Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/i386/kernel/process.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * | |
6 | * Pentium III FXSR, SSE support | |
7 | * Gareth Hughes <gareth@valinux.com>, May 2000 | |
8 | */ | |
9 | ||
10 | /* | |
11 | * This file handles the architecture-dependent parts of process handling.. | |
12 | */ | |
13 | ||
14 | #include <stdarg.h> | |
15 | ||
16 | #include <linux/errno.h> | |
17 | #include <linux/sched.h> | |
18 | #include <linux/fs.h> | |
19 | #include <linux/kernel.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/elfcore.h> | |
22 | #include <linux/smp.h> | |
23 | #include <linux/smp_lock.h> | |
24 | #include <linux/stddef.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/vmalloc.h> | |
27 | #include <linux/user.h> | |
28 | #include <linux/a.out.h> | |
29 | #include <linux/interrupt.h> | |
30 | #include <linux/config.h> | |
31 | #include <linux/utsname.h> | |
32 | #include <linux/delay.h> | |
33 | #include <linux/reboot.h> | |
34 | #include <linux/init.h> | |
35 | #include <linux/mc146818rtc.h> | |
36 | #include <linux/module.h> | |
37 | #include <linux/kallsyms.h> | |
38 | #include <linux/ptrace.h> | |
39 | #include <linux/random.h> | |
40 | ||
41 | #include <asm/uaccess.h> | |
42 | #include <asm/pgtable.h> | |
43 | #include <asm/system.h> | |
44 | #include <asm/io.h> | |
45 | #include <asm/ldt.h> | |
46 | #include <asm/processor.h> | |
47 | #include <asm/i387.h> | |
48 | #include <asm/irq.h> | |
49 | #include <asm/desc.h> | |
50 | #ifdef CONFIG_MATH_EMULATION | |
51 | #include <asm/math_emu.h> | |
52 | #endif | |
53 | ||
54 | #include <linux/irq.h> | |
55 | #include <linux/err.h> | |
56 | ||
57 | asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); | |
58 | ||
59 | static int hlt_counter; | |
60 | ||
61 | unsigned long boot_option_idle_override = 0; | |
62 | EXPORT_SYMBOL(boot_option_idle_override); | |
63 | ||
64 | /* | |
65 | * Return saved PC of a blocked thread. | |
66 | */ | |
67 | unsigned long thread_saved_pc(struct task_struct *tsk) | |
68 | { | |
69 | return ((unsigned long *)tsk->thread.esp)[3]; | |
70 | } | |
71 | ||
72 | /* | |
73 | * Powermanagement idle function, if any.. | |
74 | */ | |
75 | void (*pm_idle)(void); | |
76 | static DEFINE_PER_CPU(unsigned int, cpu_idle_state); | |
77 | ||
78 | void disable_hlt(void) | |
79 | { | |
80 | hlt_counter++; | |
81 | } | |
82 | ||
83 | EXPORT_SYMBOL(disable_hlt); | |
84 | ||
85 | void enable_hlt(void) | |
86 | { | |
87 | hlt_counter--; | |
88 | } | |
89 | ||
90 | EXPORT_SYMBOL(enable_hlt); | |
91 | ||
92 | /* | |
93 | * We use this if we don't have any better | |
94 | * idle routine.. | |
95 | */ | |
96 | void default_idle(void) | |
97 | { | |
98 | if (!hlt_counter && boot_cpu_data.hlt_works_ok) { | |
99 | local_irq_disable(); | |
100 | if (!need_resched()) | |
101 | safe_halt(); | |
102 | else | |
103 | local_irq_enable(); | |
104 | } else { | |
105 | cpu_relax(); | |
106 | } | |
107 | } | |
108 | ||
109 | /* | |
110 | * On SMP it's slightly faster (but much more power-consuming!) | |
111 | * to poll the ->work.need_resched flag instead of waiting for the | |
112 | * cross-CPU IPI to arrive. Use this option with caution. | |
113 | */ | |
114 | static void poll_idle (void) | |
115 | { | |
116 | int oldval; | |
117 | ||
118 | local_irq_enable(); | |
119 | ||
120 | /* | |
121 | * Deal with another CPU just having chosen a thread to | |
122 | * run here: | |
123 | */ | |
124 | oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED); | |
125 | ||
126 | if (!oldval) { | |
127 | set_thread_flag(TIF_POLLING_NRFLAG); | |
128 | asm volatile( | |
129 | "2:" | |
130 | "testl %0, %1;" | |
131 | "rep; nop;" | |
132 | "je 2b;" | |
133 | : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags)); | |
134 | ||
135 | clear_thread_flag(TIF_POLLING_NRFLAG); | |
136 | } else { | |
137 | set_need_resched(); | |
138 | } | |
139 | } | |
140 | ||
141 | /* | |
142 | * The idle thread. There's no useful work to be | |
143 | * done, so just try to conserve power and have a | |
144 | * low exit latency (ie sit in a loop waiting for | |
145 | * somebody to say that they'd like to reschedule) | |
146 | */ | |
147 | void cpu_idle (void) | |
148 | { | |
149 | /* endless idle loop with no priority at all */ | |
150 | while (1) { | |
151 | while (!need_resched()) { | |
152 | void (*idle)(void); | |
153 | ||
154 | if (__get_cpu_var(cpu_idle_state)) | |
155 | __get_cpu_var(cpu_idle_state) = 0; | |
156 | ||
157 | rmb(); | |
158 | idle = pm_idle; | |
159 | ||
160 | if (!idle) | |
161 | idle = default_idle; | |
162 | ||
163 | __get_cpu_var(irq_stat).idle_timestamp = jiffies; | |
164 | idle(); | |
165 | } | |
166 | schedule(); | |
167 | } | |
168 | } | |
169 | ||
170 | void cpu_idle_wait(void) | |
171 | { | |
172 | unsigned int cpu, this_cpu = get_cpu(); | |
173 | cpumask_t map; | |
174 | ||
175 | set_cpus_allowed(current, cpumask_of_cpu(this_cpu)); | |
176 | put_cpu(); | |
177 | ||
178 | cpus_clear(map); | |
179 | for_each_online_cpu(cpu) { | |
180 | per_cpu(cpu_idle_state, cpu) = 1; | |
181 | cpu_set(cpu, map); | |
182 | } | |
183 | ||
184 | __get_cpu_var(cpu_idle_state) = 0; | |
185 | ||
186 | wmb(); | |
187 | do { | |
188 | ssleep(1); | |
189 | for_each_online_cpu(cpu) { | |
190 | if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu)) | |
191 | cpu_clear(cpu, map); | |
192 | } | |
193 | cpus_and(map, map, cpu_online_map); | |
194 | } while (!cpus_empty(map)); | |
195 | } | |
196 | EXPORT_SYMBOL_GPL(cpu_idle_wait); | |
197 | ||
198 | /* | |
199 | * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, | |
200 | * which can obviate IPI to trigger checking of need_resched. | |
201 | * We execute MONITOR against need_resched and enter optimized wait state | |
202 | * through MWAIT. Whenever someone changes need_resched, we would be woken | |
203 | * up from MWAIT (without an IPI). | |
204 | */ | |
205 | static void mwait_idle(void) | |
206 | { | |
207 | local_irq_enable(); | |
208 | ||
209 | if (!need_resched()) { | |
210 | set_thread_flag(TIF_POLLING_NRFLAG); | |
211 | do { | |
212 | __monitor((void *)¤t_thread_info()->flags, 0, 0); | |
213 | if (need_resched()) | |
214 | break; | |
215 | __mwait(0, 0); | |
216 | } while (!need_resched()); | |
217 | clear_thread_flag(TIF_POLLING_NRFLAG); | |
218 | } | |
219 | } | |
220 | ||
221 | void __init select_idle_routine(const struct cpuinfo_x86 *c) | |
222 | { | |
223 | if (cpu_has(c, X86_FEATURE_MWAIT)) { | |
224 | printk("monitor/mwait feature present.\n"); | |
225 | /* | |
226 | * Skip, if setup has overridden idle. | |
227 | * One CPU supports mwait => All CPUs supports mwait | |
228 | */ | |
229 | if (!pm_idle) { | |
230 | printk("using mwait in idle threads.\n"); | |
231 | pm_idle = mwait_idle; | |
232 | } | |
233 | } | |
234 | } | |
235 | ||
236 | static int __init idle_setup (char *str) | |
237 | { | |
238 | if (!strncmp(str, "poll", 4)) { | |
239 | printk("using polling idle threads.\n"); | |
240 | pm_idle = poll_idle; | |
241 | #ifdef CONFIG_X86_SMP | |
242 | if (smp_num_siblings > 1) | |
243 | printk("WARNING: polling idle and HT enabled, performance may degrade.\n"); | |
244 | #endif | |
245 | } else if (!strncmp(str, "halt", 4)) { | |
246 | printk("using halt in idle threads.\n"); | |
247 | pm_idle = default_idle; | |
248 | } | |
249 | ||
250 | boot_option_idle_override = 1; | |
251 | return 1; | |
252 | } | |
253 | ||
254 | __setup("idle=", idle_setup); | |
255 | ||
256 | void show_regs(struct pt_regs * regs) | |
257 | { | |
258 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; | |
259 | ||
260 | printk("\n"); | |
261 | printk("Pid: %d, comm: %20s\n", current->pid, current->comm); | |
262 | printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id()); | |
263 | print_symbol("EIP is at %s\n", regs->eip); | |
264 | ||
265 | if (regs->xcs & 3) | |
266 | printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp); | |
267 | printk(" EFLAGS: %08lx %s (%s)\n", | |
268 | regs->eflags, print_tainted(), system_utsname.release); | |
269 | printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", | |
270 | regs->eax,regs->ebx,regs->ecx,regs->edx); | |
271 | printk("ESI: %08lx EDI: %08lx EBP: %08lx", | |
272 | regs->esi, regs->edi, regs->ebp); | |
273 | printk(" DS: %04x ES: %04x\n", | |
274 | 0xffff & regs->xds,0xffff & regs->xes); | |
275 | ||
276 | __asm__("movl %%cr0, %0": "=r" (cr0)); | |
277 | __asm__("movl %%cr2, %0": "=r" (cr2)); | |
278 | __asm__("movl %%cr3, %0": "=r" (cr3)); | |
279 | /* This could fault if %cr4 does not exist */ | |
280 | __asm__("1: movl %%cr4, %0 \n" | |
281 | "2: \n" | |
282 | ".section __ex_table,\"a\" \n" | |
283 | ".long 1b,2b \n" | |
284 | ".previous \n" | |
285 | : "=r" (cr4): "0" (0)); | |
286 | printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4); | |
287 | show_trace(NULL, ®s->esp); | |
288 | } | |
289 | ||
290 | /* | |
291 | * This gets run with %ebx containing the | |
292 | * function to call, and %edx containing | |
293 | * the "args". | |
294 | */ | |
295 | extern void kernel_thread_helper(void); | |
296 | __asm__(".section .text\n" | |
297 | ".align 4\n" | |
298 | "kernel_thread_helper:\n\t" | |
299 | "movl %edx,%eax\n\t" | |
300 | "pushl %edx\n\t" | |
301 | "call *%ebx\n\t" | |
302 | "pushl %eax\n\t" | |
303 | "call do_exit\n" | |
304 | ".previous"); | |
305 | ||
306 | /* | |
307 | * Create a kernel thread | |
308 | */ | |
309 | int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags) | |
310 | { | |
311 | struct pt_regs regs; | |
312 | ||
313 | memset(®s, 0, sizeof(regs)); | |
314 | ||
315 | regs.ebx = (unsigned long) fn; | |
316 | regs.edx = (unsigned long) arg; | |
317 | ||
318 | regs.xds = __USER_DS; | |
319 | regs.xes = __USER_DS; | |
320 | regs.orig_eax = -1; | |
321 | regs.eip = (unsigned long) kernel_thread_helper; | |
322 | regs.xcs = __KERNEL_CS; | |
323 | regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2; | |
324 | ||
325 | /* Ok, create the new process.. */ | |
326 | return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL); | |
327 | } | |
328 | ||
329 | /* | |
330 | * Free current thread data structures etc.. | |
331 | */ | |
332 | void exit_thread(void) | |
333 | { | |
334 | struct task_struct *tsk = current; | |
335 | struct thread_struct *t = &tsk->thread; | |
336 | ||
337 | /* The process may have allocated an io port bitmap... nuke it. */ | |
338 | if (unlikely(NULL != t->io_bitmap_ptr)) { | |
339 | int cpu = get_cpu(); | |
340 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
341 | ||
342 | kfree(t->io_bitmap_ptr); | |
343 | t->io_bitmap_ptr = NULL; | |
344 | /* | |
345 | * Careful, clear this in the TSS too: | |
346 | */ | |
347 | memset(tss->io_bitmap, 0xff, tss->io_bitmap_max); | |
348 | t->io_bitmap_max = 0; | |
349 | tss->io_bitmap_owner = NULL; | |
350 | tss->io_bitmap_max = 0; | |
351 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET; | |
352 | put_cpu(); | |
353 | } | |
354 | } | |
355 | ||
356 | void flush_thread(void) | |
357 | { | |
358 | struct task_struct *tsk = current; | |
359 | ||
360 | memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8); | |
361 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); | |
362 | /* | |
363 | * Forget coprocessor state.. | |
364 | */ | |
365 | clear_fpu(tsk); | |
366 | clear_used_math(); | |
367 | } | |
368 | ||
369 | void release_thread(struct task_struct *dead_task) | |
370 | { | |
371 | if (dead_task->mm) { | |
372 | // temporary debugging check | |
373 | if (dead_task->mm->context.size) { | |
374 | printk("WARNING: dead process %8s still has LDT? <%p/%d>\n", | |
375 | dead_task->comm, | |
376 | dead_task->mm->context.ldt, | |
377 | dead_task->mm->context.size); | |
378 | BUG(); | |
379 | } | |
380 | } | |
381 | ||
382 | release_vm86_irqs(dead_task); | |
383 | } | |
384 | ||
385 | /* | |
386 | * This gets called before we allocate a new thread and copy | |
387 | * the current task into it. | |
388 | */ | |
389 | void prepare_to_copy(struct task_struct *tsk) | |
390 | { | |
391 | unlazy_fpu(tsk); | |
392 | } | |
393 | ||
394 | int copy_thread(int nr, unsigned long clone_flags, unsigned long esp, | |
395 | unsigned long unused, | |
396 | struct task_struct * p, struct pt_regs * regs) | |
397 | { | |
398 | struct pt_regs * childregs; | |
399 | struct task_struct *tsk; | |
400 | int err; | |
401 | ||
402 | childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1; | |
403 | *childregs = *regs; | |
404 | childregs->eax = 0; | |
405 | childregs->esp = esp; | |
406 | ||
407 | p->thread.esp = (unsigned long) childregs; | |
5df24082 SS |
408 | /* |
409 | * The below -8 is to reserve 8 bytes on top of the ring0 stack. | |
410 | * This is necessary to guarantee that the entire "struct pt_regs" | |
411 | * is accessable even if the CPU haven't stored the SS/ESP registers | |
412 | * on the stack (interrupt gate does not save these registers | |
413 | * when switching to the same priv ring). | |
414 | * Therefore beware: accessing the xss/esp fields of the | |
415 | * "struct pt_regs" is possible, but they may contain the | |
416 | * completely wrong values. | |
417 | */ | |
418 | p->thread.esp0 = (unsigned long) (childregs+1) - 8; | |
1da177e4 LT |
419 | |
420 | p->thread.eip = (unsigned long) ret_from_fork; | |
421 | ||
422 | savesegment(fs,p->thread.fs); | |
423 | savesegment(gs,p->thread.gs); | |
424 | ||
425 | tsk = current; | |
426 | if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) { | |
427 | p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL); | |
428 | if (!p->thread.io_bitmap_ptr) { | |
429 | p->thread.io_bitmap_max = 0; | |
430 | return -ENOMEM; | |
431 | } | |
432 | memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr, | |
433 | IO_BITMAP_BYTES); | |
434 | } | |
435 | ||
436 | /* | |
437 | * Set a new TLS for the child thread? | |
438 | */ | |
439 | if (clone_flags & CLONE_SETTLS) { | |
440 | struct desc_struct *desc; | |
441 | struct user_desc info; | |
442 | int idx; | |
443 | ||
444 | err = -EFAULT; | |
445 | if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info))) | |
446 | goto out; | |
447 | err = -EINVAL; | |
448 | if (LDT_empty(&info)) | |
449 | goto out; | |
450 | ||
451 | idx = info.entry_number; | |
452 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
453 | goto out; | |
454 | ||
455 | desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; | |
456 | desc->a = LDT_entry_a(&info); | |
457 | desc->b = LDT_entry_b(&info); | |
458 | } | |
459 | ||
460 | err = 0; | |
461 | out: | |
462 | if (err && p->thread.io_bitmap_ptr) { | |
463 | kfree(p->thread.io_bitmap_ptr); | |
464 | p->thread.io_bitmap_max = 0; | |
465 | } | |
466 | return err; | |
467 | } | |
468 | ||
469 | /* | |
470 | * fill in the user structure for a core dump.. | |
471 | */ | |
472 | void dump_thread(struct pt_regs * regs, struct user * dump) | |
473 | { | |
474 | int i; | |
475 | ||
476 | /* changed the size calculations - should hopefully work better. lbt */ | |
477 | dump->magic = CMAGIC; | |
478 | dump->start_code = 0; | |
479 | dump->start_stack = regs->esp & ~(PAGE_SIZE - 1); | |
480 | dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT; | |
481 | dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT; | |
482 | dump->u_dsize -= dump->u_tsize; | |
483 | dump->u_ssize = 0; | |
484 | for (i = 0; i < 8; i++) | |
485 | dump->u_debugreg[i] = current->thread.debugreg[i]; | |
486 | ||
487 | if (dump->start_stack < TASK_SIZE) | |
488 | dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT; | |
489 | ||
490 | dump->regs.ebx = regs->ebx; | |
491 | dump->regs.ecx = regs->ecx; | |
492 | dump->regs.edx = regs->edx; | |
493 | dump->regs.esi = regs->esi; | |
494 | dump->regs.edi = regs->edi; | |
495 | dump->regs.ebp = regs->ebp; | |
496 | dump->regs.eax = regs->eax; | |
497 | dump->regs.ds = regs->xds; | |
498 | dump->regs.es = regs->xes; | |
499 | savesegment(fs,dump->regs.fs); | |
500 | savesegment(gs,dump->regs.gs); | |
501 | dump->regs.orig_eax = regs->orig_eax; | |
502 | dump->regs.eip = regs->eip; | |
503 | dump->regs.cs = regs->xcs; | |
504 | dump->regs.eflags = regs->eflags; | |
505 | dump->regs.esp = regs->esp; | |
506 | dump->regs.ss = regs->xss; | |
507 | ||
508 | dump->u_fpvalid = dump_fpu (regs, &dump->i387); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Capture the user space registers if the task is not running (in user space) | |
513 | */ | |
514 | int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs) | |
515 | { | |
516 | struct pt_regs ptregs; | |
517 | ||
518 | ptregs = *(struct pt_regs *) | |
519 | ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs)); | |
520 | ptregs.xcs &= 0xffff; | |
521 | ptregs.xds &= 0xffff; | |
522 | ptregs.xes &= 0xffff; | |
523 | ptregs.xss &= 0xffff; | |
524 | ||
525 | elf_core_copy_regs(regs, &ptregs); | |
526 | ||
527 | return 1; | |
528 | } | |
529 | ||
530 | static inline void | |
531 | handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss) | |
532 | { | |
533 | if (!next->io_bitmap_ptr) { | |
534 | /* | |
535 | * Disable the bitmap via an invalid offset. We still cache | |
536 | * the previous bitmap owner and the IO bitmap contents: | |
537 | */ | |
538 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET; | |
539 | return; | |
540 | } | |
541 | if (likely(next == tss->io_bitmap_owner)) { | |
542 | /* | |
543 | * Previous owner of the bitmap (hence the bitmap content) | |
544 | * matches the next task, we dont have to do anything but | |
545 | * to set a valid offset in the TSS: | |
546 | */ | |
547 | tss->io_bitmap_base = IO_BITMAP_OFFSET; | |
548 | return; | |
549 | } | |
550 | /* | |
551 | * Lazy TSS's I/O bitmap copy. We set an invalid offset here | |
552 | * and we let the task to get a GPF in case an I/O instruction | |
553 | * is performed. The handler of the GPF will verify that the | |
554 | * faulting task has a valid I/O bitmap and, it true, does the | |
555 | * real copy and restart the instruction. This will save us | |
556 | * redundant copies when the currently switched task does not | |
557 | * perform any I/O during its timeslice. | |
558 | */ | |
559 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY; | |
560 | } | |
1da177e4 LT |
561 | |
562 | /* | |
563 | * switch_to(x,yn) should switch tasks from x to y. | |
564 | * | |
565 | * We fsave/fwait so that an exception goes off at the right time | |
566 | * (as a call from the fsave or fwait in effect) rather than to | |
567 | * the wrong process. Lazy FP saving no longer makes any sense | |
568 | * with modern CPU's, and this simplifies a lot of things (SMP | |
569 | * and UP become the same). | |
570 | * | |
571 | * NOTE! We used to use the x86 hardware context switching. The | |
572 | * reason for not using it any more becomes apparent when you | |
573 | * try to recover gracefully from saved state that is no longer | |
574 | * valid (stale segment register values in particular). With the | |
575 | * hardware task-switch, there is no way to fix up bad state in | |
576 | * a reasonable manner. | |
577 | * | |
578 | * The fact that Intel documents the hardware task-switching to | |
579 | * be slow is a fairly red herring - this code is not noticeably | |
580 | * faster. However, there _is_ some room for improvement here, | |
581 | * so the performance issues may eventually be a valid point. | |
582 | * More important, however, is the fact that this allows us much | |
583 | * more flexibility. | |
584 | * | |
585 | * The return value (in %eax) will be the "prev" task after | |
586 | * the task-switch, and shows up in ret_from_fork in entry.S, | |
587 | * for example. | |
588 | */ | |
589 | struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p) | |
590 | { | |
591 | struct thread_struct *prev = &prev_p->thread, | |
592 | *next = &next_p->thread; | |
593 | int cpu = smp_processor_id(); | |
594 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
595 | ||
596 | /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ | |
597 | ||
598 | __unlazy_fpu(prev_p); | |
599 | ||
600 | /* | |
601 | * Reload esp0, LDT and the page table pointer: | |
602 | */ | |
603 | load_esp0(tss, next); | |
604 | ||
605 | /* | |
606 | * Load the per-thread Thread-Local Storage descriptor. | |
607 | */ | |
608 | load_TLS(next, cpu); | |
609 | ||
610 | /* | |
611 | * Save away %fs and %gs. No need to save %es and %ds, as | |
612 | * those are always kernel segments while inside the kernel. | |
613 | */ | |
fd51f666 L |
614 | asm volatile("mov %%fs,%0":"=m" (prev->fs)); |
615 | asm volatile("mov %%gs,%0":"=m" (prev->gs)); | |
1da177e4 LT |
616 | |
617 | /* | |
618 | * Restore %fs and %gs if needed. | |
619 | */ | |
620 | if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) { | |
621 | loadsegment(fs, next->fs); | |
622 | loadsegment(gs, next->gs); | |
623 | } | |
624 | ||
625 | /* | |
626 | * Now maybe reload the debug registers | |
627 | */ | |
628 | if (unlikely(next->debugreg[7])) { | |
629 | loaddebug(next, 0); | |
630 | loaddebug(next, 1); | |
631 | loaddebug(next, 2); | |
632 | loaddebug(next, 3); | |
633 | /* no 4 and 5 */ | |
634 | loaddebug(next, 6); | |
635 | loaddebug(next, 7); | |
636 | } | |
637 | ||
638 | if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) | |
639 | handle_io_bitmap(next, tss); | |
640 | ||
641 | return prev_p; | |
642 | } | |
643 | ||
644 | asmlinkage int sys_fork(struct pt_regs regs) | |
645 | { | |
646 | return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL); | |
647 | } | |
648 | ||
649 | asmlinkage int sys_clone(struct pt_regs regs) | |
650 | { | |
651 | unsigned long clone_flags; | |
652 | unsigned long newsp; | |
653 | int __user *parent_tidptr, *child_tidptr; | |
654 | ||
655 | clone_flags = regs.ebx; | |
656 | newsp = regs.ecx; | |
657 | parent_tidptr = (int __user *)regs.edx; | |
658 | child_tidptr = (int __user *)regs.edi; | |
659 | if (!newsp) | |
660 | newsp = regs.esp; | |
661 | return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr); | |
662 | } | |
663 | ||
664 | /* | |
665 | * This is trivial, and on the face of it looks like it | |
666 | * could equally well be done in user mode. | |
667 | * | |
668 | * Not so, for quite unobvious reasons - register pressure. | |
669 | * In user mode vfork() cannot have a stack frame, and if | |
670 | * done by calling the "clone()" system call directly, you | |
671 | * do not have enough call-clobbered registers to hold all | |
672 | * the information you need. | |
673 | */ | |
674 | asmlinkage int sys_vfork(struct pt_regs regs) | |
675 | { | |
676 | return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL); | |
677 | } | |
678 | ||
679 | /* | |
680 | * sys_execve() executes a new program. | |
681 | */ | |
682 | asmlinkage int sys_execve(struct pt_regs regs) | |
683 | { | |
684 | int error; | |
685 | char * filename; | |
686 | ||
687 | filename = getname((char __user *) regs.ebx); | |
688 | error = PTR_ERR(filename); | |
689 | if (IS_ERR(filename)) | |
690 | goto out; | |
691 | error = do_execve(filename, | |
692 | (char __user * __user *) regs.ecx, | |
693 | (char __user * __user *) regs.edx, | |
694 | ®s); | |
695 | if (error == 0) { | |
696 | task_lock(current); | |
697 | current->ptrace &= ~PT_DTRACE; | |
698 | task_unlock(current); | |
699 | /* Make sure we don't return using sysenter.. */ | |
700 | set_thread_flag(TIF_IRET); | |
701 | } | |
702 | putname(filename); | |
703 | out: | |
704 | return error; | |
705 | } | |
706 | ||
707 | #define top_esp (THREAD_SIZE - sizeof(unsigned long)) | |
708 | #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) | |
709 | ||
710 | unsigned long get_wchan(struct task_struct *p) | |
711 | { | |
712 | unsigned long ebp, esp, eip; | |
713 | unsigned long stack_page; | |
714 | int count = 0; | |
715 | if (!p || p == current || p->state == TASK_RUNNING) | |
716 | return 0; | |
717 | stack_page = (unsigned long)p->thread_info; | |
718 | esp = p->thread.esp; | |
719 | if (!stack_page || esp < stack_page || esp > top_esp+stack_page) | |
720 | return 0; | |
721 | /* include/asm-i386/system.h:switch_to() pushes ebp last. */ | |
722 | ebp = *(unsigned long *) esp; | |
723 | do { | |
724 | if (ebp < stack_page || ebp > top_ebp+stack_page) | |
725 | return 0; | |
726 | eip = *(unsigned long *) (ebp+4); | |
727 | if (!in_sched_functions(eip)) | |
728 | return eip; | |
729 | ebp = *(unsigned long *) ebp; | |
730 | } while (count++ < 16); | |
731 | return 0; | |
732 | } | |
733 | ||
734 | /* | |
735 | * sys_alloc_thread_area: get a yet unused TLS descriptor index. | |
736 | */ | |
737 | static int get_free_idx(void) | |
738 | { | |
739 | struct thread_struct *t = ¤t->thread; | |
740 | int idx; | |
741 | ||
742 | for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++) | |
743 | if (desc_empty(t->tls_array + idx)) | |
744 | return idx + GDT_ENTRY_TLS_MIN; | |
745 | return -ESRCH; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Set a given TLS descriptor: | |
750 | */ | |
751 | asmlinkage int sys_set_thread_area(struct user_desc __user *u_info) | |
752 | { | |
753 | struct thread_struct *t = ¤t->thread; | |
754 | struct user_desc info; | |
755 | struct desc_struct *desc; | |
756 | int cpu, idx; | |
757 | ||
758 | if (copy_from_user(&info, u_info, sizeof(info))) | |
759 | return -EFAULT; | |
760 | idx = info.entry_number; | |
761 | ||
762 | /* | |
763 | * index -1 means the kernel should try to find and | |
764 | * allocate an empty descriptor: | |
765 | */ | |
766 | if (idx == -1) { | |
767 | idx = get_free_idx(); | |
768 | if (idx < 0) | |
769 | return idx; | |
770 | if (put_user(idx, &u_info->entry_number)) | |
771 | return -EFAULT; | |
772 | } | |
773 | ||
774 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
775 | return -EINVAL; | |
776 | ||
777 | desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN; | |
778 | ||
779 | /* | |
780 | * We must not get preempted while modifying the TLS. | |
781 | */ | |
782 | cpu = get_cpu(); | |
783 | ||
784 | if (LDT_empty(&info)) { | |
785 | desc->a = 0; | |
786 | desc->b = 0; | |
787 | } else { | |
788 | desc->a = LDT_entry_a(&info); | |
789 | desc->b = LDT_entry_b(&info); | |
790 | } | |
791 | load_TLS(t, cpu); | |
792 | ||
793 | put_cpu(); | |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
798 | /* | |
799 | * Get the current Thread-Local Storage area: | |
800 | */ | |
801 | ||
802 | #define GET_BASE(desc) ( \ | |
803 | (((desc)->a >> 16) & 0x0000ffff) | \ | |
804 | (((desc)->b << 16) & 0x00ff0000) | \ | |
805 | ( (desc)->b & 0xff000000) ) | |
806 | ||
807 | #define GET_LIMIT(desc) ( \ | |
808 | ((desc)->a & 0x0ffff) | \ | |
809 | ((desc)->b & 0xf0000) ) | |
810 | ||
811 | #define GET_32BIT(desc) (((desc)->b >> 22) & 1) | |
812 | #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3) | |
813 | #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1) | |
814 | #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1) | |
815 | #define GET_PRESENT(desc) (((desc)->b >> 15) & 1) | |
816 | #define GET_USEABLE(desc) (((desc)->b >> 20) & 1) | |
817 | ||
818 | asmlinkage int sys_get_thread_area(struct user_desc __user *u_info) | |
819 | { | |
820 | struct user_desc info; | |
821 | struct desc_struct *desc; | |
822 | int idx; | |
823 | ||
824 | if (get_user(idx, &u_info->entry_number)) | |
825 | return -EFAULT; | |
826 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
827 | return -EINVAL; | |
828 | ||
829 | desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; | |
830 | ||
831 | info.entry_number = idx; | |
832 | info.base_addr = GET_BASE(desc); | |
833 | info.limit = GET_LIMIT(desc); | |
834 | info.seg_32bit = GET_32BIT(desc); | |
835 | info.contents = GET_CONTENTS(desc); | |
836 | info.read_exec_only = !GET_WRITABLE(desc); | |
837 | info.limit_in_pages = GET_LIMIT_PAGES(desc); | |
838 | info.seg_not_present = !GET_PRESENT(desc); | |
839 | info.useable = GET_USEABLE(desc); | |
840 | ||
841 | if (copy_to_user(u_info, &info, sizeof(info))) | |
842 | return -EFAULT; | |
843 | return 0; | |
844 | } | |
845 | ||
846 | unsigned long arch_align_stack(unsigned long sp) | |
847 | { | |
848 | if (randomize_va_space) | |
849 | sp -= get_random_int() % 8192; | |
850 | return sp & ~0xf; | |
851 | } |