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 | ||
f3705136 | 16 | #include <linux/cpu.h> |
1da177e4 LT |
17 | #include <linux/errno.h> |
18 | #include <linux/sched.h> | |
19 | #include <linux/fs.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/elfcore.h> | |
23 | #include <linux/smp.h> | |
24 | #include <linux/smp_lock.h> | |
25 | #include <linux/stddef.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/vmalloc.h> | |
28 | #include <linux/user.h> | |
29 | #include <linux/a.out.h> | |
30 | #include <linux/interrupt.h> | |
1da177e4 LT |
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> | |
c16b63e0 | 40 | #include <linux/personality.h> |
1da177e4 LT |
41 | |
42 | #include <asm/uaccess.h> | |
43 | #include <asm/pgtable.h> | |
44 | #include <asm/system.h> | |
45 | #include <asm/io.h> | |
46 | #include <asm/ldt.h> | |
47 | #include <asm/processor.h> | |
48 | #include <asm/i387.h> | |
1da177e4 | 49 | #include <asm/desc.h> |
64ca9004 | 50 | #include <asm/vm86.h> |
1da177e4 LT |
51 | #ifdef CONFIG_MATH_EMULATION |
52 | #include <asm/math_emu.h> | |
53 | #endif | |
54 | ||
1da177e4 LT |
55 | #include <linux/err.h> |
56 | ||
f3705136 ZM |
57 | #include <asm/tlbflush.h> |
58 | #include <asm/cpu.h> | |
f95d47ca | 59 | #include <asm/pda.h> |
f3705136 | 60 | |
1da177e4 LT |
61 | asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); |
62 | ||
63 | static int hlt_counter; | |
64 | ||
65 | unsigned long boot_option_idle_override = 0; | |
66 | EXPORT_SYMBOL(boot_option_idle_override); | |
67 | ||
68 | /* | |
69 | * Return saved PC of a blocked thread. | |
70 | */ | |
71 | unsigned long thread_saved_pc(struct task_struct *tsk) | |
72 | { | |
73 | return ((unsigned long *)tsk->thread.esp)[3]; | |
74 | } | |
75 | ||
76 | /* | |
77 | * Powermanagement idle function, if any.. | |
78 | */ | |
79 | void (*pm_idle)(void); | |
129f6946 | 80 | EXPORT_SYMBOL(pm_idle); |
1da177e4 LT |
81 | static DEFINE_PER_CPU(unsigned int, cpu_idle_state); |
82 | ||
83 | void disable_hlt(void) | |
84 | { | |
85 | hlt_counter++; | |
86 | } | |
87 | ||
88 | EXPORT_SYMBOL(disable_hlt); | |
89 | ||
90 | void enable_hlt(void) | |
91 | { | |
92 | hlt_counter--; | |
93 | } | |
94 | ||
95 | EXPORT_SYMBOL(enable_hlt); | |
96 | ||
97 | /* | |
98 | * We use this if we don't have any better | |
99 | * idle routine.. | |
100 | */ | |
101 | void default_idle(void) | |
102 | { | |
103 | if (!hlt_counter && boot_cpu_data.hlt_works_ok) { | |
495ab9c0 | 104 | current_thread_info()->status &= ~TS_POLLING; |
64c7c8f8 | 105 | smp_mb__after_clear_bit(); |
72690a21 AK |
106 | local_irq_disable(); |
107 | if (!need_resched()) | |
108 | safe_halt(); /* enables interrupts racelessly */ | |
109 | else | |
110 | local_irq_enable(); | |
495ab9c0 | 111 | current_thread_info()->status |= TS_POLLING; |
1da177e4 | 112 | } else { |
72690a21 AK |
113 | /* loop is done by the caller */ |
114 | cpu_relax(); | |
1da177e4 LT |
115 | } |
116 | } | |
129f6946 AD |
117 | #ifdef CONFIG_APM_MODULE |
118 | EXPORT_SYMBOL(default_idle); | |
119 | #endif | |
1da177e4 LT |
120 | |
121 | /* | |
122 | * On SMP it's slightly faster (but much more power-consuming!) | |
123 | * to poll the ->work.need_resched flag instead of waiting for the | |
124 | * cross-CPU IPI to arrive. Use this option with caution. | |
125 | */ | |
126 | static void poll_idle (void) | |
127 | { | |
72690a21 | 128 | cpu_relax(); |
1da177e4 LT |
129 | } |
130 | ||
f3705136 ZM |
131 | #ifdef CONFIG_HOTPLUG_CPU |
132 | #include <asm/nmi.h> | |
133 | /* We don't actually take CPU down, just spin without interrupts. */ | |
134 | static inline void play_dead(void) | |
135 | { | |
e1367daf LS |
136 | /* This must be done before dead CPU ack */ |
137 | cpu_exit_clear(); | |
138 | wbinvd(); | |
139 | mb(); | |
f3705136 ZM |
140 | /* Ack it */ |
141 | __get_cpu_var(cpu_state) = CPU_DEAD; | |
142 | ||
e1367daf LS |
143 | /* |
144 | * With physical CPU hotplug, we should halt the cpu | |
145 | */ | |
f3705136 | 146 | local_irq_disable(); |
e1367daf | 147 | while (1) |
f2ab4461 | 148 | halt(); |
f3705136 ZM |
149 | } |
150 | #else | |
151 | static inline void play_dead(void) | |
152 | { | |
153 | BUG(); | |
154 | } | |
155 | #endif /* CONFIG_HOTPLUG_CPU */ | |
156 | ||
1da177e4 LT |
157 | /* |
158 | * The idle thread. There's no useful work to be | |
159 | * done, so just try to conserve power and have a | |
160 | * low exit latency (ie sit in a loop waiting for | |
161 | * somebody to say that they'd like to reschedule) | |
162 | */ | |
f3705136 | 163 | void cpu_idle(void) |
1da177e4 | 164 | { |
5bfb5d69 | 165 | int cpu = smp_processor_id(); |
f3705136 | 166 | |
495ab9c0 | 167 | current_thread_info()->status |= TS_POLLING; |
64c7c8f8 | 168 | |
1da177e4 LT |
169 | /* endless idle loop with no priority at all */ |
170 | while (1) { | |
171 | while (!need_resched()) { | |
172 | void (*idle)(void); | |
173 | ||
174 | if (__get_cpu_var(cpu_idle_state)) | |
175 | __get_cpu_var(cpu_idle_state) = 0; | |
176 | ||
177 | rmb(); | |
178 | idle = pm_idle; | |
179 | ||
180 | if (!idle) | |
181 | idle = default_idle; | |
182 | ||
f3705136 ZM |
183 | if (cpu_is_offline(cpu)) |
184 | play_dead(); | |
185 | ||
1da177e4 LT |
186 | __get_cpu_var(irq_stat).idle_timestamp = jiffies; |
187 | idle(); | |
188 | } | |
5bfb5d69 | 189 | preempt_enable_no_resched(); |
1da177e4 | 190 | schedule(); |
5bfb5d69 | 191 | preempt_disable(); |
1da177e4 LT |
192 | } |
193 | } | |
194 | ||
195 | void cpu_idle_wait(void) | |
196 | { | |
197 | unsigned int cpu, this_cpu = get_cpu(); | |
dc1829a4 | 198 | cpumask_t map, tmp = current->cpus_allowed; |
1da177e4 LT |
199 | |
200 | set_cpus_allowed(current, cpumask_of_cpu(this_cpu)); | |
201 | put_cpu(); | |
202 | ||
203 | cpus_clear(map); | |
204 | for_each_online_cpu(cpu) { | |
205 | per_cpu(cpu_idle_state, cpu) = 1; | |
206 | cpu_set(cpu, map); | |
207 | } | |
208 | ||
209 | __get_cpu_var(cpu_idle_state) = 0; | |
210 | ||
211 | wmb(); | |
212 | do { | |
213 | ssleep(1); | |
214 | for_each_online_cpu(cpu) { | |
215 | if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu)) | |
216 | cpu_clear(cpu, map); | |
217 | } | |
218 | cpus_and(map, map, cpu_online_map); | |
219 | } while (!cpus_empty(map)); | |
dc1829a4 IM |
220 | |
221 | set_cpus_allowed(current, tmp); | |
1da177e4 LT |
222 | } |
223 | EXPORT_SYMBOL_GPL(cpu_idle_wait); | |
224 | ||
225 | /* | |
226 | * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, | |
227 | * which can obviate IPI to trigger checking of need_resched. | |
228 | * We execute MONITOR against need_resched and enter optimized wait state | |
229 | * through MWAIT. Whenever someone changes need_resched, we would be woken | |
230 | * up from MWAIT (without an IPI). | |
991528d7 VP |
231 | * |
232 | * New with Core Duo processors, MWAIT can take some hints based on CPU | |
233 | * capability. | |
1da177e4 | 234 | */ |
991528d7 | 235 | void mwait_idle_with_hints(unsigned long eax, unsigned long ecx) |
1da177e4 | 236 | { |
991528d7 | 237 | if (!need_resched()) { |
64c7c8f8 NP |
238 | __monitor((void *)¤t_thread_info()->flags, 0, 0); |
239 | smp_mb(); | |
991528d7 VP |
240 | if (!need_resched()) |
241 | __mwait(eax, ecx); | |
1da177e4 LT |
242 | } |
243 | } | |
244 | ||
991528d7 VP |
245 | /* Default MONITOR/MWAIT with no hints, used for default C1 state */ |
246 | static void mwait_idle(void) | |
247 | { | |
248 | local_irq_enable(); | |
72690a21 | 249 | mwait_idle_with_hints(0, 0); |
991528d7 VP |
250 | } |
251 | ||
0bb3184d | 252 | void __devinit select_idle_routine(const struct cpuinfo_x86 *c) |
1da177e4 LT |
253 | { |
254 | if (cpu_has(c, X86_FEATURE_MWAIT)) { | |
255 | printk("monitor/mwait feature present.\n"); | |
256 | /* | |
257 | * Skip, if setup has overridden idle. | |
258 | * One CPU supports mwait => All CPUs supports mwait | |
259 | */ | |
260 | if (!pm_idle) { | |
261 | printk("using mwait in idle threads.\n"); | |
262 | pm_idle = mwait_idle; | |
263 | } | |
264 | } | |
265 | } | |
266 | ||
267 | static int __init idle_setup (char *str) | |
268 | { | |
269 | if (!strncmp(str, "poll", 4)) { | |
270 | printk("using polling idle threads.\n"); | |
271 | pm_idle = poll_idle; | |
272 | #ifdef CONFIG_X86_SMP | |
273 | if (smp_num_siblings > 1) | |
274 | printk("WARNING: polling idle and HT enabled, performance may degrade.\n"); | |
275 | #endif | |
276 | } else if (!strncmp(str, "halt", 4)) { | |
277 | printk("using halt in idle threads.\n"); | |
278 | pm_idle = default_idle; | |
279 | } | |
280 | ||
281 | boot_option_idle_override = 1; | |
282 | return 1; | |
283 | } | |
284 | ||
285 | __setup("idle=", idle_setup); | |
286 | ||
287 | void show_regs(struct pt_regs * regs) | |
288 | { | |
289 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; | |
290 | ||
291 | printk("\n"); | |
292 | printk("Pid: %d, comm: %20s\n", current->pid, current->comm); | |
293 | printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id()); | |
294 | print_symbol("EIP is at %s\n", regs->eip); | |
295 | ||
db753bdf | 296 | if (user_mode_vm(regs)) |
1da177e4 | 297 | printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp); |
b53e8f68 | 298 | printk(" EFLAGS: %08lx %s (%s %.*s)\n", |
96b644bd SH |
299 | regs->eflags, print_tainted(), init_utsname()->release, |
300 | (int)strcspn(init_utsname()->version, " "), | |
301 | init_utsname()->version); | |
1da177e4 LT |
302 | printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", |
303 | regs->eax,regs->ebx,regs->ecx,regs->edx); | |
304 | printk("ESI: %08lx EDI: %08lx EBP: %08lx", | |
305 | regs->esi, regs->edi, regs->ebp); | |
66e10a44 JF |
306 | printk(" DS: %04x ES: %04x GS: %04x\n", |
307 | 0xffff & regs->xds,0xffff & regs->xes, 0xffff & regs->xgs); | |
1da177e4 | 308 | |
4bb0d3ec ZA |
309 | cr0 = read_cr0(); |
310 | cr2 = read_cr2(); | |
311 | cr3 = read_cr3(); | |
ff6e8c0d | 312 | cr4 = read_cr4_safe(); |
1da177e4 | 313 | printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4); |
176a2718 | 314 | show_trace(NULL, regs, ®s->esp); |
1da177e4 LT |
315 | } |
316 | ||
317 | /* | |
318 | * This gets run with %ebx containing the | |
319 | * function to call, and %edx containing | |
320 | * the "args". | |
321 | */ | |
322 | extern void kernel_thread_helper(void); | |
1da177e4 LT |
323 | |
324 | /* | |
325 | * Create a kernel thread | |
326 | */ | |
327 | int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags) | |
328 | { | |
329 | struct pt_regs regs; | |
330 | ||
331 | memset(®s, 0, sizeof(regs)); | |
332 | ||
333 | regs.ebx = (unsigned long) fn; | |
334 | regs.edx = (unsigned long) arg; | |
335 | ||
336 | regs.xds = __USER_DS; | |
337 | regs.xes = __USER_DS; | |
f95d47ca | 338 | regs.xgs = __KERNEL_PDA; |
1da177e4 LT |
339 | regs.orig_eax = -1; |
340 | regs.eip = (unsigned long) kernel_thread_helper; | |
78be3706 | 341 | regs.xcs = __KERNEL_CS | get_kernel_rpl(); |
1da177e4 LT |
342 | regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2; |
343 | ||
344 | /* Ok, create the new process.. */ | |
8cf2c519 | 345 | return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL); |
1da177e4 | 346 | } |
129f6946 | 347 | EXPORT_SYMBOL(kernel_thread); |
1da177e4 LT |
348 | |
349 | /* | |
350 | * Free current thread data structures etc.. | |
351 | */ | |
352 | void exit_thread(void) | |
353 | { | |
1da177e4 | 354 | /* The process may have allocated an io port bitmap... nuke it. */ |
b3cf2576 SE |
355 | if (unlikely(test_thread_flag(TIF_IO_BITMAP))) { |
356 | struct task_struct *tsk = current; | |
357 | struct thread_struct *t = &tsk->thread; | |
1da177e4 LT |
358 | int cpu = get_cpu(); |
359 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
360 | ||
361 | kfree(t->io_bitmap_ptr); | |
362 | t->io_bitmap_ptr = NULL; | |
b3cf2576 | 363 | clear_thread_flag(TIF_IO_BITMAP); |
1da177e4 LT |
364 | /* |
365 | * Careful, clear this in the TSS too: | |
366 | */ | |
367 | memset(tss->io_bitmap, 0xff, tss->io_bitmap_max); | |
368 | t->io_bitmap_max = 0; | |
369 | tss->io_bitmap_owner = NULL; | |
370 | tss->io_bitmap_max = 0; | |
371 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET; | |
372 | put_cpu(); | |
373 | } | |
374 | } | |
375 | ||
376 | void flush_thread(void) | |
377 | { | |
378 | struct task_struct *tsk = current; | |
379 | ||
380 | memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8); | |
381 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); | |
b3cf2576 | 382 | clear_tsk_thread_flag(tsk, TIF_DEBUG); |
1da177e4 LT |
383 | /* |
384 | * Forget coprocessor state.. | |
385 | */ | |
386 | clear_fpu(tsk); | |
387 | clear_used_math(); | |
388 | } | |
389 | ||
390 | void release_thread(struct task_struct *dead_task) | |
391 | { | |
2684927c | 392 | BUG_ON(dead_task->mm); |
1da177e4 LT |
393 | release_vm86_irqs(dead_task); |
394 | } | |
395 | ||
396 | /* | |
397 | * This gets called before we allocate a new thread and copy | |
398 | * the current task into it. | |
399 | */ | |
400 | void prepare_to_copy(struct task_struct *tsk) | |
401 | { | |
402 | unlazy_fpu(tsk); | |
403 | } | |
404 | ||
405 | int copy_thread(int nr, unsigned long clone_flags, unsigned long esp, | |
406 | unsigned long unused, | |
407 | struct task_struct * p, struct pt_regs * regs) | |
408 | { | |
409 | struct pt_regs * childregs; | |
410 | struct task_struct *tsk; | |
411 | int err; | |
412 | ||
07b047fc | 413 | childregs = task_pt_regs(p); |
f48d9663 AN |
414 | *childregs = *regs; |
415 | childregs->eax = 0; | |
416 | childregs->esp = esp; | |
417 | ||
418 | p->thread.esp = (unsigned long) childregs; | |
419 | p->thread.esp0 = (unsigned long) (childregs+1); | |
1da177e4 LT |
420 | |
421 | p->thread.eip = (unsigned long) ret_from_fork; | |
422 | ||
423 | savesegment(fs,p->thread.fs); | |
1da177e4 LT |
424 | |
425 | tsk = current; | |
b3cf2576 | 426 | if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { |
52978be6 AD |
427 | p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, |
428 | IO_BITMAP_BYTES, GFP_KERNEL); | |
1da177e4 LT |
429 | if (!p->thread.io_bitmap_ptr) { |
430 | p->thread.io_bitmap_max = 0; | |
431 | return -ENOMEM; | |
432 | } | |
b3cf2576 | 433 | set_tsk_thread_flag(p, TIF_IO_BITMAP); |
1da177e4 LT |
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); | |
66e10a44 | 500 | dump->regs.gs = regs->xgs; |
1da177e4 LT |
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 | } | |
129f6946 | 510 | EXPORT_SYMBOL(dump_thread); |
1da177e4 LT |
511 | |
512 | /* | |
513 | * Capture the user space registers if the task is not running (in user space) | |
514 | */ | |
515 | int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs) | |
516 | { | |
07b047fc | 517 | struct pt_regs ptregs = *task_pt_regs(tsk); |
1da177e4 LT |
518 | ptregs.xcs &= 0xffff; |
519 | ptregs.xds &= 0xffff; | |
520 | ptregs.xes &= 0xffff; | |
521 | ptregs.xss &= 0xffff; | |
522 | ||
523 | elf_core_copy_regs(regs, &ptregs); | |
524 | ||
525 | return 1; | |
526 | } | |
527 | ||
b3cf2576 SE |
528 | static noinline void __switch_to_xtra(struct task_struct *next_p, |
529 | struct tss_struct *tss) | |
1da177e4 | 530 | { |
b3cf2576 SE |
531 | struct thread_struct *next; |
532 | ||
533 | next = &next_p->thread; | |
534 | ||
535 | if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { | |
536 | set_debugreg(next->debugreg[0], 0); | |
537 | set_debugreg(next->debugreg[1], 1); | |
538 | set_debugreg(next->debugreg[2], 2); | |
539 | set_debugreg(next->debugreg[3], 3); | |
540 | /* no 4 and 5 */ | |
541 | set_debugreg(next->debugreg[6], 6); | |
542 | set_debugreg(next->debugreg[7], 7); | |
543 | } | |
544 | ||
545 | if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { | |
1da177e4 LT |
546 | /* |
547 | * Disable the bitmap via an invalid offset. We still cache | |
548 | * the previous bitmap owner and the IO bitmap contents: | |
549 | */ | |
550 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET; | |
551 | return; | |
552 | } | |
b3cf2576 | 553 | |
1da177e4 LT |
554 | if (likely(next == tss->io_bitmap_owner)) { |
555 | /* | |
556 | * Previous owner of the bitmap (hence the bitmap content) | |
557 | * matches the next task, we dont have to do anything but | |
558 | * to set a valid offset in the TSS: | |
559 | */ | |
560 | tss->io_bitmap_base = IO_BITMAP_OFFSET; | |
561 | return; | |
562 | } | |
563 | /* | |
564 | * Lazy TSS's I/O bitmap copy. We set an invalid offset here | |
565 | * and we let the task to get a GPF in case an I/O instruction | |
566 | * is performed. The handler of the GPF will verify that the | |
567 | * faulting task has a valid I/O bitmap and, it true, does the | |
568 | * real copy and restart the instruction. This will save us | |
569 | * redundant copies when the currently switched task does not | |
570 | * perform any I/O during its timeslice. | |
571 | */ | |
572 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY; | |
573 | } | |
1da177e4 | 574 | |
ffaa8bd6 AA |
575 | /* |
576 | * This function selects if the context switch from prev to next | |
577 | * has to tweak the TSC disable bit in the cr4. | |
578 | */ | |
579 | static inline void disable_tsc(struct task_struct *prev_p, | |
580 | struct task_struct *next_p) | |
581 | { | |
582 | struct thread_info *prev, *next; | |
583 | ||
584 | /* | |
585 | * gcc should eliminate the ->thread_info dereference if | |
586 | * has_secure_computing returns 0 at compile time (SECCOMP=n). | |
587 | */ | |
06b425d8 AV |
588 | prev = task_thread_info(prev_p); |
589 | next = task_thread_info(next_p); | |
ffaa8bd6 AA |
590 | |
591 | if (has_secure_computing(prev) || has_secure_computing(next)) { | |
592 | /* slow path here */ | |
593 | if (has_secure_computing(prev) && | |
594 | !has_secure_computing(next)) { | |
595 | write_cr4(read_cr4() & ~X86_CR4_TSD); | |
596 | } else if (!has_secure_computing(prev) && | |
597 | has_secure_computing(next)) | |
598 | write_cr4(read_cr4() | X86_CR4_TSD); | |
599 | } | |
600 | } | |
601 | ||
1da177e4 LT |
602 | /* |
603 | * switch_to(x,yn) should switch tasks from x to y. | |
604 | * | |
605 | * We fsave/fwait so that an exception goes off at the right time | |
606 | * (as a call from the fsave or fwait in effect) rather than to | |
607 | * the wrong process. Lazy FP saving no longer makes any sense | |
608 | * with modern CPU's, and this simplifies a lot of things (SMP | |
609 | * and UP become the same). | |
610 | * | |
611 | * NOTE! We used to use the x86 hardware context switching. The | |
612 | * reason for not using it any more becomes apparent when you | |
613 | * try to recover gracefully from saved state that is no longer | |
614 | * valid (stale segment register values in particular). With the | |
615 | * hardware task-switch, there is no way to fix up bad state in | |
616 | * a reasonable manner. | |
617 | * | |
618 | * The fact that Intel documents the hardware task-switching to | |
619 | * be slow is a fairly red herring - this code is not noticeably | |
620 | * faster. However, there _is_ some room for improvement here, | |
621 | * so the performance issues may eventually be a valid point. | |
622 | * More important, however, is the fact that this allows us much | |
623 | * more flexibility. | |
624 | * | |
625 | * The return value (in %eax) will be the "prev" task after | |
626 | * the task-switch, and shows up in ret_from_fork in entry.S, | |
627 | * for example. | |
628 | */ | |
629 | struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p) | |
630 | { | |
631 | struct thread_struct *prev = &prev_p->thread, | |
632 | *next = &next_p->thread; | |
633 | int cpu = smp_processor_id(); | |
634 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
635 | ||
636 | /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ | |
637 | ||
638 | __unlazy_fpu(prev_p); | |
639 | ||
acc20761 CE |
640 | |
641 | /* we're going to use this soon, after a few expensive things */ | |
642 | if (next_p->fpu_counter > 5) | |
643 | prefetch(&next->i387.fxsave); | |
644 | ||
1da177e4 | 645 | /* |
e7a2ff59 | 646 | * Reload esp0. |
1da177e4 LT |
647 | */ |
648 | load_esp0(tss, next); | |
649 | ||
650 | /* | |
f95d47ca JF |
651 | * Save away %fs. No need to save %gs, as it was saved on the |
652 | * stack on entry. No need to save %es and %ds, as those are | |
653 | * always kernel segments while inside the kernel. Doing this | |
654 | * before setting the new TLS descriptors avoids the situation | |
655 | * where we temporarily have non-reloadable segments in %fs | |
656 | * and %gs. This could be an issue if the NMI handler ever | |
657 | * used %fs or %gs (it does not today), or if the kernel is | |
658 | * running inside of a hypervisor layer. | |
1da177e4 | 659 | */ |
e7a2ff59 | 660 | savesegment(fs, prev->fs); |
1da177e4 LT |
661 | |
662 | /* | |
e7a2ff59 | 663 | * Load the per-thread Thread-Local Storage descriptor. |
1da177e4 | 664 | */ |
e7a2ff59 | 665 | load_TLS(next, cpu); |
1da177e4 LT |
666 | |
667 | /* | |
f95d47ca | 668 | * Restore %fs if needed. |
b339a18b | 669 | * |
f95d47ca | 670 | * Glibc normally makes %fs be zero. |
1da177e4 | 671 | */ |
b339a18b | 672 | if (unlikely(prev->fs | next->fs)) |
1da177e4 | 673 | loadsegment(fs, next->fs); |
b339a18b | 674 | |
ec7fcaab | 675 | write_pda(pcurrent, next_p); |
1da177e4 LT |
676 | |
677 | /* | |
b3cf2576 | 678 | * Now maybe handle debug registers and/or IO bitmaps |
1da177e4 | 679 | */ |
facf0147 CE |
680 | if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW) |
681 | || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))) | |
b3cf2576 | 682 | __switch_to_xtra(next_p, tss); |
1da177e4 | 683 | |
ffaa8bd6 AA |
684 | disable_tsc(prev_p, next_p); |
685 | ||
acc20761 CE |
686 | /* If the task has used fpu the last 5 timeslices, just do a full |
687 | * restore of the math state immediately to avoid the trap; the | |
688 | * chances of needing FPU soon are obviously high now | |
689 | */ | |
690 | if (next_p->fpu_counter > 5) | |
691 | math_state_restore(); | |
692 | ||
1da177e4 LT |
693 | return prev_p; |
694 | } | |
695 | ||
696 | asmlinkage int sys_fork(struct pt_regs regs) | |
697 | { | |
698 | return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL); | |
699 | } | |
700 | ||
701 | asmlinkage int sys_clone(struct pt_regs regs) | |
702 | { | |
703 | unsigned long clone_flags; | |
704 | unsigned long newsp; | |
705 | int __user *parent_tidptr, *child_tidptr; | |
706 | ||
707 | clone_flags = regs.ebx; | |
708 | newsp = regs.ecx; | |
709 | parent_tidptr = (int __user *)regs.edx; | |
710 | child_tidptr = (int __user *)regs.edi; | |
711 | if (!newsp) | |
712 | newsp = regs.esp; | |
713 | return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr); | |
714 | } | |
715 | ||
716 | /* | |
717 | * This is trivial, and on the face of it looks like it | |
718 | * could equally well be done in user mode. | |
719 | * | |
720 | * Not so, for quite unobvious reasons - register pressure. | |
721 | * In user mode vfork() cannot have a stack frame, and if | |
722 | * done by calling the "clone()" system call directly, you | |
723 | * do not have enough call-clobbered registers to hold all | |
724 | * the information you need. | |
725 | */ | |
726 | asmlinkage int sys_vfork(struct pt_regs regs) | |
727 | { | |
728 | return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL); | |
729 | } | |
730 | ||
731 | /* | |
732 | * sys_execve() executes a new program. | |
733 | */ | |
734 | asmlinkage int sys_execve(struct pt_regs regs) | |
735 | { | |
736 | int error; | |
737 | char * filename; | |
738 | ||
739 | filename = getname((char __user *) regs.ebx); | |
740 | error = PTR_ERR(filename); | |
741 | if (IS_ERR(filename)) | |
742 | goto out; | |
743 | error = do_execve(filename, | |
744 | (char __user * __user *) regs.ecx, | |
745 | (char __user * __user *) regs.edx, | |
746 | ®s); | |
747 | if (error == 0) { | |
748 | task_lock(current); | |
749 | current->ptrace &= ~PT_DTRACE; | |
750 | task_unlock(current); | |
751 | /* Make sure we don't return using sysenter.. */ | |
752 | set_thread_flag(TIF_IRET); | |
753 | } | |
754 | putname(filename); | |
755 | out: | |
756 | return error; | |
757 | } | |
758 | ||
759 | #define top_esp (THREAD_SIZE - sizeof(unsigned long)) | |
760 | #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) | |
761 | ||
762 | unsigned long get_wchan(struct task_struct *p) | |
763 | { | |
764 | unsigned long ebp, esp, eip; | |
765 | unsigned long stack_page; | |
766 | int count = 0; | |
767 | if (!p || p == current || p->state == TASK_RUNNING) | |
768 | return 0; | |
65e0fdff | 769 | stack_page = (unsigned long)task_stack_page(p); |
1da177e4 LT |
770 | esp = p->thread.esp; |
771 | if (!stack_page || esp < stack_page || esp > top_esp+stack_page) | |
772 | return 0; | |
773 | /* include/asm-i386/system.h:switch_to() pushes ebp last. */ | |
774 | ebp = *(unsigned long *) esp; | |
775 | do { | |
776 | if (ebp < stack_page || ebp > top_ebp+stack_page) | |
777 | return 0; | |
778 | eip = *(unsigned long *) (ebp+4); | |
779 | if (!in_sched_functions(eip)) | |
780 | return eip; | |
781 | ebp = *(unsigned long *) ebp; | |
782 | } while (count++ < 16); | |
783 | return 0; | |
784 | } | |
785 | ||
786 | /* | |
787 | * sys_alloc_thread_area: get a yet unused TLS descriptor index. | |
788 | */ | |
789 | static int get_free_idx(void) | |
790 | { | |
791 | struct thread_struct *t = ¤t->thread; | |
792 | int idx; | |
793 | ||
794 | for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++) | |
795 | if (desc_empty(t->tls_array + idx)) | |
796 | return idx + GDT_ENTRY_TLS_MIN; | |
797 | return -ESRCH; | |
798 | } | |
799 | ||
800 | /* | |
801 | * Set a given TLS descriptor: | |
802 | */ | |
803 | asmlinkage int sys_set_thread_area(struct user_desc __user *u_info) | |
804 | { | |
805 | struct thread_struct *t = ¤t->thread; | |
806 | struct user_desc info; | |
807 | struct desc_struct *desc; | |
808 | int cpu, idx; | |
809 | ||
810 | if (copy_from_user(&info, u_info, sizeof(info))) | |
811 | return -EFAULT; | |
812 | idx = info.entry_number; | |
813 | ||
814 | /* | |
815 | * index -1 means the kernel should try to find and | |
816 | * allocate an empty descriptor: | |
817 | */ | |
818 | if (idx == -1) { | |
819 | idx = get_free_idx(); | |
820 | if (idx < 0) | |
821 | return idx; | |
822 | if (put_user(idx, &u_info->entry_number)) | |
823 | return -EFAULT; | |
824 | } | |
825 | ||
826 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
827 | return -EINVAL; | |
828 | ||
829 | desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN; | |
830 | ||
831 | /* | |
832 | * We must not get preempted while modifying the TLS. | |
833 | */ | |
834 | cpu = get_cpu(); | |
835 | ||
836 | if (LDT_empty(&info)) { | |
837 | desc->a = 0; | |
838 | desc->b = 0; | |
839 | } else { | |
840 | desc->a = LDT_entry_a(&info); | |
841 | desc->b = LDT_entry_b(&info); | |
842 | } | |
843 | load_TLS(t, cpu); | |
844 | ||
845 | put_cpu(); | |
846 | ||
847 | return 0; | |
848 | } | |
849 | ||
850 | /* | |
851 | * Get the current Thread-Local Storage area: | |
852 | */ | |
853 | ||
854 | #define GET_BASE(desc) ( \ | |
855 | (((desc)->a >> 16) & 0x0000ffff) | \ | |
856 | (((desc)->b << 16) & 0x00ff0000) | \ | |
857 | ( (desc)->b & 0xff000000) ) | |
858 | ||
859 | #define GET_LIMIT(desc) ( \ | |
860 | ((desc)->a & 0x0ffff) | \ | |
861 | ((desc)->b & 0xf0000) ) | |
862 | ||
863 | #define GET_32BIT(desc) (((desc)->b >> 22) & 1) | |
864 | #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3) | |
865 | #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1) | |
866 | #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1) | |
867 | #define GET_PRESENT(desc) (((desc)->b >> 15) & 1) | |
868 | #define GET_USEABLE(desc) (((desc)->b >> 20) & 1) | |
869 | ||
870 | asmlinkage int sys_get_thread_area(struct user_desc __user *u_info) | |
871 | { | |
872 | struct user_desc info; | |
873 | struct desc_struct *desc; | |
874 | int idx; | |
875 | ||
876 | if (get_user(idx, &u_info->entry_number)) | |
877 | return -EFAULT; | |
878 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
879 | return -EINVAL; | |
880 | ||
71ae18ec B |
881 | memset(&info, 0, sizeof(info)); |
882 | ||
1da177e4 LT |
883 | desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; |
884 | ||
885 | info.entry_number = idx; | |
886 | info.base_addr = GET_BASE(desc); | |
887 | info.limit = GET_LIMIT(desc); | |
888 | info.seg_32bit = GET_32BIT(desc); | |
889 | info.contents = GET_CONTENTS(desc); | |
890 | info.read_exec_only = !GET_WRITABLE(desc); | |
891 | info.limit_in_pages = GET_LIMIT_PAGES(desc); | |
892 | info.seg_not_present = !GET_PRESENT(desc); | |
893 | info.useable = GET_USEABLE(desc); | |
894 | ||
895 | if (copy_to_user(u_info, &info, sizeof(info))) | |
896 | return -EFAULT; | |
897 | return 0; | |
898 | } | |
899 | ||
900 | unsigned long arch_align_stack(unsigned long sp) | |
901 | { | |
c16b63e0 | 902 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
1da177e4 LT |
903 | sp -= get_random_int() % 8192; |
904 | return sp & ~0xf; | |
905 | } |