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); | |
129f6946 | 76 | EXPORT_SYMBOL(pm_idle); |
1da177e4 LT |
77 | static DEFINE_PER_CPU(unsigned int, cpu_idle_state); |
78 | ||
79 | void disable_hlt(void) | |
80 | { | |
81 | hlt_counter++; | |
82 | } | |
83 | ||
84 | EXPORT_SYMBOL(disable_hlt); | |
85 | ||
86 | void enable_hlt(void) | |
87 | { | |
88 | hlt_counter--; | |
89 | } | |
90 | ||
91 | EXPORT_SYMBOL(enable_hlt); | |
92 | ||
93 | /* | |
94 | * We use this if we don't have any better | |
95 | * idle routine.. | |
96 | */ | |
97 | void default_idle(void) | |
98 | { | |
99 | if (!hlt_counter && boot_cpu_data.hlt_works_ok) { | |
100 | local_irq_disable(); | |
101 | if (!need_resched()) | |
102 | safe_halt(); | |
103 | else | |
104 | local_irq_enable(); | |
105 | } else { | |
106 | cpu_relax(); | |
107 | } | |
108 | } | |
129f6946 AD |
109 | #ifdef CONFIG_APM_MODULE |
110 | EXPORT_SYMBOL(default_idle); | |
111 | #endif | |
1da177e4 LT |
112 | |
113 | /* | |
114 | * On SMP it's slightly faster (but much more power-consuming!) | |
115 | * to poll the ->work.need_resched flag instead of waiting for the | |
116 | * cross-CPU IPI to arrive. Use this option with caution. | |
117 | */ | |
118 | static void poll_idle (void) | |
119 | { | |
120 | int oldval; | |
121 | ||
122 | local_irq_enable(); | |
123 | ||
124 | /* | |
125 | * Deal with another CPU just having chosen a thread to | |
126 | * run here: | |
127 | */ | |
128 | oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED); | |
129 | ||
130 | if (!oldval) { | |
131 | set_thread_flag(TIF_POLLING_NRFLAG); | |
132 | asm volatile( | |
133 | "2:" | |
134 | "testl %0, %1;" | |
135 | "rep; nop;" | |
136 | "je 2b;" | |
137 | : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags)); | |
138 | ||
139 | clear_thread_flag(TIF_POLLING_NRFLAG); | |
140 | } else { | |
141 | set_need_resched(); | |
142 | } | |
143 | } | |
144 | ||
145 | /* | |
146 | * The idle thread. There's no useful work to be | |
147 | * done, so just try to conserve power and have a | |
148 | * low exit latency (ie sit in a loop waiting for | |
149 | * somebody to say that they'd like to reschedule) | |
150 | */ | |
151 | void cpu_idle (void) | |
152 | { | |
153 | /* endless idle loop with no priority at all */ | |
154 | while (1) { | |
155 | while (!need_resched()) { | |
156 | void (*idle)(void); | |
157 | ||
158 | if (__get_cpu_var(cpu_idle_state)) | |
159 | __get_cpu_var(cpu_idle_state) = 0; | |
160 | ||
161 | rmb(); | |
162 | idle = pm_idle; | |
163 | ||
164 | if (!idle) | |
165 | idle = default_idle; | |
166 | ||
167 | __get_cpu_var(irq_stat).idle_timestamp = jiffies; | |
168 | idle(); | |
169 | } | |
170 | schedule(); | |
171 | } | |
172 | } | |
173 | ||
174 | void cpu_idle_wait(void) | |
175 | { | |
176 | unsigned int cpu, this_cpu = get_cpu(); | |
177 | cpumask_t map; | |
178 | ||
179 | set_cpus_allowed(current, cpumask_of_cpu(this_cpu)); | |
180 | put_cpu(); | |
181 | ||
182 | cpus_clear(map); | |
183 | for_each_online_cpu(cpu) { | |
184 | per_cpu(cpu_idle_state, cpu) = 1; | |
185 | cpu_set(cpu, map); | |
186 | } | |
187 | ||
188 | __get_cpu_var(cpu_idle_state) = 0; | |
189 | ||
190 | wmb(); | |
191 | do { | |
192 | ssleep(1); | |
193 | for_each_online_cpu(cpu) { | |
194 | if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu)) | |
195 | cpu_clear(cpu, map); | |
196 | } | |
197 | cpus_and(map, map, cpu_online_map); | |
198 | } while (!cpus_empty(map)); | |
199 | } | |
200 | EXPORT_SYMBOL_GPL(cpu_idle_wait); | |
201 | ||
202 | /* | |
203 | * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, | |
204 | * which can obviate IPI to trigger checking of need_resched. | |
205 | * We execute MONITOR against need_resched and enter optimized wait state | |
206 | * through MWAIT. Whenever someone changes need_resched, we would be woken | |
207 | * up from MWAIT (without an IPI). | |
208 | */ | |
209 | static void mwait_idle(void) | |
210 | { | |
211 | local_irq_enable(); | |
212 | ||
213 | if (!need_resched()) { | |
214 | set_thread_flag(TIF_POLLING_NRFLAG); | |
215 | do { | |
216 | __monitor((void *)¤t_thread_info()->flags, 0, 0); | |
217 | if (need_resched()) | |
218 | break; | |
219 | __mwait(0, 0); | |
220 | } while (!need_resched()); | |
221 | clear_thread_flag(TIF_POLLING_NRFLAG); | |
222 | } | |
223 | } | |
224 | ||
225 | void __init select_idle_routine(const struct cpuinfo_x86 *c) | |
226 | { | |
227 | if (cpu_has(c, X86_FEATURE_MWAIT)) { | |
228 | printk("monitor/mwait feature present.\n"); | |
229 | /* | |
230 | * Skip, if setup has overridden idle. | |
231 | * One CPU supports mwait => All CPUs supports mwait | |
232 | */ | |
233 | if (!pm_idle) { | |
234 | printk("using mwait in idle threads.\n"); | |
235 | pm_idle = mwait_idle; | |
236 | } | |
237 | } | |
238 | } | |
239 | ||
240 | static int __init idle_setup (char *str) | |
241 | { | |
242 | if (!strncmp(str, "poll", 4)) { | |
243 | printk("using polling idle threads.\n"); | |
244 | pm_idle = poll_idle; | |
245 | #ifdef CONFIG_X86_SMP | |
246 | if (smp_num_siblings > 1) | |
247 | printk("WARNING: polling idle and HT enabled, performance may degrade.\n"); | |
248 | #endif | |
249 | } else if (!strncmp(str, "halt", 4)) { | |
250 | printk("using halt in idle threads.\n"); | |
251 | pm_idle = default_idle; | |
252 | } | |
253 | ||
254 | boot_option_idle_override = 1; | |
255 | return 1; | |
256 | } | |
257 | ||
258 | __setup("idle=", idle_setup); | |
259 | ||
260 | void show_regs(struct pt_regs * regs) | |
261 | { | |
262 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; | |
263 | ||
264 | printk("\n"); | |
265 | printk("Pid: %d, comm: %20s\n", current->pid, current->comm); | |
266 | printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id()); | |
267 | print_symbol("EIP is at %s\n", regs->eip); | |
268 | ||
269 | if (regs->xcs & 3) | |
270 | printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp); | |
271 | printk(" EFLAGS: %08lx %s (%s)\n", | |
272 | regs->eflags, print_tainted(), system_utsname.release); | |
273 | printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", | |
274 | regs->eax,regs->ebx,regs->ecx,regs->edx); | |
275 | printk("ESI: %08lx EDI: %08lx EBP: %08lx", | |
276 | regs->esi, regs->edi, regs->ebp); | |
277 | printk(" DS: %04x ES: %04x\n", | |
278 | 0xffff & regs->xds,0xffff & regs->xes); | |
279 | ||
280 | __asm__("movl %%cr0, %0": "=r" (cr0)); | |
281 | __asm__("movl %%cr2, %0": "=r" (cr2)); | |
282 | __asm__("movl %%cr3, %0": "=r" (cr3)); | |
283 | /* This could fault if %cr4 does not exist */ | |
284 | __asm__("1: movl %%cr4, %0 \n" | |
285 | "2: \n" | |
286 | ".section __ex_table,\"a\" \n" | |
287 | ".long 1b,2b \n" | |
288 | ".previous \n" | |
289 | : "=r" (cr4): "0" (0)); | |
290 | printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4); | |
291 | show_trace(NULL, ®s->esp); | |
292 | } | |
293 | ||
294 | /* | |
295 | * This gets run with %ebx containing the | |
296 | * function to call, and %edx containing | |
297 | * the "args". | |
298 | */ | |
299 | extern void kernel_thread_helper(void); | |
300 | __asm__(".section .text\n" | |
301 | ".align 4\n" | |
302 | "kernel_thread_helper:\n\t" | |
303 | "movl %edx,%eax\n\t" | |
304 | "pushl %edx\n\t" | |
305 | "call *%ebx\n\t" | |
306 | "pushl %eax\n\t" | |
307 | "call do_exit\n" | |
308 | ".previous"); | |
309 | ||
310 | /* | |
311 | * Create a kernel thread | |
312 | */ | |
313 | int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags) | |
314 | { | |
315 | struct pt_regs regs; | |
316 | ||
317 | memset(®s, 0, sizeof(regs)); | |
318 | ||
319 | regs.ebx = (unsigned long) fn; | |
320 | regs.edx = (unsigned long) arg; | |
321 | ||
322 | regs.xds = __USER_DS; | |
323 | regs.xes = __USER_DS; | |
324 | regs.orig_eax = -1; | |
325 | regs.eip = (unsigned long) kernel_thread_helper; | |
326 | regs.xcs = __KERNEL_CS; | |
327 | regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2; | |
328 | ||
329 | /* Ok, create the new process.. */ | |
330 | return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL); | |
331 | } | |
129f6946 | 332 | EXPORT_SYMBOL(kernel_thread); |
1da177e4 LT |
333 | |
334 | /* | |
335 | * Free current thread data structures etc.. | |
336 | */ | |
337 | void exit_thread(void) | |
338 | { | |
339 | struct task_struct *tsk = current; | |
340 | struct thread_struct *t = &tsk->thread; | |
341 | ||
342 | /* The process may have allocated an io port bitmap... nuke it. */ | |
343 | if (unlikely(NULL != t->io_bitmap_ptr)) { | |
344 | int cpu = get_cpu(); | |
345 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
346 | ||
347 | kfree(t->io_bitmap_ptr); | |
348 | t->io_bitmap_ptr = NULL; | |
349 | /* | |
350 | * Careful, clear this in the TSS too: | |
351 | */ | |
352 | memset(tss->io_bitmap, 0xff, tss->io_bitmap_max); | |
353 | t->io_bitmap_max = 0; | |
354 | tss->io_bitmap_owner = NULL; | |
355 | tss->io_bitmap_max = 0; | |
356 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET; | |
357 | put_cpu(); | |
358 | } | |
359 | } | |
360 | ||
361 | void flush_thread(void) | |
362 | { | |
363 | struct task_struct *tsk = current; | |
364 | ||
365 | memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8); | |
366 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); | |
367 | /* | |
368 | * Forget coprocessor state.. | |
369 | */ | |
370 | clear_fpu(tsk); | |
371 | clear_used_math(); | |
372 | } | |
373 | ||
374 | void release_thread(struct task_struct *dead_task) | |
375 | { | |
376 | if (dead_task->mm) { | |
377 | // temporary debugging check | |
378 | if (dead_task->mm->context.size) { | |
379 | printk("WARNING: dead process %8s still has LDT? <%p/%d>\n", | |
380 | dead_task->comm, | |
381 | dead_task->mm->context.ldt, | |
382 | dead_task->mm->context.size); | |
383 | BUG(); | |
384 | } | |
385 | } | |
386 | ||
387 | release_vm86_irqs(dead_task); | |
388 | } | |
389 | ||
390 | /* | |
391 | * This gets called before we allocate a new thread and copy | |
392 | * the current task into it. | |
393 | */ | |
394 | void prepare_to_copy(struct task_struct *tsk) | |
395 | { | |
396 | unlazy_fpu(tsk); | |
397 | } | |
398 | ||
399 | int copy_thread(int nr, unsigned long clone_flags, unsigned long esp, | |
400 | unsigned long unused, | |
401 | struct task_struct * p, struct pt_regs * regs) | |
402 | { | |
403 | struct pt_regs * childregs; | |
404 | struct task_struct *tsk; | |
405 | int err; | |
406 | ||
407 | childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1; | |
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 | */ | |
f48d9663 AN |
418 | childregs = (struct pt_regs *) ((unsigned long) childregs - 8); |
419 | *childregs = *regs; | |
420 | childregs->eax = 0; | |
421 | childregs->esp = esp; | |
422 | ||
423 | p->thread.esp = (unsigned long) childregs; | |
424 | p->thread.esp0 = (unsigned long) (childregs+1); | |
1da177e4 LT |
425 | |
426 | p->thread.eip = (unsigned long) ret_from_fork; | |
427 | ||
428 | savesegment(fs,p->thread.fs); | |
429 | savesegment(gs,p->thread.gs); | |
430 | ||
431 | tsk = current; | |
432 | if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) { | |
433 | p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL); | |
434 | if (!p->thread.io_bitmap_ptr) { | |
435 | p->thread.io_bitmap_max = 0; | |
436 | return -ENOMEM; | |
437 | } | |
438 | memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr, | |
439 | IO_BITMAP_BYTES); | |
440 | } | |
441 | ||
442 | /* | |
443 | * Set a new TLS for the child thread? | |
444 | */ | |
445 | if (clone_flags & CLONE_SETTLS) { | |
446 | struct desc_struct *desc; | |
447 | struct user_desc info; | |
448 | int idx; | |
449 | ||
450 | err = -EFAULT; | |
451 | if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info))) | |
452 | goto out; | |
453 | err = -EINVAL; | |
454 | if (LDT_empty(&info)) | |
455 | goto out; | |
456 | ||
457 | idx = info.entry_number; | |
458 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
459 | goto out; | |
460 | ||
461 | desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; | |
462 | desc->a = LDT_entry_a(&info); | |
463 | desc->b = LDT_entry_b(&info); | |
464 | } | |
465 | ||
466 | err = 0; | |
467 | out: | |
468 | if (err && p->thread.io_bitmap_ptr) { | |
469 | kfree(p->thread.io_bitmap_ptr); | |
470 | p->thread.io_bitmap_max = 0; | |
471 | } | |
472 | return err; | |
473 | } | |
474 | ||
475 | /* | |
476 | * fill in the user structure for a core dump.. | |
477 | */ | |
478 | void dump_thread(struct pt_regs * regs, struct user * dump) | |
479 | { | |
480 | int i; | |
481 | ||
482 | /* changed the size calculations - should hopefully work better. lbt */ | |
483 | dump->magic = CMAGIC; | |
484 | dump->start_code = 0; | |
485 | dump->start_stack = regs->esp & ~(PAGE_SIZE - 1); | |
486 | dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT; | |
487 | dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT; | |
488 | dump->u_dsize -= dump->u_tsize; | |
489 | dump->u_ssize = 0; | |
490 | for (i = 0; i < 8; i++) | |
491 | dump->u_debugreg[i] = current->thread.debugreg[i]; | |
492 | ||
493 | if (dump->start_stack < TASK_SIZE) | |
494 | dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT; | |
495 | ||
496 | dump->regs.ebx = regs->ebx; | |
497 | dump->regs.ecx = regs->ecx; | |
498 | dump->regs.edx = regs->edx; | |
499 | dump->regs.esi = regs->esi; | |
500 | dump->regs.edi = regs->edi; | |
501 | dump->regs.ebp = regs->ebp; | |
502 | dump->regs.eax = regs->eax; | |
503 | dump->regs.ds = regs->xds; | |
504 | dump->regs.es = regs->xes; | |
505 | savesegment(fs,dump->regs.fs); | |
506 | savesegment(gs,dump->regs.gs); | |
507 | dump->regs.orig_eax = regs->orig_eax; | |
508 | dump->regs.eip = regs->eip; | |
509 | dump->regs.cs = regs->xcs; | |
510 | dump->regs.eflags = regs->eflags; | |
511 | dump->regs.esp = regs->esp; | |
512 | dump->regs.ss = regs->xss; | |
513 | ||
514 | dump->u_fpvalid = dump_fpu (regs, &dump->i387); | |
515 | } | |
129f6946 | 516 | EXPORT_SYMBOL(dump_thread); |
1da177e4 LT |
517 | |
518 | /* | |
519 | * Capture the user space registers if the task is not running (in user space) | |
520 | */ | |
521 | int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs) | |
522 | { | |
523 | struct pt_regs ptregs; | |
524 | ||
525 | ptregs = *(struct pt_regs *) | |
526 | ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs)); | |
527 | ptregs.xcs &= 0xffff; | |
528 | ptregs.xds &= 0xffff; | |
529 | ptregs.xes &= 0xffff; | |
530 | ptregs.xss &= 0xffff; | |
531 | ||
532 | elf_core_copy_regs(regs, &ptregs); | |
533 | ||
534 | return 1; | |
535 | } | |
536 | ||
537 | static inline void | |
538 | handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss) | |
539 | { | |
540 | if (!next->io_bitmap_ptr) { | |
541 | /* | |
542 | * Disable the bitmap via an invalid offset. We still cache | |
543 | * the previous bitmap owner and the IO bitmap contents: | |
544 | */ | |
545 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET; | |
546 | return; | |
547 | } | |
548 | if (likely(next == tss->io_bitmap_owner)) { | |
549 | /* | |
550 | * Previous owner of the bitmap (hence the bitmap content) | |
551 | * matches the next task, we dont have to do anything but | |
552 | * to set a valid offset in the TSS: | |
553 | */ | |
554 | tss->io_bitmap_base = IO_BITMAP_OFFSET; | |
555 | return; | |
556 | } | |
557 | /* | |
558 | * Lazy TSS's I/O bitmap copy. We set an invalid offset here | |
559 | * and we let the task to get a GPF in case an I/O instruction | |
560 | * is performed. The handler of the GPF will verify that the | |
561 | * faulting task has a valid I/O bitmap and, it true, does the | |
562 | * real copy and restart the instruction. This will save us | |
563 | * redundant copies when the currently switched task does not | |
564 | * perform any I/O during its timeslice. | |
565 | */ | |
566 | tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY; | |
567 | } | |
1da177e4 LT |
568 | |
569 | /* | |
570 | * switch_to(x,yn) should switch tasks from x to y. | |
571 | * | |
572 | * We fsave/fwait so that an exception goes off at the right time | |
573 | * (as a call from the fsave or fwait in effect) rather than to | |
574 | * the wrong process. Lazy FP saving no longer makes any sense | |
575 | * with modern CPU's, and this simplifies a lot of things (SMP | |
576 | * and UP become the same). | |
577 | * | |
578 | * NOTE! We used to use the x86 hardware context switching. The | |
579 | * reason for not using it any more becomes apparent when you | |
580 | * try to recover gracefully from saved state that is no longer | |
581 | * valid (stale segment register values in particular). With the | |
582 | * hardware task-switch, there is no way to fix up bad state in | |
583 | * a reasonable manner. | |
584 | * | |
585 | * The fact that Intel documents the hardware task-switching to | |
586 | * be slow is a fairly red herring - this code is not noticeably | |
587 | * faster. However, there _is_ some room for improvement here, | |
588 | * so the performance issues may eventually be a valid point. | |
589 | * More important, however, is the fact that this allows us much | |
590 | * more flexibility. | |
591 | * | |
592 | * The return value (in %eax) will be the "prev" task after | |
593 | * the task-switch, and shows up in ret_from_fork in entry.S, | |
594 | * for example. | |
595 | */ | |
596 | struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p) | |
597 | { | |
598 | struct thread_struct *prev = &prev_p->thread, | |
599 | *next = &next_p->thread; | |
600 | int cpu = smp_processor_id(); | |
601 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
602 | ||
603 | /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ | |
604 | ||
605 | __unlazy_fpu(prev_p); | |
606 | ||
607 | /* | |
608 | * Reload esp0, LDT and the page table pointer: | |
609 | */ | |
610 | load_esp0(tss, next); | |
611 | ||
612 | /* | |
613 | * Load the per-thread Thread-Local Storage descriptor. | |
614 | */ | |
615 | load_TLS(next, cpu); | |
616 | ||
617 | /* | |
618 | * Save away %fs and %gs. No need to save %es and %ds, as | |
619 | * those are always kernel segments while inside the kernel. | |
620 | */ | |
fd51f666 L |
621 | asm volatile("mov %%fs,%0":"=m" (prev->fs)); |
622 | asm volatile("mov %%gs,%0":"=m" (prev->gs)); | |
1da177e4 LT |
623 | |
624 | /* | |
625 | * Restore %fs and %gs if needed. | |
626 | */ | |
627 | if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) { | |
628 | loadsegment(fs, next->fs); | |
629 | loadsegment(gs, next->gs); | |
630 | } | |
631 | ||
632 | /* | |
633 | * Now maybe reload the debug registers | |
634 | */ | |
635 | if (unlikely(next->debugreg[7])) { | |
1cc6f12e VH |
636 | set_debugreg(current->thread.debugreg[0], 0); |
637 | set_debugreg(current->thread.debugreg[1], 1); | |
638 | set_debugreg(current->thread.debugreg[2], 2); | |
639 | set_debugreg(current->thread.debugreg[3], 3); | |
1da177e4 | 640 | /* no 4 and 5 */ |
1cc6f12e VH |
641 | set_debugreg(current->thread.debugreg[6], 6); |
642 | set_debugreg(current->thread.debugreg[7], 7); | |
1da177e4 LT |
643 | } |
644 | ||
645 | if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) | |
646 | handle_io_bitmap(next, tss); | |
647 | ||
648 | return prev_p; | |
649 | } | |
650 | ||
651 | asmlinkage int sys_fork(struct pt_regs regs) | |
652 | { | |
653 | return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL); | |
654 | } | |
655 | ||
656 | asmlinkage int sys_clone(struct pt_regs regs) | |
657 | { | |
658 | unsigned long clone_flags; | |
659 | unsigned long newsp; | |
660 | int __user *parent_tidptr, *child_tidptr; | |
661 | ||
662 | clone_flags = regs.ebx; | |
663 | newsp = regs.ecx; | |
664 | parent_tidptr = (int __user *)regs.edx; | |
665 | child_tidptr = (int __user *)regs.edi; | |
666 | if (!newsp) | |
667 | newsp = regs.esp; | |
668 | return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr); | |
669 | } | |
670 | ||
671 | /* | |
672 | * This is trivial, and on the face of it looks like it | |
673 | * could equally well be done in user mode. | |
674 | * | |
675 | * Not so, for quite unobvious reasons - register pressure. | |
676 | * In user mode vfork() cannot have a stack frame, and if | |
677 | * done by calling the "clone()" system call directly, you | |
678 | * do not have enough call-clobbered registers to hold all | |
679 | * the information you need. | |
680 | */ | |
681 | asmlinkage int sys_vfork(struct pt_regs regs) | |
682 | { | |
683 | return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL); | |
684 | } | |
685 | ||
686 | /* | |
687 | * sys_execve() executes a new program. | |
688 | */ | |
689 | asmlinkage int sys_execve(struct pt_regs regs) | |
690 | { | |
691 | int error; | |
692 | char * filename; | |
693 | ||
694 | filename = getname((char __user *) regs.ebx); | |
695 | error = PTR_ERR(filename); | |
696 | if (IS_ERR(filename)) | |
697 | goto out; | |
698 | error = do_execve(filename, | |
699 | (char __user * __user *) regs.ecx, | |
700 | (char __user * __user *) regs.edx, | |
701 | ®s); | |
702 | if (error == 0) { | |
703 | task_lock(current); | |
704 | current->ptrace &= ~PT_DTRACE; | |
705 | task_unlock(current); | |
706 | /* Make sure we don't return using sysenter.. */ | |
707 | set_thread_flag(TIF_IRET); | |
708 | } | |
709 | putname(filename); | |
710 | out: | |
711 | return error; | |
712 | } | |
713 | ||
714 | #define top_esp (THREAD_SIZE - sizeof(unsigned long)) | |
715 | #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) | |
716 | ||
717 | unsigned long get_wchan(struct task_struct *p) | |
718 | { | |
719 | unsigned long ebp, esp, eip; | |
720 | unsigned long stack_page; | |
721 | int count = 0; | |
722 | if (!p || p == current || p->state == TASK_RUNNING) | |
723 | return 0; | |
724 | stack_page = (unsigned long)p->thread_info; | |
725 | esp = p->thread.esp; | |
726 | if (!stack_page || esp < stack_page || esp > top_esp+stack_page) | |
727 | return 0; | |
728 | /* include/asm-i386/system.h:switch_to() pushes ebp last. */ | |
729 | ebp = *(unsigned long *) esp; | |
730 | do { | |
731 | if (ebp < stack_page || ebp > top_ebp+stack_page) | |
732 | return 0; | |
733 | eip = *(unsigned long *) (ebp+4); | |
734 | if (!in_sched_functions(eip)) | |
735 | return eip; | |
736 | ebp = *(unsigned long *) ebp; | |
737 | } while (count++ < 16); | |
738 | return 0; | |
739 | } | |
129f6946 | 740 | EXPORT_SYMBOL(get_wchan); |
1da177e4 LT |
741 | |
742 | /* | |
743 | * sys_alloc_thread_area: get a yet unused TLS descriptor index. | |
744 | */ | |
745 | static int get_free_idx(void) | |
746 | { | |
747 | struct thread_struct *t = ¤t->thread; | |
748 | int idx; | |
749 | ||
750 | for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++) | |
751 | if (desc_empty(t->tls_array + idx)) | |
752 | return idx + GDT_ENTRY_TLS_MIN; | |
753 | return -ESRCH; | |
754 | } | |
755 | ||
756 | /* | |
757 | * Set a given TLS descriptor: | |
758 | */ | |
759 | asmlinkage int sys_set_thread_area(struct user_desc __user *u_info) | |
760 | { | |
761 | struct thread_struct *t = ¤t->thread; | |
762 | struct user_desc info; | |
763 | struct desc_struct *desc; | |
764 | int cpu, idx; | |
765 | ||
766 | if (copy_from_user(&info, u_info, sizeof(info))) | |
767 | return -EFAULT; | |
768 | idx = info.entry_number; | |
769 | ||
770 | /* | |
771 | * index -1 means the kernel should try to find and | |
772 | * allocate an empty descriptor: | |
773 | */ | |
774 | if (idx == -1) { | |
775 | idx = get_free_idx(); | |
776 | if (idx < 0) | |
777 | return idx; | |
778 | if (put_user(idx, &u_info->entry_number)) | |
779 | return -EFAULT; | |
780 | } | |
781 | ||
782 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
783 | return -EINVAL; | |
784 | ||
785 | desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN; | |
786 | ||
787 | /* | |
788 | * We must not get preempted while modifying the TLS. | |
789 | */ | |
790 | cpu = get_cpu(); | |
791 | ||
792 | if (LDT_empty(&info)) { | |
793 | desc->a = 0; | |
794 | desc->b = 0; | |
795 | } else { | |
796 | desc->a = LDT_entry_a(&info); | |
797 | desc->b = LDT_entry_b(&info); | |
798 | } | |
799 | load_TLS(t, cpu); | |
800 | ||
801 | put_cpu(); | |
802 | ||
803 | return 0; | |
804 | } | |
805 | ||
806 | /* | |
807 | * Get the current Thread-Local Storage area: | |
808 | */ | |
809 | ||
810 | #define GET_BASE(desc) ( \ | |
811 | (((desc)->a >> 16) & 0x0000ffff) | \ | |
812 | (((desc)->b << 16) & 0x00ff0000) | \ | |
813 | ( (desc)->b & 0xff000000) ) | |
814 | ||
815 | #define GET_LIMIT(desc) ( \ | |
816 | ((desc)->a & 0x0ffff) | \ | |
817 | ((desc)->b & 0xf0000) ) | |
818 | ||
819 | #define GET_32BIT(desc) (((desc)->b >> 22) & 1) | |
820 | #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3) | |
821 | #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1) | |
822 | #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1) | |
823 | #define GET_PRESENT(desc) (((desc)->b >> 15) & 1) | |
824 | #define GET_USEABLE(desc) (((desc)->b >> 20) & 1) | |
825 | ||
826 | asmlinkage int sys_get_thread_area(struct user_desc __user *u_info) | |
827 | { | |
828 | struct user_desc info; | |
829 | struct desc_struct *desc; | |
830 | int idx; | |
831 | ||
832 | if (get_user(idx, &u_info->entry_number)) | |
833 | return -EFAULT; | |
834 | if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) | |
835 | return -EINVAL; | |
836 | ||
837 | desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; | |
838 | ||
839 | info.entry_number = idx; | |
840 | info.base_addr = GET_BASE(desc); | |
841 | info.limit = GET_LIMIT(desc); | |
842 | info.seg_32bit = GET_32BIT(desc); | |
843 | info.contents = GET_CONTENTS(desc); | |
844 | info.read_exec_only = !GET_WRITABLE(desc); | |
845 | info.limit_in_pages = GET_LIMIT_PAGES(desc); | |
846 | info.seg_not_present = !GET_PRESENT(desc); | |
847 | info.useable = GET_USEABLE(desc); | |
848 | ||
849 | if (copy_to_user(u_info, &info, sizeof(info))) | |
850 | return -EFAULT; | |
851 | return 0; | |
852 | } | |
853 | ||
854 | unsigned long arch_align_stack(unsigned long sp) | |
855 | { | |
856 | if (randomize_va_space) | |
857 | sp -= get_random_int() % 8192; | |
858 | return sp & ~0xf; | |
859 | } |