process.c 9.4 KB
Newer Older
J
Jeff Dike 已提交
1
/*
J
Jeff Dike 已提交
2
 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
L
Linus Torvalds 已提交
3 4 5 6
 * Copyright 2003 PathScale, Inc.
 * Licensed under the GPL
 */

J
Jeff Dike 已提交
7 8 9 10
#include <linux/stddef.h>
#include <linux/err.h>
#include <linux/hardirq.h>
#include <linux/mm.h>
11
#include <linux/module.h>
J
Jeff Dike 已提交
12 13 14 15
#include <linux/personality.h>
#include <linux/proc_fs.h>
#include <linux/ptrace.h>
#include <linux/random.h>
16
#include <linux/slab.h>
J
Jeff Dike 已提交
17
#include <linux/sched.h>
18
#include <linux/seq_file.h>
J
Jeff Dike 已提交
19 20
#include <linux/tick.h>
#include <linux/threads.h>
A
Al Viro 已提交
21
#include <linux/tracehook.h>
J
Jeff Dike 已提交
22 23
#include <asm/current.h>
#include <asm/pgtable.h>
A
Al Viro 已提交
24
#include <asm/mmu_context.h>
J
Jeff Dike 已提交
25
#include <asm/uaccess.h>
J
Jeff Dike 已提交
26
#include "as-layout.h"
J
Jeff Dike 已提交
27
#include "kern_util.h"
L
Linus Torvalds 已提交
28
#include "os.h"
29
#include "skas.h"
L
Linus Torvalds 已提交
30

J
Jeff Dike 已提交
31 32
/*
 * This is a per-cpu array.  A processor only modifies its entry and it only
L
Linus Torvalds 已提交
33 34 35 36 37
 * cares about its entry, so it's OK if another processor is modifying its
 * entry.
 */
struct cpu_task cpu_tasks[NR_CPUS] = { [0 ... NR_CPUS - 1] = { -1, NULL } };

38
static inline int external_pid(void)
L
Linus Torvalds 已提交
39
{
40
	/* FIXME: Need to look up userspace_pid by cpu */
J
Jeff Dike 已提交
41
	return userspace_pid[0];
L
Linus Torvalds 已提交
42 43 44 45 46 47
}

int pid_to_processor_id(int pid)
{
	int i;

J
Jeff Dike 已提交
48
	for (i = 0; i < ncpus; i++) {
J
Jeff Dike 已提交
49
		if (cpu_tasks[i].pid == pid)
J
Jeff Dike 已提交
50
			return i;
L
Linus Torvalds 已提交
51
	}
J
Jeff Dike 已提交
52
	return -1;
L
Linus Torvalds 已提交
53 54 55 56 57 58 59 60 61 62
}

void free_stack(unsigned long stack, int order)
{
	free_pages(stack, order);
}

unsigned long alloc_stack(int order, int atomic)
{
	unsigned long page;
A
Al Viro 已提交
63
	gfp_t flags = GFP_KERNEL;
L
Linus Torvalds 已提交
64

65 66
	if (atomic)
		flags = GFP_ATOMIC;
L
Linus Torvalds 已提交
67
	page = __get_free_pages(flags, order);
68

J
Jeff Dike 已提交
69
	return page;
L
Linus Torvalds 已提交
70 71 72 73 74 75 76 77
}

int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
{
	int pid;

	current->thread.request.u.thread.proc = fn;
	current->thread.request.u.thread.arg = arg;
J
Jeff Dike 已提交
78 79
	pid = do_fork(CLONE_VM | CLONE_UNTRACED | flags, 0,
		      &current->thread.regs, 0, NULL, NULL);
J
Jeff Dike 已提交
80
	return pid;
L
Linus Torvalds 已提交
81
}
82
EXPORT_SYMBOL(kernel_thread);
L
Linus Torvalds 已提交
83

J
Jeff Dike 已提交
84
static inline void set_current(struct task_struct *task)
L
Linus Torvalds 已提交
85
{
A
Al Viro 已提交
86
	cpu_tasks[task_thread_info(task)->cpu] = ((struct cpu_task)
87
		{ external_pid(), task });
L
Linus Torvalds 已提交
88 89
}

90
extern void arch_switch_to(struct task_struct *to);
91

92
void *__switch_to(struct task_struct *from, struct task_struct *to)
L
Linus Torvalds 已提交
93
{
J
Jeff Dike 已提交
94 95
	to->thread.prev_sched = from;
	set_current(to);
96

97
	do {
J
Jeff Dike 已提交
98
		current->thread.saved_task = NULL;
99

J
Jeff Dike 已提交
100 101
		switch_threads(&from->thread.switch_buf,
			       &to->thread.switch_buf);
102

103
		arch_switch_to(current);
104

J
Jeff Dike 已提交
105
		if (current->thread.saved_task)
106
			show_regs(&(current->thread.regs));
J
Jeff Dike 已提交
107 108
		to = current->thread.saved_task;
		from = current;
109
	} while (current->thread.saved_task);
110

J
Jeff Dike 已提交
111
	return current->thread.prev_sched;
L
Linus Torvalds 已提交
112 113 114 115
}

void interrupt_end(void)
{
J
Jeff Dike 已提交
116
	if (need_resched())
J
Jeff Dike 已提交
117
		schedule();
A
Al Viro 已提交
118
	if (test_thread_flag(TIF_SIGPENDING))
J
Jeff Dike 已提交
119
		do_signal();
120
	if (test_and_clear_thread_flag(TIF_NOTIFY_RESUME))
A
Al Viro 已提交
121
		tracehook_notify_resume(&current->thread.regs);
L
Linus Torvalds 已提交
122 123 124 125 126
}

void exit_thread(void)
{
}
J
Jeff Dike 已提交
127

A
Al Viro 已提交
128
int get_current_pid(void)
L
Linus Torvalds 已提交
129
{
A
Al Viro 已提交
130
	return task_pid_nr(current);
L
Linus Torvalds 已提交
131 132
}

J
Jeff Dike 已提交
133 134
/*
 * This is called magically, by its address being stuffed in a jmp_buf
135 136 137 138 139 140 141
 * and being longjmp-d to.
 */
void new_thread_handler(void)
{
	int (*fn)(void *), n;
	void *arg;

J
Jeff Dike 已提交
142
	if (current->thread.prev_sched != NULL)
143 144 145 146 147 148
		schedule_tail(current->thread.prev_sched);
	current->thread.prev_sched = NULL;

	fn = current->thread.request.u.thread.proc;
	arg = current->thread.request.u.thread.arg;

J
Jeff Dike 已提交
149 150
	/*
	 * The return value is 1 if the kernel thread execs a process,
151 152 153
	 * 0 if it just exits
	 */
	n = run_kernel_thread(fn, arg, &current->thread.exec_buf);
154
	if (n == 1)
155
		userspace(&current->thread.regs.regs);
156 157
	else
		do_exit(0);
158 159 160 161 162 163 164 165 166
}

/* Called magically, see new_thread_handler above */
void fork_handler(void)
{
	force_flush_all();

	schedule_tail(current->thread.prev_sched);

J
Jeff Dike 已提交
167 168
	/*
	 * XXX: if interrupt_end() calls schedule, this call to
169
	 * arch_switch_to isn't needed. We could want to apply this to
J
Jeff Dike 已提交
170 171
	 * improve performance. -bb
	 */
172
	arch_switch_to(current);
173 174 175 176 177 178

	current->thread.prev_sched = NULL;

	userspace(&current->thread.regs.regs);
}

A
Alexey Dobriyan 已提交
179
int copy_thread(unsigned long clone_flags, unsigned long sp,
J
Jeff Dike 已提交
180
		unsigned long stack_top, struct task_struct * p,
L
Linus Torvalds 已提交
181 182
		struct pt_regs *regs)
{
183
	void (*handler)(void);
A
Al Viro 已提交
184
	int kthread = current->flags & PF_KTHREAD;
185
	int ret = 0;
186

L
Linus Torvalds 已提交
187
	p->thread = (struct thread_struct) INIT_THREAD;
188

A
Al Viro 已提交
189
	if (!kthread) {
190 191
	  	memcpy(&p->thread.regs.regs, &regs->regs,
		       sizeof(p->thread.regs.regs));
192
		PT_REGS_SET_SYSCALL_RETURN(&p->thread.regs, 0);
J
Jeff Dike 已提交
193
		if (sp != 0)
194
			REGS_SP(p->thread.regs.regs.gp) = sp;
195

196
		handler = fork_handler;
197

198
		arch_copy_thread(&current->thread.arch, &p->thread.arch);
A
Al Viro 已提交
199
	} else {
200
		get_safe_registers(p->thread.regs.regs.gp, p->thread.regs.regs.fp);
201 202 203 204 205 206
		p->thread.request.u.thread = current->thread.request.u.thread;
		handler = new_thread_handler;
	}

	new_thread(task_stack_page(p), &p->thread.switch_buf, handler);

A
Al Viro 已提交
207
	if (!kthread) {
208 209 210 211 212 213 214 215
		clear_flushed_tls(p);

		/*
		 * Set a new TLS for the child thread?
		 */
		if (clone_flags & CLONE_SETTLS)
			ret = arch_copy_tls(p);
	}
216 217

	return ret;
L
Linus Torvalds 已提交
218 219 220 221 222 223 224
}

void initial_thread_cb(void (*proc)(void *), void *arg)
{
	int save_kmalloc_ok = kmalloc_ok;

	kmalloc_ok = 0;
J
Jeff Dike 已提交
225
	initial_thread_cb_skas(proc, arg);
L
Linus Torvalds 已提交
226 227
	kmalloc_ok = save_kmalloc_ok;
}
J
Jeff Dike 已提交
228

L
Linus Torvalds 已提交
229 230
void default_idle(void)
{
231 232
	unsigned long long nsecs;

J
Jeff Dike 已提交
233
	while (1) {
L
Linus Torvalds 已提交
234 235 236 237 238 239
		/* endless idle loop with no priority at all */

		/*
		 * although we are an idle CPU, we do not want to
		 * get into the scheduler unnecessarily.
		 */
J
Jeff Dike 已提交
240
		if (need_resched())
L
Linus Torvalds 已提交
241
			schedule();
J
Jeff Dike 已提交
242

243 244
		tick_nohz_idle_enter();
		rcu_idle_enter();
245 246
		nsecs = disable_timer();
		idle_sleep(nsecs);
247 248
		rcu_idle_exit();
		tick_nohz_idle_exit();
L
Linus Torvalds 已提交
249 250 251 252 253
	}
}

void cpu_idle(void)
{
J
Jeff Dike 已提交
254
	cpu_tasks[current_thread_info()->cpu].pid = os_getpid();
255
	default_idle();
L
Linus Torvalds 已提交
256 257
}

258 259 260
int __cant_sleep(void) {
	return in_atomic() || irqs_disabled() || in_interrupt();
	/* Is in_interrupt() really needed? */
L
Linus Torvalds 已提交
261 262 263 264 265 266 267
}

int user_context(unsigned long sp)
{
	unsigned long stack;

	stack = sp & (PAGE_MASK << CONFIG_KERNEL_STACK_ORDER);
J
Jeff Dike 已提交
268
	return stack != (unsigned long) current_thread_info();
L
Linus Torvalds 已提交
269 270 271 272 273 274 275 276 277 278 279 280 281
}

extern exitcall_t __uml_exitcall_begin, __uml_exitcall_end;

void do_uml_exitcalls(void)
{
	exitcall_t *call;

	call = &__uml_exitcall_end;
	while (--call >= &__uml_exitcall_begin)
		(*call)();
}

W
WANG Cong 已提交
282
char *uml_strdup(const char *string)
L
Linus Torvalds 已提交
283
{
284
	return kstrdup(string, GFP_KERNEL);
L
Linus Torvalds 已提交
285
}
286
EXPORT_SYMBOL(uml_strdup);
L
Linus Torvalds 已提交
287 288 289

int copy_to_user_proc(void __user *to, void *from, int size)
{
J
Jeff Dike 已提交
290
	return copy_to_user(to, from, size);
L
Linus Torvalds 已提交
291 292 293 294
}

int copy_from_user_proc(void *to, void __user *from, int size)
{
J
Jeff Dike 已提交
295
	return copy_from_user(to, from, size);
L
Linus Torvalds 已提交
296 297 298 299
}

int clear_user_proc(void __user *buf, int size)
{
J
Jeff Dike 已提交
300
	return clear_user(buf, size);
L
Linus Torvalds 已提交
301 302 303 304
}

int strlen_user_proc(char __user *str)
{
J
Jeff Dike 已提交
305
	return strlen_user(str);
L
Linus Torvalds 已提交
306 307 308 309 310
}

int smp_sigio_handler(void)
{
#ifdef CONFIG_SMP
J
Jeff Dike 已提交
311
	int cpu = current_thread_info()->cpu;
L
Linus Torvalds 已提交
312
	IPI_handler(cpu);
J
Jeff Dike 已提交
313
	if (cpu != 0)
J
Jeff Dike 已提交
314
		return 1;
L
Linus Torvalds 已提交
315
#endif
J
Jeff Dike 已提交
316
	return 0;
L
Linus Torvalds 已提交
317 318 319 320
}

int cpu(void)
{
J
Jeff Dike 已提交
321
	return current_thread_info()->cpu;
L
Linus Torvalds 已提交
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338
}

static atomic_t using_sysemu = ATOMIC_INIT(0);
int sysemu_supported;

void set_using_sysemu(int value)
{
	if (value > sysemu_supported)
		return;
	atomic_set(&using_sysemu, value);
}

int get_using_sysemu(void)
{
	return atomic_read(&using_sysemu);
}

339
static int sysemu_proc_show(struct seq_file *m, void *v)
L
Linus Torvalds 已提交
340
{
341 342 343
	seq_printf(m, "%d\n", get_using_sysemu());
	return 0;
}
L
Linus Torvalds 已提交
344

345 346 347
static int sysemu_proc_open(struct inode *inode, struct file *file)
{
	return single_open(file, sysemu_proc_show, NULL);
L
Linus Torvalds 已提交
348 349
}

350 351
static ssize_t sysemu_proc_write(struct file *file, const char __user *buf,
				 size_t count, loff_t *pos)
L
Linus Torvalds 已提交
352 353 354 355 356 357 358 359
{
	char tmp[2];

	if (copy_from_user(tmp, buf, 1))
		return -EFAULT;

	if (tmp[0] >= '0' && tmp[0] <= '2')
		set_using_sysemu(tmp[0] - '0');
J
Jeff Dike 已提交
360 361
	/* We use the first char, but pretend to write everything */
	return count;
L
Linus Torvalds 已提交
362 363
}

364 365 366 367 368 369 370 371 372
static const struct file_operations sysemu_proc_fops = {
	.owner		= THIS_MODULE,
	.open		= sysemu_proc_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= single_release,
	.write		= sysemu_proc_write,
};

L
Linus Torvalds 已提交
373 374 375 376 377 378
int __init make_proc_sysemu(void)
{
	struct proc_dir_entry *ent;
	if (!sysemu_supported)
		return 0;

379
	ent = proc_create("sysemu", 0600, NULL, &sysemu_proc_fops);
L
Linus Torvalds 已提交
380 381 382

	if (ent == NULL)
	{
383
		printk(KERN_WARNING "Failed to register /proc/sysemu\n");
J
Jeff Dike 已提交
384
		return 0;
L
Linus Torvalds 已提交
385 386 387 388 389 390 391 392 393 394 395
	}

	return 0;
}

late_initcall(make_proc_sysemu);

int singlestepping(void * t)
{
	struct task_struct *task = t ? t : current;

J
Jeff Dike 已提交
396
	if (!(task->ptrace & PT_DTRACE))
J
Jeff Dike 已提交
397
		return 0;
L
Linus Torvalds 已提交
398 399

	if (task->thread.singlestep_syscall)
J
Jeff Dike 已提交
400
		return 1;
L
Linus Torvalds 已提交
401 402 403 404

	return 2;
}

405 406 407 408 409 410 411 412
/*
 * Only x86 and x86_64 have an arch_align_stack().
 * All other arches have "#define arch_align_stack(x) (x)"
 * in their asm/system.h
 * As this is included in UML from asm-um/system-generic.h,
 * we can use it to behave as the subarch does.
 */
#ifndef arch_align_stack
L
Linus Torvalds 已提交
413 414
unsigned long arch_align_stack(unsigned long sp)
{
J
Jeff Dike 已提交
415
	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
L
Linus Torvalds 已提交
416 417 418
		sp -= get_random_int() % 8192;
	return sp & ~0xf;
}
419
#endif
J
Jeff Dike 已提交
420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454

unsigned long get_wchan(struct task_struct *p)
{
	unsigned long stack_page, sp, ip;
	bool seen_sched = 0;

	if ((p == NULL) || (p == current) || (p->state == TASK_RUNNING))
		return 0;

	stack_page = (unsigned long) task_stack_page(p);
	/* Bail if the process has no kernel stack for some reason */
	if (stack_page == 0)
		return 0;

	sp = p->thread.switch_buf->JB_SP;
	/*
	 * Bail if the stack pointer is below the bottom of the kernel
	 * stack for some reason
	 */
	if (sp < stack_page)
		return 0;

	while (sp < stack_page + THREAD_SIZE) {
		ip = *((unsigned long *) sp);
		if (in_sched_functions(ip))
			/* Ignore everything until we're above the scheduler */
			seen_sched = 1;
		else if (kernel_text_address(ip) && seen_sched)
			return ip;

		sp += sizeof(unsigned long);
	}

	return 0;
}
J
Jeff Dike 已提交
455 456 457 458 459 460 461 462

int elf_core_copy_fpregs(struct task_struct *t, elf_fpregset_t *fpu)
{
	int cpu = current_thread_info()->cpu;

	return save_fp_registers(userspace_pid[cpu], (unsigned long *) fpu);
}