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>
33 #include <linux/cpu.h>
37 #include <asm/bootinfo.h>
39 #include <asm/dsemul.h>
43 #include <asm/mips-cps.h>
45 #include <asm/mipsregs.h>
46 #include <asm/processor.h>
48 #include <linux/uaccess.h>
51 #include <asm/isadep.h>
53 #include <asm/stacktrace.h>
54 #include <asm/irq_regs.h>
56 #ifdef CONFIG_HOTPLUG_CPU
57 void arch_cpu_idle_dead(void)
63 asmlinkage void ret_from_fork(void);
64 asmlinkage void ret_from_kernel_thread(void);
66 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
70 /* New thread loses kernel privileges. */
71 status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
73 regs->cp0_status = status;
75 clear_thread_flag(TIF_MSA_CTX_LIVE);
77 #ifdef CONFIG_MIPS_FP_SUPPORT
78 atomic_set(¤t->thread.bd_emu_frame, BD_EMUFRAME_NONE);
85 void exit_thread(struct task_struct *tsk)
88 * User threads may have allocated a delay slot emulation frame.
89 * If so, clean up that allocation.
91 if (!(current->flags & PF_KTHREAD))
92 dsemul_thread_cleanup(tsk);
95 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
98 * Save any process state which is live in hardware registers to the
99 * parent context prior to duplication. This prevents the new child
100 * state becoming stale if the parent is preempted before copy_thread()
101 * gets a chance to save the parent's live hardware registers to the
106 if (is_msa_enabled())
108 else if (is_fpu_owner())
120 * Copy architecture-specific thread state
122 int copy_thread_tls(unsigned long clone_flags, unsigned long usp,
123 unsigned long kthread_arg, struct task_struct *p, unsigned long tls)
125 struct thread_info *ti = task_thread_info(p);
126 struct pt_regs *childregs, *regs = current_pt_regs();
127 unsigned long childksp;
129 childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
131 /* set up new TSS. */
132 childregs = (struct pt_regs *) childksp - 1;
133 /* Put the stack after the struct pt_regs. */
134 childksp = (unsigned long) childregs;
135 p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
136 if (unlikely(p->flags & PF_KTHREAD)) {
138 unsigned long status = p->thread.cp0_status;
139 memset(childregs, 0, sizeof(struct pt_regs));
140 ti->addr_limit = KERNEL_DS;
141 p->thread.reg16 = usp; /* fn */
142 p->thread.reg17 = kthread_arg;
143 p->thread.reg29 = childksp;
144 p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
145 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
146 status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
147 ((status & (ST0_KUC | ST0_IEC)) << 2);
151 childregs->cp0_status = status;
157 childregs->regs[7] = 0; /* Clear error flag */
158 childregs->regs[2] = 0; /* Child gets zero as return value */
160 childregs->regs[29] = usp;
161 ti->addr_limit = USER_DS;
163 p->thread.reg29 = (unsigned long) childregs;
164 p->thread.reg31 = (unsigned long) ret_from_fork;
167 * New tasks lose permission to use the fpu. This accelerates context
168 * switching for most programs since they don't use the fpu.
170 childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
172 clear_tsk_thread_flag(p, TIF_USEDFPU);
173 clear_tsk_thread_flag(p, TIF_USEDMSA);
174 clear_tsk_thread_flag(p, TIF_MSA_CTX_LIVE);
176 #ifdef CONFIG_MIPS_MT_FPAFF
177 clear_tsk_thread_flag(p, TIF_FPUBOUND);
178 #endif /* CONFIG_MIPS_MT_FPAFF */
180 #ifdef CONFIG_MIPS_FP_SUPPORT
181 atomic_set(&p->thread.bd_emu_frame, BD_EMUFRAME_NONE);
184 if (clone_flags & CLONE_SETTLS)
190 #ifdef CONFIG_STACKPROTECTOR
191 #include <linux/stackprotector.h>
192 unsigned long __stack_chk_guard __read_mostly;
193 EXPORT_SYMBOL(__stack_chk_guard);
196 struct mips_frame_info {
198 unsigned long func_size;
203 #define J_TARGET(pc,target) \
204 (((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
206 static inline int is_ra_save_ins(union mips_instruction *ip, int *poff)
208 #ifdef CONFIG_CPU_MICROMIPS
211 * swm16 reglist,offset(sp)
212 * swm32 reglist,offset(sp)
214 * jradiussp - NOT SUPPORTED
216 * microMIPS is way more fun...
218 if (mm_insn_16bit(ip->word >> 16)) {
219 switch (ip->mm16_r5_format.opcode) {
221 if (ip->mm16_r5_format.rt != 31)
224 *poff = ip->mm16_r5_format.imm;
225 *poff = (*poff << 2) / sizeof(ulong);
229 switch (ip->mm16_m_format.func) {
231 *poff = ip->mm16_m_format.imm;
232 *poff += 1 + ip->mm16_m_format.rlist;
233 *poff = (*poff << 2) / sizeof(ulong);
245 switch (ip->i_format.opcode) {
247 if (ip->i_format.rs != 29)
249 if (ip->i_format.rt != 31)
252 *poff = ip->i_format.simmediate / sizeof(ulong);
256 switch (ip->mm_m_format.func) {
258 if (ip->mm_m_format.rd < 0x10)
260 if (ip->mm_m_format.base != 29)
263 *poff = ip->mm_m_format.simmediate;
264 *poff += (ip->mm_m_format.rd & 0xf) * sizeof(u32);
265 *poff /= sizeof(ulong);
275 /* sw / sd $ra, offset($sp) */
276 if ((ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
277 ip->i_format.rs == 29 && ip->i_format.rt == 31) {
278 *poff = ip->i_format.simmediate / sizeof(ulong);
286 static inline int is_jump_ins(union mips_instruction *ip)
288 #ifdef CONFIG_CPU_MICROMIPS
290 * jr16,jrc,jalr16,jalr16
292 * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
293 * jraddiusp - NOT SUPPORTED
295 * microMIPS is kind of more fun...
297 if (mm_insn_16bit(ip->word >> 16)) {
298 if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
299 (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op))
304 if (ip->j_format.opcode == mm_j32_op)
306 if (ip->j_format.opcode == mm_jal32_op)
308 if (ip->r_format.opcode != mm_pool32a_op ||
309 ip->r_format.func != mm_pool32axf_op)
311 return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
313 if (ip->j_format.opcode == j_op)
315 if (ip->j_format.opcode == jal_op)
317 if (ip->r_format.opcode != spec_op)
319 return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
323 static inline int is_sp_move_ins(union mips_instruction *ip, int *frame_size)
325 #ifdef CONFIG_CPU_MICROMIPS
332 * jradiussp - NOT SUPPORTED
334 * microMIPS is not more fun...
336 if (mm_insn_16bit(ip->word >> 16)) {
337 if (ip->mm16_r3_format.opcode == mm_pool16d_op &&
338 ip->mm16_r3_format.simmediate & mm_addiusp_func) {
339 tmp = ip->mm_b0_format.simmediate >> 1;
340 tmp = ((tmp & 0x1ff) ^ 0x100) - 0x100;
341 if ((tmp + 2) < 4) /* 0x0,0x1,0x1fe,0x1ff are special */
343 *frame_size = -(signed short)(tmp << 2);
346 if (ip->mm16_r5_format.opcode == mm_pool16d_op &&
347 ip->mm16_r5_format.rt == 29) {
348 tmp = ip->mm16_r5_format.imm >> 1;
349 *frame_size = -(signed short)(tmp & 0xf);
355 if (ip->mm_i_format.opcode == mm_addiu32_op &&
356 ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29) {
357 *frame_size = -ip->i_format.simmediate;
361 /* addiu/daddiu sp,sp,-imm */
362 if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
365 if (ip->i_format.opcode == addiu_op ||
366 ip->i_format.opcode == daddiu_op) {
367 *frame_size = -ip->i_format.simmediate;
374 static int get_frame_info(struct mips_frame_info *info)
376 bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
377 union mips_instruction insn, *ip;
378 const unsigned int max_insns = 128;
379 unsigned int last_insn_size = 0;
381 bool saw_jump = false;
383 info->pc_offset = -1;
384 info->frame_size = 0;
386 ip = (void *)msk_isa16_mode((ulong)info->func);
390 for (i = 0; i < max_insns; i++) {
391 ip = (void *)ip + last_insn_size;
393 if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
394 insn.word = ip->halfword[0] << 16;
396 } else if (is_mmips) {
397 insn.word = ip->halfword[0] << 16 | ip->halfword[1];
400 insn.word = ip->word;
404 if (!info->frame_size) {
405 is_sp_move_ins(&insn, &info->frame_size);
407 } else if (!saw_jump && is_jump_ins(ip)) {
409 * If we see a jump instruction, we are finished
410 * with the frame save.
412 * Some functions can have a shortcut return at
413 * the beginning of the function, so don't start
414 * looking for jump instruction until we see the
417 * The RA save instruction can get put into the
418 * delay slot of the jump instruction, so look
419 * at the next instruction, too.
424 if (info->pc_offset == -1 &&
425 is_ra_save_ins(&insn, &info->pc_offset))
430 if (info->frame_size && info->pc_offset >= 0) /* nested */
432 if (info->pc_offset < 0) /* leaf */
434 /* prologue seems bogus... */
439 static struct mips_frame_info schedule_mfi __read_mostly;
441 #ifdef CONFIG_KALLSYMS
442 static unsigned long get___schedule_addr(void)
444 return kallsyms_lookup_name("__schedule");
447 static unsigned long get___schedule_addr(void)
449 union mips_instruction *ip = (void *)schedule;
453 for (i = 0; i < max_insns; i++, ip++) {
454 if (ip->j_format.opcode == j_op)
455 return J_TARGET(ip, ip->j_format.target);
461 static int __init frame_info_init(void)
463 unsigned long size = 0;
464 #ifdef CONFIG_KALLSYMS
469 addr = get___schedule_addr();
471 addr = (unsigned long)schedule;
473 #ifdef CONFIG_KALLSYMS
474 kallsyms_lookup_size_offset(addr, &size, &ofs);
476 schedule_mfi.func = (void *)addr;
477 schedule_mfi.func_size = size;
479 get_frame_info(&schedule_mfi);
482 * Without schedule() frame info, result given by
483 * thread_saved_pc() and get_wchan() are not reliable.
485 if (schedule_mfi.pc_offset < 0)
486 printk("Can't analyze schedule() prologue at %p\n", schedule);
491 arch_initcall(frame_info_init);
494 * Return saved PC of a blocked thread.
496 static unsigned long thread_saved_pc(struct task_struct *tsk)
498 struct thread_struct *t = &tsk->thread;
500 /* New born processes are a special case */
501 if (t->reg31 == (unsigned long) ret_from_fork)
503 if (schedule_mfi.pc_offset < 0)
505 return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
509 #ifdef CONFIG_KALLSYMS
510 /* generic stack unwinding function */
511 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
516 unsigned long low, high, irq_stack_high;
517 struct mips_frame_info info;
518 unsigned long size, ofs;
519 struct pt_regs *regs;
526 * IRQ stacks start at IRQ_STACK_START
527 * task stacks at THREAD_SIZE - 32
530 if (!preemptible() && on_irq_stack(raw_smp_processor_id(), *sp)) {
531 high = stack_page + IRQ_STACK_START;
532 irq_stack_high = high;
534 high = stack_page + THREAD_SIZE - 32;
539 * If we reached the top of the interrupt stack, start unwinding
540 * the interrupted task stack.
542 if (unlikely(*sp == irq_stack_high)) {
543 unsigned long task_sp = *(unsigned long *)*sp;
546 * Check that the pointer saved in the IRQ stack head points to
547 * something within the stack of the current task
549 if (!object_is_on_stack((void *)task_sp))
553 * Follow pointer to tasks kernel stack frame where interrupted
556 regs = (struct pt_regs *)task_sp;
558 if (!user_mode(regs) && __kernel_text_address(pc)) {
559 *sp = regs->regs[29];
560 *ra = regs->regs[31];
565 if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
568 * Return ra if an exception occurred at the first instruction
570 if (unlikely(ofs == 0)) {
576 info.func = (void *)(pc - ofs);
577 info.func_size = ofs; /* analyze from start to ofs */
578 leaf = get_frame_info(&info);
582 if (*sp < low || *sp + info.frame_size > high)
587 * For some extreme cases, get_frame_info() can
588 * consider wrongly a nested function as a leaf
589 * one. In that cases avoid to return always the
592 pc = pc != *ra ? *ra : 0;
594 pc = ((unsigned long *)(*sp))[info.pc_offset];
596 *sp += info.frame_size;
598 return __kernel_text_address(pc) ? pc : 0;
600 EXPORT_SYMBOL(unwind_stack_by_address);
602 /* used by show_backtrace() */
603 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
604 unsigned long pc, unsigned long *ra)
606 unsigned long stack_page = 0;
609 for_each_possible_cpu(cpu) {
610 if (on_irq_stack(cpu, *sp)) {
611 stack_page = (unsigned long)irq_stack[cpu];
617 stack_page = (unsigned long)task_stack_page(task);
619 return unwind_stack_by_address(stack_page, sp, pc, ra);
624 * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
626 unsigned long get_wchan(struct task_struct *task)
628 unsigned long pc = 0;
629 #ifdef CONFIG_KALLSYMS
631 unsigned long ra = 0;
634 if (!task || task == current || task->state == TASK_RUNNING)
636 if (!task_stack_page(task))
639 pc = thread_saved_pc(task);
641 #ifdef CONFIG_KALLSYMS
642 sp = task->thread.reg29 + schedule_mfi.frame_size;
644 while (in_sched_functions(pc))
645 pc = unwind_stack(task, &sp, pc, &ra);
652 unsigned long mips_stack_top(void)
654 unsigned long top = TASK_SIZE & PAGE_MASK;
656 if (IS_ENABLED(CONFIG_MIPS_FP_SUPPORT)) {
657 /* One page for branch delay slot "emulation" */
661 /* Space for the VDSO, data page & GIC user page */
662 top -= PAGE_ALIGN(current->thread.abi->vdso->size);
664 top -= mips_gic_present() ? PAGE_SIZE : 0;
666 /* Space for cache colour alignment */
667 if (cpu_has_dc_aliases)
668 top -= shm_align_mask + 1;
670 /* Space to randomize the VDSO base */
671 if (current->flags & PF_RANDOMIZE)
672 top -= VDSO_RANDOMIZE_SIZE;
678 * Don't forget that the stack pointer must be aligned on a 8 bytes
679 * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
681 unsigned long arch_align_stack(unsigned long sp)
683 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
684 sp -= get_random_int() & ~PAGE_MASK;
689 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
690 static struct cpumask backtrace_csd_busy;
692 static void handle_backtrace(void *info)
694 nmi_cpu_backtrace(get_irq_regs());
695 cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
698 static void raise_backtrace(cpumask_t *mask)
700 call_single_data_t *csd;
703 for_each_cpu(cpu, mask) {
705 * If we previously sent an IPI to the target CPU & it hasn't
706 * cleared its bit in the busy cpumask then it didn't handle
707 * our previous IPI & it's not safe for us to reuse the
708 * call_single_data_t.
710 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
711 pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
716 csd = &per_cpu(backtrace_csd, cpu);
717 csd->func = handle_backtrace;
718 smp_call_function_single_async(cpu, csd);
722 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
724 nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace);
727 int mips_get_process_fp_mode(struct task_struct *task)
731 if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
732 value |= PR_FP_MODE_FR;
733 if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
734 value |= PR_FP_MODE_FRE;
739 static long prepare_for_fp_mode_switch(void *unused)
742 * This is icky, but we use this to simply ensure that all CPUs have
743 * context switched, regardless of whether they were previously running
744 * kernel or user code. This ensures that no CPU that a mode-switching
745 * program may execute on keeps its FPU enabled (& in the old mode)
746 * throughout the mode switch.
751 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
753 const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
754 struct task_struct *t;
755 struct cpumask process_cpus;
758 /* If nothing to change, return right away, successfully. */
759 if (value == mips_get_process_fp_mode(task))
762 /* Only accept a mode change if 64-bit FP enabled for o32. */
763 if (!IS_ENABLED(CONFIG_MIPS_O32_FP64_SUPPORT))
766 /* And only for o32 tasks. */
767 if (IS_ENABLED(CONFIG_64BIT) && !test_thread_flag(TIF_32BIT_REGS))
770 /* Check the value is valid */
771 if (value & ~known_bits)
774 /* Setting FRE without FR is not supported. */
775 if ((value & (PR_FP_MODE_FR | PR_FP_MODE_FRE)) == PR_FP_MODE_FRE)
778 /* Avoid inadvertently triggering emulation */
779 if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
780 !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
782 if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
785 /* FR = 0 not supported in MIPS R6 */
786 if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
789 /* Indicate the new FP mode in each thread */
790 for_each_thread(task, t) {
791 /* Update desired FP register width */
792 if (value & PR_FP_MODE_FR) {
793 clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
795 set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
796 clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
799 /* Update desired FP single layout */
800 if (value & PR_FP_MODE_FRE)
801 set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
803 clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
807 * We need to ensure that all threads in the process have switched mode
808 * before returning, in order to allow userland to not worry about
809 * races. We can do this by forcing all CPUs that any thread in the
810 * process may be running on to schedule something else - in this case
811 * prepare_for_fp_mode_switch().
813 * We begin by generating a mask of all CPUs that any thread in the
814 * process may be running on.
816 cpumask_clear(&process_cpus);
817 for_each_thread(task, t)
818 cpumask_set_cpu(task_cpu(t), &process_cpus);
821 * Now we schedule prepare_for_fp_mode_switch() on each of those CPUs.
823 * The CPUs may have rescheduled already since we switched mode or
824 * generated the cpumask, but that doesn't matter. If the task in this
825 * process is scheduled out then our scheduling
826 * prepare_for_fp_mode_switch() will simply be redundant. If it's
827 * scheduled in then it will already have picked up the new FP mode
831 for_each_cpu_and(cpu, &process_cpus, cpu_online_mask)
832 work_on_cpu(cpu, prepare_for_fp_mode_switch, NULL);
838 #if defined(CONFIG_32BIT) || defined(CONFIG_MIPS32_O32)
839 void mips_dump_regs32(u32 *uregs, const struct pt_regs *regs)
843 for (i = MIPS32_EF_R1; i <= MIPS32_EF_R31; i++) {
844 /* k0/k1 are copied as zero. */
845 if (i == MIPS32_EF_R26 || i == MIPS32_EF_R27)
848 uregs[i] = regs->regs[i - MIPS32_EF_R0];
851 uregs[MIPS32_EF_LO] = regs->lo;
852 uregs[MIPS32_EF_HI] = regs->hi;
853 uregs[MIPS32_EF_CP0_EPC] = regs->cp0_epc;
854 uregs[MIPS32_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
855 uregs[MIPS32_EF_CP0_STATUS] = regs->cp0_status;
856 uregs[MIPS32_EF_CP0_CAUSE] = regs->cp0_cause;
858 #endif /* CONFIG_32BIT || CONFIG_MIPS32_O32 */
861 void mips_dump_regs64(u64 *uregs, const struct pt_regs *regs)
865 for (i = MIPS64_EF_R1; i <= MIPS64_EF_R31; i++) {
866 /* k0/k1 are copied as zero. */
867 if (i == MIPS64_EF_R26 || i == MIPS64_EF_R27)
870 uregs[i] = regs->regs[i - MIPS64_EF_R0];
873 uregs[MIPS64_EF_LO] = regs->lo;
874 uregs[MIPS64_EF_HI] = regs->hi;
875 uregs[MIPS64_EF_CP0_EPC] = regs->cp0_epc;
876 uregs[MIPS64_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
877 uregs[MIPS64_EF_CP0_STATUS] = regs->cp0_status;
878 uregs[MIPS64_EF_CP0_CAUSE] = regs->cp0_cause;
880 #endif /* CONFIG_64BIT */