uml: tickless support
[linux-2.6.git] / 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         if (page == 0)
64                 return 0;
65
66         return page;
67 }
68
69 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
70 {
71         int pid;
72
73         current->thread.request.u.thread.proc = fn;
74         current->thread.request.u.thread.arg = arg;
75         pid = do_fork(CLONE_VM | CLONE_UNTRACED | flags, 0,
76                       &current->thread.regs, 0, NULL, NULL);
77         return pid;
78 }
79
80 static inline void set_current(struct task_struct *task)
81 {
82         cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task)
83                 { external_pid(task), task });
84 }
85
86 extern void arch_switch_to(struct task_struct *from, struct task_struct *to);
87
88 void *_switch_to(void *prev, void *next, void *last)
89 {
90         struct task_struct *from = prev;
91         struct task_struct *to= next;
92
93         to->thread.prev_sched = from;
94         set_current(to);
95
96         do {
97                 current->thread.saved_task = NULL;
98
99                 switch_threads(&from->thread.switch_buf,
100                                &to->thread.switch_buf);
101
102                 arch_switch_to(current->thread.prev_sched, current);
103
104                 if (current->thread.saved_task)
105                         show_regs(&(current->thread.regs));
106                 next= current->thread.saved_task;
107                 prev= current;
108         } while(current->thread.saved_task);
109
110         return current->thread.prev_sched;
111
112 }
113
114 void interrupt_end(void)
115 {
116         if (need_resched())
117                 schedule();
118         if (test_tsk_thread_flag(current, TIF_SIGPENDING))
119                 do_signal();
120 }
121
122 void exit_thread(void)
123 {
124 }
125
126 void *get_current(void)
127 {
128         return current;
129 }
130
131 extern void schedule_tail(struct task_struct *prev);
132
133 /*
134  * This is called magically, by its address being stuffed in a jmp_buf
135  * and being longjmp-d to.
136  */
137 void new_thread_handler(void)
138 {
139         int (*fn)(void *), n;
140         void *arg;
141
142         if (current->thread.prev_sched != NULL)
143                 schedule_tail(current->thread.prev_sched);
144         current->thread.prev_sched = NULL;
145
146         fn = current->thread.request.u.thread.proc;
147         arg = current->thread.request.u.thread.arg;
148
149         /*
150          * The return value is 1 if the kernel thread execs a process,
151          * 0 if it just exits
152          */
153         n = run_kernel_thread(fn, arg, &current->thread.exec_buf);
154         if (n == 1) {
155                 /* Handle any immediate reschedules or signals */
156                 interrupt_end();
157                 userspace(&current->thread.regs.regs);
158         }
159         else do_exit(0);
160 }
161
162 /* Called magically, see new_thread_handler above */
163 void fork_handler(void)
164 {
165         force_flush_all();
166         if (current->thread.prev_sched == NULL)
167                 panic("blech");
168
169         schedule_tail(current->thread.prev_sched);
170
171         /*
172          * XXX: if interrupt_end() calls schedule, this call to
173          * arch_switch_to isn't needed. We could want to apply this to
174          * improve performance. -bb
175          */
176         arch_switch_to(current->thread.prev_sched, current);
177
178         current->thread.prev_sched = NULL;
179
180         /* Handle any immediate reschedules or signals */
181         interrupt_end();
182
183         userspace(&current->thread.regs.regs);
184 }
185
186 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
187                 unsigned long stack_top, struct task_struct * p,
188                 struct pt_regs *regs)
189 {
190         void (*handler)(void);
191         int ret = 0;
192
193         p->thread = (struct thread_struct) INIT_THREAD;
194
195         if (current->thread.forking) {
196                 memcpy(&p->thread.regs.regs, &regs->regs,
197                        sizeof(p->thread.regs.regs));
198                 REGS_SET_SYSCALL_RETURN(p->thread.regs.regs.gp, 0);
199                 if (sp != 0)
200                         REGS_SP(p->thread.regs.regs.gp) = sp;
201
202                 handler = fork_handler;
203
204                 arch_copy_thread(&current->thread.arch, &p->thread.arch);
205         }
206         else {
207                 init_thread_registers(&p->thread.regs.regs);
208                 p->thread.request.u.thread = current->thread.request.u.thread;
209                 handler = new_thread_handler;
210         }
211
212         new_thread(task_stack_page(p), &p->thread.switch_buf, handler);
213
214         if (current->thread.forking) {
215                 clear_flushed_tls(p);
216
217                 /*
218                  * Set a new TLS for the child thread?
219                  */
220                 if (clone_flags & CLONE_SETTLS)
221                         ret = arch_copy_tls(p);
222         }
223
224         return ret;
225 }
226
227 void initial_thread_cb(void (*proc)(void *), void *arg)
228 {
229         int save_kmalloc_ok = kmalloc_ok;
230
231         kmalloc_ok = 0;
232         initial_thread_cb_skas(proc, arg);
233         kmalloc_ok = save_kmalloc_ok;
234 }
235
236 void default_idle(void)
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                 switch_timers(1);
250                 idle_sleep(10);
251                 switch_timers(0);
252                 tick_nohz_restart_sched_tick();
253         }
254 }
255
256 void cpu_idle(void)
257 {
258         cpu_tasks[current_thread->cpu].pid = os_getpid();
259         default_idle();
260 }
261
262 void *um_virt_to_phys(struct task_struct *task, unsigned long addr,
263                       pte_t *pte_out)
264 {
265         pgd_t *pgd;
266         pud_t *pud;
267         pmd_t *pmd;
268         pte_t *pte;
269         pte_t ptent;
270
271         if (task->mm == NULL)
272                 return ERR_PTR(-EINVAL);
273         pgd = pgd_offset(task->mm, addr);
274         if (!pgd_present(*pgd))
275                 return ERR_PTR(-EINVAL);
276
277         pud = pud_offset(pgd, addr);
278         if (!pud_present(*pud))
279                 return ERR_PTR(-EINVAL);
280
281         pmd = pmd_offset(pud, addr);
282         if (!pmd_present(*pmd))
283                 return ERR_PTR(-EINVAL);
284
285         pte = pte_offset_kernel(pmd, addr);
286         ptent = *pte;
287         if (!pte_present(ptent))
288                 return ERR_PTR(-EINVAL);
289
290         if (pte_out != NULL)
291                 *pte_out = ptent;
292         return (void *) (pte_val(ptent) & PAGE_MASK) + (addr & ~PAGE_MASK);
293 }
294
295 char *current_cmd(void)
296 {
297 #if defined(CONFIG_SMP) || defined(CONFIG_HIGHMEM)
298         return "(Unknown)";
299 #else
300         void *addr = um_virt_to_phys(current, current->mm->arg_start, NULL);
301         return IS_ERR(addr) ? "(Unknown)": __va((unsigned long) addr);
302 #endif
303 }
304
305 void dump_thread(struct pt_regs *regs, struct user *u)
306 {
307 }
308
309 int __cant_sleep(void) {
310         return in_atomic() || irqs_disabled() || in_interrupt();
311         /* Is in_interrupt() really needed? */
312 }
313
314 int user_context(unsigned long sp)
315 {
316         unsigned long stack;
317
318         stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER);
319         return stack != (unsigned long) current_thread;
320 }
321
322 extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end;
323
324 void do_uml_exitcalls(void)
325 {
326         exitcall_t *call;
327
328         call = &__uml_exitcall_end;
329         while (--call >= &__uml_exitcall_begin)
330                 (*call)();
331 }
332
333 char *uml_strdup(char *string)
334 {
335         return kstrdup(string, GFP_KERNEL);
336 }
337
338 int copy_to_user_proc(void __user *to, void *from, int size)
339 {
340         return copy_to_user(to, from, size);
341 }
342
343 int copy_from_user_proc(void *to, void __user *from, int size)
344 {
345         return copy_from_user(to, from, size);
346 }
347
348 int clear_user_proc(void __user *buf, int size)
349 {
350         return clear_user(buf, size);
351 }
352
353 int strlen_user_proc(char __user *str)
354 {
355         return strlen_user(str);
356 }
357
358 int smp_sigio_handler(void)
359 {
360 #ifdef CONFIG_SMP
361         int cpu = current_thread->cpu;
362         IPI_handler(cpu);
363         if (cpu != 0)
364                 return 1;
365 #endif
366         return 0;
367 }
368
369 int cpu(void)
370 {
371         return current_thread->cpu;
372 }
373
374 static atomic_t using_sysemu = ATOMIC_INIT(0);
375 int sysemu_supported;
376
377 void set_using_sysemu(int value)
378 {
379         if (value > sysemu_supported)
380                 return;
381         atomic_set(&using_sysemu, value);
382 }
383
384 int get_using_sysemu(void)
385 {
386         return atomic_read(&using_sysemu);
387 }
388
389 static int proc_read_sysemu(char *buf, char **start, off_t offset, int size,int *eof, void *data)
390 {
391         if (snprintf(buf, size, "%d\n", get_using_sysemu()) < size)
392                 /* No overflow */
393                 *eof = 1;
394
395         return strlen(buf);
396 }
397
398 static int proc_write_sysemu(struct file *file,const char __user *buf, unsigned long count,void *data)
399 {
400         char tmp[2];
401
402         if (copy_from_user(tmp, buf, 1))
403                 return -EFAULT;
404
405         if (tmp[0] >= '0' && tmp[0] <= '2')
406                 set_using_sysemu(tmp[0] - '0');
407         /* We use the first char, but pretend to write everything */
408         return count;
409 }
410
411 int __init make_proc_sysemu(void)
412 {
413         struct proc_dir_entry *ent;
414         if (!sysemu_supported)
415                 return 0;
416
417         ent = create_proc_entry("sysemu", 0600, &proc_root);
418
419         if (ent == NULL)
420         {
421                 printk(KERN_WARNING "Failed to register /proc/sysemu\n");
422                 return 0;
423         }
424
425         ent->read_proc  = proc_read_sysemu;
426         ent->write_proc = proc_write_sysemu;
427
428         return 0;
429 }
430
431 late_initcall(make_proc_sysemu);
432
433 int singlestepping(void * t)
434 {
435         struct task_struct *task = t ? t : current;
436
437         if ( ! (task->ptrace & PT_DTRACE) )
438                 return 0;
439
440         if (task->thread.singlestep_syscall)
441                 return 1;
442
443         return 2;
444 }
445
446 /*
447  * Only x86 and x86_64 have an arch_align_stack().
448  * All other arches have "#define arch_align_stack(x) (x)"
449  * in their asm/system.h
450  * As this is included in UML from asm-um/system-generic.h,
451  * we can use it to behave as the subarch does.
452  */
453 #ifndef arch_align_stack
454 unsigned long arch_align_stack(unsigned long sp)
455 {
456         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
457                 sp -= get_random_int() % 8192;
458         return sp & ~0xf;
459 }
460 #endif