Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  *  Derived from "arch/i386/kernel/process.c"
0004  *    Copyright (C) 1995  Linus Torvalds
0005  *
0006  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
0007  *  Paul Mackerras (paulus@cs.anu.edu.au)
0008  *
0009  *  PowerPC version
0010  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
0011  */
0012 
0013 #include <linux/errno.h>
0014 #include <linux/sched.h>
0015 #include <linux/sched/debug.h>
0016 #include <linux/sched/task.h>
0017 #include <linux/sched/task_stack.h>
0018 #include <linux/kernel.h>
0019 #include <linux/mm.h>
0020 #include <linux/smp.h>
0021 #include <linux/stddef.h>
0022 #include <linux/unistd.h>
0023 #include <linux/ptrace.h>
0024 #include <linux/slab.h>
0025 #include <linux/user.h>
0026 #include <linux/elf.h>
0027 #include <linux/prctl.h>
0028 #include <linux/init_task.h>
0029 #include <linux/export.h>
0030 #include <linux/kallsyms.h>
0031 #include <linux/mqueue.h>
0032 #include <linux/hardirq.h>
0033 #include <linux/utsname.h>
0034 #include <linux/ftrace.h>
0035 #include <linux/kernel_stat.h>
0036 #include <linux/personality.h>
0037 #include <linux/hw_breakpoint.h>
0038 #include <linux/uaccess.h>
0039 #include <linux/pkeys.h>
0040 #include <linux/seq_buf.h>
0041 
0042 #include <asm/interrupt.h>
0043 #include <asm/io.h>
0044 #include <asm/processor.h>
0045 #include <asm/mmu.h>
0046 #include <asm/machdep.h>
0047 #include <asm/time.h>
0048 #include <asm/runlatch.h>
0049 #include <asm/syscalls.h>
0050 #include <asm/switch_to.h>
0051 #include <asm/tm.h>
0052 #include <asm/debug.h>
0053 #ifdef CONFIG_PPC64
0054 #include <asm/firmware.h>
0055 #include <asm/hw_irq.h>
0056 #endif
0057 #include <asm/code-patching.h>
0058 #include <asm/exec.h>
0059 #include <asm/livepatch.h>
0060 #include <asm/cpu_has_feature.h>
0061 #include <asm/asm-prototypes.h>
0062 #include <asm/stacktrace.h>
0063 #include <asm/hw_breakpoint.h>
0064 
0065 #include <linux/kprobes.h>
0066 #include <linux/kdebug.h>
0067 
0068 /* Transactional Memory debug */
0069 #ifdef TM_DEBUG_SW
0070 #define TM_DEBUG(x...) printk(KERN_INFO x)
0071 #else
0072 #define TM_DEBUG(x...) do { } while(0)
0073 #endif
0074 
0075 extern unsigned long _get_SP(void);
0076 
0077 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
0078 /*
0079  * Are we running in "Suspend disabled" mode? If so we have to block any
0080  * sigreturn that would get us into suspended state, and we also warn in some
0081  * other paths that we should never reach with suspend disabled.
0082  */
0083 bool tm_suspend_disabled __ro_after_init = false;
0084 
0085 static void check_if_tm_restore_required(struct task_struct *tsk)
0086 {
0087     /*
0088      * If we are saving the current thread's registers, and the
0089      * thread is in a transactional state, set the TIF_RESTORE_TM
0090      * bit so that we know to restore the registers before
0091      * returning to userspace.
0092      */
0093     if (tsk == current && tsk->thread.regs &&
0094         MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
0095         !test_thread_flag(TIF_RESTORE_TM)) {
0096         regs_set_return_msr(&tsk->thread.ckpt_regs,
0097                         tsk->thread.regs->msr);
0098         set_thread_flag(TIF_RESTORE_TM);
0099     }
0100 }
0101 
0102 #else
0103 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
0104 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
0105 
0106 bool strict_msr_control;
0107 EXPORT_SYMBOL(strict_msr_control);
0108 
0109 static int __init enable_strict_msr_control(char *str)
0110 {
0111     strict_msr_control = true;
0112     pr_info("Enabling strict facility control\n");
0113 
0114     return 0;
0115 }
0116 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
0117 
0118 /* notrace because it's called by restore_math */
0119 unsigned long notrace msr_check_and_set(unsigned long bits)
0120 {
0121     unsigned long oldmsr = mfmsr();
0122     unsigned long newmsr;
0123 
0124     newmsr = oldmsr | bits;
0125 
0126     if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
0127         newmsr |= MSR_VSX;
0128 
0129     if (oldmsr != newmsr)
0130         mtmsr_isync(newmsr);
0131 
0132     return newmsr;
0133 }
0134 EXPORT_SYMBOL_GPL(msr_check_and_set);
0135 
0136 /* notrace because it's called by restore_math */
0137 void notrace __msr_check_and_clear(unsigned long bits)
0138 {
0139     unsigned long oldmsr = mfmsr();
0140     unsigned long newmsr;
0141 
0142     newmsr = oldmsr & ~bits;
0143 
0144     if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
0145         newmsr &= ~MSR_VSX;
0146 
0147     if (oldmsr != newmsr)
0148         mtmsr_isync(newmsr);
0149 }
0150 EXPORT_SYMBOL(__msr_check_and_clear);
0151 
0152 #ifdef CONFIG_PPC_FPU
0153 static void __giveup_fpu(struct task_struct *tsk)
0154 {
0155     unsigned long msr;
0156 
0157     save_fpu(tsk);
0158     msr = tsk->thread.regs->msr;
0159     msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
0160     if (cpu_has_feature(CPU_FTR_VSX))
0161         msr &= ~MSR_VSX;
0162     regs_set_return_msr(tsk->thread.regs, msr);
0163 }
0164 
0165 void giveup_fpu(struct task_struct *tsk)
0166 {
0167     check_if_tm_restore_required(tsk);
0168 
0169     msr_check_and_set(MSR_FP);
0170     __giveup_fpu(tsk);
0171     msr_check_and_clear(MSR_FP);
0172 }
0173 EXPORT_SYMBOL(giveup_fpu);
0174 
0175 /*
0176  * Make sure the floating-point register state in the
0177  * the thread_struct is up to date for task tsk.
0178  */
0179 void flush_fp_to_thread(struct task_struct *tsk)
0180 {
0181     if (tsk->thread.regs) {
0182         /*
0183          * We need to disable preemption here because if we didn't,
0184          * another process could get scheduled after the regs->msr
0185          * test but before we have finished saving the FP registers
0186          * to the thread_struct.  That process could take over the
0187          * FPU, and then when we get scheduled again we would store
0188          * bogus values for the remaining FP registers.
0189          */
0190         preempt_disable();
0191         if (tsk->thread.regs->msr & MSR_FP) {
0192             /*
0193              * This should only ever be called for current or
0194              * for a stopped child process.  Since we save away
0195              * the FP register state on context switch,
0196              * there is something wrong if a stopped child appears
0197              * to still have its FP state in the CPU registers.
0198              */
0199             BUG_ON(tsk != current);
0200             giveup_fpu(tsk);
0201         }
0202         preempt_enable();
0203     }
0204 }
0205 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
0206 
0207 void enable_kernel_fp(void)
0208 {
0209     unsigned long cpumsr;
0210 
0211     WARN_ON(preemptible());
0212 
0213     cpumsr = msr_check_and_set(MSR_FP);
0214 
0215     if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
0216         check_if_tm_restore_required(current);
0217         /*
0218          * If a thread has already been reclaimed then the
0219          * checkpointed registers are on the CPU but have definitely
0220          * been saved by the reclaim code. Don't need to and *cannot*
0221          * giveup as this would save  to the 'live' structure not the
0222          * checkpointed structure.
0223          */
0224         if (!MSR_TM_ACTIVE(cpumsr) &&
0225              MSR_TM_ACTIVE(current->thread.regs->msr))
0226             return;
0227         __giveup_fpu(current);
0228     }
0229 }
0230 EXPORT_SYMBOL(enable_kernel_fp);
0231 #else
0232 static inline void __giveup_fpu(struct task_struct *tsk) { }
0233 #endif /* CONFIG_PPC_FPU */
0234 
0235 #ifdef CONFIG_ALTIVEC
0236 static void __giveup_altivec(struct task_struct *tsk)
0237 {
0238     unsigned long msr;
0239 
0240     save_altivec(tsk);
0241     msr = tsk->thread.regs->msr;
0242     msr &= ~MSR_VEC;
0243     if (cpu_has_feature(CPU_FTR_VSX))
0244         msr &= ~MSR_VSX;
0245     regs_set_return_msr(tsk->thread.regs, msr);
0246 }
0247 
0248 void giveup_altivec(struct task_struct *tsk)
0249 {
0250     check_if_tm_restore_required(tsk);
0251 
0252     msr_check_and_set(MSR_VEC);
0253     __giveup_altivec(tsk);
0254     msr_check_and_clear(MSR_VEC);
0255 }
0256 EXPORT_SYMBOL(giveup_altivec);
0257 
0258 void enable_kernel_altivec(void)
0259 {
0260     unsigned long cpumsr;
0261 
0262     WARN_ON(preemptible());
0263 
0264     cpumsr = msr_check_and_set(MSR_VEC);
0265 
0266     if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
0267         check_if_tm_restore_required(current);
0268         /*
0269          * If a thread has already been reclaimed then the
0270          * checkpointed registers are on the CPU but have definitely
0271          * been saved by the reclaim code. Don't need to and *cannot*
0272          * giveup as this would save  to the 'live' structure not the
0273          * checkpointed structure.
0274          */
0275         if (!MSR_TM_ACTIVE(cpumsr) &&
0276              MSR_TM_ACTIVE(current->thread.regs->msr))
0277             return;
0278         __giveup_altivec(current);
0279     }
0280 }
0281 EXPORT_SYMBOL(enable_kernel_altivec);
0282 
0283 /*
0284  * Make sure the VMX/Altivec register state in the
0285  * the thread_struct is up to date for task tsk.
0286  */
0287 void flush_altivec_to_thread(struct task_struct *tsk)
0288 {
0289     if (tsk->thread.regs) {
0290         preempt_disable();
0291         if (tsk->thread.regs->msr & MSR_VEC) {
0292             BUG_ON(tsk != current);
0293             giveup_altivec(tsk);
0294         }
0295         preempt_enable();
0296     }
0297 }
0298 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
0299 #endif /* CONFIG_ALTIVEC */
0300 
0301 #ifdef CONFIG_VSX
0302 static void __giveup_vsx(struct task_struct *tsk)
0303 {
0304     unsigned long msr = tsk->thread.regs->msr;
0305 
0306     /*
0307      * We should never be setting MSR_VSX without also setting
0308      * MSR_FP and MSR_VEC
0309      */
0310     WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
0311 
0312     /* __giveup_fpu will clear MSR_VSX */
0313     if (msr & MSR_FP)
0314         __giveup_fpu(tsk);
0315     if (msr & MSR_VEC)
0316         __giveup_altivec(tsk);
0317 }
0318 
0319 static void giveup_vsx(struct task_struct *tsk)
0320 {
0321     check_if_tm_restore_required(tsk);
0322 
0323     msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
0324     __giveup_vsx(tsk);
0325     msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
0326 }
0327 
0328 void enable_kernel_vsx(void)
0329 {
0330     unsigned long cpumsr;
0331 
0332     WARN_ON(preemptible());
0333 
0334     cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
0335 
0336     if (current->thread.regs &&
0337         (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
0338         check_if_tm_restore_required(current);
0339         /*
0340          * If a thread has already been reclaimed then the
0341          * checkpointed registers are on the CPU but have definitely
0342          * been saved by the reclaim code. Don't need to and *cannot*
0343          * giveup as this would save  to the 'live' structure not the
0344          * checkpointed structure.
0345          */
0346         if (!MSR_TM_ACTIVE(cpumsr) &&
0347              MSR_TM_ACTIVE(current->thread.regs->msr))
0348             return;
0349         __giveup_vsx(current);
0350     }
0351 }
0352 EXPORT_SYMBOL(enable_kernel_vsx);
0353 
0354 void flush_vsx_to_thread(struct task_struct *tsk)
0355 {
0356     if (tsk->thread.regs) {
0357         preempt_disable();
0358         if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
0359             BUG_ON(tsk != current);
0360             giveup_vsx(tsk);
0361         }
0362         preempt_enable();
0363     }
0364 }
0365 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
0366 #endif /* CONFIG_VSX */
0367 
0368 #ifdef CONFIG_SPE
0369 void giveup_spe(struct task_struct *tsk)
0370 {
0371     check_if_tm_restore_required(tsk);
0372 
0373     msr_check_and_set(MSR_SPE);
0374     __giveup_spe(tsk);
0375     msr_check_and_clear(MSR_SPE);
0376 }
0377 EXPORT_SYMBOL(giveup_spe);
0378 
0379 void enable_kernel_spe(void)
0380 {
0381     WARN_ON(preemptible());
0382 
0383     msr_check_and_set(MSR_SPE);
0384 
0385     if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
0386         check_if_tm_restore_required(current);
0387         __giveup_spe(current);
0388     }
0389 }
0390 EXPORT_SYMBOL(enable_kernel_spe);
0391 
0392 void flush_spe_to_thread(struct task_struct *tsk)
0393 {
0394     if (tsk->thread.regs) {
0395         preempt_disable();
0396         if (tsk->thread.regs->msr & MSR_SPE) {
0397             BUG_ON(tsk != current);
0398             tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
0399             giveup_spe(tsk);
0400         }
0401         preempt_enable();
0402     }
0403 }
0404 #endif /* CONFIG_SPE */
0405 
0406 static unsigned long msr_all_available;
0407 
0408 static int __init init_msr_all_available(void)
0409 {
0410     if (IS_ENABLED(CONFIG_PPC_FPU))
0411         msr_all_available |= MSR_FP;
0412     if (cpu_has_feature(CPU_FTR_ALTIVEC))
0413         msr_all_available |= MSR_VEC;
0414     if (cpu_has_feature(CPU_FTR_VSX))
0415         msr_all_available |= MSR_VSX;
0416     if (cpu_has_feature(CPU_FTR_SPE))
0417         msr_all_available |= MSR_SPE;
0418 
0419     return 0;
0420 }
0421 early_initcall(init_msr_all_available);
0422 
0423 void giveup_all(struct task_struct *tsk)
0424 {
0425     unsigned long usermsr;
0426 
0427     if (!tsk->thread.regs)
0428         return;
0429 
0430     check_if_tm_restore_required(tsk);
0431 
0432     usermsr = tsk->thread.regs->msr;
0433 
0434     if ((usermsr & msr_all_available) == 0)
0435         return;
0436 
0437     msr_check_and_set(msr_all_available);
0438 
0439     WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
0440 
0441     if (usermsr & MSR_FP)
0442         __giveup_fpu(tsk);
0443     if (usermsr & MSR_VEC)
0444         __giveup_altivec(tsk);
0445     if (usermsr & MSR_SPE)
0446         __giveup_spe(tsk);
0447 
0448     msr_check_and_clear(msr_all_available);
0449 }
0450 EXPORT_SYMBOL(giveup_all);
0451 
0452 #ifdef CONFIG_PPC_BOOK3S_64
0453 #ifdef CONFIG_PPC_FPU
0454 static bool should_restore_fp(void)
0455 {
0456     if (current->thread.load_fp) {
0457         current->thread.load_fp++;
0458         return true;
0459     }
0460     return false;
0461 }
0462 
0463 static void do_restore_fp(void)
0464 {
0465     load_fp_state(&current->thread.fp_state);
0466 }
0467 #else
0468 static bool should_restore_fp(void) { return false; }
0469 static void do_restore_fp(void) { }
0470 #endif /* CONFIG_PPC_FPU */
0471 
0472 #ifdef CONFIG_ALTIVEC
0473 static bool should_restore_altivec(void)
0474 {
0475     if (cpu_has_feature(CPU_FTR_ALTIVEC) && (current->thread.load_vec)) {
0476         current->thread.load_vec++;
0477         return true;
0478     }
0479     return false;
0480 }
0481 
0482 static void do_restore_altivec(void)
0483 {
0484     load_vr_state(&current->thread.vr_state);
0485     current->thread.used_vr = 1;
0486 }
0487 #else
0488 static bool should_restore_altivec(void) { return false; }
0489 static void do_restore_altivec(void) { }
0490 #endif /* CONFIG_ALTIVEC */
0491 
0492 static bool should_restore_vsx(void)
0493 {
0494     if (cpu_has_feature(CPU_FTR_VSX))
0495         return true;
0496     return false;
0497 }
0498 #ifdef CONFIG_VSX
0499 static void do_restore_vsx(void)
0500 {
0501     current->thread.used_vsr = 1;
0502 }
0503 #else
0504 static void do_restore_vsx(void) { }
0505 #endif /* CONFIG_VSX */
0506 
0507 /*
0508  * The exception exit path calls restore_math() with interrupts hard disabled
0509  * but the soft irq state not "reconciled". ftrace code that calls
0510  * local_irq_save/restore causes warnings.
0511  *
0512  * Rather than complicate the exit path, just don't trace restore_math. This
0513  * could be done by having ftrace entry code check for this un-reconciled
0514  * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
0515  * temporarily fix it up for the duration of the ftrace call.
0516  */
0517 void notrace restore_math(struct pt_regs *regs)
0518 {
0519     unsigned long msr;
0520     unsigned long new_msr = 0;
0521 
0522     msr = regs->msr;
0523 
0524     /*
0525      * new_msr tracks the facilities that are to be restored. Only reload
0526      * if the bit is not set in the user MSR (if it is set, the registers
0527      * are live for the user thread).
0528      */
0529     if ((!(msr & MSR_FP)) && should_restore_fp())
0530         new_msr |= MSR_FP;
0531 
0532     if ((!(msr & MSR_VEC)) && should_restore_altivec())
0533         new_msr |= MSR_VEC;
0534 
0535     if ((!(msr & MSR_VSX)) && should_restore_vsx()) {
0536         if (((msr | new_msr) & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC))
0537             new_msr |= MSR_VSX;
0538     }
0539 
0540     if (new_msr) {
0541         unsigned long fpexc_mode = 0;
0542 
0543         msr_check_and_set(new_msr);
0544 
0545         if (new_msr & MSR_FP) {
0546             do_restore_fp();
0547 
0548             // This also covers VSX, because VSX implies FP
0549             fpexc_mode = current->thread.fpexc_mode;
0550         }
0551 
0552         if (new_msr & MSR_VEC)
0553             do_restore_altivec();
0554 
0555         if (new_msr & MSR_VSX)
0556             do_restore_vsx();
0557 
0558         msr_check_and_clear(new_msr);
0559 
0560         regs_set_return_msr(regs, regs->msr | new_msr | fpexc_mode);
0561     }
0562 }
0563 #endif /* CONFIG_PPC_BOOK3S_64 */
0564 
0565 static void save_all(struct task_struct *tsk)
0566 {
0567     unsigned long usermsr;
0568 
0569     if (!tsk->thread.regs)
0570         return;
0571 
0572     usermsr = tsk->thread.regs->msr;
0573 
0574     if ((usermsr & msr_all_available) == 0)
0575         return;
0576 
0577     msr_check_and_set(msr_all_available);
0578 
0579     WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
0580 
0581     if (usermsr & MSR_FP)
0582         save_fpu(tsk);
0583 
0584     if (usermsr & MSR_VEC)
0585         save_altivec(tsk);
0586 
0587     if (usermsr & MSR_SPE)
0588         __giveup_spe(tsk);
0589 
0590     msr_check_and_clear(msr_all_available);
0591 }
0592 
0593 void flush_all_to_thread(struct task_struct *tsk)
0594 {
0595     if (tsk->thread.regs) {
0596         preempt_disable();
0597         BUG_ON(tsk != current);
0598 #ifdef CONFIG_SPE
0599         if (tsk->thread.regs->msr & MSR_SPE)
0600             tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
0601 #endif
0602         save_all(tsk);
0603 
0604         preempt_enable();
0605     }
0606 }
0607 EXPORT_SYMBOL(flush_all_to_thread);
0608 
0609 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
0610 void do_send_trap(struct pt_regs *regs, unsigned long address,
0611           unsigned long error_code, int breakpt)
0612 {
0613     current->thread.trap_nr = TRAP_HWBKPT;
0614     if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
0615             11, SIGSEGV) == NOTIFY_STOP)
0616         return;
0617 
0618     /* Deliver the signal to userspace */
0619     force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
0620                     (void __user *)address);
0621 }
0622 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
0623 
0624 static void do_break_handler(struct pt_regs *regs)
0625 {
0626     struct arch_hw_breakpoint null_brk = {0};
0627     struct arch_hw_breakpoint *info;
0628     ppc_inst_t instr = ppc_inst(0);
0629     int type = 0;
0630     int size = 0;
0631     unsigned long ea;
0632     int i;
0633 
0634     /*
0635      * If underneath hw supports only one watchpoint, we know it
0636      * caused exception. 8xx also falls into this category.
0637      */
0638     if (nr_wp_slots() == 1) {
0639         __set_breakpoint(0, &null_brk);
0640         current->thread.hw_brk[0] = null_brk;
0641         current->thread.hw_brk[0].flags |= HW_BRK_FLAG_DISABLED;
0642         return;
0643     }
0644 
0645     /* Otherwise find out which DAWR caused exception and disable it. */
0646     wp_get_instr_detail(regs, &instr, &type, &size, &ea);
0647 
0648     for (i = 0; i < nr_wp_slots(); i++) {
0649         info = &current->thread.hw_brk[i];
0650         if (!info->address)
0651             continue;
0652 
0653         if (wp_check_constraints(regs, instr, ea, type, size, info)) {
0654             __set_breakpoint(i, &null_brk);
0655             current->thread.hw_brk[i] = null_brk;
0656             current->thread.hw_brk[i].flags |= HW_BRK_FLAG_DISABLED;
0657         }
0658     }
0659 }
0660 
0661 DEFINE_INTERRUPT_HANDLER(do_break)
0662 {
0663     current->thread.trap_nr = TRAP_HWBKPT;
0664     if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, regs->dsisr,
0665             11, SIGSEGV) == NOTIFY_STOP)
0666         return;
0667 
0668     if (debugger_break_match(regs))
0669         return;
0670 
0671     /*
0672      * We reach here only when watchpoint exception is generated by ptrace
0673      * event (or hw is buggy!). Now if CONFIG_HAVE_HW_BREAKPOINT is set,
0674      * watchpoint is already handled by hw_breakpoint_handler() so we don't
0675      * have to do anything. But when CONFIG_HAVE_HW_BREAKPOINT is not set,
0676      * we need to manually handle the watchpoint here.
0677      */
0678     if (!IS_ENABLED(CONFIG_HAVE_HW_BREAKPOINT))
0679         do_break_handler(regs);
0680 
0681     /* Deliver the signal to userspace */
0682     force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)regs->dar);
0683 }
0684 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
0685 
0686 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
0687 
0688 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
0689 /*
0690  * Set the debug registers back to their default "safe" values.
0691  */
0692 static void set_debug_reg_defaults(struct thread_struct *thread)
0693 {
0694     thread->debug.iac1 = thread->debug.iac2 = 0;
0695 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
0696     thread->debug.iac3 = thread->debug.iac4 = 0;
0697 #endif
0698     thread->debug.dac1 = thread->debug.dac2 = 0;
0699 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
0700     thread->debug.dvc1 = thread->debug.dvc2 = 0;
0701 #endif
0702     thread->debug.dbcr0 = 0;
0703 #ifdef CONFIG_BOOKE
0704     /*
0705      * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
0706      */
0707     thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
0708             DBCR1_IAC3US | DBCR1_IAC4US;
0709     /*
0710      * Force Data Address Compare User/Supervisor bits to be User-only
0711      * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
0712      */
0713     thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
0714 #else
0715     thread->debug.dbcr1 = 0;
0716 #endif
0717 }
0718 
0719 static void prime_debug_regs(struct debug_reg *debug)
0720 {
0721     /*
0722      * We could have inherited MSR_DE from userspace, since
0723      * it doesn't get cleared on exception entry.  Make sure
0724      * MSR_DE is clear before we enable any debug events.
0725      */
0726     mtmsr(mfmsr() & ~MSR_DE);
0727 
0728     mtspr(SPRN_IAC1, debug->iac1);
0729     mtspr(SPRN_IAC2, debug->iac2);
0730 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
0731     mtspr(SPRN_IAC3, debug->iac3);
0732     mtspr(SPRN_IAC4, debug->iac4);
0733 #endif
0734     mtspr(SPRN_DAC1, debug->dac1);
0735     mtspr(SPRN_DAC2, debug->dac2);
0736 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
0737     mtspr(SPRN_DVC1, debug->dvc1);
0738     mtspr(SPRN_DVC2, debug->dvc2);
0739 #endif
0740     mtspr(SPRN_DBCR0, debug->dbcr0);
0741     mtspr(SPRN_DBCR1, debug->dbcr1);
0742 #ifdef CONFIG_BOOKE
0743     mtspr(SPRN_DBCR2, debug->dbcr2);
0744 #endif
0745 }
0746 /*
0747  * Unless neither the old or new thread are making use of the
0748  * debug registers, set the debug registers from the values
0749  * stored in the new thread.
0750  */
0751 void switch_booke_debug_regs(struct debug_reg *new_debug)
0752 {
0753     if ((current->thread.debug.dbcr0 & DBCR0_IDM)
0754         || (new_debug->dbcr0 & DBCR0_IDM))
0755             prime_debug_regs(new_debug);
0756 }
0757 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
0758 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
0759 #ifndef CONFIG_HAVE_HW_BREAKPOINT
0760 static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
0761 {
0762     preempt_disable();
0763     __set_breakpoint(i, brk);
0764     preempt_enable();
0765 }
0766 
0767 static void set_debug_reg_defaults(struct thread_struct *thread)
0768 {
0769     int i;
0770     struct arch_hw_breakpoint null_brk = {0};
0771 
0772     for (i = 0; i < nr_wp_slots(); i++) {
0773         thread->hw_brk[i] = null_brk;
0774         if (ppc_breakpoint_available())
0775             set_breakpoint(i, &thread->hw_brk[i]);
0776     }
0777 }
0778 
0779 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
0780                 struct arch_hw_breakpoint *b)
0781 {
0782     if (a->address != b->address)
0783         return false;
0784     if (a->type != b->type)
0785         return false;
0786     if (a->len != b->len)
0787         return false;
0788     /* no need to check hw_len. it's calculated from address and len */
0789     return true;
0790 }
0791 
0792 static void switch_hw_breakpoint(struct task_struct *new)
0793 {
0794     int i;
0795 
0796     for (i = 0; i < nr_wp_slots(); i++) {
0797         if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
0798                     &new->thread.hw_brk[i])))
0799             continue;
0800 
0801         __set_breakpoint(i, &new->thread.hw_brk[i]);
0802     }
0803 }
0804 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
0805 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
0806 
0807 static inline int set_dabr(struct arch_hw_breakpoint *brk)
0808 {
0809     unsigned long dabr, dabrx;
0810 
0811     dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
0812     dabrx = ((brk->type >> 3) & 0x7);
0813 
0814     if (ppc_md.set_dabr)
0815         return ppc_md.set_dabr(dabr, dabrx);
0816 
0817     if (IS_ENABLED(CONFIG_PPC_ADV_DEBUG_REGS)) {
0818         mtspr(SPRN_DAC1, dabr);
0819         if (IS_ENABLED(CONFIG_PPC_47x))
0820             isync();
0821         return 0;
0822     } else if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
0823         mtspr(SPRN_DABR, dabr);
0824         if (cpu_has_feature(CPU_FTR_DABRX))
0825             mtspr(SPRN_DABRX, dabrx);
0826         return 0;
0827     } else {
0828         return -EINVAL;
0829     }
0830 }
0831 
0832 static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
0833 {
0834     unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
0835                    LCTRL1_CRWF_RW;
0836     unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
0837     unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
0838     unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
0839 
0840     if (start_addr == 0)
0841         lctrl2 |= LCTRL2_LW0LA_F;
0842     else if (end_addr == 0)
0843         lctrl2 |= LCTRL2_LW0LA_E;
0844     else
0845         lctrl2 |= LCTRL2_LW0LA_EandF;
0846 
0847     mtspr(SPRN_LCTRL2, 0);
0848 
0849     if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
0850         return 0;
0851 
0852     if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
0853         lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
0854     if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
0855         lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
0856 
0857     mtspr(SPRN_CMPE, start_addr - 1);
0858     mtspr(SPRN_CMPF, end_addr);
0859     mtspr(SPRN_LCTRL1, lctrl1);
0860     mtspr(SPRN_LCTRL2, lctrl2);
0861 
0862     return 0;
0863 }
0864 
0865 void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
0866 {
0867     memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
0868 
0869     if (dawr_enabled())
0870         // Power8 or later
0871         set_dawr(nr, brk);
0872     else if (IS_ENABLED(CONFIG_PPC_8xx))
0873         set_breakpoint_8xx(brk);
0874     else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
0875         // Power7 or earlier
0876         set_dabr(brk);
0877     else
0878         // Shouldn't happen due to higher level checks
0879         WARN_ON_ONCE(1);
0880 }
0881 
0882 /* Check if we have DAWR or DABR hardware */
0883 bool ppc_breakpoint_available(void)
0884 {
0885     if (dawr_enabled())
0886         return true; /* POWER8 DAWR or POWER9 forced DAWR */
0887     if (cpu_has_feature(CPU_FTR_ARCH_207S))
0888         return false; /* POWER9 with DAWR disabled */
0889     /* DABR: Everything but POWER8 and POWER9 */
0890     return true;
0891 }
0892 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
0893 
0894 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
0895 
0896 static inline bool tm_enabled(struct task_struct *tsk)
0897 {
0898     return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
0899 }
0900 
0901 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
0902 {
0903     /*
0904      * Use the current MSR TM suspended bit to track if we have
0905      * checkpointed state outstanding.
0906      * On signal delivery, we'd normally reclaim the checkpointed
0907      * state to obtain stack pointer (see:get_tm_stackpointer()).
0908      * This will then directly return to userspace without going
0909      * through __switch_to(). However, if the stack frame is bad,
0910      * we need to exit this thread which calls __switch_to() which
0911      * will again attempt to reclaim the already saved tm state.
0912      * Hence we need to check that we've not already reclaimed
0913      * this state.
0914      * We do this using the current MSR, rather tracking it in
0915      * some specific thread_struct bit, as it has the additional
0916      * benefit of checking for a potential TM bad thing exception.
0917      */
0918     if (!MSR_TM_SUSPENDED(mfmsr()))
0919         return;
0920 
0921     giveup_all(container_of(thr, struct task_struct, thread));
0922 
0923     tm_reclaim(thr, cause);
0924 
0925     /*
0926      * If we are in a transaction and FP is off then we can't have
0927      * used FP inside that transaction. Hence the checkpointed
0928      * state is the same as the live state. We need to copy the
0929      * live state to the checkpointed state so that when the
0930      * transaction is restored, the checkpointed state is correct
0931      * and the aborted transaction sees the correct state. We use
0932      * ckpt_regs.msr here as that's what tm_reclaim will use to
0933      * determine if it's going to write the checkpointed state or
0934      * not. So either this will write the checkpointed registers,
0935      * or reclaim will. Similarly for VMX.
0936      */
0937     if ((thr->ckpt_regs.msr & MSR_FP) == 0)
0938         memcpy(&thr->ckfp_state, &thr->fp_state,
0939                sizeof(struct thread_fp_state));
0940     if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
0941         memcpy(&thr->ckvr_state, &thr->vr_state,
0942                sizeof(struct thread_vr_state));
0943 }
0944 
0945 void tm_reclaim_current(uint8_t cause)
0946 {
0947     tm_enable();
0948     tm_reclaim_thread(&current->thread, cause);
0949 }
0950 
0951 static inline void tm_reclaim_task(struct task_struct *tsk)
0952 {
0953     /* We have to work out if we're switching from/to a task that's in the
0954      * middle of a transaction.
0955      *
0956      * In switching we need to maintain a 2nd register state as
0957      * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
0958      * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
0959      * ckvr_state
0960      *
0961      * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
0962      */
0963     struct thread_struct *thr = &tsk->thread;
0964 
0965     if (!thr->regs)
0966         return;
0967 
0968     if (!MSR_TM_ACTIVE(thr->regs->msr))
0969         goto out_and_saveregs;
0970 
0971     WARN_ON(tm_suspend_disabled);
0972 
0973     TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
0974          "ccr=%lx, msr=%lx, trap=%lx)\n",
0975          tsk->pid, thr->regs->nip,
0976          thr->regs->ccr, thr->regs->msr,
0977          thr->regs->trap);
0978 
0979     tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
0980 
0981     TM_DEBUG("--- tm_reclaim on pid %d complete\n",
0982          tsk->pid);
0983 
0984 out_and_saveregs:
0985     /* Always save the regs here, even if a transaction's not active.
0986      * This context-switches a thread's TM info SPRs.  We do it here to
0987      * be consistent with the restore path (in recheckpoint) which
0988      * cannot happen later in _switch().
0989      */
0990     tm_save_sprs(thr);
0991 }
0992 
0993 extern void __tm_recheckpoint(struct thread_struct *thread);
0994 
0995 void tm_recheckpoint(struct thread_struct *thread)
0996 {
0997     unsigned long flags;
0998 
0999     if (!(thread->regs->msr & MSR_TM))
1000         return;
1001 
1002     /* We really can't be interrupted here as the TEXASR registers can't
1003      * change and later in the trecheckpoint code, we have a userspace R1.
1004      * So let's hard disable over this region.
1005      */
1006     local_irq_save(flags);
1007     hard_irq_disable();
1008 
1009     /* The TM SPRs are restored here, so that TEXASR.FS can be set
1010      * before the trecheckpoint and no explosion occurs.
1011      */
1012     tm_restore_sprs(thread);
1013 
1014     __tm_recheckpoint(thread);
1015 
1016     local_irq_restore(flags);
1017 }
1018 
1019 static inline void tm_recheckpoint_new_task(struct task_struct *new)
1020 {
1021     if (!cpu_has_feature(CPU_FTR_TM))
1022         return;
1023 
1024     /* Recheckpoint the registers of the thread we're about to switch to.
1025      *
1026      * If the task was using FP, we non-lazily reload both the original and
1027      * the speculative FP register states.  This is because the kernel
1028      * doesn't see if/when a TM rollback occurs, so if we take an FP
1029      * unavailable later, we are unable to determine which set of FP regs
1030      * need to be restored.
1031      */
1032     if (!tm_enabled(new))
1033         return;
1034 
1035     if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
1036         tm_restore_sprs(&new->thread);
1037         return;
1038     }
1039     /* Recheckpoint to restore original checkpointed register state. */
1040     TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1041          new->pid, new->thread.regs->msr);
1042 
1043     tm_recheckpoint(&new->thread);
1044 
1045     /*
1046      * The checkpointed state has been restored but the live state has
1047      * not, ensure all the math functionality is turned off to trigger
1048      * restore_math() to reload.
1049      */
1050     new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1051 
1052     TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1053          "(kernel msr 0x%lx)\n",
1054          new->pid, mfmsr());
1055 }
1056 
1057 static inline void __switch_to_tm(struct task_struct *prev,
1058         struct task_struct *new)
1059 {
1060     if (cpu_has_feature(CPU_FTR_TM)) {
1061         if (tm_enabled(prev) || tm_enabled(new))
1062             tm_enable();
1063 
1064         if (tm_enabled(prev)) {
1065             prev->thread.load_tm++;
1066             tm_reclaim_task(prev);
1067             if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1068                 prev->thread.regs->msr &= ~MSR_TM;
1069         }
1070 
1071         tm_recheckpoint_new_task(new);
1072     }
1073 }
1074 
1075 /*
1076  * This is called if we are on the way out to userspace and the
1077  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1078  * FP and/or vector state and does so if necessary.
1079  * If userspace is inside a transaction (whether active or
1080  * suspended) and FP/VMX/VSX instructions have ever been enabled
1081  * inside that transaction, then we have to keep them enabled
1082  * and keep the FP/VMX/VSX state loaded while ever the transaction
1083  * continues.  The reason is that if we didn't, and subsequently
1084  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1085  * we don't know whether it's the same transaction, and thus we
1086  * don't know which of the checkpointed state and the transactional
1087  * state to use.
1088  */
1089 void restore_tm_state(struct pt_regs *regs)
1090 {
1091     unsigned long msr_diff;
1092 
1093     /*
1094      * This is the only moment we should clear TIF_RESTORE_TM as
1095      * it is here that ckpt_regs.msr and pt_regs.msr become the same
1096      * again, anything else could lead to an incorrect ckpt_msr being
1097      * saved and therefore incorrect signal contexts.
1098      */
1099     clear_thread_flag(TIF_RESTORE_TM);
1100     if (!MSR_TM_ACTIVE(regs->msr))
1101         return;
1102 
1103     msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1104     msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1105 
1106     /* Ensure that restore_math() will restore */
1107     if (msr_diff & MSR_FP)
1108         current->thread.load_fp = 1;
1109 #ifdef CONFIG_ALTIVEC
1110     if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1111         current->thread.load_vec = 1;
1112 #endif
1113     restore_math(regs);
1114 
1115     regs_set_return_msr(regs, regs->msr | msr_diff);
1116 }
1117 
1118 #else /* !CONFIG_PPC_TRANSACTIONAL_MEM */
1119 #define tm_recheckpoint_new_task(new)
1120 #define __switch_to_tm(prev, new)
1121 void tm_reclaim_current(uint8_t cause) {}
1122 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1123 
1124 static inline void save_sprs(struct thread_struct *t)
1125 {
1126 #ifdef CONFIG_ALTIVEC
1127     if (cpu_has_feature(CPU_FTR_ALTIVEC))
1128         t->vrsave = mfspr(SPRN_VRSAVE);
1129 #endif
1130 #ifdef CONFIG_SPE
1131     if (cpu_has_feature(CPU_FTR_SPE))
1132         t->spefscr = mfspr(SPRN_SPEFSCR);
1133 #endif
1134 #ifdef CONFIG_PPC_BOOK3S_64
1135     if (cpu_has_feature(CPU_FTR_DSCR))
1136         t->dscr = mfspr(SPRN_DSCR);
1137 
1138     if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1139         t->bescr = mfspr(SPRN_BESCR);
1140         t->ebbhr = mfspr(SPRN_EBBHR);
1141         t->ebbrr = mfspr(SPRN_EBBRR);
1142 
1143         t->fscr = mfspr(SPRN_FSCR);
1144 
1145         /*
1146          * Note that the TAR is not available for use in the kernel.
1147          * (To provide this, the TAR should be backed up/restored on
1148          * exception entry/exit instead, and be in pt_regs.  FIXME,
1149          * this should be in pt_regs anyway (for debug).)
1150          */
1151         t->tar = mfspr(SPRN_TAR);
1152     }
1153 #endif
1154 }
1155 
1156 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1157 void kvmppc_save_user_regs(void)
1158 {
1159     unsigned long usermsr;
1160 
1161     if (!current->thread.regs)
1162         return;
1163 
1164     usermsr = current->thread.regs->msr;
1165 
1166     if (usermsr & MSR_FP)
1167         save_fpu(current);
1168 
1169     if (usermsr & MSR_VEC)
1170         save_altivec(current);
1171 
1172 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1173     if (usermsr & MSR_TM) {
1174         current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
1175         current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
1176         current->thread.tm_texasr = mfspr(SPRN_TEXASR);
1177         current->thread.regs->msr &= ~MSR_TM;
1178     }
1179 #endif
1180 }
1181 EXPORT_SYMBOL_GPL(kvmppc_save_user_regs);
1182 
1183 void kvmppc_save_current_sprs(void)
1184 {
1185     save_sprs(&current->thread);
1186 }
1187 EXPORT_SYMBOL_GPL(kvmppc_save_current_sprs);
1188 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1189 
1190 static inline void restore_sprs(struct thread_struct *old_thread,
1191                 struct thread_struct *new_thread)
1192 {
1193 #ifdef CONFIG_ALTIVEC
1194     if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1195         old_thread->vrsave != new_thread->vrsave)
1196         mtspr(SPRN_VRSAVE, new_thread->vrsave);
1197 #endif
1198 #ifdef CONFIG_SPE
1199     if (cpu_has_feature(CPU_FTR_SPE) &&
1200         old_thread->spefscr != new_thread->spefscr)
1201         mtspr(SPRN_SPEFSCR, new_thread->spefscr);
1202 #endif
1203 #ifdef CONFIG_PPC_BOOK3S_64
1204     if (cpu_has_feature(CPU_FTR_DSCR)) {
1205         u64 dscr = get_paca()->dscr_default;
1206         if (new_thread->dscr_inherit)
1207             dscr = new_thread->dscr;
1208 
1209         if (old_thread->dscr != dscr)
1210             mtspr(SPRN_DSCR, dscr);
1211     }
1212 
1213     if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1214         if (old_thread->bescr != new_thread->bescr)
1215             mtspr(SPRN_BESCR, new_thread->bescr);
1216         if (old_thread->ebbhr != new_thread->ebbhr)
1217             mtspr(SPRN_EBBHR, new_thread->ebbhr);
1218         if (old_thread->ebbrr != new_thread->ebbrr)
1219             mtspr(SPRN_EBBRR, new_thread->ebbrr);
1220 
1221         if (old_thread->fscr != new_thread->fscr)
1222             mtspr(SPRN_FSCR, new_thread->fscr);
1223 
1224         if (old_thread->tar != new_thread->tar)
1225             mtspr(SPRN_TAR, new_thread->tar);
1226     }
1227 
1228     if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1229         old_thread->tidr != new_thread->tidr)
1230         mtspr(SPRN_TIDR, new_thread->tidr);
1231 #endif
1232 
1233 }
1234 
1235 struct task_struct *__switch_to(struct task_struct *prev,
1236     struct task_struct *new)
1237 {
1238     struct thread_struct *new_thread, *old_thread;
1239     struct task_struct *last;
1240 #ifdef CONFIG_PPC_64S_HASH_MMU
1241     struct ppc64_tlb_batch *batch;
1242 #endif
1243 
1244     new_thread = &new->thread;
1245     old_thread = &current->thread;
1246 
1247     WARN_ON(!irqs_disabled());
1248 
1249 #ifdef CONFIG_PPC_64S_HASH_MMU
1250     batch = this_cpu_ptr(&ppc64_tlb_batch);
1251     if (batch->active) {
1252         current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1253         if (batch->index)
1254             __flush_tlb_pending(batch);
1255         batch->active = 0;
1256     }
1257 
1258     /*
1259      * On POWER9 the copy-paste buffer can only paste into
1260      * foreign real addresses, so unprivileged processes can not
1261      * see the data or use it in any way unless they have
1262      * foreign real mappings. If the new process has the foreign
1263      * real address mappings, we must issue a cp_abort to clear
1264      * any state and prevent snooping, corruption or a covert
1265      * channel. ISA v3.1 supports paste into local memory.
1266      */
1267     if (new->mm && (cpu_has_feature(CPU_FTR_ARCH_31) ||
1268             atomic_read(&new->mm->context.vas_windows)))
1269         asm volatile(PPC_CP_ABORT);
1270 #endif /* CONFIG_PPC_BOOK3S_64 */
1271 
1272 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1273     switch_booke_debug_regs(&new->thread.debug);
1274 #else
1275 /*
1276  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1277  * schedule DABR
1278  */
1279 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1280     switch_hw_breakpoint(new);
1281 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1282 #endif
1283 
1284     /*
1285      * We need to save SPRs before treclaim/trecheckpoint as these will
1286      * change a number of them.
1287      */
1288     save_sprs(&prev->thread);
1289 
1290     /* Save FPU, Altivec, VSX and SPE state */
1291     giveup_all(prev);
1292 
1293     __switch_to_tm(prev, new);
1294 
1295     if (!radix_enabled()) {
1296         /*
1297          * We can't take a PMU exception inside _switch() since there
1298          * is a window where the kernel stack SLB and the kernel stack
1299          * are out of sync. Hard disable here.
1300          */
1301         hard_irq_disable();
1302     }
1303 
1304     /*
1305      * Call restore_sprs() and set_return_regs_changed() before calling
1306      * _switch(). If we move it after _switch() then we miss out on calling
1307      * it for new tasks. The reason for this is we manually create a stack
1308      * frame for new tasks that directly returns through ret_from_fork() or
1309      * ret_from_kernel_thread(). See copy_thread() for details.
1310      */
1311     restore_sprs(old_thread, new_thread);
1312 
1313     set_return_regs_changed(); /* _switch changes stack (and regs) */
1314 
1315     if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
1316         kuap_assert_locked();
1317 
1318     last = _switch(old_thread, new_thread);
1319 
1320     /*
1321      * Nothing after _switch will be run for newly created tasks,
1322      * because they switch directly to ret_from_fork/ret_from_kernel_thread
1323      * etc. Code added here should have a comment explaining why that is
1324      * okay.
1325      */
1326 
1327 #ifdef CONFIG_PPC_BOOK3S_64
1328 #ifdef CONFIG_PPC_64S_HASH_MMU
1329     /*
1330      * This applies to a process that was context switched while inside
1331      * arch_enter_lazy_mmu_mode(), to re-activate the batch that was
1332      * deactivated above, before _switch(). This will never be the case
1333      * for new tasks.
1334      */
1335     if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1336         current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1337         batch = this_cpu_ptr(&ppc64_tlb_batch);
1338         batch->active = 1;
1339     }
1340 #endif
1341 
1342     /*
1343      * Math facilities are masked out of the child MSR in copy_thread.
1344      * A new task does not need to restore_math because it will
1345      * demand fault them.
1346      */
1347     if (current->thread.regs)
1348         restore_math(current->thread.regs);
1349 #endif /* CONFIG_PPC_BOOK3S_64 */
1350 
1351     return last;
1352 }
1353 
1354 #define NR_INSN_TO_PRINT    16
1355 
1356 static void show_instructions(struct pt_regs *regs)
1357 {
1358     int i;
1359     unsigned long nip = regs->nip;
1360     unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1361 
1362     printk("Instruction dump:");
1363 
1364     /*
1365      * If we were executing with the MMU off for instructions, adjust pc
1366      * rather than printing XXXXXXXX.
1367      */
1368     if (!IS_ENABLED(CONFIG_BOOKE) && !(regs->msr & MSR_IR)) {
1369         pc = (unsigned long)phys_to_virt(pc);
1370         nip = (unsigned long)phys_to_virt(regs->nip);
1371     }
1372 
1373     for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1374         int instr;
1375 
1376         if (!(i % 8))
1377             pr_cont("\n");
1378 
1379         if (!__kernel_text_address(pc) ||
1380             get_kernel_nofault(instr, (const void *)pc)) {
1381             pr_cont("XXXXXXXX ");
1382         } else {
1383             if (nip == pc)
1384                 pr_cont("<%08x> ", instr);
1385             else
1386                 pr_cont("%08x ", instr);
1387         }
1388 
1389         pc += sizeof(int);
1390     }
1391 
1392     pr_cont("\n");
1393 }
1394 
1395 void show_user_instructions(struct pt_regs *regs)
1396 {
1397     unsigned long pc;
1398     int n = NR_INSN_TO_PRINT;
1399     struct seq_buf s;
1400     char buf[96]; /* enough for 8 times 9 + 2 chars */
1401 
1402     pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1403 
1404     seq_buf_init(&s, buf, sizeof(buf));
1405 
1406     while (n) {
1407         int i;
1408 
1409         seq_buf_clear(&s);
1410 
1411         for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1412             int instr;
1413 
1414             if (copy_from_user_nofault(&instr, (void __user *)pc,
1415                     sizeof(instr))) {
1416                 seq_buf_printf(&s, "XXXXXXXX ");
1417                 continue;
1418             }
1419             seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1420         }
1421 
1422         if (!seq_buf_has_overflowed(&s))
1423             pr_info("%s[%d]: code: %s\n", current->comm,
1424                 current->pid, s.buffer);
1425     }
1426 }
1427 
1428 struct regbit {
1429     unsigned long bit;
1430     const char *name;
1431 };
1432 
1433 static struct regbit msr_bits[] = {
1434 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1435     {MSR_SF,    "SF"},
1436     {MSR_HV,    "HV"},
1437 #endif
1438     {MSR_VEC,   "VEC"},
1439     {MSR_VSX,   "VSX"},
1440 #ifdef CONFIG_BOOKE
1441     {MSR_CE,    "CE"},
1442 #endif
1443     {MSR_EE,    "EE"},
1444     {MSR_PR,    "PR"},
1445     {MSR_FP,    "FP"},
1446     {MSR_ME,    "ME"},
1447 #ifdef CONFIG_BOOKE
1448     {MSR_DE,    "DE"},
1449 #else
1450     {MSR_SE,    "SE"},
1451     {MSR_BE,    "BE"},
1452 #endif
1453     {MSR_IR,    "IR"},
1454     {MSR_DR,    "DR"},
1455     {MSR_PMM,   "PMM"},
1456 #ifndef CONFIG_BOOKE
1457     {MSR_RI,    "RI"},
1458     {MSR_LE,    "LE"},
1459 #endif
1460     {0,     NULL}
1461 };
1462 
1463 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1464 {
1465     const char *s = "";
1466 
1467     for (; bits->bit; ++bits)
1468         if (val & bits->bit) {
1469             pr_cont("%s%s", s, bits->name);
1470             s = sep;
1471         }
1472 }
1473 
1474 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1475 static struct regbit msr_tm_bits[] = {
1476     {MSR_TS_T,  "T"},
1477     {MSR_TS_S,  "S"},
1478     {MSR_TM,    "E"},
1479     {0,     NULL}
1480 };
1481 
1482 static void print_tm_bits(unsigned long val)
1483 {
1484 /*
1485  * This only prints something if at least one of the TM bit is set.
1486  * Inside the TM[], the output means:
1487  *   E: Enabled     (bit 32)
1488  *   S: Suspended   (bit 33)
1489  *   T: Transactional   (bit 34)
1490  */
1491     if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1492         pr_cont(",TM[");
1493         print_bits(val, msr_tm_bits, "");
1494         pr_cont("]");
1495     }
1496 }
1497 #else
1498 static void print_tm_bits(unsigned long val) {}
1499 #endif
1500 
1501 static void print_msr_bits(unsigned long val)
1502 {
1503     pr_cont("<");
1504     print_bits(val, msr_bits, ",");
1505     print_tm_bits(val);
1506     pr_cont(">");
1507 }
1508 
1509 #ifdef CONFIG_PPC64
1510 #define REG     "%016lx"
1511 #define REGS_PER_LINE   4
1512 #else
1513 #define REG     "%08lx"
1514 #define REGS_PER_LINE   8
1515 #endif
1516 
1517 static void __show_regs(struct pt_regs *regs)
1518 {
1519     int i, trap;
1520 
1521     printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1522            regs->nip, regs->link, regs->ctr);
1523     printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1524            regs, regs->trap, print_tainted(), init_utsname()->release);
1525     printk("MSR:  "REG" ", regs->msr);
1526     print_msr_bits(regs->msr);
1527     pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1528     trap = TRAP(regs);
1529     if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
1530         pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1531     if (trap == INTERRUPT_MACHINE_CHECK ||
1532         trap == INTERRUPT_DATA_STORAGE ||
1533         trap == INTERRUPT_ALIGNMENT) {
1534         if (IS_ENABLED(CONFIG_4xx) || IS_ENABLED(CONFIG_BOOKE))
1535             pr_cont("DEAR: "REG" ESR: "REG" ", regs->dear, regs->esr);
1536         else
1537             pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1538     }
1539 
1540 #ifdef CONFIG_PPC64
1541     pr_cont("IRQMASK: %lx ", regs->softe);
1542 #endif
1543 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1544     if (MSR_TM_ACTIVE(regs->msr))
1545         pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1546 #endif
1547 
1548     for (i = 0;  i < 32;  i++) {
1549         if ((i % REGS_PER_LINE) == 0)
1550             pr_cont("\nGPR%02d: ", i);
1551         pr_cont(REG " ", regs->gpr[i]);
1552     }
1553     pr_cont("\n");
1554     /*
1555      * Lookup NIP late so we have the best change of getting the
1556      * above info out without failing
1557      */
1558     if (IS_ENABLED(CONFIG_KALLSYMS)) {
1559         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1560         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1561     }
1562 }
1563 
1564 void show_regs(struct pt_regs *regs)
1565 {
1566     show_regs_print_info(KERN_DEFAULT);
1567     __show_regs(regs);
1568     show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
1569     if (!user_mode(regs))
1570         show_instructions(regs);
1571 }
1572 
1573 void flush_thread(void)
1574 {
1575 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1576     flush_ptrace_hw_breakpoint(current);
1577 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1578     set_debug_reg_defaults(&current->thread);
1579 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1580 }
1581 
1582 void arch_setup_new_exec(void)
1583 {
1584 
1585 #ifdef CONFIG_PPC_BOOK3S_64
1586     if (!radix_enabled())
1587         hash__setup_new_exec();
1588 #endif
1589     /*
1590      * If we exec out of a kernel thread then thread.regs will not be
1591      * set.  Do it now.
1592      */
1593     if (!current->thread.regs) {
1594         struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1595         current->thread.regs = regs - 1;
1596     }
1597 
1598 #ifdef CONFIG_PPC_MEM_KEYS
1599     current->thread.regs->amr  = default_amr;
1600     current->thread.regs->iamr  = default_iamr;
1601 #endif
1602 }
1603 
1604 #ifdef CONFIG_PPC64
1605 /**
1606  * Assign a TIDR (thread ID) for task @t and set it in the thread
1607  * structure. For now, we only support setting TIDR for 'current' task.
1608  *
1609  * Since the TID value is a truncated form of it PID, it is possible
1610  * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1611  * that 2 threads share the same TID and are waiting, one of the following
1612  * cases will happen:
1613  *
1614  * 1. The correct thread is running, the wrong thread is not
1615  * In this situation, the correct thread is woken and proceeds to pass it's
1616  * condition check.
1617  *
1618  * 2. Neither threads are running
1619  * In this situation, neither thread will be woken. When scheduled, the waiting
1620  * threads will execute either a wait, which will return immediately, followed
1621  * by a condition check, which will pass for the correct thread and fail
1622  * for the wrong thread, or they will execute the condition check immediately.
1623  *
1624  * 3. The wrong thread is running, the correct thread is not
1625  * The wrong thread will be woken, but will fail it's condition check and
1626  * re-execute wait. The correct thread, when scheduled, will execute either
1627  * it's condition check (which will pass), or wait, which returns immediately
1628  * when called the first time after the thread is scheduled, followed by it's
1629  * condition check (which will pass).
1630  *
1631  * 4. Both threads are running
1632  * Both threads will be woken. The wrong thread will fail it's condition check
1633  * and execute another wait, while the correct thread will pass it's condition
1634  * check.
1635  *
1636  * @t: the task to set the thread ID for
1637  */
1638 int set_thread_tidr(struct task_struct *t)
1639 {
1640     if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1641         return -EINVAL;
1642 
1643     if (t != current)
1644         return -EINVAL;
1645 
1646     if (t->thread.tidr)
1647         return 0;
1648 
1649     t->thread.tidr = (u16)task_pid_nr(t);
1650     mtspr(SPRN_TIDR, t->thread.tidr);
1651 
1652     return 0;
1653 }
1654 EXPORT_SYMBOL_GPL(set_thread_tidr);
1655 
1656 #endif /* CONFIG_PPC64 */
1657 
1658 void
1659 release_thread(struct task_struct *t)
1660 {
1661 }
1662 
1663 /*
1664  * this gets called so that we can store coprocessor state into memory and
1665  * copy the current task into the new thread.
1666  */
1667 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1668 {
1669     flush_all_to_thread(src);
1670     /*
1671      * Flush TM state out so we can copy it.  __switch_to_tm() does this
1672      * flush but it removes the checkpointed state from the current CPU and
1673      * transitions the CPU out of TM mode.  Hence we need to call
1674      * tm_recheckpoint_new_task() (on the same task) to restore the
1675      * checkpointed state back and the TM mode.
1676      *
1677      * Can't pass dst because it isn't ready. Doesn't matter, passing
1678      * dst is only important for __switch_to()
1679      */
1680     __switch_to_tm(src, src);
1681 
1682     *dst = *src;
1683 
1684     clear_task_ebb(dst);
1685 
1686     return 0;
1687 }
1688 
1689 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1690 {
1691 #ifdef CONFIG_PPC_64S_HASH_MMU
1692     unsigned long sp_vsid;
1693     unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1694 
1695     if (radix_enabled())
1696         return;
1697 
1698     if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1699         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1700             << SLB_VSID_SHIFT_1T;
1701     else
1702         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1703             << SLB_VSID_SHIFT;
1704     sp_vsid |= SLB_VSID_KERNEL | llp;
1705     p->thread.ksp_vsid = sp_vsid;
1706 #endif
1707 }
1708 
1709 /*
1710  * Copy a thread..
1711  */
1712 
1713 /*
1714  * Copy architecture-specific thread state
1715  */
1716 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
1717 {
1718     unsigned long clone_flags = args->flags;
1719     unsigned long usp = args->stack;
1720     unsigned long tls = args->tls;
1721     struct pt_regs *childregs, *kregs;
1722     extern void ret_from_fork(void);
1723     extern void ret_from_fork_scv(void);
1724     extern void ret_from_kernel_thread(void);
1725     void (*f)(void);
1726     unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1727     struct thread_info *ti = task_thread_info(p);
1728 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1729     int i;
1730 #endif
1731 
1732     klp_init_thread_info(p);
1733 
1734     /* Copy registers */
1735     sp -= sizeof(struct pt_regs);
1736     childregs = (struct pt_regs *) sp;
1737     if (unlikely(args->fn)) {
1738         /* kernel thread */
1739         memset(childregs, 0, sizeof(struct pt_regs));
1740         childregs->gpr[1] = sp + sizeof(struct pt_regs);
1741         /* function */
1742         if (args->fn)
1743             childregs->gpr[14] = ppc_function_entry((void *)args->fn);
1744 #ifdef CONFIG_PPC64
1745         clear_tsk_thread_flag(p, TIF_32BIT);
1746         childregs->softe = IRQS_ENABLED;
1747 #endif
1748         childregs->gpr[15] = (unsigned long)args->fn_arg;
1749         p->thread.regs = NULL;  /* no user register state */
1750         ti->flags |= _TIF_RESTOREALL;
1751         f = ret_from_kernel_thread;
1752     } else {
1753         /* user thread */
1754         struct pt_regs *regs = current_pt_regs();
1755         *childregs = *regs;
1756         if (usp)
1757             childregs->gpr[1] = usp;
1758         p->thread.regs = childregs;
1759         /* 64s sets this in ret_from_fork */
1760         if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
1761             childregs->gpr[3] = 0;  /* Result from fork() */
1762         if (clone_flags & CLONE_SETTLS) {
1763             if (!is_32bit_task())
1764                 childregs->gpr[13] = tls;
1765             else
1766                 childregs->gpr[2] = tls;
1767         }
1768 
1769         if (trap_is_scv(regs))
1770             f = ret_from_fork_scv;
1771         else
1772             f = ret_from_fork;
1773     }
1774     childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1775     sp -= STACK_FRAME_OVERHEAD;
1776 
1777     /*
1778      * The way this works is that at some point in the future
1779      * some task will call _switch to switch to the new task.
1780      * That will pop off the stack frame created below and start
1781      * the new task running at ret_from_fork.  The new task will
1782      * do some house keeping and then return from the fork or clone
1783      * system call, using the stack frame created above.
1784      */
1785     ((unsigned long *)sp)[0] = 0;
1786     sp -= sizeof(struct pt_regs);
1787     kregs = (struct pt_regs *) sp;
1788     sp -= STACK_FRAME_OVERHEAD;
1789     p->thread.ksp = sp;
1790 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1791     for (i = 0; i < nr_wp_slots(); i++)
1792         p->thread.ptrace_bps[i] = NULL;
1793 #endif
1794 
1795 #ifdef CONFIG_PPC_FPU_REGS
1796     p->thread.fp_save_area = NULL;
1797 #endif
1798 #ifdef CONFIG_ALTIVEC
1799     p->thread.vr_save_area = NULL;
1800 #endif
1801 #if defined(CONFIG_PPC_BOOK3S_32) && defined(CONFIG_PPC_KUAP)
1802     p->thread.kuap = KUAP_NONE;
1803 #endif
1804 #if defined(CONFIG_BOOKE_OR_40x) && defined(CONFIG_PPC_KUAP)
1805     p->thread.pid = MMU_NO_CONTEXT;
1806 #endif
1807 
1808     setup_ksp_vsid(p, sp);
1809 
1810 #ifdef CONFIG_PPC64 
1811     if (cpu_has_feature(CPU_FTR_DSCR)) {
1812         p->thread.dscr_inherit = current->thread.dscr_inherit;
1813         p->thread.dscr = mfspr(SPRN_DSCR);
1814     }
1815     if (cpu_has_feature(CPU_FTR_HAS_PPR))
1816         childregs->ppr = DEFAULT_PPR;
1817 
1818     p->thread.tidr = 0;
1819 #endif
1820     /*
1821      * Run with the current AMR value of the kernel
1822      */
1823 #ifdef CONFIG_PPC_PKEY
1824     if (mmu_has_feature(MMU_FTR_BOOK3S_KUAP))
1825         kregs->amr = AMR_KUAP_BLOCKED;
1826 
1827     if (mmu_has_feature(MMU_FTR_BOOK3S_KUEP))
1828         kregs->iamr = AMR_KUEP_BLOCKED;
1829 #endif
1830     kregs->nip = ppc_function_entry(f);
1831     return 0;
1832 }
1833 
1834 void preload_new_slb_context(unsigned long start, unsigned long sp);
1835 
1836 /*
1837  * Set up a thread for executing a new program
1838  */
1839 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1840 {
1841 #ifdef CONFIG_PPC64
1842     unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1843 
1844     if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled())
1845         preload_new_slb_context(start, sp);
1846 #endif
1847 
1848 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1849     /*
1850      * Clear any transactional state, we're exec()ing. The cause is
1851      * not important as there will never be a recheckpoint so it's not
1852      * user visible.
1853      */
1854     if (MSR_TM_SUSPENDED(mfmsr()))
1855         tm_reclaim_current(0);
1856 #endif
1857 
1858     memset(&regs->gpr[1], 0, sizeof(regs->gpr) - sizeof(regs->gpr[0]));
1859     regs->ctr = 0;
1860     regs->link = 0;
1861     regs->xer = 0;
1862     regs->ccr = 0;
1863     regs->gpr[1] = sp;
1864 
1865 #ifdef CONFIG_PPC32
1866     regs->mq = 0;
1867     regs->nip = start;
1868     regs->msr = MSR_USER;
1869 #else
1870     if (!is_32bit_task()) {
1871         unsigned long entry;
1872 
1873         if (is_elf2_task()) {
1874             /* Look ma, no function descriptors! */
1875             entry = start;
1876 
1877             /*
1878              * Ulrich says:
1879              *   The latest iteration of the ABI requires that when
1880              *   calling a function (at its global entry point),
1881              *   the caller must ensure r12 holds the entry point
1882              *   address (so that the function can quickly
1883              *   establish addressability).
1884              */
1885             regs->gpr[12] = start;
1886             /* Make sure that's restored on entry to userspace. */
1887             set_thread_flag(TIF_RESTOREALL);
1888         } else {
1889             unsigned long toc;
1890 
1891             /* start is a relocated pointer to the function
1892              * descriptor for the elf _start routine.  The first
1893              * entry in the function descriptor is the entry
1894              * address of _start and the second entry is the TOC
1895              * value we need to use.
1896              */
1897             __get_user(entry, (unsigned long __user *)start);
1898             __get_user(toc, (unsigned long __user *)start+1);
1899 
1900             /* Check whether the e_entry function descriptor entries
1901              * need to be relocated before we can use them.
1902              */
1903             if (load_addr != 0) {
1904                 entry += load_addr;
1905                 toc   += load_addr;
1906             }
1907             regs->gpr[2] = toc;
1908         }
1909         regs_set_return_ip(regs, entry);
1910         regs_set_return_msr(regs, MSR_USER64);
1911     } else {
1912         regs->gpr[2] = 0;
1913         regs_set_return_ip(regs, start);
1914         regs_set_return_msr(regs, MSR_USER32);
1915     }
1916 
1917 #endif
1918 #ifdef CONFIG_VSX
1919     current->thread.used_vsr = 0;
1920 #endif
1921     current->thread.load_slb = 0;
1922     current->thread.load_fp = 0;
1923 #ifdef CONFIG_PPC_FPU_REGS
1924     memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1925     current->thread.fp_save_area = NULL;
1926 #endif
1927 #ifdef CONFIG_ALTIVEC
1928     memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1929     current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1930     current->thread.vr_save_area = NULL;
1931     current->thread.vrsave = 0;
1932     current->thread.used_vr = 0;
1933     current->thread.load_vec = 0;
1934 #endif /* CONFIG_ALTIVEC */
1935 #ifdef CONFIG_SPE
1936     memset(current->thread.evr, 0, sizeof(current->thread.evr));
1937     current->thread.acc = 0;
1938     current->thread.spefscr = 0;
1939     current->thread.used_spe = 0;
1940 #endif /* CONFIG_SPE */
1941 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1942     current->thread.tm_tfhar = 0;
1943     current->thread.tm_texasr = 0;
1944     current->thread.tm_tfiar = 0;
1945     current->thread.load_tm = 0;
1946 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1947 }
1948 EXPORT_SYMBOL(start_thread);
1949 
1950 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1951         | PR_FP_EXC_RES | PR_FP_EXC_INV)
1952 
1953 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1954 {
1955     struct pt_regs *regs = tsk->thread.regs;
1956 
1957     /* This is a bit hairy.  If we are an SPE enabled  processor
1958      * (have embedded fp) we store the IEEE exception enable flags in
1959      * fpexc_mode.  fpexc_mode is also used for setting FP exception
1960      * mode (asyn, precise, disabled) for 'Classic' FP. */
1961     if (val & PR_FP_EXC_SW_ENABLE) {
1962         if (cpu_has_feature(CPU_FTR_SPE)) {
1963             /*
1964              * When the sticky exception bits are set
1965              * directly by userspace, it must call prctl
1966              * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1967              * in the existing prctl settings) or
1968              * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1969              * the bits being set).  <fenv.h> functions
1970              * saving and restoring the whole
1971              * floating-point environment need to do so
1972              * anyway to restore the prctl settings from
1973              * the saved environment.
1974              */
1975 #ifdef CONFIG_SPE
1976             tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1977             tsk->thread.fpexc_mode = val &
1978                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1979 #endif
1980             return 0;
1981         } else {
1982             return -EINVAL;
1983         }
1984     }
1985 
1986     /* on a CONFIG_SPE this does not hurt us.  The bits that
1987      * __pack_fe01 use do not overlap with bits used for
1988      * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1989      * on CONFIG_SPE implementations are reserved so writing to
1990      * them does not change anything */
1991     if (val > PR_FP_EXC_PRECISE)
1992         return -EINVAL;
1993     tsk->thread.fpexc_mode = __pack_fe01(val);
1994     if (regs != NULL && (regs->msr & MSR_FP) != 0) {
1995         regs_set_return_msr(regs, (regs->msr & ~(MSR_FE0|MSR_FE1))
1996                         | tsk->thread.fpexc_mode);
1997     }
1998     return 0;
1999 }
2000 
2001 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
2002 {
2003     unsigned int val = 0;
2004 
2005     if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE) {
2006         if (cpu_has_feature(CPU_FTR_SPE)) {
2007             /*
2008              * When the sticky exception bits are set
2009              * directly by userspace, it must call prctl
2010              * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
2011              * in the existing prctl settings) or
2012              * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
2013              * the bits being set).  <fenv.h> functions
2014              * saving and restoring the whole
2015              * floating-point environment need to do so
2016              * anyway to restore the prctl settings from
2017              * the saved environment.
2018              */
2019 #ifdef CONFIG_SPE
2020             tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
2021             val = tsk->thread.fpexc_mode;
2022 #endif
2023         } else
2024             return -EINVAL;
2025     } else {
2026         val = __unpack_fe01(tsk->thread.fpexc_mode);
2027     }
2028     return put_user(val, (unsigned int __user *) adr);
2029 }
2030 
2031 int set_endian(struct task_struct *tsk, unsigned int val)
2032 {
2033     struct pt_regs *regs = tsk->thread.regs;
2034 
2035     if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
2036         (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
2037         return -EINVAL;
2038 
2039     if (regs == NULL)
2040         return -EINVAL;
2041 
2042     if (val == PR_ENDIAN_BIG)
2043         regs_set_return_msr(regs, regs->msr & ~MSR_LE);
2044     else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
2045         regs_set_return_msr(regs, regs->msr | MSR_LE);
2046     else
2047         return -EINVAL;
2048 
2049     return 0;
2050 }
2051 
2052 int get_endian(struct task_struct *tsk, unsigned long adr)
2053 {
2054     struct pt_regs *regs = tsk->thread.regs;
2055     unsigned int val;
2056 
2057     if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
2058         !cpu_has_feature(CPU_FTR_REAL_LE))
2059         return -EINVAL;
2060 
2061     if (regs == NULL)
2062         return -EINVAL;
2063 
2064     if (regs->msr & MSR_LE) {
2065         if (cpu_has_feature(CPU_FTR_REAL_LE))
2066             val = PR_ENDIAN_LITTLE;
2067         else
2068             val = PR_ENDIAN_PPC_LITTLE;
2069     } else
2070         val = PR_ENDIAN_BIG;
2071 
2072     return put_user(val, (unsigned int __user *)adr);
2073 }
2074 
2075 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
2076 {
2077     tsk->thread.align_ctl = val;
2078     return 0;
2079 }
2080 
2081 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
2082 {
2083     return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
2084 }
2085 
2086 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
2087                   unsigned long nbytes)
2088 {
2089     unsigned long stack_page;
2090     unsigned long cpu = task_cpu(p);
2091 
2092     stack_page = (unsigned long)hardirq_ctx[cpu];
2093     if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2094         return 1;
2095 
2096     stack_page = (unsigned long)softirq_ctx[cpu];
2097     if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2098         return 1;
2099 
2100     return 0;
2101 }
2102 
2103 static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
2104                     unsigned long nbytes)
2105 {
2106 #ifdef CONFIG_PPC64
2107     unsigned long stack_page;
2108     unsigned long cpu = task_cpu(p);
2109 
2110     if (!paca_ptrs)
2111         return 0;
2112 
2113     stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
2114     if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2115         return 1;
2116 
2117 # ifdef CONFIG_PPC_BOOK3S_64
2118     stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
2119     if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2120         return 1;
2121 
2122     stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
2123     if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2124         return 1;
2125 # endif
2126 #endif
2127 
2128     return 0;
2129 }
2130 
2131 
2132 int validate_sp(unsigned long sp, struct task_struct *p,
2133                unsigned long nbytes)
2134 {
2135     unsigned long stack_page = (unsigned long)task_stack_page(p);
2136 
2137     if (sp < THREAD_SIZE)
2138         return 0;
2139 
2140     if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2141         return 1;
2142 
2143     if (valid_irq_stack(sp, p, nbytes))
2144         return 1;
2145 
2146     return valid_emergency_stack(sp, p, nbytes);
2147 }
2148 
2149 EXPORT_SYMBOL(validate_sp);
2150 
2151 static unsigned long ___get_wchan(struct task_struct *p)
2152 {
2153     unsigned long ip, sp;
2154     int count = 0;
2155 
2156     sp = p->thread.ksp;
2157     if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2158         return 0;
2159 
2160     do {
2161         sp = READ_ONCE_NOCHECK(*(unsigned long *)sp);
2162         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2163             task_is_running(p))
2164             return 0;
2165         if (count > 0) {
2166             ip = READ_ONCE_NOCHECK(((unsigned long *)sp)[STACK_FRAME_LR_SAVE]);
2167             if (!in_sched_functions(ip))
2168                 return ip;
2169         }
2170     } while (count++ < 16);
2171     return 0;
2172 }
2173 
2174 unsigned long __get_wchan(struct task_struct *p)
2175 {
2176     unsigned long ret;
2177 
2178     if (!try_get_task_stack(p))
2179         return 0;
2180 
2181     ret = ___get_wchan(p);
2182 
2183     put_task_stack(p);
2184 
2185     return ret;
2186 }
2187 
2188 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2189 
2190 void __no_sanitize_address show_stack(struct task_struct *tsk,
2191                       unsigned long *stack,
2192                       const char *loglvl)
2193 {
2194     unsigned long sp, ip, lr, newsp;
2195     int count = 0;
2196     int firstframe = 1;
2197     unsigned long ret_addr;
2198     int ftrace_idx = 0;
2199 
2200     if (tsk == NULL)
2201         tsk = current;
2202 
2203     if (!try_get_task_stack(tsk))
2204         return;
2205 
2206     sp = (unsigned long) stack;
2207     if (sp == 0) {
2208         if (tsk == current)
2209             sp = current_stack_frame();
2210         else
2211             sp = tsk->thread.ksp;
2212     }
2213 
2214     lr = 0;
2215     printk("%sCall Trace:\n", loglvl);
2216     do {
2217         if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2218             break;
2219 
2220         stack = (unsigned long *) sp;
2221         newsp = stack[0];
2222         ip = stack[STACK_FRAME_LR_SAVE];
2223         if (!firstframe || ip != lr) {
2224             printk("%s["REG"] ["REG"] %pS",
2225                 loglvl, sp, ip, (void *)ip);
2226             ret_addr = ftrace_graph_ret_addr(current,
2227                         &ftrace_idx, ip, stack);
2228             if (ret_addr != ip)
2229                 pr_cont(" (%pS)", (void *)ret_addr);
2230             if (firstframe)
2231                 pr_cont(" (unreliable)");
2232             pr_cont("\n");
2233         }
2234         firstframe = 0;
2235 
2236         /*
2237          * See if this is an exception frame.
2238          * We look for the "regshere" marker in the current frame.
2239          */
2240         if (validate_sp(sp, tsk, STACK_FRAME_WITH_PT_REGS)
2241             && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2242             struct pt_regs *regs = (struct pt_regs *)
2243                 (sp + STACK_FRAME_OVERHEAD);
2244 
2245             lr = regs->link;
2246             printk("%s--- interrupt: %lx at %pS\n",
2247                    loglvl, regs->trap, (void *)regs->nip);
2248             __show_regs(regs);
2249             printk("%s--- interrupt: %lx\n",
2250                    loglvl, regs->trap);
2251 
2252             firstframe = 1;
2253         }
2254 
2255         sp = newsp;
2256     } while (count++ < kstack_depth_to_print);
2257 
2258     put_task_stack(tsk);
2259 }
2260 
2261 #ifdef CONFIG_PPC64
2262 /* Called with hard IRQs off */
2263 void notrace __ppc64_runlatch_on(void)
2264 {
2265     struct thread_info *ti = current_thread_info();
2266 
2267     if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2268         /*
2269          * Least significant bit (RUN) is the only writable bit of
2270          * the CTRL register, so we can avoid mfspr. 2.06 is not the
2271          * earliest ISA where this is the case, but it's convenient.
2272          */
2273         mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2274     } else {
2275         unsigned long ctrl;
2276 
2277         /*
2278          * Some architectures (e.g., Cell) have writable fields other
2279          * than RUN, so do the read-modify-write.
2280          */
2281         ctrl = mfspr(SPRN_CTRLF);
2282         ctrl |= CTRL_RUNLATCH;
2283         mtspr(SPRN_CTRLT, ctrl);
2284     }
2285 
2286     ti->local_flags |= _TLF_RUNLATCH;
2287 }
2288 
2289 /* Called with hard IRQs off */
2290 void notrace __ppc64_runlatch_off(void)
2291 {
2292     struct thread_info *ti = current_thread_info();
2293 
2294     ti->local_flags &= ~_TLF_RUNLATCH;
2295 
2296     if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2297         mtspr(SPRN_CTRLT, 0);
2298     } else {
2299         unsigned long ctrl;
2300 
2301         ctrl = mfspr(SPRN_CTRLF);
2302         ctrl &= ~CTRL_RUNLATCH;
2303         mtspr(SPRN_CTRLT, ctrl);
2304     }
2305 }
2306 #endif /* CONFIG_PPC64 */
2307 
2308 unsigned long arch_align_stack(unsigned long sp)
2309 {
2310     if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2311         sp -= get_random_int() & ~PAGE_MASK;
2312     return sp & ~0xf;
2313 }