 * saved and therefore incorrect signal contexts.
1061 clear_thread_flag(TIF_RESTORE_TM);
1062 if (!MSR_TM_ACTIVE(regs->msr))
1065 msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1066 msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1068 /* Ensure that restore_math() will restore */
1069 if (msr_diff & MSR_FP)
1070 current->thread.load_fp = 1;
1071 #ifdef CONFIG_ALTIVEC
1072 if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1073 current->thread.load_vec = 1;
1077 regs->msr |= msr_diff;
1081 #define tm_recheckpoint_new_task(new)
1082 #define __switch_to_tm(prev, new)
1083 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1085 static inline void save_sprs(struct thread_struct *t)
1087 #ifdef CONFIG_ALTIVEC
1088 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1089 t->vrsave = mfspr(SPRN_VRSAVE);
1091 #ifdef CONFIG_PPC_BOOK3S_64
1092 if (cpu_has_feature(CPU_FTR_DSCR))
1093 t->dscr = mfspr(SPRN_DSCR);
1095 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1096 t->bescr = mfspr(SPRN_BESCR);
1097 t->ebbhr = mfspr(SPRN_EBBHR);
1098 t->ebbrr = mfspr(SPRN_EBBRR);
1100 t->fscr = mfspr(SPRN_FSCR);
1103 * Note that the TAR is not available for use in the kernel.
1104 * (To provide this, the TAR should be backed up/restored on
1105 * exception entry/exit instead, and be in pt_regs. FIXME,
1106 * this should be in pt_regs anyway (for debug).)
1108 t->tar = mfspr(SPRN_TAR);
1112 thread_pkey_regs_save(t);
1115 static inline void restore_sprs(struct thread_struct *old_thread,
1116 struct thread_struct *new_thread)
1118 #ifdef CONFIG_ALTIVEC
1119 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1120 old_thread->vrsave != new_thread->vrsave)
1121 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1123 #ifdef CONFIG_PPC_BOOK3S_64
1124 if (cpu_has_feature(CPU_FTR_DSCR)) {
1125 u64 dscr = get_paca()->dscr_default;
1126 if (new_thread->dscr_inherit)
1127 dscr = new_thread->dscr;
1129 if (old_thread->dscr != dscr)
1130 mtspr(SPRN_DSCR, dscr);
1133 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1134 if (old_thread->bescr != new_thread->bescr)
1135 mtspr(SPRN_BESCR, new_thread->bescr);
1136 if (old_thread->ebbhr != new_thread->ebbhr)
1137 mtspr(SPRN_EBBHR, new_thread->ebbhr);
1138 if (old_thread->ebbrr != new_thread->ebbrr)
1139 mtspr(SPRN_EBBRR, new_thread->ebbrr);
1141 if (old_thread->fscr != new_thread->fscr)
1142 mtspr(SPRN_FSCR, new_thread->fscr);
1144 if (old_thread->tar != new_thread->tar)
1145 mtspr(SPRN_TAR, new_thread->tar);
1148 if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1149 old_thread->tidr != new_thread->tidr)
1150 mtspr(SPRN_TIDR, new_thread->tidr);
1153 thread_pkey_regs_restore(new_thread, old_thread);
1156 #ifdef CONFIG_PPC_BOOK3S_64
1158 static const u8 dummy_copy_buffer[CP_SIZE] __attribute__((aligned(CP_SIZE)));
1161 struct task_struct *__switch_to(struct task_struct *prev,
1162 struct task_struct *new)
1164 struct thread_struct *new_thread, *old_thread;
1165 struct task_struct *last;
1166 #ifdef CONFIG_PPC_BOOK3S_64
1167 struct ppc64_tlb_batch *batch;
1170 new_thread = &new->thread;
1171 old_thread = ¤t->thread;
1173 WARN_ON(!irqs_disabled());
1175 #ifdef CONFIG_PPC_BOOK3S_64
1176 batch = this_cpu_ptr(&ppc64_tlb_batch);
1177 if (batch->active) {
1178 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1180 __flush_tlb_pending(batch);
1183 #endif /* CONFIG_PPC_BOOK3S_64 */
1185 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1186 switch_booke_debug_regs(&new->thread.debug);
1189 * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1192 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1193 if (unlikely(!hw_brk_match(this_cpu_ptr(¤t_brk), &new->thread.hw_brk)))
1194 __set_breakpoint(&new->thread.hw_brk);
1195 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1199 * We need to save SPRs before treclaim/trecheckpoint as these will
1200 * change a number of them.
1202 save_sprs(&prev->thread);
1204 /* Save FPU, Altivec, VSX and SPE state */
1207 __switch_to_tm(prev, new);
1209 if (!radix_enabled()) {
1211 * We can't take a PMU exception inside _switch() since there
1212 * is a window where the kernel stack SLB and the kernel stack
1213 * are out of sync. Hard disable here.
1219 * Call restore_sprs() before calling _switch(). If we move it after
1220 * _switch() then we miss out on calling it for new tasks. The reason
1221 * for this is we manually create a stack frame for new tasks that
1222 * directly returns through ret_from_fork() or
1223 * ret_from_kernel_thread(). See copy_thread() for details.
1225 restore_sprs(old_thread, new_thread);
1227 last = _switch(old_thread, new_thread);
1229 #ifdef CONFIG_PPC_BOOK3S_64
1230 if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1231 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1232 batch = this_cpu_ptr(&ppc64_tlb_batch);
1236 if (current_thread_info()->task->thread.regs) {
1237 restore_math(current_thread_info()->task->thread.regs);
1240 * The copy-paste buffer can only store into foreign real
1241 * addresses, so unprivileged processes can not see the
1242 * data or use it in any way unless they have foreign real
1243 * mappings. If the new process has the foreign real address
1244 * mappings, we must issue a cp_abort to clear any state and
1245 * prevent snooping, corruption or a covert channel.
1247 if (current_thread_info()->task->thread.used_vas)
1248 asm volatile(PPC_CP_ABORT);
1250 #endif /* CONFIG_PPC_BOOK3S_64 */
1255 static int instructions_to_print = 16;
1257 static void show_instructions(struct pt_regs *regs)
1260 unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
1263 printk("Instruction dump:");
1265 for (i = 0; i < instructions_to_print; i++) {
1271 #if !defined(CONFIG_BOOKE)
1272 /* If executing with the IMMU off, adjust pc rather
1273 * than print XXXXXXXX.
1275 if (!(regs->msr & MSR_IR))
1276 pc = (unsigned long)phys_to_virt(pc);
1279 if (!__kernel_text_address(pc) ||
1280 probe_kernel_address((unsigned int __user *)pc, instr)) {
1281 pr_cont("XXXXXXXX ");
1283 if (regs->nip == pc)
1284 pr_cont("<%08x> ", instr);
1286 pr_cont("%08x ", instr);
1295 void show_user_instructions(struct pt_regs *regs)
1300 pc = regs->nip - (instructions_to_print * 3 / 4 * sizeof(int));
1303 * Make sure the NIP points at userspace, not kernel text/data or
1306 if (!__access_ok(pc, instructions_to_print * sizeof(int), USER_DS)) {
1307 pr_info("%s[%d]: Bad NIP, not dumping instructions.\n",
1308 current->comm, current->pid);
1312 pr_info("%s[%d]: code: ", current->comm, current->pid);
1314 for (i = 0; i < instructions_to_print; i++) {
1317 if (!(i % 8) && (i > 0)) {
1319 pr_info("%s[%d]: code: ", current->comm, current->pid);
1322 if (probe_kernel_address((unsigned int __user *)pc, instr)) {
1323 pr_cont("XXXXXXXX ");
1325 if (regs->nip == pc)
1326 pr_cont("<%08x> ", instr);
1328 pr_cont("%08x ", instr);
1342 static struct regbit msr_bits[] = {
1343 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1365 #ifndef CONFIG_BOOKE
1372 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1376 for (; bits->bit; ++bits)
1377 if (val & bits->bit) {
1378 pr_cont("%s%s", s, bits->name);
1383 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1384 static struct regbit msr_tm_bits[] = {
1391 static void print_tm_bits(unsigned long val)
1394 * This only prints something if at least one of the TM bit is set.
1395 * Inside the TM[], the output means:
1396 * E: Enabled (bit 32)
1397 * S: Suspended (bit 33)
1398 * T: Transactional (bit 34)
1400 if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1402 print_bits(val, msr_tm_bits, "");
1407 static void print_tm_bits(unsigned long val) {}
1410 static void print_msr_bits(unsigned long val)
1413 print_bits(val, msr_bits, ",");
1419 #define REG "%016lx"
1420 #define REGS_PER_LINE 4
1421 #define LAST_VOLATILE 13
1424 #define REGS_PER_LINE 8
1425 #define LAST_VOLATILE 12
1428 void show_regs(struct pt_regs * regs)
1432 show_regs_print_info(KERN_DEFAULT);
1434 printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1435 regs->nip, regs->link, regs->ctr);
1436 printk("REGS: %px TRAP: %04lx %s (%s)\n",
1437 regs, regs->trap, print_tainted(), init_utsname()->release);
1438 printk("MSR: "REG" ", regs->msr);
1439 print_msr_bits(regs->msr);
1440 pr_cont(" CR: %08lx XER: %08lx\n", regs->ccr, regs->xer);
1442 if ((TRAP(regs) != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1443 pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1444 if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1445 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1446 pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1448 pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1451 pr_cont("IRQMASK: %lx ", regs->softe);
1453 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1454 if (MSR_TM_ACTIVE(regs->msr))
1455 pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1458 for (i = 0; i < 32; i++) {
1459 if ((i % REGS_PER_LINE) == 0)
1460 pr_cont("\nGPR%02d: ", i);
1461 pr_cont(REG " ", regs->gpr[i]);
1462 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1466 #ifdef CONFIG_KALLSYMS
1468 * Lookup NIP late so we have the best change of getting the
1469 * above info out without failing
1471 printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1472 printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1474 show_stack(current, (unsigned long *) regs->gpr[1]);
1475 if (!user_mode(regs))
1476 show_instructions(regs);
1479 void flush_thread(void)
1481 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1482 flush_ptrace_hw_breakpoint(current);
1483 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1484 set_debug_reg_defaults(¤t->thread);
1485 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1488 int set_thread_uses_vas(void)
1490 #ifdef CONFIG_PPC_BOOK3S_64
1491 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1494 current->thread.used_vas = 1;
1497 * Even a process that has no foreign real address mapping can use
1498 * an unpaired COPY instruction (to no real effect). Issue CP_ABORT
1499 * to clear any pending COPY and prevent a covert channel.
1501 * __switch_to() will issue CP_ABORT on future context switches.
1503 asm volatile(PPC_CP_ABORT);
1505 #endif /* CONFIG_PPC_BOOK3S_64 */
1511 * Assign a TIDR (thread ID) for task @t and set it in the thread
1512 * structure. For now, we only support setting TIDR for 'current' task.
1514 * Since the TID value is a truncated form of it PID, it is possible
1515 * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1516 * that 2 threads share the same TID and are waiting, one of the following
1517 * cases will happen:
1519 * 1. The correct thread is running, the wrong thread is not
1520 * In this situation, the correct thread is woken and proceeds to pass it's
1523 * 2. Neither threads are running
1524 * In this situation, neither thread will be woken. When scheduled, the waiting
1525 * threads will execute either a wait, which will return immediately, followed
1526 * by a condition check, which will pass for the correct thread and fail
1527 * for the wrong thread, or they will execute the condition check immediately.
1529 * 3. The wrong thread is running, the correct thread is not
1530 * The wrong thread will be woken, but will fail it's condition check and
1531 * re-execute wait. The correct thread, when scheduled, will execute either
1532 * it's condition check (which will pass), or wait, which returns immediately
1533 * when called the first time after the thread is scheduled, followed by it's
1534 * condition check (which will pass).
1536 * 4. Both threads are running
1537 * Both threads will be woken. The wrong thread will fail it's condition check
1538 * and execute another wait, while the correct thread will pass it's condition
1541 * @t: the task to set the thread ID for
1543 int set_thread_tidr(struct task_struct *t)
1545 if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1554 t->thread.tidr = (u16)task_pid_nr(t);
1555 mtspr(SPRN_TIDR, t->thread.tidr);
1559 EXPORT_SYMBOL_GPL(set_thread_tidr);
1561 #endif /* CONFIG_PPC64 */
1564 release_thread(struct task_struct *t)
1569 * this gets called so that we can store coprocessor state into memory and
1570 * copy the current task into the new thread.
1572 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1574 flush_all_to_thread(src);
1576 * Flush TM state out so we can copy it. __switch_to_tm() does this
1577 * flush but it removes the checkpointed state from the current CPU and
1578 * transitions the CPU out of TM mode. Hence we need to call
1579 * tm_recheckpoint_new_task() (on the same task) to restore the
1580 * checkpointed state back and the TM mode.
1582 * Can't pass dst because it isn't ready. Doesn't matter, passing
1583 * dst is only important for __switch_to()
1585 __switch_to_tm(src, src);
1589 clear_task_ebb(dst);
1594 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1596 #ifdef CONFIG_PPC_BOOK3S_64
1597 unsigned long sp_vsid;
1598 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1600 if (radix_enabled())
1603 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1604 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1605 << SLB_VSID_SHIFT_1T;
1607 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1609 sp_vsid |= SLB_VSID_KERNEL | llp;
1610 p->thread.ksp_vsid = sp_vsid;
1619 * Copy architecture-specific thread state
1621 int copy_thread(unsigned long clone_flags, unsigned long usp,
1622 unsigned long kthread_arg, struct task_struct *p)
1624 struct pt_regs *childregs, *kregs;
1625 extern void ret_from_fork(void);
1626 extern void ret_from_kernel_thread(void);
1628 unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1629 struct thread_info *ti = task_thread_info(p);
1631 klp_init_thread_info(ti);
1633 /* Copy registers */
1634 sp -= sizeof(struct pt_regs);
1635 childregs = (struct pt_regs *) sp;
1636 if (unlikely(p->flags & PF_KTHREAD)) {
1638 memset(childregs, 0, sizeof(struct pt_regs));
1639 childregs->gpr[1] = sp + sizeof(struct pt_regs);
1642 childregs->gpr[14] = ppc_function_entry((void *)usp);
1644 clear_tsk_thread_flag(p, TIF_32BIT);
1645 childregs->softe = IRQS_ENABLED;
1647 childregs->gpr[15] = kthread_arg;
1648 p->thread.regs = NULL; /* no user register state */
1649 ti->flags |= _TIF_RESTOREALL;
1650 f = ret_from_kernel_thread;
1653 struct pt_regs *regs = current_pt_regs();
1654 CHECK_FULL_REGS(regs);
1657 childregs->gpr[1] = usp;
1658 p->thread.regs = childregs;
1659 childregs->gpr[3] = 0; /* Result from fork() */
1660 if (clone_flags & CLONE_SETTLS) {
1662 if (!is_32bit_task())
1663 childregs->gpr[13] = childregs->gpr[6];
1666 childregs->gpr[2] = childregs->gpr[6];
1671 childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1672 sp -= STACK_FRAME_OVERHEAD;
1675 * The way this works is that at some point in the future
1676 * some task will call _switch to switch to the new task.
1677 * That will pop off the stack frame created below and start
1678 * the new task running at ret_from_fork. The new task will
1679 * do some house keeping and then return from the fork or clone
1680 * system call, using the stack frame created above.
1682 ((unsigned long *)sp)[0] = 0;
1683 sp -= sizeof(struct pt_regs);
1684 kregs = (struct pt_regs *) sp;
1685 sp -= STACK_FRAME_OVERHEAD;
1688 p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1689 _ALIGN_UP(sizeof(struct thread_info), 16);
1691 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1692 p->thread.ptrace_bps[0] = NULL;
1695 p->thread.fp_save_area = NULL;
1696 #ifdef CONFIG_ALTIVEC
1697 p->thread.vr_save_area = NULL;
1700 setup_ksp_vsid(p, sp);
1703 if (cpu_has_feature(CPU_FTR_DSCR)) {
1704 p->thread.dscr_inherit = current->thread.dscr_inherit;
1705 p->thread.dscr = mfspr(SPRN_DSCR);
1707 if (cpu_has_feature(CPU_FTR_HAS_PPR))
1708 p->thread.ppr = INIT_PPR;
1712 kregs->nip = ppc_function_entry(f);
1717 * Set up a thread for executing a new program
1719 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1722 unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1726 * If we exec out of a kernel thread then thread.regs will not be
1729 if (!current->thread.regs) {
1730 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1731 current->thread.regs = regs - 1;
1734 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1736 * Clear any transactional state, we're exec()ing. The cause is
1737 * not important as there will never be a recheckpoint so it's not
1740 if (MSR_TM_SUSPENDED(mfmsr()))
1741 tm_reclaim_current(0);
1744 memset(regs->gpr, 0, sizeof(regs->gpr));
1752 * We have just cleared all the nonvolatile GPRs, so make
1753 * FULL_REGS(regs) return true. This is necessary to allow
1754 * ptrace to examine the thread immediately after exec.
1761 regs->msr = MSR_USER;
1763 if (!is_32bit_task()) {
1764 unsigned long entry;
1766 if (is_elf2_task()) {
1767 /* Look ma, no function descriptors! */
1772 * The latest iteration of the ABI requires that when
1773 * calling a function (at its global entry point),
1774 * the caller must ensure r12 holds the entry point
1775 * address (so that the function can quickly
1776 * establish addressability).
1778 regs->gpr[12] = start;
1779 /* Make sure that's restored on entry to userspace. */
1780 set_thread_flag(TIF_RESTOREALL);
1784 /* start is a relocated pointer to the function
1785 * descriptor for the elf _start routine. The first
1786 * entry in the function descriptor is the entry
1787 * address of _start and the second entry is the TOC
1788 * value we need to use.
1790 __get_user(entry, (unsigned long __user *)start);
1791 __get_user(toc, (unsigned long __user *)start+1);
1793 /* Check whether the e_entry function descriptor entries
1794 * need to be relocated before we can use them.
1796 if (load_addr != 0) {
1803 regs->msr = MSR_USER64;
1807 regs->msr = MSR_USER32;
1811 current->thread.used_vsr = 0;
1813 current->thread.load_fp = 0;
1814 memset(¤t->thread.fp_state, 0, sizeof(current->thread.fp_state));
1815 current->thread.fp_save_area = NULL;
1816 #ifdef CONFIG_ALTIVEC
1817 memset(¤t->thread.vr_state, 0, sizeof(current->thread.vr_state));
1818 current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1819 current->thread.vr_save_area = NULL;
1820 current->thread.vrsave = 0;
1821 current->thread.used_vr = 0;
1822 current->thread.load_vec = 0;
1823 #endif /* CONFIG_ALTIVEC */
1825 memset(current->thread.evr, 0, sizeof(current->thread.evr));
1826 current->thread.acc = 0;
1827 current->thread.spefscr = 0;
1828 current->thread.used_spe = 0;
1829 #endif /* CONFIG_SPE */
1830 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1831 current->thread.tm_tfhar = 0;
1832 current->thread.tm_texasr = 0;
1833 current->thread.tm_tfiar = 0;
1834 current->thread.load_tm = 0;
1835 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1837 thread_pkey_regs_init(¤t->thread);
1839 EXPORT_SYMBOL(start_thread);
1841 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1842 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1844 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1846 struct pt_regs *regs = tsk->thread.regs;
1848 /* This is a bit hairy. If we are an SPE enabled processor
1849 * (have embedded fp) we store the IEEE exception enable flags in
1850 * fpexc_mode. fpexc_mode is also used for setting FP exception
1851 * mode (asyn, precise, disabled) for 'Classic' FP. */
1852 if (val & PR_FP_EXC_SW_ENABLE) {
1854 if (cpu_has_feature(CPU_FTR_SPE)) {
1856 * When the sticky exception bits are set
1857 * directly by userspace, it must call prctl
1858 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1859 * in the existing prctl settings) or
1860 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1861 * the bits being set). <fenv.h> functions
1862 * saving and restoring the whole
1863 * floating-point environment need to do so
1864 * anyway to restore the prctl settings from
1865 * the saved environment.
1867 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1868 tsk->thread.fpexc_mode = val &
1869 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1879 /* on a CONFIG_SPE this does not hurt us. The bits that
1880 * __pack_fe01 use do not overlap with bits used for
1881 * PR_FP_EXC_SW_ENABLE. Additionally, the MSR[FE0,FE1] bits
1882 * on CONFIG_SPE implementations are reserved so writing to
1883 * them does not change anything */
1884 if (val > PR_FP_EXC_PRECISE)
1886 tsk->thread.fpexc_mode = __pack_fe01(val);
1887 if (regs != NULL && (regs->msr & MSR_FP) != 0)
1888 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1889 | tsk->thread.fpexc_mode;
1893 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1897 if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1899 if (cpu_has_feature(CPU_FTR_SPE)) {
1901 * When the sticky exception bits are set
1902 * directly by userspace, it must call prctl
1903 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1904 * in the existing prctl settings) or
1905 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1906 * the bits being set). <fenv.h> functions
1907 * saving and restoring the whole
1908 * floating-point environment need to do so
1909 * anyway to restore the prctl settings from
1910 * the saved environment.
1912 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1913 val = tsk->thread.fpexc_mode;
1920 val = __unpack_fe01(tsk->thread.fpexc_mode);
1921 return put_user(val, (unsigned int __user *) adr);
1924 int set_endian(struct task_struct *tsk, unsigned int val)
1926 struct pt_regs *regs = tsk->thread.regs;
1928 if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1929 (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1935 if (val == PR_ENDIAN_BIG)
1936 regs->msr &= ~MSR_LE;
1937 else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1938 regs->msr |= MSR_LE;
1945 int get_endian(struct task_struct *tsk, unsigned long adr)
1947 struct pt_regs *regs = tsk->thread.regs;
1950 if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1951 !cpu_has_feature(CPU_FTR_REAL_LE))
1957 if (regs->msr & MSR_LE) {
1958 if (cpu_has_feature(CPU_FTR_REAL_LE))
1959 val = PR_ENDIAN_LITTLE;
1961 val = PR_ENDIAN_PPC_LITTLE;
1963 val = PR_ENDIAN_BIG;
1965 return put_user(val, (unsigned int __user *)adr);
1968 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1970 tsk->thread.align_ctl = val;
1974 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1976 return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1979 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1980 unsigned long nbytes)
1982 unsigned long stack_page;
1983 unsigned long cpu = task_cpu(p);
1986 * Avoid crashing if the stack has overflowed and corrupted
1987 * task_cpu(p), which is in the thread_info struct.
1989 if (cpu < NR_CPUS && cpu_possible(cpu)) {
1990 stack_page = (unsigned long) hardirq_ctx[cpu];
1991 if (sp >= stack_page + sizeof(struct thread_struct)
1992 && sp <= stack_page + THREAD_SIZE - nbytes)
1995 stack_page = (unsigned long) softirq_ctx[cpu];
1996 if (sp >= stack_page + sizeof(struct thread_struct)
1997 && sp <= stack_page + THREAD_SIZE - nbytes)
2003 int validate_sp(unsigned long sp, struct task_struct *p,
2004 unsigned long nbytes)
2006 unsigned long stack_page = (unsigned long)task_stack_page(p);
2008 if (sp >= stack_page + sizeof(struct thread_struct)
2009 && sp <= stack_page + THREAD_SIZE - nbytes)
2012 return valid_irq_stack(sp, p, nbytes);
2015 EXPORT_SYMBOL(validate_sp);
2017 unsigned long get_wchan(struct task_struct *p)
2019 unsigned long ip, sp;
2022 if (!p || p == current || p->state == TASK_RUNNING)
2026 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2030 sp = *(unsigned long *)sp;
2031 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2032 p->state == TASK_RUNNING)
2035 ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2036 if (!in_sched_functions(ip))
2039 } while (count++ < 16);
2043 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2045 void show_stack(struct task_struct *tsk, unsigned long *stack)
2047 unsigned long sp, ip, lr, newsp;
2050 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2051 int curr_frame = current->curr_ret_stack;
2052 extern void return_to_handler(void);
2053 unsigned long rth = (unsigned long)return_to_handler;
2056 sp = (unsigned long) stack;
2061 sp = current_stack_pointer();
2063 sp = tsk->thread.ksp;
2067 printk("Call Trace:\n");
2069 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2072 stack = (unsigned long *) sp;
2074 ip = stack[STACK_FRAME_LR_SAVE];
2075 if (!firstframe || ip != lr) {
2076 printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
2077 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2078 if ((ip == rth) && curr_frame >= 0) {
2080 (void *)current->ret_stack[curr_frame].ret);
2085 pr_cont(" (unreliable)");
2091 * See if this is an exception frame.
2092 * We look for the "regshere" marker in the current frame.
2094 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2095 && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2096 struct pt_regs *regs = (struct pt_regs *)
2097 (sp + STACK_FRAME_OVERHEAD);
2099 printk("--- interrupt: %lx at %pS\n LR = %pS\n",
2100 regs->trap, (void *)regs->nip, (void *)lr);
2105 } while (count++ < kstack_depth_to_print);
2109 /* Called with hard IRQs off */
2110 void notrace __ppc64_runlatch_on(void)
2112 struct thread_info *ti = current_thread_info();
2114 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2116 * Least significant bit (RUN) is the only writable bit of
2117 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2118 * earliest ISA where this is the case, but it's convenient.
2120 mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2125 * Some architectures (e.g., Cell) have writable fields other
2126 * than RUN, so do the read-modify-write.
2128 ctrl = mfspr(SPRN_CTRLF);
2129 ctrl |= CTRL_RUNLATCH;
2130 mtspr(SPRN_CTRLT, ctrl);
2133 ti->local_flags |= _TLF_RUNLATCH;
2136 /* Called with hard IRQs off */
2137 void notrace __ppc64_runlatch_off(void)
2139 struct thread_info *ti = current_thread_info();
2141 ti->local_flags &= ~_TLF_RUNLATCH;
2143 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2144 mtspr(SPRN_CTRLT, 0);
2148 ctrl = mfspr(SPRN_CTRLF);
2149 ctrl &= ~CTRL_RUNLATCH;
2150 mtspr(SPRN_CTRLT, ctrl);
2153 #endif /* CONFIG_PPC64 */
2155 unsigned long arch_align_stack(unsigned long sp)
2157 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2158 sp -= get_random_int() & ~PAGE_MASK;
2162 static inline unsigned long brk_rnd(void)
2164 unsigned long rnd = 0;
2166 /* 8MB for 32bit, 1GB for 64bit */
2167 if (is_32bit_task())
2168 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2170 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2172 return rnd << PAGE_SHIFT;
2175 unsigned long arch_randomize_brk(struct mm_struct *mm)
2177 unsigned long base = mm->brk;
2180 #ifdef CONFIG_PPC_BOOK3S_64
2182 * If we are using 1TB segments and we are allowed to randomise
2183 * the heap, we can put it above 1TB so it is backed by a 1TB
2184 * segment. Otherwise the heap will be in the bottom 1TB
2185 * which always uses 256MB segments and this may result in a
2186 * performance penalty. We don't need to worry about radix. For
2187 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2189 if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2190 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2193 ret = PAGE_ALIGN(base + brk_rnd());