7bd0af0ea86d710e26f0a5e649723d080a8ea317
[firefly-linux-kernel-4.4.55.git] / arch / mips / kernel / process.c
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         clear_used_math();
69         clear_fpu_owner();
70         init_dsp();
71         clear_thread_flag(TIF_USEDMSA);
72         clear_thread_flag(TIF_MSA_CTX_LIVE);
73         disable_msa();
74         regs->cp0_epc = pc;
75         regs->regs[29] = sp;
76 }
77
78 void exit_thread(void)
79 {
80 }
81
82 void flush_thread(void)
83 {
84 }
85
86 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
87 {
88         /*
89          * Save any process state which is live in hardware registers to the
90          * parent context prior to duplication. This prevents the new child
91          * state becoming stale if the parent is preempted before copy_thread()
92          * gets a chance to save the parent's live hardware registers to the
93          * child context.
94          */
95         preempt_disable();
96
97         if (is_msa_enabled())
98                 save_msa(current);
99         else if (is_fpu_owner())
100                 _save_fp(current);
101
102         save_dsp(current);
103
104         preempt_enable();
105
106         *dst = *src;
107         return 0;
108 }
109
110 /*
111  * Copy architecture-specific thread state
112  */
113 int copy_thread(unsigned long clone_flags, unsigned long usp,
114         unsigned long kthread_arg, struct task_struct *p)
115 {
116         struct thread_info *ti = task_thread_info(p);
117         struct pt_regs *childregs, *regs = current_pt_regs();
118         unsigned long childksp;
119         p->set_child_tid = p->clear_child_tid = NULL;
120
121         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
122
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)) {
129                 /* kernel thread */
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);
140 #else
141                 status |= ST0_EXL;
142 #endif
143                 childregs->cp0_status = status;
144                 return 0;
145         }
146
147         /* user thread */
148         *childregs = *regs;
149         childregs->regs[7] = 0; /* Clear error flag */
150         childregs->regs[2] = 0; /* Child gets zero as return value */
151         if (usp)
152                 childregs->regs[29] = usp;
153         ti->addr_limit = USER_DS;
154
155         p->thread.reg29 = (unsigned long) childregs;
156         p->thread.reg31 = (unsigned long) ret_from_fork;
157
158         /*
159          * New tasks lose permission to use the fpu. This accelerates context
160          * switching for most programs since they don't use the fpu.
161          */
162         childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
163
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);
167
168 #ifdef CONFIG_MIPS_MT_FPAFF
169         clear_tsk_thread_flag(p, TIF_FPUBOUND);
170 #endif /* CONFIG_MIPS_MT_FPAFF */
171
172         if (clone_flags & CLONE_SETTLS)
173                 ti->tp_value = regs->regs[7];
174
175         return 0;
176 }
177
178 #ifdef CONFIG_CC_STACKPROTECTOR
179 #include <linux/stackprotector.h>
180 unsigned long __stack_chk_guard __read_mostly;
181 EXPORT_SYMBOL(__stack_chk_guard);
182 #endif
183
184 struct mips_frame_info {
185         void            *func;
186         unsigned long   func_size;
187         int             frame_size;
188         int             pc_offset;
189 };
190
191 #define J_TARGET(pc,target)     \
192                 (((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
193
194 static inline int is_ra_save_ins(union mips_instruction *ip)
195 {
196 #ifdef CONFIG_CPU_MICROMIPS
197         /*
198          * swsp ra,offset
199          * swm16 reglist,offset(sp)
200          * swm32 reglist,offset(sp)
201          * sw32 ra,offset(sp)
202          * jradiussp - NOT SUPPORTED
203          *
204          * microMIPS is way more fun...
205          */
206         if (mm_insn_16bit(ip->halfword[1])) {
207                 return (ip->mm16_r5_format.opcode == mm_swsp16_op &&
208                         ip->mm16_r5_format.rt == 31) ||
209                        (ip->mm16_m_format.opcode == mm_pool16c_op &&
210                         ip->mm16_m_format.func == mm_swm16_op);
211         }
212         else {
213                 return (ip->mm_m_format.opcode == mm_pool32b_op &&
214                         ip->mm_m_format.rd > 9 &&
215                         ip->mm_m_format.base == 29 &&
216                         ip->mm_m_format.func == mm_swm32_func) ||
217                        (ip->i_format.opcode == mm_sw32_op &&
218                         ip->i_format.rs == 29 &&
219                         ip->i_format.rt == 31);
220         }
221 #else
222         /* sw / sd $ra, offset($sp) */
223         return (ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
224                 ip->i_format.rs == 29 &&
225                 ip->i_format.rt == 31;
226 #endif
227 }
228
229 static inline int is_jump_ins(union mips_instruction *ip)
230 {
231 #ifdef CONFIG_CPU_MICROMIPS
232         /*
233          * jr16,jrc,jalr16,jalr16
234          * jal
235          * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
236          * jraddiusp - NOT SUPPORTED
237          *
238          * microMIPS is kind of more fun...
239          */
240         if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
241             (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op) ||
242             ip->j_format.opcode == mm_jal32_op)
243                 return 1;
244         if (ip->r_format.opcode != mm_pool32a_op ||
245                         ip->r_format.func != mm_pool32axf_op)
246                 return 0;
247         return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
248 #else
249         if (ip->j_format.opcode == j_op)
250                 return 1;
251         if (ip->j_format.opcode == jal_op)
252                 return 1;
253         if (ip->r_format.opcode != spec_op)
254                 return 0;
255         return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
256 #endif
257 }
258
259 static inline int is_sp_move_ins(union mips_instruction *ip)
260 {
261 #ifdef CONFIG_CPU_MICROMIPS
262         /*
263          * addiusp -imm
264          * addius5 sp,-imm
265          * addiu32 sp,sp,-imm
266          * jradiussp - NOT SUPPORTED
267          *
268          * microMIPS is not more fun...
269          */
270         if (mm_insn_16bit(ip->halfword[1])) {
271                 return (ip->mm16_r3_format.opcode == mm_pool16d_op &&
272                         ip->mm16_r3_format.simmediate && mm_addiusp_func) ||
273                        (ip->mm16_r5_format.opcode == mm_pool16d_op &&
274                         ip->mm16_r5_format.rt == 29);
275         }
276
277         return ip->mm_i_format.opcode == mm_addiu32_op &&
278                ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29;
279 #else
280         /* addiu/daddiu sp,sp,-imm */
281         if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
282                 return 0;
283         if (ip->i_format.opcode == addiu_op || ip->i_format.opcode == daddiu_op)
284                 return 1;
285 #endif
286         return 0;
287 }
288
289 static int get_frame_info(struct mips_frame_info *info)
290 {
291         bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
292         union mips_instruction insn, *ip;
293         unsigned max_insns = info->func_size / sizeof(union mips_instruction);
294         unsigned i;
295
296         info->pc_offset = -1;
297         info->frame_size = 0;
298
299         ip = (void *)msk_isa16_mode((ulong)info->func);
300         if (!ip)
301                 goto err;
302
303         if (max_insns == 0)
304                 max_insns = 128U;       /* unknown function size */
305         max_insns = min(128U, max_insns);
306
307         for (i = 0; i < max_insns; i++, ip++) {
308                 if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
309                         insn.halfword[0] = 0;
310                         insn.halfword[1] = ip->halfword[0];
311                 } else if (is_mmips) {
312                         insn.halfword[0] = ip->halfword[1];
313                         insn.halfword[1] = ip->halfword[0];
314                 } else {
315                         insn.word = ip->word;
316                 }
317
318                 if (is_jump_ins(&insn))
319                         break;
320
321                 if (!info->frame_size) {
322                         if (is_sp_move_ins(&insn))
323                         {
324 #ifdef CONFIG_CPU_MICROMIPS
325                                 if (mm_insn_16bit(ip->halfword[0]))
326                                 {
327                                         unsigned short tmp;
328
329                                         if (ip->halfword[0] & mm_addiusp_func)
330                                         {
331                                                 tmp = (((ip->halfword[0] >> 1) & 0x1ff) << 2);
332                                                 info->frame_size = -(signed short)(tmp | ((tmp & 0x100) ? 0xfe00 : 0));
333                                         } else {
334                                                 tmp = (ip->halfword[0] >> 1);
335                                                 info->frame_size = -(signed short)(tmp & 0xf);
336                                         }
337                                         ip = (void *) &ip->halfword[1];
338                                         ip--;
339                                 } else
340 #endif
341                                 info->frame_size = - ip->i_format.simmediate;
342                         }
343                         continue;
344                 }
345                 if (info->pc_offset == -1 && is_ra_save_ins(&insn)) {
346                         info->pc_offset =
347                                 ip->i_format.simmediate / sizeof(long);
348                         break;
349                 }
350         }
351         if (info->frame_size && info->pc_offset >= 0) /* nested */
352                 return 0;
353         if (info->pc_offset < 0) /* leaf */
354                 return 1;
355         /* prologue seems boggus... */
356 err:
357         return -1;
358 }
359
360 static struct mips_frame_info schedule_mfi __read_mostly;
361
362 #ifdef CONFIG_KALLSYMS
363 static unsigned long get___schedule_addr(void)
364 {
365         return kallsyms_lookup_name("__schedule");
366 }
367 #else
368 static unsigned long get___schedule_addr(void)
369 {
370         union mips_instruction *ip = (void *)schedule;
371         int max_insns = 8;
372         int i;
373
374         for (i = 0; i < max_insns; i++, ip++) {
375                 if (ip->j_format.opcode == j_op)
376                         return J_TARGET(ip, ip->j_format.target);
377         }
378         return 0;
379 }
380 #endif
381
382 static int __init frame_info_init(void)
383 {
384         unsigned long size = 0;
385 #ifdef CONFIG_KALLSYMS
386         unsigned long ofs;
387 #endif
388         unsigned long addr;
389
390         addr = get___schedule_addr();
391         if (!addr)
392                 addr = (unsigned long)schedule;
393
394 #ifdef CONFIG_KALLSYMS
395         kallsyms_lookup_size_offset(addr, &size, &ofs);
396 #endif
397         schedule_mfi.func = (void *)addr;
398         schedule_mfi.func_size = size;
399
400         get_frame_info(&schedule_mfi);
401
402         /*
403          * Without schedule() frame info, result given by
404          * thread_saved_pc() and get_wchan() are not reliable.
405          */
406         if (schedule_mfi.pc_offset < 0)
407                 printk("Can't analyze schedule() prologue at %p\n", schedule);
408
409         return 0;
410 }
411
412 arch_initcall(frame_info_init);
413
414 /*
415  * Return saved PC of a blocked thread.
416  */
417 unsigned long thread_saved_pc(struct task_struct *tsk)
418 {
419         struct thread_struct *t = &tsk->thread;
420
421         /* New born processes are a special case */
422         if (t->reg31 == (unsigned long) ret_from_fork)
423                 return t->reg31;
424         if (schedule_mfi.pc_offset < 0)
425                 return 0;
426         return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
427 }
428
429
430 #ifdef CONFIG_KALLSYMS
431 /* generic stack unwinding function */
432 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
433                                               unsigned long *sp,
434                                               unsigned long pc,
435                                               unsigned long *ra)
436 {
437         struct mips_frame_info info;
438         unsigned long size, ofs;
439         int leaf;
440         extern void ret_from_irq(void);
441         extern void ret_from_exception(void);
442
443         if (!stack_page)
444                 return 0;
445
446         /*
447          * If we reached the bottom of interrupt context,
448          * return saved pc in pt_regs.
449          */
450         if (pc == (unsigned long)ret_from_irq ||
451             pc == (unsigned long)ret_from_exception) {
452                 struct pt_regs *regs;
453                 if (*sp >= stack_page &&
454                     *sp + sizeof(*regs) <= stack_page + THREAD_SIZE - 32) {
455                         regs = (struct pt_regs *)*sp;
456                         pc = regs->cp0_epc;
457                         if (!user_mode(regs) && __kernel_text_address(pc)) {
458                                 *sp = regs->regs[29];
459                                 *ra = regs->regs[31];
460                                 return pc;
461                         }
462                 }
463                 return 0;
464         }
465         if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
466                 return 0;
467         /*
468          * Return ra if an exception occurred at the first instruction
469          */
470         if (unlikely(ofs == 0)) {
471                 pc = *ra;
472                 *ra = 0;
473                 return pc;
474         }
475
476         info.func = (void *)(pc - ofs);
477         info.func_size = ofs;   /* analyze from start to ofs */
478         leaf = get_frame_info(&info);
479         if (leaf < 0)
480                 return 0;
481
482         if (*sp < stack_page ||
483             *sp + info.frame_size > stack_page + THREAD_SIZE - 32)
484                 return 0;
485
486         if (leaf)
487                 /*
488                  * For some extreme cases, get_frame_info() can
489                  * consider wrongly a nested function as a leaf
490                  * one. In that cases avoid to return always the
491                  * same value.
492                  */
493                 pc = pc != *ra ? *ra : 0;
494         else
495                 pc = ((unsigned long *)(*sp))[info.pc_offset];
496
497         *sp += info.frame_size;
498         *ra = 0;
499         return __kernel_text_address(pc) ? pc : 0;
500 }
501 EXPORT_SYMBOL(unwind_stack_by_address);
502
503 /* used by show_backtrace() */
504 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
505                            unsigned long pc, unsigned long *ra)
506 {
507         unsigned long stack_page = (unsigned long)task_stack_page(task);
508         return unwind_stack_by_address(stack_page, sp, pc, ra);
509 }
510 #endif
511
512 /*
513  * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
514  */
515 unsigned long get_wchan(struct task_struct *task)
516 {
517         unsigned long pc = 0;
518 #ifdef CONFIG_KALLSYMS
519         unsigned long sp;
520         unsigned long ra = 0;
521 #endif
522
523         if (!task || task == current || task->state == TASK_RUNNING)
524                 goto out;
525         if (!task_stack_page(task))
526                 goto out;
527
528         pc = thread_saved_pc(task);
529
530 #ifdef CONFIG_KALLSYMS
531         sp = task->thread.reg29 + schedule_mfi.frame_size;
532
533         while (in_sched_functions(pc))
534                 pc = unwind_stack(task, &sp, pc, &ra);
535 #endif
536
537 out:
538         return pc;
539 }
540
541 /*
542  * Don't forget that the stack pointer must be aligned on a 8 bytes
543  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
544  */
545 unsigned long arch_align_stack(unsigned long sp)
546 {
547         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
548                 sp -= get_random_int() & ~PAGE_MASK;
549
550         return sp & ALMASK;
551 }
552
553 static void arch_dump_stack(void *info)
554 {
555         struct pt_regs *regs;
556
557         regs = get_irq_regs();
558
559         if (regs)
560                 show_regs(regs);
561
562         dump_stack();
563 }
564
565 void arch_trigger_all_cpu_backtrace(bool include_self)
566 {
567         smp_call_function(arch_dump_stack, NULL, 1);
568 }
569
570 int mips_get_process_fp_mode(struct task_struct *task)
571 {
572         int value = 0;
573
574         if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
575                 value |= PR_FP_MODE_FR;
576         if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
577                 value |= PR_FP_MODE_FRE;
578
579         return value;
580 }
581
582 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
583 {
584         const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
585         unsigned long switch_count;
586         struct task_struct *t;
587
588         /* Check the value is valid */
589         if (value & ~known_bits)
590                 return -EOPNOTSUPP;
591
592         /* Avoid inadvertently triggering emulation */
593         if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
594             !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
595                 return -EOPNOTSUPP;
596         if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
597                 return -EOPNOTSUPP;
598
599         /* FR = 0 not supported in MIPS R6 */
600         if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
601                 return -EOPNOTSUPP;
602
603         /* Proceed with the mode switch */
604         preempt_disable();
605
606         /* Save FP & vector context, then disable FPU & MSA */
607         if (task->signal == current->signal)
608                 lose_fpu(1);
609
610         /* Prevent any threads from obtaining live FP context */
611         atomic_set(&task->mm->context.fp_mode_switching, 1);
612         smp_mb__after_atomic();
613
614         /*
615          * If there are multiple online CPUs then wait until all threads whose
616          * FP mode is about to change have been context switched. This approach
617          * allows us to only worry about whether an FP mode switch is in
618          * progress when FP is first used in a tasks time slice. Pretty much all
619          * of the mode switch overhead can thus be confined to cases where mode
620          * switches are actually occuring. That is, to here. However for the
621          * thread performing the mode switch it may take a while...
622          */
623         if (num_online_cpus() > 1) {
624                 spin_lock_irq(&task->sighand->siglock);
625
626                 for_each_thread(task, t) {
627                         if (t == current)
628                                 continue;
629
630                         switch_count = t->nvcsw + t->nivcsw;
631
632                         do {
633                                 spin_unlock_irq(&task->sighand->siglock);
634                                 cond_resched();
635                                 spin_lock_irq(&task->sighand->siglock);
636                         } while ((t->nvcsw + t->nivcsw) == switch_count);
637                 }
638
639                 spin_unlock_irq(&task->sighand->siglock);
640         }
641
642         /*
643          * There are now no threads of the process with live FP context, so it
644          * is safe to proceed with the FP mode switch.
645          */
646         for_each_thread(task, t) {
647                 /* Update desired FP register width */
648                 if (value & PR_FP_MODE_FR) {
649                         clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
650                 } else {
651                         set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
652                         clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
653                 }
654
655                 /* Update desired FP single layout */
656                 if (value & PR_FP_MODE_FRE)
657                         set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
658                 else
659                         clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
660         }
661
662         /* Allow threads to use FP again */
663         atomic_set(&task->mm->context.fp_mode_switching, 0);
664         preempt_enable();
665
666         return 0;
667 }