2 * Derived from "arch/i386/kernel/process.c"
3 * Copyright (C) 1995 Linus Torvalds
5 * Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
6 * Paul Mackerras (paulus@cs.anu.edu.au)
9 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
17 #include <linux/errno.h>
18 #include <linux/sched.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/kernel.h>
24 #include <linux/smp.h>
25 #include <linux/stddef.h>
26 #include <linux/unistd.h>
27 #include <linux/ptrace.h>
28 #include <linux/slab.h>
29 #include <linux/user.h>
30 #include <linux/elf.h>
31 #include <linux/prctl.h>
32 #include <linux/init_task.h>
33 #include <linux/export.h>
34 #include <linux/kallsyms.h>
35 #include <linux/mqueue.h>
36 #include <linux/hardirq.h>
37 #include <linux/utsname.h>
38 #include <linux/ftrace.h>
39 #include <linux/kernel_stat.h>
40 #include <linux/personality.h>
41 #include <linux/random.h>
42 #include <linux/hw_breakpoint.h>
43 #include <linux/uaccess.h>
44 #include <linux/elf-randomize.h>
46 #include <asm/pgtable.h>
48 #include <asm/processor.h>
51 #include <asm/machdep.h>
53 #include <asm/runlatch.h>
54 #include <asm/syscalls.h>
55 #include <asm/switch_to.h>
57 #include <asm/debug.h>
59 #include <asm/firmware.h>
61 #include <asm/code-patching.h>
63 #include <asm/livepatch.h>
64 #include <asm/cpu_has_feature.h>
65 #include <asm/asm-prototypes.h>
67 #include <linux/kprobes.h>
68 #include <linux/kdebug.h>
70 /* Transactional Memory debug */
72 #define TM_DEBUG(x...) printk(KERN_INFO x)
74 #define TM_DEBUG(x...) do { } while(0)
77 extern unsigned long _get_SP(void);
79 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
81 * Are we running in "Suspend disabled" mode? If so we have to block any
82 * sigreturn that would get us into suspended state, and we also warn in some
83 * other paths that we should never reach with suspend disabled.
85 bool tm_suspend_disabled __ro_after_init = false;
87 static void check_if_tm_restore_required(struct task_struct *tsk)
90 * If we are saving the current thread's registers, and the
91 * thread is in a transactional state, set the TIF_RESTORE_TM
92 * bit so that we know to restore the registers before
93 * returning to userspace.
95 if (tsk == current && tsk->thread.regs &&
96 MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
97 !test_thread_flag(TIF_RESTORE_TM)) {
98 tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
99 set_thread_flag(TIF_RESTORE_TM);
103 static inline bool msr_tm_active(unsigned long msr)
105 return MSR_TM_ACTIVE(msr);
108 static bool tm_active_with_fp(struct task_struct *tsk)
110 return msr_tm_active(tsk->thread.regs->msr) &&
111 (tsk->thread.ckpt_regs.msr & MSR_FP);
114 static bool tm_active_with_altivec(struct task_struct *tsk)
116 return msr_tm_active(tsk->thread.regs->msr) &&
117 (tsk->thread.ckpt_regs.msr & MSR_VEC);
120 static inline bool msr_tm_active(unsigned long msr) { return false; }
121 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
122 static inline bool tm_active_with_fp(struct task_struct *tsk) { return false; }
123 static inline bool tm_active_with_altivec(struct task_struct *tsk) { return false; }
124 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
126 bool strict_msr_control;
127 EXPORT_SYMBOL(strict_msr_control);
129 static int __init enable_strict_msr_control(char *str)
131 strict_msr_control = true;
132 pr_info("Enabling strict facility control\n");
136 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
138 unsigned long msr_check_and_set(unsigned long bits)
140 unsigned long oldmsr = mfmsr();
141 unsigned long newmsr;
143 newmsr = oldmsr | bits;
146 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
150 if (oldmsr != newmsr)
156 void __msr_check_and_clear(unsigned long bits)
158 unsigned long oldmsr = mfmsr();
159 unsigned long newmsr;
161 newmsr = oldmsr & ~bits;
164 if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
168 if (oldmsr != newmsr)
171 EXPORT_SYMBOL(__msr_check_and_clear);
173 #ifdef CONFIG_PPC_FPU
174 void __giveup_fpu(struct task_struct *tsk)
179 msr = tsk->thread.regs->msr;
182 if (cpu_has_feature(CPU_FTR_VSX))
185 tsk->thread.regs->msr = msr;
188 void giveup_fpu(struct task_struct *tsk)
190 check_if_tm_restore_required(tsk);
192 msr_check_and_set(MSR_FP);
194 msr_check_and_clear(MSR_FP);
196 EXPORT_SYMBOL(giveup_fpu);
199 * Make sure the floating-point register state in the
200 * the thread_struct is up to date for task tsk.
202 void flush_fp_to_thread(struct task_struct *tsk)
204 if (tsk->thread.regs) {
206 * We need to disable preemption here because if we didn't,
207 * another process could get scheduled after the regs->msr
208 * test but before we have finished saving the FP registers
209 * to the thread_struct. That process could take over the
210 * FPU, and then when we get scheduled again we would store
211 * bogus values for the remaining FP registers.
214 if (tsk->thread.regs->msr & MSR_FP) {
216 * This should only ever be called for current or
217 * for a stopped child process. Since we save away
218 * the FP register state on context switch,
219 * there is something wrong if a stopped child appears
220 * to still have its FP state in the CPU registers.
222 BUG_ON(tsk != current);
228 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
230 void enable_kernel_fp(void)
232 unsigned long cpumsr;
234 WARN_ON(preemptible());
236 cpumsr = msr_check_and_set(MSR_FP);
238 if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
239 check_if_tm_restore_required(current);
241 * If a thread has already been reclaimed then the
242 * checkpointed registers are on the CPU but have definitely
243 * been saved by the reclaim code. Don't need to and *cannot*
244 * giveup as this would save to the 'live' structure not the
245 * checkpointed structure.
247 if(!msr_tm_active(cpumsr) && 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) && msr_tm_active(current->thread.regs->msr))
313 __giveup_altivec(current);
316 EXPORT_SYMBOL(enable_kernel_altivec);
319 * Make sure the VMX/Altivec register state in the
320 * the thread_struct is up to date for task tsk.
322 void flush_altivec_to_thread(struct task_struct *tsk)
324 if (tsk->thread.regs) {
326 if (tsk->thread.regs->msr & MSR_VEC) {
327 BUG_ON(tsk != current);
333 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
335 static int restore_altivec(struct task_struct *tsk)
337 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
338 (tsk->thread.load_vec || tm_active_with_altivec(tsk))) {
339 load_vr_state(&tsk->thread.vr_state);
340 tsk->thread.used_vr = 1;
341 tsk->thread.load_vec++;
348 #define loadvec(thr) 0
349 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
350 #endif /* CONFIG_ALTIVEC */
353 static void __giveup_vsx(struct task_struct *tsk)
355 unsigned long msr = tsk->thread.regs->msr;
358 * We should never be ssetting MSR_VSX without also setting
361 WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
363 /* __giveup_fpu will clear MSR_VSX */
367 __giveup_altivec(tsk);
370 static void giveup_vsx(struct task_struct *tsk)
372 check_if_tm_restore_required(tsk);
374 msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
376 msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
379 void enable_kernel_vsx(void)
381 unsigned long cpumsr;
383 WARN_ON(preemptible());
385 cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
387 if (current->thread.regs &&
388 (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
389 check_if_tm_restore_required(current);
391 * If a thread has already been reclaimed then the
392 * checkpointed registers are on the CPU but have definitely
393 * been saved by the reclaim code. Don't need to and *cannot*
394 * giveup as this would save to the 'live' structure not the
395 * checkpointed structure.
397 if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
399 __giveup_vsx(current);
402 EXPORT_SYMBOL(enable_kernel_vsx);
404 void flush_vsx_to_thread(struct task_struct *tsk)
406 if (tsk->thread.regs) {
408 if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
409 BUG_ON(tsk != current);
415 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
417 static int restore_vsx(struct task_struct *tsk)
419 if (cpu_has_feature(CPU_FTR_VSX)) {
420 tsk->thread.used_vsr = 1;
427 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
428 #endif /* CONFIG_VSX */
431 void giveup_spe(struct task_struct *tsk)
433 check_if_tm_restore_required(tsk);
435 msr_check_and_set(MSR_SPE);
437 msr_check_and_clear(MSR_SPE);
439 EXPORT_SYMBOL(giveup_spe);
441 void enable_kernel_spe(void)
443 WARN_ON(preemptible());
445 msr_check_and_set(MSR_SPE);
447 if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
448 check_if_tm_restore_required(current);
449 __giveup_spe(current);
452 EXPORT_SYMBOL(enable_kernel_spe);
454 void flush_spe_to_thread(struct task_struct *tsk)
456 if (tsk->thread.regs) {
458 if (tsk->thread.regs->msr & MSR_SPE) {
459 BUG_ON(tsk != current);
460 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
466 #endif /* CONFIG_SPE */
468 static unsigned long msr_all_available;
470 static int __init init_msr_all_available(void)
472 #ifdef CONFIG_PPC_FPU
473 msr_all_available |= MSR_FP;
475 #ifdef CONFIG_ALTIVEC
476 if (cpu_has_feature(CPU_FTR_ALTIVEC))
477 msr_all_available |= MSR_VEC;
480 if (cpu_has_feature(CPU_FTR_VSX))
481 msr_all_available |= MSR_VSX;
484 if (cpu_has_feature(CPU_FTR_SPE))
485 msr_all_available |= MSR_SPE;
490 early_initcall(init_msr_all_available);
492 void giveup_all(struct task_struct *tsk)
494 unsigned long usermsr;
496 if (!tsk->thread.regs)
499 usermsr = tsk->thread.regs->msr;
501 if ((usermsr & msr_all_available) == 0)
504 msr_check_and_set(msr_all_available);
505 check_if_tm_restore_required(tsk);
507 WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
509 #ifdef CONFIG_PPC_FPU
510 if (usermsr & MSR_FP)
513 #ifdef CONFIG_ALTIVEC
514 if (usermsr & MSR_VEC)
515 __giveup_altivec(tsk);
518 if (usermsr & MSR_SPE)
522 msr_check_and_clear(msr_all_available);
524 EXPORT_SYMBOL(giveup_all);
526 void restore_math(struct pt_regs *regs)
530 if (!msr_tm_active(regs->msr) &&
531 !current->thread.load_fp && !loadvec(current->thread))
535 msr_check_and_set(msr_all_available);
538 * Only reload if the bit is not set in the user MSR, the bit BEING set
539 * indicates that the registers are hot
541 if ((!(msr & MSR_FP)) && restore_fp(current))
542 msr |= MSR_FP | current->thread.fpexc_mode;
544 if ((!(msr & MSR_VEC)) && restore_altivec(current))
547 if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
548 restore_vsx(current)) {
552 msr_check_and_clear(msr_all_available);
557 void save_all(struct task_struct *tsk)
559 unsigned long usermsr;
561 if (!tsk->thread.regs)
564 usermsr = tsk->thread.regs->msr;
566 if ((usermsr & msr_all_available) == 0)
569 msr_check_and_set(msr_all_available);
571 WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
573 if (usermsr & MSR_FP)
576 if (usermsr & MSR_VEC)
579 if (usermsr & MSR_SPE)
582 msr_check_and_clear(msr_all_available);
585 void flush_all_to_thread(struct task_struct *tsk)
587 if (tsk->thread.regs) {
589 BUG_ON(tsk != current);
593 if (tsk->thread.regs->msr & MSR_SPE)
594 tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
600 EXPORT_SYMBOL(flush_all_to_thread);
602 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
603 void do_send_trap(struct pt_regs *regs, unsigned long address,
604 unsigned long error_code, int breakpt)
606 current->thread.trap_nr = TRAP_HWBKPT;
607 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
608 11, SIGSEGV) == NOTIFY_STOP)
611 /* Deliver the signal to userspace */
612 force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
613 (void __user *)address);
615 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
616 void do_break (struct pt_regs *regs, unsigned long address,
617 unsigned long error_code)
621 current->thread.trap_nr = TRAP_HWBKPT;
622 if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
623 11, SIGSEGV) == NOTIFY_STOP)
626 if (debugger_break_match(regs))
629 /* Clear the breakpoint */
630 hw_breakpoint_disable();
632 /* Deliver the signal to userspace */
633 info.si_signo = SIGTRAP;
635 info.si_code = TRAP_HWBKPT;
636 info.si_addr = (void __user *)address;
637 force_sig_info(SIGTRAP, &info, current);
639 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
641 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
643 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
645 * Set the debug registers back to their default "safe" values.
647 static void set_debug_reg_defaults(struct thread_struct *thread)
649 thread->debug.iac1 = thread->debug.iac2 = 0;
650 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
651 thread->debug.iac3 = thread->debug.iac4 = 0;
653 thread->debug.dac1 = thread->debug.dac2 = 0;
654 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
655 thread->debug.dvc1 = thread->debug.dvc2 = 0;
657 thread->debug.dbcr0 = 0;
660 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
662 thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
663 DBCR1_IAC3US | DBCR1_IAC4US;
665 * Force Data Address Compare User/Supervisor bits to be User-only
666 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
668 thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
670 thread->debug.dbcr1 = 0;
674 static void prime_debug_regs(struct debug_reg *debug)
677 * We could have inherited MSR_DE from userspace, since
678 * it doesn't get cleared on exception entry. Make sure
679 * MSR_DE is clear before we enable any debug events.
681 mtmsr(mfmsr() & ~MSR_DE);
683 mtspr(SPRN_IAC1, debug->iac1);
684 mtspr(SPRN_IAC2, debug->iac2);
685 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
686 mtspr(SPRN_IAC3, debug->iac3);
687 mtspr(SPRN_IAC4, debug->iac4);
689 mtspr(SPRN_DAC1, debug->dac1);
690 mtspr(SPRN_DAC2, debug->dac2);
691 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
692 mtspr(SPRN_DVC1, debug->dvc1);
693 mtspr(SPRN_DVC2, debug->dvc2);
695 mtspr(SPRN_DBCR0, debug->dbcr0);
696 mtspr(SPRN_DBCR1, debug->dbcr1);
698 mtspr(SPRN_DBCR2, debug->dbcr2);
702 * Unless neither the old or new thread are making use of the
703 * debug registers, set the debug registers from the values
704 * stored in the new thread.
706 void switch_booke_debug_regs(struct debug_reg *new_debug)
708 if ((current->thread.debug.dbcr0 & DBCR0_IDM)
709 || (new_debug->dbcr0 & DBCR0_IDM))
710 prime_debug_regs(new_debug);
712 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
713 #else /* !CONFIG_PPC_ADV_DEBUG_REGS */
714 #ifndef CONFIG_HAVE_HW_BREAKPOINT
715 static void set_debug_reg_defaults(struct thread_struct *thread)
717 thread->hw_brk.address = 0;
718 thread->hw_brk.type = 0;
719 set_breakpoint(&thread->hw_brk);
721 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
722 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
724 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
725 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
727 mtspr(SPRN_DAC1, dabr);
728 #ifdef CONFIG_PPC_47x
733 #elif defined(CONFIG_PPC_BOOK3S)
734 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
736 mtspr(SPRN_DABR, dabr);
737 if (cpu_has_feature(CPU_FTR_DABRX))
738 mtspr(SPRN_DABRX, dabrx);
741 #elif defined(CONFIG_PPC_8xx)
742 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
744 unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
745 unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
746 unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
748 if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
750 else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
752 else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
755 mtspr(SPRN_LCTRL2, 0);
756 mtspr(SPRN_CMPE, addr);
757 mtspr(SPRN_CMPF, addr + 4);
758 mtspr(SPRN_LCTRL1, lctrl1);
759 mtspr(SPRN_LCTRL2, lctrl2);
764 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
770 static inline int set_dabr(struct arch_hw_breakpoint *brk)
772 unsigned long dabr, dabrx;
774 dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
775 dabrx = ((brk->type >> 3) & 0x7);
778 return ppc_md.set_dabr(dabr, dabrx);
780 return __set_dabr(dabr, dabrx);
783 static inline int set_dawr(struct arch_hw_breakpoint *brk)
785 unsigned long dawr, dawrx, mrd;
789 dawrx = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
790 << (63 - 58); //* read/write bits */
791 dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
792 << (63 - 59); //* translate */
793 dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
794 >> 3; //* PRIM bits */
795 /* dawr length is stored in field MDR bits 48:53. Matches range in
796 doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
798 brk->len is in bytes.
799 This aligns up to double word size, shifts and does the bias.
801 mrd = ((brk->len + 7) >> 3) - 1;
802 dawrx |= (mrd & 0x3f) << (63 - 53);
805 return ppc_md.set_dawr(dawr, dawrx);
806 mtspr(SPRN_DAWR, dawr);
807 mtspr(SPRN_DAWRX, dawrx);
811 void __set_breakpoint(struct arch_hw_breakpoint *brk)
813 memcpy(this_cpu_ptr(¤t_brk), brk, sizeof(*brk));
815 if (cpu_has_feature(CPU_FTR_DAWR))
821 void set_breakpoint(struct arch_hw_breakpoint *brk)
824 __set_breakpoint(brk);
829 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
832 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
833 struct arch_hw_breakpoint *b)
835 if (a->address != b->address)
837 if (a->type != b->type)
839 if (a->len != b->len)
844 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
846 static inline bool tm_enabled(struct task_struct *tsk)
848 return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
851 static void tm_reclaim_thread(struct thread_struct *thr,
852 struct thread_info *ti, uint8_t cause)
855 * Use the current MSR TM suspended bit to track if we have
856 * checkpointed state outstanding.
857 * On signal delivery, we'd normally reclaim the checkpointed
858 * state to obtain stack pointer (see:get_tm_stackpointer()).
859 * This will then directly return to userspace without going
860 * through __switch_to(). However, if the stack frame is bad,
861 * we need to exit this thread which calls __switch_to() which
862 * will again attempt to reclaim the already saved tm state.
863 * Hence we need to check that we've not already reclaimed
865 * We do this using the current MSR, rather tracking it in
866 * some specific thread_struct bit, as it has the additional
867 * benefit of checking for a potential TM bad thing exception.
869 if (!MSR_TM_SUSPENDED(mfmsr()))
872 giveup_all(container_of(thr, struct task_struct, thread));
874 tm_reclaim(thr, cause);
877 * If we are in a transaction and FP is off then we can't have
878 * used FP inside that transaction. Hence the checkpointed
879 * state is the same as the live state. We need to copy the
880 * live state to the checkpointed state so that when the
881 * transaction is restored, the checkpointed state is correct
882 * and the aborted transaction sees the correct state. We use
883 * ckpt_regs.msr here as that's what tm_reclaim will use to
884 * determine if it's going to write the checkpointed state or
885 * not. So either this will write the checkpointed registers,
886 * or reclaim will. Similarly for VMX.
888 if ((thr->ckpt_regs.msr & MSR_FP) == 0)
889 memcpy(&thr->ckfp_state, &thr->fp_state,
890 sizeof(struct thread_fp_state));
891 if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
892 memcpy(&thr->ckvr_state, &thr->vr_state,
893 sizeof(struct thread_vr_state));
896 void tm_reclaim_current(uint8_t cause)
899 tm_reclaim_thread(¤t->thread, current_thread_info(), cause);
902 static inline void tm_reclaim_task(struct task_struct *tsk)
904 /* We have to work out if we're switching from/to a task that's in the
905 * middle of a transaction.
907 * In switching we need to maintain a 2nd register state as
908 * oldtask->thread.ckpt_regs. We tm_reclaim(oldproc); this saves the
909 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
912 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
914 struct thread_struct *thr = &tsk->thread;
919 if (!MSR_TM_ACTIVE(thr->regs->msr))
920 goto out_and_saveregs;
922 WARN_ON(tm_suspend_disabled);
924 TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
925 "ccr=%lx, msr=%lx, trap=%lx)\n",
926 tsk->pid, thr->regs->nip,
927 thr->regs->ccr, thr->regs->msr,
930 tm_reclaim_thread(thr, task_thread_info(tsk), TM_CAUSE_RESCHED);
932 TM_DEBUG("--- tm_reclaim on pid %d complete\n",
936 /* Always save the regs here, even if a transaction's not active.
937 * This context-switches a thread's TM info SPRs. We do it here to
938 * be consistent with the restore path (in recheckpoint) which
939 * cannot happen later in _switch().
944 extern void __tm_recheckpoint(struct thread_struct *thread);
946 void tm_recheckpoint(struct thread_struct *thread)
950 if (!(thread->regs->msr & MSR_TM))
953 /* We really can't be interrupted here as the TEXASR registers can't
954 * change and later in the trecheckpoint code, we have a userspace R1.
955 * So let's hard disable over this region.
957 local_irq_save(flags);
960 /* The TM SPRs are restored here, so that TEXASR.FS can be set
961 * before the trecheckpoint and no explosion occurs.
963 tm_restore_sprs(thread);
965 __tm_recheckpoint(thread);
967 local_irq_restore(flags);
970 static inline void tm_recheckpoint_new_task(struct task_struct *new)
972 if (!cpu_has_feature(CPU_FTR_TM))
975 /* Recheckpoint the registers of the thread we're about to switch to.
977 * If the task was using FP, we non-lazily reload both the original and
978 * the speculative FP register states. This is because the kernel
979 * doesn't see if/when a TM rollback occurs, so if we take an FP
980 * unavailable later, we are unable to determine which set of FP regs
981 * need to be restored.
983 if (!tm_enabled(new))
986 if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
987 tm_restore_sprs(&new->thread);
990 /* Recheckpoint to restore original checkpointed register state. */
991 TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
992 new->pid, new->thread.regs->msr);
994 tm_recheckpoint(&new->thread);
997 * The checkpointed state has been restored but the live state has
998 * not, ensure all the math functionality is turned off to trigger
999 * restore_math() to reload.
1001 new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1003 TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1004 "(kernel msr 0x%lx)\n",
1008 static inline void __switch_to_tm(struct task_struct *prev,
1009 struct task_struct *new)
1011 if (cpu_has_feature(CPU_FTR_TM)) {
1012 if (tm_enabled(prev) || tm_enabled(new))
1015 if (tm_enabled(prev)) {
1016 prev->thread.load_tm++;
1017 tm_reclaim_task(prev);
1018 if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1019 prev->thread.regs->msr &= ~MSR_TM;
1022 tm_recheckpoint_new_task(new);
1027 * This is called if we are on the way out to userspace and the
1028 * TIF_RESTORE_TM flag is set. It checks if we need to reload
1029 * FP and/or vector state and does so if necessary.
1030 * If userspace is inside a transaction (whether active or
1031 * suspended) and FP/VMX/VSX instructions have ever been enabled
1032 * inside that transaction, then we have to keep them enabled
1033 * and keep the FP/VMX/VSX state loaded while ever the transaction
1034 * continues. The reason is that if we didn't, and subsequently
1035 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1036 * we don't know whether it's the same transaction, and thus we
1037 * don't know which of the checkpointed state and the transactional
1040 void restore_tm_state(struct pt_regs *regs)
1042 unsigned long msr_diff;
1045 * This is the only moment we should clear TIF_RESTORE_TM as
1046 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1047 * again, anything else could lead to an incorrect ckpt_msr being
1048 * saved and therefore incorrect signal contexts.
1050 clear_thread_flag(TIF_RESTORE_TM);
1051 if (!MSR_TM_ACTIVE(regs->msr))
1054 msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1055 msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1057 /* Ensure that restore_math() will restore */
1058 if (msr_diff & MSR_FP)
1059 current->thread.load_fp = 1;
1060 #ifdef CONFIG_ALTIVEC
1061 if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1062 current->thread.load_vec = 1;
1066 regs->msr |= msr_diff;
1070 #define tm_recheckpoint_new_task(new)
1071 #define __switch_to_tm(prev, new)
1072 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1074 static inline void save_sprs(struct thread_struct *t)
1076 #ifdef CONFIG_ALTIVEC
1077 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1078 t->vrsave = mfspr(SPRN_VRSAVE);
1080 #ifdef CONFIG_PPC_BOOK3S_64
1081 if (cpu_has_feature(CPU_FTR_DSCR))
1082 t->dscr = mfspr(SPRN_DSCR);
1084 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1085 t->bescr = mfspr(SPRN_BESCR);
1086 t->ebbhr = mfspr(SPRN_EBBHR);
1087 t->ebbrr = mfspr(SPRN_EBBRR);
1089 t->fscr = mfspr(SPRN_FSCR);
1092 * Note that the TAR is not available for use in the kernel.
1093 * (To provide this, the TAR should be backed up/restored on
1094 * exception entry/exit instead, and be in pt_regs. FIXME,
1095 * this should be in pt_regs anyway (for debug).)
1097 t->tar = mfspr(SPRN_TAR);
1102 static inline void restore_sprs(struct thread_struct *old_thread,
1103 struct thread_struct *new_thread)
1105 #ifdef CONFIG_ALTIVEC
1106 if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1107 old_thread->vrsave != new_thread->vrsave)
1108 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1110 #ifdef CONFIG_PPC_BOOK3S_64
1111 if (cpu_has_feature(CPU_FTR_DSCR)) {
1112 u64 dscr = get_paca()->dscr_default;
1113 if (new_thread->dscr_inherit)
1114 dscr = new_thread->dscr;
1116 if (old_thread->dscr != dscr)
1117 mtspr(SPRN_DSCR, dscr);
1120 if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1121 if (old_thread->bescr != new_thread->bescr)
1122 mtspr(SPRN_BESCR, new_thread->bescr);
1123 if (old_thread->ebbhr != new_thread->ebbhr)
1124 mtspr(SPRN_EBBHR, new_thread->ebbhr);
1125 if (old_thread->ebbrr != new_thread->ebbrr)
1126 mtspr(SPRN_EBBRR, new_thread->ebbrr);
1128 if (old_thread->fscr != new_thread->fscr)
1129 mtspr(SPRN_FSCR, new_thread->fscr);
1131 if (old_thread->tar != new_thread->tar)
1132 mtspr(SPRN_TAR, new_thread->tar);
1135 if (cpu_has_feature(CPU_FTR_ARCH_300) &&
1136 old_thread->tidr != new_thread->tidr)
1137 mtspr(SPRN_TIDR, new_thread->tidr);
1141 #ifdef CONFIG_PPC_BOOK3S_64
1143 static const u8 dummy_copy_buffer[CP_SIZE] __attribute__((aligned(CP_SIZE)));
1146 struct task_struct *__switch_to(struct task_struct *prev,
1147 struct task_struct *new)
1149 struct thread_struct *new_thread, *old_thread;
1150 struct task_struct *last;
1151 #ifdef CONFIG_PPC_BOOK3S_64
1152 struct ppc64_tlb_batch *batch;
1155 new_thread = &new->thread;
1156 old_thread = ¤t->thread;
1158 WARN_ON(!irqs_disabled());
1162 * Collect processor utilization data per process
1164 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
1165 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
1166 long unsigned start_tb, current_tb;
1167 start_tb = old_thread->start_tb;
1168 cu->current_tb = current_tb = mfspr(SPRN_PURR);
1169 old_thread->accum_tb += (current_tb - start_tb);
1170 new_thread->start_tb = current_tb;
1172 #endif /* CONFIG_PPC64 */
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 * DD1 allows paste into normal system memory so we do an
1247 * unpaired copy, rather than cp_abort, to clear the buffer,
1248 * since cp_abort is quite expensive.
1250 if (current_thread_info()->task->thread.used_vas) {
1251 asm volatile(PPC_CP_ABORT);
1252 } else if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
1253 asm volatile(PPC_COPY(%0, %1)
1254 : : "r"(dummy_copy_buffer), "r"(0));
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);
1307 static struct regbit msr_bits[] = {
1308 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1330 #ifndef CONFIG_BOOKE
1337 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1341 for (; bits->bit; ++bits)
1342 if (val & bits->bit) {
1343 pr_cont("%s%s", s, bits->name);
1348 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1349 static struct regbit msr_tm_bits[] = {
1356 static void print_tm_bits(unsigned long val)
1359 * This only prints something if at least one of the TM bit is set.
1360 * Inside the TM[], the output means:
1361 * E: Enabled (bit 32)
1362 * S: Suspended (bit 33)
1363 * T: Transactional (bit 34)
1365 if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1367 print_bits(val, msr_tm_bits, "");
1372 static void print_tm_bits(unsigned long val) {}
1375 static void print_msr_bits(unsigned long val)
1378 print_bits(val, msr_bits, ",");
1384 #define REG "%016lx"
1385 #define REGS_PER_LINE 4
1386 #define LAST_VOLATILE 13
1389 #define REGS_PER_LINE 8
1390 #define LAST_VOLATILE 12
1393 void show_regs(struct pt_regs * regs)
1397 show_regs_print_info(KERN_DEFAULT);
1399 printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1400 regs->nip, regs->link, regs->ctr);
1401 printk("REGS: %px TRAP: %04lx %s (%s)\n",
1402 regs, regs->trap, print_tainted(), init_utsname()->release);
1403 printk("MSR: "REG" ", regs->msr);
1404 print_msr_bits(regs->msr);
1405 pr_cont(" CR: %08lx XER: %08lx\n", regs->ccr, regs->xer);
1407 if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1408 pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1409 if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1410 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1411 pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1413 pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1416 pr_cont("SOFTE: %ld ", regs->softe);
1418 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1419 if (MSR_TM_ACTIVE(regs->msr))
1420 pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1423 for (i = 0; i < 32; i++) {
1424 if ((i % REGS_PER_LINE) == 0)
1425 pr_cont("\nGPR%02d: ", i);
1426 pr_cont(REG " ", regs->gpr[i]);
1427 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1431 #ifdef CONFIG_KALLSYMS
1433 * Lookup NIP late so we have the best change of getting the
1434 * above info out without failing
1436 printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1437 printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1439 show_stack(current, (unsigned long *) regs->gpr[1]);
1440 if (!user_mode(regs))
1441 show_instructions(regs);
1444 void flush_thread(void)
1446 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1447 flush_ptrace_hw_breakpoint(current);
1448 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1449 set_debug_reg_defaults(¤t->thread);
1450 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1453 int set_thread_uses_vas(void)
1455 #ifdef CONFIG_PPC_BOOK3S_64
1456 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1459 current->thread.used_vas = 1;
1462 * Even a process that has no foreign real address mapping can use
1463 * an unpaired COPY instruction (to no real effect). Issue CP_ABORT
1464 * to clear any pending COPY and prevent a covert channel.
1466 * __switch_to() will issue CP_ABORT on future context switches.
1468 asm volatile(PPC_CP_ABORT);
1470 #endif /* CONFIG_PPC_BOOK3S_64 */
1475 static DEFINE_SPINLOCK(vas_thread_id_lock);
1476 static DEFINE_IDA(vas_thread_ida);
1479 * We need to assign a unique thread id to each thread in a process.
1481 * This thread id, referred to as TIDR, and separate from the Linux's tgid,
1482 * is intended to be used to direct an ASB_Notify from the hardware to the
1483 * thread, when a suitable event occurs in the system.
1485 * One such event is a "paste" instruction in the context of Fast Thread
1486 * Wakeup (aka Core-to-core wake up in the Virtual Accelerator Switchboard
1489 * To get a unique TIDR per process we could simply reuse task_pid_nr() but
1490 * the problem is that task_pid_nr() is not yet available copy_thread() is
1491 * called. Fixing that would require changing more intrusive arch-neutral
1492 * code in code path in copy_process()?.
1494 * Further, to assign unique TIDRs within each process, we need an atomic
1495 * field (or an IDR) in task_struct, which again intrudes into the arch-
1496 * neutral code. So try to assign globally unique TIDRs for now.
1498 * NOTE: TIDR 0 indicates that the thread does not need a TIDR value.
1499 * For now, only threads that expect to be notified by the VAS
1500 * hardware need a TIDR value and we assign values > 0 for those.
1502 #define MAX_THREAD_CONTEXT ((1 << 16) - 1)
1503 static int assign_thread_tidr(void)
1509 if (!ida_pre_get(&vas_thread_ida, GFP_KERNEL))
1512 spin_lock(&vas_thread_id_lock);
1513 err = ida_get_new_above(&vas_thread_ida, 1, &index);
1514 spin_unlock(&vas_thread_id_lock);
1521 if (index > MAX_THREAD_CONTEXT) {
1522 spin_lock(&vas_thread_id_lock);
1523 ida_remove(&vas_thread_ida, index);
1524 spin_unlock(&vas_thread_id_lock);
1531 static void free_thread_tidr(int id)
1533 spin_lock(&vas_thread_id_lock);
1534 ida_remove(&vas_thread_ida, id);
1535 spin_unlock(&vas_thread_id_lock);
1539 * Clear any TIDR value assigned to this thread.
1541 void clear_thread_tidr(struct task_struct *t)
1543 if (!t->thread.tidr)
1546 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1551 mtspr(SPRN_TIDR, 0);
1552 free_thread_tidr(t->thread.tidr);
1556 void arch_release_task_struct(struct task_struct *t)
1558 clear_thread_tidr(t);
1562 * Assign a unique TIDR (thread id) for task @t and set it in the thread
1563 * structure. For now, we only support setting TIDR for 'current' task.
1565 int set_thread_tidr(struct task_struct *t)
1569 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1578 rc = assign_thread_tidr();
1582 t->thread.tidr = rc;
1583 mtspr(SPRN_TIDR, t->thread.tidr);
1588 #endif /* CONFIG_PPC64 */
1591 release_thread(struct task_struct *t)
1596 * this gets called so that we can store coprocessor state into memory and
1597 * copy the current task into the new thread.
1599 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1601 flush_all_to_thread(src);
1603 * Flush TM state out so we can copy it. __switch_to_tm() does this
1604 * flush but it removes the checkpointed state from the current CPU and
1605 * transitions the CPU out of TM mode. Hence we need to call
1606 * tm_recheckpoint_new_task() (on the same task) to restore the
1607 * checkpointed state back and the TM mode.
1609 * Can't pass dst because it isn't ready. Doesn't matter, passing
1610 * dst is only important for __switch_to()
1612 __switch_to_tm(src, src);
1616 clear_task_ebb(dst);
1621 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1623 #ifdef CONFIG_PPC_BOOK3S_64
1624 unsigned long sp_vsid;
1625 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1627 if (radix_enabled())
1630 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1631 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1632 << SLB_VSID_SHIFT_1T;
1634 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1636 sp_vsid |= SLB_VSID_KERNEL | llp;
1637 p->thread.ksp_vsid = sp_vsid;
1646 * Copy architecture-specific thread state
1648 int copy_thread(unsigned long clone_flags, unsigned long usp,
1649 unsigned long kthread_arg, struct task_struct *p)
1651 struct pt_regs *childregs, *kregs;
1652 extern void ret_from_fork(void);
1653 extern void ret_from_kernel_thread(void);
1655 unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1656 struct thread_info *ti = task_thread_info(p);
1658 klp_init_thread_info(ti);
1660 /* Copy registers */
1661 sp -= sizeof(struct pt_regs);
1662 childregs = (struct pt_regs *) sp;
1663 if (unlikely(p->flags & PF_KTHREAD)) {
1665 memset(childregs, 0, sizeof(struct pt_regs));
1666 childregs->gpr[1] = sp + sizeof(struct pt_regs);
1669 childregs->gpr[14] = ppc_function_entry((void *)usp);
1671 clear_tsk_thread_flag(p, TIF_32BIT);
1672 childregs->softe = 1;
1674 childregs->gpr[15] = kthread_arg;
1675 p->thread.regs = NULL; /* no user register state */
1676 ti->flags |= _TIF_RESTOREALL;
1677 f = ret_from_kernel_thread;
1680 struct pt_regs *regs = current_pt_regs();
1681 CHECK_FULL_REGS(regs);
1684 childregs->gpr[1] = usp;
1685 p->thread.regs = childregs;
1686 childregs->gpr[3] = 0; /* Result from fork() */
1687 if (clone_flags & CLONE_SETTLS) {
1689 if (!is_32bit_task())
1690 childregs->gpr[13] = childregs->gpr[6];
1693 childregs->gpr[2] = childregs->gpr[6];
1698 childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1699 sp -= STACK_FRAME_OVERHEAD;
1702 * The way this works is that at some point in the future
1703 * some task will call _switch to switch to the new task.
1704 * That will pop off the stack frame created below and start
1705 * the new task running at ret_from_fork. The new task will
1706 * do some house keeping and then return from the fork or clone
1707 * system call, using the stack frame created above.
1709 ((unsigned long *)sp)[0] = 0;
1710 sp -= sizeof(struct pt_regs);
1711 kregs = (struct pt_regs *) sp;
1712 sp -= STACK_FRAME_OVERHEAD;
1715 p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1716 _ALIGN_UP(sizeof(struct thread_info), 16);
1718 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1719 p->thread.ptrace_bps[0] = NULL;
1722 p->thread.fp_save_area = NULL;
1723 #ifdef CONFIG_ALTIVEC
1724 p->thread.vr_save_area = NULL;
1727 setup_ksp_vsid(p, sp);
1730 if (cpu_has_feature(CPU_FTR_DSCR)) {
1731 p->thread.dscr_inherit = current->thread.dscr_inherit;
1732 p->thread.dscr = mfspr(SPRN_DSCR);
1734 if (cpu_has_feature(CPU_FTR_HAS_PPR))
1735 p->thread.ppr = INIT_PPR;
1739 kregs->nip = ppc_function_entry(f);
1744 * Set up a thread for executing a new program
1746 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1749 unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1753 * If we exec out of a kernel thread then thread.regs will not be
1756 if (!current->thread.regs) {
1757 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1758 current->thread.regs = regs - 1;
1761 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1763 * Clear any transactional state, we're exec()ing. The cause is
1764 * not important as there will never be a recheckpoint so it's not
1767 if (MSR_TM_SUSPENDED(mfmsr()))
1768 tm_reclaim_current(0);
1771 memset(regs->gpr, 0, sizeof(regs->gpr));
1779 * We have just cleared all the nonvolatile GPRs, so make
1780 * FULL_REGS(regs) return true. This is necessary to allow
1781 * ptrace to examine the thread immediately after exec.
1788 regs->msr = MSR_USER;
1790 if (!is_32bit_task()) {
1791 unsigned long entry;
1793 if (is_elf2_task()) {
1794 /* Look ma, no function descriptors! */
1799 * The latest iteration of the ABI requires that when
1800 * calling a function (at its global entry point),
1801 * the caller must ensure r12 holds the entry point
1802 * address (so that the function can quickly
1803 * establish addressability).
1805 regs->gpr[12] = start;
1806 /* Make sure that's restored on entry to userspace. */
1807 set_thread_flag(TIF_RESTOREALL);
1811 /* start is a relocated pointer to the function
1812 * descriptor for the elf _start routine. The first
1813 * entry in the function descriptor is the entry
1814 * address of _start and the second entry is the TOC
1815 * value we need to use.
1817 __get_user(entry, (unsigned long __user *)start);
1818 __get_user(toc, (unsigned long __user *)start+1);
1820 /* Check whether the e_entry function descriptor entries
1821 * need to be relocated before we can use them.
1823 if (load_addr != 0) {
1830 regs->msr = MSR_USER64;
1834 regs->msr = MSR_USER32;
1838 current->thread.used_vsr = 0;
1840 current->thread.load_fp = 0;
1841 memset(¤t->thread.fp_state, 0, sizeof(current->thread.fp_state));
1842 current->thread.fp_save_area = NULL;
1843 #ifdef CONFIG_ALTIVEC
1844 memset(¤t->thread.vr_state, 0, sizeof(current->thread.vr_state));
1845 current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1846 current->thread.vr_save_area = NULL;
1847 current->thread.vrsave = 0;
1848 current->thread.used_vr = 0;
1849 current->thread.load_vec = 0;
1850 #endif /* CONFIG_ALTIVEC */
1852 memset(current->thread.evr, 0, sizeof(current->thread.evr));
1853 current->thread.acc = 0;
1854 current->thread.spefscr = 0;
1855 current->thread.used_spe = 0;
1856 #endif /* CONFIG_SPE */
1857 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1858 current->thread.tm_tfhar = 0;
1859 current->thread.tm_texasr = 0;
1860 current->thread.tm_tfiar = 0;
1861 current->thread.load_tm = 0;
1862 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1864 EXPORT_SYMBOL(start_thread);
1866 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1867 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1869 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1871 struct pt_regs *regs = tsk->thread.regs;
1873 /* This is a bit hairy. If we are an SPE enabled processor
1874 * (have embedded fp) we store the IEEE exception enable flags in
1875 * fpexc_mode. fpexc_mode is also used for setting FP exception
1876 * mode (asyn, precise, disabled) for 'Classic' FP. */
1877 if (val & PR_FP_EXC_SW_ENABLE) {
1879 if (cpu_has_feature(CPU_FTR_SPE)) {
1881 * When the sticky exception bits are set
1882 * directly by userspace, it must call prctl
1883 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1884 * in the existing prctl settings) or
1885 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1886 * the bits being set). <fenv.h> functions
1887 * saving and restoring the whole
1888 * floating-point environment need to do so
1889 * anyway to restore the prctl settings from
1890 * the saved environment.
1892 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1893 tsk->thread.fpexc_mode = val &
1894 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1904 /* on a CONFIG_SPE this does not hurt us. The bits that
1905 * __pack_fe01 use do not overlap with bits used for
1906 * PR_FP_EXC_SW_ENABLE. Additionally, the MSR[FE0,FE1] bits
1907 * on CONFIG_SPE implementations are reserved so writing to
1908 * them does not change anything */
1909 if (val > PR_FP_EXC_PRECISE)
1911 tsk->thread.fpexc_mode = __pack_fe01(val);
1912 if (regs != NULL && (regs->msr & MSR_FP) != 0)
1913 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1914 | tsk->thread.fpexc_mode;
1918 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1922 if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1924 if (cpu_has_feature(CPU_FTR_SPE)) {
1926 * When the sticky exception bits are set
1927 * directly by userspace, it must call prctl
1928 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1929 * in the existing prctl settings) or
1930 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1931 * the bits being set). <fenv.h> functions
1932 * saving and restoring the whole
1933 * floating-point environment need to do so
1934 * anyway to restore the prctl settings from
1935 * the saved environment.
1937 tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1938 val = tsk->thread.fpexc_mode;
1945 val = __unpack_fe01(tsk->thread.fpexc_mode);
1946 return put_user(val, (unsigned int __user *) adr);
1949 int set_endian(struct task_struct *tsk, unsigned int val)
1951 struct pt_regs *regs = tsk->thread.regs;
1953 if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1954 (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1960 if (val == PR_ENDIAN_BIG)
1961 regs->msr &= ~MSR_LE;
1962 else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1963 regs->msr |= MSR_LE;
1970 int get_endian(struct task_struct *tsk, unsigned long adr)
1972 struct pt_regs *regs = tsk->thread.regs;
1975 if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1976 !cpu_has_feature(CPU_FTR_REAL_LE))
1982 if (regs->msr & MSR_LE) {
1983 if (cpu_has_feature(CPU_FTR_REAL_LE))
1984 val = PR_ENDIAN_LITTLE;
1986 val = PR_ENDIAN_PPC_LITTLE;
1988 val = PR_ENDIAN_BIG;
1990 return put_user(val, (unsigned int __user *)adr);
1993 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1995 tsk->thread.align_ctl = val;
1999 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
2001 return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
2004 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
2005 unsigned long nbytes)
2007 unsigned long stack_page;
2008 unsigned long cpu = task_cpu(p);
2011 * Avoid crashing if the stack has overflowed and corrupted
2012 * task_cpu(p), which is in the thread_info struct.
2014 if (cpu < NR_CPUS && cpu_possible(cpu)) {
2015 stack_page = (unsigned long) hardirq_ctx[cpu];
2016 if (sp >= stack_page + sizeof(struct thread_struct)
2017 && sp <= stack_page + THREAD_SIZE - nbytes)
2020 stack_page = (unsigned long) softirq_ctx[cpu];
2021 if (sp >= stack_page + sizeof(struct thread_struct)
2022 && sp <= stack_page + THREAD_SIZE - nbytes)
2028 int validate_sp(unsigned long sp, struct task_struct *p,
2029 unsigned long nbytes)
2031 unsigned long stack_page = (unsigned long)task_stack_page(p);
2033 if (sp >= stack_page + sizeof(struct thread_struct)
2034 && sp <= stack_page + THREAD_SIZE - nbytes)
2037 return valid_irq_stack(sp, p, nbytes);
2040 EXPORT_SYMBOL(validate_sp);
2042 unsigned long get_wchan(struct task_struct *p)
2044 unsigned long ip, sp;
2047 if (!p || p == current || p->state == TASK_RUNNING)
2051 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2055 sp = *(unsigned long *)sp;
2056 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2057 p->state == TASK_RUNNING)
2060 ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2061 if (!in_sched_functions(ip))
2064 } while (count++ < 16);
2068 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2070 void show_stack(struct task_struct *tsk, unsigned long *stack)
2072 unsigned long sp, ip, lr, newsp;
2075 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2076 int curr_frame = current->curr_ret_stack;
2077 extern void return_to_handler(void);
2078 unsigned long rth = (unsigned long)return_to_handler;
2081 sp = (unsigned long) stack;
2086 sp = current_stack_pointer();
2088 sp = tsk->thread.ksp;
2092 printk("Call Trace:\n");
2094 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2097 stack = (unsigned long *) sp;
2099 ip = stack[STACK_FRAME_LR_SAVE];
2100 if (!firstframe || ip != lr) {
2101 printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
2102 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2103 if ((ip == rth) && curr_frame >= 0) {
2105 (void *)current->ret_stack[curr_frame].ret);
2110 pr_cont(" (unreliable)");
2116 * See if this is an exception frame.
2117 * We look for the "regshere" marker in the current frame.
2119 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2120 && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2121 struct pt_regs *regs = (struct pt_regs *)
2122 (sp + STACK_FRAME_OVERHEAD);
2124 printk("--- interrupt: %lx at %pS\n LR = %pS\n",
2125 regs->trap, (void *)regs->nip, (void *)lr);
2130 } while (count++ < kstack_depth_to_print);
2134 /* Called with hard IRQs off */
2135 void notrace __ppc64_runlatch_on(void)
2137 struct thread_info *ti = current_thread_info();
2139 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2141 * Least significant bit (RUN) is the only writable bit of
2142 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2143 * earliest ISA where this is the case, but it's convenient.
2145 mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2150 * Some architectures (e.g., Cell) have writable fields other
2151 * than RUN, so do the read-modify-write.
2153 ctrl = mfspr(SPRN_CTRLF);
2154 ctrl |= CTRL_RUNLATCH;
2155 mtspr(SPRN_CTRLT, ctrl);
2158 ti->local_flags |= _TLF_RUNLATCH;
2161 /* Called with hard IRQs off */
2162 void notrace __ppc64_runlatch_off(void)
2164 struct thread_info *ti = current_thread_info();
2166 ti->local_flags &= ~_TLF_RUNLATCH;
2168 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2169 mtspr(SPRN_CTRLT, 0);
2173 ctrl = mfspr(SPRN_CTRLF);
2174 ctrl &= ~CTRL_RUNLATCH;
2175 mtspr(SPRN_CTRLT, ctrl);
2178 #endif /* CONFIG_PPC64 */
2180 unsigned long arch_align_stack(unsigned long sp)
2182 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2183 sp -= get_random_int() & ~PAGE_MASK;
2187 static inline unsigned long brk_rnd(void)
2189 unsigned long rnd = 0;
2191 /* 8MB for 32bit, 1GB for 64bit */
2192 if (is_32bit_task())
2193 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2195 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2197 return rnd << PAGE_SHIFT;
2200 unsigned long arch_randomize_brk(struct mm_struct *mm)
2202 unsigned long base = mm->brk;
2205 #ifdef CONFIG_PPC_BOOK3S_64
2207 * If we are using 1TB segments and we are allowed to randomise
2208 * the heap, we can put it above 1TB so it is backed by a 1TB
2209 * segment. Otherwise the heap will be in the bottom 1TB
2210 * which always uses 256MB segments and this may result in a
2211 * performance penalty. We don't need to worry about radix. For
2212 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2214 if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2215 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2218 ret = PAGE_ALIGN(base + brk_rnd());