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/tick.h>
15 #include <linux/kernel.h>
17 #include <linux/stddef.h>
18 #include <linux/unistd.h>
19 #include <linux/export.h>
20 #include <linux/ptrace.h>
21 #include <linux/mman.h>
22 #include <linux/personality.h>
23 #include <linux/sys.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/kallsyms.h>
27 #include <linux/random.h>
28 #include <linux/prctl.h>
31 #include <asm/bootinfo.h>
33 #include <asm/dsemul.h>
38 #include <asm/pgtable.h>
39 #include <asm/mipsregs.h>
40 #include <asm/processor.h>
42 #include <asm/uaccess.h>
45 #include <asm/isadep.h>
47 #include <asm/stacktrace.h>
48 #include <asm/irq_regs.h>
50 #ifdef CONFIG_HOTPLUG_CPU
51 void arch_cpu_idle_dead(void)
57 asmlinkage void ret_from_fork(void);
58 asmlinkage void ret_from_kernel_thread(void);
60 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
64 /* New thread loses kernel privileges. */
65 status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
67 regs->cp0_status = status;
69 clear_thread_flag(TIF_MSA_CTX_LIVE);
71 atomic_set(¤t->thread.bd_emu_frame, BD_EMUFRAME_NONE);
77 void exit_thread(struct task_struct *tsk)
80 * User threads may have allocated a delay slot emulation frame.
81 * If so, clean up that allocation.
83 if (!(current->flags & PF_KTHREAD))
84 dsemul_thread_cleanup(tsk);
87 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
90 * Save any process state which is live in hardware registers to the
91 * parent context prior to duplication. This prevents the new child
92 * state becoming stale if the parent is preempted before copy_thread()
93 * gets a chance to save the parent's live hardware registers to the
100 else if (is_fpu_owner())
112 * Copy architecture-specific thread state
114 int copy_thread(unsigned long clone_flags, unsigned long usp,
115 unsigned long kthread_arg, struct task_struct *p)
117 struct thread_info *ti = task_thread_info(p);
118 struct pt_regs *childregs, *regs = current_pt_regs();
119 unsigned long childksp;
121 childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
123 /* set up new TSS. */
124 childregs = (struct pt_regs *) childksp - 1;
125 /* Put the stack after the struct pt_regs. */
126 childksp = (unsigned long) childregs;
127 p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
128 if (unlikely(p->flags & PF_KTHREAD)) {
130 unsigned long status = p->thread.cp0_status;
131 memset(childregs, 0, sizeof(struct pt_regs));
132 ti->addr_limit = KERNEL_DS;
133 p->thread.reg16 = usp; /* fn */
134 p->thread.reg17 = kthread_arg;
135 p->thread.reg29 = childksp;
136 p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
137 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
138 status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
139 ((status & (ST0_KUC | ST0_IEC)) << 2);
143 childregs->cp0_status = status;
149 childregs->regs[7] = 0; /* Clear error flag */
150 childregs->regs[2] = 0; /* Child gets zero as return value */
152 childregs->regs[29] = usp;
153 ti->addr_limit = USER_DS;
155 p->thread.reg29 = (unsigned long) childregs;
156 p->thread.reg31 = (unsigned long) ret_from_fork;
159 * New tasks lose permission to use the fpu. This accelerates context
160 * switching for most programs since they don't use the fpu.
162 childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
164 clear_tsk_thread_flag(p, TIF_USEDFPU);
165 clear_tsk_thread_flag(p, TIF_USEDMSA);
166 clear_tsk_thread_flag(p, TIF_MSA_CTX_LIVE);
168 #ifdef CONFIG_MIPS_MT_FPAFF
169 clear_tsk_thread_flag(p, TIF_FPUBOUND);
170 #endif /* CONFIG_MIPS_MT_FPAFF */
172 atomic_set(&p->thread.bd_emu_frame, BD_EMUFRAME_NONE);
174 if (clone_flags & CLONE_SETTLS)
175 ti->tp_value = regs->regs[7];
180 #ifdef CONFIG_CC_STACKPROTECTOR
181 #include <linux/stackprotector.h>
182 unsigned long __stack_chk_guard __read_mostly;
183 EXPORT_SYMBOL(__stack_chk_guard);
186 struct mips_frame_info {
188 unsigned long func_size;
193 #define J_TARGET(pc,target) \
194 (((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
196 static inline int is_ra_save_ins(union mips_instruction *ip, int *poff)
198 #ifdef CONFIG_CPU_MICROMIPS
201 * swm16 reglist,offset(sp)
202 * swm32 reglist,offset(sp)
204 * jradiussp - NOT SUPPORTED
206 * microMIPS is way more fun...
208 if (mm_insn_16bit(ip->halfword[1])) {
209 switch (ip->mm16_r5_format.opcode) {
211 if (ip->mm16_r5_format.rt != 31)
214 *poff = ip->mm16_r5_format.simmediate;
215 *poff = (*poff << 2) / sizeof(ulong);
219 switch (ip->mm16_m_format.func) {
221 *poff = ip->mm16_m_format.imm;
222 *poff += 1 + ip->mm16_m_format.rlist;
223 *poff = (*poff << 2) / sizeof(ulong);
235 switch (ip->i_format.opcode) {
237 if (ip->i_format.rs != 29)
239 if (ip->i_format.rt != 31)
242 *poff = ip->i_format.simmediate / sizeof(ulong);
246 switch (ip->mm_m_format.func) {
248 if (ip->mm_m_format.rd < 0x10)
250 if (ip->mm_m_format.base != 29)
253 *poff = ip->mm_m_format.simmediate;
254 *poff += (ip->mm_m_format.rd & 0xf) * sizeof(u32);
255 *poff /= sizeof(ulong);
265 /* sw / sd $ra, offset($sp) */
266 if ((ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
267 ip->i_format.rs == 29 && ip->i_format.rt == 31) {
268 *poff = ip->i_format.simmediate / sizeof(ulong);
276 static inline int is_jump_ins(union mips_instruction *ip)
278 #ifdef CONFIG_CPU_MICROMIPS
280 * jr16,jrc,jalr16,jalr16
282 * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
283 * jraddiusp - NOT SUPPORTED
285 * microMIPS is kind of more fun...
287 if (mm_insn_16bit(ip->halfword[1])) {
288 if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
289 (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op))
294 if (ip->j_format.opcode == mm_j32_op)
296 if (ip->j_format.opcode == mm_jal32_op)
298 if (ip->r_format.opcode != mm_pool32a_op ||
299 ip->r_format.func != mm_pool32axf_op)
301 return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
303 if (ip->j_format.opcode == j_op)
305 if (ip->j_format.opcode == jal_op)
307 if (ip->r_format.opcode != spec_op)
309 return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
313 static inline int is_sp_move_ins(union mips_instruction *ip)
315 #ifdef CONFIG_CPU_MICROMIPS
320 * jradiussp - NOT SUPPORTED
322 * microMIPS is not more fun...
324 if (mm_insn_16bit(ip->halfword[1])) {
325 return (ip->mm16_r3_format.opcode == mm_pool16d_op &&
326 ip->mm16_r3_format.simmediate && mm_addiusp_func) ||
327 (ip->mm16_r5_format.opcode == mm_pool16d_op &&
328 ip->mm16_r5_format.rt == 29);
331 return ip->mm_i_format.opcode == mm_addiu32_op &&
332 ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29;
334 /* addiu/daddiu sp,sp,-imm */
335 if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
337 if (ip->i_format.opcode == addiu_op || ip->i_format.opcode == daddiu_op)
343 static int get_frame_info(struct mips_frame_info *info)
345 bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
346 union mips_instruction insn, *ip, *ip_end;
347 const unsigned int max_insns = 128;
350 info->pc_offset = -1;
351 info->frame_size = 0;
353 ip = (void *)msk_isa16_mode((ulong)info->func);
357 ip_end = (void *)ip + info->func_size;
359 for (i = 0; i < max_insns && ip < ip_end; i++, ip++) {
360 if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
361 insn.halfword[0] = 0;
362 insn.halfword[1] = ip->halfword[0];
363 } else if (is_mmips) {
364 insn.halfword[0] = ip->halfword[1];
365 insn.halfword[1] = ip->halfword[0];
367 insn.word = ip->word;
370 if (is_jump_ins(&insn))
373 if (!info->frame_size) {
374 if (is_sp_move_ins(&insn))
376 #ifdef CONFIG_CPU_MICROMIPS
377 if (mm_insn_16bit(ip->halfword[0]))
381 if (ip->halfword[0] & mm_addiusp_func)
383 tmp = (((ip->halfword[0] >> 1) & 0x1ff) << 2);
384 info->frame_size = -(signed short)(tmp | ((tmp & 0x100) ? 0xfe00 : 0));
386 tmp = (ip->halfword[0] >> 1);
387 info->frame_size = -(signed short)(tmp & 0xf);
389 ip = (void *) &ip->halfword[1];
393 info->frame_size = - ip->i_format.simmediate;
397 if (info->pc_offset == -1 &&
398 is_ra_save_ins(&insn, &info->pc_offset))
401 if (info->frame_size && info->pc_offset >= 0) /* nested */
403 if (info->pc_offset < 0) /* leaf */
405 /* prologue seems boggus... */
410 static struct mips_frame_info schedule_mfi __read_mostly;
412 #ifdef CONFIG_KALLSYMS
413 static unsigned long get___schedule_addr(void)
415 return kallsyms_lookup_name("__schedule");
418 static unsigned long get___schedule_addr(void)
420 union mips_instruction *ip = (void *)schedule;
424 for (i = 0; i < max_insns; i++, ip++) {
425 if (ip->j_format.opcode == j_op)
426 return J_TARGET(ip, ip->j_format.target);
432 static int __init frame_info_init(void)
434 unsigned long size = 0;
435 #ifdef CONFIG_KALLSYMS
440 addr = get___schedule_addr();
442 addr = (unsigned long)schedule;
444 #ifdef CONFIG_KALLSYMS
445 kallsyms_lookup_size_offset(addr, &size, &ofs);
447 schedule_mfi.func = (void *)addr;
448 schedule_mfi.func_size = size;
450 get_frame_info(&schedule_mfi);
453 * Without schedule() frame info, result given by
454 * thread_saved_pc() and get_wchan() are not reliable.
456 if (schedule_mfi.pc_offset < 0)
457 printk("Can't analyze schedule() prologue at %p\n", schedule);
462 arch_initcall(frame_info_init);
465 * Return saved PC of a blocked thread.
467 unsigned long thread_saved_pc(struct task_struct *tsk)
469 struct thread_struct *t = &tsk->thread;
471 /* New born processes are a special case */
472 if (t->reg31 == (unsigned long) ret_from_fork)
474 if (schedule_mfi.pc_offset < 0)
476 return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
480 #ifdef CONFIG_KALLSYMS
481 /* generic stack unwinding function */
482 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
487 unsigned long low, high, irq_stack_high;
488 struct mips_frame_info info;
489 unsigned long size, ofs;
490 struct pt_regs *regs;
497 * IRQ stacks start at IRQ_STACK_START
498 * task stacks at THREAD_SIZE - 32
501 if (!preemptible() && on_irq_stack(raw_smp_processor_id(), *sp)) {
502 high = stack_page + IRQ_STACK_START;
503 irq_stack_high = high;
505 high = stack_page + THREAD_SIZE - 32;
510 * If we reached the top of the interrupt stack, start unwinding
511 * the interrupted task stack.
513 if (unlikely(*sp == irq_stack_high)) {
514 unsigned long task_sp = *(unsigned long *)*sp;
517 * Check that the pointer saved in the IRQ stack head points to
518 * something within the stack of the current task
520 if (!object_is_on_stack((void *)task_sp))
524 * Follow pointer to tasks kernel stack frame where interrupted
527 regs = (struct pt_regs *)task_sp;
529 if (!user_mode(regs) && __kernel_text_address(pc)) {
530 *sp = regs->regs[29];
531 *ra = regs->regs[31];
536 if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
539 * Return ra if an exception occurred at the first instruction
541 if (unlikely(ofs == 0)) {
547 info.func = (void *)(pc - ofs);
548 info.func_size = ofs; /* analyze from start to ofs */
549 leaf = get_frame_info(&info);
553 if (*sp < low || *sp + info.frame_size > high)
558 * For some extreme cases, get_frame_info() can
559 * consider wrongly a nested function as a leaf
560 * one. In that cases avoid to return always the
563 pc = pc != *ra ? *ra : 0;
565 pc = ((unsigned long *)(*sp))[info.pc_offset];
567 *sp += info.frame_size;
569 return __kernel_text_address(pc) ? pc : 0;
571 EXPORT_SYMBOL(unwind_stack_by_address);
573 /* used by show_backtrace() */
574 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
575 unsigned long pc, unsigned long *ra)
577 unsigned long stack_page = 0;
580 for_each_possible_cpu(cpu) {
581 if (on_irq_stack(cpu, *sp)) {
582 stack_page = (unsigned long)irq_stack[cpu];
588 stack_page = (unsigned long)task_stack_page(task);
590 return unwind_stack_by_address(stack_page, sp, pc, ra);
595 * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
597 unsigned long get_wchan(struct task_struct *task)
599 unsigned long pc = 0;
600 #ifdef CONFIG_KALLSYMS
602 unsigned long ra = 0;
605 if (!task || task == current || task->state == TASK_RUNNING)
607 if (!task_stack_page(task))
610 pc = thread_saved_pc(task);
612 #ifdef CONFIG_KALLSYMS
613 sp = task->thread.reg29 + schedule_mfi.frame_size;
615 while (in_sched_functions(pc))
616 pc = unwind_stack(task, &sp, pc, &ra);
624 * Don't forget that the stack pointer must be aligned on a 8 bytes
625 * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
627 unsigned long arch_align_stack(unsigned long sp)
629 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
630 sp -= get_random_int() & ~PAGE_MASK;
635 static DEFINE_PER_CPU(struct call_single_data, backtrace_csd);
636 static struct cpumask backtrace_csd_busy;
638 static void arch_dump_stack(void *info)
640 struct pt_regs *regs;
641 static arch_spinlock_t lock = __ARCH_SPIN_LOCK_UNLOCKED;
643 arch_spin_lock(&lock);
644 regs = get_irq_regs();
650 arch_spin_unlock(&lock);
652 cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
655 void arch_trigger_all_cpu_backtrace(bool include_self)
657 struct call_single_data *csd;
660 for_each_cpu(cpu, cpu_online_mask) {
662 * If we previously sent an IPI to the target CPU & it hasn't
663 * cleared its bit in the busy cpumask then it didn't handle
664 * our previous IPI & it's not safe for us to reuse the
665 * call_single_data_t.
667 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
668 pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
673 csd = &per_cpu(backtrace_csd, cpu);
674 csd->func = arch_dump_stack;
675 smp_call_function_single_async(cpu, csd);
679 int mips_get_process_fp_mode(struct task_struct *task)
683 if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
684 value |= PR_FP_MODE_FR;
685 if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
686 value |= PR_FP_MODE_FRE;
691 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
693 const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
694 unsigned long switch_count;
695 struct task_struct *t;
697 /* If nothing to change, return right away, successfully. */
698 if (value == mips_get_process_fp_mode(task))
701 /* Only accept a mode change if 64-bit FP enabled for o32. */
702 if (!IS_ENABLED(CONFIG_MIPS_O32_FP64_SUPPORT))
705 /* And only for o32 tasks. */
706 if (IS_ENABLED(CONFIG_64BIT) && !test_thread_flag(TIF_32BIT_REGS))
709 /* Check the value is valid */
710 if (value & ~known_bits)
713 /* Setting FRE without FR is not supported. */
714 if ((value & (PR_FP_MODE_FR | PR_FP_MODE_FRE)) == PR_FP_MODE_FRE)
717 /* Avoid inadvertently triggering emulation */
718 if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
719 !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
721 if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
724 /* FR = 0 not supported in MIPS R6 */
725 if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
728 /* Proceed with the mode switch */
731 /* Save FP & vector context, then disable FPU & MSA */
732 if (task->signal == current->signal)
735 /* Prevent any threads from obtaining live FP context */
736 atomic_set(&task->mm->context.fp_mode_switching, 1);
737 smp_mb__after_atomic();
740 * If there are multiple online CPUs then wait until all threads whose
741 * FP mode is about to change have been context switched. This approach
742 * allows us to only worry about whether an FP mode switch is in
743 * progress when FP is first used in a tasks time slice. Pretty much all
744 * of the mode switch overhead can thus be confined to cases where mode
745 * switches are actually occurring. That is, to here. However for the
746 * thread performing the mode switch it may take a while...
748 if (num_online_cpus() > 1) {
749 spin_lock_irq(&task->sighand->siglock);
751 for_each_thread(task, t) {
755 switch_count = t->nvcsw + t->nivcsw;
758 spin_unlock_irq(&task->sighand->siglock);
760 spin_lock_irq(&task->sighand->siglock);
761 } while ((t->nvcsw + t->nivcsw) == switch_count);
764 spin_unlock_irq(&task->sighand->siglock);
768 * There are now no threads of the process with live FP context, so it
769 * is safe to proceed with the FP mode switch.
771 for_each_thread(task, t) {
772 /* Update desired FP register width */
773 if (value & PR_FP_MODE_FR) {
774 clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
776 set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
777 clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
780 /* Update desired FP single layout */
781 if (value & PR_FP_MODE_FRE)
782 set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
784 clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
787 /* Allow threads to use FP again */
788 atomic_set(&task->mm->context.fp_mode_switching, 0);