arch/um/kernel/process.c

   1 /*
   2  * Copyright (C) 2015 Anton Ivanov (aivanov@{brocade.com,kot-begemot.co.uk})
   3  * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
   4  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   5  * Copyright 2003 PathScale, Inc.
   6  * Licensed under the GPL
   7  */
   8
   9 #include <linux/stddef.h>
  10 #include <linux/err.h>
  11 #include <linux/hardirq.h>
  12 #include <linux/mm.h>
  13 #include <linux/module.h>
  14 #include <linux/personality.h>
  15 #include <linux/proc_fs.h>
  16 #include <linux/ptrace.h>
  17 #include <linux/random.h>
  18 #include <linux/slab.h>
  19 #include <linux/sched.h>
  20 #include <linux/sched/debug.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/tick.h>
  25 #include <linux/threads.h>
  26 #include <linux/tracehook.h>
  27 #include <asm/current.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/mmu_context.h>
  30 #include <linux/uaccess.h>
  31 #include <as-layout.h>
  32 #include <kern_util.h>
  33 #include <os.h>
  34 #include <skas.h>
  35 #include <timer-internal.h>
  36
  37 /*
  38  * This is a per-cpu array.  A processor only modifies its entry and it only
  39  * cares about its entry, so it's OK if another processor is modifying its
  40  * entry.
  41  */
  42 struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } };
  43
  44 static inline int external_pid(void)
  45 {
  46         /* FIXME: Need to look up userspace_pid by cpu */
  47         return userspace_pid[0];
  48 }
  49
  50 int pid_to_processor_id(int pid)
  51 {
  52         int i;
  53
  54         for (i = 0; i < ncpus; i++) {
  55                 if (cpu_tasks[i].pid == pid)
  56                         return i;
  57         }
  58         return -1;
  59 }
  60
  61 void free_stack(unsigned long stack, int order)
  62 {
  63         free_pages(stack, order);
  64 }
  65
  66 unsigned long alloc_stack(int order, int atomic)
  67 {
  68         unsigned long page;
  69         gfp_t flags = GFP_KERNEL;
  70
  71         if (atomic)
  72                 flags = GFP_ATOMIC;
  73         page = __get_free_pages(flags, order);
  74
  75         return page;
  76 }
  77
  78 static inline void set_current(struct task_struct *task)
  79 {
  80         cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task)
  81                 { external_pid(), task });
  82 }
  83
  84 extern void arch_switch_to(struct task_struct *to);
  85
  86 void *__switch_to(struct task_struct *from, struct task_struct *to)
  87 {
  88         to->thread.prev_sched = from;
  89         set_current(to);
  90
  91         switch_threads(&from->thread.switch_buf, &to->thread.switch_buf);
  92         arch_switch_to(current);
  93
  94         return current->thread.prev_sched;
  95 }
  96
  97 void interrupt_end(void)
  98 {
  99         struct pt_regs *regs = &current->thread.regs;
 100
 101         if (need_resched())
 102                 schedule();
 103         if (test_thread_flag(TIF_SIGPENDING))
 104                 do_signal(regs);
 105         if (test_and_clear_thread_flag(TIF_NOTIFY_RESUME))
 106                 tracehook_notify_resume(regs);
 107 }
 108
 109 int get_current_pid(void)
 110 {
 111         return task_pid_nr(current);
 112 }
 113
 114 /*
 115  * This is called magically, by its address being stuffed in a jmp_buf
 116  * and being longjmp-d to.
 117  */
 118 void new_thread_handler(void)
 119 {
 120         int (*fn)(void *), n;
 121         void *arg;
 122
 123         if (current->thread.prev_sched != NULL)
 124                 schedule_tail(current->thread.prev_sched);
 125         current->thread.prev_sched = NULL;
 126
 127         fn = current->thread.request.u.thread.proc;
 128         arg = current->thread.request.u.thread.arg;
 129
 130         /*
 131          * callback returns only if the kernel thread execs a process
 132          */
 133         n = fn(arg);
 134         userspace(&current->thread.regs.regs, current_thread_info()->aux_fp_regs);
 135 }
 136
 137 /* Called magically, see new_thread_handler above */
 138 void fork_handler(void)
 139 {
 140         force_flush_all();
 141
 142         schedule_tail(current->thread.prev_sched);
 143
 144         /*
 145          * XXX: if interrupt_end() calls schedule, this call to
 146          * arch_switch_to isn't needed. We could want to apply this to
 147          * improve performance. -bb
 148          */
 149         arch_switch_to(current);
 150
 151         current->thread.prev_sched = NULL;
 152
 153         userspace(&current->thread.regs.regs, current_thread_info()->aux_fp_regs);
 154 }
 155
 156 int copy_thread(unsigned long clone_flags, unsigned long sp,
 157                 unsigned long arg, struct task_struct * p)
 158 {
 159         void (*handler)(void);
 160         int kthread = current->flags & PF_KTHREAD;
 161         int ret = 0;
 162
 163         p->thread = (struct thread_struct) INIT_THREAD;
 164
 165         if (!kthread) {
 166                 memcpy(&p->thread.regs.regs, current_pt_regs(),
 167                        sizeof(p->thread.regs.regs));
 168                 PT_REGS_SET_SYSCALL_RETURN(&p->thread.regs, 0);
 169                 if (sp != 0)
 170                         REGS_SP(p->thread.regs.regs.gp) = sp;
 171
 172                 handler = fork_handler;
 173
 174                 arch_copy_thread(&current->thread.arch, &p->thread.arch);
 175         } else {
 176                 get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp);
 177                 p->thread.request.u.thread.proc = (int (*)(void *))sp;
 178                 p->thread.request.u.thread.arg = (void *)arg;
 179                 handler = new_thread_handler;
 180         }
 181
 182         new_thread(task_stack_page(p), &p->thread.switch_buf, handler);
 183
 184         if (!kthread) {
 185                 clear_flushed_tls(p);
 186
 187                 /*
 188                  * Set a new TLS for the child thread?
 189                  */
 190                 if (clone_flags & CLONE_SETTLS)
 191                         ret = arch_copy_tls(p);
 192         }
 193
 194         return ret;
 195 }
 196
 197 void initial_thread_cb(void (*proc)(void *), void *arg)
 198 {
 199         int save_kmalloc_ok = kmalloc_ok;
 200
 201         kmalloc_ok = 0;
 202         initial_thread_cb_skas(proc, arg);
 203         kmalloc_ok = save_kmalloc_ok;
 204 }
 205
 206 static void time_travel_sleep(unsigned long long duration)
 207 {
 208         unsigned long long next = time_travel_time + duration;
 209
 210         if (time_travel_mode != TT_MODE_INFCPU)
 211                 os_timer_disable();
 212
 213         if (time_travel_timer_mode != TT_TMR_DISABLED ||
 214             time_travel_timer_expiry < next) {
 215                 if (time_travel_timer_mode == TT_TMR_ONESHOT)
 216                         time_travel_set_timer_mode(TT_TMR_DISABLED);
 217                 /*
 218                  * time_travel_time will be adjusted in the timer
 219                  * IRQ handler so it works even when the signal
 220                  * comes from the OS timer
 221                  */
 222                 deliver_alarm();
 223         } else {
 224                 time_travel_set_time(next);
 225         }
 226
 227         if (time_travel_mode != TT_MODE_INFCPU) {
 228                 if (time_travel_timer_mode == TT_TMR_PERIODIC)
 229                         os_timer_set_interval(time_travel_timer_interval);
 230                 else if (time_travel_timer_mode == TT_TMR_ONESHOT)
 231                         os_timer_one_shot(time_travel_timer_expiry - next);
 232         }
 233 }
 234
 235 static void um_idle_sleep(void)
 236 {
 237         unsigned long long duration = UM_NSEC_PER_SEC;
 238
 239         if (time_travel_mode != TT_MODE_OFF) {
 240                 time_travel_sleep(duration);
 241         } else {
 242                 os_idle_sleep(duration);
 243         }
 244 }
 245
 246 void arch_cpu_idle(void)
 247 {
 248         cpu_tasks[current_thread_info()->cpu].pid = os_getpid();
 249         um_idle_sleep();
 250         local_irq_enable();
 251 }
 252
 253 int __cant_sleep(void) {
 254         return in_atomic() || irqs_disabled() || in_interrupt();
 255         /* Is in_interrupt() really needed? */
 256 }
 257
 258 int user_context(unsigned long sp)
 259 {
 260         unsigned long stack;
 261
 262         stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER);
 263         return stack != (unsigned long) current_thread_info();
 264 }
 265
 266 extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end;
 267
 268 void do_uml_exitcalls(void)
 269 {
 270         exitcall_t *call;
 271
 272         call = &__uml_exitcall_end;
 273         while (--call >= &__uml_exitcall_begin)
 274                 (*call)();
 275 }
 276
 277 char *uml_strdup(const char *string)
 278 {
 279         return kstrdup(string, GFP_KERNEL);
 280 }
 281 EXPORT_SYMBOL(uml_strdup);
 282
 283 int copy_to_user_proc(void __user *to, void *from, int size)
 284 {
 285         return copy_to_user(to, from, size);
 286 }
 287
 288 int copy_from_user_proc(void *to, void __user *from, int size)
 289 {
 290         return copy_from_user(to, from, size);
 291 }
 292
 293 int clear_user_proc(void __user *buf, int size)
 294 {
 295         return clear_user(buf, size);
 296 }
 297
 298 int cpu(void)
 299 {
 300         return current_thread_info()->cpu;
 301 }
 302
 303 static atomic_t using_sysemu = ATOMIC_INIT(0);
 304 int sysemu_supported;
 305
 306 void set_using_sysemu(int value)
 307 {
 308         if (value > sysemu_supported)
 309                 return;
 310         atomic_set(&using_sysemu, value);
 311 }
 312
 313 int get_using_sysemu(void)
 314 {
 315         return atomic_read(&using_sysemu);
 316 }
 317
 318 static int sysemu_proc_show(struct seq_file *m, void *v)
 319 {
 320         seq_printf(m, "%d\n", get_using_sysemu());
 321         return 0;
 322 }
 323
 324 static int sysemu_proc_open(struct inode *inode, struct file *file)
 325 {
 326         return single_open(file, sysemu_proc_show, NULL);
 327 }
 328
 329 static ssize_t sysemu_proc_write(struct file *file, const char __user *buf,
 330                                  size_t count, loff_t *pos)
 331 {
 332         char tmp[2];
 333
 334         if (copy_from_user(tmp, buf, 1))
 335                 return -EFAULT;
 336
 337         if (tmp[0] >= '0' && tmp[0] <= '2')
 338                 set_using_sysemu(tmp[0] - '0');
 339         /* We use the first char, but pretend to write everything */
 340         return count;
 341 }
 342
 343 static const struct file_operations sysemu_proc_fops = {
 344         .owner          = THIS_MODULE,
 345         .open           = sysemu_proc_open,
 346         .read           = seq_read,
 347         .llseek         = seq_lseek,
 348         .release        = single_release,
 349         .write          = sysemu_proc_write,
 350 };
 351
 352 int __init make_proc_sysemu(void)
 353 {
 354         struct proc_dir_entry *ent;
 355         if (!sysemu_supported)
 356                 return 0;
 357
 358         ent = proc_create("sysemu", 0600, NULL, &sysemu_proc_fops);
 359
 360         if (ent == NULL)
 361         {
 362                 printk(KERN_WARNING "Failed to register /proc/sysemu\n");
 363                 return 0;
 364         }
 365
 366         return 0;
 367 }
 368
 369 late_initcall(make_proc_sysemu);
 370
 371 int singlestepping(void * t)
 372 {
 373         struct task_struct *task = t ? t : current;
 374
 375         if (!(task->ptrace & PT_DTRACE))
 376                 return 0;
 377
 378         if (task->thread.singlestep_syscall)
 379                 return 1;
 380
 381         return 2;
 382 }
 383
 384 /*
 385  * Only x86 and x86_64 have an arch_align_stack().
 386  * All other arches have "#define arch_align_stack(x) (x)"
 387  * in their asm/exec.h
 388  * As this is included in UML from asm-um/system-generic.h,
 389  * we can use it to behave as the subarch does.
 390  */
 391 #ifndef arch_align_stack
 392 unsigned long arch_align_stack(unsigned long sp)
 393 {
 394         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 395                 sp -= get_random_int() % 8192;
 396         return sp & ~0xf;
 397 }
 398 #endif
 399
 400 unsigned long get_wchan(struct task_struct *p)
 401 {
 402         unsigned long stack_page, sp, ip;
 403         bool seen_sched = 0;
 404
 405         if ((p == NULL) || (p == current) || (p->state == TASK_RUNNING))
 406                 return 0;
 407
 408         stack_page = (unsigned long) task_stack_page(p);
 409         /* Bail if the process has no kernel stack for some reason */
 410         if (stack_page == 0)
 411                 return 0;
 412
 413         sp = p->thread.switch_buf->JB_SP;
 414         /*
 415          * Bail if the stack pointer is below the bottom of the kernel
 416          * stack for some reason
 417          */
 418         if (sp < stack_page)
 419                 return 0;
 420
 421         while (sp < stack_page + THREAD_SIZE) {
 422                 ip = *((unsigned long *) sp);
 423                 if (in_sched_functions(ip))
 424                         /* Ignore everything until we're above the scheduler */
 425                         seen_sched = 1;
 426                 else if (kernel_text_address(ip) && seen_sched)
 427                         return ip;
 428
 429                 sp += sizeof(unsigned long);
 430         }
 431
 432         return 0;
 433 }
 434
 435 int elf_core_copy_fpregs(struct task_struct *t, elf_fpregset_t *fpu)
 436 {
 437         int cpu = current_thread_info()->cpu;
 438
 439         return save_i387_registers(userspace_pid[cpu], (unsigned long *) fpu);
 440 }
 441