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>
40 #include <linux/dmi.h>
41 #include <linux/ftrace.h>
42 #include <linux/uaccess.h>
44 #include <linux/kdebug.h>
46 #include <asm/pgtable.h>
47 #include <asm/system.h>
49 #include <asm/processor.h>
52 #ifdef CONFIG_MATH_EMULATION
53 #include <asm/math_emu.h>
56 #include <linux/err.h>
58 #include <asm/tlbflush.h>
61 #include <asm/syscalls.h>
64 asmlinkage
void ret_from_fork(void) __asm__("ret_from_fork");
66 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
67 EXPORT_PER_CPU_SYMBOL(current_task
);
69 DEFINE_PER_CPU(int, cpu_number
);
70 EXPORT_PER_CPU_SYMBOL(cpu_number
);
73 * Return saved PC of a blocked thread.
75 unsigned long thread_saved_pc(struct task_struct
*tsk
)
77 return ((unsigned long *)tsk
->thread
.sp
)[3];
81 static inline void play_dead(void)
88 * The idle thread. There's no useful work to be
89 * done, so just try to conserve power and have a
90 * low exit latency (ie sit in a loop waiting for
91 * somebody to say that they'd like to reschedule)
95 int cpu
= smp_processor_id();
97 current_thread_info()->status
|= TS_POLLING
;
99 /* endless idle loop with no priority at all */
101 tick_nohz_stop_sched_tick(1);
102 while (!need_resched()) {
107 if (rcu_pending(cpu
))
108 rcu_check_callbacks(cpu
, 0);
110 if (cpu_is_offline(cpu
))
114 __get_cpu_var(irq_stat
).idle_timestamp
= jiffies
;
115 /* Don't trace irqs off for idle */
116 stop_critical_timings();
118 start_critical_timings();
120 tick_nohz_restart_sched_tick();
121 preempt_enable_no_resched();
127 void __show_regs(struct pt_regs
*regs
, int all
)
129 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
130 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
132 unsigned short ss
, gs
;
135 if (user_mode_vm(regs
)) {
137 ss
= regs
->ss
& 0xffff;
140 sp
= (unsigned long) (®s
->sp
);
147 board
= dmi_get_system_info(DMI_PRODUCT_NAME
);
150 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
151 task_pid_nr(current
), current
->comm
,
152 print_tainted(), init_utsname()->release
,
153 (int)strcspn(init_utsname()->version
, " "),
154 init_utsname()->version
, board
);
156 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
157 (u16
)regs
->cs
, regs
->ip
, regs
->flags
,
159 print_symbol("EIP is at %s\n", regs
->ip
);
161 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
162 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
163 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
164 regs
->si
, regs
->di
, regs
->bp
, sp
);
165 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
166 (u16
)regs
->ds
, (u16
)regs
->es
, (u16
)regs
->fs
, gs
, ss
);
174 cr4
= read_cr4_safe();
175 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
182 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
187 printk("DR6: %08lx DR7: %08lx\n",
191 void show_regs(struct pt_regs
*regs
)
193 __show_regs(regs
, 1);
194 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
198 * This gets run with %bx containing the
199 * function to call, and %dx containing
202 extern void kernel_thread_helper(void);
205 * Create a kernel thread
207 int kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
211 memset(®s
, 0, sizeof(regs
));
213 regs
.bx
= (unsigned long) fn
;
214 regs
.dx
= (unsigned long) arg
;
218 regs
.fs
= __KERNEL_PERCPU
;
220 regs
.ip
= (unsigned long) kernel_thread_helper
;
221 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
222 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
224 /* Ok, create the new process.. */
225 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
227 EXPORT_SYMBOL(kernel_thread
);
230 * Free current thread data structures etc..
232 void exit_thread(void)
234 /* The process may have allocated an io port bitmap... nuke it. */
235 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
236 struct task_struct
*tsk
= current
;
237 struct thread_struct
*t
= &tsk
->thread
;
239 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
241 kfree(t
->io_bitmap_ptr
);
242 t
->io_bitmap_ptr
= NULL
;
243 clear_thread_flag(TIF_IO_BITMAP
);
245 * Careful, clear this in the TSS too:
247 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
248 t
->io_bitmap_max
= 0;
249 tss
->io_bitmap_owner
= NULL
;
250 tss
->io_bitmap_max
= 0;
251 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
255 ds_exit_thread(current
);
258 void flush_thread(void)
260 struct task_struct
*tsk
= current
;
262 tsk
->thread
.debugreg0
= 0;
263 tsk
->thread
.debugreg1
= 0;
264 tsk
->thread
.debugreg2
= 0;
265 tsk
->thread
.debugreg3
= 0;
266 tsk
->thread
.debugreg6
= 0;
267 tsk
->thread
.debugreg7
= 0;
268 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
269 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
271 * Forget coprocessor state..
273 tsk
->fpu_counter
= 0;
278 void release_thread(struct task_struct
*dead_task
)
280 BUG_ON(dead_task
->mm
);
281 release_vm86_irqs(dead_task
);
285 * This gets called before we allocate a new thread and copy
286 * the current task into it.
288 void prepare_to_copy(struct task_struct
*tsk
)
293 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
294 unsigned long unused
,
295 struct task_struct
*p
, struct pt_regs
*regs
)
297 struct pt_regs
*childregs
;
298 struct task_struct
*tsk
;
301 childregs
= task_pt_regs(p
);
306 p
->thread
.sp
= (unsigned long) childregs
;
307 p
->thread
.sp0
= (unsigned long) (childregs
+1);
309 p
->thread
.ip
= (unsigned long) ret_from_fork
;
311 savesegment(gs
, p
->thread
.gs
);
314 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
315 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
316 IO_BITMAP_BYTES
, GFP_KERNEL
);
317 if (!p
->thread
.io_bitmap_ptr
) {
318 p
->thread
.io_bitmap_max
= 0;
321 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
327 * Set a new TLS for the child thread?
329 if (clone_flags
& CLONE_SETTLS
)
330 err
= do_set_thread_area(p
, -1,
331 (struct user_desc __user
*)childregs
->si
, 0);
333 if (err
&& p
->thread
.io_bitmap_ptr
) {
334 kfree(p
->thread
.io_bitmap_ptr
);
335 p
->thread
.io_bitmap_max
= 0;
338 ds_copy_thread(p
, current
);
340 clear_tsk_thread_flag(p
, TIF_DEBUGCTLMSR
);
341 p
->thread
.debugctlmsr
= 0;
347 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
349 __asm__("movl %0, %%gs" : : "r"(0));
352 regs
->ds
= __USER_DS
;
353 regs
->es
= __USER_DS
;
354 regs
->ss
= __USER_DS
;
355 regs
->cs
= __USER_CS
;
359 * Free the old FP and other extended state
361 free_thread_xstate(current
);
363 EXPORT_SYMBOL_GPL(start_thread
);
365 static void hard_disable_TSC(void)
367 write_cr4(read_cr4() | X86_CR4_TSD
);
370 void disable_TSC(void)
373 if (!test_and_set_thread_flag(TIF_NOTSC
))
375 * Must flip the CPU state synchronously with
376 * TIF_NOTSC in the current running context.
382 static void hard_enable_TSC(void)
384 write_cr4(read_cr4() & ~X86_CR4_TSD
);
387 static void enable_TSC(void)
390 if (test_and_clear_thread_flag(TIF_NOTSC
))
392 * Must flip the CPU state synchronously with
393 * TIF_NOTSC in the current running context.
399 int get_tsc_mode(unsigned long adr
)
403 if (test_thread_flag(TIF_NOTSC
))
404 val
= PR_TSC_SIGSEGV
;
408 return put_user(val
, (unsigned int __user
*)adr
);
411 int set_tsc_mode(unsigned int val
)
413 if (val
== PR_TSC_SIGSEGV
)
415 else if (val
== PR_TSC_ENABLE
)
424 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
425 struct tss_struct
*tss
)
427 struct thread_struct
*prev
, *next
;
429 prev
= &prev_p
->thread
;
430 next
= &next_p
->thread
;
432 if (test_tsk_thread_flag(next_p
, TIF_DS_AREA_MSR
) ||
433 test_tsk_thread_flag(prev_p
, TIF_DS_AREA_MSR
))
434 ds_switch_to(prev_p
, next_p
);
435 else if (next
->debugctlmsr
!= prev
->debugctlmsr
)
436 update_debugctlmsr(next
->debugctlmsr
);
438 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
439 set_debugreg(next
->debugreg0
, 0);
440 set_debugreg(next
->debugreg1
, 1);
441 set_debugreg(next
->debugreg2
, 2);
442 set_debugreg(next
->debugreg3
, 3);
444 set_debugreg(next
->debugreg6
, 6);
445 set_debugreg(next
->debugreg7
, 7);
448 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
449 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
450 /* prev and next are different */
451 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
457 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
459 * Disable the bitmap via an invalid offset. We still cache
460 * the previous bitmap owner and the IO bitmap contents:
462 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
466 if (likely(next
== tss
->io_bitmap_owner
)) {
468 * Previous owner of the bitmap (hence the bitmap content)
469 * matches the next task, we dont have to do anything but
470 * to set a valid offset in the TSS:
472 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
476 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
477 * and we let the task to get a GPF in case an I/O instruction
478 * is performed. The handler of the GPF will verify that the
479 * faulting task has a valid I/O bitmap and, it true, does the
480 * real copy and restart the instruction. This will save us
481 * redundant copies when the currently switched task does not
482 * perform any I/O during its timeslice.
484 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
488 * switch_to(x,yn) should switch tasks from x to y.
490 * We fsave/fwait so that an exception goes off at the right time
491 * (as a call from the fsave or fwait in effect) rather than to
492 * the wrong process. Lazy FP saving no longer makes any sense
493 * with modern CPU's, and this simplifies a lot of things (SMP
494 * and UP become the same).
496 * NOTE! We used to use the x86 hardware context switching. The
497 * reason for not using it any more becomes apparent when you
498 * try to recover gracefully from saved state that is no longer
499 * valid (stale segment register values in particular). With the
500 * hardware task-switch, there is no way to fix up bad state in
501 * a reasonable manner.
503 * The fact that Intel documents the hardware task-switching to
504 * be slow is a fairly red herring - this code is not noticeably
505 * faster. However, there _is_ some room for improvement here,
506 * so the performance issues may eventually be a valid point.
507 * More important, however, is the fact that this allows us much
510 * The return value (in %ax) will be the "prev" task after
511 * the task-switch, and shows up in ret_from_fork in entry.S,
514 __notrace_funcgraph
struct task_struct
*
515 __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
517 struct thread_struct
*prev
= &prev_p
->thread
,
518 *next
= &next_p
->thread
;
519 int cpu
= smp_processor_id();
520 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
522 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
524 __unlazy_fpu(prev_p
);
527 /* we're going to use this soon, after a few expensive things */
528 if (next_p
->fpu_counter
> 5)
529 prefetch(next
->xstate
);
537 * Save away %gs. No need to save %fs, as it was saved on the
538 * stack on entry. No need to save %es and %ds, as those are
539 * always kernel segments while inside the kernel. Doing this
540 * before setting the new TLS descriptors avoids the situation
541 * where we temporarily have non-reloadable segments in %fs
542 * and %gs. This could be an issue if the NMI handler ever
543 * used %fs or %gs (it does not today), or if the kernel is
544 * running inside of a hypervisor layer.
546 savesegment(gs
, prev
->gs
);
549 * Load the per-thread Thread-Local Storage descriptor.
554 * Restore IOPL if needed. In normal use, the flags restore
555 * in the switch assembly will handle this. But if the kernel
556 * is running virtualized at a non-zero CPL, the popf will
557 * not restore flags, so it must be done in a separate step.
559 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
560 set_iopl_mask(next
->iopl
);
563 * Now maybe handle debug registers and/or IO bitmaps
565 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
566 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
567 __switch_to_xtra(prev_p
, next_p
, tss
);
570 * Leave lazy mode, flushing any hypercalls made here.
571 * This must be done before restoring TLS segments so
572 * the GDT and LDT are properly updated, and must be
573 * done before math_state_restore, so the TS bit is up
576 arch_leave_lazy_cpu_mode();
578 /* If the task has used fpu the last 5 timeslices, just do a full
579 * restore of the math state immediately to avoid the trap; the
580 * chances of needing FPU soon are obviously high now
582 * tsk_used_math() checks prevent calling math_state_restore(),
583 * which can sleep in the case of !tsk_used_math()
585 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
586 math_state_restore();
589 * Restore %gs if needed (which is common)
591 if (prev
->gs
| next
->gs
)
592 loadsegment(gs
, next
->gs
);
594 x86_write_percpu(current_task
, next_p
);
599 asmlinkage
int sys_fork(struct pt_regs regs
)
601 return do_fork(SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
604 asmlinkage
int sys_clone(struct pt_regs regs
)
606 unsigned long clone_flags
;
608 int __user
*parent_tidptr
, *child_tidptr
;
610 clone_flags
= regs
.bx
;
612 parent_tidptr
= (int __user
*)regs
.dx
;
613 child_tidptr
= (int __user
*)regs
.di
;
616 return do_fork(clone_flags
, newsp
, ®s
, 0, parent_tidptr
, child_tidptr
);
620 * This is trivial, and on the face of it looks like it
621 * could equally well be done in user mode.
623 * Not so, for quite unobvious reasons - register pressure.
624 * In user mode vfork() cannot have a stack frame, and if
625 * done by calling the "clone()" system call directly, you
626 * do not have enough call-clobbered registers to hold all
627 * the information you need.
629 asmlinkage
int sys_vfork(struct pt_regs regs
)
631 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
635 * sys_execve() executes a new program.
637 asmlinkage
int sys_execve(struct pt_regs regs
)
642 filename
= getname((char __user
*) regs
.bx
);
643 error
= PTR_ERR(filename
);
644 if (IS_ERR(filename
))
646 error
= do_execve(filename
,
647 (char __user
* __user
*) regs
.cx
,
648 (char __user
* __user
*) regs
.dx
,
651 /* Make sure we don't return using sysenter.. */
652 set_thread_flag(TIF_IRET
);
659 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
660 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
662 unsigned long get_wchan(struct task_struct
*p
)
664 unsigned long bp
, sp
, ip
;
665 unsigned long stack_page
;
667 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
669 stack_page
= (unsigned long)task_stack_page(p
);
671 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
673 /* include/asm-i386/system.h:switch_to() pushes bp last. */
674 bp
= *(unsigned long *) sp
;
676 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
678 ip
= *(unsigned long *) (bp
+4);
679 if (!in_sched_functions(ip
))
681 bp
= *(unsigned long *) bp
;
682 } while (count
++ < 16);
686 unsigned long arch_align_stack(unsigned long sp
)
688 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
689 sp
-= get_random_int() % 8192;
693 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
695 unsigned long range_end
= mm
->brk
+ 0x02000000;
696 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;