Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86-64/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 | * X86-64 port | |
10 | * Andi Kleen. | |
76e4f660 AR |
11 | * |
12 | * CPU hotplug support - ashok.raj@intel.com | |
1da177e4 LT |
13 | */ |
14 | ||
15 | /* | |
16 | * This file handles the architecture-dependent parts of process handling.. | |
17 | */ | |
18 | ||
19 | #include <stdarg.h> | |
20 | ||
76e4f660 | 21 | #include <linux/cpu.h> |
1da177e4 LT |
22 | #include <linux/errno.h> |
23 | #include <linux/sched.h> | |
24 | #include <linux/kernel.h> | |
25 | #include <linux/mm.h> | |
26 | #include <linux/elfcore.h> | |
27 | #include <linux/smp.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/user.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/a.out.h> | |
32 | #include <linux/interrupt.h> | |
33 | #include <linux/delay.h> | |
1da177e4 LT |
34 | #include <linux/ptrace.h> |
35 | #include <linux/utsname.h> | |
36 | #include <linux/random.h> | |
95833c83 | 37 | #include <linux/notifier.h> |
c6fd91f0 | 38 | #include <linux/kprobes.h> |
1da177e4 LT |
39 | |
40 | #include <asm/uaccess.h> | |
41 | #include <asm/pgtable.h> | |
42 | #include <asm/system.h> | |
43 | #include <asm/io.h> | |
44 | #include <asm/processor.h> | |
45 | #include <asm/i387.h> | |
46 | #include <asm/mmu_context.h> | |
47 | #include <asm/pda.h> | |
48 | #include <asm/prctl.h> | |
49 | #include <asm/kdebug.h> | |
50 | #include <asm/desc.h> | |
51 | #include <asm/proto.h> | |
52 | #include <asm/ia32.h> | |
95833c83 | 53 | #include <asm/idle.h> |
1da177e4 LT |
54 | |
55 | asmlinkage extern void ret_from_fork(void); | |
56 | ||
57 | unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED; | |
58 | ||
1da177e4 LT |
59 | unsigned long boot_option_idle_override = 0; |
60 | EXPORT_SYMBOL(boot_option_idle_override); | |
61 | ||
62 | /* | |
63 | * Powermanagement idle function, if any.. | |
64 | */ | |
65 | void (*pm_idle)(void); | |
2ee60e17 | 66 | EXPORT_SYMBOL(pm_idle); |
1da177e4 LT |
67 | static DEFINE_PER_CPU(unsigned int, cpu_idle_state); |
68 | ||
e041c683 | 69 | static ATOMIC_NOTIFIER_HEAD(idle_notifier); |
95833c83 AK |
70 | |
71 | void idle_notifier_register(struct notifier_block *n) | |
72 | { | |
e041c683 | 73 | atomic_notifier_chain_register(&idle_notifier, n); |
95833c83 AK |
74 | } |
75 | EXPORT_SYMBOL_GPL(idle_notifier_register); | |
76 | ||
77 | void idle_notifier_unregister(struct notifier_block *n) | |
78 | { | |
e041c683 | 79 | atomic_notifier_chain_unregister(&idle_notifier, n); |
95833c83 AK |
80 | } |
81 | EXPORT_SYMBOL(idle_notifier_unregister); | |
82 | ||
95833c83 AK |
83 | void enter_idle(void) |
84 | { | |
a15da49d | 85 | write_pda(isidle, 1); |
e041c683 | 86 | atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL); |
95833c83 AK |
87 | } |
88 | ||
89 | static void __exit_idle(void) | |
90 | { | |
a15da49d AK |
91 | if (read_pda(isidle) == 0) |
92 | return; | |
93 | write_pda(isidle, 0); | |
e041c683 | 94 | atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL); |
95833c83 AK |
95 | } |
96 | ||
97 | /* Called from interrupts to signify idle end */ | |
98 | void exit_idle(void) | |
99 | { | |
a15da49d AK |
100 | /* idle loop has pid 0 */ |
101 | if (current->pid) | |
95833c83 AK |
102 | return; |
103 | __exit_idle(); | |
104 | } | |
105 | ||
1da177e4 LT |
106 | /* |
107 | * We use this if we don't have any better | |
108 | * idle routine.. | |
109 | */ | |
cdb04527 | 110 | static void default_idle(void) |
1da177e4 | 111 | { |
64c7c8f8 NP |
112 | local_irq_enable(); |
113 | ||
495ab9c0 | 114 | current_thread_info()->status &= ~TS_POLLING; |
2d52ede9 AK |
115 | smp_mb__after_clear_bit(); |
116 | while (!need_resched()) { | |
117 | local_irq_disable(); | |
118 | if (!need_resched()) | |
119 | safe_halt(); | |
120 | else | |
121 | local_irq_enable(); | |
1da177e4 | 122 | } |
495ab9c0 | 123 | current_thread_info()->status |= TS_POLLING; |
1da177e4 LT |
124 | } |
125 | ||
126 | /* | |
127 | * On SMP it's slightly faster (but much more power-consuming!) | |
128 | * to poll the ->need_resched flag instead of waiting for the | |
129 | * cross-CPU IPI to arrive. Use this option with caution. | |
130 | */ | |
131 | static void poll_idle (void) | |
132 | { | |
1da177e4 LT |
133 | local_irq_enable(); |
134 | ||
64c7c8f8 NP |
135 | asm volatile( |
136 | "2:" | |
137 | "testl %0,%1;" | |
138 | "rep; nop;" | |
139 | "je 2b;" | |
140 | : : | |
141 | "i" (_TIF_NEED_RESCHED), | |
142 | "m" (current_thread_info()->flags)); | |
1da177e4 LT |
143 | } |
144 | ||
145 | void cpu_idle_wait(void) | |
146 | { | |
147 | unsigned int cpu, this_cpu = get_cpu(); | |
148 | cpumask_t map; | |
149 | ||
150 | set_cpus_allowed(current, cpumask_of_cpu(this_cpu)); | |
151 | put_cpu(); | |
152 | ||
153 | cpus_clear(map); | |
154 | for_each_online_cpu(cpu) { | |
155 | per_cpu(cpu_idle_state, cpu) = 1; | |
156 | cpu_set(cpu, map); | |
157 | } | |
158 | ||
159 | __get_cpu_var(cpu_idle_state) = 0; | |
160 | ||
161 | wmb(); | |
162 | do { | |
163 | ssleep(1); | |
164 | for_each_online_cpu(cpu) { | |
a88cde13 AK |
165 | if (cpu_isset(cpu, map) && |
166 | !per_cpu(cpu_idle_state, cpu)) | |
1da177e4 LT |
167 | cpu_clear(cpu, map); |
168 | } | |
169 | cpus_and(map, map, cpu_online_map); | |
170 | } while (!cpus_empty(map)); | |
171 | } | |
172 | EXPORT_SYMBOL_GPL(cpu_idle_wait); | |
173 | ||
76e4f660 AR |
174 | #ifdef CONFIG_HOTPLUG_CPU |
175 | DECLARE_PER_CPU(int, cpu_state); | |
176 | ||
177 | #include <asm/nmi.h> | |
1fa744e6 | 178 | /* We halt the CPU with physical CPU hotplug */ |
76e4f660 AR |
179 | static inline void play_dead(void) |
180 | { | |
181 | idle_task_exit(); | |
182 | wbinvd(); | |
183 | mb(); | |
184 | /* Ack it */ | |
185 | __get_cpu_var(cpu_state) = CPU_DEAD; | |
186 | ||
1fa744e6 | 187 | local_irq_disable(); |
76e4f660 | 188 | while (1) |
1fa744e6 | 189 | halt(); |
76e4f660 AR |
190 | } |
191 | #else | |
192 | static inline void play_dead(void) | |
193 | { | |
194 | BUG(); | |
195 | } | |
196 | #endif /* CONFIG_HOTPLUG_CPU */ | |
197 | ||
1da177e4 LT |
198 | /* |
199 | * The idle thread. There's no useful work to be | |
200 | * done, so just try to conserve power and have a | |
201 | * low exit latency (ie sit in a loop waiting for | |
202 | * somebody to say that they'd like to reschedule) | |
203 | */ | |
204 | void cpu_idle (void) | |
205 | { | |
495ab9c0 | 206 | current_thread_info()->status |= TS_POLLING; |
1da177e4 LT |
207 | /* endless idle loop with no priority at all */ |
208 | while (1) { | |
209 | while (!need_resched()) { | |
210 | void (*idle)(void); | |
211 | ||
212 | if (__get_cpu_var(cpu_idle_state)) | |
213 | __get_cpu_var(cpu_idle_state) = 0; | |
214 | ||
215 | rmb(); | |
216 | idle = pm_idle; | |
217 | if (!idle) | |
218 | idle = default_idle; | |
76e4f660 AR |
219 | if (cpu_is_offline(smp_processor_id())) |
220 | play_dead(); | |
95833c83 | 221 | enter_idle(); |
1da177e4 | 222 | idle(); |
a15da49d AK |
223 | /* In many cases the interrupt that ended idle |
224 | has already called exit_idle. But some idle | |
225 | loops can be woken up without interrupt. */ | |
95833c83 | 226 | __exit_idle(); |
1da177e4 LT |
227 | } |
228 | ||
5bfb5d69 | 229 | preempt_enable_no_resched(); |
1da177e4 | 230 | schedule(); |
5bfb5d69 | 231 | preempt_disable(); |
1da177e4 LT |
232 | } |
233 | } | |
234 | ||
235 | /* | |
236 | * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, | |
237 | * which can obviate IPI to trigger checking of need_resched. | |
238 | * We execute MONITOR against need_resched and enter optimized wait state | |
239 | * through MWAIT. Whenever someone changes need_resched, we would be woken | |
240 | * up from MWAIT (without an IPI). | |
241 | */ | |
242 | static void mwait_idle(void) | |
243 | { | |
244 | local_irq_enable(); | |
245 | ||
64c7c8f8 NP |
246 | while (!need_resched()) { |
247 | __monitor((void *)¤t_thread_info()->flags, 0, 0); | |
248 | smp_mb(); | |
249 | if (need_resched()) | |
250 | break; | |
251 | __mwait(0, 0); | |
1da177e4 LT |
252 | } |
253 | } | |
254 | ||
e6982c67 | 255 | void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c) |
1da177e4 LT |
256 | { |
257 | static int printed; | |
258 | if (cpu_has(c, X86_FEATURE_MWAIT)) { | |
259 | /* | |
260 | * Skip, if setup has overridden idle. | |
261 | * One CPU supports mwait => All CPUs supports mwait | |
262 | */ | |
263 | if (!pm_idle) { | |
264 | if (!printed) { | |
265 | printk("using mwait in idle threads.\n"); | |
266 | printed = 1; | |
267 | } | |
268 | pm_idle = mwait_idle; | |
269 | } | |
270 | } | |
271 | } | |
272 | ||
273 | static int __init idle_setup (char *str) | |
274 | { | |
275 | if (!strncmp(str, "poll", 4)) { | |
276 | printk("using polling idle threads.\n"); | |
277 | pm_idle = poll_idle; | |
278 | } | |
279 | ||
280 | boot_option_idle_override = 1; | |
281 | return 1; | |
282 | } | |
283 | ||
284 | __setup("idle=", idle_setup); | |
285 | ||
286 | /* Prints also some state that isn't saved in the pt_regs */ | |
287 | void __show_regs(struct pt_regs * regs) | |
288 | { | |
289 | unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs; | |
290 | unsigned int fsindex,gsindex; | |
291 | unsigned int ds,cs,es; | |
292 | ||
293 | printk("\n"); | |
294 | print_modules(); | |
9acf23c4 AK |
295 | printk("Pid: %d, comm: %.20s %s %s %.*s\n", |
296 | current->pid, current->comm, print_tainted(), | |
297 | system_utsname.release, | |
298 | (int)strcspn(system_utsname.version, " "), | |
299 | system_utsname.version); | |
1da177e4 LT |
300 | printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip); |
301 | printk_address(regs->rip); | |
3ac94932 | 302 | printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp, |
a88cde13 | 303 | regs->eflags); |
1da177e4 LT |
304 | printk("RAX: %016lx RBX: %016lx RCX: %016lx\n", |
305 | regs->rax, regs->rbx, regs->rcx); | |
306 | printk("RDX: %016lx RSI: %016lx RDI: %016lx\n", | |
307 | regs->rdx, regs->rsi, regs->rdi); | |
308 | printk("RBP: %016lx R08: %016lx R09: %016lx\n", | |
309 | regs->rbp, regs->r8, regs->r9); | |
310 | printk("R10: %016lx R11: %016lx R12: %016lx\n", | |
311 | regs->r10, regs->r11, regs->r12); | |
312 | printk("R13: %016lx R14: %016lx R15: %016lx\n", | |
313 | regs->r13, regs->r14, regs->r15); | |
314 | ||
315 | asm("movl %%ds,%0" : "=r" (ds)); | |
316 | asm("movl %%cs,%0" : "=r" (cs)); | |
317 | asm("movl %%es,%0" : "=r" (es)); | |
318 | asm("movl %%fs,%0" : "=r" (fsindex)); | |
319 | asm("movl %%gs,%0" : "=r" (gsindex)); | |
320 | ||
321 | rdmsrl(MSR_FS_BASE, fs); | |
322 | rdmsrl(MSR_GS_BASE, gs); | |
323 | rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); | |
324 | ||
325 | asm("movq %%cr0, %0": "=r" (cr0)); | |
326 | asm("movq %%cr2, %0": "=r" (cr2)); | |
327 | asm("movq %%cr3, %0": "=r" (cr3)); | |
328 | asm("movq %%cr4, %0": "=r" (cr4)); | |
329 | ||
330 | printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", | |
331 | fs,fsindex,gs,gsindex,shadowgs); | |
332 | printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0); | |
333 | printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4); | |
334 | } | |
335 | ||
336 | void show_regs(struct pt_regs *regs) | |
337 | { | |
c078d326 | 338 | printk("CPU %d:", smp_processor_id()); |
1da177e4 | 339 | __show_regs(regs); |
b538ed27 | 340 | show_trace(NULL, regs, (void *)(regs + 1)); |
1da177e4 LT |
341 | } |
342 | ||
343 | /* | |
344 | * Free current thread data structures etc.. | |
345 | */ | |
346 | void exit_thread(void) | |
347 | { | |
348 | struct task_struct *me = current; | |
349 | struct thread_struct *t = &me->thread; | |
73649dab | 350 | |
1da177e4 LT |
351 | if (me->thread.io_bitmap_ptr) { |
352 | struct tss_struct *tss = &per_cpu(init_tss, get_cpu()); | |
353 | ||
354 | kfree(t->io_bitmap_ptr); | |
355 | t->io_bitmap_ptr = NULL; | |
d3a4f48d | 356 | clear_thread_flag(TIF_IO_BITMAP); |
1da177e4 LT |
357 | /* |
358 | * Careful, clear this in the TSS too: | |
359 | */ | |
360 | memset(tss->io_bitmap, 0xff, t->io_bitmap_max); | |
361 | t->io_bitmap_max = 0; | |
362 | put_cpu(); | |
363 | } | |
364 | } | |
365 | ||
366 | void flush_thread(void) | |
367 | { | |
368 | struct task_struct *tsk = current; | |
369 | struct thread_info *t = current_thread_info(); | |
370 | ||
4d9bc79c | 371 | if (t->flags & _TIF_ABI_PENDING) { |
1da177e4 | 372 | t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32); |
4d9bc79c AK |
373 | if (t->flags & _TIF_IA32) |
374 | current_thread_info()->status |= TS_COMPAT; | |
375 | } | |
d3a4f48d | 376 | t->flags &= ~_TIF_DEBUG; |
1da177e4 LT |
377 | |
378 | tsk->thread.debugreg0 = 0; | |
379 | tsk->thread.debugreg1 = 0; | |
380 | tsk->thread.debugreg2 = 0; | |
381 | tsk->thread.debugreg3 = 0; | |
382 | tsk->thread.debugreg6 = 0; | |
383 | tsk->thread.debugreg7 = 0; | |
384 | memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); | |
385 | /* | |
386 | * Forget coprocessor state.. | |
387 | */ | |
388 | clear_fpu(tsk); | |
389 | clear_used_math(); | |
390 | } | |
391 | ||
392 | void release_thread(struct task_struct *dead_task) | |
393 | { | |
394 | if (dead_task->mm) { | |
395 | if (dead_task->mm->context.size) { | |
396 | printk("WARNING: dead process %8s still has LDT? <%p/%d>\n", | |
397 | dead_task->comm, | |
398 | dead_task->mm->context.ldt, | |
399 | dead_task->mm->context.size); | |
400 | BUG(); | |
401 | } | |
402 | } | |
403 | } | |
404 | ||
405 | static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr) | |
406 | { | |
407 | struct user_desc ud = { | |
408 | .base_addr = addr, | |
409 | .limit = 0xfffff, | |
410 | .seg_32bit = 1, | |
411 | .limit_in_pages = 1, | |
412 | .useable = 1, | |
413 | }; | |
414 | struct n_desc_struct *desc = (void *)t->thread.tls_array; | |
415 | desc += tls; | |
416 | desc->a = LDT_entry_a(&ud); | |
417 | desc->b = LDT_entry_b(&ud); | |
418 | } | |
419 | ||
420 | static inline u32 read_32bit_tls(struct task_struct *t, int tls) | |
421 | { | |
422 | struct desc_struct *desc = (void *)t->thread.tls_array; | |
423 | desc += tls; | |
424 | return desc->base0 | | |
425 | (((u32)desc->base1) << 16) | | |
426 | (((u32)desc->base2) << 24); | |
427 | } | |
428 | ||
429 | /* | |
430 | * This gets called before we allocate a new thread and copy | |
431 | * the current task into it. | |
432 | */ | |
433 | void prepare_to_copy(struct task_struct *tsk) | |
434 | { | |
435 | unlazy_fpu(tsk); | |
436 | } | |
437 | ||
438 | int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp, | |
439 | unsigned long unused, | |
440 | struct task_struct * p, struct pt_regs * regs) | |
441 | { | |
442 | int err; | |
443 | struct pt_regs * childregs; | |
444 | struct task_struct *me = current; | |
445 | ||
a88cde13 | 446 | childregs = ((struct pt_regs *) |
57eafdc2 | 447 | (THREAD_SIZE + task_stack_page(p))) - 1; |
1da177e4 LT |
448 | *childregs = *regs; |
449 | ||
450 | childregs->rax = 0; | |
451 | childregs->rsp = rsp; | |
a88cde13 | 452 | if (rsp == ~0UL) |
1da177e4 | 453 | childregs->rsp = (unsigned long)childregs; |
1da177e4 LT |
454 | |
455 | p->thread.rsp = (unsigned long) childregs; | |
456 | p->thread.rsp0 = (unsigned long) (childregs+1); | |
457 | p->thread.userrsp = me->thread.userrsp; | |
458 | ||
e4f17c43 | 459 | set_tsk_thread_flag(p, TIF_FORK); |
1da177e4 LT |
460 | |
461 | p->thread.fs = me->thread.fs; | |
462 | p->thread.gs = me->thread.gs; | |
463 | ||
fd51f666 L |
464 | asm("mov %%gs,%0" : "=m" (p->thread.gsindex)); |
465 | asm("mov %%fs,%0" : "=m" (p->thread.fsindex)); | |
466 | asm("mov %%es,%0" : "=m" (p->thread.es)); | |
467 | asm("mov %%ds,%0" : "=m" (p->thread.ds)); | |
1da177e4 | 468 | |
d3a4f48d | 469 | if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { |
1da177e4 LT |
470 | p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL); |
471 | if (!p->thread.io_bitmap_ptr) { | |
472 | p->thread.io_bitmap_max = 0; | |
473 | return -ENOMEM; | |
474 | } | |
a88cde13 AK |
475 | memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, |
476 | IO_BITMAP_BYTES); | |
d3a4f48d | 477 | set_tsk_thread_flag(p, TIF_IO_BITMAP); |
1da177e4 LT |
478 | } |
479 | ||
480 | /* | |
481 | * Set a new TLS for the child thread? | |
482 | */ | |
483 | if (clone_flags & CLONE_SETTLS) { | |
484 | #ifdef CONFIG_IA32_EMULATION | |
485 | if (test_thread_flag(TIF_IA32)) | |
486 | err = ia32_child_tls(p, childregs); | |
487 | else | |
488 | #endif | |
489 | err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8); | |
490 | if (err) | |
491 | goto out; | |
492 | } | |
493 | err = 0; | |
494 | out: | |
495 | if (err && p->thread.io_bitmap_ptr) { | |
496 | kfree(p->thread.io_bitmap_ptr); | |
497 | p->thread.io_bitmap_max = 0; | |
498 | } | |
499 | return err; | |
500 | } | |
501 | ||
502 | /* | |
503 | * This special macro can be used to load a debugging register | |
504 | */ | |
2b514e74 | 505 | #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r) |
1da177e4 | 506 | |
d3a4f48d SE |
507 | static inline void __switch_to_xtra(struct task_struct *prev_p, |
508 | struct task_struct *next_p, | |
509 | struct tss_struct *tss) | |
510 | { | |
511 | struct thread_struct *prev, *next; | |
512 | ||
513 | prev = &prev_p->thread, | |
514 | next = &next_p->thread; | |
515 | ||
516 | if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { | |
517 | loaddebug(next, 0); | |
518 | loaddebug(next, 1); | |
519 | loaddebug(next, 2); | |
520 | loaddebug(next, 3); | |
521 | /* no 4 and 5 */ | |
522 | loaddebug(next, 6); | |
523 | loaddebug(next, 7); | |
524 | } | |
525 | ||
526 | if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { | |
527 | /* | |
528 | * Copy the relevant range of the IO bitmap. | |
529 | * Normally this is 128 bytes or less: | |
530 | */ | |
531 | memcpy(tss->io_bitmap, next->io_bitmap_ptr, | |
532 | max(prev->io_bitmap_max, next->io_bitmap_max)); | |
533 | } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) { | |
534 | /* | |
535 | * Clear any possible leftover bits: | |
536 | */ | |
537 | memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); | |
538 | } | |
539 | } | |
540 | ||
1da177e4 LT |
541 | /* |
542 | * switch_to(x,y) should switch tasks from x to y. | |
543 | * | |
544 | * This could still be optimized: | |
545 | * - fold all the options into a flag word and test it with a single test. | |
546 | * - could test fs/gs bitsliced | |
099f318b AK |
547 | * |
548 | * Kprobes not supported here. Set the probe on schedule instead. | |
1da177e4 | 549 | */ |
099f318b | 550 | __kprobes struct task_struct * |
a88cde13 | 551 | __switch_to(struct task_struct *prev_p, struct task_struct *next_p) |
1da177e4 LT |
552 | { |
553 | struct thread_struct *prev = &prev_p->thread, | |
554 | *next = &next_p->thread; | |
555 | int cpu = smp_processor_id(); | |
556 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
557 | ||
e07e23e1 AV |
558 | /* we're going to use this soon, after a few expensive things */ |
559 | if (next_p->fpu_counter>5) | |
560 | prefetch(&next->i387.fxsave); | |
561 | ||
1da177e4 LT |
562 | /* |
563 | * Reload esp0, LDT and the page table pointer: | |
564 | */ | |
565 | tss->rsp0 = next->rsp0; | |
566 | ||
567 | /* | |
568 | * Switch DS and ES. | |
569 | * This won't pick up thread selector changes, but I guess that is ok. | |
570 | */ | |
fd51f666 | 571 | asm volatile("mov %%es,%0" : "=m" (prev->es)); |
1da177e4 LT |
572 | if (unlikely(next->es | prev->es)) |
573 | loadsegment(es, next->es); | |
574 | ||
fd51f666 | 575 | asm volatile ("mov %%ds,%0" : "=m" (prev->ds)); |
1da177e4 LT |
576 | if (unlikely(next->ds | prev->ds)) |
577 | loadsegment(ds, next->ds); | |
578 | ||
579 | load_TLS(next, cpu); | |
580 | ||
581 | /* | |
582 | * Switch FS and GS. | |
583 | */ | |
584 | { | |
585 | unsigned fsindex; | |
586 | asm volatile("movl %%fs,%0" : "=r" (fsindex)); | |
587 | /* segment register != 0 always requires a reload. | |
588 | also reload when it has changed. | |
589 | when prev process used 64bit base always reload | |
590 | to avoid an information leak. */ | |
591 | if (unlikely(fsindex | next->fsindex | prev->fs)) { | |
592 | loadsegment(fs, next->fsindex); | |
593 | /* check if the user used a selector != 0 | |
594 | * if yes clear 64bit base, since overloaded base | |
595 | * is always mapped to the Null selector | |
596 | */ | |
597 | if (fsindex) | |
598 | prev->fs = 0; | |
599 | } | |
600 | /* when next process has a 64bit base use it */ | |
601 | if (next->fs) | |
602 | wrmsrl(MSR_FS_BASE, next->fs); | |
603 | prev->fsindex = fsindex; | |
604 | } | |
605 | { | |
606 | unsigned gsindex; | |
607 | asm volatile("movl %%gs,%0" : "=r" (gsindex)); | |
608 | if (unlikely(gsindex | next->gsindex | prev->gs)) { | |
609 | load_gs_index(next->gsindex); | |
610 | if (gsindex) | |
611 | prev->gs = 0; | |
612 | } | |
613 | if (next->gs) | |
614 | wrmsrl(MSR_KERNEL_GS_BASE, next->gs); | |
615 | prev->gsindex = gsindex; | |
616 | } | |
617 | ||
618 | /* | |
45948d77 | 619 | * Switch the PDA and FPU contexts. |
1da177e4 LT |
620 | */ |
621 | prev->userrsp = read_pda(oldrsp); | |
622 | write_pda(oldrsp, next->userrsp); | |
623 | write_pda(pcurrent, next_p); | |
18bd057b | 624 | |
45948d77 | 625 | /* This must be here to ensure both math_state_restore() and |
18bd057b AK |
626 | kernel_fpu_begin() work consistently. |
627 | And the AMD workaround requires it to be after DS reload. */ | |
45948d77 | 628 | unlazy_fpu(prev_p); |
a88cde13 | 629 | write_pda(kernelstack, |
7b0bda74 | 630 | (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET); |
0a425405 AV |
631 | #ifdef CONFIG_CC_STACKPROTECTOR |
632 | write_pda(stack_canary, next_p->stack_canary); | |
633 | /* | |
634 | * Build time only check to make sure the stack_canary is at | |
635 | * offset 40 in the pda; this is a gcc ABI requirement | |
636 | */ | |
637 | BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40); | |
638 | #endif | |
1da177e4 LT |
639 | |
640 | /* | |
d3a4f48d | 641 | * Now maybe reload the debug registers and handle I/O bitmaps |
1da177e4 | 642 | */ |
d3a4f48d SE |
643 | if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW)) |
644 | || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) | |
645 | __switch_to_xtra(prev_p, next_p, tss); | |
1da177e4 | 646 | |
e07e23e1 AV |
647 | /* If the task has used fpu the last 5 timeslices, just do a full |
648 | * restore of the math state immediately to avoid the trap; the | |
649 | * chances of needing FPU soon are obviously high now | |
650 | */ | |
651 | if (next_p->fpu_counter>5) | |
652 | math_state_restore(); | |
1da177e4 LT |
653 | return prev_p; |
654 | } | |
655 | ||
656 | /* | |
657 | * sys_execve() executes a new program. | |
658 | */ | |
659 | asmlinkage | |
660 | long sys_execve(char __user *name, char __user * __user *argv, | |
661 | char __user * __user *envp, struct pt_regs regs) | |
662 | { | |
663 | long error; | |
664 | char * filename; | |
665 | ||
666 | filename = getname(name); | |
667 | error = PTR_ERR(filename); | |
668 | if (IS_ERR(filename)) | |
669 | return error; | |
670 | error = do_execve(filename, argv, envp, ®s); | |
671 | if (error == 0) { | |
672 | task_lock(current); | |
673 | current->ptrace &= ~PT_DTRACE; | |
674 | task_unlock(current); | |
675 | } | |
676 | putname(filename); | |
677 | return error; | |
678 | } | |
679 | ||
680 | void set_personality_64bit(void) | |
681 | { | |
682 | /* inherit personality from parent */ | |
683 | ||
684 | /* Make sure to be in 64bit mode */ | |
685 | clear_thread_flag(TIF_IA32); | |
686 | ||
687 | /* TBD: overwrites user setup. Should have two bits. | |
688 | But 64bit processes have always behaved this way, | |
689 | so it's not too bad. The main problem is just that | |
690 | 32bit childs are affected again. */ | |
691 | current->personality &= ~READ_IMPLIES_EXEC; | |
692 | } | |
693 | ||
694 | asmlinkage long sys_fork(struct pt_regs *regs) | |
695 | { | |
696 | return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL); | |
697 | } | |
698 | ||
a88cde13 AK |
699 | asmlinkage long |
700 | sys_clone(unsigned long clone_flags, unsigned long newsp, | |
701 | void __user *parent_tid, void __user *child_tid, struct pt_regs *regs) | |
1da177e4 LT |
702 | { |
703 | if (!newsp) | |
704 | newsp = regs->rsp; | |
705 | return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid); | |
706 | } | |
707 | ||
708 | /* | |
709 | * This is trivial, and on the face of it looks like it | |
710 | * could equally well be done in user mode. | |
711 | * | |
712 | * Not so, for quite unobvious reasons - register pressure. | |
713 | * In user mode vfork() cannot have a stack frame, and if | |
714 | * done by calling the "clone()" system call directly, you | |
715 | * do not have enough call-clobbered registers to hold all | |
716 | * the information you need. | |
717 | */ | |
718 | asmlinkage long sys_vfork(struct pt_regs *regs) | |
719 | { | |
720 | return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0, | |
721 | NULL, NULL); | |
722 | } | |
723 | ||
724 | unsigned long get_wchan(struct task_struct *p) | |
725 | { | |
726 | unsigned long stack; | |
727 | u64 fp,rip; | |
728 | int count = 0; | |
729 | ||
730 | if (!p || p == current || p->state==TASK_RUNNING) | |
731 | return 0; | |
57eafdc2 | 732 | stack = (unsigned long)task_stack_page(p); |
1da177e4 LT |
733 | if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE) |
734 | return 0; | |
735 | fp = *(u64 *)(p->thread.rsp); | |
736 | do { | |
a88cde13 AK |
737 | if (fp < (unsigned long)stack || |
738 | fp > (unsigned long)stack+THREAD_SIZE) | |
1da177e4 LT |
739 | return 0; |
740 | rip = *(u64 *)(fp+8); | |
741 | if (!in_sched_functions(rip)) | |
742 | return rip; | |
743 | fp = *(u64 *)fp; | |
744 | } while (count++ < 16); | |
745 | return 0; | |
746 | } | |
747 | ||
748 | long do_arch_prctl(struct task_struct *task, int code, unsigned long addr) | |
749 | { | |
750 | int ret = 0; | |
751 | int doit = task == current; | |
752 | int cpu; | |
753 | ||
754 | switch (code) { | |
755 | case ARCH_SET_GS: | |
84929801 | 756 | if (addr >= TASK_SIZE_OF(task)) |
1da177e4 LT |
757 | return -EPERM; |
758 | cpu = get_cpu(); | |
759 | /* handle small bases via the GDT because that's faster to | |
760 | switch. */ | |
761 | if (addr <= 0xffffffff) { | |
762 | set_32bit_tls(task, GS_TLS, addr); | |
763 | if (doit) { | |
764 | load_TLS(&task->thread, cpu); | |
765 | load_gs_index(GS_TLS_SEL); | |
766 | } | |
767 | task->thread.gsindex = GS_TLS_SEL; | |
768 | task->thread.gs = 0; | |
769 | } else { | |
770 | task->thread.gsindex = 0; | |
771 | task->thread.gs = addr; | |
772 | if (doit) { | |
a88cde13 AK |
773 | load_gs_index(0); |
774 | ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr); | |
1da177e4 LT |
775 | } |
776 | } | |
777 | put_cpu(); | |
778 | break; | |
779 | case ARCH_SET_FS: | |
780 | /* Not strictly needed for fs, but do it for symmetry | |
781 | with gs */ | |
84929801 | 782 | if (addr >= TASK_SIZE_OF(task)) |
1da177e4 LT |
783 | return -EPERM; |
784 | cpu = get_cpu(); | |
785 | /* handle small bases via the GDT because that's faster to | |
786 | switch. */ | |
787 | if (addr <= 0xffffffff) { | |
788 | set_32bit_tls(task, FS_TLS, addr); | |
789 | if (doit) { | |
790 | load_TLS(&task->thread, cpu); | |
a88cde13 | 791 | asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL)); |
1da177e4 LT |
792 | } |
793 | task->thread.fsindex = FS_TLS_SEL; | |
794 | task->thread.fs = 0; | |
795 | } else { | |
796 | task->thread.fsindex = 0; | |
797 | task->thread.fs = addr; | |
798 | if (doit) { | |
799 | /* set the selector to 0 to not confuse | |
800 | __switch_to */ | |
a88cde13 AK |
801 | asm volatile("movl %0,%%fs" :: "r" (0)); |
802 | ret = checking_wrmsrl(MSR_FS_BASE, addr); | |
1da177e4 LT |
803 | } |
804 | } | |
805 | put_cpu(); | |
806 | break; | |
807 | case ARCH_GET_FS: { | |
808 | unsigned long base; | |
809 | if (task->thread.fsindex == FS_TLS_SEL) | |
810 | base = read_32bit_tls(task, FS_TLS); | |
a88cde13 | 811 | else if (doit) |
1da177e4 | 812 | rdmsrl(MSR_FS_BASE, base); |
a88cde13 | 813 | else |
1da177e4 LT |
814 | base = task->thread.fs; |
815 | ret = put_user(base, (unsigned long __user *)addr); | |
816 | break; | |
817 | } | |
818 | case ARCH_GET_GS: { | |
819 | unsigned long base; | |
97c2803c | 820 | unsigned gsindex; |
1da177e4 LT |
821 | if (task->thread.gsindex == GS_TLS_SEL) |
822 | base = read_32bit_tls(task, GS_TLS); | |
97c2803c JB |
823 | else if (doit) { |
824 | asm("movl %%gs,%0" : "=r" (gsindex)); | |
825 | if (gsindex) | |
826 | rdmsrl(MSR_KERNEL_GS_BASE, base); | |
827 | else | |
828 | base = task->thread.gs; | |
829 | } | |
a88cde13 | 830 | else |
1da177e4 LT |
831 | base = task->thread.gs; |
832 | ret = put_user(base, (unsigned long __user *)addr); | |
833 | break; | |
834 | } | |
835 | ||
836 | default: | |
837 | ret = -EINVAL; | |
838 | break; | |
839 | } | |
840 | ||
841 | return ret; | |
842 | } | |
843 | ||
844 | long sys_arch_prctl(int code, unsigned long addr) | |
845 | { | |
846 | return do_arch_prctl(current, code, addr); | |
847 | } | |
848 | ||
849 | /* | |
850 | * Capture the user space registers if the task is not running (in user space) | |
851 | */ | |
852 | int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs) | |
853 | { | |
854 | struct pt_regs *pp, ptregs; | |
855 | ||
bb049232 | 856 | pp = task_pt_regs(tsk); |
1da177e4 LT |
857 | |
858 | ptregs = *pp; | |
859 | ptregs.cs &= 0xffff; | |
860 | ptregs.ss &= 0xffff; | |
861 | ||
862 | elf_core_copy_regs(regs, &ptregs); | |
863 | ||
864 | return 1; | |
865 | } | |
866 | ||
867 | unsigned long arch_align_stack(unsigned long sp) | |
868 | { | |
c16b63e0 | 869 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
1da177e4 LT |
870 | sp -= get_random_int() % 8192; |
871 | return sp & ~0xf; | |
872 | } |