xref: /linux/arch/x86/kernel/process_32.c (revision 5bdef865eb358b6f3760e25e591ae115e9eeddef)
1 /*
2  *  Copyright (C) 1995  Linus Torvalds
3  *
4  *  Pentium III FXSR, SSE support
5  *	Gareth Hughes <gareth@valinux.com>, May 2000
6  */
7 
8 /*
9  * This file handles the architecture-dependent parts of process handling..
10  */
11 
12 #include <linux/stackprotector.h>
13 #include <linux/cpu.h>
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/fs.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/elfcore.h>
20 #include <linux/smp.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/user.h>
25 #include <linux/interrupt.h>
26 #include <linux/utsname.h>
27 #include <linux/delay.h>
28 #include <linux/reboot.h>
29 #include <linux/init.h>
30 #include <linux/mc146818rtc.h>
31 #include <linux/module.h>
32 #include <linux/kallsyms.h>
33 #include <linux/ptrace.h>
34 #include <linux/personality.h>
35 #include <linux/tick.h>
36 #include <linux/percpu.h>
37 #include <linux/prctl.h>
38 #include <linux/dmi.h>
39 #include <linux/ftrace.h>
40 #include <linux/uaccess.h>
41 #include <linux/io.h>
42 #include <linux/kdebug.h>
43 
44 #include <asm/pgtable.h>
45 #include <asm/system.h>
46 #include <asm/ldt.h>
47 #include <asm/processor.h>
48 #include <asm/i387.h>
49 #include <asm/desc.h>
50 #ifdef CONFIG_MATH_EMULATION
51 #include <asm/math_emu.h>
52 #endif
53 
54 #include <linux/err.h>
55 
56 #include <asm/tlbflush.h>
57 #include <asm/cpu.h>
58 #include <asm/idle.h>
59 #include <asm/syscalls.h>
60 #include <asm/ds.h>
61 
62 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
63 
64 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
65 EXPORT_PER_CPU_SYMBOL(current_task);
66 
67 /*
68  * Return saved PC of a blocked thread.
69  */
70 unsigned long thread_saved_pc(struct task_struct *tsk)
71 {
72 	return ((unsigned long *)tsk->thread.sp)[3];
73 }
74 
75 #ifndef CONFIG_SMP
76 static inline void play_dead(void)
77 {
78 	BUG();
79 }
80 #endif
81 
82 /*
83  * The idle thread. There's no useful work to be
84  * done, so just try to conserve power and have a
85  * low exit latency (ie sit in a loop waiting for
86  * somebody to say that they'd like to reschedule)
87  */
88 void cpu_idle(void)
89 {
90 	int cpu = smp_processor_id();
91 
92 	/*
93 	 * If we're the non-boot CPU, nothing set the stack canary up
94 	 * for us.  CPU0 already has it initialized but no harm in
95 	 * doing it again.  This is a good place for updating it, as
96 	 * we wont ever return from this function (so the invalid
97 	 * canaries already on the stack wont ever trigger).
98 	 */
99 	boot_init_stack_canary();
100 
101 	current_thread_info()->status |= TS_POLLING;
102 
103 	/* endless idle loop with no priority at all */
104 	while (1) {
105 		tick_nohz_stop_sched_tick(1);
106 		while (!need_resched()) {
107 
108 			check_pgt_cache();
109 			rmb();
110 
111 			if (cpu_is_offline(cpu))
112 				play_dead();
113 
114 			local_irq_disable();
115 			/* Don't trace irqs off for idle */
116 			stop_critical_timings();
117 			pm_idle();
118 			start_critical_timings();
119 		}
120 		tick_nohz_restart_sched_tick();
121 		preempt_enable_no_resched();
122 		schedule();
123 		preempt_disable();
124 	}
125 }
126 
127 void __show_regs(struct pt_regs *regs, int all)
128 {
129 	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
130 	unsigned long d0, d1, d2, d3, d6, d7;
131 	unsigned long sp;
132 	unsigned short ss, gs;
133 	const char *board;
134 
135 	if (user_mode_vm(regs)) {
136 		sp = regs->sp;
137 		ss = regs->ss & 0xffff;
138 		gs = get_user_gs(regs);
139 	} else {
140 		sp = (unsigned long) (&regs->sp);
141 		savesegment(ss, ss);
142 		savesegment(gs, gs);
143 	}
144 
145 	printk("\n");
146 
147 	board = dmi_get_system_info(DMI_PRODUCT_NAME);
148 	if (!board)
149 		board = "";
150 	printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
151 			task_pid_nr(current), current->comm,
152 			print_tainted(), init_utsname()->release,
153 			(int)strcspn(init_utsname()->version, " "),
154 			init_utsname()->version, board);
155 
156 	printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
157 			(u16)regs->cs, regs->ip, regs->flags,
158 			smp_processor_id());
159 	print_symbol("EIP is at %s\n", regs->ip);
160 
161 	printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
162 		regs->ax, regs->bx, regs->cx, regs->dx);
163 	printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
164 		regs->si, regs->di, regs->bp, sp);
165 	printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
166 	       (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
167 
168 	if (!all)
169 		return;
170 
171 	cr0 = read_cr0();
172 	cr2 = read_cr2();
173 	cr3 = read_cr3();
174 	cr4 = read_cr4_safe();
175 	printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
176 			cr0, cr2, cr3, cr4);
177 
178 	get_debugreg(d0, 0);
179 	get_debugreg(d1, 1);
180 	get_debugreg(d2, 2);
181 	get_debugreg(d3, 3);
182 	printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
183 			d0, d1, d2, d3);
184 
185 	get_debugreg(d6, 6);
186 	get_debugreg(d7, 7);
187 	printk("DR6: %08lx DR7: %08lx\n",
188 			d6, d7);
189 }
190 
191 void show_regs(struct pt_regs *regs)
192 {
193 	__show_regs(regs, 1);
194 	show_trace(NULL, regs, &regs->sp, regs->bp);
195 }
196 
197 /*
198  * This gets run with %bx containing the
199  * function to call, and %dx containing
200  * the "args".
201  */
202 extern void kernel_thread_helper(void);
203 
204 /*
205  * Create a kernel thread
206  */
207 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
208 {
209 	struct pt_regs regs;
210 
211 	memset(&regs, 0, sizeof(regs));
212 
213 	regs.bx = (unsigned long) fn;
214 	regs.dx = (unsigned long) arg;
215 
216 	regs.ds = __USER_DS;
217 	regs.es = __USER_DS;
218 	regs.fs = __KERNEL_PERCPU;
219 	regs.gs = __KERNEL_STACK_CANARY;
220 	regs.orig_ax = -1;
221 	regs.ip = (unsigned long) kernel_thread_helper;
222 	regs.cs = __KERNEL_CS | get_kernel_rpl();
223 	regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
224 
225 	/* Ok, create the new process.. */
226 	return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
227 }
228 EXPORT_SYMBOL(kernel_thread);
229 
230 void release_thread(struct task_struct *dead_task)
231 {
232 	BUG_ON(dead_task->mm);
233 	release_vm86_irqs(dead_task);
234 }
235 
236 /*
237  * This gets called before we allocate a new thread and copy
238  * the current task into it.
239  */
240 void prepare_to_copy(struct task_struct *tsk)
241 {
242 	unlazy_fpu(tsk);
243 }
244 
245 int copy_thread(unsigned long clone_flags, unsigned long sp,
246 	unsigned long unused,
247 	struct task_struct *p, struct pt_regs *regs)
248 {
249 	struct pt_regs *childregs;
250 	struct task_struct *tsk;
251 	int err;
252 
253 	childregs = task_pt_regs(p);
254 	*childregs = *regs;
255 	childregs->ax = 0;
256 	childregs->sp = sp;
257 
258 	p->thread.sp = (unsigned long) childregs;
259 	p->thread.sp0 = (unsigned long) (childregs+1);
260 
261 	p->thread.ip = (unsigned long) ret_from_fork;
262 
263 	task_user_gs(p) = get_user_gs(regs);
264 
265 	tsk = current;
266 	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
267 		p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
268 						IO_BITMAP_BYTES, GFP_KERNEL);
269 		if (!p->thread.io_bitmap_ptr) {
270 			p->thread.io_bitmap_max = 0;
271 			return -ENOMEM;
272 		}
273 		set_tsk_thread_flag(p, TIF_IO_BITMAP);
274 	}
275 
276 	err = 0;
277 
278 	/*
279 	 * Set a new TLS for the child thread?
280 	 */
281 	if (clone_flags & CLONE_SETTLS)
282 		err = do_set_thread_area(p, -1,
283 			(struct user_desc __user *)childregs->si, 0);
284 
285 	if (err && p->thread.io_bitmap_ptr) {
286 		kfree(p->thread.io_bitmap_ptr);
287 		p->thread.io_bitmap_max = 0;
288 	}
289 
290 	clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
291 	p->thread.ds_ctx = NULL;
292 
293 	clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
294 	p->thread.debugctlmsr = 0;
295 
296 	return err;
297 }
298 
299 void
300 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
301 {
302 	set_user_gs(regs, 0);
303 	regs->fs		= 0;
304 	set_fs(USER_DS);
305 	regs->ds		= __USER_DS;
306 	regs->es		= __USER_DS;
307 	regs->ss		= __USER_DS;
308 	regs->cs		= __USER_CS;
309 	regs->ip		= new_ip;
310 	regs->sp		= new_sp;
311 	/*
312 	 * Free the old FP and other extended state
313 	 */
314 	free_thread_xstate(current);
315 }
316 EXPORT_SYMBOL_GPL(start_thread);
317 
318 
319 /*
320  *	switch_to(x,yn) should switch tasks from x to y.
321  *
322  * We fsave/fwait so that an exception goes off at the right time
323  * (as a call from the fsave or fwait in effect) rather than to
324  * the wrong process. Lazy FP saving no longer makes any sense
325  * with modern CPU's, and this simplifies a lot of things (SMP
326  * and UP become the same).
327  *
328  * NOTE! We used to use the x86 hardware context switching. The
329  * reason for not using it any more becomes apparent when you
330  * try to recover gracefully from saved state that is no longer
331  * valid (stale segment register values in particular). With the
332  * hardware task-switch, there is no way to fix up bad state in
333  * a reasonable manner.
334  *
335  * The fact that Intel documents the hardware task-switching to
336  * be slow is a fairly red herring - this code is not noticeably
337  * faster. However, there _is_ some room for improvement here,
338  * so the performance issues may eventually be a valid point.
339  * More important, however, is the fact that this allows us much
340  * more flexibility.
341  *
342  * The return value (in %ax) will be the "prev" task after
343  * the task-switch, and shows up in ret_from_fork in entry.S,
344  * for example.
345  */
346 __notrace_funcgraph struct task_struct *
347 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
348 {
349 	struct thread_struct *prev = &prev_p->thread,
350 				 *next = &next_p->thread;
351 	int cpu = smp_processor_id();
352 	struct tss_struct *tss = &per_cpu(init_tss, cpu);
353 
354 	/* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
355 
356 	__unlazy_fpu(prev_p);
357 
358 
359 	/* we're going to use this soon, after a few expensive things */
360 	if (next_p->fpu_counter > 5)
361 		prefetch(next->xstate);
362 
363 	/*
364 	 * Reload esp0.
365 	 */
366 	load_sp0(tss, next);
367 
368 	/*
369 	 * Save away %gs. No need to save %fs, as it was saved on the
370 	 * stack on entry.  No need to save %es and %ds, as those are
371 	 * always kernel segments while inside the kernel.  Doing this
372 	 * before setting the new TLS descriptors avoids the situation
373 	 * where we temporarily have non-reloadable segments in %fs
374 	 * and %gs.  This could be an issue if the NMI handler ever
375 	 * used %fs or %gs (it does not today), or if the kernel is
376 	 * running inside of a hypervisor layer.
377 	 */
378 	lazy_save_gs(prev->gs);
379 
380 	/*
381 	 * Load the per-thread Thread-Local Storage descriptor.
382 	 */
383 	load_TLS(next, cpu);
384 
385 	/*
386 	 * Restore IOPL if needed.  In normal use, the flags restore
387 	 * in the switch assembly will handle this.  But if the kernel
388 	 * is running virtualized at a non-zero CPL, the popf will
389 	 * not restore flags, so it must be done in a separate step.
390 	 */
391 	if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
392 		set_iopl_mask(next->iopl);
393 
394 	/*
395 	 * Now maybe handle debug registers and/or IO bitmaps
396 	 */
397 	if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
398 		     task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
399 		__switch_to_xtra(prev_p, next_p, tss);
400 
401 	/*
402 	 * Leave lazy mode, flushing any hypercalls made here.
403 	 * This must be done before restoring TLS segments so
404 	 * the GDT and LDT are properly updated, and must be
405 	 * done before math_state_restore, so the TS bit is up
406 	 * to date.
407 	 */
408 	arch_end_context_switch(next_p);
409 
410 	/* If the task has used fpu the last 5 timeslices, just do a full
411 	 * restore of the math state immediately to avoid the trap; the
412 	 * chances of needing FPU soon are obviously high now
413 	 *
414 	 * tsk_used_math() checks prevent calling math_state_restore(),
415 	 * which can sleep in the case of !tsk_used_math()
416 	 */
417 	if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
418 		math_state_restore();
419 
420 	/*
421 	 * Restore %gs if needed (which is common)
422 	 */
423 	if (prev->gs | next->gs)
424 		lazy_load_gs(next->gs);
425 
426 	percpu_write(current_task, next_p);
427 
428 	return prev_p;
429 }
430 
431 int sys_clone(struct pt_regs *regs)
432 {
433 	unsigned long clone_flags;
434 	unsigned long newsp;
435 	int __user *parent_tidptr, *child_tidptr;
436 
437 	clone_flags = regs->bx;
438 	newsp = regs->cx;
439 	parent_tidptr = (int __user *)regs->dx;
440 	child_tidptr = (int __user *)regs->di;
441 	if (!newsp)
442 		newsp = regs->sp;
443 	return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr);
444 }
445 
446 /*
447  * sys_execve() executes a new program.
448  */
449 int sys_execve(struct pt_regs *regs)
450 {
451 	int error;
452 	char *filename;
453 
454 	filename = getname((char __user *) regs->bx);
455 	error = PTR_ERR(filename);
456 	if (IS_ERR(filename))
457 		goto out;
458 	error = do_execve(filename,
459 			(char __user * __user *) regs->cx,
460 			(char __user * __user *) regs->dx,
461 			regs);
462 	if (error == 0) {
463 		/* Make sure we don't return using sysenter.. */
464 		set_thread_flag(TIF_IRET);
465 	}
466 	putname(filename);
467 out:
468 	return error;
469 }
470 
471 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
472 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
473 
474 unsigned long get_wchan(struct task_struct *p)
475 {
476 	unsigned long bp, sp, ip;
477 	unsigned long stack_page;
478 	int count = 0;
479 	if (!p || p == current || p->state == TASK_RUNNING)
480 		return 0;
481 	stack_page = (unsigned long)task_stack_page(p);
482 	sp = p->thread.sp;
483 	if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
484 		return 0;
485 	/* include/asm-i386/system.h:switch_to() pushes bp last. */
486 	bp = *(unsigned long *) sp;
487 	do {
488 		if (bp < stack_page || bp > top_ebp+stack_page)
489 			return 0;
490 		ip = *(unsigned long *) (bp+4);
491 		if (!in_sched_functions(ip))
492 			return ip;
493 		bp = *(unsigned long *) bp;
494 	} while (count++ < 16);
495 	return 0;
496 }
497 
498