xref: /openbmc/linux/arch/mips/kernel/process.c (revision 8571e645)
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/errno.h>
13 #include <linux/sched.h>
14 #include <linux/tick.h>
15 #include <linux/kernel.h>
16 #include <linux/mm.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>
29 
30 #include <asm/asm.h>
31 #include <asm/bootinfo.h>
32 #include <asm/cpu.h>
33 #include <asm/dsp.h>
34 #include <asm/fpu.h>
35 #include <asm/msa.h>
36 #include <asm/pgtable.h>
37 #include <asm/mipsregs.h>
38 #include <asm/processor.h>
39 #include <asm/reg.h>
40 #include <asm/uaccess.h>
41 #include <asm/io.h>
42 #include <asm/elf.h>
43 #include <asm/isadep.h>
44 #include <asm/inst.h>
45 #include <asm/stacktrace.h>
46 #include <asm/irq_regs.h>
47 
48 #ifdef CONFIG_HOTPLUG_CPU
49 void arch_cpu_idle_dead(void)
50 {
51 	/* What the heck is this check doing ? */
52 	if (!cpumask_test_cpu(smp_processor_id(), &cpu_callin_map))
53 		play_dead();
54 }
55 #endif
56 
57 asmlinkage void ret_from_fork(void);
58 asmlinkage void ret_from_kernel_thread(void);
59 
60 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
61 {
62 	unsigned long status;
63 
64 	/* New thread loses kernel privileges. */
65 	status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
66 	status |= KU_USER;
67 	regs->cp0_status = status;
68 	lose_fpu(0);
69 	clear_thread_flag(TIF_MSA_CTX_LIVE);
70 	clear_used_math();
71 	init_dsp();
72 	regs->cp0_epc = pc;
73 	regs->regs[29] = sp;
74 }
75 
76 void exit_thread(void)
77 {
78 }
79 
80 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
81 {
82 	/*
83 	 * Save any process state which is live in hardware registers to the
84 	 * parent context prior to duplication. This prevents the new child
85 	 * state becoming stale if the parent is preempted before copy_thread()
86 	 * gets a chance to save the parent's live hardware registers to the
87 	 * child context.
88 	 */
89 	preempt_disable();
90 
91 	if (is_msa_enabled())
92 		save_msa(current);
93 	else if (is_fpu_owner())
94 		_save_fp(current);
95 
96 	save_dsp(current);
97 
98 	preempt_enable();
99 
100 	*dst = *src;
101 	return 0;
102 }
103 
104 /*
105  * Copy architecture-specific thread state
106  */
107 int copy_thread(unsigned long clone_flags, unsigned long usp,
108 	unsigned long kthread_arg, struct task_struct *p)
109 {
110 	struct thread_info *ti = task_thread_info(p);
111 	struct pt_regs *childregs, *regs = current_pt_regs();
112 	unsigned long childksp;
113 	p->set_child_tid = p->clear_child_tid = NULL;
114 
115 	childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
116 
117 	/* set up new TSS. */
118 	childregs = (struct pt_regs *) childksp - 1;
119 	/*  Put the stack after the struct pt_regs.  */
120 	childksp = (unsigned long) childregs;
121 	p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
122 	if (unlikely(p->flags & PF_KTHREAD)) {
123 		/* kernel thread */
124 		unsigned long status = p->thread.cp0_status;
125 		memset(childregs, 0, sizeof(struct pt_regs));
126 		ti->addr_limit = KERNEL_DS;
127 		p->thread.reg16 = usp; /* fn */
128 		p->thread.reg17 = kthread_arg;
129 		p->thread.reg29 = childksp;
130 		p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
131 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
132 		status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
133 			 ((status & (ST0_KUC | ST0_IEC)) << 2);
134 #else
135 		status |= ST0_EXL;
136 #endif
137 		childregs->cp0_status = status;
138 		return 0;
139 	}
140 
141 	/* user thread */
142 	*childregs = *regs;
143 	childregs->regs[7] = 0; /* Clear error flag */
144 	childregs->regs[2] = 0; /* Child gets zero as return value */
145 	if (usp)
146 		childregs->regs[29] = usp;
147 	ti->addr_limit = USER_DS;
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 	if (clone_flags & CLONE_SETTLS)
167 		ti->tp_value = regs->regs[7];
168 
169 	return 0;
170 }
171 
172 #ifdef CONFIG_CC_STACKPROTECTOR
173 #include <linux/stackprotector.h>
174 unsigned long __stack_chk_guard __read_mostly;
175 EXPORT_SYMBOL(__stack_chk_guard);
176 #endif
177 
178 struct mips_frame_info {
179 	void		*func;
180 	unsigned long	func_size;
181 	int		frame_size;
182 	int		pc_offset;
183 };
184 
185 #define J_TARGET(pc,target)	\
186 		(((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
187 
188 static inline int is_ra_save_ins(union mips_instruction *ip)
189 {
190 #ifdef CONFIG_CPU_MICROMIPS
191 	union mips_instruction mmi;
192 
193 	/*
194 	 * swsp ra,offset
195 	 * swm16 reglist,offset(sp)
196 	 * swm32 reglist,offset(sp)
197 	 * sw32 ra,offset(sp)
198 	 * jradiussp - NOT SUPPORTED
199 	 *
200 	 * microMIPS is way more fun...
201 	 */
202 	if (mm_insn_16bit(ip->halfword[0])) {
203 		mmi.word = (ip->halfword[0] << 16);
204 		return (mmi.mm16_r5_format.opcode == mm_swsp16_op &&
205 			mmi.mm16_r5_format.rt == 31) ||
206 		       (mmi.mm16_m_format.opcode == mm_pool16c_op &&
207 			mmi.mm16_m_format.func == mm_swm16_op);
208 	}
209 	else {
210 		mmi.halfword[0] = ip->halfword[1];
211 		mmi.halfword[1] = ip->halfword[0];
212 		return (mmi.mm_m_format.opcode == mm_pool32b_op &&
213 			mmi.mm_m_format.rd > 9 &&
214 			mmi.mm_m_format.base == 29 &&
215 			mmi.mm_m_format.func == mm_swm32_func) ||
216 		       (mmi.i_format.opcode == mm_sw32_op &&
217 			mmi.i_format.rs == 29 &&
218 			mmi.i_format.rt == 31);
219 	}
220 #else
221 	/* sw / sd $ra, offset($sp) */
222 	return (ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
223 		ip->i_format.rs == 29 &&
224 		ip->i_format.rt == 31;
225 #endif
226 }
227 
228 static inline int is_jump_ins(union mips_instruction *ip)
229 {
230 #ifdef CONFIG_CPU_MICROMIPS
231 	/*
232 	 * jr16,jrc,jalr16,jalr16
233 	 * jal
234 	 * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
235 	 * jraddiusp - NOT SUPPORTED
236 	 *
237 	 * microMIPS is kind of more fun...
238 	 */
239 	union mips_instruction mmi;
240 
241 	mmi.word = (ip->halfword[0] << 16);
242 
243 	if ((mmi.mm16_r5_format.opcode == mm_pool16c_op &&
244 	    (mmi.mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op) ||
245 	    ip->j_format.opcode == mm_jal32_op)
246 		return 1;
247 	if (ip->r_format.opcode != mm_pool32a_op ||
248 			ip->r_format.func != mm_pool32axf_op)
249 		return 0;
250 	return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
251 #else
252 	if (ip->j_format.opcode == j_op)
253 		return 1;
254 	if (ip->j_format.opcode == jal_op)
255 		return 1;
256 	if (ip->r_format.opcode != spec_op)
257 		return 0;
258 	return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
259 #endif
260 }
261 
262 static inline int is_sp_move_ins(union mips_instruction *ip)
263 {
264 #ifdef CONFIG_CPU_MICROMIPS
265 	/*
266 	 * addiusp -imm
267 	 * addius5 sp,-imm
268 	 * addiu32 sp,sp,-imm
269 	 * jradiussp - NOT SUPPORTED
270 	 *
271 	 * microMIPS is not more fun...
272 	 */
273 	if (mm_insn_16bit(ip->halfword[0])) {
274 		union mips_instruction mmi;
275 
276 		mmi.word = (ip->halfword[0] << 16);
277 		return (mmi.mm16_r3_format.opcode == mm_pool16d_op &&
278 			mmi.mm16_r3_format.simmediate && mm_addiusp_func) ||
279 		       (mmi.mm16_r5_format.opcode == mm_pool16d_op &&
280 			mmi.mm16_r5_format.rt == 29);
281 	}
282 	return ip->mm_i_format.opcode == mm_addiu32_op &&
283 	       ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29;
284 #else
285 	/* addiu/daddiu sp,sp,-imm */
286 	if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
287 		return 0;
288 	if (ip->i_format.opcode == addiu_op || ip->i_format.opcode == daddiu_op)
289 		return 1;
290 #endif
291 	return 0;
292 }
293 
294 static int get_frame_info(struct mips_frame_info *info)
295 {
296 #ifdef CONFIG_CPU_MICROMIPS
297 	union mips_instruction *ip = (void *) (((char *) info->func) - 1);
298 #else
299 	union mips_instruction *ip = info->func;
300 #endif
301 	unsigned max_insns = info->func_size / sizeof(union mips_instruction);
302 	unsigned i;
303 
304 	info->pc_offset = -1;
305 	info->frame_size = 0;
306 
307 	if (!ip)
308 		goto err;
309 
310 	if (max_insns == 0)
311 		max_insns = 128U;	/* unknown function size */
312 	max_insns = min(128U, max_insns);
313 
314 	for (i = 0; i < max_insns; i++, ip++) {
315 
316 		if (is_jump_ins(ip))
317 			break;
318 		if (!info->frame_size) {
319 			if (is_sp_move_ins(ip))
320 			{
321 #ifdef CONFIG_CPU_MICROMIPS
322 				if (mm_insn_16bit(ip->halfword[0]))
323 				{
324 					unsigned short tmp;
325 
326 					if (ip->halfword[0] & mm_addiusp_func)
327 					{
328 						tmp = (((ip->halfword[0] >> 1) & 0x1ff) << 2);
329 						info->frame_size = -(signed short)(tmp | ((tmp & 0x100) ? 0xfe00 : 0));
330 					} else {
331 						tmp = (ip->halfword[0] >> 1);
332 						info->frame_size = -(signed short)(tmp & 0xf);
333 					}
334 					ip = (void *) &ip->halfword[1];
335 					ip--;
336 				} else
337 #endif
338 				info->frame_size = - ip->i_format.simmediate;
339 			}
340 			continue;
341 		}
342 		if (info->pc_offset == -1 && is_ra_save_ins(ip)) {
343 			info->pc_offset =
344 				ip->i_format.simmediate / sizeof(long);
345 			break;
346 		}
347 	}
348 	if (info->frame_size && info->pc_offset >= 0) /* nested */
349 		return 0;
350 	if (info->pc_offset < 0) /* leaf */
351 		return 1;
352 	/* prologue seems boggus... */
353 err:
354 	return -1;
355 }
356 
357 static struct mips_frame_info schedule_mfi __read_mostly;
358 
359 #ifdef CONFIG_KALLSYMS
360 static unsigned long get___schedule_addr(void)
361 {
362 	return kallsyms_lookup_name("__schedule");
363 }
364 #else
365 static unsigned long get___schedule_addr(void)
366 {
367 	union mips_instruction *ip = (void *)schedule;
368 	int max_insns = 8;
369 	int i;
370 
371 	for (i = 0; i < max_insns; i++, ip++) {
372 		if (ip->j_format.opcode == j_op)
373 			return J_TARGET(ip, ip->j_format.target);
374 	}
375 	return 0;
376 }
377 #endif
378 
379 static int __init frame_info_init(void)
380 {
381 	unsigned long size = 0;
382 #ifdef CONFIG_KALLSYMS
383 	unsigned long ofs;
384 #endif
385 	unsigned long addr;
386 
387 	addr = get___schedule_addr();
388 	if (!addr)
389 		addr = (unsigned long)schedule;
390 
391 #ifdef CONFIG_KALLSYMS
392 	kallsyms_lookup_size_offset(addr, &size, &ofs);
393 #endif
394 	schedule_mfi.func = (void *)addr;
395 	schedule_mfi.func_size = size;
396 
397 	get_frame_info(&schedule_mfi);
398 
399 	/*
400 	 * Without schedule() frame info, result given by
401 	 * thread_saved_pc() and get_wchan() are not reliable.
402 	 */
403 	if (schedule_mfi.pc_offset < 0)
404 		printk("Can't analyze schedule() prologue at %p\n", schedule);
405 
406 	return 0;
407 }
408 
409 arch_initcall(frame_info_init);
410 
411 /*
412  * Return saved PC of a blocked thread.
413  */
414 unsigned long thread_saved_pc(struct task_struct *tsk)
415 {
416 	struct thread_struct *t = &tsk->thread;
417 
418 	/* New born processes are a special case */
419 	if (t->reg31 == (unsigned long) ret_from_fork)
420 		return t->reg31;
421 	if (schedule_mfi.pc_offset < 0)
422 		return 0;
423 	return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
424 }
425 
426 
427 #ifdef CONFIG_KALLSYMS
428 /* generic stack unwinding function */
429 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
430 					      unsigned long *sp,
431 					      unsigned long pc,
432 					      unsigned long *ra)
433 {
434 	struct mips_frame_info info;
435 	unsigned long size, ofs;
436 	int leaf;
437 	extern void ret_from_irq(void);
438 	extern void ret_from_exception(void);
439 
440 	if (!stack_page)
441 		return 0;
442 
443 	/*
444 	 * If we reached the bottom of interrupt context,
445 	 * return saved pc in pt_regs.
446 	 */
447 	if (pc == (unsigned long)ret_from_irq ||
448 	    pc == (unsigned long)ret_from_exception) {
449 		struct pt_regs *regs;
450 		if (*sp >= stack_page &&
451 		    *sp + sizeof(*regs) <= stack_page + THREAD_SIZE - 32) {
452 			regs = (struct pt_regs *)*sp;
453 			pc = regs->cp0_epc;
454 			if (!user_mode(regs) && __kernel_text_address(pc)) {
455 				*sp = regs->regs[29];
456 				*ra = regs->regs[31];
457 				return pc;
458 			}
459 		}
460 		return 0;
461 	}
462 	if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
463 		return 0;
464 	/*
465 	 * Return ra if an exception occurred at the first instruction
466 	 */
467 	if (unlikely(ofs == 0)) {
468 		pc = *ra;
469 		*ra = 0;
470 		return pc;
471 	}
472 
473 	info.func = (void *)(pc - ofs);
474 	info.func_size = ofs;	/* analyze from start to ofs */
475 	leaf = get_frame_info(&info);
476 	if (leaf < 0)
477 		return 0;
478 
479 	if (*sp < stack_page ||
480 	    *sp + info.frame_size > stack_page + THREAD_SIZE - 32)
481 		return 0;
482 
483 	if (leaf)
484 		/*
485 		 * For some extreme cases, get_frame_info() can
486 		 * consider wrongly a nested function as a leaf
487 		 * one. In that cases avoid to return always the
488 		 * same value.
489 		 */
490 		pc = pc != *ra ? *ra : 0;
491 	else
492 		pc = ((unsigned long *)(*sp))[info.pc_offset];
493 
494 	*sp += info.frame_size;
495 	*ra = 0;
496 	return __kernel_text_address(pc) ? pc : 0;
497 }
498 EXPORT_SYMBOL(unwind_stack_by_address);
499 
500 /* used by show_backtrace() */
501 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
502 			   unsigned long pc, unsigned long *ra)
503 {
504 	unsigned long stack_page = (unsigned long)task_stack_page(task);
505 	return unwind_stack_by_address(stack_page, sp, pc, ra);
506 }
507 #endif
508 
509 /*
510  * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
511  */
512 unsigned long get_wchan(struct task_struct *task)
513 {
514 	unsigned long pc = 0;
515 #ifdef CONFIG_KALLSYMS
516 	unsigned long sp;
517 	unsigned long ra = 0;
518 #endif
519 
520 	if (!task || task == current || task->state == TASK_RUNNING)
521 		goto out;
522 	if (!task_stack_page(task))
523 		goto out;
524 
525 	pc = thread_saved_pc(task);
526 
527 #ifdef CONFIG_KALLSYMS
528 	sp = task->thread.reg29 + schedule_mfi.frame_size;
529 
530 	while (in_sched_functions(pc))
531 		pc = unwind_stack(task, &sp, pc, &ra);
532 #endif
533 
534 out:
535 	return pc;
536 }
537 
538 /*
539  * Don't forget that the stack pointer must be aligned on a 8 bytes
540  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
541  */
542 unsigned long arch_align_stack(unsigned long sp)
543 {
544 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
545 		sp -= get_random_int() & ~PAGE_MASK;
546 
547 	return sp & ALMASK;
548 }
549 
550 static void arch_dump_stack(void *info)
551 {
552 	struct pt_regs *regs;
553 
554 	regs = get_irq_regs();
555 
556 	if (regs)
557 		show_regs(regs);
558 
559 	dump_stack();
560 }
561 
562 void arch_trigger_all_cpu_backtrace(bool include_self)
563 {
564 	smp_call_function(arch_dump_stack, NULL, 1);
565 }
566 
567 int mips_get_process_fp_mode(struct task_struct *task)
568 {
569 	int value = 0;
570 
571 	if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
572 		value |= PR_FP_MODE_FR;
573 	if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
574 		value |= PR_FP_MODE_FRE;
575 
576 	return value;
577 }
578 
579 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
580 {
581 	const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
582 	unsigned long switch_count;
583 	struct task_struct *t;
584 
585 	/* Check the value is valid */
586 	if (value & ~known_bits)
587 		return -EOPNOTSUPP;
588 
589 	/* Avoid inadvertently triggering emulation */
590 	if ((value & PR_FP_MODE_FR) && cpu_has_fpu &&
591 	    !(current_cpu_data.fpu_id & MIPS_FPIR_F64))
592 		return -EOPNOTSUPP;
593 	if ((value & PR_FP_MODE_FRE) && cpu_has_fpu && !cpu_has_fre)
594 		return -EOPNOTSUPP;
595 
596 	/* FR = 0 not supported in MIPS R6 */
597 	if (!(value & PR_FP_MODE_FR) && cpu_has_fpu && cpu_has_mips_r6)
598 		return -EOPNOTSUPP;
599 
600 	/* Save FP & vector context, then disable FPU & MSA */
601 	if (task->signal == current->signal)
602 		lose_fpu(1);
603 
604 	/* Prevent any threads from obtaining live FP context */
605 	atomic_set(&task->mm->context.fp_mode_switching, 1);
606 	smp_mb__after_atomic();
607 
608 	/*
609 	 * If there are multiple online CPUs then wait until all threads whose
610 	 * FP mode is about to change have been context switched. This approach
611 	 * allows us to only worry about whether an FP mode switch is in
612 	 * progress when FP is first used in a tasks time slice. Pretty much all
613 	 * of the mode switch overhead can thus be confined to cases where mode
614 	 * switches are actually occurring. That is, to here. However for the
615 	 * thread performing the mode switch it may take a while...
616 	 */
617 	if (num_online_cpus() > 1) {
618 		spin_lock_irq(&task->sighand->siglock);
619 
620 		for_each_thread(task, t) {
621 			if (t == current)
622 				continue;
623 
624 			switch_count = t->nvcsw + t->nivcsw;
625 
626 			do {
627 				spin_unlock_irq(&task->sighand->siglock);
628 				cond_resched();
629 				spin_lock_irq(&task->sighand->siglock);
630 			} while ((t->nvcsw + t->nivcsw) == switch_count);
631 		}
632 
633 		spin_unlock_irq(&task->sighand->siglock);
634 	}
635 
636 	/*
637 	 * There are now no threads of the process with live FP context, so it
638 	 * is safe to proceed with the FP mode switch.
639 	 */
640 	for_each_thread(task, t) {
641 		/* Update desired FP register width */
642 		if (value & PR_FP_MODE_FR) {
643 			clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
644 		} else {
645 			set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
646 			clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
647 		}
648 
649 		/* Update desired FP single layout */
650 		if (value & PR_FP_MODE_FRE)
651 			set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
652 		else
653 			clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
654 	}
655 
656 	/* Allow threads to use FP again */
657 	atomic_set(&task->mm->context.fp_mode_switching, 0);
658 
659 	return 0;
660 }
661