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>
41 #include <linux/cpuidle.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
46 #include <asm/processor.h>
49 #ifdef CONFIG_MATH_EMULATION
50 #include <asm/math_emu.h>
53 #include <xen/interface/physdev.h>
55 #include <linux/err.h>
57 #include <asm/tlbflush.h>
60 #include <asm/syscalls.h>
61 #include <asm/debugreg.h>
64 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
65 asmlinkage void cstar_ret_from_fork(void) __asm__("cstar_ret_from_fork");
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];
76 static inline void play_dead(void)
83 * The idle thread. There's no useful work to be
84 * done, so just try to conserve power and have a
85 * low exit latency (ie sit in a loop waiting for
86 * somebody to say that they'd like to reschedule)
90 int cpu = smp_processor_id();
93 * If we're the non-boot CPU, nothing set the stack canary up
94 * for us. CPU0 already has it initialized but no harm in
95 * doing it again. This is a good place for updating it, as
96 * we wont ever return from this function (so the invalid
97 * canaries already on the stack wont ever trigger).
99 boot_init_stack_canary();
101 current_thread_info()->status |= TS_POLLING;
103 /* endless idle loop with no priority at all */
105 tick_nohz_idle_enter();
107 while (!need_resched()) {
112 if (cpu_is_offline(cpu))
117 /* Don't trace irqs off for idle */
118 stop_critical_timings();
119 if (cpuidle_idle_call())
121 start_critical_timings();
124 tick_nohz_idle_exit();
125 preempt_enable_no_resched();
131 void __show_regs(struct pt_regs *regs, int all)
133 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
134 unsigned long d0, d1, d2, d3, d6, d7;
136 unsigned short ss, gs;
138 if (user_mode_vm(regs)) {
140 ss = regs->ss & 0xffff;
141 gs = get_user_gs(regs);
143 sp = kernel_stack_pointer(regs);
150 printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
151 (u16)regs->cs, regs->ip, regs->flags,
153 print_symbol("EIP is at %s\n", regs->ip);
155 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
156 regs->ax, regs->bx, regs->cx, regs->dx);
157 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
158 regs->si, regs->di, regs->bp, sp);
159 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
160 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
168 cr4 = read_cr4_safe();
169 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
176 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
181 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
185 void release_thread(struct task_struct *dead_task)
187 BUG_ON(dead_task->mm);
188 release_vm86_irqs(dead_task);
192 * This gets called before we allocate a new thread and copy
193 * the current task into it.
195 void prepare_to_copy(struct task_struct *tsk)
200 int copy_thread(unsigned long clone_flags, unsigned long sp,
201 unsigned long unused,
202 struct task_struct *p, struct pt_regs *regs)
204 struct pt_regs *childregs;
205 struct task_struct *tsk;
208 childregs = task_pt_regs(p);
213 p->thread.sp = (unsigned long) childregs;
214 p->thread.sp0 = (unsigned long) (childregs+1);
216 p->thread.ip = (unsigned long) ret_from_fork;
218 task_user_gs(p) = get_user_gs(regs);
221 p->thread.io_bitmap_ptr = NULL;
225 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
228 if (test_tsk_thread_flag(tsk, TIF_CSTAR))
229 p->thread.ip = (unsigned long) cstar_ret_from_fork;
231 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
232 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
233 IO_BITMAP_BYTES, GFP_KERNEL);
234 if (!p->thread.io_bitmap_ptr) {
235 p->thread.io_bitmap_max = 0;
238 set_tsk_thread_flag(p, TIF_IO_BITMAP);
244 * Set a new TLS for the child thread?
246 if (clone_flags & CLONE_SETTLS)
247 err = do_set_thread_area(p, -1,
248 (struct user_desc __user *)childregs->si, 0);
250 p->thread.iopl = current->thread.iopl;
252 if (err && p->thread.io_bitmap_ptr) {
253 kfree(p->thread.io_bitmap_ptr);
254 p->thread.io_bitmap_max = 0;
260 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
262 set_user_gs(regs, 0);
264 regs->ds = __USER_DS;
265 regs->es = __USER_DS;
266 regs->ss = __USER_DS;
267 regs->cs = __USER_CS;
271 * Free the old FP and other extended state
273 free_thread_xstate(current);
275 EXPORT_SYMBOL_GPL(start_thread);
278 * switch_to(x,y) should switch tasks from x to y.
280 * We fsave/fwait so that an exception goes off at the right time
281 * (as a call from the fsave or fwait in effect) rather than to
282 * the wrong process. Lazy FP saving no longer makes any sense
283 * with modern CPU's, and this simplifies a lot of things (SMP
284 * and UP become the same).
286 * NOTE! We used to use the x86 hardware context switching. The
287 * reason for not using it any more becomes apparent when you
288 * try to recover gracefully from saved state that is no longer
289 * valid (stale segment register values in particular). With the
290 * hardware task-switch, there is no way to fix up bad state in
291 * a reasonable manner.
293 * The fact that Intel documents the hardware task-switching to
294 * be slow is a fairly red herring - this code is not noticeably
295 * faster. However, there _is_ some room for improvement here,
296 * so the performance issues may eventually be a valid point.
297 * More important, however, is the fact that this allows us much
300 * The return value (in %ax) will be the "prev" task after
301 * the task-switch, and shows up in ret_from_fork in entry.S,
304 __notrace_funcgraph struct task_struct *
305 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
307 struct thread_struct *prev = &prev_p->thread,
308 *next = &next_p->thread;
309 int cpu = smp_processor_id();
310 #ifndef CONFIG_X86_NO_TSS
311 struct tss_struct *tss = &per_cpu(init_tss, cpu);
314 #if CONFIG_XEN_COMPAT > 0x030002
315 struct physdev_set_iopl iopl_op;
316 struct physdev_set_iobitmap iobmp_op;
318 struct physdev_op _pdo[2], *pdo = _pdo;
319 #define iopl_op pdo->u.set_iopl
320 #define iobmp_op pdo->u.set_iobitmap
322 multicall_entry_t _mcl[8], *mcl = _mcl;
324 /* XEN NOTE: FS/GS saved in switch_mm(), not here. */
326 fpu = xen_switch_fpu_prepare(prev_p, next_p, cpu, &mcl);
330 * This is load_sp0(tss, next) with a multicall.
332 mcl->op = __HYPERVISOR_stack_switch;
333 mcl->args[0] = __KERNEL_DS;
334 mcl->args[1] = next->sp0;
338 * Load the per-thread Thread-Local Storage descriptor.
339 * This is load_TLS(next, cpu) with multicalls.
342 if (unlikely(next->tls_array[i].a != prev->tls_array[i].a || \
343 next->tls_array[i].b != prev->tls_array[i].b)) { \
344 mcl->op = __HYPERVISOR_update_descriptor; \
345 *(u64 *)&mcl->args[0] = arbitrary_virt_to_machine( \
346 &get_cpu_gdt_table(cpu)[GDT_ENTRY_TLS_MIN + i]);\
347 *(u64 *)&mcl->args[2] = *(u64 *)&next->tls_array[i]; \
354 if (unlikely(prev->iopl != next->iopl)) {
355 iopl_op.iopl = (next->iopl == 0) ? 1 : (next->iopl >> 12) & 3;
356 #if CONFIG_XEN_COMPAT > 0x030002
357 mcl->op = __HYPERVISOR_physdev_op;
358 mcl->args[0] = PHYSDEVOP_set_iopl;
359 mcl->args[1] = (unsigned long)&iopl_op;
361 mcl->op = __HYPERVISOR_physdev_op_compat;
362 pdo->cmd = PHYSDEVOP_set_iopl;
363 mcl->args[0] = (unsigned long)pdo++;
368 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
369 set_xen_guest_handle(iobmp_op.bitmap,
370 (char *)next->io_bitmap_ptr);
371 iobmp_op.nr_ports = next->io_bitmap_ptr ? IO_BITMAP_BITS : 0;
372 #if CONFIG_XEN_COMPAT > 0x030002
373 mcl->op = __HYPERVISOR_physdev_op;
374 mcl->args[0] = PHYSDEVOP_set_iobitmap;
375 mcl->args[1] = (unsigned long)&iobmp_op;
377 mcl->op = __HYPERVISOR_physdev_op_compat;
378 pdo->cmd = PHYSDEVOP_set_iobitmap;
379 mcl->args[0] = (unsigned long)pdo++;
384 #if CONFIG_XEN_COMPAT <= 0x030002
385 BUG_ON(pdo > _pdo + ARRAY_SIZE(_pdo));
387 BUG_ON(mcl > _mcl + ARRAY_SIZE(_mcl));
388 if (unlikely(HYPERVISOR_multicall_check(_mcl, mcl - _mcl, NULL)))
392 * Now maybe handle debug registers
394 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
395 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
396 __switch_to_xtra(prev_p, next_p);
399 * Leave lazy mode, flushing any hypercalls made here.
400 * This must be done before restoring TLS segments so
401 * the GDT and LDT are properly updated, and must be
402 * done before math_state_restore, so the TS bit is up
405 arch_end_context_switch(next_p);
408 * Restore %gs if needed (which is common)
410 if (prev->gs | next->gs)
411 lazy_load_gs(next->gs);
413 switch_fpu_finish(next_p, fpu);
415 percpu_write(current_task, next_p);
420 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
421 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
423 unsigned long get_wchan(struct task_struct *p)
425 unsigned long bp, sp, ip;
426 unsigned long stack_page;
428 if (!p || p == current || p->state == TASK_RUNNING)
430 stack_page = (unsigned long)task_stack_page(p);
432 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
434 /* include/asm-i386/system.h:switch_to() pushes bp last. */
435 bp = *(unsigned long *) sp;
437 if (bp < stack_page || bp > top_ebp+stack_page)
439 ip = *(unsigned long *) (bp+4);
440 if (!in_sched_functions(ip))
442 bp = *(unsigned long *) bp;
443 } while (count++ < 16);