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.
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>
33 #include <linux/cpu.h>
37 #include <asm/bootinfo.h>
39 #include <asm/dsemul.h>
43 #include <asm/mips-cps.h>
45 #include <asm/pgtable.h>
46 #include <asm/mipsregs.h>
47 #include <asm/processor.h>
49 #include <linux/uaccess.h>
52 #include <asm/isadep.h>
54 #include <asm/stacktrace.h>
55 #include <asm/irq_regs.h>
57 #ifdef CONFIG_HOTPLUG_CPU
58 void arch_cpu_idle_dead(void)
64 asmlinkage void ret_from_fork(void);
65 asmlinkage void ret_from_kernel_thread(void);
67 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
71 /* New thread loses kernel privileges. */
72 status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
74 regs->cp0_status = status;
76 clear_thread_flag(TIF_MSA_CTX_LIVE);
78 atomic_set(¤t->thread.bd_emu_frame, BD_EMUFRAME_NONE);
84 void exit_thread(struct task_struct *tsk)
87 * User threads may have allocated a delay slot emulation frame.
88 * If so, clean up that allocation.
90 if (!(current->flags & PF_KTHREAD))
91 dsemul_thread_cleanup(tsk);
94 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
97 * Save any process state which is live in hardware registers to the
98 * parent context prior to duplication. This prevents the new child
99 * state becoming stale if the parent is preempted before copy_thread()
100 * gets a chance to save the parent's live hardware registers to the
105 if (is_msa_enabled())
107 else if (is_fpu_owner())
119 * Copy architecture-specific thread state
121 int copy_thread_tls(unsigned long clone_flags, unsigned long usp,
122 unsigned long kthread_arg, struct task_struct *p, unsigned long tls)
124 struct thread_info *ti = task_thread_info(p);
125 struct pt_regs *childregs, *regs = current_pt_regs();
126 unsigned long childksp;
128 childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
130 /* set up new TSS. */
131 childregs = (struct pt_regs *) childksp - 1;
132 /* Put the stack after the struct pt_regs. */
133 childksp = (unsigned long) childregs;
134 p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
135 if (unlikely(p->flags & PF_KTHREAD)) {
137 unsigned long status = p->thread.cp0_status;
138 memset(childregs, 0, sizeof(struct pt_regs));
139 ti->addr_limit = KERNEL_DS;
140 p->thread.reg16 = usp; /* fn */
141 p->thread.reg17 = kthread_arg;
142 p->thread.reg29 = childksp;
143 p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
144 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
145 status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
146 ((status & (ST0_KUC | ST0_IEC)) << 2);
150 childregs->cp0_status = status;
156 childregs->regs[7] = 0; /* Clear error flag */
157 childregs->regs[2] = 0; /* Child gets zero as return value */
159 childregs->regs[29] = usp;
160 ti->addr_limit = USER_DS;
162 p->thread.reg29 = (unsigned long) childregs;
163 p->thread.reg31 = (unsigned long) ret_from_fork;
166 * New tasks lose permission to use the fpu. This accelerates context
167 * switching for most programs since they don't use the fpu.
169 childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
171 clear_tsk_thread_flag(p, TIF_USEDFPU);
172 clear_tsk_thread_flag(p, TIF_USEDMSA);
173 clear_tsk_thread_flag(p, TIF_MSA_CTX_LIVE);
175 #ifdef CONFIG_MIPS_MT_FPAFF
176 clear_tsk_thread_flag(p, TIF_FPUBOUND);
177 #endif /* CONFIG_MIPS_MT_FPAFF */
179 atomic_set(&p->thread.bd_emu_frame, BD_EMUFRAME_NONE);
181 if (clone_flags & CLONE_SETTLS)
187 #ifdef CONFIG_STACKPROTECTOR
188 #include <linux/stackprotector.h>
189 unsigned long __stack_chk_guard __read_mostly;
190 EXPORT_SYMBOL(__stack_chk_guard);
193 struct mips_frame_info {
195 unsigned long func_size;
200 #define J_TARGET(pc,target) \
201 (((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
203 static inline int is_ra_save_ins(union mips_instruction *ip, int *poff)
205 #ifdef CONFIG_CPU_MICROMIPS
208 * swm16 reglist,offset(sp)
209 * swm32 reglist,offset(sp)
211 * jradiussp - NOT SUPPORTED
213 * microMIPS is way more fun...
215 if (mm_insn_16bit(ip->word >> 16)) {
216 switch (ip->mm16_r5_format.opcode) {
218 if (ip->mm16_r5_format.rt != 31)
221 *poff = ip->mm16_r5_format.imm;
222 *poff = (*poff << 2) / sizeof(ulong);
226 switch (ip->mm16_m_format.func) {
228 *poff = ip->mm16_m_format.imm;
229 *poff += 1 + ip->mm16_m_format.rlist;
230 *poff = (*poff << 2) / sizeof(ulong);
242 switch (ip->i_format.opcode) {
244 if (ip->i_format.rs != 29)
246 if (ip->i_format.rt != 31)
249 *poff = ip->i_format.simmediate / sizeof(ulong);
253 switch (ip->mm_m_format.func) {
255 if (ip->mm_m_format.rd < 0x10)
257 if (ip->mm_m_format.base != 29)
260 *poff = ip->mm_m_format.simmediate;
261 *poff += (ip->mm_m_format.rd & 0xf) * sizeof(u32);
262 *poff /= sizeof(ulong);
272 /* sw / sd $ra, offset($sp) */
273 if ((ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
274 ip->i_format.rs == 29 && ip->i_format.rt == 31) {
275 *poff = ip->i_format.simmediate / sizeof(ulong);
283 static inline int is_jump_ins(union mips_instruction *ip)
285 #ifdef CONFIG_CPU_MICROMIPS
287 * jr16,jrc,jalr16,jalr16
289 * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
290 * jraddiusp - NOT SUPPORTED
292 * microMIPS is kind of more fun...
294 if (mm_insn_16bit(ip->word >> 16)) {
295 if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
296 (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op))
301 if (ip->j_format.opcode == mm_j32_op)
303 if (ip->j_format.opcode == mm_jal32_op)
305 if (ip->r_format.opcode != mm_pool32a_op ||
306 ip->r_format.func != mm_pool32axf_op)
308 return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
310 if (ip->j_format.opcode == j_op)
312 if (ip->j_format.opcode == jal_op)
314 if (ip->r_format.opcode != spec_op)
316 return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
320 static inline int is_sp_move_ins(union mips_instruction *ip, int *frame_size)
322 #ifdef CONFIG_CPU_MICROMIPS
329 * jradiussp - NOT SUPPORTED
331 * microMIPS is not more fun...
333 if (mm_insn_16bit(ip->word >> 16)) {
334 if (ip->mm16_r3_format.opcode == mm_pool16d_op &&
335 ip->mm16_r3_format.simmediate & mm_addiusp_func) {
336 tmp = ip->mm_b0_format.simmediate >> 1;
337 tmp = ((tmp & 0x1ff) ^ 0x100) - 0x100;
338 if ((tmp + 2) < 4) /* 0x0,0x1,0x1fe,0x1ff are special */
340 *frame_size = -(signed short)(tmp << 2);
343 if (ip->mm16_r5_format.opcode == mm_pool16d_op &&
344 ip->mm16_r5_format.rt == 29) {
345 tmp = ip->mm16_r5_format.imm >> 1;
346 *frame_size = -(signed short)(tmp & 0xf);
352 if (ip->mm_i_format.opcode == mm_addiu32_op &&
353 ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29) {
354 *frame_size = -ip->i_format.simmediate;
358 /* addiu/daddiu sp,sp,-imm */
359 if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
362 if (ip->i_format.opcode == addiu_op ||
363 ip->i_format.opcode == daddiu_op) {
364 *frame_size = -ip->i_format.simmediate;
371 static int get_frame_info(struct mips_frame_info *info)
373 bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
374 union mips_instruction insn, *ip;
375 const unsigned int max_insns = 128;
376 unsigned int last_insn_size = 0;
378 bool saw_jump = false;
380 info->pc_offset = -1;
381 info->frame_size = 0;
383 ip = (void *)msk_isa16_mode((ulong)info->func);
387 for (i = 0; i < max_insns; i++) {
388 ip = (void *)ip + last_insn_size;
390 if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
391 insn.word = ip->halfword[0] << 16;
393 } else if (is_mmips) {
394 insn.word = ip->halfword[0] << 16 | ip->halfword[1];
397 insn.word = ip->word;
401 if (!info->frame_size) {
402 is_sp_move_ins(&insn, &info->frame_size);
404 } else if (!saw_jump && is_jump_ins(ip)) {
406 * If we see a jump instruction, we are finished
407 * with the frame save.
409 * Some functions can have a shortcut return at
410 * the beginning of the function, so don't start
411 * looking for jump instruction until we see the
414 * The RA save instruction can get put into the
415 * delay slot of the jump instruction, so look
416 * at the next instruction, too.
421 if (info->pc_offset == -1 &&
422 is_ra_save_ins(&insn, &info->pc_offset))
427 if (info->frame_size && info->pc_offset >= 0) /* nested */
429 if (info->pc_offset < 0) /* leaf */
431 /* prologue seems bogus... */
436 static struct mips_frame_info schedule_mfi __read_mostly;
438 #ifdef CONFIG_KALLSYMS
439 static unsigned long get___schedule_addr(void)
441 return kallsyms_lookup_name("__schedule");
444 static unsigned long get___schedule_addr(void)
446 union mips_instruction *ip = (void *)schedule;
450 for (i = 0; i < max_insns; i++, ip++) {
451 if (ip->j_format.opcode == j_op)
452 return J_TARGET(ip, ip->j_format.target);
458 static int __init frame_info_init(void)
460 unsigned long size = 0;
461 #ifdef CONFIG_KALLSYMS
466 addr = get___schedule_addr();
468 addr = (unsigned long)schedule;
470 #ifdef CONFIG_KALLSYMS
471 kallsyms_lookup_size_offset(addr, &size, &ofs);
473 schedule_mfi.func = (void *)addr;
474 schedule_mfi.func_size = size;
476 get_frame_info(&schedule_mfi);
479 * Without schedule() frame info, result given by
480 * thread_saved_pc() and get_wchan() are not reliable.
482 if (schedule_mfi.pc_offset < 0)
483 printk("Can't analyze schedule() prologue at %p\n", schedule);
488 arch_initcall(frame_info_init);
491 * Return saved PC of a blocked thread.
493 static unsigned long thread_saved_pc(struct task_struct *tsk)
495 struct thread_struct *t = &tsk->thread;
497 /* New born processes are a special case */
498 if (t->reg31 == (unsigned long) ret_from_fork)
500 if (schedule_mfi.pc_offset < 0)
502 return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
506 #ifdef CONFIG_KALLSYMS
507 /* generic stack unwinding function */
508 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
513 unsigned long low, high, irq_stack_high;
514 struct mips_frame_info info;
515 unsigned long size, ofs;
516 struct pt_regs *regs;
523 * IRQ stacks start at IRQ_STACK_START
524 * task stacks at THREAD_SIZE - 32
527 if (!preemptible() && on_irq_stack(raw_smp_processor_id(), *sp)) {
528 high = stack_page + IRQ_STACK_START;
529 irq_stack_high = high;
531 high = stack_page + THREAD_SIZE - 32;
536 * If we reached the top of the interrupt stack, start unwinding
537 * the interrupted task stack.
539 if (unlikely(*sp == irq_stack_high)) {
540 unsigned long task_sp = *(unsigned long *)*sp;
543 * Check that the pointer saved in the IRQ stack head points to
544 * something within the stack of the current task
546 if (!object_is_on_stack((void *)task_sp))
550 * Follow pointer to tasks kernel stack frame where interrupted
553 regs = (struct pt_regs *)task_sp;
555 if (!user_mode(regs) && __kernel_text_address(pc)) {
556 *sp = regs->regs[29];
557 *ra = regs->regs[31];
562 if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
565 * Return ra if an exception occurred at the first instruction
567 if (unlikely(ofs == 0)) {
573 info.func = (void *)(pc - ofs);
574 info.func_size = ofs; /* analyze from start to ofs */
575 leaf = get_frame_info(&info);
579 if (*sp < low || *sp + info.frame_size > high)
584 * For some extreme cases, get_frame_info() can
585 * consider wrongly a nested function as a leaf
586 * one. In that cases avoid to return always the
589 pc = pc != *ra ? *ra : 0;
591 pc = ((unsigned long *)(*sp))[info.pc_offset];
593 *sp += info.frame_size;
595 return __kernel_text_address(pc) ? pc : 0;
597 EXPORT_SYMBOL(unwind_stack_by_address);
599 /* used by show_backtrace() */
600 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
601 unsigned long pc, unsigned long *ra)
603 unsigned long stack_page = 0;
606 for_each_possible_cpu(cpu) {
607 if (on_irq_stack(cpu, *sp)) {
608 stack_page = (unsigned long)irq_stack[cpu];
614 stack_page = (unsigned long)task_stack_page(task);
616 return unwind_stack_by_address(stack_page, sp, pc, ra);
621 * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
623 unsigned long get_wchan(struct task_struct *task)
625 unsigned long pc = 0;
626 #ifdef CONFIG_KALLSYMS
628 unsigned long ra = 0;
631 if (!task || task == current || task->state == TASK_RUNNING)
633 if (!task_stack_page(task))
636 pc = thread_saved_pc(task);
638 #ifdef CONFIG_KALLSYMS
639 sp = task->thread.reg29 + schedule_mfi.frame_size;
641 while (in_sched_functions(pc))
642 pc = unwind_stack(task, &sp, pc, &ra);
649 unsigned long mips_stack_top(void)
651 unsigned long top = TASK_SIZE & PAGE_MASK;
653 /* One page for branch delay slot "emulation" */
656 /* Space for the VDSO, data page & GIC user page */
657 top -= PAGE_ALIGN(current->thread.abi->vdso->size);
659 top -= mips_gic_present() ? PAGE_SIZE : 0;
661 /* Space for cache colour alignment */
662 if (cpu_has_dc_aliases)
663 top -= shm_align_mask + 1;
665 /* Space to randomize the VDSO base */
666 if (current->flags & PF_RANDOMIZE)
667 top -= VDSO_RANDOMIZE_SIZE;
673 * Don't forget that the stack pointer must be aligned on a 8 bytes
674 * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
676 unsigned long arch_align_stack(unsigned long sp)
678 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
679 sp -= get_random_int() & ~PAGE_MASK;
684 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
685 static struct cpumask backtrace_csd_busy;
687 static void handle_backtrace(void *info)
689 nmi_cpu_backtrace(get_irq_regs());
690 cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
693 static void raise_backtrace(cpumask_t *mask)
695 call_single_data_t *csd;
698 for_each_cpu(cpu, mask) {
700 * If we previously sent an IPI to the target CPU & it hasn't
701 * cleared its bit in the busy cpumask then it didn't handle
702 * our previous IPI & it's not safe for us to reuse the
703 * call_single_data_t.
705 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
706 pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
711 csd = &per_cpu(backtrace_csd, cpu);
712 csd->func = handle_backtrace;
713 smp_call_function_single_async(cpu, csd);
717 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
719 nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace);
722 int mips_get_process_fp_mode(struct task_struct *task)
726 if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
727 value |= PR_FP_MODE_FR;
728 if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
729 value |= PR_FP_MODE_FRE;
734 static long prepare_for_fp_mode_switch(void *unused)
737 * This is icky, but we use this to simply ensure that all CPUs have
738 * context switched, regardless of whether they were previously running
739 * kernel or user code. This ensures that no CPU that a mode-switching
740 * program may execute on keeps its FPU enabled (& in the old mode)
741 * throughout the mode switch.
746 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
748 const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
749 struct task_struct *t;
750 struct cpumask process_cpus;
753 /* If nothing to change, return right away, successfully. */
754 if (value == mips_get_process_fp_mode(task))
757 /* Only accept a mode change if 64-bit FP enabled for o32. */
758 if (!IS_ENABLED(CONFIG_MIPS_O32_FP64_SUPPORT))
761 /* And only for o32 tasks. */
762 if (IS_ENABLED(CONFIG_64BIT) && !test_thread_flag(TIF_32BIT_REGS))
765 /* Check the value is valid */
766 if (value & ~known_bits)
769 /* Setting FRE without FR is not supported. */
770 if ((value & (PR_FP_MODE_FR | PR_FP_MODE_FRE)) == PR_FP_MODE_FRE)
773 /* Avoid inadvertently triggering emulation */
774 if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
775 !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
777 if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
780 /* FR = 0 not supported in MIPS R6 */
781 if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
784 /* Indicate the new FP mode in each thread */
785 for_each_thread(task, t) {
786 /* Update desired FP register width */
787 if (value & PR_FP_MODE_FR) {
788 clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
790 set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
791 clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
794 /* Update desired FP single layout */
795 if (value & PR_FP_MODE_FRE)
796 set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
798 clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
802 * We need to ensure that all threads in the process have switched mode
803 * before returning, in order to allow userland to not worry about
804 * races. We can do this by forcing all CPUs that any thread in the
805 * process may be running on to schedule something else - in this case
806 * prepare_for_fp_mode_switch().
808 * We begin by generating a mask of all CPUs that any thread in the
809 * process may be running on.
811 cpumask_clear(&process_cpus);
812 for_each_thread(task, t)
813 cpumask_set_cpu(task_cpu(t), &process_cpus);
816 * Now we schedule prepare_for_fp_mode_switch() on each of those CPUs.
818 * The CPUs may have rescheduled already since we switched mode or
819 * generated the cpumask, but that doesn't matter. If the task in this
820 * process is scheduled out then our scheduling
821 * prepare_for_fp_mode_switch() will simply be redundant. If it's
822 * scheduled in then it will already have picked up the new FP mode
826 for_each_cpu_and(cpu, &process_cpus, cpu_online_mask)
827 work_on_cpu(cpu, prepare_for_fp_mode_switch, NULL);
833 #if defined(CONFIG_32BIT) || defined(CONFIG_MIPS32_O32)
834 void mips_dump_regs32(u32 *uregs, const struct pt_regs *regs)
838 for (i = MIPS32_EF_R1; i <= MIPS32_EF_R31; i++) {
839 /* k0/k1 are copied as zero. */
840 if (i == MIPS32_EF_R26 || i == MIPS32_EF_R27)
843 uregs[i] = regs->regs[i - MIPS32_EF_R0];
846 uregs[MIPS32_EF_LO] = regs->lo;
847 uregs[MIPS32_EF_HI] = regs->hi;
848 uregs[MIPS32_EF_CP0_EPC] = regs->cp0_epc;
849 uregs[MIPS32_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
850 uregs[MIPS32_EF_CP0_STATUS] = regs->cp0_status;
851 uregs[MIPS32_EF_CP0_CAUSE] = regs->cp0_cause;
853 #endif /* CONFIG_32BIT || CONFIG_MIPS32_O32 */
856 void mips_dump_regs64(u64 *uregs, const struct pt_regs *regs)
860 for (i = MIPS64_EF_R1; i <= MIPS64_EF_R31; i++) {
861 /* k0/k1 are copied as zero. */
862 if (i == MIPS64_EF_R26 || i == MIPS64_EF_R27)
865 uregs[i] = regs->regs[i - MIPS64_EF_R0];
868 uregs[MIPS64_EF_LO] = regs->lo;
869 uregs[MIPS64_EF_HI] = regs->hi;
870 uregs[MIPS64_EF_CP0_EPC] = regs->cp0_epc;
871 uregs[MIPS64_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
872 uregs[MIPS64_EF_CP0_STATUS] = regs->cp0_status;
873 uregs[MIPS64_EF_CP0_CAUSE] = regs->cp0_cause;
875 #endif /* CONFIG_64BIT */