Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  1 /*
  2  *  Copyright (C) 1995  Linus Torvalds
  3  *
  4  *  Pentium III FXSR, SSE support
  5  *      Gareth Hughes <gareth@valinux.com>, May 2000
  6  */
  7 
  8 /*
  9  * This file handles the architecture-dependent parts of process handling..
 10  */
 11 
 12 #include <stdarg.h>
 13 
 14 #include <linux/cpu.h>
 15 #include <linux/errno.h>
 16 #include <linux/sched.h>
 17 #include <linux/fs.h>
 18 #include <linux/kernel.h>
 19 #include <linux/mm.h>
 20 #include <linux/elfcore.h>
 21 #include <linux/smp.h>
 22 #include <linux/stddef.h>
 23 #include <linux/slab.h>
 24 #include <linux/vmalloc.h>
 25 #include <linux/user.h>
 26 #include <linux/interrupt.h>
 27 #include <linux/utsname.h>
 28 #include <linux/delay.h>
 29 #include <linux/reboot.h>
 30 #include <linux/init.h>
 31 #include <linux/mc146818rtc.h>
 32 #include <linux/module.h>
 33 #include <linux/kallsyms.h>
 34 #include <linux/ptrace.h>
 35 #include <linux/random.h>
 36 #include <linux/personality.h>
 37 #include <linux/tick.h>
 38 #include <linux/percpu.h>
 39 
 40 #include <asm/uaccess.h>
 41 #include <asm/pgtable.h>
 42 #include <asm/system.h>
 43 #include <asm/io.h>
 44 #include <asm/ldt.h>
 45 #include <asm/processor.h>
 46 #include <asm/i387.h>
 47 #include <asm/desc.h>
 48 #include <asm/vm86.h>
 49 #ifdef CONFIG_MATH_EMULATION
 50 #include <asm/math_emu.h>
 51 #endif
 52 
 53 #include <linux/err.h>
 54 
 55 #include <asm/tlbflush.h>
 56 #include <asm/cpu.h>
 57 #include <asm/kdebug.h>
 58 
 59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
 60 
 61 static int hlt_counter;
 62 
 63 unsigned long boot_option_idle_override = 0;
 64 EXPORT_SYMBOL(boot_option_idle_override);
 65 
 66 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
 67 EXPORT_PER_CPU_SYMBOL(current_task);
 68 
 69 DEFINE_PER_CPU(int, cpu_number);
 70 EXPORT_PER_CPU_SYMBOL(cpu_number);
 71 
 72 /*
 73  * Return saved PC of a blocked thread.
 74  */
 75 unsigned long thread_saved_pc(struct task_struct *tsk)
 76 {
 77         return ((unsigned long *)tsk->thread.sp)[3];
 78 }
 79 
 80 /*
 81  * Powermanagement idle function, if any..
 82  */
 83 void (*pm_idle)(void);
 84 EXPORT_SYMBOL(pm_idle);
 85 
 86 void disable_hlt(void)
 87 {
 88         hlt_counter++;
 89 }
 90 
 91 EXPORT_SYMBOL(disable_hlt);
 92 
 93 void enable_hlt(void)
 94 {
 95         hlt_counter--;
 96 }
 97 
 98 EXPORT_SYMBOL(enable_hlt);
 99 
100 /*
101  * We use this if we don't have any better
102  * idle routine..
103  */
104 void default_idle(void)
105 {
106         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
107                 current_thread_info()->status &= ~TS_POLLING;
108                 /*
109                  * TS_POLLING-cleared state must be visible before we
110                  * test NEED_RESCHED:
111                  */
112                 smp_mb();
113 
114                 local_irq_disable();
115                 if (!need_resched() && !need_resched_delayed()) {
116                         ktime_t t0, t1;
117                         u64 t0n, t1n;
118 
119                         t0 = ktime_get();
120                         t0n = ktime_to_ns(t0);
121                         safe_halt();    /* enables interrupts racelessly */
122                         local_irq_disable();
123                         t1 = ktime_get();
124                         t1n = ktime_to_ns(t1);
125                         sched_clock_idle_wakeup_event(t1n - t0n);
126                 }
127                 local_irq_enable();
128                 current_thread_info()->status |= TS_POLLING;
129         } else {
130                 /* loop is done by the caller */
131                 cpu_relax();
132         }
133 }
134 #ifdef CONFIG_APM_MODULE
135 EXPORT_SYMBOL(default_idle);
136 #endif
137 
138 /*
139  * On SMP it's slightly faster (but much more power-consuming!)
140  * to poll the ->work.need_resched flag instead of waiting for the
141  * cross-CPU IPI to arrive. Use this option with caution.
142  */
143 static void poll_idle(void)
144 {
145         do {
146                 cpu_relax();
147         } while (!need_resched() && !need_resched_delayed());
148 }
149 
150 #ifdef CONFIG_HOTPLUG_CPU
151 #include <asm/nmi.h>
152 /* We don't actually take CPU down, just spin without interrupts. */
153 static inline void play_dead(void)
154 {
155         /* This must be done before dead CPU ack */
156         cpu_exit_clear();
157         wbinvd();
158         mb();
159         /* Ack it */
160         __get_cpu_var(cpu_state) = CPU_DEAD;
161 
162         /*
163          * With physical CPU hotplug, we should halt the cpu
164          */
165         local_irq_disable();
166         while (1)
167                 halt();
168 }
169 #else
170 static inline void play_dead(void)
171 {
172         BUG();
173 }
174 #endif /* CONFIG_HOTPLUG_CPU */
175 
176 /*
177  * The idle thread. There's no useful work to be
178  * done, so just try to conserve power and have a
179  * low exit latency (ie sit in a loop waiting for
180  * somebody to say that they'd like to reschedule)
181  */
182 void cpu_idle(void)
183 {
184         int cpu = smp_processor_id();
185 
186         current_thread_info()->status |= TS_POLLING;
187 
188         /* endless idle loop with no priority at all */
189         while (1) {
190                 tick_nohz_stop_sched_tick();
191                 while (!need_resched() && !need_resched_delayed()) {
192                         void (*idle)(void);
193 
194                         rmb();
195                         idle = pm_idle;
196 
197                         if (rcu_pending(cpu))
198                                 rcu_check_callbacks(cpu, 0);
199 
200                         if (!idle)
201                                 idle = default_idle;
202 
203                         if (cpu_is_offline(cpu))
204                                 play_dead();
205 
206                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
207                         /* Don't trace irqs off for idle */
208                         stop_critical_timings();
209                         idle();
210                         start_critical_timings();
211                 }
212                 local_irq_disable();
213                 tick_nohz_restart_sched_tick();
214                 __preempt_enable_no_resched();
215                 __schedule();
216                 preempt_disable();
217                 local_irq_enable();
218         }
219 }
220 
221 static void do_nothing(void *unused)
222 {
223 }
224 
225 /*
226  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
227  * pm_idle and update to new pm_idle value. Required while changing pm_idle
228  * handler on SMP systems.
229  *
230  * Caller must have changed pm_idle to the new value before the call. Old
231  * pm_idle value will not be used by any CPU after the return of this function.
232  */
233 void cpu_idle_wait(void)
234 {
235         smp_mb();
236         /* kick all the CPUs so that they exit out of pm_idle */
237         smp_call_function(do_nothing, NULL, 0, 1);
238 }
239 EXPORT_SYMBOL_GPL(cpu_idle_wait);
240 
241 /*
242  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
243  * which can obviate IPI to trigger checking of need_resched.
244  * We execute MONITOR against need_resched and enter optimized wait state
245  * through MWAIT. Whenever someone changes need_resched, we would be woken
246  * up from MWAIT (without an IPI).
247  *
248  * New with Core Duo processors, MWAIT can take some hints based on CPU
249  * capability.
250  */
251 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
252 {
253         if (!need_resched() && !need_resched_delayed()) {
254                 __monitor((void *)&current_thread_info()->flags, 0, 0);
255                 smp_mb();
256                 if (!need_resched() && !need_resched_delayed())
257                         __mwait(ax, cx);
258         }
259 }
260 
261 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
262 static void mwait_idle(void)
263 {
264         local_irq_enable();
265         mwait_idle_with_hints(0, 0);
266 }
267 
268 /*
269  * mwait selection logic:
270  *
271  * It depends on the CPU. For AMD CPUs that support MWAIT this is
272  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
273  * then depend on a clock divisor and current Pstate of the core. If
274  * all cores of a processor are in halt state (C1) the processor can
275  * enter the C1E (C1 enhanced) state. If mwait is used this will never
276  * happen.
277  *
278  * idle=mwait overrides this decision and forces the usage of mwait.
279  */
280 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
281 {
282         if (force_mwait)
283                 return 1;
284 
285         if (c->x86_vendor == X86_VENDOR_AMD) {
286                 switch(c->x86) {
287                 case 0x10:
288                 case 0x11:
289                         return 0;
290                 }
291         }
292         return 1;
293 }
294 
295 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
296 {
297         static int selected;
298 
299         if (selected)
300                 return;
301 #ifdef CONFIG_X86_SMP
302         if (pm_idle == poll_idle && smp_num_siblings > 1) {
303                 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
304                         " performance may degrade.\n");
305         }
306 #endif
307         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
308                 /*
309                  * Skip, if setup has overridden idle.
310                  * One CPU supports mwait => All CPUs supports mwait
311                  */
312                 if (!pm_idle) {
313                         printk(KERN_INFO "using mwait in idle threads.\n");
314                         pm_idle = mwait_idle;
315                 }
316         }
317         selected = 1;
318 }
319 
320 static int __init idle_setup(char *str)
321 {
322         if (!strcmp(str, "poll")) {
323                 printk("using polling idle threads.\n");
324                 pm_idle = poll_idle;
325         } else if (!strcmp(str, "mwait"))
326                 force_mwait = 1;
327         else
328                 return -1;
329 
330         boot_option_idle_override = 1;
331         return 0;
332 }
333 early_param("idle", idle_setup);
334 
335 void __show_registers(struct pt_regs *regs, int all)
336 {
337         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
338         unsigned long d0, d1, d2, d3, d6, d7;
339         unsigned long sp;
340         unsigned short ss, gs;
341 
342         if (user_mode_vm(regs)) {
343                 sp = regs->sp;
344                 ss = regs->ss & 0xffff;
345                 savesegment(gs, gs);
346         } else {
347                 sp = (unsigned long) (&regs->sp);
348                 savesegment(ss, ss);
349                 savesegment(gs, gs);
350         }
351 
352         printk("\n");
353         printk("Pid: %d, comm: %s %s (%s %.*s)\n",
354                         task_pid_nr(current), current->comm,
355                         print_tainted(), init_utsname()->release,
356                         (int)strcspn(init_utsname()->version, " "),
357                         init_utsname()->version);
358 
359         printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
360                         0xffff & regs->cs, regs->ip, regs->flags,
361                         smp_processor_id());
362         print_symbol("EIP is at %s\n", regs->ip);
363 
364         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
365                 regs->ax, regs->bx, regs->cx, regs->dx);
366         printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
367                 regs->si, regs->di, regs->bp, sp);
368         printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x"
369                " preempt:%08x\n",
370                regs->ds & 0xffff, regs->es & 0xffff,
371                regs->fs & 0xffff, gs, ss, preempt_count());
372 
373         if (!all)
374                 return;
375 
376         cr0 = read_cr0();
377         cr2 = read_cr2();
378         cr3 = read_cr3();
379         cr4 = read_cr4_safe();
380         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
381                         cr0, cr2, cr3, cr4);
382 
383         get_debugreg(d0, 0);
384         get_debugreg(d1, 1);
385         get_debugreg(d2, 2);
386         get_debugreg(d3, 3);
387         printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
388                         d0, d1, d2, d3);
389 
390         get_debugreg(d6, 6);
391         get_debugreg(d7, 7);
392         printk("DR6: %08lx DR7: %08lx\n",
393                         d6, d7);
394 }
395 
396 void show_regs(struct pt_regs *regs)
397 {
398         __show_registers(regs, 1);
399         show_trace(NULL, regs, &regs->sp, regs->bp);
400 }
401 
402 /*
403  * This gets run with %bx containing the
404  * function to call, and %dx containing
405  * the "args".
406  */
407 extern void kernel_thread_helper(void);
408 
409 /*
410  * Create a kernel thread
411  */
412 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
413 {
414         struct pt_regs regs;
415 
416         memset(&regs, 0, sizeof(regs));
417 
418         regs.bx = (unsigned long) fn;
419         regs.dx = (unsigned long) arg;
420 
421         regs.ds = __USER_DS;
422         regs.es = __USER_DS;
423         regs.fs = __KERNEL_PERCPU;
424         regs.orig_ax = -1;
425         regs.ip = (unsigned long) kernel_thread_helper;
426         regs.cs = __KERNEL_CS | get_kernel_rpl();
427         regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
428 
429         /* Ok, create the new process.. */
430         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
431 }
432 EXPORT_SYMBOL(kernel_thread);
433 
434 /*
435  * Free current thread data structures etc..
436  */
437 void exit_thread(void)
438 {
439         /* The process may have allocated an io port bitmap... nuke it. */
440         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
441                 struct task_struct *tsk = current;
442                 struct thread_struct *t = &tsk->thread;
443                 void *io_bitmap_ptr = t->io_bitmap_ptr;
444                 int cpu;
445                 struct tss_struct *tss;
446 
447                 /*
448                  * On PREEMPT_RT we must not call kfree() with
449                  * preemption disabled, so we first zap the pointer:
450                  */
451                 t->io_bitmap_ptr = NULL;
452                 kfree(io_bitmap_ptr);
453 
454                 clear_thread_flag(TIF_IO_BITMAP);
455                 /*
456                  * Careful, clear this in the TSS too:
457                  */
458                 cpu = get_cpu();
459                 tss = &per_cpu(init_tss, cpu);
460                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
461                 t->io_bitmap_max = 0;
462                 tss->io_bitmap_owner = NULL;
463                 tss->io_bitmap_max = 0;
464                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
465                 put_cpu();
466         }
467 }
468 
469 void flush_thread(void)
470 {
471         struct task_struct *tsk = current;
472 
473         tsk->thread.debugreg0 = 0;
474         tsk->thread.debugreg1 = 0;
475         tsk->thread.debugreg2 = 0;
476         tsk->thread.debugreg3 = 0;
477         tsk->thread.debugreg6 = 0;
478         tsk->thread.debugreg7 = 0;
479         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));        
480         clear_tsk_thread_flag(tsk, TIF_DEBUG);
481         /*
482          * Forget coprocessor state..
483          */
484         clear_fpu(tsk);
485         clear_used_math();
486 }
487 
488 void release_thread(struct task_struct *dead_task)
489 {
490         BUG_ON(dead_task->mm);
491         release_vm86_irqs(dead_task);
492 }
493 
494 /*
495  * This gets called before we allocate a new thread and copy
496  * the current task into it.
497  */
498 void prepare_to_copy(struct task_struct *tsk)
499 {
500         unlazy_fpu(tsk);
501 }
502 
503 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
504         unsigned long unused,
505         struct task_struct * p, struct pt_regs * regs)
506 {
507         struct pt_regs * childregs;
508         struct task_struct *tsk;
509         int err;
510 
511         childregs = task_pt_regs(p);
512         *childregs = *regs;
513         childregs->ax = 0;
514         childregs->sp = sp;
515 
516         p->thread.sp = (unsigned long) childregs;
517         p->thread.sp0 = (unsigned long) (childregs+1);
518 
519         p->thread.ip = (unsigned long) ret_from_fork;
520 
521         savesegment(gs, p->thread.gs);
522 
523         tsk = current;
524         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
525                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
526                                                 IO_BITMAP_BYTES, GFP_KERNEL);
527                 if (!p->thread.io_bitmap_ptr) {
528                         p->thread.io_bitmap_max = 0;
529                         return -ENOMEM;
530                 }
531                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
532         }
533 
534         err = 0;
535 
536         /*
537          * Set a new TLS for the child thread?
538          */
539         if (clone_flags & CLONE_SETTLS)
540                 err = do_set_thread_area(p, -1,
541                         (struct user_desc __user *)childregs->si, 0);
542 
543         if (err && p->thread.io_bitmap_ptr) {
544                 kfree(p->thread.io_bitmap_ptr);
545                 p->thread.io_bitmap_max = 0;
546         }
547         return err;
548 }
549 
550 #ifdef CONFIG_SECCOMP
551 static void hard_disable_TSC(void)
552 {
553         write_cr4(read_cr4() | X86_CR4_TSD);
554 }
555 void disable_TSC(void)
556 {
557         preempt_disable();
558         if (!test_and_set_thread_flag(TIF_NOTSC))
559                 /*
560                  * Must flip the CPU state synchronously with
561                  * TIF_NOTSC in the current running context.
562                  */
563                 hard_disable_TSC();
564         preempt_enable();
565 }
566 static void hard_enable_TSC(void)
567 {
568         write_cr4(read_cr4() & ~X86_CR4_TSD);
569 }
570 #endif /* CONFIG_SECCOMP */
571 
572 static noinline void
573 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
574                  struct tss_struct *tss)
575 {
576         struct thread_struct *prev, *next;
577         unsigned long debugctl;
578 
579         prev = &prev_p->thread;
580         next = &next_p->thread;
581 
582         debugctl = prev->debugctlmsr;
583         if (next->ds_area_msr != prev->ds_area_msr) {
584                 /* we clear debugctl to make sure DS
585                  * is not in use when we change it */
586                 debugctl = 0;
587                 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
588                 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
589         }
590 
591         if (next->debugctlmsr != debugctl)
592                 wrmsr(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr, 0);
593 
594         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
595                 set_debugreg(next->debugreg0, 0);
596                 set_debugreg(next->debugreg1, 1);
597                 set_debugreg(next->debugreg2, 2);
598                 set_debugreg(next->debugreg3, 3);
599                 /* no 4 and 5 */
600                 set_debugreg(next->debugreg6, 6);
601                 set_debugreg(next->debugreg7, 7);
602         }
603 
604 #ifdef CONFIG_SECCOMP
605         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
606             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
607                 /* prev and next are different */
608                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
609                         hard_disable_TSC();
610                 else
611                         hard_enable_TSC();
612         }
613 #endif
614 
615 #ifdef X86_BTS
616         if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
617                 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
618 
619         if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
620                 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
621 #endif
622 
623 
624         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
625                 /*
626                  * Disable the bitmap via an invalid offset. We still cache
627                  * the previous bitmap owner and the IO bitmap contents:
628                  */
629                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
630                 return;
631         }
632 
633         if (likely(next == tss->io_bitmap_owner)) {
634                 /*
635                  * Previous owner of the bitmap (hence the bitmap content)
636                  * matches the next task, we dont have to do anything but
637                  * to set a valid offset in the TSS:
638                  */
639                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
640                 return;
641         }
642         /*
643          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
644          * and we let the task to get a GPF in case an I/O instruction
645          * is performed.  The handler of the GPF will verify that the
646          * faulting task has a valid I/O bitmap and, it true, does the
647          * real copy and restart the instruction.  This will save us
648          * redundant copies when the currently switched task does not
649          * perform any I/O during its timeslice.
650          */
651         tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
652 }
653 
654 /*
655  *      switch_to(x,yn) should switch tasks from x to y.
656  *
657  * We fsave/fwait so that an exception goes off at the right time
658  * (as a call from the fsave or fwait in effect) rather than to
659  * the wrong process. Lazy FP saving no longer makes any sense
660  * with modern CPU's, and this simplifies a lot of things (SMP
661  * and UP become the same).
662  *
663  * NOTE! We used to use the x86 hardware context switching. The
664  * reason for not using it any more becomes apparent when you
665  * try to recover gracefully from saved state that is no longer
666  * valid (stale segment register values in particular). With the
667  * hardware task-switch, there is no way to fix up bad state in
668  * a reasonable manner.
669  *
670  * The fact that Intel documents the hardware task-switching to
671  * be slow is a fairly red herring - this code is not noticeably
672  * faster. However, there _is_ some room for improvement here,
673  * so the performance issues may eventually be a valid point.
674  * More important, however, is the fact that this allows us much
675  * more flexibility.
676  *
677  * The return value (in %ax) will be the "prev" task after
678  * the task-switch, and shows up in ret_from_fork in entry.S,
679  * for example.
680  */
681 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
682 {
683         struct thread_struct *prev = &prev_p->thread,
684                                  *next = &next_p->thread;
685         int cpu = smp_processor_id();
686         struct tss_struct *tss = &per_cpu(init_tss, cpu);
687 
688         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
689 
690         __unlazy_fpu(prev_p);
691 
692 
693         /* we're going to use this soon, after a few expensive things */
694         if (next_p->fpu_counter > 5)
695                 prefetch(&next->i387.fxsave);
696 
697         /*
698          * Reload esp0.
699          */
700         load_sp0(tss, next);
701 
702         /*
703          * Save away %gs. No need to save %fs, as it was saved on the
704          * stack on entry.  No need to save %es and %ds, as those are
705          * always kernel segments while inside the kernel.  Doing this
706          * before setting the new TLS descriptors avoids the situation
707          * where we temporarily have non-reloadable segments in %fs
708          * and %gs.  This could be an issue if the NMI handler ever
709          * used %fs or %gs (it does not today), or if the kernel is
710          * running inside of a hypervisor layer.
711          */
712         savesegment(gs, prev->gs);
713 
714         /*
715          * Load the per-thread Thread-Local Storage descriptor.
716          */
717         load_TLS(next, cpu);
718 
719         /*
720          * Restore IOPL if needed.  In normal use, the flags restore
721          * in the switch assembly will handle this.  But if the kernel
722          * is running virtualized at a non-zero CPL, the popf will
723          * not restore flags, so it must be done in a separate step.
724          */
725         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
726                 set_iopl_mask(next->iopl);
727 
728         /*
729          * Now maybe handle debug registers and/or IO bitmaps
730          */
731         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
732                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
733                 __switch_to_xtra(prev_p, next_p, tss);
734 
735         /*
736          * Leave lazy mode, flushing any hypercalls made here.
737          * This must be done before restoring TLS segments so
738          * the GDT and LDT are properly updated, and must be
739          * done before math_state_restore, so the TS bit is up
740          * to date.
741          */
742         arch_leave_lazy_cpu_mode();
743 
744         /* If the task has used fpu the last 5 timeslices, just do a full
745          * restore of the math state immediately to avoid the trap; the
746          * chances of needing FPU soon are obviously high now
747          *
748          * tsk_used_math() checks prevent calling math_state_restore(),
749          * which can sleep in the case of !tsk_used_math()
750          */
751         if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
752                 math_state_restore();
753 
754         /*
755          * Restore %gs if needed (which is common)
756          */
757         if (prev->gs | next->gs)
758                 loadsegment(gs, next->gs);
759 
760         x86_write_percpu(current_task, next_p);
761 
762         return prev_p;
763 }
764 
765 asmlinkage int sys_fork(struct pt_regs regs)
766 {
767         return do_fork(SIGCHLD, regs.sp, &regs, 0, NULL, NULL);
768 }
769 
770 asmlinkage int sys_clone(struct pt_regs regs)
771 {
772         unsigned long clone_flags;
773         unsigned long newsp;
774         int __user *parent_tidptr, *child_tidptr;
775 
776         clone_flags = regs.bx;
777         newsp = regs.cx;
778         parent_tidptr = (int __user *)regs.dx;
779         child_tidptr = (int __user *)regs.di;
780         if (!newsp)
781                 newsp = regs.sp;
782         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
783 }
784 
785 /*
786  * This is trivial, and on the face of it looks like it
787  * could equally well be done in user mode.
788  *
789  * Not so, for quite unobvious reasons - register pressure.
790  * In user mode vfork() cannot have a stack frame, and if
791  * done by calling the "clone()" system call directly, you
792  * do not have enough call-clobbered registers to hold all
793  * the information you need.
794  */
795 asmlinkage int sys_vfork(struct pt_regs regs)
796 {
797         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, &regs, 0, NULL, NULL);
798 }
799 
800 /*
801  * sys_execve() executes a new program.
802  */
803 asmlinkage int sys_execve(struct pt_regs regs)
804 {
805         int error;
806         char * filename;
807 
808         filename = getname((char __user *) regs.bx);
809         error = PTR_ERR(filename);
810         if (IS_ERR(filename))
811                 goto out;
812         error = do_execve(filename,
813                         (char __user * __user *) regs.cx,
814                         (char __user * __user *) regs.dx,
815                         &regs);
816         if (error == 0) {
817                 /* Make sure we don't return using sysenter.. */
818                 set_thread_flag(TIF_IRET);
819         }
820         putname(filename);
821 out:
822         return error;
823 }
824 
825 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
826 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
827 
828 unsigned long get_wchan(struct task_struct *p)
829 {
830         unsigned long bp, sp, ip;
831         unsigned long stack_page;
832         int count = 0;
833         if (!p || p == current || p->state == TASK_RUNNING)
834                 return 0;
835         stack_page = (unsigned long)task_stack_page(p);
836         sp = p->thread.sp;
837         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
838                 return 0;
839         /* include/asm-i386/system.h:switch_to() pushes bp last. */
840         bp = *(unsigned long *) sp;
841         do {
842                 if (bp < stack_page || bp > top_ebp+stack_page)
843                         return 0;
844                 ip = *(unsigned long *) (bp+4);
845                 if (!in_sched_functions(ip))
846                         return ip;
847                 bp = *(unsigned long *) bp;
848         } while (count++ < 16);
849         return 0;
850 }
851 
852 unsigned long arch_align_stack(unsigned long sp)
853 {
854         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
855                 sp -= get_random_int() % 8192;
856         return sp & ~0xf;
857 }
858 
859 unsigned long arch_randomize_brk(struct mm_struct *mm)
860 {
861         unsigned long range_end = mm->brk + 0x02000000;
862         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
863 }
864 
  This page was automatically generated by the LXR engine.