arch/um/kernel/process.c

   1 /*
   2  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   3  * Copyright 2003 PathScale, Inc.
   4  * Licensed under the GPL
   5  */
   6
   7 #include "linux/stddef.h"
   8 #include "linux/err.h"
   9 #include "linux/hardirq.h"
  10 #include "linux/mm.h"
  11 #include "linux/personality.h"
  12 #include "linux/proc_fs.h"
  13 #include "linux/ptrace.h"
  14 #include "linux/random.h"
  15 #include "linux/sched.h"
  16 #include "linux/tick.h"
  17 #include "linux/threads.h"
  18 #include "asm/pgtable.h"
  19 #include "asm/uaccess.h"
  20 #include "as-layout.h"
  21 #include "kern_util.h"
  22 #include "os.h"
  23 #include "skas.h"
  24 #include "tlb.h"
  25
  26 /*
  27  * This is a per-cpu array.  A processor only modifies its entry and it only
  28  * cares about its entry, so it's OK if another processor is modifying its
  29  * entry.
  30  */
  31 struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } };
  32
  33 static inline int external_pid(struct task_struct *task)
  34 {
  35         /* FIXME: Need to look up userspace_pid by cpu */
  36         return userspace_pid[0];
  37 }
  38
  39 int pid_to_processor_id(int pid)
  40 {
  41         int i;
  42
  43         for(i = 0; i < ncpus; i++) {
  44                 if (cpu_tasks[i].pid == pid)
  45                         return i;
  46         }
  47         return -1;
  48 }
  49
  50 void free_stack(unsigned long stack, int order)
  51 {
  52         free_pages(stack, order);
  53 }
  54
  55 unsigned long alloc_stack(int order, int atomic)
  56 {
  57         unsigned long page;
  58         gfp_t flags = GFP_KERNEL;
  59
  60         if (atomic)
  61                 flags = GFP_ATOMIC;
  62         page = __get_free_pages(flags, order);
  63
  64         return page;
  65 }
  66
  67 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
  68 {
  69         int pid;
  70
  71         current->thread.request.u.thread.proc = fn;
  72         current->thread.request.u.thread.arg = arg;
  73         pid = do_fork(CLONE_VM | CLONE_UNTRACED | flags, 0,
  74                       &current->thread.regs, 0, NULL, NULL);
  75         return pid;
  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), task });
  82 }
  83
  84 extern void arch_switch_to(struct task_struct *from, struct task_struct *to);
  85
  86 void *_switch_to(void *prev, void *next, void *last)
  87 {
  88         struct task_struct *from = prev;
  89         struct task_struct *to= next;
  90
  91         to->thread.prev_sched = from;
  92         set_current(to);
  93
  94         do {
  95                 current->thread.saved_task = NULL;
  96
  97                 switch_threads(&from->thread.switch_buf,
  98                                &to->thread.switch_buf);
  99
 100                 arch_switch_to(current->thread.prev_sched, current);
 101
 102                 if (current->thread.saved_task)
 103                         show_regs(&(current->thread.regs));
 104                 next= current->thread.saved_task;
 105                 prev= current;
 106         } while(current->thread.saved_task);
 107
 108         return current->thread.prev_sched;
 109
 110 }
 111
 112 void interrupt_end(void)
 113 {
 114         if (need_resched())
 115                 schedule();
 116         if (test_tsk_thread_flag(current, TIF_SIGPENDING))
 117                 do_signal();
 118 }
 119
 120 void exit_thread(void)
 121 {
 122 }
 123
 124 void *get_current(void)
 125 {
 126         return current;
 127 }
 128
 129 extern void schedule_tail(struct task_struct *prev);
 130
 131 /*
 132  * This is called magically, by its address being stuffed in a jmp_buf
 133  * and being longjmp-d to.
 134  */
 135 void new_thread_handler(void)
 136 {
 137         int (*fn)(void *), n;
 138         void *arg;
 139
 140         if (current->thread.prev_sched != NULL)
 141                 schedule_tail(current->thread.prev_sched);
 142         current->thread.prev_sched = NULL;
 143
 144         fn = current->thread.request.u.thread.proc;
 145         arg = current->thread.request.u.thread.arg;
 146
 147         /*
 148          * The return value is 1 if the kernel thread execs a process,
 149          * 0 if it just exits
 150          */
 151         n = run_kernel_thread(fn, arg, &current->thread.exec_buf);
 152         if (n == 1) {
 153                 /* Handle any immediate reschedules or signals */
 154                 interrupt_end();
 155                 userspace(&current->thread.regs.regs);
 156         }
 157         else do_exit(0);
 158 }
 159
 160 /* Called magically, see new_thread_handler above */
 161 void fork_handler(void)
 162 {
 163         force_flush_all();
 164         if (current->thread.prev_sched == NULL)
 165                 panic("blech");
 166
 167         schedule_tail(current->thread.prev_sched);
 168
 169         /*
 170          * XXX: if interrupt_end() calls schedule, this call to
 171          * arch_switch_to isn't needed. We could want to apply this to
 172          * improve performance. -bb
 173          */
 174         arch_switch_to(current->thread.prev_sched, current);
 175
 176         current->thread.prev_sched = NULL;
 177
 178         /* Handle any immediate reschedules or signals */
 179         interrupt_end();
 180
 181         userspace(&current->thread.regs.regs);
 182 }
 183
 184 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
 185                 unsigned long stack_top, struct task_struct * p,
 186                 struct pt_regs *regs)
 187 {
 188         void (*handler)(void);
 189         int ret = 0;
 190
 191         p->thread = (struct thread_struct) INIT_THREAD;
 192
 193         if (current->thread.forking) {
 194                 memcpy(&p->thread.regs.regs, &regs->regs,
 195                        sizeof(p->thread.regs.regs));
 196                 REGS_SET_SYSCALL_RETURN(p->thread.regs.regs.gp, 0);
 197                 if (sp != 0)
 198                         REGS_SP(p->thread.regs.regs.gp) = sp;
 199
 200                 handler = fork_handler;
 201
 202                 arch_copy_thread(&current->thread.arch, &p->thread.arch);
 203         }
 204         else {
 205                 init_thread_registers(&p->thread.regs.regs);
 206                 p->thread.request.u.thread = current->thread.request.u.thread;
 207                 handler = new_thread_handler;
 208         }
 209
 210         new_thread(task_stack_page(p), &p->thread.switch_buf, handler);
 211
 212         if (current->thread.forking) {
 213                 clear_flushed_tls(p);
 214
 215                 /*
 216                  * Set a new TLS for the child thread?
 217                  */
 218                 if (clone_flags & CLONE_SETTLS)
 219                         ret = arch_copy_tls(p);
 220         }
 221
 222         return ret;
 223 }
 224
 225 void initial_thread_cb(void (*proc)(void *), void *arg)
 226 {
 227         int save_kmalloc_ok = kmalloc_ok;
 228
 229         kmalloc_ok = 0;
 230         initial_thread_cb_skas(proc, arg);
 231         kmalloc_ok = save_kmalloc_ok;
 232 }
 233
 234 void default_idle(void)
 235 {
 236         unsigned long long nsecs;
 237
 238         while(1) {
 239                 /* endless idle loop with no priority at all */
 240
 241                 /*
 242                  * although we are an idle CPU, we do not want to
 243                  * get into the scheduler unnecessarily.
 244                  */
 245                 if (need_resched())
 246                         schedule();
 247
 248                 tick_nohz_stop_sched_tick();
 249                 nsecs = disable_timer();
 250                 idle_sleep(nsecs);
 251                 tick_nohz_restart_sched_tick();
 252         }
 253 }
 254
 255 void cpu_idle(void)
 256 {
 257         cpu_tasks[current_thread->cpu].pid = os_getpid();
 258         default_idle();
 259 }
 260
 261 void *um_virt_to_phys(struct task_struct *task, unsigned long addr,
 262                       pte_t *pte_out)
 263 {
 264         pgd_t *pgd;
 265         pud_t *pud;
 266         pmd_t *pmd;
 267         pte_t *pte;
 268         pte_t ptent;
 269
 270         if (task->mm == NULL)
 271                 return ERR_PTR(-EINVAL);
 272         pgd = pgd_offset(task->mm, addr);
 273         if (!pgd_present(*pgd))
 274                 return ERR_PTR(-EINVAL);
 275
 276         pud = pud_offset(pgd, addr);
 277         if (!pud_present(*pud))
 278                 return ERR_PTR(-EINVAL);
 279
 280         pmd = pmd_offset(pud, addr);
 281         if (!pmd_present(*pmd))
 282                 return ERR_PTR(-EINVAL);
 283
 284         pte = pte_offset_kernel(pmd, addr);
 285         ptent = *pte;
 286         if (!pte_present(ptent))
 287                 return ERR_PTR(-EINVAL);
 288
 289         if (pte_out != NULL)
 290                 *pte_out = ptent;
 291         return (void *) (pte_val(ptent) & PAGE_MASK) + (addr & ~PAGE_MASK);
 292 }
 293
 294 char *current_cmd(void)
 295 {
 296 #if defined(CONFIG_SMP) || defined(CONFIG_HIGHMEM)
 297         return "(Unknown)";
 298 #else
 299         void *addr = um_virt_to_phys(current, current->mm->arg_start, NULL);
 300         return IS_ERR(addr) ? "(Unknown)": __va((unsigned long) addr);
 301 #endif
 302 }
 303
 304 void dump_thread(struct pt_regs *regs, struct user *u)
 305 {
 306 }
 307
 308 int __cant_sleep(void) {
 309         return in_atomic() || irqs_disabled() || in_interrupt();
 310         /* Is in_interrupt() really needed? */
 311 }
 312
 313 int user_context(unsigned long sp)
 314 {
 315         unsigned long stack;
 316
 317         stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER);
 318         return stack != (unsigned long) current_thread;
 319 }
 320
 321 extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end;
 322
 323 void do_uml_exitcalls(void)
 324 {
 325         exitcall_t *call;
 326
 327         call = &__uml_exitcall_end;
 328         while (--call >= &__uml_exitcall_begin)
 329                 (*call)();
 330 }
 331
 332 char *uml_strdup(const char *string)
 333 {
 334         return kstrdup(string, GFP_KERNEL);
 335 }
 336
 337 int copy_to_user_proc(void __user *to, void *from, int size)
 338 {
 339         return copy_to_user(to, from, size);
 340 }
 341
 342 int copy_from_user_proc(void *to, void __user *from, int size)
 343 {
 344         return copy_from_user(to, from, size);
 345 }
 346
 347 int clear_user_proc(void __user *buf, int size)
 348 {
 349         return clear_user(buf, size);
 350 }
 351
 352 int strlen_user_proc(char __user *str)
 353 {
 354         return strlen_user(str);
 355 }
 356
 357 int smp_sigio_handler(void)
 358 {
 359 #ifdef CONFIG_SMP
 360         int cpu = current_thread->cpu;
 361         IPI_handler(cpu);
 362         if (cpu != 0)
 363                 return 1;
 364 #endif
 365         return 0;
 366 }
 367
 368 int cpu(void)
 369 {
 370         return current_thread->cpu;
 371 }
 372
 373 static atomic_t using_sysemu = ATOMIC_INIT(0);
 374 int sysemu_supported;
 375
 376 void set_using_sysemu(int value)
 377 {
 378         if (value > sysemu_supported)
 379                 return;
 380         atomic_set(&using_sysemu, value);
 381 }
 382
 383 int get_using_sysemu(void)
 384 {
 385         return atomic_read(&using_sysemu);
 386 }
 387
 388 static int proc_read_sysemu(char *buf, char **start, off_t offset, int size,int *eof, void *data)
 389 {
 390         if (snprintf(buf, size, "%d\n", get_using_sysemu()) < size)
 391                 /* No overflow */
 392                 *eof = 1;
 393
 394         return strlen(buf);
 395 }
 396
 397 static int proc_write_sysemu(struct file *file,const char __user *buf, unsigned long count,void *data)
 398 {
 399         char tmp[2];
 400
 401         if (copy_from_user(tmp, buf, 1))
 402                 return -EFAULT;
 403
 404         if (tmp[0] >= '0' && tmp[0] <= '2')
 405                 set_using_sysemu(tmp[0] - '0');
 406         /* We use the first char, but pretend to write everything */
 407         return count;
 408 }
 409
 410 int __init make_proc_sysemu(void)
 411 {
 412         struct proc_dir_entry *ent;
 413         if (!sysemu_supported)
 414                 return 0;
 415
 416         ent = create_proc_entry("sysemu", 0600, &proc_root);
 417
 418         if (ent == NULL)
 419         {
 420                 printk(KERN_WARNING "Failed to register /proc/sysemu\n");
 421                 return 0;
 422         }
 423
 424         ent->read_proc  = proc_read_sysemu;
 425         ent->write_proc = proc_write_sysemu;
 426
 427         return 0;
 428 }
 429
 430 late_initcall(make_proc_sysemu);
 431
 432 int singlestepping(void * t)
 433 {
 434         struct task_struct *task = t ? t : current;
 435
 436         if ( ! (task->ptrace & PT_DTRACE) )
 437                 return 0;
 438
 439         if (task->thread.singlestep_syscall)
 440                 return 1;
 441
 442         return 2;
 443 }
 444
 445 /*
 446  * Only x86 and x86_64 have an arch_align_stack().
 447  * All other arches have "#define arch_align_stack(x) (x)"
 448  * in their asm/system.h
 449  * As this is included in UML from asm-um/system-generic.h,
 450  * we can use it to behave as the subarch does.
 451  */
 452 #ifndef arch_align_stack
 453 unsigned long arch_align_stack(unsigned long sp)
 454 {
 455         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 456                 sp -= get_random_int() % 8192;
 457         return sp & ~0xf;
 458 }
 459 #endif
 460
 461 unsigned long get_wchan(struct task_struct *p)
 462 {
 463         unsigned long stack_page, sp, ip;
 464         bool seen_sched = 0;
 465
 466         if ((p == NULL) || (p == current) || (p->state == TASK_RUNNING))
 467                 return 0;
 468
 469         stack_page = (unsigned long) task_stack_page(p);
 470         /* Bail if the process has no kernel stack for some reason */
 471         if (stack_page == 0)
 472                 return 0;
 473
 474         sp = p->thread.switch_buf->JB_SP;
 475         /*
 476          * Bail if the stack pointer is below the bottom of the kernel
 477          * stack for some reason
 478          */
 479         if (sp < stack_page)
 480                 return 0;
 481
 482         while (sp < stack_page + THREAD_SIZE) {
 483                 ip = *((unsigned long *) sp);
 484                 if (in_sched_functions(ip))
 485                         /* Ignore everything until we're above the scheduler */
 486                         seen_sched = 1;
 487                 else if (kernel_text_address(ip) && seen_sched)
 488                         return ip;
 489
 490                 sp += sizeof(unsigned long);
 491         }
 492
 493         return 0;
 494 }