]> git.karo-electronics.de Git - karo-tx-linux.git/blob - arch/mips/kernel/process.c
MIPS: Fix get_frame_info() handling of microMIPS function size
[karo-tx-linux.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/dsemul.h>
34 #include <asm/dsp.h>
35 #include <asm/fpu.h>
36 #include <asm/irq.h>
37 #include <asm/msa.h>
38 #include <asm/pgtable.h>
39 #include <asm/mipsregs.h>
40 #include <asm/processor.h>
41 #include <asm/reg.h>
42 #include <linux/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/elf.h>
45 #include <asm/isadep.h>
46 #include <asm/inst.h>
47 #include <asm/stacktrace.h>
48 #include <asm/irq_regs.h>
49
50 #ifdef CONFIG_HOTPLUG_CPU
51 void arch_cpu_idle_dead(void)
52 {
53         /* What the heck is this check doing ? */
54         if (!cpumask_test_cpu(smp_processor_id(), &cpu_callin_map))
55                 play_dead();
56 }
57 #endif
58
59 asmlinkage void ret_from_fork(void);
60 asmlinkage void ret_from_kernel_thread(void);
61
62 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
63 {
64         unsigned long status;
65
66         /* New thread loses kernel privileges. */
67         status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
68         status |= KU_USER;
69         regs->cp0_status = status;
70         lose_fpu(0);
71         clear_thread_flag(TIF_MSA_CTX_LIVE);
72         clear_used_math();
73         atomic_set(&current->thread.bd_emu_frame, BD_EMUFRAME_NONE);
74         init_dsp();
75         regs->cp0_epc = pc;
76         regs->regs[29] = sp;
77 }
78
79 void exit_thread(struct task_struct *tsk)
80 {
81         /*
82          * User threads may have allocated a delay slot emulation frame.
83          * If so, clean up that allocation.
84          */
85         if (!(current->flags & PF_KTHREAD))
86                 dsemul_thread_cleanup(tsk);
87 }
88
89 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
90 {
91         /*
92          * Save any process state which is live in hardware registers to the
93          * parent context prior to duplication. This prevents the new child
94          * state becoming stale if the parent is preempted before copy_thread()
95          * gets a chance to save the parent's live hardware registers to the
96          * child context.
97          */
98         preempt_disable();
99
100         if (is_msa_enabled())
101                 save_msa(current);
102         else if (is_fpu_owner())
103                 _save_fp(current);
104
105         save_dsp(current);
106
107         preempt_enable();
108
109         *dst = *src;
110         return 0;
111 }
112
113 /*
114  * Copy architecture-specific thread state
115  */
116 int copy_thread(unsigned long clone_flags, unsigned long usp,
117         unsigned long kthread_arg, struct task_struct *p)
118 {
119         struct thread_info *ti = task_thread_info(p);
120         struct pt_regs *childregs, *regs = current_pt_regs();
121         unsigned long childksp;
122         p->set_child_tid = p->clear_child_tid = NULL;
123
124         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
125
126         /* set up new TSS. */
127         childregs = (struct pt_regs *) childksp - 1;
128         /*  Put the stack after the struct pt_regs.  */
129         childksp = (unsigned long) childregs;
130         p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
131         if (unlikely(p->flags & PF_KTHREAD)) {
132                 /* kernel thread */
133                 unsigned long status = p->thread.cp0_status;
134                 memset(childregs, 0, sizeof(struct pt_regs));
135                 ti->addr_limit = KERNEL_DS;
136                 p->thread.reg16 = usp; /* fn */
137                 p->thread.reg17 = kthread_arg;
138                 p->thread.reg29 = childksp;
139                 p->thread.reg31 = (unsigned long) ret_from_kernel_thread;
140 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
141                 status = (status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
142                          ((status & (ST0_KUC | ST0_IEC)) << 2);
143 #else
144                 status |= ST0_EXL;
145 #endif
146                 childregs->cp0_status = status;
147                 return 0;
148         }
149
150         /* user thread */
151         *childregs = *regs;
152         childregs->regs[7] = 0; /* Clear error flag */
153         childregs->regs[2] = 0; /* Child gets zero as return value */
154         if (usp)
155                 childregs->regs[29] = usp;
156         ti->addr_limit = USER_DS;
157
158         p->thread.reg29 = (unsigned long) childregs;
159         p->thread.reg31 = (unsigned long) ret_from_fork;
160
161         /*
162          * New tasks lose permission to use the fpu. This accelerates context
163          * switching for most programs since they don't use the fpu.
164          */
165         childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
166
167         clear_tsk_thread_flag(p, TIF_USEDFPU);
168         clear_tsk_thread_flag(p, TIF_USEDMSA);
169         clear_tsk_thread_flag(p, TIF_MSA_CTX_LIVE);
170
171 #ifdef CONFIG_MIPS_MT_FPAFF
172         clear_tsk_thread_flag(p, TIF_FPUBOUND);
173 #endif /* CONFIG_MIPS_MT_FPAFF */
174
175         atomic_set(&p->thread.bd_emu_frame, BD_EMUFRAME_NONE);
176
177         if (clone_flags & CLONE_SETTLS)
178                 ti->tp_value = regs->regs[7];
179
180         return 0;
181 }
182
183 #ifdef CONFIG_CC_STACKPROTECTOR
184 #include <linux/stackprotector.h>
185 unsigned long __stack_chk_guard __read_mostly;
186 EXPORT_SYMBOL(__stack_chk_guard);
187 #endif
188
189 struct mips_frame_info {
190         void            *func;
191         unsigned long   func_size;
192         int             frame_size;
193         int             pc_offset;
194 };
195
196 #define J_TARGET(pc,target)     \
197                 (((unsigned long)(pc) & 0xf0000000) | ((target) << 2))
198
199 static inline int is_ra_save_ins(union mips_instruction *ip)
200 {
201 #ifdef CONFIG_CPU_MICROMIPS
202         /*
203          * swsp ra,offset
204          * swm16 reglist,offset(sp)
205          * swm32 reglist,offset(sp)
206          * sw32 ra,offset(sp)
207          * jradiussp - NOT SUPPORTED
208          *
209          * microMIPS is way more fun...
210          */
211         if (mm_insn_16bit(ip->halfword[1])) {
212                 return (ip->mm16_r5_format.opcode == mm_swsp16_op &&
213                         ip->mm16_r5_format.rt == 31) ||
214                        (ip->mm16_m_format.opcode == mm_pool16c_op &&
215                         ip->mm16_m_format.func == mm_swm16_op);
216         }
217         else {
218                 return (ip->mm_m_format.opcode == mm_pool32b_op &&
219                         ip->mm_m_format.rd > 9 &&
220                         ip->mm_m_format.base == 29 &&
221                         ip->mm_m_format.func == mm_swm32_func) ||
222                        (ip->i_format.opcode == mm_sw32_op &&
223                         ip->i_format.rs == 29 &&
224                         ip->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         if ((ip->mm16_r5_format.opcode == mm_pool16c_op &&
246             (ip->mm16_r5_format.rt & mm_jr16_op) == mm_jr16_op) ||
247             ip->j_format.opcode == mm_jal32_op)
248                 return 1;
249         if (ip->r_format.opcode != mm_pool32a_op ||
250                         ip->r_format.func != mm_pool32axf_op)
251                 return 0;
252         return ((ip->u_format.uimmediate >> 6) & mm_jalr_op) == mm_jalr_op;
253 #else
254         if (ip->j_format.opcode == j_op)
255                 return 1;
256         if (ip->j_format.opcode == jal_op)
257                 return 1;
258         if (ip->r_format.opcode != spec_op)
259                 return 0;
260         return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
261 #endif
262 }
263
264 static inline int is_sp_move_ins(union mips_instruction *ip)
265 {
266 #ifdef CONFIG_CPU_MICROMIPS
267         /*
268          * addiusp -imm
269          * addius5 sp,-imm
270          * addiu32 sp,sp,-imm
271          * jradiussp - NOT SUPPORTED
272          *
273          * microMIPS is not more fun...
274          */
275         if (mm_insn_16bit(ip->halfword[1])) {
276                 return (ip->mm16_r3_format.opcode == mm_pool16d_op &&
277                         ip->mm16_r3_format.simmediate && mm_addiusp_func) ||
278                        (ip->mm16_r5_format.opcode == mm_pool16d_op &&
279                         ip->mm16_r5_format.rt == 29);
280         }
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         bool is_mmips = IS_ENABLED(CONFIG_CPU_MICROMIPS);
297         union mips_instruction insn, *ip, *ip_end;
298         const unsigned int max_insns = 128;
299         unsigned int i;
300
301         info->pc_offset = -1;
302         info->frame_size = 0;
303
304         ip = (void *)msk_isa16_mode((ulong)info->func);
305         if (!ip)
306                 goto err;
307
308         ip_end = (void *)ip + info->func_size;
309
310         for (i = 0; i < max_insns && ip < ip_end; i++, ip++) {
311                 if (is_mmips && mm_insn_16bit(ip->halfword[0])) {
312                         insn.halfword[0] = 0;
313                         insn.halfword[1] = ip->halfword[0];
314                 } else if (is_mmips) {
315                         insn.halfword[0] = ip->halfword[1];
316                         insn.halfword[1] = ip->halfword[0];
317                 } else {
318                         insn.word = ip->word;
319                 }
320
321                 if (is_jump_ins(&insn))
322                         break;
323
324                 if (!info->frame_size) {
325                         if (is_sp_move_ins(&insn))
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(&insn)) {
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 bogus... */
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 (!user_mode(regs) && __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 = 0;
511         int cpu;
512
513         for_each_possible_cpu(cpu) {
514                 if (on_irq_stack(cpu, *sp)) {
515                         stack_page = (unsigned long)irq_stack[cpu];
516                         break;
517                 }
518         }
519
520         if (!stack_page)
521                 stack_page = (unsigned long)task_stack_page(task);
522
523         return unwind_stack_by_address(stack_page, sp, pc, ra);
524 }
525 #endif
526
527 /*
528  * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
529  */
530 unsigned long get_wchan(struct task_struct *task)
531 {
532         unsigned long pc = 0;
533 #ifdef CONFIG_KALLSYMS
534         unsigned long sp;
535         unsigned long ra = 0;
536 #endif
537
538         if (!task || task == current || task->state == TASK_RUNNING)
539                 goto out;
540         if (!task_stack_page(task))
541                 goto out;
542
543         pc = thread_saved_pc(task);
544
545 #ifdef CONFIG_KALLSYMS
546         sp = task->thread.reg29 + schedule_mfi.frame_size;
547
548         while (in_sched_functions(pc))
549                 pc = unwind_stack(task, &sp, pc, &ra);
550 #endif
551
552 out:
553         return pc;
554 }
555
556 /*
557  * Don't forget that the stack pointer must be aligned on a 8 bytes
558  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
559  */
560 unsigned long arch_align_stack(unsigned long sp)
561 {
562         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
563                 sp -= get_random_int() & ~PAGE_MASK;
564
565         return sp & ALMASK;
566 }
567
568 static void arch_dump_stack(void *info)
569 {
570         struct pt_regs *regs;
571
572         regs = get_irq_regs();
573
574         if (regs)
575                 show_regs(regs);
576
577         dump_stack();
578 }
579
580 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
581 {
582         long this_cpu = get_cpu();
583
584         if (cpumask_test_cpu(this_cpu, mask) && !exclude_self)
585                 dump_stack();
586
587         smp_call_function_many(mask, arch_dump_stack, NULL, 1);
588
589         put_cpu();
590 }
591
592 int mips_get_process_fp_mode(struct task_struct *task)
593 {
594         int value = 0;
595
596         if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
597                 value |= PR_FP_MODE_FR;
598         if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
599                 value |= PR_FP_MODE_FRE;
600
601         return value;
602 }
603
604 static void prepare_for_fp_mode_switch(void *info)
605 {
606         struct mm_struct *mm = info;
607
608         if (current->mm == mm)
609                 lose_fpu(1);
610 }
611
612 int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
613 {
614         const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
615         struct task_struct *t;
616         int max_users;
617
618         /* Check the value is valid */
619         if (value & ~known_bits)
620                 return -EOPNOTSUPP;
621
622         /* Avoid inadvertently triggering emulation */
623         if ((value & PR_FP_MODE_FR) && raw_cpu_has_fpu &&
624             !(raw_current_cpu_data.fpu_id & MIPS_FPIR_F64))
625                 return -EOPNOTSUPP;
626         if ((value & PR_FP_MODE_FRE) && raw_cpu_has_fpu && !cpu_has_fre)
627                 return -EOPNOTSUPP;
628
629         /* FR = 0 not supported in MIPS R6 */
630         if (!(value & PR_FP_MODE_FR) && raw_cpu_has_fpu && cpu_has_mips_r6)
631                 return -EOPNOTSUPP;
632
633         /* Proceed with the mode switch */
634         preempt_disable();
635
636         /* Save FP & vector context, then disable FPU & MSA */
637         if (task->signal == current->signal)
638                 lose_fpu(1);
639
640         /* Prevent any threads from obtaining live FP context */
641         atomic_set(&task->mm->context.fp_mode_switching, 1);
642         smp_mb__after_atomic();
643
644         /*
645          * If there are multiple online CPUs then force any which are running
646          * threads in this process to lose their FPU context, which they can't
647          * regain until fp_mode_switching is cleared later.
648          */
649         if (num_online_cpus() > 1) {
650                 /* No need to send an IPI for the local CPU */
651                 max_users = (task->mm == current->mm) ? 1 : 0;
652
653                 if (atomic_read(&current->mm->mm_users) > max_users)
654                         smp_call_function(prepare_for_fp_mode_switch,
655                                           (void *)current->mm, 1);
656         }
657
658         /*
659          * There are now no threads of the process with live FP context, so it
660          * is safe to proceed with the FP mode switch.
661          */
662         for_each_thread(task, t) {
663                 /* Update desired FP register width */
664                 if (value & PR_FP_MODE_FR) {
665                         clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
666                 } else {
667                         set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
668                         clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
669                 }
670
671                 /* Update desired FP single layout */
672                 if (value & PR_FP_MODE_FRE)
673                         set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
674                 else
675                         clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
676         }
677
678         /* Allow threads to use FP again */
679         atomic_set(&task->mm->context.fp_mode_switching, 0);
680         preempt_enable();
681
682         return 0;
683 }