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/delay.h>
27 #include <linux/reboot.h>
28 #include <linux/init.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/module.h>
31 #include <linux/kallsyms.h>
32 #include <linux/ptrace.h>
33 #include <linux/personality.h>
34 #include <linux/tick.h>
35 #include <linux/percpu.h>
36 #include <linux/prctl.h>
37 #include <linux/ftrace.h>
38 #include <linux/uaccess.h>
40 #include <linux/kdebug.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
45 #include <asm/processor.h>
48 #ifdef CONFIG_MATH_EMULATION
49 #include <asm/math_emu.h>
52 #include <linux/err.h>
54 #include <asm/tlbflush.h>
57 #include <asm/syscalls.h>
58 #include <asm/debugreg.h>
60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
63 * Return saved PC of a blocked thread.
65 unsigned long thread_saved_pc(struct task_struct *tsk)
67 return ((unsigned long *)tsk->thread.sp)[3];
71 static inline void play_dead(void)
78 * The idle thread. There's no useful work to be
79 * done, so just try to conserve power and have a
80 * low exit latency (ie sit in a loop waiting for
81 * somebody to say that they'd like to reschedule)
85 int cpu = smp_processor_id();
88 * If we're the non-boot CPU, nothing set the stack canary up
89 * for us. CPU0 already has it initialized but no harm in
90 * doing it again. This is a good place for updating it, as
91 * we wont ever return from this function (so the invalid
92 * canaries already on the stack wont ever trigger).
94 boot_init_stack_canary();
96 current_thread_info()->status |= TS_POLLING;
98 /* endless idle loop with no priority at all */
100 tick_nohz_stop_sched_tick(1);
101 while (!need_resched()) {
106 if (cpu_is_offline(cpu))
110 /* Don't trace irqs off for idle */
111 stop_critical_timings();
113 start_critical_timings();
115 tick_nohz_restart_sched_tick();
116 preempt_enable_no_resched();
122 void __show_regs(struct pt_regs *regs, int all)
124 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
125 unsigned long d0, d1, d2, d3, d6, d7;
127 unsigned short ss, gs;
129 if (user_mode_vm(regs)) {
131 ss = regs->ss & 0xffff;
132 gs = get_user_gs(regs);
134 sp = kernel_stack_pointer(regs);
141 printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
142 (u16)regs->cs, regs->ip, regs->flags,
144 print_symbol("EIP is at %s\n", regs->ip);
146 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
147 regs->ax, regs->bx, regs->cx, regs->dx);
148 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
149 regs->si, regs->di, regs->bp, sp);
150 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
151 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
159 cr4 = read_cr4_safe();
160 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
167 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
172 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
176 void release_thread(struct task_struct *dead_task)
178 BUG_ON(dead_task->mm);
179 release_vm86_irqs(dead_task);
183 * This gets called before we allocate a new thread and copy
184 * the current task into it.
186 void prepare_to_copy(struct task_struct *tsk)
191 int copy_thread(unsigned long clone_flags, unsigned long sp,
192 unsigned long unused,
193 struct task_struct *p, struct pt_regs *regs)
195 struct pt_regs *childregs;
196 struct task_struct *tsk;
199 childregs = task_pt_regs(p);
204 p->thread.sp = (unsigned long) childregs;
205 p->thread.sp0 = (unsigned long) (childregs+1);
207 p->thread.ip = (unsigned long) ret_from_fork;
209 task_user_gs(p) = get_user_gs(regs);
211 p->thread.io_bitmap_ptr = NULL;
215 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
217 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
218 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
219 IO_BITMAP_BYTES, GFP_KERNEL);
220 if (!p->thread.io_bitmap_ptr) {
221 p->thread.io_bitmap_max = 0;
224 set_tsk_thread_flag(p, TIF_IO_BITMAP);
230 * Set a new TLS for the child thread?
232 if (clone_flags & CLONE_SETTLS)
233 err = do_set_thread_area(p, -1,
234 (struct user_desc __user *)childregs->si, 0);
236 if (err && p->thread.io_bitmap_ptr) {
237 kfree(p->thread.io_bitmap_ptr);
238 p->thread.io_bitmap_max = 0;
244 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
248 set_user_gs(regs, 0);
252 regs->ds = __USER_DS;
253 regs->es = __USER_DS;
254 regs->ss = __USER_DS;
255 regs->cs = __USER_CS;
260 load_user_cs_desc(cpu, current->mm);
264 * Free the old FP and other extended state
266 free_thread_xstate(current);
268 EXPORT_SYMBOL_GPL(start_thread);
272 * switch_to(x,yn) should switch tasks from x to y.
274 * We fsave/fwait so that an exception goes off at the right time
275 * (as a call from the fsave or fwait in effect) rather than to
276 * the wrong process. Lazy FP saving no longer makes any sense
277 * with modern CPU's, and this simplifies a lot of things (SMP
278 * and UP become the same).
280 * NOTE! We used to use the x86 hardware context switching. The
281 * reason for not using it any more becomes apparent when you
282 * try to recover gracefully from saved state that is no longer
283 * valid (stale segment register values in particular). With the
284 * hardware task-switch, there is no way to fix up bad state in
285 * a reasonable manner.
287 * The fact that Intel documents the hardware task-switching to
288 * be slow is a fairly red herring - this code is not noticeably
289 * faster. However, there _is_ some room for improvement here,
290 * so the performance issues may eventually be a valid point.
291 * More important, however, is the fact that this allows us much
294 * The return value (in %ax) will be the "prev" task after
295 * the task-switch, and shows up in ret_from_fork in entry.S,
298 __notrace_funcgraph struct task_struct *
299 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
301 struct thread_struct *prev = &prev_p->thread,
302 *next = &next_p->thread;
303 int cpu = smp_processor_id();
304 struct tss_struct *tss = &per_cpu(init_tss, cpu);
307 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
310 * If the task has used fpu the last 5 timeslices, just do a full
311 * restore of the math state immediately to avoid the trap; the
312 * chances of needing FPU soon are obviously high now
314 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
316 __unlazy_fpu(prev_p);
318 /* we're going to use this soon, after a few expensive things */
320 prefetch(next->fpu.state);
323 load_user_cs_desc(cpu, next_p->mm);
331 * Save away %gs. No need to save %fs, as it was saved on the
332 * stack on entry. No need to save %es and %ds, as those are
333 * always kernel segments while inside the kernel. Doing this
334 * before setting the new TLS descriptors avoids the situation
335 * where we temporarily have non-reloadable segments in %fs
336 * and %gs. This could be an issue if the NMI handler ever
337 * used %fs or %gs (it does not today), or if the kernel is
338 * running inside of a hypervisor layer.
340 lazy_save_gs(prev->gs);
343 * Load the per-thread Thread-Local Storage descriptor.
348 * Restore IOPL if needed. In normal use, the flags restore
349 * in the switch assembly will handle this. But if the kernel
350 * is running virtualized at a non-zero CPL, the popf will
351 * not restore flags, so it must be done in a separate step.
353 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
354 set_iopl_mask(next->iopl);
357 * Now maybe handle debug registers and/or IO bitmaps
359 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
360 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
361 __switch_to_xtra(prev_p, next_p, tss);
363 /* If we're going to preload the fpu context, make sure clts
364 is run while we're batching the cpu state updates. */
369 * Leave lazy mode, flushing any hypercalls made here.
370 * This must be done before restoring TLS segments so
371 * the GDT and LDT are properly updated, and must be
372 * done before math_state_restore, so the TS bit is up
375 arch_end_context_switch(next_p);
378 __math_state_restore();
381 * Restore %gs if needed (which is common)
383 if (prev->gs | next->gs)
384 lazy_load_gs(next->gs);
386 percpu_write(current_task, next_p);
391 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
392 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
394 unsigned long get_wchan(struct task_struct *p)
396 unsigned long bp, sp, ip;
397 unsigned long stack_page;
399 if (!p || p == current || p->state == TASK_RUNNING)
401 stack_page = (unsigned long)task_stack_page(p);
403 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
405 /* include/asm-i386/system.h:switch_to() pushes bp last. */
406 bp = *(unsigned long *) sp;
408 if (bp < stack_page || bp > top_ebp+stack_page)
410 ip = *(unsigned long *) (bp+4);
411 if (!in_sched_functions(ip))
413 bp = *(unsigned long *) bp;
414 } while (count++ < 16);
418 static void modify_cs(struct mm_struct *mm, unsigned long limit)
420 mm->context.exec_limit = limit;
421 set_user_cs(&mm->context.user_cs, limit);
422 if (mm == current->mm) {
426 load_user_cs_desc(cpu, mm);
431 void arch_add_exec_range(struct mm_struct *mm, unsigned long limit)
433 if (limit > mm->context.exec_limit)
434 modify_cs(mm, limit);
437 void arch_remove_exec_range(struct mm_struct *mm, unsigned long old_end)
439 struct vm_area_struct *vma;
440 unsigned long limit = PAGE_SIZE;
442 if (old_end == mm->context.exec_limit) {
443 for (vma = mm->mmap; vma; vma = vma->vm_next)
444 if ((vma->vm_flags & VM_EXEC) && (vma->vm_end > limit))
446 modify_cs(mm, limit);
450 void arch_flush_exec_range(struct mm_struct *mm)
452 mm->context.exec_limit = 0;
453 set_user_cs(&mm->context.user_cs, 0);