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..
12 #include <linux/stackprotector.h>
13 #include <linux/cpu.h>
14 #include <linux/errno.h>
15 #include <linux/sched.h>
17 #include <linux/kernel.h>
19 #include <linux/elfcore.h>
20 #include <linux/smp.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/user.h>
25 #include <linux/interrupt.h>
26 #include <linux/utsname.h>
27 #include <linux/delay.h>
28 #include <linux/reboot.h>
29 #include <linux/init.h>
30 #include <linux/mc146818rtc.h>
31 #include <linux/module.h>
32 #include <linux/kallsyms.h>
33 #include <linux/ptrace.h>
34 #include <linux/personality.h>
35 #include <linux/tick.h>
36 #include <linux/percpu.h>
37 #include <linux/prctl.h>
38 #include <linux/dmi.h>
39 #include <linux/ftrace.h>
40 #include <linux/uaccess.h>
42 #include <linux/kdebug.h>
44 #include <asm/pgtable.h>
45 #include <asm/system.h>
47 #include <asm/processor.h>
50 #ifdef CONFIG_MATH_EMULATION
51 #include <asm/math_emu.h>
54 #include <linux/err.h>
56 #include <asm/tlbflush.h>
59 #include <asm/syscalls.h>
61 #include <asm/debugreg.h>
63 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
66 * Return saved PC of a blocked thread.
68 unsigned long thread_saved_pc(struct task_struct *tsk)
70 return ((unsigned long *)tsk->thread.sp)[3];
74 static inline void play_dead(void)
81 * The idle thread. There's no useful work to be
82 * done, so just try to conserve power and have a
83 * low exit latency (ie sit in a loop waiting for
84 * somebody to say that they'd like to reschedule)
88 int cpu = smp_processor_id();
91 * If we're the non-boot CPU, nothing set the stack canary up
92 * for us. CPU0 already has it initialized but no harm in
93 * doing it again. This is a good place for updating it, as
94 * we wont ever return from this function (so the invalid
95 * canaries already on the stack wont ever trigger).
97 boot_init_stack_canary();
99 current_thread_info()->status |= TS_POLLING;
101 /* endless idle loop with no priority at all */
103 tick_nohz_stop_sched_tick(1);
104 while (!need_resched()) {
109 if (cpu_is_offline(cpu))
113 /* Don't trace irqs off for idle */
114 stop_critical_timings();
116 start_critical_timings();
118 tick_nohz_restart_sched_tick();
119 preempt_enable_no_resched();
125 void __show_regs(struct pt_regs *regs, int all)
127 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
128 unsigned long d0, d1, d2, d3, d6, d7;
130 unsigned short ss, gs;
133 if (user_mode_vm(regs)) {
135 ss = regs->ss & 0xffff;
136 gs = get_user_gs(regs);
138 sp = kernel_stack_pointer(regs);
145 board = dmi_get_system_info(DMI_PRODUCT_NAME);
148 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
149 task_pid_nr(current), current->comm,
150 print_tainted(), init_utsname()->release,
151 (int)strcspn(init_utsname()->version, " "),
152 init_utsname()->version, board);
154 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
155 (u16)regs->cs, regs->ip, regs->flags,
157 print_symbol("EIP is at %s\n", regs->ip);
159 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
160 regs->ax, regs->bx, regs->cx, regs->dx);
161 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
162 regs->si, regs->di, regs->bp, sp);
163 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
164 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
172 cr4 = read_cr4_safe();
173 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
180 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
185 printk("DR6: %08lx DR7: %08lx\n",
189 void show_regs(struct pt_regs *regs)
191 show_registers(regs);
192 show_trace(NULL, regs, ®s->sp, regs->bp);
196 * This gets run with %bx containing the
197 * function to call, and %dx containing
200 extern void kernel_thread_helper(void);
203 * Create a kernel thread
205 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
209 memset(®s, 0, sizeof(regs));
211 regs.bx = (unsigned long) fn;
212 regs.dx = (unsigned long) arg;
216 regs.fs = __KERNEL_PERCPU;
217 regs.gs = __KERNEL_STACK_CANARY;
219 regs.ip = (unsigned long) kernel_thread_helper;
220 regs.cs = __KERNEL_CS | get_kernel_rpl();
221 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
223 /* Ok, create the new process.. */
224 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
226 EXPORT_SYMBOL(kernel_thread);
228 void release_thread(struct task_struct *dead_task)
230 BUG_ON(dead_task->mm);
231 release_vm86_irqs(dead_task);
235 * This gets called before we allocate a new thread and copy
236 * the current task into it.
238 void prepare_to_copy(struct task_struct *tsk)
243 int copy_thread(unsigned long clone_flags, unsigned long sp,
244 unsigned long unused,
245 struct task_struct *p, struct pt_regs *regs)
247 struct pt_regs *childregs;
248 struct task_struct *tsk;
251 childregs = task_pt_regs(p);
256 p->thread.sp = (unsigned long) childregs;
257 p->thread.sp0 = (unsigned long) (childregs+1);
259 p->thread.ip = (unsigned long) ret_from_fork;
261 task_user_gs(p) = get_user_gs(regs);
263 p->thread.io_bitmap_ptr = NULL;
267 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
269 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
270 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
271 IO_BITMAP_BYTES, GFP_KERNEL);
272 if (!p->thread.io_bitmap_ptr) {
273 p->thread.io_bitmap_max = 0;
276 set_tsk_thread_flag(p, TIF_IO_BITMAP);
282 * Set a new TLS for the child thread?
284 if (clone_flags & CLONE_SETTLS)
285 err = do_set_thread_area(p, -1,
286 (struct user_desc __user *)childregs->si, 0);
288 if (err && p->thread.io_bitmap_ptr) {
289 kfree(p->thread.io_bitmap_ptr);
290 p->thread.io_bitmap_max = 0;
293 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
294 p->thread.ds_ctx = NULL;
296 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
297 p->thread.debugctlmsr = 0;
303 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
305 set_user_gs(regs, 0);
308 regs->ds = __USER_DS;
309 regs->es = __USER_DS;
310 regs->ss = __USER_DS;
311 regs->cs = __USER_CS;
315 * Free the old FP and other extended state
317 free_thread_xstate(current);
319 EXPORT_SYMBOL_GPL(start_thread);
323 * switch_to(x,yn) should switch tasks from x to y.
325 * We fsave/fwait so that an exception goes off at the right time
326 * (as a call from the fsave or fwait in effect) rather than to
327 * the wrong process. Lazy FP saving no longer makes any sense
328 * with modern CPU's, and this simplifies a lot of things (SMP
329 * and UP become the same).
331 * NOTE! We used to use the x86 hardware context switching. The
332 * reason for not using it any more becomes apparent when you
333 * try to recover gracefully from saved state that is no longer
334 * valid (stale segment register values in particular). With the
335 * hardware task-switch, there is no way to fix up bad state in
336 * a reasonable manner.
338 * The fact that Intel documents the hardware task-switching to
339 * be slow is a fairly red herring - this code is not noticeably
340 * faster. However, there _is_ some room for improvement here,
341 * so the performance issues may eventually be a valid point.
342 * More important, however, is the fact that this allows us much
345 * The return value (in %ax) will be the "prev" task after
346 * the task-switch, and shows up in ret_from_fork in entry.S,
349 __notrace_funcgraph struct task_struct *
350 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
352 struct thread_struct *prev = &prev_p->thread,
353 *next = &next_p->thread;
354 int cpu = smp_processor_id();
355 struct tss_struct *tss = &per_cpu(init_tss, cpu);
358 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
361 * If the task has used fpu the last 5 timeslices, just do a full
362 * restore of the math state immediately to avoid the trap; the
363 * chances of needing FPU soon are obviously high now
365 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
367 __unlazy_fpu(prev_p);
369 /* we're going to use this soon, after a few expensive things */
371 prefetch(next->xstate);
379 * Save away %gs. No need to save %fs, as it was saved on the
380 * stack on entry. No need to save %es and %ds, as those are
381 * always kernel segments while inside the kernel. Doing this
382 * before setting the new TLS descriptors avoids the situation
383 * where we temporarily have non-reloadable segments in %fs
384 * and %gs. This could be an issue if the NMI handler ever
385 * used %fs or %gs (it does not today), or if the kernel is
386 * running inside of a hypervisor layer.
388 lazy_save_gs(prev->gs);
391 * Load the per-thread Thread-Local Storage descriptor.
396 * Restore IOPL if needed. In normal use, the flags restore
397 * in the switch assembly will handle this. But if the kernel
398 * is running virtualized at a non-zero CPL, the popf will
399 * not restore flags, so it must be done in a separate step.
401 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
402 set_iopl_mask(next->iopl);
405 * Now maybe handle debug registers and/or IO bitmaps
407 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
408 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
409 __switch_to_xtra(prev_p, next_p, tss);
411 /* If we're going to preload the fpu context, make sure clts
412 is run while we're batching the cpu state updates. */
417 * Leave lazy mode, flushing any hypercalls made here.
418 * This must be done before restoring TLS segments so
419 * the GDT and LDT are properly updated, and must be
420 * done before math_state_restore, so the TS bit is up
423 arch_end_context_switch(next_p);
426 __math_state_restore();
429 * Restore %gs if needed (which is common)
431 if (prev->gs | next->gs)
432 lazy_load_gs(next->gs);
434 percpu_write(current_task, next_p);
439 int sys_clone(struct pt_regs *regs)
441 unsigned long clone_flags;
443 int __user *parent_tidptr, *child_tidptr;
445 clone_flags = regs->bx;
447 parent_tidptr = (int __user *)regs->dx;
448 child_tidptr = (int __user *)regs->di;
451 return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr);
455 * sys_execve() executes a new program.
457 int sys_execve(struct pt_regs *regs)
462 filename = getname((char __user *) regs->bx);
463 error = PTR_ERR(filename);
464 if (IS_ERR(filename))
466 error = do_execve(filename,
467 (char __user * __user *) regs->cx,
468 (char __user * __user *) regs->dx,
471 /* Make sure we don't return using sysenter.. */
472 set_thread_flag(TIF_IRET);
479 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
480 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
482 unsigned long get_wchan(struct task_struct *p)
484 unsigned long bp, sp, ip;
485 unsigned long stack_page;
487 if (!p || p == current || p->state == TASK_RUNNING)
489 stack_page = (unsigned long)task_stack_page(p);
491 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
493 /* include/asm-i386/system.h:switch_to() pushes bp last. */
494 bp = *(unsigned long *) sp;
496 if (bp < stack_page || bp > top_ebp+stack_page)
498 ip = *(unsigned long *) (bp+4);
499 if (!in_sched_functions(ip))
501 bp = *(unsigned long *) bp;
502 } while (count++ < 16);