2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * Copyright (C) 1994 - 1999, 2000 by Ralf Baechle and others.
7 * Copyright (C) 2005, 2006 by Ralf Baechle (ralf@linux-mips.org)
8 * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
9 * Copyright (C) 2004 Thiemo Seufer
10 * Copyright (C) 2013 Imagination Technologies Ltd.
12 #include <linux/errno.h>
13 #include <linux/sched.h>
14 #include <linux/sched/debug.h>
15 #include <linux/sched/task.h>
16 #include <linux/sched/task_stack.h>
17 #include <linux/tick.h>
18 #include <linux/kernel.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/export.h>
23 #include <linux/ptrace.h>
24 #include <linux/mman.h>
25 #include <linux/personality.h>
26 #include <linux/sys.h>
27 #include <linux/init.h>
28 #include <linux/completion.h>
29 #include <linux/kallsyms.h>
30 #include <linux/random.h>
31 #include <linux/prctl.h>
32 #include <linux/nmi.h>
35 #include <asm/bootinfo.h>
37 #include <asm/dsemul.h>
42 #include <asm/pgtable.h>
43 #include <asm/mipsregs.h>
44 #include <asm/processor.h>
46 #include <linux/uaccess.h>
49 #include <asm/isadep.h>
51 #include <asm/stacktrace.h>
52 #include <asm/irq_regs.h>
54 #ifdef CONFIG_HOTPLUG_CPU
55 void arch_cpu_idle_dead(void)
61 asmlinkage void ret_from_fork(void);
62 asmlinkage void ret_from_kernel_thread(void);
64 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
68 /* New thread loses kernel privileges. */
69 status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
71 regs->cp0_status = status;
73 clear_thread_flag(TIF_MSA_CTX_LIVE);
75 atomic_set(¤t->thread.bd_emu_frame, BD_EMUFRAME_NONE);
81 void exit_thread(struct task_struct *tsk)
84 * User threads may have allocated a delay slot emulation frame.
85 * If so, clean up that allocation.
87 if (!(current->flags & PF_KTHREAD))
88 dsemul_thread_cleanup(tsk);
91 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
94 * Save any process state which is live in hardware registers to the
95 * parent context prior to duplication. This prevents the new child
96 * state becoming stale if the parent is preempted before copy_thread()
97 * gets a chance to save the parent's live hardware registers to the
102 if (is_msa_enabled())
104 else if (is_fpu_owner())
116 * Copy architecture-specific thread state
118 int copy_thread_tls(unsigned long clone_flags, unsigned long usp,
119 unsigned long kthread_arg, struct task_struct *p, unsigned long tls)
121 struct thread_info *ti = task_thread_info(p);
122 struct pt_regs *childregs, *regs = current_pt_regs();
123 unsigned long childksp;
125 childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
127 /* set up new TSS. */
128 childregs = (struct pt_regs *) childksp - 1;
129 /* Put the stack after the struct pt_regs. */
130 childksp = (unsigned long) childregs;
131 p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
132 if (unlikely(p->flags & PF_KTHREAD)) {
134 unsigned long status = p->thread.cp0_status;
135 memset(childregs, 0, sizeof(struct pt_regs));
136 ti->addr_limit = KERNEL_DS;
137 p->thread.reg16 = usp; /* fn */
138 p->thread.reg17 = kthread_arg;
139 p->thread.reg29 = childksp;
140 p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
141 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
142 status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
143 ((status & (ST0_KUC | ST0_IEC)) << 2);
147 childregs->cp0_status = status;
153 childregs->regs[7] = 0; /* Clear error flag */
154 childregs->regs[2] = 0; /* Child gets zero as return value */
156 childregs->regs[29] = usp;
157 ti->addr_limit = USER_DS;
159 p->thread.reg29 = (unsigned long) childregs;
160 p->thread.reg31 = (unsigned long) ret_from_fork;
163 * New tasks lose permission to use the fpu. This accelerates context
164 * switching for most programs since they don't use the fpu.
166 childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
168 clear_tsk_thread_flag(p, TIF_USEDFPU);
169 clear_tsk_thread_flag(p, TIF_USEDMSA);
170 clear_tsk_thread_flag(p, TIF_MSA_CTX_LIVE);
172 #ifdef CONFIG_MIPS_MT_FPAFF
173 clear_tsk_thread_flag(p, TIF_FPUBOUND);
174 #endif /* CONFIG_MIPS_MT_FPAFF */
176 atomic_set(&p->thread.bd_emu_frame, BD_EMUFRAME_NONE);
178 if (clone_flags & CLONE_SETTLS)
184 #ifdef CONFIG_STACKPROTECTOR
185 #include <linux/stackprotector.h>
186 unsigned long __stack_chk_guard __read_mostly;
187 EXPORT_SYMBOL(__stack_chk_guard);
190 struct mips_frame_info {
192 unsigned long func_size;
197 #define J_TARGET(pc,target) \
198 (((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
200 static inline int is_ra_save_ins(union mips_instruction *ip, int *poff)
202 #ifdef CONFIG_CPU_MICROMIPS
205 * swm16 reglist,offset(sp)
206 * swm32 reglist,offset(sp)
208 * jradiussp - NOT SUPPORTED
210 * microMIPS is way more fun...
212 if (mm_insn_16bit(ip->word >> 16)) {
213 switch (ip->mm16_r5_format.opcode) {
215 if (ip->mm16_r5_format.rt != 31)
218 *poff = ip->mm16_r5_format.imm;
219 *poff = (*poff << 2) / sizeof(ulong);
223 switch (ip->mm16_m_format.func) {
225 *poff = ip->mm16_m_format.imm;
226 *poff += 1 + ip->mm16_m_format.rlist;
227 *poff = (*poff << 2) / sizeof(ulong);
239 switch (ip->i_format.opcode) {
241 if (ip->i_format.rs != 29)
243 if (ip->i_format.rt != 31)
246 *poff = ip->i_format.simmediate / sizeof(ulong);
250 switch (ip->mm_m_format.func) {
252 if (ip->mm_m_format.rd < 0x10)
254 if (ip->mm_m_format.base != 29)
257 *poff = ip->mm_m_format.simmediate;
258 *poff += (ip->mm_m_format.rd & 0xf) * sizeof(u32);
259 *poff /= sizeof(ulong);
269 /* sw / sd $ra, offset($sp) */
270 if ((ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
271 ip->i_format.rs == 29 && ip->i_format.rt == 31) {
272 *poff = ip->i_format.simmediate / sizeof(ulong);
280 static inline int is_jump_ins(union mips_instruction *ip)
282 #ifdef CONFIG_CPU_MICROMIPS
284 * jr16,jrc,jalr16,jalr16
286 * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
287 * jraddiusp - NOT SUPPORTED
289 * microMIPS is kind of more fun...
291 if (mm_insn_16bit(ip->word >> 16)) {
292 if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
293 (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op))
298 if (ip->j_format.opcode == mm_j32_op)
300 if (ip->j_format.opcode == mm_jal32_op)
302 if (ip->r_format.opcode != mm_pool32a_op ||
303 ip->r_format.func != mm_pool32axf_op)
305 return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
307 if (ip->j_format.opcode == j_op)
309 if (ip->j_format.opcode == jal_op)
311 if (ip->r_format.opcode != spec_op)
313 return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
317 static inline int is_sp_move_ins(union mips_instruction *ip, int *frame_size)
319 #ifdef CONFIG_CPU_MICROMIPS
326 * jradiussp - NOT SUPPORTED
328 * microMIPS is not more fun...
330 if (mm_insn_16bit(ip->word >> 16)) {
331 if (ip->mm16_r3_format.opcode == mm_pool16d_op &&
332 ip->mm16_r3_format.simmediate & mm_addiusp_func) {
333 tmp = ip->mm_b0_format.simmediate >> 1;
334 tmp = ((tmp & 0x1ff) ^ 0x100) - 0x100;
335 if ((tmp + 2) < 4) /* 0x0,0x1,0x1fe,0x1ff are special */
337 *frame_size = -(signed short)(tmp << 2);
340 if (ip->mm16_r5_format.opcode == mm_pool16d_op &&
341 ip->mm16_r5_format.rt == 29) {
342 tmp = ip->mm16_r5_format.imm >> 1;
343 *frame_size = -(signed short)(tmp & 0xf);
349 if (ip->mm_i_format.opcode == mm_addiu32_op &&
350 ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29) {
351 *frame_size = -ip->i_format.simmediate;
355 /* addiu/daddiu sp,sp,-imm */
356 if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
359 if (ip->i_format.opcode == addiu_op ||
360 ip->i_format.opcode == daddiu_op) {
361 *frame_size = -ip->i_format.simmediate;
368 static int get_frame_info(struct mips_frame_info *info)
370 bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
371 union mips_instruction insn, *ip, *ip_end;
372 const unsigned int max_insns = 128;
373 unsigned int last_insn_size = 0;
375 bool saw_jump = false;
377 info->pc_offset = -1;
378 info->frame_size = 0;
380 ip = (void *)msk_isa16_mode((ulong)info->func);
384 ip_end = (void *)ip + info->func_size;
386 for (i = 0; i < max_insns && ip < ip_end; i++) {
387 ip = (void *)ip + last_insn_size;
388 if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
389 insn.word = ip->halfword[0] << 16;
391 } else if (is_mmips) {
392 insn.word = ip->halfword[0] << 16 | ip->halfword[1];
395 insn.word = ip->word;
399 if (!info->frame_size) {
400 is_sp_move_ins(&insn, &info->frame_size);
402 } else if (!saw_jump && is_jump_ins(ip)) {
404 * If we see a jump instruction, we are finished
405 * with the frame save.
407 * Some functions can have a shortcut return at
408 * the beginning of the function, so don't start
409 * looking for jump instruction until we see the
412 * The RA save instruction can get put into the
413 * delay slot of the jump instruction, so look
414 * at the next instruction, too.
419 if (info->pc_offset == -1 &&
420 is_ra_save_ins(&insn, &info->pc_offset))
425 if (info->frame_size && info->pc_offset >= 0) /* nested */
427 if (info->pc_offset < 0) /* leaf */
429 /* prologue seems bogus... */
434 static struct mips_frame_info schedule_mfi __read_mostly;
436 #ifdef CONFIG_KALLSYMS
437 static unsigned long get___schedule_addr(void)
439 return kallsyms_lookup_name("__schedule");
442 static unsigned long get___schedule_addr(void)
444 union mips_instruction *ip = (void *)schedule;
448 for (i = 0; i < max_insns; i++, ip++) {
449 if (ip->j_format.opcode == j_op)
450 return J_TARGET(ip, ip->j_format.target);
456 static int __init frame_info_init(void)
458 unsigned long size = 0;
459 #ifdef CONFIG_KALLSYMS
464 addr = get___schedule_addr();
466 addr = (unsigned long)schedule;
468 #ifdef CONFIG_KALLSYMS
469 kallsyms_lookup_size_offset(addr, &size, &ofs);
471 schedule_mfi.func = (void *)addr;
472 schedule_mfi.func_size = size;
474 get_frame_info(&schedule_mfi);
477 * Without schedule() frame info, result given by
478 * thread_saved_pc() and get_wchan() are not reliable.
480 if (schedule_mfi.pc_offset < 0)
481 printk("Can't analyze schedule() prologue at %p\n", schedule);
486 arch_initcall(frame_info_init);
489 * Return saved PC of a blocked thread.
491 static unsigned long thread_saved_pc(struct task_struct *tsk)
493 struct thread_struct *t = &tsk->thread;
495 /* New born processes are a special case */
496 if (t->reg31 == (unsigned long) ret_from_fork)
498 if (schedule_mfi.pc_offset < 0)
500 return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
504 #ifdef CONFIG_KALLSYMS
505 /* generic stack unwinding function */
506 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
511 unsigned long low, high, irq_stack_high;
512 struct mips_frame_info info;
513 unsigned long size, ofs;
514 struct pt_regs *regs;
521 * IRQ stacks start at IRQ_STACK_START
522 * task stacks at THREAD_SIZE - 32
525 if (!preemptible() && on_irq_stack(raw_smp_processor_id(), *sp)) {
526 high = stack_page + IRQ_STACK_START;
527 irq_stack_high = high;
529 high = stack_page + THREAD_SIZE - 32;
534 * If we reached the top of the interrupt stack, start unwinding
535 * the interrupted task stack.
537 if (unlikely(*sp == irq_stack_high)) {
538 unsigned long task_sp = *(unsigned long *)*sp;
541 * Check that the pointer saved in the IRQ stack head points to
542 * something within the stack of the current task
544 if (!object_is_on_stack((void *)task_sp))
548 * Follow pointer to tasks kernel stack frame where interrupted
551 regs = (struct pt_regs *)task_sp;
553 if (!user_mode(regs) && __kernel_text_address(pc)) {
554 *sp = regs->regs[29];
555 *ra = regs->regs[31];
560 if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
563 * Return ra if an exception occurred at the first instruction
565 if (unlikely(ofs == 0)) {
571 info.func = (void *)(pc - ofs);
572 info.func_size = ofs; /* analyze from start to ofs */
573 leaf = get_frame_info(&info);
577 if (*sp < low || *sp + info.frame_size > high)
582 * For some extreme cases, get_frame_info() can
583 * consider wrongly a nested function as a leaf
584 * one. In that cases avoid to return always the
587 pc = pc != *ra ? *ra : 0;
589 pc = ((unsigned long *)(*sp))[info.pc_offset];
591 *sp += info.frame_size;
593 return __kernel_text_address(pc) ? pc : 0;
595 EXPORT_SYMBOL(unwind_stack_by_address);
597 /* used by show_backtrace() */
598 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
599 unsigned long pc, unsigned long *ra)
601 unsigned long stack_page = 0;
604 for_each_possible_cpu(cpu) {
605 if (on_irq_stack(cpu, *sp)) {
606 stack_page = (unsigned long)irq_stack[cpu];
612 stack_page = (unsigned long)task_stack_page(task);
614 return unwind_stack_by_address(stack_page, sp, pc, ra);
619 * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
621 unsigned long get_wchan(struct task_struct *task)
623 unsigned long pc = 0;
624 #ifdef CONFIG_KALLSYMS
626 unsigned long ra = 0;
629 if (!task || task == current || task->state == TASK_RUNNING)
631 if (!task_stack_page(task))
634 pc = thread_saved_pc(task);
636 #ifdef CONFIG_KALLSYMS
637 sp = task->thread.reg29 + schedule_mfi.frame_size;
639 while (in_sched_functions(pc))
640 pc = unwind_stack(task, &sp, pc, &ra);
648 * Don't forget that the stack pointer must be aligned on a 8 bytes
649 * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
651 unsigned long arch_align_stack(unsigned long sp)
653 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
654 sp -= get_random_int() & ~PAGE_MASK;
659 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
660 static struct cpumask backtrace_csd_busy;
662 static void handle_backtrace(void *info)
664 nmi_cpu_backtrace(get_irq_regs());
665 cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
668 static void raise_backtrace(cpumask_t *mask)
670 call_single_data_t *csd;
673 for_each_cpu(cpu, mask) {
675 * If we previously sent an IPI to the target CPU & it hasn't
676 * cleared its bit in the busy cpumask then it didn't handle
677 * our previous IPI & it's not safe for us to reuse the
678 * call_single_data_t.
680 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
681 pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
686 csd = &per_cpu(backtrace_csd, cpu);
687 csd->func = handle_backtrace;
688 smp_call_function_single_async(cpu, csd);
692 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
694 nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace);
697 int mips_get_process_fp_mode(struct task_struct *task)
701 if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
702 value |= PR_FP_MODE_FR;
703 if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
704 value |= PR_FP_MODE_FRE;
709 static void prepare_for_fp_mode_switch(void *info)
711 struct mm_struct *mm = info;
713 if (current->mm == mm)
717 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
719 const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
720 struct task_struct *t;
723 /* If nothing to change, return right away, successfully. */
724 if (value == mips_get_process_fp_mode(task))
727 /* Only accept a mode change if 64-bit FP enabled for o32. */
728 if (!IS_ENABLED(CONFIG_MIPS_O32_FP64_SUPPORT))
731 /* And only for o32 tasks. */
732 if (IS_ENABLED(CONFIG_64BIT) && !test_thread_flag(TIF_32BIT_REGS))
735 /* Check the value is valid */
736 if (value & ~known_bits)
739 /* Setting FRE without FR is not supported. */
740 if ((value & (PR_FP_MODE_FR | PR_FP_MODE_FRE)) == PR_FP_MODE_FRE)
743 /* Avoid inadvertently triggering emulation */
744 if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
745 !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
747 if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
750 /* FR = 0 not supported in MIPS R6 */
751 if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
754 /* Proceed with the mode switch */
757 /* Save FP & vector context, then disable FPU & MSA */
758 if (task->signal == current->signal)
761 /* Prevent any threads from obtaining live FP context */
762 atomic_set(&task->mm->context.fp_mode_switching, 1);
763 smp_mb__after_atomic();
766 * If there are multiple online CPUs then force any which are running
767 * threads in this process to lose their FPU context, which they can't
768 * regain until fp_mode_switching is cleared later.
770 if (num_online_cpus() > 1) {
771 /* No need to send an IPI for the local CPU */
772 max_users = (task->mm == current->mm) ? 1 : 0;
774 if (atomic_read(¤t->mm->mm_users) > max_users)
775 smp_call_function(prepare_for_fp_mode_switch,
776 (void *)current->mm, 1);
780 * There are now no threads of the process with live FP context, so it
781 * is safe to proceed with the FP mode switch.
783 for_each_thread(task, t) {
784 /* Update desired FP register width */
785 if (value & PR_FP_MODE_FR) {
786 clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
788 set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
789 clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
792 /* Update desired FP single layout */
793 if (value & PR_FP_MODE_FRE)
794 set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
796 clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
799 /* Allow threads to use FP again */
800 atomic_set(&task->mm->context.fp_mode_switching, 0);
803 wake_up_var(&task->mm->context.fp_mode_switching);
808 #if defined(CONFIG_32BIT) || defined(CONFIG_MIPS32_O32)
809 void mips_dump_regs32(u32 *uregs, const struct pt_regs *regs)
813 for (i = MIPS32_EF_R1; i <= MIPS32_EF_R31; i++) {
814 /* k0/k1 are copied as zero. */
815 if (i == MIPS32_EF_R26 || i == MIPS32_EF_R27)
818 uregs[i] = regs->regs[i - MIPS32_EF_R0];
821 uregs[MIPS32_EF_LO] = regs->lo;
822 uregs[MIPS32_EF_HI] = regs->hi;
823 uregs[MIPS32_EF_CP0_EPC] = regs->cp0_epc;
824 uregs[MIPS32_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
825 uregs[MIPS32_EF_CP0_STATUS] = regs->cp0_status;
826 uregs[MIPS32_EF_CP0_CAUSE] = regs->cp0_cause;
828 #endif /* CONFIG_32BIT || CONFIG_MIPS32_O32 */
831 void mips_dump_regs64(u64 *uregs, const struct pt_regs *regs)
835 for (i = MIPS64_EF_R1; i <= MIPS64_EF_R31; i++) {
836 /* k0/k1 are copied as zero. */
837 if (i == MIPS64_EF_R26 || i == MIPS64_EF_R27)
840 uregs[i] = regs->regs[i - MIPS64_EF_R0];
843 uregs[MIPS64_EF_LO] = regs->lo;
844 uregs[MIPS64_EF_HI] = regs->hi;
845 uregs[MIPS64_EF_CP0_EPC] = regs->cp0_epc;
846 uregs[MIPS64_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
847 uregs[MIPS64_EF_CP0_STATUS] = regs->cp0_status;
848 uregs[MIPS64_EF_CP0_CAUSE] = regs->cp0_cause;
850 #endif /* CONFIG_64BIT */