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)
625 current->thread.trap_nr = TRAP_HWBKPT;
626 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
627 11, SIGSEGV) == NOTIFY_STOP)
630 if (debugger_break_match(regs))
633 /* Clear the breakpoint */
634 hw_breakpoint_disable();
636 /* Deliver the signal to userspace */
637 clear_siginfo(&info);
638 info.si_signo = SIGTRAP;
640 info.si_code = TRAP_HWBKPT;
641 info.si_addr = (void __user *)address;
642 force_sig_info(SIGTRAP, &info, current);
644 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
646 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
648 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
650 * Set the debug registers back to their default "safe" values.
652 static void set_debug_reg_defaults(struct thread_struct *thread)
654 thread->debug.iac1 = thread->debug.iac2 = 0;
655 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
656 thread->debug.iac3 = thread->debug.iac4 = 0;
658 thread->debug.dac1 = thread->debug.dac2 = 0;
659 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
660 thread->debug.dvc1 = thread->debug.dvc2 = 0;
662 thread->debug.dbcr0 = 0;
665 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
667 thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
668 DBCR1_IAC3US | DBCR1_IAC4US;
670 * Force Data Address Compare User/Supervisor bits to be User-only
671 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
673 thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
675 thread->debug.dbcr1 = 0;
679 static void prime_debug_regs(struct debug_reg *debug)
682 * We could have inherited MSR_DE from userspace, since
683 * it doesn't get cleared on exception entry. Make sure
684 * MSR_DE is clear before we enable any debug events.
686 mtmsr(mfmsr() & ~MSR_DE);
688 mtspr(SPRN_IAC1, debug->iac1);
689 mtspr(SPRN_IAC2, debug->iac2);
690 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
691 mtspr(SPRN_IAC3, debug->iac3);
692 mtspr(SPRN_IAC4, debug->iac4);
694 mtspr(SPRN_DAC1, debug->dac1);
695 mtspr(SPRN_DAC2, debug->dac2);
696 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
697 mtspr(SPRN_DVC1, debug->dvc1);
698 mtspr(SPRN_DVC2, debug->dvc2);
700 mtspr(SPRN_DBCR0, debug->dbcr0);
701 mtspr(SPRN_DBCR1, debug->dbcr1);
703 mtspr(SPRN_DBCR2, debug->dbcr2);
707 * Unless neither the old or new thread are making use of the
708 * debug registers, set the debug registers from the values
709 * stored in the new thread.
711 void switch_booke_debug_regs(struct debug_reg *new_debug)
713 if ((current->thread.debug.dbcr0 & DBCR0_IDM)
714 || (new_debug->dbcr0 & DBCR0_IDM))
715 prime_debug_regs(new_debug);
717 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
718 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
719 #ifndef CONFIG_HAVE_HW_BREAKPOINT
720 static void set_breakpoint(struct arch_hw_breakpoint *brk)
723 __set_breakpoint(brk);
727 static void set_debug_reg_defaults(struct thread_struct *thread)
729 thread->hw_brk.address = 0;
730 thread->hw_brk.type = 0;
731 if (ppc_breakpoint_available())
732 set_breakpoint(&thread->hw_brk);
734 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
735 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
737 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
738 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
740 mtspr(SPRN_DAC1, dabr);
741 #ifdef CONFIG_PPC_47x
746 #elif defined(CONFIG_PPC_BOOK3S)
747 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
749 mtspr(SPRN_DABR, dabr);
750 if (cpu_has_feature(CPU_FTR_DABRX))
751 mtspr(SPRN_DABRX, dabrx);
754 #elif defined(CONFIG_PPC_8xx)
755 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
757 unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
758 unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
759 unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
761 if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
763 else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
765 else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
768 mtspr(SPRN_LCTRL2, 0);
769 mtspr(SPRN_CMPE, addr);
770 mtspr(SPRN_CMPF, addr + 4);
771 mtspr(SPRN_LCTRL1, lctrl1);
772 mtspr(SPRN_LCTRL2, lctrl2);
777 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
783 static inline int set_dabr(struct arch_hw_breakpoint *brk)
785 unsigned long dabr, dabrx;
787 dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
788 dabrx = ((brk->type >> 3) & 0x7);
791 return ppc_md.set_dabr(dabr, dabrx);
793 return __set_dabr(dabr, dabrx);
796 static inline int set_dawr(struct arch_hw_breakpoint *brk)
798 unsigned long dawr, dawrx, mrd;
802 dawrx = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
803 << (63 - 58); //* read/write bits */
804 dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
805 << (63 - 59); //* translate */
806 dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
807 >> 3; //* PRIM bits */
808 /* dawr length is stored in field MDR bits 48:53. Matches range in
809 doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
811 brk->len is in bytes.
812 This aligns up to double word size, shifts and does the bias.
814 mrd = ((brk->len + 7) >> 3) - 1;
815 dawrx |= (mrd & 0x3f) << (63 - 53);
818 return ppc_md.set_dawr(dawr, dawrx);
819 mtspr(SPRN_DAWR, dawr);
820 mtspr(SPRN_DAWRX, dawrx);
824 void __set_breakpoint(struct arch_hw_breakpoint *brk)
826 memcpy(this_cpu_ptr(¤t_brk), brk, sizeof(*brk));
828 if (cpu_has_feature(CPU_FTR_DAWR))
831 else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
835 // Shouldn't happen due to higher level checks
839 /* Check if we have DAWR or DABR hardware */
840 bool ppc_breakpoint_available(void)
842 if (cpu_has_feature(CPU_FTR_DAWR))
843 return true; /* POWER8 DAWR */
844 if (cpu_has_feature(CPU_FTR_ARCH_207S))
845 return false; /* POWER9 with DAWR disabled */
846 /* DABR: Everything but POWER8 and POWER9 */
849 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
851 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
852 struct arch_hw_breakpoint *b)
854 if (a->address != b->address)
856 if (a->type != b->type)
858 if (a->len != b->len)
863 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
865 static inline bool tm_enabled(struct task_struct *tsk)
867 return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
870 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
873 * Use the current MSR TM suspended bit to track if we have
874 * checkpointed state outstanding.
875 * On signal delivery, we'd normally reclaim the checkpointed
876 * state to obtain stack pointer (see:get_tm_stackpointer()).
877 * This will then directly return to userspace without going
878 * through __switch_to(). However, if the stack frame is bad,
879 * we need to exit this thread which calls __switch_to() which
880 * will again attempt to reclaim the already saved tm state.
881 * Hence we need to check that we've not already reclaimed
883 * We do this using the current MSR, rather tracking it in
884 * some specific thread_struct bit, as it has the additional
885 * benefit of checking for a potential TM bad thing exception.
887 if (!MSR_TM_SUSPENDED(mfmsr()))
890 giveup_all(container_of(thr, struct task_struct, thread));
892 tm_reclaim(thr, cause);
895 * If we are in a transaction and FP is off then we can't have
896 * used FP inside that transaction. Hence the checkpointed
897 * state is the same as the live state. We need to copy the
898 * live state to the checkpointed state so that when the
899 * transaction is restored, the checkpointed state is correct
900 * and the aborted transaction sees the correct state. We use
901 * ckpt_regs.msr here as that's what tm_reclaim will use to
902 * determine if it's going to write the checkpointed state or
903 * not. So either this will write the checkpointed registers,
904 * or reclaim will. Similarly for VMX.
906 if ((thr->ckpt_regs.msr & MSR_FP) == 0)
907 memcpy(&thr->ckfp_state, &thr->fp_state,
908 sizeof(struct thread_fp_state));
909 if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
910 memcpy(&thr->ckvr_state, &thr->vr_state,
911 sizeof(struct thread_vr_state));
914 void tm_reclaim_current(uint8_t cause)
917 tm_reclaim_thread(¤t->thread, cause);
920 static inline void tm_reclaim_task(struct task_struct *tsk)
922 /* We have to work out if we're switching from/to a task that's in the
923 * middle of a transaction.
925 * In switching we need to maintain a 2nd register state as
926 * oldtask->thread.ckpt_regs. We tm_reclaim(oldproc); this saves the
927 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
930 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
932 struct thread_struct *thr = &tsk->thread;
937 if (!MSR_TM_ACTIVE(thr->regs->msr))
938 goto out_and_saveregs;
940 WARN_ON(tm_suspend_disabled);
942 TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
943 "ccr=%lx, msr=%lx, trap=%lx)\n",
944 tsk->pid, thr->regs->nip,
945 thr->regs->ccr, thr->regs->msr,
948 tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
950 TM_DEBUG("--- tm_reclaim on pid %d complete\n",
954 /* Always save the regs here, even if a transaction's not active.
955 * This context-switches a thread's TM info SPRs. We do it here to
956 * be consistent with the restore path (in recheckpoint) which
957 * cannot happen later in _switch().
962 extern void __tm_recheckpoint(struct thread_struct *thread);
964 void tm_recheckpoint(struct thread_struct *thread)
968 if (!(thread->regs->msr & MSR_TM))
971 /* We really can't be interrupted here as the TEXASR registers can't
972 * change and later in the trecheckpoint code, we have a userspace R1.
973 * So let's hard disable over this region.
975 local_irq_save(flags);
978 /* The TM SPRs are restored here, so that TEXASR.FS can be set
979 * before the trecheckpoint and no explosion occurs.
981 tm_restore_sprs(thread);
983 __tm_recheckpoint(thread);
985 local_irq_restore(flags);
988 static inline void tm_recheckpoint_new_task(struct task_struct *new)
990 if (!cpu_has_feature(CPU_FTR_TM))
993 /* Recheckpoint the registers of the thread we're about to switch to.
995 * If the task was using FP, we non-lazily reload both the original and
996 * the speculative FP register states. This is because the kernel
997 * doesn't see if/when a TM rollback occurs, so if we take an FP
998 * unavailable later, we are unable to determine which set of FP regs
999 * need to be restored.
1001 if (!tm_enabled(new))
1004 if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
1005 tm_restore_sprs(&new->thread);
1008 /* Recheckpoint to restore original checkpointed register state. */
1009 TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1010 new->pid, new->thread.regs->msr);
1012 tm_recheckpoint(&new->thread);
1015 * The checkpointed state has been restored but the live state has
1016 * not, ensure all the math functionality is turned off to trigger
1017 * restore_math() to reload.
1019 new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1021 TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1022 "(kernel msr 0x%lx)\n",
1026 static inline void __switch_to_tm(struct task_struct *prev,
1027 struct task_struct *new)
1029 if (cpu_has_feature(CPU_FTR_TM)) {
1030 if (tm_enabled(prev) || tm_enabled(new))
1033 if (tm_enabled(prev)) {
1034 prev->thread.load_tm++;
1035 tm_reclaim_task(prev);
1036 if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1037 prev->thread.regs->msr &= ~MSR_TM;
1040 tm_recheckpoint_new_task(new);
1045 * This is called if we are on the way out to userspace and the
1046 * TIF_RESTORE_TM flag is set. It checks if we need to reload
1047 * FP and/or vector state and does so if necessary.
1048 * If userspace is inside a transaction (whether active or
1049 * suspended) and FP/VMX/VSX instructions have ever been enabled
1050 * inside that transaction, then we have to keep them enabled
1051 * and keep the FP/VMX/VSX state loaded while ever the transaction
1052 * continues. The reason is that if we didn't, and subsequently
1053 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1054 * we don't know whether it's the same transaction, and thus we
1055 * don't know which of the checkpointed state and the transactional
1058 void restore_tm_state(struct pt_regs *regs)
1060 unsigned long msr_diff;
1063 * This is the only moment we should clear TIF_RESTORE_TM as
1064 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1065 * again, anything else could lead to an incorrect ckpt_msr being
1066 * saved and therefore incorrect signal contexts.
1068 clear_thread_flag(TIF_RESTORE_TM);
1069 if (!MSR_TM_ACTIVE(regs->msr))
1072 msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1073 msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1075 /* Ensure that restore_math() will restore */
1076 if (msr_diff & MSR_FP)
1077 current->thread.load_fp = 1;
1078 #ifdef CONFIG_ALTIVEC
1079 if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1080 current->thread.load_vec = 1;
1084 regs->msr |= msr_diff;
1088 #define tm_recheckpoint_new_task(new)
1089 #define __switch_to_tm(prev, new)
1090 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1092 static inline void save_sprs(struct thread_struct *t)
1094 #ifdef CONFIG_ALTIVEC
1095 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1096 t->vrsave = mfspr(SPRN_VRSAVE);
1098 #ifdef CONFIG_PPC_BOOK3S_64
1099 if (cpu_has_feature(CPU_FTR_DSCR))
1100 t->dscr = mfspr(SPRN_DSCR);
1102 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1103 t->bescr = mfspr(SPRN_BESCR);
1104 t->ebbhr = mfspr(SPRN_EBBHR);
1105 t->ebbrr = mfspr(SPRN_EBBRR);
1107 t->fscr = mfspr(SPRN_FSCR);
1110 * Note that the TAR is not available for use in the kernel.
1111 * (To provide this, the TAR should be backed up/restored on
1112 * exception entry/exit instead, and be in pt_regs. FIXME,
1113 * this should be in pt_regs anyway (for debug).)
1115 t->tar = mfspr(SPRN_TAR);
1119 thread_pkey_regs_save(t);
1122 static inline void restore_sprs(struct thread_struct *old_thread,
1123 struct thread_struct *new_thread)
1125 #ifdef CONFIG_ALTIVEC
1126 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1127 old_thread->vrsave != new_thread->vrsave)
1128 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1130 #ifdef CONFIG_PPC_BOOK3S_64
1131 if (cpu_has_feature(CPU_FTR_DSCR)) {
1132 u64 dscr = get_paca()->dscr_default;
1133 if (new_thread->dscr_inherit)
1134 dscr = new_thread->dscr;
1136 if (old_thread->dscr != dscr)
1137 mtspr(SPRN_DSCR, dscr);
1140 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1141 if (old_thread->bescr != new_thread->bescr)
1142 mtspr(SPRN_BESCR, new_thread->bescr);
1143 if (old_thread->ebbhr != new_thread->ebbhr)
1144 mtspr(SPRN_EBBHR, new_thread->ebbhr);
1145 if (old_thread->ebbrr != new_thread->ebbrr)
1146 mtspr(SPRN_EBBRR, new_thread->ebbrr);
1148 if (old_thread->fscr != new_thread->fscr)
1149 mtspr(SPRN_FSCR, new_thread->fscr);
1151 if (old_thread->tar != new_thread->tar)
1152 mtspr(SPRN_TAR, new_thread->tar);
1155 if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1156 old_thread->tidr != new_thread->tidr)
1157 mtspr(SPRN_TIDR, new_thread->tidr);
1160 thread_pkey_regs_restore(new_thread, old_thread);
1163 #ifdef CONFIG_PPC_BOOK3S_64
1165 static const u8 dummy_copy_buffer[CP_SIZE] __attribute__((aligned(CP_SIZE)));
1168 struct task_struct *__switch_to(struct task_struct *prev,
1169 struct task_struct *new)
1171 struct thread_struct *new_thread, *old_thread;
1172 struct task_struct *last;
1173 #ifdef CONFIG_PPC_BOOK3S_64
1174 struct ppc64_tlb_batch *batch;
1177 new_thread = &new->thread;
1178 old_thread = ¤t->thread;
1180 WARN_ON(!irqs_disabled());
1182 #ifdef CONFIG_PPC_BOOK3S_64
1183 batch = this_cpu_ptr(&ppc64_tlb_batch);
1184 if (batch->active) {
1185 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1187 __flush_tlb_pending(batch);
1190 #endif /* CONFIG_PPC_BOOK3S_64 */
1192 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1193 switch_booke_debug_regs(&new->thread.debug);
1196 * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1199 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1200 if (unlikely(!hw_brk_match(this_cpu_ptr(¤t_brk), &new->thread.hw_brk)))
1201 __set_breakpoint(&new->thread.hw_brk);
1202 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1206 * We need to save SPRs before treclaim/trecheckpoint as these will
1207 * change a number of them.
1209 save_sprs(&prev->thread);
1211 /* Save FPU, Altivec, VSX and SPE state */
1214 __switch_to_tm(prev, new);
1216 if (!radix_enabled()) {
1218 * We can't take a PMU exception inside _switch() since there
1219 * is a window where the kernel stack SLB and the kernel stack
1220 * are out of sync. Hard disable here.
1226 * Call restore_sprs() before calling _switch(). If we move it after
1227 * _switch() then we miss out on calling it for new tasks. The reason
1228 * for this is we manually create a stack frame for new tasks that
1229 * directly returns through ret_from_fork() or
1230 * ret_from_kernel_thread(). See copy_thread() for details.
1232 restore_sprs(old_thread, new_thread);
1234 last = _switch(old_thread, new_thread);
1236 #ifdef CONFIG_PPC_BOOK3S_64
1237 if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1238 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1239 batch = this_cpu_ptr(&ppc64_tlb_batch);
1243 if (current_thread_info()->task->thread.regs) {
1244 restore_math(current_thread_info()->task->thread.regs);
1247 * The copy-paste buffer can only store into foreign real
1248 * addresses, so unprivileged processes can not see the
1249 * data or use it in any way unless they have foreign real
1250 * mappings. If the new process has the foreign real address
1251 * mappings, we must issue a cp_abort to clear any state and
1252 * prevent snooping, corruption or a covert channel.
1254 if (current_thread_info()->task->thread.used_vas)
1255 asm volatile(PPC_CP_ABORT);
1257 #endif /* CONFIG_PPC_BOOK3S_64 */
1262 static int instructions_to_print = 16;
1264 static void show_instructions(struct pt_regs *regs)
1267 unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
1270 printk("Instruction dump:");
1272 for (i = 0; i < instructions_to_print; i++) {
1278 #if !defined(CONFIG_BOOKE)
1279 /* If executing with the IMMU off, adjust pc rather
1280 * than print XXXXXXXX.
1282 if (!(regs->msr & MSR_IR))
1283 pc = (unsigned long)phys_to_virt(pc);
1286 if (!__kernel_text_address(pc) ||
1287 probe_kernel_address((unsigned int __user *)pc, instr)) {
1288 pr_cont("XXXXXXXX ");
1290 if (regs->nip == pc)
1291 pr_cont("<%08x> ", instr);
1293 pr_cont("%08x ", instr);
1302 void show_user_instructions(struct pt_regs *regs)
1307 pc = regs->nip - (instructions_to_print * 3 / 4 * sizeof(int));
1310 * Make sure the NIP points at userspace, not kernel text/data or
1313 if (!__access_ok(pc, instructions_to_print * sizeof(int), USER_DS)) {
1314 pr_info("%s[%d]: Bad NIP, not dumping instructions.\n",
1315 current->comm, current->pid);
1319 pr_info("%s[%d]: code: ", current->comm, current->pid);
1321 for (i = 0; i < instructions_to_print; i++) {
1324 if (!(i % 8) && (i > 0)) {
1326 pr_info("%s[%d]: code: ", current->comm, current->pid);
1329 if (probe_kernel_address((unsigned int __user *)pc, instr)) {
1330 pr_cont("XXXXXXXX ");
1332 if (regs->nip == pc)
1333 pr_cont("<%08x> ", instr);
1335 pr_cont("%08x ", instr);
1349 static struct regbit msr_bits[] = {
1350 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1372 #ifndef CONFIG_BOOKE
1379 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1383 for (; bits->bit; ++bits)
1384 if (val & bits->bit) {
1385 pr_cont("%s%s", s, bits->name);
1390 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1391 static struct regbit msr_tm_bits[] = {
1398 static void print_tm_bits(unsigned long val)
1401 * This only prints something if at least one of the TM bit is set.
1402 * Inside the TM[], the output means:
1403 * E: Enabled (bit 32)
1404 * S: Suspended (bit 33)
1405 * T: Transactional (bit 34)
1407 if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1409 print_bits(val, msr_tm_bits, "");
1414 static void print_tm_bits(unsigned long val) {}
1417 static void print_msr_bits(unsigned long val)
1420 print_bits(val, msr_bits, ",");
1426 #define REG "%016lx"
1427 #define REGS_PER_LINE 4
1428 #define LAST_VOLATILE 13
1431 #define REGS_PER_LINE 8
1432 #define LAST_VOLATILE 12
1435 void show_regs(struct pt_regs * regs)
1439 show_regs_print_info(KERN_DEFAULT);
1441 printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1442 regs->nip, regs->link, regs->ctr);
1443 printk("REGS: %px TRAP: %04lx %s (%s)\n",
1444 regs, regs->trap, print_tainted(), init_utsname()->release);
1445 printk("MSR: "REG" ", regs->msr);
1446 print_msr_bits(regs->msr);
1447 pr_cont(" CR: %08lx XER: %08lx\n", regs->ccr, regs->xer);
1449 if ((TRAP(regs) != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1450 pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1451 if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1452 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1453 pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1455 pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1458 pr_cont("IRQMASK: %lx ", regs->softe);
1460 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1461 if (MSR_TM_ACTIVE(regs->msr))
1462 pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1465 for (i = 0; i < 32; i++) {
1466 if ((i % REGS_PER_LINE) == 0)
1467 pr_cont("\nGPR%02d: ", i);
1468 pr_cont(REG " ", regs->gpr[i]);
1469 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1473 #ifdef CONFIG_KALLSYMS
1475 * Lookup NIP late so we have the best change of getting the
1476 * above info out without failing
1478 printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1479 printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1481 show_stack(current, (unsigned long *) regs->gpr[1]);
1482 if (!user_mode(regs))
1483 show_instructions(regs);
1486 void flush_thread(void)
1488 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1489 flush_ptrace_hw_breakpoint(current);
1490 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1491 set_debug_reg_defaults(¤t->thread);
1492 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
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 p->thread.ppr = INIT_PPR;
1719 kregs->nip = ppc_function_entry(f);
1724 * Set up a thread for executing a new program
1726 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1729 unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1733 * If we exec out of a kernel thread then thread.regs will not be
1736 if (!current->thread.regs) {
1737 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1738 current->thread.regs = regs - 1;
1741 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1743 * Clear any transactional state, we're exec()ing. The cause is
1744 * not important as there will never be a recheckpoint so it's not
1747 if (MSR_TM_SUSPENDED(mfmsr()))
1748 tm_reclaim_current(0);
1751 memset(regs->gpr, 0, sizeof(regs->gpr));
1759 * We have just cleared all the nonvolatile GPRs, so make
1760 * FULL_REGS(regs) return true. This is necessary to allow
1761 * ptrace to examine the thread immediately after exec.
1768 regs->msr = MSR_USER;
1770 if (!is_32bit_task()) {
1771 unsigned long entry;
1773 if (is_elf2_task()) {
1774 /* Look ma, no function descriptors! */
1779 * The latest iteration of the ABI requires that when
1780 * calling a function (at its global entry point),
1781 * the caller must ensure r12 holds the entry point
1782 * address (so that the function can quickly
1783 * establish addressability).
1785 regs->gpr[12] = start;
1786 /* Make sure that's restored on entry to userspace. */
1787 set_thread_flag(TIF_RESTOREALL);
1791 /* start is a relocated pointer to the function
1792 * descriptor for the elf _start routine. The first
1793 * entry in the function descriptor is the entry
1794 * address of _start and the second entry is the TOC
1795 * value we need to use.
1797 __get_user(entry, (unsigned long __user *)start);
1798 __get_user(toc, (unsigned long __user *)start+1);
1800 /* Check whether the e_entry function descriptor entries
1801 * need to be relocated before we can use them.
1803 if (load_addr != 0) {
1810 regs->msr = MSR_USER64;
1814 regs->msr = MSR_USER32;
1818 current->thread.used_vsr = 0;
1820 current->thread.load_fp = 0;
1821 memset(¤t->thread.fp_state, 0, sizeof(current->thread.fp_state));
1822 current->thread.fp_save_area = NULL;
1823 #ifdef CONFIG_ALTIVEC
1824 memset(¤t->thread.vr_state, 0, sizeof(current->thread.vr_state));
1825 current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1826 current->thread.vr_save_area = NULL;
1827 current->thread.vrsave = 0;
1828 current->thread.used_vr = 0;
1829 current->thread.load_vec = 0;
1830 #endif /* CONFIG_ALTIVEC */
1832 memset(current->thread.evr, 0, sizeof(current->thread.evr));
1833 current->thread.acc = 0;
1834 current->thread.spefscr = 0;
1835 current->thread.used_spe = 0;
1836 #endif /* CONFIG_SPE */
1837 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1838 current->thread.tm_tfhar = 0;
1839 current->thread.tm_texasr = 0;
1840 current->thread.tm_tfiar = 0;
1841 current->thread.load_tm = 0;
1842 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1844 thread_pkey_regs_init(¤t->thread);
1846 EXPORT_SYMBOL(start_thread);
1848 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1849 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1851 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1853 struct pt_regs *regs = tsk->thread.regs;
1855 /* This is a bit hairy. If we are an SPE enabled processor
1856 * (have embedded fp) we store the IEEE exception enable flags in
1857 * fpexc_mode. fpexc_mode is also used for setting FP exception
1858 * mode (asyn, precise, disabled) for 'Classic' FP. */
1859 if (val & PR_FP_EXC_SW_ENABLE) {
1861 if (cpu_has_feature(CPU_FTR_SPE)) {
1863 * When the sticky exception bits are set
1864 * directly by userspace, it must call prctl
1865 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1866 * in the existing prctl settings) or
1867 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1868 * the bits being set). <fenv.h> functions
1869 * saving and restoring the whole
1870 * floating-point environment need to do so
1871 * anyway to restore the prctl settings from
1872 * the saved environment.
1874 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1875 tsk->thread.fpexc_mode = val &
1876 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1886 /* on a CONFIG_SPE this does not hurt us. The bits that
1887 * __pack_fe01 use do not overlap with bits used for
1888 * PR_FP_EXC_SW_ENABLE. Additionally, the MSR[FE0,FE1] bits
1889 * on CONFIG_SPE implementations are reserved so writing to
1890 * them does not change anything */
1891 if (val > PR_FP_EXC_PRECISE)
1893 tsk->thread.fpexc_mode = __pack_fe01(val);
1894 if (regs != NULL && (regs->msr & MSR_FP) != 0)
1895 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1896 | tsk->thread.fpexc_mode;
1900 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1904 if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1906 if (cpu_has_feature(CPU_FTR_SPE)) {
1908 * When the sticky exception bits are set
1909 * directly by userspace, it must call prctl
1910 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1911 * in the existing prctl settings) or
1912 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1913 * the bits being set). <fenv.h> functions
1914 * saving and restoring the whole
1915 * floating-point environment need to do so
1916 * anyway to restore the prctl settings from
1917 * the saved environment.
1919 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1920 val = tsk->thread.fpexc_mode;
1927 val = __unpack_fe01(tsk->thread.fpexc_mode);
1928 return put_user(val, (unsigned int __user *) adr);
1931 int set_endian(struct task_struct *tsk, unsigned int val)
1933 struct pt_regs *regs = tsk->thread.regs;
1935 if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1936 (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1942 if (val == PR_ENDIAN_BIG)
1943 regs->msr &= ~MSR_LE;
1944 else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1945 regs->msr |= MSR_LE;
1952 int get_endian(struct task_struct *tsk, unsigned long adr)
1954 struct pt_regs *regs = tsk->thread.regs;
1957 if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1958 !cpu_has_feature(CPU_FTR_REAL_LE))
1964 if (regs->msr & MSR_LE) {
1965 if (cpu_has_feature(CPU_FTR_REAL_LE))
1966 val = PR_ENDIAN_LITTLE;
1968 val = PR_ENDIAN_PPC_LITTLE;
1970 val = PR_ENDIAN_BIG;
1972 return put_user(val, (unsigned int __user *)adr);
1975 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1977 tsk->thread.align_ctl = val;
1981 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1983 return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1986 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1987 unsigned long nbytes)
1989 unsigned long stack_page;
1990 unsigned long cpu = task_cpu(p);
1993 * Avoid crashing if the stack has overflowed and corrupted
1994 * task_cpu(p), which is in the thread_info struct.
1996 if (cpu < NR_CPUS && cpu_possible(cpu)) {
1997 stack_page = (unsigned long) hardirq_ctx[cpu];
1998 if (sp >= stack_page + sizeof(struct thread_struct)
1999 && sp <= stack_page + THREAD_SIZE - nbytes)
2002 stack_page = (unsigned long) softirq_ctx[cpu];
2003 if (sp >= stack_page + sizeof(struct thread_struct)
2004 && sp <= stack_page + THREAD_SIZE - nbytes)
2010 int validate_sp(unsigned long sp, struct task_struct *p,
2011 unsigned long nbytes)
2013 unsigned long stack_page = (unsigned long)task_stack_page(p);
2015 if (sp >= stack_page + sizeof(struct thread_struct)
2016 && sp <= stack_page + THREAD_SIZE - nbytes)
2019 return valid_irq_stack(sp, p, nbytes);
2022 EXPORT_SYMBOL(validate_sp);
2024 unsigned long get_wchan(struct task_struct *p)
2026 unsigned long ip, sp;
2029 if (!p || p == current || p->state == TASK_RUNNING)
2033 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2037 sp = *(unsigned long *)sp;
2038 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2039 p->state == TASK_RUNNING)
2042 ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2043 if (!in_sched_functions(ip))
2046 } while (count++ < 16);
2050 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2052 void show_stack(struct task_struct *tsk, unsigned long *stack)
2054 unsigned long sp, ip, lr, newsp;
2057 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2058 int curr_frame = current->curr_ret_stack;
2059 extern void return_to_handler(void);
2060 unsigned long rth = (unsigned long)return_to_handler;
2063 sp = (unsigned long) stack;
2068 sp = current_stack_pointer();
2070 sp = tsk->thread.ksp;
2074 printk("Call Trace:\n");
2076 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2079 stack = (unsigned long *) sp;
2081 ip = stack[STACK_FRAME_LR_SAVE];
2082 if (!firstframe || ip != lr) {
2083 printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
2084 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2085 if ((ip == rth) && curr_frame >= 0) {
2087 (void *)current->ret_stack[curr_frame].ret);
2092 pr_cont(" (unreliable)");
2098 * See if this is an exception frame.
2099 * We look for the "regshere" marker in the current frame.
2101 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2102 && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2103 struct pt_regs *regs = (struct pt_regs *)
2104 (sp + STACK_FRAME_OVERHEAD);
2106 printk("--- interrupt: %lx at %pS\n LR = %pS\n",
2107 regs->trap, (void *)regs->nip, (void *)lr);
2112 } while (count++ < kstack_depth_to_print);
2116 /* Called with hard IRQs off */
2117 void notrace __ppc64_runlatch_on(void)
2119 struct thread_info *ti = current_thread_info();
2121 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2123 * Least significant bit (RUN) is the only writable bit of
2124 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2125 * earliest ISA where this is the case, but it's convenient.
2127 mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2132 * Some architectures (e.g., Cell) have writable fields other
2133 * than RUN, so do the read-modify-write.
2135 ctrl = mfspr(SPRN_CTRLF);
2136 ctrl |= CTRL_RUNLATCH;
2137 mtspr(SPRN_CTRLT, ctrl);
2140 ti->local_flags |= _TLF_RUNLATCH;
2143 /* Called with hard IRQs off */
2144 void notrace __ppc64_runlatch_off(void)
2146 struct thread_info *ti = current_thread_info();
2148 ti->local_flags &= ~_TLF_RUNLATCH;
2150 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2151 mtspr(SPRN_CTRLT, 0);
2155 ctrl = mfspr(SPRN_CTRLF);
2156 ctrl &= ~CTRL_RUNLATCH;
2157 mtspr(SPRN_CTRLT, ctrl);
2160 #endif /* CONFIG_PPC64 */
2162 unsigned long arch_align_stack(unsigned long sp)
2164 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2165 sp -= get_random_int() & ~PAGE_MASK;
2169 static inline unsigned long brk_rnd(void)
2171 unsigned long rnd = 0;
2173 /* 8MB for 32bit, 1GB for 64bit */
2174 if (is_32bit_task())
2175 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2177 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2179 return rnd << PAGE_SHIFT;
2182 unsigned long arch_randomize_brk(struct mm_struct *mm)
2184 unsigned long base = mm->brk;
2187 #ifdef CONFIG_PPC_BOOK3S_64
2189 * If we are using 1TB segments and we are allowed to randomise
2190 * the heap, we can put it above 1TB so it is backed by a 1TB
2191 * segment. Otherwise the heap will be in the bottom 1TB
2192 * which always uses 256MB segments and this may result in a
2193 * performance penalty. We don't need to worry about radix. For
2194 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2196 if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2197 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2200 ret = PAGE_ALIGN(base + brk_rnd());