2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * This file handles the architecture-dependent parts of process handling..
14 #include <linux/cpu.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
18 #include <linux/kernel.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 #include <linux/prctl.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
46 #include <asm/processor.h>
49 #ifdef CONFIG_MATH_EMULATION
50 #include <asm/math_emu.h>
53 #include <linux/err.h>
55 #include <asm/tlbflush.h>
57 #include <asm/kdebug.h>
59 asmlinkage
void ret_from_fork(void) __asm__("ret_from_fork");
61 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
62 EXPORT_PER_CPU_SYMBOL(current_task
);
64 DEFINE_PER_CPU(int, cpu_number
);
65 EXPORT_PER_CPU_SYMBOL(cpu_number
);
68 * Return saved PC of a blocked thread.
70 unsigned long thread_saved_pc(struct task_struct
*tsk
)
72 return ((unsigned long *)tsk
->thread
.sp
)[3];
75 #ifdef CONFIG_HOTPLUG_CPU
77 /* We don't actually take CPU down, just spin without interrupts. */
78 static inline void play_dead(void)
80 /* This must be done before dead CPU ack */
85 __get_cpu_var(cpu_state
) = CPU_DEAD
;
88 * With physical CPU hotplug, we should halt the cpu
95 static inline void play_dead(void)
99 #endif /* CONFIG_HOTPLUG_CPU */
102 * The idle thread. There's no useful work to be
103 * done, so just try to conserve power and have a
104 * low exit latency (ie sit in a loop waiting for
105 * somebody to say that they'd like to reschedule)
109 int cpu
= smp_processor_id();
111 current_thread_info()->status
|= TS_POLLING
;
113 /* endless idle loop with no priority at all */
115 tick_nohz_stop_sched_tick();
116 while (!need_resched()) {
121 if (rcu_pending(cpu
))
122 rcu_check_callbacks(cpu
, 0);
124 if (cpu_is_offline(cpu
))
128 __get_cpu_var(irq_stat
).idle_timestamp
= jiffies
;
131 tick_nohz_restart_sched_tick();
132 preempt_enable_no_resched();
138 void __show_registers(struct pt_regs
*regs
, int all
)
140 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
141 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
143 unsigned short ss
, gs
;
145 if (user_mode_vm(regs
)) {
147 ss
= regs
->ss
& 0xffff;
150 sp
= (unsigned long) (®s
->sp
);
156 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
157 task_pid_nr(current
), current
->comm
,
158 print_tainted(), init_utsname()->release
,
159 (int)strcspn(init_utsname()->version
, " "),
160 init_utsname()->version
);
162 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
163 (u16
)regs
->cs
, regs
->ip
, regs
->flags
,
165 print_symbol("EIP is at %s\n", regs
->ip
);
167 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
168 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
169 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
170 regs
->si
, regs
->di
, regs
->bp
, sp
);
171 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
172 (u16
)regs
->ds
, (u16
)regs
->es
, (u16
)regs
->fs
, gs
, ss
);
180 cr4
= read_cr4_safe();
181 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
188 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
193 printk("DR6: %08lx DR7: %08lx\n",
197 void show_regs(struct pt_regs
*regs
)
199 __show_registers(regs
, 1);
200 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
204 * This gets run with %bx containing the
205 * function to call, and %dx containing
208 extern void kernel_thread_helper(void);
211 * Create a kernel thread
213 int kernel_thread(int (*fn
)(void *), void * arg
, unsigned long flags
)
217 memset(®s
, 0, sizeof(regs
));
219 regs
.bx
= (unsigned long) fn
;
220 regs
.dx
= (unsigned long) arg
;
224 regs
.fs
= __KERNEL_PERCPU
;
226 regs
.ip
= (unsigned long) kernel_thread_helper
;
227 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
228 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
230 /* Ok, create the new process.. */
231 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
233 EXPORT_SYMBOL(kernel_thread
);
236 * Free current thread data structures etc..
238 void exit_thread(void)
240 /* The process may have allocated an io port bitmap... nuke it. */
241 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
242 struct task_struct
*tsk
= current
;
243 struct thread_struct
*t
= &tsk
->thread
;
245 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
247 kfree(t
->io_bitmap_ptr
);
248 t
->io_bitmap_ptr
= NULL
;
249 clear_thread_flag(TIF_IO_BITMAP
);
251 * Careful, clear this in the TSS too:
253 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
254 t
->io_bitmap_max
= 0;
255 tss
->io_bitmap_owner
= NULL
;
256 tss
->io_bitmap_max
= 0;
257 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
262 void flush_thread(void)
264 struct task_struct
*tsk
= current
;
266 tsk
->thread
.debugreg0
= 0;
267 tsk
->thread
.debugreg1
= 0;
268 tsk
->thread
.debugreg2
= 0;
269 tsk
->thread
.debugreg3
= 0;
270 tsk
->thread
.debugreg6
= 0;
271 tsk
->thread
.debugreg7
= 0;
272 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
273 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
275 * Forget coprocessor state..
277 tsk
->fpu_counter
= 0;
282 void release_thread(struct task_struct
*dead_task
)
284 BUG_ON(dead_task
->mm
);
285 release_vm86_irqs(dead_task
);
289 * This gets called before we allocate a new thread and copy
290 * the current task into it.
292 void prepare_to_copy(struct task_struct
*tsk
)
297 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
298 unsigned long unused
,
299 struct task_struct
* p
, struct pt_regs
* regs
)
301 struct pt_regs
* childregs
;
302 struct task_struct
*tsk
;
305 childregs
= task_pt_regs(p
);
310 p
->thread
.sp
= (unsigned long) childregs
;
311 p
->thread
.sp0
= (unsigned long) (childregs
+1);
313 p
->thread
.ip
= (unsigned long) ret_from_fork
;
315 savesegment(gs
, p
->thread
.gs
);
318 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
319 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
320 IO_BITMAP_BYTES
, GFP_KERNEL
);
321 if (!p
->thread
.io_bitmap_ptr
) {
322 p
->thread
.io_bitmap_max
= 0;
325 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
331 * Set a new TLS for the child thread?
333 if (clone_flags
& CLONE_SETTLS
)
334 err
= do_set_thread_area(p
, -1,
335 (struct user_desc __user
*)childregs
->si
, 0);
337 if (err
&& p
->thread
.io_bitmap_ptr
) {
338 kfree(p
->thread
.io_bitmap_ptr
);
339 p
->thread
.io_bitmap_max
= 0;
345 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
347 __asm__("movl %0, %%gs" :: "r"(0));
350 regs
->ds
= __USER_DS
;
351 regs
->es
= __USER_DS
;
352 regs
->ss
= __USER_DS
;
353 regs
->cs
= __USER_CS
;
357 * Free the old FP and other extended state
359 free_thread_xstate(current
);
361 EXPORT_SYMBOL_GPL(start_thread
);
363 static void hard_disable_TSC(void)
365 write_cr4(read_cr4() | X86_CR4_TSD
);
368 void disable_TSC(void)
371 if (!test_and_set_thread_flag(TIF_NOTSC
))
373 * Must flip the CPU state synchronously with
374 * TIF_NOTSC in the current running context.
380 static void hard_enable_TSC(void)
382 write_cr4(read_cr4() & ~X86_CR4_TSD
);
385 static void enable_TSC(void)
388 if (test_and_clear_thread_flag(TIF_NOTSC
))
390 * Must flip the CPU state synchronously with
391 * TIF_NOTSC in the current running context.
397 int get_tsc_mode(unsigned long adr
)
401 if (test_thread_flag(TIF_NOTSC
))
402 val
= PR_TSC_SIGSEGV
;
406 return put_user(val
, (unsigned int __user
*)adr
);
409 int set_tsc_mode(unsigned int val
)
411 if (val
== PR_TSC_SIGSEGV
)
413 else if (val
== PR_TSC_ENABLE
)
422 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
423 struct tss_struct
*tss
)
425 struct thread_struct
*prev
, *next
;
426 unsigned long debugctl
;
428 prev
= &prev_p
->thread
;
429 next
= &next_p
->thread
;
431 debugctl
= prev
->debugctlmsr
;
432 if (next
->ds_area_msr
!= prev
->ds_area_msr
) {
433 /* we clear debugctl to make sure DS
434 * is not in use when we change it */
436 update_debugctlmsr(0);
437 wrmsr(MSR_IA32_DS_AREA
, next
->ds_area_msr
, 0);
440 if (next
->debugctlmsr
!= debugctl
)
441 update_debugctlmsr(next
->debugctlmsr
);
443 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
444 set_debugreg(next
->debugreg0
, 0);
445 set_debugreg(next
->debugreg1
, 1);
446 set_debugreg(next
->debugreg2
, 2);
447 set_debugreg(next
->debugreg3
, 3);
449 set_debugreg(next
->debugreg6
, 6);
450 set_debugreg(next
->debugreg7
, 7);
453 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
454 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
455 /* prev and next are different */
456 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
463 if (test_tsk_thread_flag(prev_p
, TIF_BTS_TRACE_TS
))
464 ptrace_bts_take_timestamp(prev_p
, BTS_TASK_DEPARTS
);
466 if (test_tsk_thread_flag(next_p
, TIF_BTS_TRACE_TS
))
467 ptrace_bts_take_timestamp(next_p
, BTS_TASK_ARRIVES
);
471 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
473 * Disable the bitmap via an invalid offset. We still cache
474 * the previous bitmap owner and the IO bitmap contents:
476 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
480 if (likely(next
== tss
->io_bitmap_owner
)) {
482 * Previous owner of the bitmap (hence the bitmap content)
483 * matches the next task, we dont have to do anything but
484 * to set a valid offset in the TSS:
486 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
490 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
491 * and we let the task to get a GPF in case an I/O instruction
492 * is performed. The handler of the GPF will verify that the
493 * faulting task has a valid I/O bitmap and, it true, does the
494 * real copy and restart the instruction. This will save us
495 * redundant copies when the currently switched task does not
496 * perform any I/O during its timeslice.
498 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
502 * switch_to(x,yn) should switch tasks from x to y.
504 * We fsave/fwait so that an exception goes off at the right time
505 * (as a call from the fsave or fwait in effect) rather than to
506 * the wrong process. Lazy FP saving no longer makes any sense
507 * with modern CPU's, and this simplifies a lot of things (SMP
508 * and UP become the same).
510 * NOTE! We used to use the x86 hardware context switching. The
511 * reason for not using it any more becomes apparent when you
512 * try to recover gracefully from saved state that is no longer
513 * valid (stale segment register values in particular). With the
514 * hardware task-switch, there is no way to fix up bad state in
515 * a reasonable manner.
517 * The fact that Intel documents the hardware task-switching to
518 * be slow is a fairly red herring - this code is not noticeably
519 * faster. However, there _is_ some room for improvement here,
520 * so the performance issues may eventually be a valid point.
521 * More important, however, is the fact that this allows us much
524 * The return value (in %ax) will be the "prev" task after
525 * the task-switch, and shows up in ret_from_fork in entry.S,
528 struct task_struct
* __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
530 struct thread_struct
*prev
= &prev_p
->thread
,
531 *next
= &next_p
->thread
;
532 int cpu
= smp_processor_id();
533 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
535 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
537 __unlazy_fpu(prev_p
);
540 /* we're going to use this soon, after a few expensive things */
541 if (next_p
->fpu_counter
> 5)
542 prefetch(next
->xstate
);
550 * Save away %gs. No need to save %fs, as it was saved on the
551 * stack on entry. No need to save %es and %ds, as those are
552 * always kernel segments while inside the kernel. Doing this
553 * before setting the new TLS descriptors avoids the situation
554 * where we temporarily have non-reloadable segments in %fs
555 * and %gs. This could be an issue if the NMI handler ever
556 * used %fs or %gs (it does not today), or if the kernel is
557 * running inside of a hypervisor layer.
559 savesegment(gs
, prev
->gs
);
562 * Load the per-thread Thread-Local Storage descriptor.
567 * Restore IOPL if needed. In normal use, the flags restore
568 * in the switch assembly will handle this. But if the kernel
569 * is running virtualized at a non-zero CPL, the popf will
570 * not restore flags, so it must be done in a separate step.
572 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
573 set_iopl_mask(next
->iopl
);
576 * Now maybe handle debug registers and/or IO bitmaps
578 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
579 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
580 __switch_to_xtra(prev_p
, next_p
, tss
);
583 * Leave lazy mode, flushing any hypercalls made here.
584 * This must be done before restoring TLS segments so
585 * the GDT and LDT are properly updated, and must be
586 * done before math_state_restore, so the TS bit is up
589 arch_leave_lazy_cpu_mode();
591 /* If the task has used fpu the last 5 timeslices, just do a full
592 * restore of the math state immediately to avoid the trap; the
593 * chances of needing FPU soon are obviously high now
595 * tsk_used_math() checks prevent calling math_state_restore(),
596 * which can sleep in the case of !tsk_used_math()
598 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
599 math_state_restore();
602 * Restore %gs if needed (which is common)
604 if (prev
->gs
| next
->gs
)
605 loadsegment(gs
, next
->gs
);
607 x86_write_percpu(current_task
, next_p
);
612 asmlinkage
int sys_fork(struct pt_regs regs
)
614 return do_fork(SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
617 asmlinkage
int sys_clone(struct pt_regs regs
)
619 unsigned long clone_flags
;
621 int __user
*parent_tidptr
, *child_tidptr
;
623 clone_flags
= regs
.bx
;
625 parent_tidptr
= (int __user
*)regs
.dx
;
626 child_tidptr
= (int __user
*)regs
.di
;
629 return do_fork(clone_flags
, newsp
, ®s
, 0, parent_tidptr
, child_tidptr
);
633 * This is trivial, and on the face of it looks like it
634 * could equally well be done in user mode.
636 * Not so, for quite unobvious reasons - register pressure.
637 * In user mode vfork() cannot have a stack frame, and if
638 * done by calling the "clone()" system call directly, you
639 * do not have enough call-clobbered registers to hold all
640 * the information you need.
642 asmlinkage
int sys_vfork(struct pt_regs regs
)
644 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
648 * sys_execve() executes a new program.
650 asmlinkage
int sys_execve(struct pt_regs regs
)
655 filename
= getname((char __user
*) regs
.bx
);
656 error
= PTR_ERR(filename
);
657 if (IS_ERR(filename
))
659 error
= do_execve(filename
,
660 (char __user
* __user
*) regs
.cx
,
661 (char __user
* __user
*) regs
.dx
,
664 /* Make sure we don't return using sysenter.. */
665 set_thread_flag(TIF_IRET
);
672 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
673 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
675 unsigned long get_wchan(struct task_struct
*p
)
677 unsigned long bp
, sp
, ip
;
678 unsigned long stack_page
;
680 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
682 stack_page
= (unsigned long)task_stack_page(p
);
684 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
686 /* include/asm-i386/system.h:switch_to() pushes bp last. */
687 bp
= *(unsigned long *) sp
;
689 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
691 ip
= *(unsigned long *) (bp
+4);
692 if (!in_sched_functions(ip
))
694 bp
= *(unsigned long *) bp
;
695 } while (count
++ < 16);
699 unsigned long arch_align_stack(unsigned long sp
)
701 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
702 sp
-= get_random_int() % 8192;
706 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
708 unsigned long range_end
= mm
->brk
+ 0x02000000;
709 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;