cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

process.c (22850B)


      1/*
      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
      4 * for more details.
      5 *
      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.
     11 */
     12#include <linux/cpu.h>
     13#include <linux/errno.h>
     14#include <linux/init.h>
     15#include <linux/kallsyms.h>
     16#include <linux/kernel.h>
     17#include <linux/nmi.h>
     18#include <linux/personality.h>
     19#include <linux/prctl.h>
     20#include <linux/random.h>
     21#include <linux/sched.h>
     22#include <linux/sched/debug.h>
     23#include <linux/sched/task_stack.h>
     24
     25#include <asm/abi.h>
     26#include <asm/asm.h>
     27#include <asm/dsemul.h>
     28#include <asm/dsp.h>
     29#include <asm/exec.h>
     30#include <asm/fpu.h>
     31#include <asm/inst.h>
     32#include <asm/irq.h>
     33#include <asm/irq_regs.h>
     34#include <asm/isadep.h>
     35#include <asm/msa.h>
     36#include <asm/mips-cps.h>
     37#include <asm/mipsregs.h>
     38#include <asm/processor.h>
     39#include <asm/reg.h>
     40#include <asm/stacktrace.h>
     41
     42#ifdef CONFIG_HOTPLUG_CPU
     43void arch_cpu_idle_dead(void)
     44{
     45	play_dead();
     46}
     47#endif
     48
     49asmlinkage void ret_from_fork(void);
     50asmlinkage void ret_from_kernel_thread(void);
     51
     52void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
     53{
     54	unsigned long status;
     55
     56	/* New thread loses kernel privileges. */
     57	status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_CU2|ST0_FR|KU_MASK);
     58	status |= KU_USER;
     59	regs->cp0_status = status;
     60	lose_fpu(0);
     61	clear_thread_flag(TIF_MSA_CTX_LIVE);
     62	clear_used_math();
     63#ifdef CONFIG_MIPS_FP_SUPPORT
     64	atomic_set(&current->thread.bd_emu_frame, BD_EMUFRAME_NONE);
     65#endif
     66	init_dsp();
     67	regs->cp0_epc = pc;
     68	regs->regs[29] = sp;
     69}
     70
     71void exit_thread(struct task_struct *tsk)
     72{
     73	/*
     74	 * User threads may have allocated a delay slot emulation frame.
     75	 * If so, clean up that allocation.
     76	 */
     77	if (!(current->flags & PF_KTHREAD))
     78		dsemul_thread_cleanup(tsk);
     79}
     80
     81int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
     82{
     83	/*
     84	 * Save any process state which is live in hardware registers to the
     85	 * parent context prior to duplication. This prevents the new child
     86	 * state becoming stale if the parent is preempted before copy_thread()
     87	 * gets a chance to save the parent's live hardware registers to the
     88	 * child context.
     89	 */
     90	preempt_disable();
     91
     92	if (is_msa_enabled())
     93		save_msa(current);
     94	else if (is_fpu_owner())
     95		_save_fp(current);
     96
     97	save_dsp(current);
     98
     99	preempt_enable();
    100
    101	*dst = *src;
    102	return 0;
    103}
    104
    105/*
    106 * Copy architecture-specific thread state
    107 */
    108int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
    109{
    110	unsigned long clone_flags = args->flags;
    111	unsigned long usp = args->stack;
    112	unsigned long tls = args->tls;
    113	struct thread_info *ti = task_thread_info(p);
    114	struct pt_regs *childregs, *regs = current_pt_regs();
    115	unsigned long childksp;
    116
    117	childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
    118
    119	/* set up new TSS. */
    120	childregs = (struct pt_regs *) childksp - 1;
    121	/*  Put the stack after the struct pt_regs.  */
    122	childksp = (unsigned long) childregs;
    123	p->thread.cp0_status = (read_c0_status() & ~(ST0_CU2|ST0_CU1)) | ST0_KERNEL_CUMASK;
    124	if (unlikely(args->fn)) {
    125		/* kernel thread */
    126		unsigned long status = p->thread.cp0_status;
    127		memset(childregs, 0, sizeof(struct pt_regs));
    128		p->thread.reg16 = (unsigned long)args->fn;
    129		p->thread.reg17 = (unsigned long)args->fn_arg;
    130		p->thread.reg29 = childksp;
    131		p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
    132#if defined(CONFIG_CPU_R3000)
    133		status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
    134			 ((status & (ST0_KUC | ST0_IEC)) << 2);
    135#else
    136		status |= ST0_EXL;
    137#endif
    138		childregs->cp0_status = status;
    139		return 0;
    140	}
    141
    142	/* user thread */
    143	*childregs = *regs;
    144	childregs->regs[7] = 0; /* Clear error flag */
    145	childregs->regs[2] = 0; /* Child gets zero as return value */
    146	if (usp)
    147		childregs->regs[29] = usp;
    148
    149	p->thread.reg29 = (unsigned long) childregs;
    150	p->thread.reg31 = (unsigned long) ret_from_fork;
    151
    152	/*
    153	 * New tasks lose permission to use the fpu. This accelerates context
    154	 * switching for most programs since they don't use the fpu.
    155	 */
    156	childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
    157
    158	clear_tsk_thread_flag(p, TIF_USEDFPU);
    159	clear_tsk_thread_flag(p, TIF_USEDMSA);
    160	clear_tsk_thread_flag(p, TIF_MSA_CTX_LIVE);
    161
    162#ifdef CONFIG_MIPS_MT_FPAFF
    163	clear_tsk_thread_flag(p, TIF_FPUBOUND);
    164#endif /* CONFIG_MIPS_MT_FPAFF */
    165
    166#ifdef CONFIG_MIPS_FP_SUPPORT
    167	atomic_set(&p->thread.bd_emu_frame, BD_EMUFRAME_NONE);
    168#endif
    169
    170	if (clone_flags & CLONE_SETTLS)
    171		ti->tp_value = tls;
    172
    173	return 0;
    174}
    175
    176#ifdef CONFIG_STACKPROTECTOR
    177#include <linux/stackprotector.h>
    178unsigned long __stack_chk_guard __read_mostly;
    179EXPORT_SYMBOL(__stack_chk_guard);
    180#endif
    181
    182struct mips_frame_info {
    183	void		*func;
    184	unsigned long	func_size;
    185	int		frame_size;
    186	int		pc_offset;
    187};
    188
    189#define J_TARGET(pc,target)	\
    190		(((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
    191
    192static inline int is_jr_ra_ins(union mips_instruction *ip)
    193{
    194#ifdef CONFIG_CPU_MICROMIPS
    195	/*
    196	 * jr16 ra
    197	 * jr ra
    198	 */
    199	if (mm_insn_16bit(ip->word >> 16)) {
    200		if (ip->mm16_r5_format.opcode == mm_pool16c_op &&
    201		    ip->mm16_r5_format.rt == mm_jr16_op &&
    202		    ip->mm16_r5_format.imm == 31)
    203			return 1;
    204		return 0;
    205	}
    206
    207	if (ip->r_format.opcode == mm_pool32a_op &&
    208	    ip->r_format.func == mm_pool32axf_op &&
    209	    ((ip->u_format.uimmediate >> 6) & GENMASK(9, 0)) == mm_jalr_op &&
    210	    ip->r_format.rt == 31)
    211		return 1;
    212	return 0;
    213#else
    214	if (ip->r_format.opcode == spec_op &&
    215	    ip->r_format.func == jr_op &&
    216	    ip->r_format.rs == 31)
    217		return 1;
    218	return 0;
    219#endif
    220}
    221
    222static inline int is_ra_save_ins(union mips_instruction *ip, int *poff)
    223{
    224#ifdef CONFIG_CPU_MICROMIPS
    225	/*
    226	 * swsp ra,offset
    227	 * swm16 reglist,offset(sp)
    228	 * swm32 reglist,offset(sp)
    229	 * sw32 ra,offset(sp)
    230	 * jradiussp - NOT SUPPORTED
    231	 *
    232	 * microMIPS is way more fun...
    233	 */
    234	if (mm_insn_16bit(ip->word >> 16)) {
    235		switch (ip->mm16_r5_format.opcode) {
    236		case mm_swsp16_op:
    237			if (ip->mm16_r5_format.rt != 31)
    238				return 0;
    239
    240			*poff = ip->mm16_r5_format.imm;
    241			*poff = (*poff << 2) / sizeof(ulong);
    242			return 1;
    243
    244		case mm_pool16c_op:
    245			switch (ip->mm16_m_format.func) {
    246			case mm_swm16_op:
    247				*poff = ip->mm16_m_format.imm;
    248				*poff += 1 + ip->mm16_m_format.rlist;
    249				*poff = (*poff << 2) / sizeof(ulong);
    250				return 1;
    251
    252			default:
    253				return 0;
    254			}
    255
    256		default:
    257			return 0;
    258		}
    259	}
    260
    261	switch (ip->i_format.opcode) {
    262	case mm_sw32_op:
    263		if (ip->i_format.rs != 29)
    264			return 0;
    265		if (ip->i_format.rt != 31)
    266			return 0;
    267
    268		*poff = ip->i_format.simmediate / sizeof(ulong);
    269		return 1;
    270
    271	case mm_pool32b_op:
    272		switch (ip->mm_m_format.func) {
    273		case mm_swm32_func:
    274			if (ip->mm_m_format.rd < 0x10)
    275				return 0;
    276			if (ip->mm_m_format.base != 29)
    277				return 0;
    278
    279			*poff = ip->mm_m_format.simmediate;
    280			*poff += (ip->mm_m_format.rd & 0xf) * sizeof(u32);
    281			*poff /= sizeof(ulong);
    282			return 1;
    283		default:
    284			return 0;
    285		}
    286
    287	default:
    288		return 0;
    289	}
    290#else
    291	/* sw / sd $ra, offset($sp) */
    292	if ((ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
    293		ip->i_format.rs == 29 && ip->i_format.rt == 31) {
    294		*poff = ip->i_format.simmediate / sizeof(ulong);
    295		return 1;
    296	}
    297#ifdef CONFIG_CPU_LOONGSON64
    298	if ((ip->loongson3_lswc2_format.opcode == swc2_op) &&
    299		      (ip->loongson3_lswc2_format.ls == 1) &&
    300		      (ip->loongson3_lswc2_format.fr == 0) &&
    301		      (ip->loongson3_lswc2_format.base == 29)) {
    302		if (ip->loongson3_lswc2_format.rt == 31) {
    303			*poff = ip->loongson3_lswc2_format.offset << 1;
    304			return 1;
    305		}
    306		if (ip->loongson3_lswc2_format.rq == 31) {
    307			*poff = (ip->loongson3_lswc2_format.offset << 1) + 1;
    308			return 1;
    309		}
    310	}
    311#endif
    312	return 0;
    313#endif
    314}
    315
    316static inline int is_jump_ins(union mips_instruction *ip)
    317{
    318#ifdef CONFIG_CPU_MICROMIPS
    319	/*
    320	 * jr16,jrc,jalr16,jalr16
    321	 * jal
    322	 * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
    323	 * jraddiusp - NOT SUPPORTED
    324	 *
    325	 * microMIPS is kind of more fun...
    326	 */
    327	if (mm_insn_16bit(ip->word >> 16)) {
    328		if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
    329		    (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op))
    330			return 1;
    331		return 0;
    332	}
    333
    334	if (ip->j_format.opcode == mm_j32_op)
    335		return 1;
    336	if (ip->j_format.opcode == mm_jal32_op)
    337		return 1;
    338	if (ip->r_format.opcode != mm_pool32a_op ||
    339			ip->r_format.func != mm_pool32axf_op)
    340		return 0;
    341	return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
    342#else
    343	if (ip->j_format.opcode == j_op)
    344		return 1;
    345	if (ip->j_format.opcode == jal_op)
    346		return 1;
    347	if (ip->r_format.opcode != spec_op)
    348		return 0;
    349	return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
    350#endif
    351}
    352
    353static inline int is_sp_move_ins(union mips_instruction *ip, int *frame_size)
    354{
    355#ifdef CONFIG_CPU_MICROMIPS
    356	unsigned short tmp;
    357
    358	/*
    359	 * addiusp -imm
    360	 * addius5 sp,-imm
    361	 * addiu32 sp,sp,-imm
    362	 * jradiussp - NOT SUPPORTED
    363	 *
    364	 * microMIPS is not more fun...
    365	 */
    366	if (mm_insn_16bit(ip->word >> 16)) {
    367		if (ip->mm16_r3_format.opcode == mm_pool16d_op &&
    368		    ip->mm16_r3_format.simmediate & mm_addiusp_func) {
    369			tmp = ip->mm_b0_format.simmediate >> 1;
    370			tmp = ((tmp & 0x1ff) ^ 0x100) - 0x100;
    371			if ((tmp + 2) < 4) /* 0x0,0x1,0x1fe,0x1ff are special */
    372				tmp ^= 0x100;
    373			*frame_size = -(signed short)(tmp << 2);
    374			return 1;
    375		}
    376		if (ip->mm16_r5_format.opcode == mm_pool16d_op &&
    377		    ip->mm16_r5_format.rt == 29) {
    378			tmp = ip->mm16_r5_format.imm >> 1;
    379			*frame_size = -(signed short)(tmp & 0xf);
    380			return 1;
    381		}
    382		return 0;
    383	}
    384
    385	if (ip->mm_i_format.opcode == mm_addiu32_op &&
    386	    ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29) {
    387		*frame_size = -ip->i_format.simmediate;
    388		return 1;
    389	}
    390#else
    391	/* addiu/daddiu sp,sp,-imm */
    392	if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
    393		return 0;
    394
    395	if (ip->i_format.opcode == addiu_op ||
    396	    ip->i_format.opcode == daddiu_op) {
    397		*frame_size = -ip->i_format.simmediate;
    398		return 1;
    399	}
    400#endif
    401	return 0;
    402}
    403
    404static int get_frame_info(struct mips_frame_info *info)
    405{
    406	bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
    407	union mips_instruction insn, *ip, *ip_end;
    408	unsigned int last_insn_size = 0;
    409	bool saw_jump = false;
    410
    411	info->pc_offset = -1;
    412	info->frame_size = 0;
    413
    414	ip = (void *)msk_isa16_mode((ulong)info->func);
    415	if (!ip)
    416		goto err;
    417
    418	ip_end = (void *)ip + (info->func_size ? info->func_size : 512);
    419
    420	while (ip < ip_end) {
    421		ip = (void *)ip + last_insn_size;
    422
    423		if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
    424			insn.word = ip->halfword[0] << 16;
    425			last_insn_size = 2;
    426		} else if (is_mmips) {
    427			insn.word = ip->halfword[0] << 16 | ip->halfword[1];
    428			last_insn_size = 4;
    429		} else {
    430			insn.word = ip->word;
    431			last_insn_size = 4;
    432		}
    433
    434		if (is_jr_ra_ins(ip)) {
    435			break;
    436		} else if (!info->frame_size) {
    437			is_sp_move_ins(&insn, &info->frame_size);
    438			continue;
    439		} else if (!saw_jump && is_jump_ins(ip)) {
    440			/*
    441			 * If we see a jump instruction, we are finished
    442			 * with the frame save.
    443			 *
    444			 * Some functions can have a shortcut return at
    445			 * the beginning of the function, so don't start
    446			 * looking for jump instruction until we see the
    447			 * frame setup.
    448			 *
    449			 * The RA save instruction can get put into the
    450			 * delay slot of the jump instruction, so look
    451			 * at the next instruction, too.
    452			 */
    453			saw_jump = true;
    454			continue;
    455		}
    456		if (info->pc_offset == -1 &&
    457		    is_ra_save_ins(&insn, &info->pc_offset))
    458			break;
    459		if (saw_jump)
    460			break;
    461	}
    462	if (info->frame_size && info->pc_offset >= 0) /* nested */
    463		return 0;
    464	if (info->pc_offset < 0) /* leaf */
    465		return 1;
    466	/* prologue seems bogus... */
    467err:
    468	return -1;
    469}
    470
    471static struct mips_frame_info schedule_mfi __read_mostly;
    472
    473#ifdef CONFIG_KALLSYMS
    474static unsigned long get___schedule_addr(void)
    475{
    476	return kallsyms_lookup_name("__schedule");
    477}
    478#else
    479static unsigned long get___schedule_addr(void)
    480{
    481	union mips_instruction *ip = (void *)schedule;
    482	int max_insns = 8;
    483	int i;
    484
    485	for (i = 0; i < max_insns; i++, ip++) {
    486		if (ip->j_format.opcode == j_op)
    487			return J_TARGET(ip, ip->j_format.target);
    488	}
    489	return 0;
    490}
    491#endif
    492
    493static int __init frame_info_init(void)
    494{
    495	unsigned long size = 0;
    496#ifdef CONFIG_KALLSYMS
    497	unsigned long ofs;
    498#endif
    499	unsigned long addr;
    500
    501	addr = get___schedule_addr();
    502	if (!addr)
    503		addr = (unsigned long)schedule;
    504
    505#ifdef CONFIG_KALLSYMS
    506	kallsyms_lookup_size_offset(addr, &size, &ofs);
    507#endif
    508	schedule_mfi.func = (void *)addr;
    509	schedule_mfi.func_size = size;
    510
    511	get_frame_info(&schedule_mfi);
    512
    513	/*
    514	 * Without schedule() frame info, result given by
    515	 * thread_saved_pc() and __get_wchan() are not reliable.
    516	 */
    517	if (schedule_mfi.pc_offset < 0)
    518		printk("Can't analyze schedule() prologue at %p\n", schedule);
    519
    520	return 0;
    521}
    522
    523arch_initcall(frame_info_init);
    524
    525/*
    526 * Return saved PC of a blocked thread.
    527 */
    528static unsigned long thread_saved_pc(struct task_struct *tsk)
    529{
    530	struct thread_struct *t = &tsk->thread;
    531
    532	/* New born processes are a special case */
    533	if (t->reg31 == (unsigned long) ret_from_fork)
    534		return t->reg31;
    535	if (schedule_mfi.pc_offset < 0)
    536		return 0;
    537	return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
    538}
    539
    540
    541#ifdef CONFIG_KALLSYMS
    542/* generic stack unwinding function */
    543unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
    544					      unsigned long *sp,
    545					      unsigned long pc,
    546					      unsigned long *ra)
    547{
    548	unsigned long low, high, irq_stack_high;
    549	struct mips_frame_info info;
    550	unsigned long size, ofs;
    551	struct pt_regs *regs;
    552	int leaf;
    553
    554	if (!stack_page)
    555		return 0;
    556
    557	/*
    558	 * IRQ stacks start at IRQ_STACK_START
    559	 * task stacks at THREAD_SIZE - 32
    560	 */
    561	low = stack_page;
    562	if (!preemptible() && on_irq_stack(raw_smp_processor_id(), *sp)) {
    563		high = stack_page + IRQ_STACK_START;
    564		irq_stack_high = high;
    565	} else {
    566		high = stack_page + THREAD_SIZE - 32;
    567		irq_stack_high = 0;
    568	}
    569
    570	/*
    571	 * If we reached the top of the interrupt stack, start unwinding
    572	 * the interrupted task stack.
    573	 */
    574	if (unlikely(*sp == irq_stack_high)) {
    575		unsigned long task_sp = *(unsigned long *)*sp;
    576
    577		/*
    578		 * Check that the pointer saved in the IRQ stack head points to
    579		 * something within the stack of the current task
    580		 */
    581		if (!object_is_on_stack((void *)task_sp))
    582			return 0;
    583
    584		/*
    585		 * Follow pointer to tasks kernel stack frame where interrupted
    586		 * state was saved.
    587		 */
    588		regs = (struct pt_regs *)task_sp;
    589		pc = regs->cp0_epc;
    590		if (!user_mode(regs) && __kernel_text_address(pc)) {
    591			*sp = regs->regs[29];
    592			*ra = regs->regs[31];
    593			return pc;
    594		}
    595		return 0;
    596	}
    597	if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
    598		return 0;
    599	/*
    600	 * Return ra if an exception occurred at the first instruction
    601	 */
    602	if (unlikely(ofs == 0)) {
    603		pc = *ra;
    604		*ra = 0;
    605		return pc;
    606	}
    607
    608	info.func = (void *)(pc - ofs);
    609	info.func_size = ofs;	/* analyze from start to ofs */
    610	leaf = get_frame_info(&info);
    611	if (leaf < 0)
    612		return 0;
    613
    614	if (*sp < low || *sp + info.frame_size > high)
    615		return 0;
    616
    617	if (leaf)
    618		/*
    619		 * For some extreme cases, get_frame_info() can
    620		 * consider wrongly a nested function as a leaf
    621		 * one. In that cases avoid to return always the
    622		 * same value.
    623		 */
    624		pc = pc != *ra ? *ra : 0;
    625	else
    626		pc = ((unsigned long *)(*sp))[info.pc_offset];
    627
    628	*sp += info.frame_size;
    629	*ra = 0;
    630	return __kernel_text_address(pc) ? pc : 0;
    631}
    632EXPORT_SYMBOL(unwind_stack_by_address);
    633
    634/* used by show_backtrace() */
    635unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
    636			   unsigned long pc, unsigned long *ra)
    637{
    638	unsigned long stack_page = 0;
    639	int cpu;
    640
    641	for_each_possible_cpu(cpu) {
    642		if (on_irq_stack(cpu, *sp)) {
    643			stack_page = (unsigned long)irq_stack[cpu];
    644			break;
    645		}
    646	}
    647
    648	if (!stack_page)
    649		stack_page = (unsigned long)task_stack_page(task);
    650
    651	return unwind_stack_by_address(stack_page, sp, pc, ra);
    652}
    653#endif
    654
    655/*
    656 * __get_wchan - a maintenance nightmare^W^Wpain in the ass ...
    657 */
    658unsigned long __get_wchan(struct task_struct *task)
    659{
    660	unsigned long pc = 0;
    661#ifdef CONFIG_KALLSYMS
    662	unsigned long sp;
    663	unsigned long ra = 0;
    664#endif
    665
    666	if (!task_stack_page(task))
    667		goto out;
    668
    669	pc = thread_saved_pc(task);
    670
    671#ifdef CONFIG_KALLSYMS
    672	sp = task->thread.reg29 + schedule_mfi.frame_size;
    673
    674	while (in_sched_functions(pc))
    675		pc = unwind_stack(task, &sp, pc, &ra);
    676#endif
    677
    678out:
    679	return pc;
    680}
    681
    682unsigned long mips_stack_top(void)
    683{
    684	unsigned long top = TASK_SIZE & PAGE_MASK;
    685
    686	if (IS_ENABLED(CONFIG_MIPS_FP_SUPPORT)) {
    687		/* One page for branch delay slot "emulation" */
    688		top -= PAGE_SIZE;
    689	}
    690
    691	/* Space for the VDSO, data page & GIC user page */
    692	top -= PAGE_ALIGN(current->thread.abi->vdso->size);
    693	top -= PAGE_SIZE;
    694	top -= mips_gic_present() ? PAGE_SIZE : 0;
    695
    696	/* Space for cache colour alignment */
    697	if (cpu_has_dc_aliases)
    698		top -= shm_align_mask + 1;
    699
    700	/* Space to randomize the VDSO base */
    701	if (current->flags & PF_RANDOMIZE)
    702		top -= VDSO_RANDOMIZE_SIZE;
    703
    704	return top;
    705}
    706
    707/*
    708 * Don't forget that the stack pointer must be aligned on a 8 bytes
    709 * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
    710 */
    711unsigned long arch_align_stack(unsigned long sp)
    712{
    713	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
    714		sp -= get_random_int() & ~PAGE_MASK;
    715
    716	return sp & ALMASK;
    717}
    718
    719static struct cpumask backtrace_csd_busy;
    720
    721static void handle_backtrace(void *info)
    722{
    723	nmi_cpu_backtrace(get_irq_regs());
    724	cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
    725}
    726
    727static DEFINE_PER_CPU(call_single_data_t, backtrace_csd) =
    728	CSD_INIT(handle_backtrace, NULL);
    729
    730static void raise_backtrace(cpumask_t *mask)
    731{
    732	call_single_data_t *csd;
    733	int cpu;
    734
    735	for_each_cpu(cpu, mask) {
    736		/*
    737		 * If we previously sent an IPI to the target CPU & it hasn't
    738		 * cleared its bit in the busy cpumask then it didn't handle
    739		 * our previous IPI & it's not safe for us to reuse the
    740		 * call_single_data_t.
    741		 */
    742		if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
    743			pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
    744				cpu);
    745			continue;
    746		}
    747
    748		csd = &per_cpu(backtrace_csd, cpu);
    749		smp_call_function_single_async(cpu, csd);
    750	}
    751}
    752
    753void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
    754{
    755	nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace);
    756}
    757
    758int mips_get_process_fp_mode(struct task_struct *task)
    759{
    760	int value = 0;
    761
    762	if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
    763		value |= PR_FP_MODE_FR;
    764	if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
    765		value |= PR_FP_MODE_FRE;
    766
    767	return value;
    768}
    769
    770static long prepare_for_fp_mode_switch(void *unused)
    771{
    772	/*
    773	 * This is icky, but we use this to simply ensure that all CPUs have
    774	 * context switched, regardless of whether they were previously running
    775	 * kernel or user code. This ensures that no CPU that a mode-switching
    776	 * program may execute on keeps its FPU enabled (& in the old mode)
    777	 * throughout the mode switch.
    778	 */
    779	return 0;
    780}
    781
    782int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
    783{
    784	const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
    785	struct task_struct *t;
    786	struct cpumask process_cpus;
    787	int cpu;
    788
    789	/* If nothing to change, return right away, successfully.  */
    790	if (value == mips_get_process_fp_mode(task))
    791		return 0;
    792
    793	/* Only accept a mode change if 64-bit FP enabled for o32.  */
    794	if (!IS_ENABLED(CONFIG_MIPS_O32_FP64_SUPPORT))
    795		return -EOPNOTSUPP;
    796
    797	/* And only for o32 tasks.  */
    798	if (IS_ENABLED(CONFIG_64BIT) && !test_thread_flag(TIF_32BIT_REGS))
    799		return -EOPNOTSUPP;
    800
    801	/* Check the value is valid */
    802	if (value & ~known_bits)
    803		return -EOPNOTSUPP;
    804
    805	/* Setting FRE without FR is not supported.  */
    806	if ((value & (PR_FP_MODE_FR | PR_FP_MODE_FRE)) == PR_FP_MODE_FRE)
    807		return -EOPNOTSUPP;
    808
    809	/* Avoid inadvertently triggering emulation */
    810	if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
    811	    !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
    812		return -EOPNOTSUPP;
    813	if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
    814		return -EOPNOTSUPP;
    815
    816	/* FR = 0 not supported in MIPS R6 */
    817	if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
    818		return -EOPNOTSUPP;
    819
    820	/* Indicate the new FP mode in each thread */
    821	for_each_thread(task, t) {
    822		/* Update desired FP register width */
    823		if (value & PR_FP_MODE_FR) {
    824			clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
    825		} else {
    826			set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
    827			clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
    828		}
    829
    830		/* Update desired FP single layout */
    831		if (value & PR_FP_MODE_FRE)
    832			set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
    833		else
    834			clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
    835	}
    836
    837	/*
    838	 * We need to ensure that all threads in the process have switched mode
    839	 * before returning, in order to allow userland to not worry about
    840	 * races. We can do this by forcing all CPUs that any thread in the
    841	 * process may be running on to schedule something else - in this case
    842	 * prepare_for_fp_mode_switch().
    843	 *
    844	 * We begin by generating a mask of all CPUs that any thread in the
    845	 * process may be running on.
    846	 */
    847	cpumask_clear(&process_cpus);
    848	for_each_thread(task, t)
    849		cpumask_set_cpu(task_cpu(t), &process_cpus);
    850
    851	/*
    852	 * Now we schedule prepare_for_fp_mode_switch() on each of those CPUs.
    853	 *
    854	 * The CPUs may have rescheduled already since we switched mode or
    855	 * generated the cpumask, but that doesn't matter. If the task in this
    856	 * process is scheduled out then our scheduling
    857	 * prepare_for_fp_mode_switch() will simply be redundant. If it's
    858	 * scheduled in then it will already have picked up the new FP mode
    859	 * whilst doing so.
    860	 */
    861	cpus_read_lock();
    862	for_each_cpu_and(cpu, &process_cpus, cpu_online_mask)
    863		work_on_cpu(cpu, prepare_for_fp_mode_switch, NULL);
    864	cpus_read_unlock();
    865
    866	return 0;
    867}
    868
    869#if defined(CONFIG_32BIT) || defined(CONFIG_MIPS32_O32)
    870void mips_dump_regs32(u32 *uregs, const struct pt_regs *regs)
    871{
    872	unsigned int i;
    873
    874	for (i = MIPS32_EF_R1; i <= MIPS32_EF_R31; i++) {
    875		/* k0/k1 are copied as zero. */
    876		if (i == MIPS32_EF_R26 || i == MIPS32_EF_R27)
    877			uregs[i] = 0;
    878		else
    879			uregs[i] = regs->regs[i - MIPS32_EF_R0];
    880	}
    881
    882	uregs[MIPS32_EF_LO] = regs->lo;
    883	uregs[MIPS32_EF_HI] = regs->hi;
    884	uregs[MIPS32_EF_CP0_EPC] = regs->cp0_epc;
    885	uregs[MIPS32_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
    886	uregs[MIPS32_EF_CP0_STATUS] = regs->cp0_status;
    887	uregs[MIPS32_EF_CP0_CAUSE] = regs->cp0_cause;
    888}
    889#endif /* CONFIG_32BIT || CONFIG_MIPS32_O32 */
    890
    891#ifdef CONFIG_64BIT
    892void mips_dump_regs64(u64 *uregs, const struct pt_regs *regs)
    893{
    894	unsigned int i;
    895
    896	for (i = MIPS64_EF_R1; i <= MIPS64_EF_R31; i++) {
    897		/* k0/k1 are copied as zero. */
    898		if (i == MIPS64_EF_R26 || i == MIPS64_EF_R27)
    899			uregs[i] = 0;
    900		else
    901			uregs[i] = regs->regs[i - MIPS64_EF_R0];
    902	}
    903
    904	uregs[MIPS64_EF_LO] = regs->lo;
    905	uregs[MIPS64_EF_HI] = regs->hi;
    906	uregs[MIPS64_EF_CP0_EPC] = regs->cp0_epc;
    907	uregs[MIPS64_EF_CP0_BADVADDR] = regs->cp0_badvaddr;
    908	uregs[MIPS64_EF_CP0_STATUS] = regs->cp0_status;
    909	uregs[MIPS64_EF_CP0_CAUSE] = regs->cp0_cause;
    910}
    911#endif /* CONFIG_64BIT */