Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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         union mips_instruction mmi;
198
199         /*
200          * swsp ra,offset
201          * swm16 reglist,offset(sp)
202          * swm32 reglist,offset(sp)
203          * sw32 ra,offset(sp)
204          * jradiussp - NOT SUPPORTED
205          *
206          * microMIPS is way more fun...
207          */
208         if (mm_insn_16bit(ip->halfword[0])) {
209                 mmi.word = (ip->halfword[0] << 16);
210                 return (mmi.mm16_r5_format.opcode == mm_swsp16_op &&
211                         mmi.mm16_r5_format.rt == 31) ||
212                        (mmi.mm16_m_format.opcode == mm_pool16c_op &&
213                         mmi.mm16_m_format.func == mm_swm16_op);
214         }
215         else {
216                 mmi.halfword[0] = ip->halfword[1];
217                 mmi.halfword[1] = ip->halfword[0];
218                 return (mmi.mm_m_format.opcode == mm_pool32b_op &&
219                         mmi.mm_m_format.rd > 9 &&
220                         mmi.mm_m_format.base == 29 &&
221                         mmi.mm_m_format.func == mm_swm32_func) ||
222                        (mmi.i_format.opcode == mm_sw32_op &&
223                         mmi.i_format.rs == 29 &&
224                         mmi.i_format.rt == 31);
225         }
226 #else
227         /* sw / sd $ra, offset($sp) */
228         return (ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
229                 ip->i_format.rs == 29 &&
230                 ip->i_format.rt == 31;
231 #endif
232 }
233
234 static inline int is_jump_ins(union mips_instruction *ip)
235 {
236 #ifdef CONFIG_CPU_MICROMIPS
237         /*
238          * jr16,jrc,jalr16,jalr16
239          * jal
240          * jalr/jr,jalr.hb/jr.hb,jalrs,jalrs.hb
241          * jraddiusp - NOT SUPPORTED
242          *
243          * microMIPS is kind of more fun...
244          */
245         union mips_instruction mmi;
246
247         mmi.word = (ip->halfword[0] << 16);
248
249         if ((mmi.mm16_r5_format.opcode == mm_pool16c_op &&
250             (mmi.mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op) ||
251             ip->j_format.opcode == mm_jal32_op)
252                 return 1;
253         if (ip->r_format.opcode != mm_pool32a_op ||
254                         ip->r_format.func != mm_pool32axf_op)
255                 return 0;
256         return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
257 #else
258         if (ip->j_format.opcode == j_op)
259                 return 1;
260         if (ip->j_format.opcode == jal_op)
261                 return 1;
262         if (ip->r_format.opcode != spec_op)
263                 return 0;
264         return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
265 #endif
266 }
267
268 static inline int is_sp_move_ins(union mips_instruction *ip)
269 {
270 #ifdef CONFIG_CPU_MICROMIPS
271         /*
272          * addiusp -imm
273          * addius5 sp,-imm
274          * addiu32 sp,sp,-imm
275          * jradiussp - NOT SUPPORTED
276          *
277          * microMIPS is not more fun...
278          */
279         if (mm_insn_16bit(ip->halfword[0])) {
280                 union mips_instruction mmi;
281
282                 mmi.word = (ip->halfword[0] << 16);
283                 return (mmi.mm16_r3_format.opcode == mm_pool16d_op &&
284                         mmi.mm16_r3_format.simmediate && mm_addiusp_func) ||
285                        (mmi.mm16_r5_format.opcode == mm_pool16d_op &&
286                         mmi.mm16_r5_format.rt == 29);
287         }
288         return ip->mm_i_format.opcode == mm_addiu32_op &&
289                ip->mm_i_format.rt == 29 && ip->mm_i_format.rs == 29;
290 #else
291         /* addiu/daddiu sp,sp,-imm */
292         if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
293                 return 0;
294         if (ip->i_format.opcode == addiu_op || ip->i_format.opcode == daddiu_op)
295                 return 1;
296 #endif
297         return 0;
298 }
299
300 static int get_frame_info(struct mips_frame_info *info)
301 {
302 #ifdef CONFIG_CPU_MICROMIPS
303         union mips_instruction *ip = (void *) (((char *) info->func) - 1);
304 #else
305         union mips_instruction *ip = info->func;
306 #endif
307         unsigned max_insns = info->func_size / sizeof(union mips_instruction);
308         unsigned i;
309
310         info->pc_offset = -1;
311         info->frame_size = 0;
312
313         if (!ip)
314                 goto err;
315
316         if (max_insns == 0)
317                 max_insns = 128U;       /* unknown function size */
318         max_insns = min(128U, max_insns);
319
320         for (i = 0; i < max_insns; i++, ip++) {
321
322                 if (is_jump_ins(ip))
323                         break;
324                 if (!info->frame_size) {
325                         if (is_sp_move_ins(ip))
326                         {
327 #ifdef CONFIG_CPU_MICROMIPS
328                                 if (mm_insn_16bit(ip->halfword[0]))
329                                 {
330                                         unsigned short tmp;
331
332                                         if (ip->halfword[0] & mm_addiusp_func)
333                                         {
334                                                 tmp = (((ip->halfword[0] >> 1) & 0x1ff) << 2);
335                                                 info->frame_size = -(signed short)(tmp | ((tmp & 0x100) ? 0xfe00 : 0));
336                                         } else {
337                                                 tmp = (ip->halfword[0] >> 1);
338                                                 info->frame_size = -(signed short)(tmp & 0xf);
339                                         }
340                                         ip = (void *) &ip->halfword[1];
341                                         ip--;
342                                 } else
343 #endif
344                                 info->frame_size = - ip->i_format.simmediate;
345                         }
346                         continue;
347                 }
348                 if (info->pc_offset == -1 && is_ra_save_ins(ip)) {
349                         info->pc_offset =
350                                 ip->i_format.simmediate / sizeof(long);
351                         break;
352                 }
353         }
354         if (info->frame_size && info->pc_offset >= 0) /* nested */
355                 return 0;
356         if (info->pc_offset < 0) /* leaf */
357                 return 1;
358         /* prologue seems boggus... */
359 err:
360         return -1;
361 }
362
363 static struct mips_frame_info schedule_mfi __read_mostly;
364
365 #ifdef CONFIG_KALLSYMS
366 static unsigned long get___schedule_addr(void)
367 {
368         return kallsyms_lookup_name("__schedule");
369 }
370 #else
371 static unsigned long get___schedule_addr(void)
372 {
373         union mips_instruction *ip = (void *)schedule;
374         int max_insns = 8;
375         int i;
376
377         for (i = 0; i < max_insns; i++, ip++) {
378                 if (ip->j_format.opcode == j_op)
379                         return J_TARGET(ip, ip->j_format.target);
380         }
381         return 0;
382 }
383 #endif
384
385 static int __init frame_info_init(void)
386 {
387         unsigned long size = 0;
388 #ifdef CONFIG_KALLSYMS
389         unsigned long ofs;
390 #endif
391         unsigned long addr;
392
393         addr = get___schedule_addr();
394         if (!addr)
395                 addr = (unsigned long)schedule;
396
397 #ifdef CONFIG_KALLSYMS
398         kallsyms_lookup_size_offset(addr, &size, &ofs);
399 #endif
400         schedule_mfi.func = (void *)addr;
401         schedule_mfi.func_size = size;
402
403         get_frame_info(&schedule_mfi);
404
405         /*
406          * Without schedule() frame info, result given by
407          * thread_saved_pc() and get_wchan() are not reliable.
408          */
409         if (schedule_mfi.pc_offset < 0)
410                 printk("Can't analyze schedule() prologue at %p\n", schedule);
411
412         return 0;
413 }
414
415 arch_initcall(frame_info_init);
416
417 /*
418  * Return saved PC of a blocked thread.
419  */
420 unsigned long thread_saved_pc(struct task_struct *tsk)
421 {
422         struct thread_struct *t = &tsk->thread;
423
424         /* New born processes are a special case */
425         if (t->reg31 == (unsigned long) ret_from_fork)
426                 return t->reg31;
427         if (schedule_mfi.pc_offset < 0)
428                 return 0;
429         return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
430 }
431
432
433 #ifdef CONFIG_KALLSYMS
434 /* generic stack unwinding function */
435 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
436                                               unsigned long *sp,
437                                               unsigned long pc,
438                                               unsigned long *ra)
439 {
440         struct mips_frame_info info;
441         unsigned long size, ofs;
442         int leaf;
443         extern void ret_from_irq(void);
444         extern void ret_from_exception(void);
445
446         if (!stack_page)
447                 return 0;
448
449         /*
450          * If we reached the bottom of interrupt context,
451          * return saved pc in pt_regs.
452          */
453         if (pc == (unsigned long)ret_from_irq ||
454             pc == (unsigned long)ret_from_exception) {
455                 struct pt_regs *regs;
456                 if (*sp >= stack_page &&
457                     *sp + sizeof(*regs) <= stack_page + THREAD_SIZE - 32) {
458                         regs = (struct pt_regs *)*sp;
459                         pc = regs->cp0_epc;
460                         if (__kernel_text_address(pc)) {
461                                 *sp = regs->regs[29];
462                                 *ra = regs->regs[31];
463                                 return pc;
464                         }
465                 }
466                 return 0;
467         }
468         if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
469                 return 0;
470         /*
471          * Return ra if an exception occurred at the first instruction
472          */
473         if (unlikely(ofs == 0)) {
474                 pc = *ra;
475                 *ra = 0;
476                 return pc;
477         }
478
479         info.func = (void *)(pc - ofs);
480         info.func_size = ofs;   /* analyze from start to ofs */
481         leaf = get_frame_info(&info);
482         if (leaf < 0)
483                 return 0;
484
485         if (*sp < stack_page ||
486             *sp + info.frame_size > stack_page + THREAD_SIZE - 32)
487                 return 0;
488
489         if (leaf)
490                 /*
491                  * For some extreme cases, get_frame_info() can
492                  * consider wrongly a nested function as a leaf
493                  * one. In that cases avoid to return always the
494                  * same value.
495                  */
496                 pc = pc != *ra ? *ra : 0;
497         else
498                 pc = ((unsigned long *)(*sp))[info.pc_offset];
499
500         *sp += info.frame_size;
501         *ra = 0;
502         return __kernel_text_address(pc) ? pc : 0;
503 }
504 EXPORT_SYMBOL(unwind_stack_by_address);
505
506 /* used by show_backtrace() */
507 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
508                            unsigned long pc, unsigned long *ra)
509 {
510         unsigned long stack_page = (unsigned long)task_stack_page(task);
511         return unwind_stack_by_address(stack_page, sp, pc, ra);
512 }
513 #endif
514
515 /*
516  * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
517  */
518 unsigned long get_wchan(struct task_struct *task)
519 {
520         unsigned long pc = 0;
521 #ifdef CONFIG_KALLSYMS
522         unsigned long sp;
523         unsigned long ra = 0;
524 #endif
525
526         if (!task || task == current || task->state == TASK_RUNNING)
527                 goto out;
528         if (!task_stack_page(task))
529                 goto out;
530
531         pc = thread_saved_pc(task);
532
533 #ifdef CONFIG_KALLSYMS
534         sp = task->thread.reg29 + schedule_mfi.frame_size;
535
536         while (in_sched_functions(pc))
537                 pc = unwind_stack(task, &sp, pc, &ra);
538 #endif
539
540 out:
541         return pc;
542 }
543
544 /*
545  * Don't forget that the stack pointer must be aligned on a 8 bytes
546  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
547  */
548 unsigned long arch_align_stack(unsigned long sp)
549 {
550         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
551                 sp -= get_random_int() & ~PAGE_MASK;
552
553         return sp & ALMASK;
554 }
555
556 static void arch_dump_stack(void *info)
557 {
558         struct pt_regs *regs;
559
560         regs = get_irq_regs();
561
562         if (regs)
563                 show_regs(regs);
564
565         dump_stack();
566 }
567
568 void arch_trigger_all_cpu_backtrace(bool include_self)
569 {
570         smp_call_function(arch_dump_stack, NULL, 1);
571 }
572
573 int mips_get_process_fp_mode(struct task_struct *task)
574 {
575         int value = 0;
576
577         if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
578                 value |= PR_FP_MODE_FR;
579         if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
580                 value |= PR_FP_MODE_FRE;
581
582         return value;
583 }
584
585 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
586 {
587         const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
588         unsigned long switch_count;
589         struct task_struct *t;
590
591         /* Check the value is valid */
592         if (value & ~known_bits)
593                 return -EOPNOTSUPP;
594
595         /* Avoid inadvertently triggering emulation */
596         if ((value & PR_FP_MODE_FR) && cpu_has_fpu &&
597             !(current_cpu_data.fpu_id & MIPS_FPIR_F64))
598                 return -EOPNOTSUPP;
599         if ((value & PR_FP_MODE_FRE) && cpu_has_fpu && !cpu_has_fre)
600                 return -EOPNOTSUPP;
601
602         /* FR = 0 not supported in MIPS R6 */
603         if (!(value & PR_FP_MODE_FR) && cpu_has_fpu && cpu_has_mips_r6)
604                 return -EOPNOTSUPP;
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
665         return 0;
666 }