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/cpu.h> 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/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/utsname.h> 31 #include <linux/delay.h> 32 #include <linux/reboot.h> 33 #include <linux/init.h> 34 #include <linux/mc146818rtc.h> 35 #include <linux/module.h> 36 #include <linux/kallsyms.h> 37 #include <linux/ptrace.h> 38 #include <linux/random.h> 39 #include <linux/personality.h> 40 #include <linux/tick.h> 41 #include <linux/percpu.h> 42 43 #include <asm/uaccess.h> 44 #include <asm/pgtable.h> 45 #include <asm/system.h> 46 #include <asm/io.h> 47 #include <asm/ldt.h> 48 #include <asm/processor.h> 49 #include <asm/i387.h> 50 #include <asm/desc.h> 51 #include <asm/vm86.h> 52 #ifdef CONFIG_MATH_EMULATION 53 #include <asm/math_emu.h> 54 #endif 55 56 #include <linux/err.h> 57 58 #include <asm/tlbflush.h> 59 #include <asm/cpu.h> 60 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 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; 69 EXPORT_PER_CPU_SYMBOL(current_task); 70 71 DEFINE_PER_CPU(int, cpu_number); 72 EXPORT_PER_CPU_SYMBOL(cpu_number); 73 74 /* 75 * Return saved PC of a blocked thread. 76 */ 77 unsigned long thread_saved_pc(struct task_struct *tsk) 78 { 79 return ((unsigned long *)tsk->thread.esp)[3]; 80 } 81 82 /* 83 * Powermanagement idle function, if any.. 84 */ 85 void (*pm_idle)(void); 86 EXPORT_SYMBOL(pm_idle); 87 static DEFINE_PER_CPU(unsigned int, cpu_idle_state); 88 89 void disable_hlt(void) 90 { 91 hlt_counter++; 92 } 93 94 EXPORT_SYMBOL(disable_hlt); 95 96 void enable_hlt(void) 97 { 98 hlt_counter--; 99 } 100 101 EXPORT_SYMBOL(enable_hlt); 102 103 /* 104 * We use this if we don't have any better 105 * idle routine.. 106 */ 107 void default_idle(void) 108 { 109 if (!hlt_counter && boot_cpu_data.hlt_works_ok) { 110 current_thread_info()->status &= ~TS_POLLING; 111 /* 112 * TS_POLLING-cleared state must be visible before we 113 * test NEED_RESCHED: 114 */ 115 smp_mb(); 116 117 local_irq_disable(); 118 if (!need_resched()) 119 safe_halt(); /* enables interrupts racelessly */ 120 else 121 local_irq_enable(); 122 current_thread_info()->status |= TS_POLLING; 123 } else { 124 /* loop is done by the caller */ 125 cpu_relax(); 126 } 127 } 128 #ifdef CONFIG_APM_MODULE 129 EXPORT_SYMBOL(default_idle); 130 #endif 131 132 /* 133 * On SMP it's slightly faster (but much more power-consuming!) 134 * to poll the ->work.need_resched flag instead of waiting for the 135 * cross-CPU IPI to arrive. Use this option with caution. 136 */ 137 static void poll_idle (void) 138 { 139 cpu_relax(); 140 } 141 142 #ifdef CONFIG_HOTPLUG_CPU 143 #include <asm/nmi.h> 144 /* We don't actually take CPU down, just spin without interrupts. */ 145 static inline void play_dead(void) 146 { 147 /* This must be done before dead CPU ack */ 148 cpu_exit_clear(); 149 wbinvd(); 150 mb(); 151 /* Ack it */ 152 __get_cpu_var(cpu_state) = CPU_DEAD; 153 154 /* 155 * With physical CPU hotplug, we should halt the cpu 156 */ 157 local_irq_disable(); 158 while (1) 159 halt(); 160 } 161 #else 162 static inline void play_dead(void) 163 { 164 BUG(); 165 } 166 #endif /* CONFIG_HOTPLUG_CPU */ 167 168 /* 169 * The idle thread. There's no useful work to be 170 * done, so just try to conserve power and have a 171 * low exit latency (ie sit in a loop waiting for 172 * somebody to say that they'd like to reschedule) 173 */ 174 void cpu_idle(void) 175 { 176 int cpu = smp_processor_id(); 177 178 current_thread_info()->status |= TS_POLLING; 179 180 /* endless idle loop with no priority at all */ 181 while (1) { 182 tick_nohz_stop_sched_tick(); 183 while (!need_resched()) { 184 void (*idle)(void); 185 186 if (__get_cpu_var(cpu_idle_state)) 187 __get_cpu_var(cpu_idle_state) = 0; 188 189 check_pgt_cache(); 190 rmb(); 191 idle = pm_idle; 192 193 if (!idle) 194 idle = default_idle; 195 196 if (cpu_is_offline(cpu)) 197 play_dead(); 198 199 __get_cpu_var(irq_stat).idle_timestamp = jiffies; 200 idle(); 201 } 202 tick_nohz_restart_sched_tick(); 203 preempt_enable_no_resched(); 204 schedule(); 205 preempt_disable(); 206 } 207 } 208 209 void cpu_idle_wait(void) 210 { 211 unsigned int cpu, this_cpu = get_cpu(); 212 cpumask_t map, tmp = current->cpus_allowed; 213 214 set_cpus_allowed(current, cpumask_of_cpu(this_cpu)); 215 put_cpu(); 216 217 cpus_clear(map); 218 for_each_online_cpu(cpu) { 219 per_cpu(cpu_idle_state, cpu) = 1; 220 cpu_set(cpu, map); 221 } 222 223 __get_cpu_var(cpu_idle_state) = 0; 224 225 wmb(); 226 do { 227 ssleep(1); 228 for_each_online_cpu(cpu) { 229 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu)) 230 cpu_clear(cpu, map); 231 } 232 cpus_and(map, map, cpu_online_map); 233 } while (!cpus_empty(map)); 234 235 set_cpus_allowed(current, tmp); 236 } 237 EXPORT_SYMBOL_GPL(cpu_idle_wait); 238 239 /* 240 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI, 241 * which can obviate IPI to trigger checking of need_resched. 242 * We execute MONITOR against need_resched and enter optimized wait state 243 * through MWAIT. Whenever someone changes need_resched, we would be woken 244 * up from MWAIT (without an IPI). 245 * 246 * New with Core Duo processors, MWAIT can take some hints based on CPU 247 * capability. 248 */ 249 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx) 250 { 251 if (!need_resched()) { 252 __monitor((void *)¤t_thread_info()->flags, 0, 0); 253 smp_mb(); 254 if (!need_resched()) 255 __mwait(eax, ecx); 256 } 257 } 258 259 /* Default MONITOR/MWAIT with no hints, used for default C1 state */ 260 static void mwait_idle(void) 261 { 262 local_irq_enable(); 263 mwait_idle_with_hints(0, 0); 264 } 265 266 void __devinit select_idle_routine(const struct cpuinfo_x86 *c) 267 { 268 if (cpu_has(c, X86_FEATURE_MWAIT)) { 269 printk("monitor/mwait feature present.\n"); 270 /* 271 * Skip, if setup has overridden idle. 272 * One CPU supports mwait => All CPUs supports mwait 273 */ 274 if (!pm_idle) { 275 printk("using mwait in idle threads.\n"); 276 pm_idle = mwait_idle; 277 } 278 } 279 } 280 281 static int __init idle_setup(char *str) 282 { 283 if (!strcmp(str, "poll")) { 284 printk("using polling idle threads.\n"); 285 pm_idle = poll_idle; 286 #ifdef CONFIG_X86_SMP 287 if (smp_num_siblings > 1) 288 printk("WARNING: polling idle and HT enabled, performance may degrade.\n"); 289 #endif 290 } else if (!strcmp(str, "mwait")) 291 force_mwait = 1; 292 else 293 return -1; 294 295 boot_option_idle_override = 1; 296 return 0; 297 } 298 early_param("idle", idle_setup); 299 300 void show_regs(struct pt_regs * regs) 301 { 302 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; 303 unsigned long d0, d1, d2, d3, d6, d7; 304 305 printk("\n"); 306 printk("Pid: %d, comm: %20s\n", current->pid, current->comm); 307 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id()); 308 print_symbol("EIP is at %s\n", regs->eip); 309 310 if (user_mode_vm(regs)) 311 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp); 312 printk(" EFLAGS: %08lx %s (%s %.*s)\n", 313 regs->eflags, print_tainted(), init_utsname()->release, 314 (int)strcspn(init_utsname()->version, " "), 315 init_utsname()->version); 316 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", 317 regs->eax,regs->ebx,regs->ecx,regs->edx); 318 printk("ESI: %08lx EDI: %08lx EBP: %08lx", 319 regs->esi, regs->edi, regs->ebp); 320 printk(" DS: %04x ES: %04x FS: %04x\n", 321 0xffff & regs->xds,0xffff & regs->xes, 0xffff & regs->xfs); 322 323 cr0 = read_cr0(); 324 cr2 = read_cr2(); 325 cr3 = read_cr3(); 326 cr4 = read_cr4_safe(); 327 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4); 328 329 get_debugreg(d0, 0); 330 get_debugreg(d1, 1); 331 get_debugreg(d2, 2); 332 get_debugreg(d3, 3); 333 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n", 334 d0, d1, d2, d3); 335 get_debugreg(d6, 6); 336 get_debugreg(d7, 7); 337 printk("DR6: %08lx DR7: %08lx\n", d6, d7); 338 339 show_trace(NULL, regs, ®s->esp); 340 } 341 342 /* 343 * This gets run with %ebx containing the 344 * function to call, and %edx containing 345 * the "args". 346 */ 347 extern void kernel_thread_helper(void); 348 349 /* 350 * Create a kernel thread 351 */ 352 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags) 353 { 354 struct pt_regs regs; 355 356 memset(®s, 0, sizeof(regs)); 357 358 regs.ebx = (unsigned long) fn; 359 regs.edx = (unsigned long) arg; 360 361 regs.xds = __USER_DS; 362 regs.xes = __USER_DS; 363 regs.xfs = __KERNEL_PERCPU; 364 regs.orig_eax = -1; 365 regs.eip = (unsigned long) kernel_thread_helper; 366 regs.xcs = __KERNEL_CS | get_kernel_rpl(); 367 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2; 368 369 /* Ok, create the new process.. */ 370 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL); 371 } 372 EXPORT_SYMBOL(kernel_thread); 373 374 /* 375 * Free current thread data structures etc.. 376 */ 377 void exit_thread(void) 378 { 379 /* The process may have allocated an io port bitmap... nuke it. */ 380 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) { 381 struct task_struct *tsk = current; 382 struct thread_struct *t = &tsk->thread; 383 int cpu = get_cpu(); 384 struct tss_struct *tss = &per_cpu(init_tss, cpu); 385 386 kfree(t->io_bitmap_ptr); 387 t->io_bitmap_ptr = NULL; 388 clear_thread_flag(TIF_IO_BITMAP); 389 /* 390 * Careful, clear this in the TSS too: 391 */ 392 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max); 393 t->io_bitmap_max = 0; 394 tss->io_bitmap_owner = NULL; 395 tss->io_bitmap_max = 0; 396 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET; 397 put_cpu(); 398 } 399 } 400 401 void flush_thread(void) 402 { 403 struct task_struct *tsk = current; 404 405 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8); 406 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); 407 clear_tsk_thread_flag(tsk, TIF_DEBUG); 408 /* 409 * Forget coprocessor state.. 410 */ 411 clear_fpu(tsk); 412 clear_used_math(); 413 } 414 415 void release_thread(struct task_struct *dead_task) 416 { 417 BUG_ON(dead_task->mm); 418 release_vm86_irqs(dead_task); 419 } 420 421 /* 422 * This gets called before we allocate a new thread and copy 423 * the current task into it. 424 */ 425 void prepare_to_copy(struct task_struct *tsk) 426 { 427 unlazy_fpu(tsk); 428 } 429 430 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp, 431 unsigned long unused, 432 struct task_struct * p, struct pt_regs * regs) 433 { 434 struct pt_regs * childregs; 435 struct task_struct *tsk; 436 int err; 437 438 childregs = task_pt_regs(p); 439 *childregs = *regs; 440 childregs->eax = 0; 441 childregs->esp = esp; 442 443 p->thread.esp = (unsigned long) childregs; 444 p->thread.esp0 = (unsigned long) (childregs+1); 445 446 p->thread.eip = (unsigned long) ret_from_fork; 447 448 savesegment(gs,p->thread.gs); 449 450 tsk = current; 451 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { 452 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, 453 IO_BITMAP_BYTES, GFP_KERNEL); 454 if (!p->thread.io_bitmap_ptr) { 455 p->thread.io_bitmap_max = 0; 456 return -ENOMEM; 457 } 458 set_tsk_thread_flag(p, TIF_IO_BITMAP); 459 } 460 461 /* 462 * Set a new TLS for the child thread? 463 */ 464 if (clone_flags & CLONE_SETTLS) { 465 struct desc_struct *desc; 466 struct user_desc info; 467 int idx; 468 469 err = -EFAULT; 470 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info))) 471 goto out; 472 err = -EINVAL; 473 if (LDT_empty(&info)) 474 goto out; 475 476 idx = info.entry_number; 477 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) 478 goto out; 479 480 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; 481 desc->a = LDT_entry_a(&info); 482 desc->b = LDT_entry_b(&info); 483 } 484 485 err = 0; 486 out: 487 if (err && p->thread.io_bitmap_ptr) { 488 kfree(p->thread.io_bitmap_ptr); 489 p->thread.io_bitmap_max = 0; 490 } 491 return err; 492 } 493 494 /* 495 * fill in the user structure for a core dump.. 496 */ 497 void dump_thread(struct pt_regs * regs, struct user * dump) 498 { 499 int i; 500 501 /* changed the size calculations - should hopefully work better. lbt */ 502 dump->magic = CMAGIC; 503 dump->start_code = 0; 504 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1); 505 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT; 506 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT; 507 dump->u_dsize -= dump->u_tsize; 508 dump->u_ssize = 0; 509 for (i = 0; i < 8; i++) 510 dump->u_debugreg[i] = current->thread.debugreg[i]; 511 512 if (dump->start_stack < TASK_SIZE) 513 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT; 514 515 dump->regs.ebx = regs->ebx; 516 dump->regs.ecx = regs->ecx; 517 dump->regs.edx = regs->edx; 518 dump->regs.esi = regs->esi; 519 dump->regs.edi = regs->edi; 520 dump->regs.ebp = regs->ebp; 521 dump->regs.eax = regs->eax; 522 dump->regs.ds = regs->xds; 523 dump->regs.es = regs->xes; 524 dump->regs.fs = regs->xfs; 525 savesegment(gs,dump->regs.gs); 526 dump->regs.orig_eax = regs->orig_eax; 527 dump->regs.eip = regs->eip; 528 dump->regs.cs = regs->xcs; 529 dump->regs.eflags = regs->eflags; 530 dump->regs.esp = regs->esp; 531 dump->regs.ss = regs->xss; 532 533 dump->u_fpvalid = dump_fpu (regs, &dump->i387); 534 } 535 EXPORT_SYMBOL(dump_thread); 536 537 /* 538 * Capture the user space registers if the task is not running (in user space) 539 */ 540 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs) 541 { 542 struct pt_regs ptregs = *task_pt_regs(tsk); 543 ptregs.xcs &= 0xffff; 544 ptregs.xds &= 0xffff; 545 ptregs.xes &= 0xffff; 546 ptregs.xss &= 0xffff; 547 548 elf_core_copy_regs(regs, &ptregs); 549 550 return 1; 551 } 552 553 #ifdef CONFIG_SECCOMP 554 void hard_disable_TSC(void) 555 { 556 write_cr4(read_cr4() | X86_CR4_TSD); 557 } 558 void disable_TSC(void) 559 { 560 preempt_disable(); 561 if (!test_and_set_thread_flag(TIF_NOTSC)) 562 /* 563 * Must flip the CPU state synchronously with 564 * TIF_NOTSC in the current running context. 565 */ 566 hard_disable_TSC(); 567 preempt_enable(); 568 } 569 void hard_enable_TSC(void) 570 { 571 write_cr4(read_cr4() & ~X86_CR4_TSD); 572 } 573 #endif /* CONFIG_SECCOMP */ 574 575 static noinline void 576 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, 577 struct tss_struct *tss) 578 { 579 struct thread_struct *next; 580 581 next = &next_p->thread; 582 583 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) { 584 set_debugreg(next->debugreg[0], 0); 585 set_debugreg(next->debugreg[1], 1); 586 set_debugreg(next->debugreg[2], 2); 587 set_debugreg(next->debugreg[3], 3); 588 /* no 4 and 5 */ 589 set_debugreg(next->debugreg[6], 6); 590 set_debugreg(next->debugreg[7], 7); 591 } 592 593 #ifdef CONFIG_SECCOMP 594 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^ 595 test_tsk_thread_flag(next_p, TIF_NOTSC)) { 596 /* prev and next are different */ 597 if (test_tsk_thread_flag(next_p, TIF_NOTSC)) 598 hard_disable_TSC(); 599 else 600 hard_enable_TSC(); 601 } 602 #endif 603 604 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) { 605 /* 606 * Disable the bitmap via an invalid offset. We still cache 607 * the previous bitmap owner and the IO bitmap contents: 608 */ 609 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET; 610 return; 611 } 612 613 if (likely(next == tss->io_bitmap_owner)) { 614 /* 615 * Previous owner of the bitmap (hence the bitmap content) 616 * matches the next task, we dont have to do anything but 617 * to set a valid offset in the TSS: 618 */ 619 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET; 620 return; 621 } 622 /* 623 * Lazy TSS's I/O bitmap copy. We set an invalid offset here 624 * and we let the task to get a GPF in case an I/O instruction 625 * is performed. The handler of the GPF will verify that the 626 * faulting task has a valid I/O bitmap and, it true, does the 627 * real copy and restart the instruction. This will save us 628 * redundant copies when the currently switched task does not 629 * perform any I/O during its timeslice. 630 */ 631 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY; 632 } 633 634 /* 635 * switch_to(x,yn) should switch tasks from x to y. 636 * 637 * We fsave/fwait so that an exception goes off at the right time 638 * (as a call from the fsave or fwait in effect) rather than to 639 * the wrong process. Lazy FP saving no longer makes any sense 640 * with modern CPU's, and this simplifies a lot of things (SMP 641 * and UP become the same). 642 * 643 * NOTE! We used to use the x86 hardware context switching. The 644 * reason for not using it any more becomes apparent when you 645 * try to recover gracefully from saved state that is no longer 646 * valid (stale segment register values in particular). With the 647 * hardware task-switch, there is no way to fix up bad state in 648 * a reasonable manner. 649 * 650 * The fact that Intel documents the hardware task-switching to 651 * be slow is a fairly red herring - this code is not noticeably 652 * faster. However, there _is_ some room for improvement here, 653 * so the performance issues may eventually be a valid point. 654 * More important, however, is the fact that this allows us much 655 * more flexibility. 656 * 657 * The return value (in %eax) will be the "prev" task after 658 * the task-switch, and shows up in ret_from_fork in entry.S, 659 * for example. 660 */ 661 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 662 { 663 struct thread_struct *prev = &prev_p->thread, 664 *next = &next_p->thread; 665 int cpu = smp_processor_id(); 666 struct tss_struct *tss = &per_cpu(init_tss, cpu); 667 668 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ 669 670 __unlazy_fpu(prev_p); 671 672 673 /* we're going to use this soon, after a few expensive things */ 674 if (next_p->fpu_counter > 5) 675 prefetch(&next->i387.fxsave); 676 677 /* 678 * Reload esp0. 679 */ 680 load_esp0(tss, next); 681 682 /* 683 * Save away %gs. No need to save %fs, as it was saved on the 684 * stack on entry. No need to save %es and %ds, as those are 685 * always kernel segments while inside the kernel. Doing this 686 * before setting the new TLS descriptors avoids the situation 687 * where we temporarily have non-reloadable segments in %fs 688 * and %gs. This could be an issue if the NMI handler ever 689 * used %fs or %gs (it does not today), or if the kernel is 690 * running inside of a hypervisor layer. 691 */ 692 savesegment(gs, prev->gs); 693 694 /* 695 * Load the per-thread Thread-Local Storage descriptor. 696 */ 697 load_TLS(next, cpu); 698 699 /* 700 * Restore IOPL if needed. In normal use, the flags restore 701 * in the switch assembly will handle this. But if the kernel 702 * is running virtualized at a non-zero CPL, the popf will 703 * not restore flags, so it must be done in a separate step. 704 */ 705 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl)) 706 set_iopl_mask(next->iopl); 707 708 /* 709 * Now maybe handle debug registers and/or IO bitmaps 710 */ 711 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV || 712 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT)) 713 __switch_to_xtra(prev_p, next_p, tss); 714 715 /* 716 * Leave lazy mode, flushing any hypercalls made here. 717 * This must be done before restoring TLS segments so 718 * the GDT and LDT are properly updated, and must be 719 * done before math_state_restore, so the TS bit is up 720 * to date. 721 */ 722 arch_leave_lazy_cpu_mode(); 723 724 /* If the task has used fpu the last 5 timeslices, just do a full 725 * restore of the math state immediately to avoid the trap; the 726 * chances of needing FPU soon are obviously high now 727 */ 728 if (next_p->fpu_counter > 5) 729 math_state_restore(); 730 731 /* 732 * Restore %gs if needed (which is common) 733 */ 734 if (prev->gs | next->gs) 735 loadsegment(gs, next->gs); 736 737 x86_write_percpu(current_task, next_p); 738 739 return prev_p; 740 } 741 742 asmlinkage int sys_fork(struct pt_regs regs) 743 { 744 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL); 745 } 746 747 asmlinkage int sys_clone(struct pt_regs regs) 748 { 749 unsigned long clone_flags; 750 unsigned long newsp; 751 int __user *parent_tidptr, *child_tidptr; 752 753 clone_flags = regs.ebx; 754 newsp = regs.ecx; 755 parent_tidptr = (int __user *)regs.edx; 756 child_tidptr = (int __user *)regs.edi; 757 if (!newsp) 758 newsp = regs.esp; 759 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr); 760 } 761 762 /* 763 * This is trivial, and on the face of it looks like it 764 * could equally well be done in user mode. 765 * 766 * Not so, for quite unobvious reasons - register pressure. 767 * In user mode vfork() cannot have a stack frame, and if 768 * done by calling the "clone()" system call directly, you 769 * do not have enough call-clobbered registers to hold all 770 * the information you need. 771 */ 772 asmlinkage int sys_vfork(struct pt_regs regs) 773 { 774 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL); 775 } 776 777 /* 778 * sys_execve() executes a new program. 779 */ 780 asmlinkage int sys_execve(struct pt_regs regs) 781 { 782 int error; 783 char * filename; 784 785 filename = getname((char __user *) regs.ebx); 786 error = PTR_ERR(filename); 787 if (IS_ERR(filename)) 788 goto out; 789 error = do_execve(filename, 790 (char __user * __user *) regs.ecx, 791 (char __user * __user *) regs.edx, 792 ®s); 793 if (error == 0) { 794 task_lock(current); 795 current->ptrace &= ~PT_DTRACE; 796 task_unlock(current); 797 /* Make sure we don't return using sysenter.. */ 798 set_thread_flag(TIF_IRET); 799 } 800 putname(filename); 801 out: 802 return error; 803 } 804 805 #define top_esp (THREAD_SIZE - sizeof(unsigned long)) 806 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) 807 808 unsigned long get_wchan(struct task_struct *p) 809 { 810 unsigned long ebp, esp, eip; 811 unsigned long stack_page; 812 int count = 0; 813 if (!p || p == current || p->state == TASK_RUNNING) 814 return 0; 815 stack_page = (unsigned long)task_stack_page(p); 816 esp = p->thread.esp; 817 if (!stack_page || esp < stack_page || esp > top_esp+stack_page) 818 return 0; 819 /* include/asm-i386/system.h:switch_to() pushes ebp last. */ 820 ebp = *(unsigned long *) esp; 821 do { 822 if (ebp < stack_page || ebp > top_ebp+stack_page) 823 return 0; 824 eip = *(unsigned long *) (ebp+4); 825 if (!in_sched_functions(eip)) 826 return eip; 827 ebp = *(unsigned long *) ebp; 828 } while (count++ < 16); 829 return 0; 830 } 831 832 /* 833 * sys_alloc_thread_area: get a yet unused TLS descriptor index. 834 */ 835 static int get_free_idx(void) 836 { 837 struct thread_struct *t = ¤t->thread; 838 int idx; 839 840 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++) 841 if (desc_empty(t->tls_array + idx)) 842 return idx + GDT_ENTRY_TLS_MIN; 843 return -ESRCH; 844 } 845 846 /* 847 * Set a given TLS descriptor: 848 */ 849 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info) 850 { 851 struct thread_struct *t = ¤t->thread; 852 struct user_desc info; 853 struct desc_struct *desc; 854 int cpu, idx; 855 856 if (copy_from_user(&info, u_info, sizeof(info))) 857 return -EFAULT; 858 idx = info.entry_number; 859 860 /* 861 * index -1 means the kernel should try to find and 862 * allocate an empty descriptor: 863 */ 864 if (idx == -1) { 865 idx = get_free_idx(); 866 if (idx < 0) 867 return idx; 868 if (put_user(idx, &u_info->entry_number)) 869 return -EFAULT; 870 } 871 872 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) 873 return -EINVAL; 874 875 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN; 876 877 /* 878 * We must not get preempted while modifying the TLS. 879 */ 880 cpu = get_cpu(); 881 882 if (LDT_empty(&info)) { 883 desc->a = 0; 884 desc->b = 0; 885 } else { 886 desc->a = LDT_entry_a(&info); 887 desc->b = LDT_entry_b(&info); 888 } 889 load_TLS(t, cpu); 890 891 put_cpu(); 892 893 return 0; 894 } 895 896 /* 897 * Get the current Thread-Local Storage area: 898 */ 899 900 #define GET_BASE(desc) ( \ 901 (((desc)->a >> 16) & 0x0000ffff) | \ 902 (((desc)->b << 16) & 0x00ff0000) | \ 903 ( (desc)->b & 0xff000000) ) 904 905 #define GET_LIMIT(desc) ( \ 906 ((desc)->a & 0x0ffff) | \ 907 ((desc)->b & 0xf0000) ) 908 909 #define GET_32BIT(desc) (((desc)->b >> 22) & 1) 910 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3) 911 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1) 912 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1) 913 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1) 914 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1) 915 916 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info) 917 { 918 struct user_desc info; 919 struct desc_struct *desc; 920 int idx; 921 922 if (get_user(idx, &u_info->entry_number)) 923 return -EFAULT; 924 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX) 925 return -EINVAL; 926 927 memset(&info, 0, sizeof(info)); 928 929 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN; 930 931 info.entry_number = idx; 932 info.base_addr = GET_BASE(desc); 933 info.limit = GET_LIMIT(desc); 934 info.seg_32bit = GET_32BIT(desc); 935 info.contents = GET_CONTENTS(desc); 936 info.read_exec_only = !GET_WRITABLE(desc); 937 info.limit_in_pages = GET_LIMIT_PAGES(desc); 938 info.seg_not_present = !GET_PRESENT(desc); 939 info.useable = GET_USEABLE(desc); 940 941 if (copy_to_user(u_info, &info, sizeof(info))) 942 return -EFAULT; 943 return 0; 944 } 945 946 unsigned long arch_align_stack(unsigned long sp) 947 { 948 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) 949 sp -= get_random_int() % 8192; 950 return sp & ~0xf; 951 } 952