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/module.h>
 12#include <linux/personality.h>
 13#include <linux/proc_fs.h>
 14#include <linux/ptrace.h>
 15#include <linux/random.h>
 16#include <linux/slab.h>
 17#include <linux/sched.h>
 18#include <linux/seq_file.h>
 19#include <linux/tick.h>
 20#include <linux/threads.h>
 21#include <linux/tracehook.h>
 22#include <asm/current.h>
 23#include <asm/pgtable.h>
 24#include <asm/mmu_context.h>
 25#include <asm/uaccess.h>
 26#include "as-layout.h"
 27#include "kern_util.h"
 28#include "os.h"
 29#include "skas.h"
 30
 31/*
 32 * This is a per-cpu array.  A processor only modifies its entry and it only
 33 * cares about its entry, so it's OK if another processor is modifying its
 34 * entry.
 35 */
 36struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } };
 37
 38static inline int external_pid(void)
 39{
 40	/* FIXME: Need to look up userspace_pid by cpu */
 41	return userspace_pid[0];
 42}
 43
 44int pid_to_processor_id(int pid)
 45{
 46	int i;
 47
 48	for (i = 0; i < ncpus; i++) {
 49		if (cpu_tasks[i].pid == pid)
 50			return i;
 51	}
 52	return -1;
 53}
 54
 55void free_stack(unsigned long stack, int order)
 56{
 57	free_pages(stack, order);
 58}
 59
 60unsigned long alloc_stack(int order, int atomic)
 61{
 62	unsigned long page;
 63	gfp_t flags = GFP_KERNEL;
 64
 65	if (atomic)
 66		flags = GFP_ATOMIC;
 67	page = __get_free_pages(flags, order);
 68
 69	return page;
 70}
 71
 72int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 73{
 74	int pid;
 75
 76	current->thread.request.u.thread.proc = fn;
 77	current->thread.request.u.thread.arg = arg;
 78	pid = do_fork(CLONE_VM | CLONE_UNTRACED | flags, 0,
 79		      &current->thread.regs, 0, NULL, NULL);
 80	return pid;
 81}
 82EXPORT_SYMBOL(kernel_thread);
 83
 84static inline void set_current(struct task_struct *task)
 85{
 86	cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task)
 87		{ external_pid(), task });
 88}
 89
 90extern void arch_switch_to(struct task_struct *to);
 91
 92void *__switch_to(struct task_struct *from, struct task_struct *to)
 93{
 94	to->thread.prev_sched = from;
 95	set_current(to);
 96
 97	do {
 98		current->thread.saved_task = NULL;
 99
100		switch_threads(&from->thread.switch_buf,
101			       &to->thread.switch_buf);
102
103		arch_switch_to(current);
104
105		if (current->thread.saved_task)
106			show_regs(&(current->thread.regs));
107		to = current->thread.saved_task;
108		from = current;
109	} while (current->thread.saved_task);
110
111	return current->thread.prev_sched;
112}
113
114void interrupt_end(void)
115{
116	if (need_resched())
117		schedule();
118	if (test_thread_flag(TIF_SIGPENDING))
119		do_signal();
120	if (test_and_clear_thread_flag(TIF_NOTIFY_RESUME))
121		tracehook_notify_resume(&current->thread.regs);
122}
123
124void exit_thread(void)
125{
126}
127
128int get_current_pid(void)
129{
130	return task_pid_nr(current);
131}
132
133/*
134 * This is called magically, by its address being stuffed in a jmp_buf
135 * and being longjmp-d to.
136 */
137void 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 */
163void fork_handler(void)
164{
165	force_flush_all();
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);
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
184int copy_thread(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		UPT_SET_SYSCALL_RETURN(&p->thread.regs.regs, 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		get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp);
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
225void 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
234void 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_idle_enter();
249		rcu_idle_enter();
250		nsecs = disable_timer();
251		idle_sleep(nsecs);
252		rcu_idle_exit();
253		tick_nohz_idle_exit();
254	}
255}
256
257void cpu_idle(void)
258{
259	cpu_tasks[current_thread_info()->cpu].pid = os_getpid();
260	default_idle();
261}
262
263int __cant_sleep(void) {
264	return in_atomic() || irqs_disabled() || in_interrupt();
265	/* Is in_interrupt() really needed? */
266}
267
268int user_context(unsigned long sp)
269{
270	unsigned long stack;
271
272	stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER);
273	return stack != (unsigned long) current_thread_info();
274}
275
276extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end;
277
278void do_uml_exitcalls(void)
279{
280	exitcall_t *call;
281
282	call = &__uml_exitcall_end;
283	while (--call >= &__uml_exitcall_begin)
284		(*call)();
285}
286
287char *uml_strdup(const char *string)
288{
289	return kstrdup(string, GFP_KERNEL);
290}
291EXPORT_SYMBOL(uml_strdup);
292
293int copy_to_user_proc(void __user *to, void *from, int size)
294{
295	return copy_to_user(to, from, size);
296}
297
298int copy_from_user_proc(void *to, void __user *from, int size)
299{
300	return copy_from_user(to, from, size);
301}
302
303int clear_user_proc(void __user *buf, int size)
304{
305	return clear_user(buf, size);
306}
307
308int strlen_user_proc(char __user *str)
309{
310	return strlen_user(str);
311}
312
313int smp_sigio_handler(void)
314{
315#ifdef CONFIG_SMP
316	int cpu = current_thread_info()->cpu;
317	IPI_handler(cpu);
318	if (cpu != 0)
319		return 1;
320#endif
321	return 0;
322}
323
324int cpu(void)
325{
326	return current_thread_info()->cpu;
327}
328
329static atomic_t using_sysemu = ATOMIC_INIT(0);
330int sysemu_supported;
331
332void set_using_sysemu(int value)
333{
334	if (value > sysemu_supported)
335		return;
336	atomic_set(&using_sysemu, value);
337}
338
339int get_using_sysemu(void)
340{
341	return atomic_read(&using_sysemu);
342}
343
344static int sysemu_proc_show(struct seq_file *m, void *v)
345{
346	seq_printf(m, "%d\n", get_using_sysemu());
347	return 0;
348}
349
350static int sysemu_proc_open(struct inode *inode, struct file *file)
351{
352	return single_open(file, sysemu_proc_show, NULL);
353}
354
355static ssize_t sysemu_proc_write(struct file *file, const char __user *buf,
356				 size_t count, loff_t *pos)
357{
358	char tmp[2];
359
360	if (copy_from_user(tmp, buf, 1))
361		return -EFAULT;
362
363	if (tmp[0] >= '0' && tmp[0] <= '2')
364		set_using_sysemu(tmp[0] - '0');
365	/* We use the first char, but pretend to write everything */
366	return count;
367}
368
369static const struct file_operations sysemu_proc_fops = {
370	.owner		= THIS_MODULE,
371	.open		= sysemu_proc_open,
372	.read		= seq_read,
373	.llseek		= seq_lseek,
374	.release	= single_release,
375	.write		= sysemu_proc_write,
376};
377
378int __init make_proc_sysemu(void)
379{
380	struct proc_dir_entry *ent;
381	if (!sysemu_supported)
382		return 0;
383
384	ent = proc_create("sysemu", 0600, NULL, &sysemu_proc_fops);
385
386	if (ent == NULL)
387	{
388		printk(KERN_WARNING "Failed to register /proc/sysemu\n");
389		return 0;
390	}
391
392	return 0;
393}
394
395late_initcall(make_proc_sysemu);
396
397int singlestepping(void * t)
398{
399	struct task_struct *task = t ? t : current;
400
401	if (!(task->ptrace & PT_DTRACE))
402		return 0;
403
404	if (task->thread.singlestep_syscall)
405		return 1;
406
407	return 2;
408}
409
410/*
411 * Only x86 and x86_64 have an arch_align_stack().
412 * All other arches have "#define arch_align_stack(x) (x)"
413 * in their asm/system.h
414 * As this is included in UML from asm-um/system-generic.h,
415 * we can use it to behave as the subarch does.
416 */
417#ifndef arch_align_stack
418unsigned long arch_align_stack(unsigned long sp)
419{
420	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
421		sp -= get_random_int() % 8192;
422	return sp & ~0xf;
423}
424#endif
425
426unsigned long get_wchan(struct task_struct *p)
427{
428	unsigned long stack_page, sp, ip;
429	bool seen_sched = 0;
430
431	if ((p == NULL) || (p == current) || (p->state == TASK_RUNNING))
432		return 0;
433
434	stack_page = (unsigned long) task_stack_page(p);
435	/* Bail if the process has no kernel stack for some reason */
436	if (stack_page == 0)
437		return 0;
438
439	sp = p->thread.switch_buf->JB_SP;
440	/*
441	 * Bail if the stack pointer is below the bottom of the kernel
442	 * stack for some reason
443	 */
444	if (sp < stack_page)
445		return 0;
446
447	while (sp < stack_page + THREAD_SIZE) {
448		ip = *((unsigned long *) sp);
449		if (in_sched_functions(ip))
450			/* Ignore everything until we're above the scheduler */
451			seen_sched = 1;
452		else if (kernel_text_address(ip) && seen_sched)
453			return ip;
454
455		sp += sizeof(unsigned long);
456	}
457
458	return 0;
459}
460
461int elf_core_copy_fpregs(struct task_struct *t, elf_fpregset_t *fpu)
462{
463	int cpu = current_thread_info()->cpu;
464
465	return save_fp_registers(userspace_pid[cpu], (unsigned long *) fpu);
466}
467