]>
git.proxmox.com Git - mirror_ubuntu-kernels.git/blob - arch/x86/kernel/process_64.c
2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
23 #include <linux/kernel.h>
25 #include <linux/elfcore.h>
26 #include <linux/smp.h>
27 #include <linux/slab.h>
28 #include <linux/user.h>
29 #include <linux/interrupt.h>
30 #include <linux/utsname.h>
31 #include <linux/delay.h>
32 #include <linux/module.h>
33 #include <linux/ptrace.h>
34 #include <linux/random.h>
35 #include <linux/notifier.h>
36 #include <linux/kprobes.h>
37 #include <linux/kdebug.h>
38 #include <linux/tick.h>
39 #include <linux/prctl.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
45 #include <asm/processor.h>
47 #include <asm/mmu_context.h>
49 #include <asm/prctl.h>
51 #include <asm/proto.h>
55 asmlinkage
extern void ret_from_fork(void);
57 unsigned long kernel_thread_flags
= CLONE_VM
| CLONE_UNTRACED
;
59 static ATOMIC_NOTIFIER_HEAD(idle_notifier
);
61 void idle_notifier_register(struct notifier_block
*n
)
63 atomic_notifier_chain_register(&idle_notifier
, n
);
69 atomic_notifier_call_chain(&idle_notifier
, IDLE_START
, NULL
);
72 static void __exit_idle(void)
74 if (test_and_clear_bit_pda(0, isidle
) == 0)
76 atomic_notifier_call_chain(&idle_notifier
, IDLE_END
, NULL
);
79 /* Called from interrupts to signify idle end */
82 /* idle loop has pid 0 */
88 #ifdef CONFIG_HOTPLUG_CPU
89 DECLARE_PER_CPU(int, cpu_state
);
92 /* We halt the CPU with physical CPU hotplug */
93 static inline void play_dead(void)
99 __get_cpu_var(cpu_state
) = CPU_DEAD
;
106 static inline void play_dead(void)
110 #endif /* CONFIG_HOTPLUG_CPU */
113 * The idle thread. There's no useful work to be
114 * done, so just try to conserve power and have a
115 * low exit latency (ie sit in a loop waiting for
116 * somebody to say that they'd like to reschedule)
120 current_thread_info()->status
|= TS_POLLING
;
121 /* endless idle loop with no priority at all */
123 tick_nohz_stop_sched_tick(1);
124 while (!need_resched()) {
128 if (cpu_is_offline(smp_processor_id()))
131 * Idle routines should keep interrupts disabled
132 * from here on, until they go to idle.
133 * Otherwise, idle callbacks can misfire.
137 /* Don't trace irqs off for idle */
138 stop_critical_timings();
140 start_critical_timings();
141 /* In many cases the interrupt that ended idle
142 has already called exit_idle. But some idle
143 loops can be woken up without interrupt. */
147 tick_nohz_restart_sched_tick();
148 preempt_enable_no_resched();
154 /* Prints also some state that isn't saved in the pt_regs */
155 void __show_regs(struct pt_regs
* regs
)
157 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L, fs
, gs
, shadowgs
;
158 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
159 unsigned int fsindex
, gsindex
;
160 unsigned int ds
, cs
, es
;
164 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
165 current
->pid
, current
->comm
, print_tainted(),
166 init_utsname()->release
,
167 (int)strcspn(init_utsname()->version
, " "),
168 init_utsname()->version
);
169 printk("RIP: %04lx:[<%016lx>] ", regs
->cs
& 0xffff, regs
->ip
);
170 printk_address(regs
->ip
, 1);
171 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs
->ss
, regs
->sp
,
173 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
174 regs
->ax
, regs
->bx
, regs
->cx
);
175 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
176 regs
->dx
, regs
->si
, regs
->di
);
177 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
178 regs
->bp
, regs
->r8
, regs
->r9
);
179 printk("R10: %016lx R11: %016lx R12: %016lx\n",
180 regs
->r10
, regs
->r11
, regs
->r12
);
181 printk("R13: %016lx R14: %016lx R15: %016lx\n",
182 regs
->r13
, regs
->r14
, regs
->r15
);
184 asm("movl %%ds,%0" : "=r" (ds
));
185 asm("movl %%cs,%0" : "=r" (cs
));
186 asm("movl %%es,%0" : "=r" (es
));
187 asm("movl %%fs,%0" : "=r" (fsindex
));
188 asm("movl %%gs,%0" : "=r" (gsindex
));
190 rdmsrl(MSR_FS_BASE
, fs
);
191 rdmsrl(MSR_GS_BASE
, gs
);
192 rdmsrl(MSR_KERNEL_GS_BASE
, shadowgs
);
199 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
200 fs
,fsindex
,gs
,gsindex
,shadowgs
);
201 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs
, ds
, es
, cr0
);
202 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2
, cr3
, cr4
);
207 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0
, d1
, d2
);
211 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3
, d6
, d7
);
214 void show_regs(struct pt_regs
*regs
)
216 printk("CPU %d:", smp_processor_id());
218 show_trace(NULL
, regs
, (void *)(regs
+ 1), regs
->bp
);
222 * Free current thread data structures etc..
224 void exit_thread(void)
226 struct task_struct
*me
= current
;
227 struct thread_struct
*t
= &me
->thread
;
229 if (me
->thread
.io_bitmap_ptr
) {
230 struct tss_struct
*tss
= &per_cpu(init_tss
, get_cpu());
232 kfree(t
->io_bitmap_ptr
);
233 t
->io_bitmap_ptr
= NULL
;
234 clear_thread_flag(TIF_IO_BITMAP
);
236 * Careful, clear this in the TSS too:
238 memset(tss
->io_bitmap
, 0xff, t
->io_bitmap_max
);
239 t
->io_bitmap_max
= 0;
243 /* Free any DS contexts that have not been properly released. */
244 if (unlikely(t
->ds_ctx
)) {
245 /* we clear debugctl to make sure DS is not used. */
246 update_debugctlmsr(0);
249 #endif /* CONFIG_X86_DS */
252 void flush_thread(void)
254 struct task_struct
*tsk
= current
;
256 if (test_tsk_thread_flag(tsk
, TIF_ABI_PENDING
)) {
257 clear_tsk_thread_flag(tsk
, TIF_ABI_PENDING
);
258 if (test_tsk_thread_flag(tsk
, TIF_IA32
)) {
259 clear_tsk_thread_flag(tsk
, TIF_IA32
);
261 set_tsk_thread_flag(tsk
, TIF_IA32
);
262 current_thread_info()->status
|= TS_COMPAT
;
265 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
267 tsk
->thread
.debugreg0
= 0;
268 tsk
->thread
.debugreg1
= 0;
269 tsk
->thread
.debugreg2
= 0;
270 tsk
->thread
.debugreg3
= 0;
271 tsk
->thread
.debugreg6
= 0;
272 tsk
->thread
.debugreg7
= 0;
273 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
275 * Forget coprocessor state..
277 tsk
->fpu_counter
= 0;
282 void release_thread(struct task_struct
*dead_task
)
285 if (dead_task
->mm
->context
.size
) {
286 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
288 dead_task
->mm
->context
.ldt
,
289 dead_task
->mm
->context
.size
);
295 static inline void set_32bit_tls(struct task_struct
*t
, int tls
, u32 addr
)
297 struct user_desc ud
= {
304 struct desc_struct
*desc
= t
->thread
.tls_array
;
309 static inline u32
read_32bit_tls(struct task_struct
*t
, int tls
)
311 return get_desc_base(&t
->thread
.tls_array
[tls
]);
315 * This gets called before we allocate a new thread and copy
316 * the current task into it.
318 void prepare_to_copy(struct task_struct
*tsk
)
323 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
324 unsigned long unused
,
325 struct task_struct
* p
, struct pt_regs
* regs
)
328 struct pt_regs
* childregs
;
329 struct task_struct
*me
= current
;
331 childregs
= ((struct pt_regs
*)
332 (THREAD_SIZE
+ task_stack_page(p
))) - 1;
338 childregs
->sp
= (unsigned long)childregs
;
340 p
->thread
.sp
= (unsigned long) childregs
;
341 p
->thread
.sp0
= (unsigned long) (childregs
+1);
342 p
->thread
.usersp
= me
->thread
.usersp
;
344 set_tsk_thread_flag(p
, TIF_FORK
);
346 p
->thread
.fs
= me
->thread
.fs
;
347 p
->thread
.gs
= me
->thread
.gs
;
349 savesegment(gs
, p
->thread
.gsindex
);
350 savesegment(fs
, p
->thread
.fsindex
);
351 savesegment(es
, p
->thread
.es
);
352 savesegment(ds
, p
->thread
.ds
);
354 if (unlikely(test_tsk_thread_flag(me
, TIF_IO_BITMAP
))) {
355 p
->thread
.io_bitmap_ptr
= kmalloc(IO_BITMAP_BYTES
, GFP_KERNEL
);
356 if (!p
->thread
.io_bitmap_ptr
) {
357 p
->thread
.io_bitmap_max
= 0;
360 memcpy(p
->thread
.io_bitmap_ptr
, me
->thread
.io_bitmap_ptr
,
362 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
366 * Set a new TLS for the child thread?
368 if (clone_flags
& CLONE_SETTLS
) {
369 #ifdef CONFIG_IA32_EMULATION
370 if (test_thread_flag(TIF_IA32
))
371 err
= do_set_thread_area(p
, -1,
372 (struct user_desc __user
*)childregs
->si
, 0);
375 err
= do_arch_prctl(p
, ARCH_SET_FS
, childregs
->r8
);
381 if (err
&& p
->thread
.io_bitmap_ptr
) {
382 kfree(p
->thread
.io_bitmap_ptr
);
383 p
->thread
.io_bitmap_max
= 0;
389 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
397 write_pda(oldrsp
, new_sp
);
398 regs
->cs
= __USER_CS
;
399 regs
->ss
= __USER_DS
;
403 * Free the old FP and other extended state
405 free_thread_xstate(current
);
407 EXPORT_SYMBOL_GPL(start_thread
);
409 static void hard_disable_TSC(void)
411 write_cr4(read_cr4() | X86_CR4_TSD
);
414 void disable_TSC(void)
417 if (!test_and_set_thread_flag(TIF_NOTSC
))
419 * Must flip the CPU state synchronously with
420 * TIF_NOTSC in the current running context.
426 static void hard_enable_TSC(void)
428 write_cr4(read_cr4() & ~X86_CR4_TSD
);
431 static void enable_TSC(void)
434 if (test_and_clear_thread_flag(TIF_NOTSC
))
436 * Must flip the CPU state synchronously with
437 * TIF_NOTSC in the current running context.
443 int get_tsc_mode(unsigned long adr
)
447 if (test_thread_flag(TIF_NOTSC
))
448 val
= PR_TSC_SIGSEGV
;
452 return put_user(val
, (unsigned int __user
*)adr
);
455 int set_tsc_mode(unsigned int val
)
457 if (val
== PR_TSC_SIGSEGV
)
459 else if (val
== PR_TSC_ENABLE
)
468 * This special macro can be used to load a debugging register
470 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
472 static inline void __switch_to_xtra(struct task_struct
*prev_p
,
473 struct task_struct
*next_p
,
474 struct tss_struct
*tss
)
476 struct thread_struct
*prev
, *next
;
477 unsigned long debugctl
;
479 prev
= &prev_p
->thread
,
480 next
= &next_p
->thread
;
482 debugctl
= prev
->debugctlmsr
;
486 unsigned long ds_prev
= 0, ds_next
= 0;
489 ds_prev
= (unsigned long)prev
->ds_ctx
->ds
;
491 ds_next
= (unsigned long)next
->ds_ctx
->ds
;
493 if (ds_next
!= ds_prev
) {
495 * We clear debugctl to make sure DS
496 * is not in use when we change it:
499 update_debugctlmsr(0);
500 wrmsrl(MSR_IA32_DS_AREA
, ds_next
);
503 #endif /* CONFIG_X86_DS */
505 if (next
->debugctlmsr
!= debugctl
)
506 update_debugctlmsr(next
->debugctlmsr
);
508 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
518 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
519 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
520 /* prev and next are different */
521 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
527 if (test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
529 * Copy the relevant range of the IO bitmap.
530 * Normally this is 128 bytes or less:
532 memcpy(tss
->io_bitmap
, next
->io_bitmap_ptr
,
533 max(prev
->io_bitmap_max
, next
->io_bitmap_max
));
534 } else if (test_tsk_thread_flag(prev_p
, TIF_IO_BITMAP
)) {
536 * Clear any possible leftover bits:
538 memset(tss
->io_bitmap
, 0xff, prev
->io_bitmap_max
);
541 #ifdef CONFIG_X86_PTRACE_BTS
542 if (test_tsk_thread_flag(prev_p
, TIF_BTS_TRACE_TS
))
543 ptrace_bts_take_timestamp(prev_p
, BTS_TASK_DEPARTS
);
545 if (test_tsk_thread_flag(next_p
, TIF_BTS_TRACE_TS
))
546 ptrace_bts_take_timestamp(next_p
, BTS_TASK_ARRIVES
);
547 #endif /* CONFIG_X86_PTRACE_BTS */
551 * switch_to(x,y) should switch tasks from x to y.
553 * This could still be optimized:
554 * - fold all the options into a flag word and test it with a single test.
555 * - could test fs/gs bitsliced
557 * Kprobes not supported here. Set the probe on schedule instead.
560 __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
562 struct thread_struct
*prev
= &prev_p
->thread
;
563 struct thread_struct
*next
= &next_p
->thread
;
564 int cpu
= smp_processor_id();
565 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
566 unsigned fsindex
, gsindex
;
568 /* we're going to use this soon, after a few expensive things */
569 if (next_p
->fpu_counter
>5)
570 prefetch(next
->xstate
);
573 * Reload esp0, LDT and the page table pointer:
579 * This won't pick up thread selector changes, but I guess that is ok.
581 savesegment(es
, prev
->es
);
582 if (unlikely(next
->es
| prev
->es
))
583 loadsegment(es
, next
->es
);
585 savesegment(ds
, prev
->ds
);
586 if (unlikely(next
->ds
| prev
->ds
))
587 loadsegment(ds
, next
->ds
);
590 /* We must save %fs and %gs before load_TLS() because
591 * %fs and %gs may be cleared by load_TLS().
593 * (e.g. xen_load_tls())
595 savesegment(fs
, fsindex
);
596 savesegment(gs
, gsindex
);
601 * Leave lazy mode, flushing any hypercalls made here.
602 * This must be done before restoring TLS segments so
603 * the GDT and LDT are properly updated, and must be
604 * done before math_state_restore, so the TS bit is up
607 arch_leave_lazy_cpu_mode();
612 * Segment register != 0 always requires a reload. Also
613 * reload when it has changed. When prev process used 64bit
614 * base always reload to avoid an information leak.
616 if (unlikely(fsindex
| next
->fsindex
| prev
->fs
)) {
617 loadsegment(fs
, next
->fsindex
);
619 * Check if the user used a selector != 0; if yes
620 * clear 64bit base, since overloaded base is always
621 * mapped to the Null selector
626 /* when next process has a 64bit base use it */
628 wrmsrl(MSR_FS_BASE
, next
->fs
);
629 prev
->fsindex
= fsindex
;
631 if (unlikely(gsindex
| next
->gsindex
| prev
->gs
)) {
632 load_gs_index(next
->gsindex
);
637 wrmsrl(MSR_KERNEL_GS_BASE
, next
->gs
);
638 prev
->gsindex
= gsindex
;
640 /* Must be after DS reload */
644 * Switch the PDA and FPU contexts.
646 prev
->usersp
= read_pda(oldrsp
);
647 write_pda(oldrsp
, next
->usersp
);
648 write_pda(pcurrent
, next_p
);
650 write_pda(kernelstack
,
651 (unsigned long)task_stack_page(next_p
) +
652 THREAD_SIZE
- PDA_STACKOFFSET
);
653 #ifdef CONFIG_CC_STACKPROTECTOR
654 write_pda(stack_canary
, next_p
->stack_canary
);
656 * Build time only check to make sure the stack_canary is at
657 * offset 40 in the pda; this is a gcc ABI requirement
659 BUILD_BUG_ON(offsetof(struct x8664_pda
, stack_canary
) != 40);
663 * Now maybe reload the debug registers and handle I/O bitmaps
665 if (unlikely(task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
||
666 task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
))
667 __switch_to_xtra(prev_p
, next_p
, tss
);
669 /* If the task has used fpu the last 5 timeslices, just do a full
670 * restore of the math state immediately to avoid the trap; the
671 * chances of needing FPU soon are obviously high now
673 * tsk_used_math() checks prevent calling math_state_restore(),
674 * which can sleep in the case of !tsk_used_math()
676 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
677 math_state_restore();
682 * sys_execve() executes a new program.
685 long sys_execve(char __user
*name
, char __user
* __user
*argv
,
686 char __user
* __user
*envp
, struct pt_regs
*regs
)
691 filename
= getname(name
);
692 error
= PTR_ERR(filename
);
693 if (IS_ERR(filename
))
695 error
= do_execve(filename
, argv
, envp
, regs
);
700 void set_personality_64bit(void)
702 /* inherit personality from parent */
704 /* Make sure to be in 64bit mode */
705 clear_thread_flag(TIF_IA32
);
707 /* TBD: overwrites user setup. Should have two bits.
708 But 64bit processes have always behaved this way,
709 so it's not too bad. The main problem is just that
710 32bit childs are affected again. */
711 current
->personality
&= ~READ_IMPLIES_EXEC
;
714 asmlinkage
long sys_fork(struct pt_regs
*regs
)
716 return do_fork(SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
720 sys_clone(unsigned long clone_flags
, unsigned long newsp
,
721 void __user
*parent_tid
, void __user
*child_tid
, struct pt_regs
*regs
)
725 return do_fork(clone_flags
, newsp
, regs
, 0, parent_tid
, child_tid
);
729 * This is trivial, and on the face of it looks like it
730 * could equally well be done in user mode.
732 * Not so, for quite unobvious reasons - register pressure.
733 * In user mode vfork() cannot have a stack frame, and if
734 * done by calling the "clone()" system call directly, you
735 * do not have enough call-clobbered registers to hold all
736 * the information you need.
738 asmlinkage
long sys_vfork(struct pt_regs
*regs
)
740 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
->sp
, regs
, 0,
744 unsigned long get_wchan(struct task_struct
*p
)
750 if (!p
|| p
== current
|| p
->state
==TASK_RUNNING
)
752 stack
= (unsigned long)task_stack_page(p
);
753 if (p
->thread
.sp
< stack
|| p
->thread
.sp
> stack
+THREAD_SIZE
)
755 fp
= *(u64
*)(p
->thread
.sp
);
757 if (fp
< (unsigned long)stack
||
758 fp
> (unsigned long)stack
+THREAD_SIZE
)
761 if (!in_sched_functions(ip
))
764 } while (count
++ < 16);
768 long do_arch_prctl(struct task_struct
*task
, int code
, unsigned long addr
)
771 int doit
= task
== current
;
776 if (addr
>= TASK_SIZE_OF(task
))
779 /* handle small bases via the GDT because that's faster to
781 if (addr
<= 0xffffffff) {
782 set_32bit_tls(task
, GS_TLS
, addr
);
784 load_TLS(&task
->thread
, cpu
);
785 load_gs_index(GS_TLS_SEL
);
787 task
->thread
.gsindex
= GS_TLS_SEL
;
790 task
->thread
.gsindex
= 0;
791 task
->thread
.gs
= addr
;
794 ret
= checking_wrmsrl(MSR_KERNEL_GS_BASE
, addr
);
800 /* Not strictly needed for fs, but do it for symmetry
802 if (addr
>= TASK_SIZE_OF(task
))
805 /* handle small bases via the GDT because that's faster to
807 if (addr
<= 0xffffffff) {
808 set_32bit_tls(task
, FS_TLS
, addr
);
810 load_TLS(&task
->thread
, cpu
);
811 loadsegment(fs
, FS_TLS_SEL
);
813 task
->thread
.fsindex
= FS_TLS_SEL
;
816 task
->thread
.fsindex
= 0;
817 task
->thread
.fs
= addr
;
819 /* set the selector to 0 to not confuse
822 ret
= checking_wrmsrl(MSR_FS_BASE
, addr
);
829 if (task
->thread
.fsindex
== FS_TLS_SEL
)
830 base
= read_32bit_tls(task
, FS_TLS
);
832 rdmsrl(MSR_FS_BASE
, base
);
834 base
= task
->thread
.fs
;
835 ret
= put_user(base
, (unsigned long __user
*)addr
);
841 if (task
->thread
.gsindex
== GS_TLS_SEL
)
842 base
= read_32bit_tls(task
, GS_TLS
);
844 savesegment(gs
, gsindex
);
846 rdmsrl(MSR_KERNEL_GS_BASE
, base
);
848 base
= task
->thread
.gs
;
851 base
= task
->thread
.gs
;
852 ret
= put_user(base
, (unsigned long __user
*)addr
);
864 long sys_arch_prctl(int code
, unsigned long addr
)
866 return do_arch_prctl(current
, code
, addr
);
869 unsigned long arch_align_stack(unsigned long sp
)
871 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
872 sp
-= get_random_int() % 8192;
876 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
878 unsigned long range_end
= mm
->brk
+ 0x02000000;
879 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;