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>
46 #include <linux/seq_buf.h>
48 #include <asm/pgtable.h>
50 #include <asm/processor.h>
53 #include <asm/machdep.h>
55 #include <asm/runlatch.h>
56 #include <asm/syscalls.h>
57 #include <asm/switch_to.h>
59 #include <asm/debug.h>
61 #include <asm/firmware.h>
62 #include <asm/hw_irq.h>
64 #include <asm/code-patching.h>
66 #include <asm/livepatch.h>
67 #include <asm/cpu_has_feature.h>
68 #include <asm/asm-prototypes.h>
69 #include <asm/stacktrace.h>
71 #include <linux/kprobes.h>
72 #include <linux/kdebug.h>
74 /* Transactional Memory debug */
76 #define TM_DEBUG(x...) printk(KERN_INFO x)
78 #define TM_DEBUG(x...) do { } while(0)
81 extern unsigned long _get_SP(void);
83 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
85 * Are we running in "Suspend disabled" mode? If so we have to block any
86 * sigreturn that would get us into suspended state, and we also warn in some
87 * other paths that we should never reach with suspend disabled.
89 bool tm_suspend_disabled __ro_after_init = false;
91 static void check_if_tm_restore_required(struct task_struct *tsk)
94 * If we are saving the current thread's registers, and the
95 * thread is in a transactional state, set the TIF_RESTORE_TM
96 * bit so that we know to restore the registers before
97 * returning to userspace.
99 if (tsk == current && tsk->thread.regs &&
100 MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
101 !test_thread_flag(TIF_RESTORE_TM)) {
102 tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
103 set_thread_flag(TIF_RESTORE_TM);
107 static bool tm_active_with_fp(struct task_struct *tsk)
109 return MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
110 (tsk->thread.ckpt_regs.msr & MSR_FP);
113 static bool tm_active_with_altivec(struct task_struct *tsk)
115 return MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
116 (tsk->thread.ckpt_regs.msr & MSR_VEC);
119 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
120 static inline bool tm_active_with_fp(struct task_struct *tsk) { return false; }
121 static inline bool tm_active_with_altivec(struct task_struct *tsk) { return false; }
122 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
124 bool strict_msr_control;
125 EXPORT_SYMBOL(strict_msr_control);
127 static int __init enable_strict_msr_control(char *str)
129 strict_msr_control = true;
130 pr_info("Enabling strict facility control\n");
134 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
136 unsigned long msr_check_and_set(unsigned long bits)
138 unsigned long oldmsr = mfmsr();
139 unsigned long newmsr;
141 newmsr = oldmsr | bits;
144 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
148 if (oldmsr != newmsr)
153 EXPORT_SYMBOL_GPL(msr_check_and_set);
155 void __msr_check_and_clear(unsigned long bits)
157 unsigned long oldmsr = mfmsr();
158 unsigned long newmsr;
160 newmsr = oldmsr & ~bits;
163 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
167 if (oldmsr != newmsr)
170 EXPORT_SYMBOL(__msr_check_and_clear);
172 #ifdef CONFIG_PPC_FPU
173 static void __giveup_fpu(struct task_struct *tsk)
178 msr = tsk->thread.regs->msr;
181 if (cpu_has_feature(CPU_FTR_VSX))
184 tsk->thread.regs->msr = msr;
187 void giveup_fpu(struct task_struct *tsk)
189 check_if_tm_restore_required(tsk);
191 msr_check_and_set(MSR_FP);
193 msr_check_and_clear(MSR_FP);
195 EXPORT_SYMBOL(giveup_fpu);
198 * Make sure the floating-point register state in the
199 * the thread_struct is up to date for task tsk.
201 void flush_fp_to_thread(struct task_struct *tsk)
203 if (tsk->thread.regs) {
205 * We need to disable preemption here because if we didn't,
206 * another process could get scheduled after the regs->msr
207 * test but before we have finished saving the FP registers
208 * to the thread_struct. That process could take over the
209 * FPU, and then when we get scheduled again we would store
210 * bogus values for the remaining FP registers.
213 if (tsk->thread.regs->msr & MSR_FP) {
215 * This should only ever be called for current or
216 * for a stopped child process. Since we save away
217 * the FP register state on context switch,
218 * there is something wrong if a stopped child appears
219 * to still have its FP state in the CPU registers.
221 BUG_ON(tsk != current);
227 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
229 void enable_kernel_fp(void)
231 unsigned long cpumsr;
233 WARN_ON(preemptible());
235 cpumsr = msr_check_and_set(MSR_FP);
237 if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
238 check_if_tm_restore_required(current);
240 * If a thread has already been reclaimed then the
241 * checkpointed registers are on the CPU but have definitely
242 * been saved by the reclaim code. Don't need to and *cannot*
243 * giveup as this would save to the 'live' structure not the
244 * checkpointed structure.
246 if (!MSR_TM_ACTIVE(cpumsr) &&
247 MSR_TM_ACTIVE(current->thread.regs->msr))
249 __giveup_fpu(current);
252 EXPORT_SYMBOL(enable_kernel_fp);
254 static int restore_fp(struct task_struct *tsk)
256 if (tsk->thread.load_fp || tm_active_with_fp(tsk)) {
257 load_fp_state(¤t->thread.fp_state);
258 current->thread.load_fp++;
264 static int restore_fp(struct task_struct *tsk) { return 0; }
265 #endif /* CONFIG_PPC_FPU */
267 #ifdef CONFIG_ALTIVEC
268 #define loadvec(thr) ((thr).load_vec)
270 static void __giveup_altivec(struct task_struct *tsk)
275 msr = tsk->thread.regs->msr;
278 if (cpu_has_feature(CPU_FTR_VSX))
281 tsk->thread.regs->msr = msr;
284 void giveup_altivec(struct task_struct *tsk)
286 check_if_tm_restore_required(tsk);
288 msr_check_and_set(MSR_VEC);
289 __giveup_altivec(tsk);
290 msr_check_and_clear(MSR_VEC);
292 EXPORT_SYMBOL(giveup_altivec);
294 void enable_kernel_altivec(void)
296 unsigned long cpumsr;
298 WARN_ON(preemptible());
300 cpumsr = msr_check_and_set(MSR_VEC);
302 if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
303 check_if_tm_restore_required(current);
305 * If a thread has already been reclaimed then the
306 * checkpointed registers are on the CPU but have definitely
307 * been saved by the reclaim code. Don't need to and *cannot*
308 * giveup as this would save to the 'live' structure not the
309 * checkpointed structure.
311 if (!MSR_TM_ACTIVE(cpumsr) &&
312 MSR_TM_ACTIVE(current->thread.regs->msr))
314 __giveup_altivec(current);
317 EXPORT_SYMBOL(enable_kernel_altivec);
320 * Make sure the VMX/Altivec register state in the
321 * the thread_struct is up to date for task tsk.
323 void flush_altivec_to_thread(struct task_struct *tsk)
325 if (tsk->thread.regs) {
327 if (tsk->thread.regs->msr & MSR_VEC) {
328 BUG_ON(tsk != current);
334 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
336 static int restore_altivec(struct task_struct *tsk)
338 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
339 (tsk->thread.load_vec || tm_active_with_altivec(tsk))) {
340 load_vr_state(&tsk->thread.vr_state);
341 tsk->thread.used_vr = 1;
342 tsk->thread.load_vec++;
349 #define loadvec(thr) 0
350 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
351 #endif /* CONFIG_ALTIVEC */
354 static void __giveup_vsx(struct task_struct *tsk)
356 unsigned long msr = tsk->thread.regs->msr;
359 * We should never be ssetting MSR_VSX without also setting
362 WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
364 /* __giveup_fpu will clear MSR_VSX */
368 __giveup_altivec(tsk);
371 static void giveup_vsx(struct task_struct *tsk)
373 check_if_tm_restore_required(tsk);
375 msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
377 msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
380 void enable_kernel_vsx(void)
382 unsigned long cpumsr;
384 WARN_ON(preemptible());
386 cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
388 if (current->thread.regs &&
389 (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
390 check_if_tm_restore_required(current);
392 * If a thread has already been reclaimed then the
393 * checkpointed registers are on the CPU but have definitely
394 * been saved by the reclaim code. Don't need to and *cannot*
395 * giveup as this would save to the 'live' structure not the
396 * checkpointed structure.
398 if (!MSR_TM_ACTIVE(cpumsr) &&
399 MSR_TM_ACTIVE(current->thread.regs->msr))
401 __giveup_vsx(current);
404 EXPORT_SYMBOL(enable_kernel_vsx);
406 void flush_vsx_to_thread(struct task_struct *tsk)
408 if (tsk->thread.regs) {
410 if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
411 BUG_ON(tsk != current);
417 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
419 static int restore_vsx(struct task_struct *tsk)
421 if (cpu_has_feature(CPU_FTR_VSX)) {
422 tsk->thread.used_vsr = 1;
429 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
430 #endif /* CONFIG_VSX */
433 void giveup_spe(struct task_struct *tsk)
435 check_if_tm_restore_required(tsk);
437 msr_check_and_set(MSR_SPE);
439 msr_check_and_clear(MSR_SPE);
441 EXPORT_SYMBOL(giveup_spe);
443 void enable_kernel_spe(void)
445 WARN_ON(preemptible());
447 msr_check_and_set(MSR_SPE);
449 if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
450 check_if_tm_restore_required(current);
451 __giveup_spe(current);
454 EXPORT_SYMBOL(enable_kernel_spe);
456 void flush_spe_to_thread(struct task_struct *tsk)
458 if (tsk->thread.regs) {
460 if (tsk->thread.regs->msr & MSR_SPE) {
461 BUG_ON(tsk != current);
462 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
468 #endif /* CONFIG_SPE */
470 static unsigned long msr_all_available;
472 static int __init init_msr_all_available(void)
474 #ifdef CONFIG_PPC_FPU
475 msr_all_available |= MSR_FP;
477 #ifdef CONFIG_ALTIVEC
478 if (cpu_has_feature(CPU_FTR_ALTIVEC))
479 msr_all_available |= MSR_VEC;
482 if (cpu_has_feature(CPU_FTR_VSX))
483 msr_all_available |= MSR_VSX;
486 if (cpu_has_feature(CPU_FTR_SPE))
487 msr_all_available |= MSR_SPE;
492 early_initcall(init_msr_all_available);
494 void giveup_all(struct task_struct *tsk)
496 unsigned long usermsr;
498 if (!tsk->thread.regs)
501 usermsr = tsk->thread.regs->msr;
503 if ((usermsr & msr_all_available) == 0)
506 msr_check_and_set(msr_all_available);
507 check_if_tm_restore_required(tsk);
509 WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
511 #ifdef CONFIG_PPC_FPU
512 if (usermsr & MSR_FP)
515 #ifdef CONFIG_ALTIVEC
516 if (usermsr & MSR_VEC)
517 __giveup_altivec(tsk);
520 if (usermsr & MSR_SPE)
524 msr_check_and_clear(msr_all_available);
526 EXPORT_SYMBOL(giveup_all);
528 void restore_math(struct pt_regs *regs)
532 if (!MSR_TM_ACTIVE(regs->msr) &&
533 !current->thread.load_fp && !loadvec(current->thread))
537 msr_check_and_set(msr_all_available);
540 * Only reload if the bit is not set in the user MSR, the bit BEING set
541 * indicates that the registers are hot
543 if ((!(msr & MSR_FP)) && restore_fp(current))
544 msr |= MSR_FP | current->thread.fpexc_mode;
546 if ((!(msr & MSR_VEC)) && restore_altivec(current))
549 if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
550 restore_vsx(current)) {
554 msr_check_and_clear(msr_all_available);
559 static void save_all(struct task_struct *tsk)
561 unsigned long usermsr;
563 if (!tsk->thread.regs)
566 usermsr = tsk->thread.regs->msr;
568 if ((usermsr & msr_all_available) == 0)
571 msr_check_and_set(msr_all_available);
573 WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
575 if (usermsr & MSR_FP)
578 if (usermsr & MSR_VEC)
581 if (usermsr & MSR_SPE)
584 msr_check_and_clear(msr_all_available);
585 thread_pkey_regs_save(&tsk->thread);
588 void flush_all_to_thread(struct task_struct *tsk)
590 if (tsk->thread.regs) {
592 BUG_ON(tsk != current);
596 if (tsk->thread.regs->msr & MSR_SPE)
597 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
603 EXPORT_SYMBOL(flush_all_to_thread);
605 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
606 void do_send_trap(struct pt_regs *regs, unsigned long address,
607 unsigned long error_code, int breakpt)
609 current->thread.trap_nr = TRAP_HWBKPT;
610 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
611 11, SIGSEGV) == NOTIFY_STOP)
614 /* Deliver the signal to userspace */
615 force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
616 (void __user *)address);
618 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
619 void do_break (struct pt_regs *regs, unsigned long address,
620 unsigned long error_code)
622 current->thread.trap_nr = TRAP_HWBKPT;
623 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
624 11, SIGSEGV) == NOTIFY_STOP)
627 if (debugger_break_match(regs))
630 /* Clear the breakpoint */
631 hw_breakpoint_disable();
633 /* Deliver the signal to userspace */
634 force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)address, current);
636 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
638 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
640 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
642 * Set the debug registers back to their default "safe" values.
644 static void set_debug_reg_defaults(struct thread_struct *thread)
646 thread->debug.iac1 = thread->debug.iac2 = 0;
647 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
648 thread->debug.iac3 = thread->debug.iac4 = 0;
650 thread->debug.dac1 = thread->debug.dac2 = 0;
651 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
652 thread->debug.dvc1 = thread->debug.dvc2 = 0;
654 thread->debug.dbcr0 = 0;
657 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
659 thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
660 DBCR1_IAC3US | DBCR1_IAC4US;
662 * Force Data Address Compare User/Supervisor bits to be User-only
663 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
665 thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
667 thread->debug.dbcr1 = 0;
671 static void prime_debug_regs(struct debug_reg *debug)
674 * We could have inherited MSR_DE from userspace, since
675 * it doesn't get cleared on exception entry. Make sure
676 * MSR_DE is clear before we enable any debug events.
678 mtmsr(mfmsr() & ~MSR_DE);
680 mtspr(SPRN_IAC1, debug->iac1);
681 mtspr(SPRN_IAC2, debug->iac2);
682 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
683 mtspr(SPRN_IAC3, debug->iac3);
684 mtspr(SPRN_IAC4, debug->iac4);
686 mtspr(SPRN_DAC1, debug->dac1);
687 mtspr(SPRN_DAC2, debug->dac2);
688 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
689 mtspr(SPRN_DVC1, debug->dvc1);
690 mtspr(SPRN_DVC2, debug->dvc2);
692 mtspr(SPRN_DBCR0, debug->dbcr0);
693 mtspr(SPRN_DBCR1, debug->dbcr1);
695 mtspr(SPRN_DBCR2, debug->dbcr2);
699 * Unless neither the old or new thread are making use of the
700 * debug registers, set the debug registers from the values
701 * stored in the new thread.
703 void switch_booke_debug_regs(struct debug_reg *new_debug)
705 if ((current->thread.debug.dbcr0 & DBCR0_IDM)
706 || (new_debug->dbcr0 & DBCR0_IDM))
707 prime_debug_regs(new_debug);
709 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
710 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
711 #ifndef CONFIG_HAVE_HW_BREAKPOINT
712 static void set_breakpoint(struct arch_hw_breakpoint *brk)
715 __set_breakpoint(brk);
719 static void set_debug_reg_defaults(struct thread_struct *thread)
721 thread->hw_brk.address = 0;
722 thread->hw_brk.type = 0;
723 if (ppc_breakpoint_available())
724 set_breakpoint(&thread->hw_brk);
726 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
727 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
729 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
730 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
732 mtspr(SPRN_DAC1, dabr);
733 #ifdef CONFIG_PPC_47x
738 #elif defined(CONFIG_PPC_BOOK3S)
739 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
741 mtspr(SPRN_DABR, dabr);
742 if (cpu_has_feature(CPU_FTR_DABRX))
743 mtspr(SPRN_DABRX, dabrx);
746 #elif defined(CONFIG_PPC_8xx)
747 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
749 unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
750 unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
751 unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
753 if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
755 else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
757 else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
760 mtspr(SPRN_LCTRL2, 0);
761 mtspr(SPRN_CMPE, addr);
762 mtspr(SPRN_CMPF, addr + 4);
763 mtspr(SPRN_LCTRL1, lctrl1);
764 mtspr(SPRN_LCTRL2, lctrl2);
769 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
775 static inline int set_dabr(struct arch_hw_breakpoint *brk)
777 unsigned long dabr, dabrx;
779 dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
780 dabrx = ((brk->type >> 3) & 0x7);
783 return ppc_md.set_dabr(dabr, dabrx);
785 return __set_dabr(dabr, dabrx);
788 static inline int set_dawr(struct arch_hw_breakpoint *brk)
790 unsigned long dawr, dawrx, mrd;
794 dawrx = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
795 << (63 - 58); //* read/write bits */
796 dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
797 << (63 - 59); //* translate */
798 dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
799 >> 3; //* PRIM bits */
800 /* dawr length is stored in field MDR bits 48:53. Matches range in
801 doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
803 brk->len is in bytes.
804 This aligns up to double word size, shifts and does the bias.
806 mrd = ((brk->len + 7) >> 3) - 1;
807 dawrx |= (mrd & 0x3f) << (63 - 53);
810 return ppc_md.set_dawr(dawr, dawrx);
811 mtspr(SPRN_DAWR, dawr);
812 mtspr(SPRN_DAWRX, dawrx);
816 void __set_breakpoint(struct arch_hw_breakpoint *brk)
818 memcpy(this_cpu_ptr(¤t_brk), brk, sizeof(*brk));
820 if (cpu_has_feature(CPU_FTR_DAWR))
823 else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
827 // Shouldn't happen due to higher level checks
831 /* Check if we have DAWR or DABR hardware */
832 bool ppc_breakpoint_available(void)
834 if (cpu_has_feature(CPU_FTR_DAWR))
835 return true; /* POWER8 DAWR */
836 if (cpu_has_feature(CPU_FTR_ARCH_207S))
837 return false; /* POWER9 with DAWR disabled */
838 /* DABR: Everything but POWER8 and POWER9 */
841 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
843 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
844 struct arch_hw_breakpoint *b)
846 if (a->address != b->address)
848 if (a->type != b->type)
850 if (a->len != b->len)
855 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
857 static inline bool tm_enabled(struct task_struct *tsk)
859 return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
862 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
865 * Use the current MSR TM suspended bit to track if we have
866 * checkpointed state outstanding.
867 * On signal delivery, we'd normally reclaim the checkpointed
868 * state to obtain stack pointer (see:get_tm_stackpointer()).
869 * This will then directly return to userspace without going
870 * through __switch_to(). However, if the stack frame is bad,
871 * we need to exit this thread which calls __switch_to() which
872 * will again attempt to reclaim the already saved tm state.
873 * Hence we need to check that we've not already reclaimed
875 * We do this using the current MSR, rather tracking it in
876 * some specific thread_struct bit, as it has the additional
877 * benefit of checking for a potential TM bad thing exception.
879 if (!MSR_TM_SUSPENDED(mfmsr()))
882 giveup_all(container_of(thr, struct task_struct, thread));
884 tm_reclaim(thr, cause);
887 * If we are in a transaction and FP is off then we can't have
888 * used FP inside that transaction. Hence the checkpointed
889 * state is the same as the live state. We need to copy the
890 * live state to the checkpointed state so that when the
891 * transaction is restored, the checkpointed state is correct
892 * and the aborted transaction sees the correct state. We use
893 * ckpt_regs.msr here as that's what tm_reclaim will use to
894 * determine if it's going to write the checkpointed state or
895 * not. So either this will write the checkpointed registers,
896 * or reclaim will. Similarly for VMX.
898 if ((thr->ckpt_regs.msr & MSR_FP) == 0)
899 memcpy(&thr->ckfp_state, &thr->fp_state,
900 sizeof(struct thread_fp_state));
901 if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
902 memcpy(&thr->ckvr_state, &thr->vr_state,
903 sizeof(struct thread_vr_state));
906 void tm_reclaim_current(uint8_t cause)
909 tm_reclaim_thread(¤t->thread, cause);
912 static inline void tm_reclaim_task(struct task_struct *tsk)
914 /* We have to work out if we're switching from/to a task that's in the
915 * middle of a transaction.
917 * In switching we need to maintain a 2nd register state as
918 * oldtask->thread.ckpt_regs. We tm_reclaim(oldproc); this saves the
919 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
922 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
924 struct thread_struct *thr = &tsk->thread;
929 if (!MSR_TM_ACTIVE(thr->regs->msr))
930 goto out_and_saveregs;
932 WARN_ON(tm_suspend_disabled);
934 TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
935 "ccr=%lx, msr=%lx, trap=%lx)\n",
936 tsk->pid, thr->regs->nip,
937 thr->regs->ccr, thr->regs->msr,
940 tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
942 TM_DEBUG("--- tm_reclaim on pid %d complete\n",
946 /* Always save the regs here, even if a transaction's not active.
947 * This context-switches a thread's TM info SPRs. We do it here to
948 * be consistent with the restore path (in recheckpoint) which
949 * cannot happen later in _switch().
954 extern void __tm_recheckpoint(struct thread_struct *thread);
956 void tm_recheckpoint(struct thread_struct *thread)
960 if (!(thread->regs->msr & MSR_TM))
963 /* We really can't be interrupted here as the TEXASR registers can't
964 * change and later in the trecheckpoint code, we have a userspace R1.
965 * So let's hard disable over this region.
967 local_irq_save(flags);
970 /* The TM SPRs are restored here, so that TEXASR.FS can be set
971 * before the trecheckpoint and no explosion occurs.
973 tm_restore_sprs(thread);
975 __tm_recheckpoint(thread);
977 local_irq_restore(flags);
980 static inline void tm_recheckpoint_new_task(struct task_struct *new)
982 if (!cpu_has_feature(CPU_FTR_TM))
985 /* Recheckpoint the registers of the thread we're about to switch to.
987 * If the task was using FP, we non-lazily reload both the original and
988 * the speculative FP register states. This is because the kernel
989 * doesn't see if/when a TM rollback occurs, so if we take an FP
990 * unavailable later, we are unable to determine which set of FP regs
991 * need to be restored.
993 if (!tm_enabled(new))
996 if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
997 tm_restore_sprs(&new->thread);
1000 /* Recheckpoint to restore original checkpointed register state. */
1001 TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1002 new->pid, new->thread.regs->msr);
1004 tm_recheckpoint(&new->thread);
1007 * The checkpointed state has been restored but the live state has
1008 * not, ensure all the math functionality is turned off to trigger
1009 * restore_math() to reload.
1011 new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1013 TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1014 "(kernel msr 0x%lx)\n",
1018 static inline void __switch_to_tm(struct task_struct *prev,
1019 struct task_struct *new)
1021 if (cpu_has_feature(CPU_FTR_TM)) {
1022 if (tm_enabled(prev) || tm_enabled(new))
1025 if (tm_enabled(prev)) {
1026 prev->thread.load_tm++;
1027 tm_reclaim_task(prev);
1028 if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1029 prev->thread.regs->msr &= ~MSR_TM;
1032 tm_recheckpoint_new_task(new);
1037 * This is called if we are on the way out to userspace and the
1038 * TIF_RESTORE_TM flag is set. It checks if we need to reload
1039 * FP and/or vector state and does so if necessary.
1040 * If userspace is inside a transaction (whether active or
1041 * suspended) and FP/VMX/VSX instructions have ever been enabled
1042 * inside that transaction, then we have to keep them enabled
1043 * and keep the FP/VMX/VSX state loaded while ever the transaction
1044 * continues. The reason is that if we didn't, and subsequently
1045 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1046 * we don't know whether it's the same transaction, and thus we
1047 * don't know which of the checkpointed state and the transactional
1050 void restore_tm_state(struct pt_regs *regs)
1052 unsigned long msr_diff;
1055 * This is the only moment we should clear TIF_RESTORE_TM as
1056 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1057 * again, anything else could lead to an incorrect ckpt_msr being
1058 * saved and therefore incorrect signal contexts.
1060 clear_thread_flag(TIF_RESTORE_TM);
1061 if (!MSR_TM_ACTIVE(regs->msr))
1064 msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1065 msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1067 /* Ensure that restore_math() will restore */
1068 if (msr_diff & MSR_FP)
1069 current->thread.load_fp = 1;
1070 #ifdef CONFIG_ALTIVEC
1071 if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1072 current->thread.load_vec = 1;
1076 regs->msr |= msr_diff;
1080 #define tm_recheckpoint_new_task(new)
1081 #define __switch_to_tm(prev, new)
1082 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1084 static inline void save_sprs(struct thread_struct *t)
1086 #ifdef CONFIG_ALTIVEC
1087 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1088 t->vrsave = mfspr(SPRN_VRSAVE);
1090 #ifdef CONFIG_PPC_BOOK3S_64
1091 if (cpu_has_feature(CPU_FTR_DSCR))
1092 t->dscr = mfspr(SPRN_DSCR);
1094 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1095 t->bescr = mfspr(SPRN_BESCR);
1096 t->ebbhr = mfspr(SPRN_EBBHR);
1097 t->ebbrr = mfspr(SPRN_EBBRR);
1099 t->fscr = mfspr(SPRN_FSCR);
1102 * Note that the TAR is not available for use in the kernel.
1103 * (To provide this, the TAR should be backed up/restored on
1104 * exception entry/exit instead, and be in pt_regs. FIXME,
1105 * this should be in pt_regs anyway (for debug).)
1107 t->tar = mfspr(SPRN_TAR);
1111 thread_pkey_regs_save(t);
1114 static inline void restore_sprs(struct thread_struct *old_thread,
1115 struct thread_struct *new_thread)
1117 #ifdef CONFIG_ALTIVEC
1118 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1119 old_thread->vrsave != new_thread->vrsave)
1120 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1122 #ifdef CONFIG_PPC_BOOK3S_64
1123 if (cpu_has_feature(CPU_FTR_DSCR)) {
1124 u64 dscr = get_paca()->dscr_default;
1125 if (new_thread->dscr_inherit)
1126 dscr = new_thread->dscr;
1128 if (old_thread->dscr != dscr)
1129 mtspr(SPRN_DSCR, dscr);
1132 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1133 if (old_thread->bescr != new_thread->bescr)
1134 mtspr(SPRN_BESCR, new_thread->bescr);
1135 if (old_thread->ebbhr != new_thread->ebbhr)
1136 mtspr(SPRN_EBBHR, new_thread->ebbhr);
1137 if (old_thread->ebbrr != new_thread->ebbrr)
1138 mtspr(SPRN_EBBRR, new_thread->ebbrr);
1140 if (old_thread->fscr != new_thread->fscr)
1141 mtspr(SPRN_FSCR, new_thread->fscr);
1143 if (old_thread->tar != new_thread->tar)
1144 mtspr(SPRN_TAR, new_thread->tar);
1147 if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1148 old_thread->tidr != new_thread->tidr)
1149 mtspr(SPRN_TIDR, new_thread->tidr);
1152 thread_pkey_regs_restore(new_thread, old_thread);
1155 #ifdef CONFIG_PPC_BOOK3S_64
1157 static const u8 dummy_copy_buffer[CP_SIZE] __attribute__((aligned(CP_SIZE)));
1160 struct task_struct *__switch_to(struct task_struct *prev,
1161 struct task_struct *new)
1163 struct thread_struct *new_thread, *old_thread;
1164 struct task_struct *last;
1165 #ifdef CONFIG_PPC_BOOK3S_64
1166 struct ppc64_tlb_batch *batch;
1169 new_thread = &new->thread;
1170 old_thread = ¤t->thread;
1172 WARN_ON(!irqs_disabled());
1174 #ifdef CONFIG_PPC_BOOK3S_64
1175 batch = this_cpu_ptr(&ppc64_tlb_batch);
1176 if (batch->active) {
1177 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1179 __flush_tlb_pending(batch);
1182 #endif /* CONFIG_PPC_BOOK3S_64 */
1184 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1185 switch_booke_debug_regs(&new->thread.debug);
1188 * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1191 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1192 if (unlikely(!hw_brk_match(this_cpu_ptr(¤t_brk), &new->thread.hw_brk)))
1193 __set_breakpoint(&new->thread.hw_brk);
1194 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1198 * We need to save SPRs before treclaim/trecheckpoint as these will
1199 * change a number of them.
1201 save_sprs(&prev->thread);
1203 /* Save FPU, Altivec, VSX and SPE state */
1206 __switch_to_tm(prev, new);
1208 if (!radix_enabled()) {
1210 * We can't take a PMU exception inside _switch() since there
1211 * is a window where the kernel stack SLB and the kernel stack
1212 * are out of sync. Hard disable here.
1218 * Call restore_sprs() before calling _switch(). If we move it after
1219 * _switch() then we miss out on calling it for new tasks. The reason
1220 * for this is we manually create a stack frame for new tasks that
1221 * directly returns through ret_from_fork() or
1222 * ret_from_kernel_thread(). See copy_thread() for details.
1224 restore_sprs(old_thread, new_thread);
1226 last = _switch(old_thread, new_thread);
1228 #ifdef CONFIG_PPC_BOOK3S_64
1229 if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1230 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1231 batch = this_cpu_ptr(&ppc64_tlb_batch);
1235 if (current_thread_info()->task->thread.regs) {
1236 restore_math(current_thread_info()->task->thread.regs);
1239 * The copy-paste buffer can only store into foreign real
1240 * addresses, so unprivileged processes can not see the
1241 * data or use it in any way unless they have foreign real
1242 * mappings. If the new process has the foreign real address
1243 * mappings, we must issue a cp_abort to clear any state and
1244 * prevent snooping, corruption or a covert channel.
1246 if (current_thread_info()->task->thread.used_vas)
1247 asm volatile(PPC_CP_ABORT);
1249 #endif /* CONFIG_PPC_BOOK3S_64 */
1254 #define NR_INSN_TO_PRINT 16
1256 static void show_instructions(struct pt_regs *regs)
1259 unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1261 printk("Instruction dump:");
1263 for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1269 #if !defined(CONFIG_BOOKE)
1270 /* If executing with the IMMU off, adjust pc rather
1271 * than print XXXXXXXX.
1273 if (!(regs->msr & MSR_IR))
1274 pc = (unsigned long)phys_to_virt(pc);
1277 if (!__kernel_text_address(pc) ||
1278 probe_kernel_address((const void *)pc, instr)) {
1279 pr_cont("XXXXXXXX ");
1281 if (regs->nip == pc)
1282 pr_cont("<%08x> ", instr);
1284 pr_cont("%08x ", instr);
1293 void show_user_instructions(struct pt_regs *regs)
1296 int n = NR_INSN_TO_PRINT;
1298 char buf[96]; /* enough for 8 times 9 + 2 chars */
1300 pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1303 * Make sure the NIP points at userspace, not kernel text/data or
1306 if (!__access_ok(pc, NR_INSN_TO_PRINT * sizeof(int), USER_DS)) {
1307 pr_info("%s[%d]: Bad NIP, not dumping instructions.\n",
1308 current->comm, current->pid);
1312 seq_buf_init(&s, buf, sizeof(buf));
1319 for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1322 if (probe_kernel_address((const void *)pc, instr)) {
1323 seq_buf_printf(&s, "XXXXXXXX ");
1326 seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1329 if (!seq_buf_has_overflowed(&s))
1330 pr_info("%s[%d]: code: %s\n", current->comm,
1331 current->pid, s.buffer);
1340 static struct regbit msr_bits[] = {
1341 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1363 #ifndef CONFIG_BOOKE
1370 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1374 for (; bits->bit; ++bits)
1375 if (val & bits->bit) {
1376 pr_cont("%s%s", s, bits->name);
1381 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1382 static struct regbit msr_tm_bits[] = {
1389 static void print_tm_bits(unsigned long val)
1392 * This only prints something if at least one of the TM bit is set.
1393 * Inside the TM[], the output means:
1394 * E: Enabled (bit 32)
1395 * S: Suspended (bit 33)
1396 * T: Transactional (bit 34)
1398 if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1400 print_bits(val, msr_tm_bits, "");
1405 static void print_tm_bits(unsigned long val) {}
1408 static void print_msr_bits(unsigned long val)
1411 print_bits(val, msr_bits, ",");
1417 #define REG "%016lx"
1418 #define REGS_PER_LINE 4
1419 #define LAST_VOLATILE 13
1422 #define REGS_PER_LINE 8
1423 #define LAST_VOLATILE 12
1426 void show_regs(struct pt_regs * regs)
1430 show_regs_print_info(KERN_DEFAULT);
1432 printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1433 regs->nip, regs->link, regs->ctr);
1434 printk("REGS: %px TRAP: %04lx %s (%s)\n",
1435 regs, regs->trap, print_tainted(), init_utsname()->release);
1436 printk("MSR: "REG" ", regs->msr);
1437 print_msr_bits(regs->msr);
1438 pr_cont(" CR: %08lx XER: %08lx\n", regs->ccr, regs->xer);
1440 if ((TRAP(regs) != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1441 pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1442 if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1443 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1444 pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1446 pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1449 pr_cont("IRQMASK: %lx ", regs->softe);
1451 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1452 if (MSR_TM_ACTIVE(regs->msr))
1453 pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1456 for (i = 0; i < 32; i++) {
1457 if ((i % REGS_PER_LINE) == 0)
1458 pr_cont("\nGPR%02d: ", i);
1459 pr_cont(REG " ", regs->gpr[i]);
1460 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1464 #ifdef CONFIG_KALLSYMS
1466 * Lookup NIP late so we have the best change of getting the
1467 * above info out without failing
1469 printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1470 printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1472 show_stack(current, (unsigned long *) regs->gpr[1]);
1473 if (!user_mode(regs))
1474 show_instructions(regs);
1477 void flush_thread(void)
1479 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1480 flush_ptrace_hw_breakpoint(current);
1481 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1482 set_debug_reg_defaults(¤t->thread);
1483 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1486 #ifdef CONFIG_PPC_BOOK3S_64
1487 void arch_setup_new_exec(void)
1489 if (radix_enabled())
1491 hash__setup_new_exec();
1495 int set_thread_uses_vas(void)
1497 #ifdef CONFIG_PPC_BOOK3S_64
1498 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1501 current->thread.used_vas = 1;
1504 * Even a process that has no foreign real address mapping can use
1505 * an unpaired COPY instruction (to no real effect). Issue CP_ABORT
1506 * to clear any pending COPY and prevent a covert channel.
1508 * __switch_to() will issue CP_ABORT on future context switches.
1510 asm volatile(PPC_CP_ABORT);
1512 #endif /* CONFIG_PPC_BOOK3S_64 */
1518 * Assign a TIDR (thread ID) for task @t and set it in the thread
1519 * structure. For now, we only support setting TIDR for 'current' task.
1521 * Since the TID value is a truncated form of it PID, it is possible
1522 * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1523 * that 2 threads share the same TID and are waiting, one of the following
1524 * cases will happen:
1526 * 1. The correct thread is running, the wrong thread is not
1527 * In this situation, the correct thread is woken and proceeds to pass it's
1530 * 2. Neither threads are running
1531 * In this situation, neither thread will be woken. When scheduled, the waiting
1532 * threads will execute either a wait, which will return immediately, followed
1533 * by a condition check, which will pass for the correct thread and fail
1534 * for the wrong thread, or they will execute the condition check immediately.
1536 * 3. The wrong thread is running, the correct thread is not
1537 * The wrong thread will be woken, but will fail it's condition check and
1538 * re-execute wait. The correct thread, when scheduled, will execute either
1539 * it's condition check (which will pass), or wait, which returns immediately
1540 * when called the first time after the thread is scheduled, followed by it's
1541 * condition check (which will pass).
1543 * 4. Both threads are running
1544 * Both threads will be woken. The wrong thread will fail it's condition check
1545 * and execute another wait, while the correct thread will pass it's condition
1548 * @t: the task to set the thread ID for
1550 int set_thread_tidr(struct task_struct *t)
1552 if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1561 t->thread.tidr = (u16)task_pid_nr(t);
1562 mtspr(SPRN_TIDR, t->thread.tidr);
1566 EXPORT_SYMBOL_GPL(set_thread_tidr);
1568 #endif /* CONFIG_PPC64 */
1571 release_thread(struct task_struct *t)
1576 * this gets called so that we can store coprocessor state into memory and
1577 * copy the current task into the new thread.
1579 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1581 flush_all_to_thread(src);
1583 * Flush TM state out so we can copy it. __switch_to_tm() does this
1584 * flush but it removes the checkpointed state from the current CPU and
1585 * transitions the CPU out of TM mode. Hence we need to call
1586 * tm_recheckpoint_new_task() (on the same task) to restore the
1587 * checkpointed state back and the TM mode.
1589 * Can't pass dst because it isn't ready. Doesn't matter, passing
1590 * dst is only important for __switch_to()
1592 __switch_to_tm(src, src);
1596 clear_task_ebb(dst);
1601 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1603 #ifdef CONFIG_PPC_BOOK3S_64
1604 unsigned long sp_vsid;
1605 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1607 if (radix_enabled())
1610 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1611 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1612 << SLB_VSID_SHIFT_1T;
1614 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1616 sp_vsid |= SLB_VSID_KERNEL | llp;
1617 p->thread.ksp_vsid = sp_vsid;
1626 * Copy architecture-specific thread state
1628 int copy_thread(unsigned long clone_flags, unsigned long usp,
1629 unsigned long kthread_arg, struct task_struct *p)
1631 struct pt_regs *childregs, *kregs;
1632 extern void ret_from_fork(void);
1633 extern void ret_from_kernel_thread(void);
1635 unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1636 struct thread_info *ti = task_thread_info(p);
1638 klp_init_thread_info(ti);
1640 /* Copy registers */
1641 sp -= sizeof(struct pt_regs);
1642 childregs = (struct pt_regs *) sp;
1643 if (unlikely(p->flags & PF_KTHREAD)) {
1645 memset(childregs, 0, sizeof(struct pt_regs));
1646 childregs->gpr[1] = sp + sizeof(struct pt_regs);
1649 childregs->gpr[14] = ppc_function_entry((void *)usp);
1651 clear_tsk_thread_flag(p, TIF_32BIT);
1652 childregs->softe = IRQS_ENABLED;
1654 childregs->gpr[15] = kthread_arg;
1655 p->thread.regs = NULL; /* no user register state */
1656 ti->flags |= _TIF_RESTOREALL;
1657 f = ret_from_kernel_thread;
1660 struct pt_regs *regs = current_pt_regs();
1661 CHECK_FULL_REGS(regs);
1664 childregs->gpr[1] = usp;
1665 p->thread.regs = childregs;
1666 childregs->gpr[3] = 0; /* Result from fork() */
1667 if (clone_flags & CLONE_SETTLS) {
1669 if (!is_32bit_task())
1670 childregs->gpr[13] = childregs->gpr[6];
1673 childregs->gpr[2] = childregs->gpr[6];
1678 childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1679 sp -= STACK_FRAME_OVERHEAD;
1682 * The way this works is that at some point in the future
1683 * some task will call _switch to switch to the new task.
1684 * That will pop off the stack frame created below and start
1685 * the new task running at ret_from_fork. The new task will
1686 * do some house keeping and then return from the fork or clone
1687 * system call, using the stack frame created above.
1689 ((unsigned long *)sp)[0] = 0;
1690 sp -= sizeof(struct pt_regs);
1691 kregs = (struct pt_regs *) sp;
1692 sp -= STACK_FRAME_OVERHEAD;
1695 p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1696 _ALIGN_UP(sizeof(struct thread_info), 16);
1698 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1699 p->thread.ptrace_bps[0] = NULL;
1702 p->thread.fp_save_area = NULL;
1703 #ifdef CONFIG_ALTIVEC
1704 p->thread.vr_save_area = NULL;
1707 setup_ksp_vsid(p, sp);
1710 if (cpu_has_feature(CPU_FTR_DSCR)) {
1711 p->thread.dscr_inherit = current->thread.dscr_inherit;
1712 p->thread.dscr = mfspr(SPRN_DSCR);
1714 if (cpu_has_feature(CPU_FTR_HAS_PPR))
1715 childregs->ppr = DEFAULT_PPR;
1719 kregs->nip = ppc_function_entry(f);
1723 void preload_new_slb_context(unsigned long start, unsigned long sp);
1726 * Set up a thread for executing a new program
1728 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1731 unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1733 #ifdef CONFIG_PPC_BOOK3S_64
1734 preload_new_slb_context(start, sp);
1739 * If we exec out of a kernel thread then thread.regs will not be
1742 if (!current->thread.regs) {
1743 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1744 current->thread.regs = regs - 1;
1747 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1749 * Clear any transactional state, we're exec()ing. The cause is
1750 * not important as there will never be a recheckpoint so it's not
1753 if (MSR_TM_SUSPENDED(mfmsr()))
1754 tm_reclaim_current(0);
1757 memset(regs->gpr, 0, sizeof(regs->gpr));
1765 * We have just cleared all the nonvolatile GPRs, so make
1766 * FULL_REGS(regs) return true. This is necessary to allow
1767 * ptrace to examine the thread immediately after exec.
1774 regs->msr = MSR_USER;
1776 if (!is_32bit_task()) {
1777 unsigned long entry;
1779 if (is_elf2_task()) {
1780 /* Look ma, no function descriptors! */
1785 * The latest iteration of the ABI requires that when
1786 * calling a function (at its global entry point),
1787 * the caller must ensure r12 holds the entry point
1788 * address (so that the function can quickly
1789 * establish addressability).
1791 regs->gpr[12] = start;
1792 /* Make sure that's restored on entry to userspace. */
1793 set_thread_flag(TIF_RESTOREALL);
1797 /* start is a relocated pointer to the function
1798 * descriptor for the elf _start routine. The first
1799 * entry in the function descriptor is the entry
1800 * address of _start and the second entry is the TOC
1801 * value we need to use.
1803 __get_user(entry, (unsigned long __user *)start);
1804 __get_user(toc, (unsigned long __user *)start+1);
1806 /* Check whether the e_entry function descriptor entries
1807 * need to be relocated before we can use them.
1809 if (load_addr != 0) {
1816 regs->msr = MSR_USER64;
1820 regs->msr = MSR_USER32;
1824 current->thread.used_vsr = 0;
1826 current->thread.load_slb = 0;
1827 current->thread.load_fp = 0;
1828 memset(¤t->thread.fp_state, 0, sizeof(current->thread.fp_state));
1829 current->thread.fp_save_area = NULL;
1830 #ifdef CONFIG_ALTIVEC
1831 memset(¤t->thread.vr_state, 0, sizeof(current->thread.vr_state));
1832 current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1833 current->thread.vr_save_area = NULL;
1834 current->thread.vrsave = 0;
1835 current->thread.used_vr = 0;
1836 current->thread.load_vec = 0;
1837 #endif /* CONFIG_ALTIVEC */
1839 memset(current->thread.evr, 0, sizeof(current->thread.evr));
1840 current->thread.acc = 0;
1841 current->thread.spefscr = 0;
1842 current->thread.used_spe = 0;
1843 #endif /* CONFIG_SPE */
1844 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1845 current->thread.tm_tfhar = 0;
1846 current->thread.tm_texasr = 0;
1847 current->thread.tm_tfiar = 0;
1848 current->thread.load_tm = 0;
1849 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1851 thread_pkey_regs_init(¤t->thread);
1853 EXPORT_SYMBOL(start_thread);
1855 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1856 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1858 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1860 struct pt_regs *regs = tsk->thread.regs;
1862 /* This is a bit hairy. If we are an SPE enabled processor
1863 * (have embedded fp) we store the IEEE exception enable flags in
1864 * fpexc_mode. fpexc_mode is also used for setting FP exception
1865 * mode (asyn, precise, disabled) for 'Classic' FP. */
1866 if (val & PR_FP_EXC_SW_ENABLE) {
1868 if (cpu_has_feature(CPU_FTR_SPE)) {
1870 * When the sticky exception bits are set
1871 * directly by userspace, it must call prctl
1872 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1873 * in the existing prctl settings) or
1874 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1875 * the bits being set). <fenv.h> functions
1876 * saving and restoring the whole
1877 * floating-point environment need to do so
1878 * anyway to restore the prctl settings from
1879 * the saved environment.
1881 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1882 tsk->thread.fpexc_mode = val &
1883 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1893 /* on a CONFIG_SPE this does not hurt us. The bits that
1894 * __pack_fe01 use do not overlap with bits used for
1895 * PR_FP_EXC_SW_ENABLE. Additionally, the MSR[FE0,FE1] bits
1896 * on CONFIG_SPE implementations are reserved so writing to
1897 * them does not change anything */
1898 if (val > PR_FP_EXC_PRECISE)
1900 tsk->thread.fpexc_mode = __pack_fe01(val);
1901 if (regs != NULL && (regs->msr & MSR_FP) != 0)
1902 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1903 | tsk->thread.fpexc_mode;
1907 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1911 if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1913 if (cpu_has_feature(CPU_FTR_SPE)) {
1915 * When the sticky exception bits are set
1916 * directly by userspace, it must call prctl
1917 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1918 * in the existing prctl settings) or
1919 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1920 * the bits being set). <fenv.h> functions
1921 * saving and restoring the whole
1922 * floating-point environment need to do so
1923 * anyway to restore the prctl settings from
1924 * the saved environment.
1926 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1927 val = tsk->thread.fpexc_mode;
1934 val = __unpack_fe01(tsk->thread.fpexc_mode);
1935 return put_user(val, (unsigned int __user *) adr);
1938 int set_endian(struct task_struct *tsk, unsigned int val)
1940 struct pt_regs *regs = tsk->thread.regs;
1942 if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1943 (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1949 if (val == PR_ENDIAN_BIG)
1950 regs->msr &= ~MSR_LE;
1951 else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1952 regs->msr |= MSR_LE;
1959 int get_endian(struct task_struct *tsk, unsigned long adr)
1961 struct pt_regs *regs = tsk->thread.regs;
1964 if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1965 !cpu_has_feature(CPU_FTR_REAL_LE))
1971 if (regs->msr & MSR_LE) {
1972 if (cpu_has_feature(CPU_FTR_REAL_LE))
1973 val = PR_ENDIAN_LITTLE;
1975 val = PR_ENDIAN_PPC_LITTLE;
1977 val = PR_ENDIAN_BIG;
1979 return put_user(val, (unsigned int __user *)adr);
1982 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1984 tsk->thread.align_ctl = val;
1988 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1990 return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1993 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1994 unsigned long nbytes)
1996 unsigned long stack_page;
1997 unsigned long cpu = task_cpu(p);
2000 * Avoid crashing if the stack has overflowed and corrupted
2001 * task_cpu(p), which is in the thread_info struct.
2003 if (cpu < NR_CPUS && cpu_possible(cpu)) {
2004 stack_page = (unsigned long) hardirq_ctx[cpu];
2005 if (sp >= stack_page + sizeof(struct thread_struct)
2006 && sp <= stack_page + THREAD_SIZE - nbytes)
2009 stack_page = (unsigned long) softirq_ctx[cpu];
2010 if (sp >= stack_page + sizeof(struct thread_struct)
2011 && sp <= stack_page + THREAD_SIZE - nbytes)
2017 int validate_sp(unsigned long sp, struct task_struct *p,
2018 unsigned long nbytes)
2020 unsigned long stack_page = (unsigned long)task_stack_page(p);
2022 if (sp >= stack_page + sizeof(struct thread_struct)
2023 && sp <= stack_page + THREAD_SIZE - nbytes)
2026 return valid_irq_stack(sp, p, nbytes);
2029 EXPORT_SYMBOL(validate_sp);
2031 unsigned long get_wchan(struct task_struct *p)
2033 unsigned long ip, sp;
2036 if (!p || p == current || p->state == TASK_RUNNING)
2040 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2044 sp = *(unsigned long *)sp;
2045 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2046 p->state == TASK_RUNNING)
2049 ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2050 if (!in_sched_functions(ip))
2053 } while (count++ < 16);
2057 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2059 void show_stack(struct task_struct *tsk, unsigned long *stack)
2061 unsigned long sp, ip, lr, newsp;
2064 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2065 int curr_frame = current->curr_ret_stack;
2066 extern void return_to_handler(void);
2067 unsigned long rth = (unsigned long)return_to_handler;
2070 sp = (unsigned long) stack;
2075 sp = current_stack_pointer();
2077 sp = tsk->thread.ksp;
2081 printk("Call Trace:\n");
2083 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2086 stack = (unsigned long *) sp;
2088 ip = stack[STACK_FRAME_LR_SAVE];
2089 if (!firstframe || ip != lr) {
2090 printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
2091 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2092 if ((ip == rth) && curr_frame >= 0) {
2094 (void *)current->ret_stack[curr_frame].ret);
2099 pr_cont(" (unreliable)");
2105 * See if this is an exception frame.
2106 * We look for the "regshere" marker in the current frame.
2108 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2109 && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2110 struct pt_regs *regs = (struct pt_regs *)
2111 (sp + STACK_FRAME_OVERHEAD);
2113 printk("--- interrupt: %lx at %pS\n LR = %pS\n",
2114 regs->trap, (void *)regs->nip, (void *)lr);
2119 } while (count++ < kstack_depth_to_print);
2123 /* Called with hard IRQs off */
2124 void notrace __ppc64_runlatch_on(void)
2126 struct thread_info *ti = current_thread_info();
2128 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2130 * Least significant bit (RUN) is the only writable bit of
2131 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2132 * earliest ISA where this is the case, but it's convenient.
2134 mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2139 * Some architectures (e.g., Cell) have writable fields other
2140 * than RUN, so do the read-modify-write.
2142 ctrl = mfspr(SPRN_CTRLF);
2143 ctrl |= CTRL_RUNLATCH;
2144 mtspr(SPRN_CTRLT, ctrl);
2147 ti->local_flags |= _TLF_RUNLATCH;
2150 /* Called with hard IRQs off */
2151 void notrace __ppc64_runlatch_off(void)
2153 struct thread_info *ti = current_thread_info();
2155 ti->local_flags &= ~_TLF_RUNLATCH;
2157 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2158 mtspr(SPRN_CTRLT, 0);
2162 ctrl = mfspr(SPRN_CTRLF);
2163 ctrl &= ~CTRL_RUNLATCH;
2164 mtspr(SPRN_CTRLT, ctrl);
2167 #endif /* CONFIG_PPC64 */
2169 unsigned long arch_align_stack(unsigned long sp)
2171 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2172 sp -= get_random_int() & ~PAGE_MASK;
2176 static inline unsigned long brk_rnd(void)
2178 unsigned long rnd = 0;
2180 /* 8MB for 32bit, 1GB for 64bit */
2181 if (is_32bit_task())
2182 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2184 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2186 return rnd << PAGE_SHIFT;
2189 unsigned long arch_randomize_brk(struct mm_struct *mm)
2191 unsigned long base = mm->brk;
2194 #ifdef CONFIG_PPC_BOOK3S_64
2196 * If we are using 1TB segments and we are allowed to randomise
2197 * the heap, we can put it above 1TB so it is backed by a 1TB
2198 * segment. Otherwise the heap will be in the bottom 1TB
2199 * which always uses 256MB segments and this may result in a
2200 * performance penalty. We don't need to worry about radix. For
2201 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2203 if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2204 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2207 ret = PAGE_ALIGN(base + brk_rnd());