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 <xen/interface/physdev.h>
53 #include <xen/interface/vcpu.h>
54 #include <xen/cpu_hotplug.h>
56 #include <linux/err.h>
58 #include <asm/tlbflush.h>
61 #include <asm/syscalls.h>
63 #include <asm/debugreg.h>
65 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
66 asmlinkage void cstar_ret_from_fork(void) __asm__("cstar_ret_from_fork");
69 * Return saved PC of a blocked thread.
71 unsigned long thread_saved_pc(struct task_struct *tsk)
73 return ((unsigned long *)tsk->thread.sp)[3];
77 static inline void play_dead(void)
84 * The idle thread. There's no useful work to be
85 * done, so just try to conserve power and have a
86 * low exit latency (ie sit in a loop waiting for
87 * somebody to say that they'd like to reschedule)
91 int cpu = smp_processor_id();
94 * If we're the non-boot CPU, nothing set the stack canary up
95 * for us. CPU0 already has it initialized but no harm in
96 * doing it again. This is a good place for updating it, as
97 * we wont ever return from this function (so the invalid
98 * canaries already on the stack wont ever trigger).
100 boot_init_stack_canary();
102 current_thread_info()->status |= TS_POLLING;
104 /* endless idle loop with no priority at all */
106 tick_nohz_stop_sched_tick(1);
107 while (!need_resched()) {
112 if (cpu_is_offline(cpu))
116 /* Don't trace irqs off for idle */
117 stop_critical_timings();
119 start_critical_timings();
121 tick_nohz_restart_sched_tick();
122 preempt_enable_no_resched();
128 void __show_regs(struct pt_regs *regs, int all)
130 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
131 unsigned long d0, d1, d2, d3, d6, d7;
133 unsigned short ss, gs;
135 if (user_mode_vm(regs)) {
137 ss = regs->ss & 0xffff;
138 gs = get_user_gs(regs);
140 sp = kernel_stack_pointer(regs);
147 printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
148 (u16)regs->cs, regs->ip, regs->flags,
150 print_symbol("EIP is at %s\n", regs->ip);
152 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
153 regs->ax, regs->bx, regs->cx, regs->dx);
154 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
155 regs->si, regs->di, regs->bp, sp);
156 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
157 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
165 cr4 = read_cr4_safe();
166 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
173 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
178 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
182 void release_thread(struct task_struct *dead_task)
184 BUG_ON(dead_task->mm);
185 release_vm86_irqs(dead_task);
189 * This gets called before we allocate a new thread and copy
190 * the current task into it.
192 void prepare_to_copy(struct task_struct *tsk)
197 int copy_thread(unsigned long clone_flags, unsigned long sp,
198 unsigned long unused,
199 struct task_struct *p, struct pt_regs *regs)
201 struct pt_regs *childregs;
202 struct task_struct *tsk;
205 childregs = task_pt_regs(p);
210 p->thread.sp = (unsigned long) childregs;
211 p->thread.sp0 = (unsigned long) (childregs+1);
213 p->thread.ip = (unsigned long) ret_from_fork;
215 task_user_gs(p) = get_user_gs(regs);
217 p->thread.io_bitmap_ptr = NULL;
221 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
223 if (test_tsk_thread_flag(tsk, TIF_CSTAR))
224 p->thread.ip = (unsigned long) cstar_ret_from_fork;
225 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
226 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
227 IO_BITMAP_BYTES, GFP_KERNEL);
228 if (!p->thread.io_bitmap_ptr) {
229 p->thread.io_bitmap_max = 0;
232 set_tsk_thread_flag(p, TIF_IO_BITMAP);
238 * Set a new TLS for the child thread?
240 if (clone_flags & CLONE_SETTLS)
241 err = do_set_thread_area(p, -1,
242 (struct user_desc __user *)childregs->si, 0);
244 p->thread.iopl = current->thread.iopl;
246 if (err && p->thread.io_bitmap_ptr) {
247 kfree(p->thread.io_bitmap_ptr);
248 p->thread.io_bitmap_max = 0;
251 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
252 p->thread.ds_ctx = NULL;
254 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
255 p->thread.debugctlmsr = 0;
261 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
263 set_user_gs(regs, 0);
266 regs->ds = __USER_DS;
267 regs->es = __USER_DS;
268 regs->ss = __USER_DS;
269 regs->cs = __USER_CS;
273 * Free the old FP and other extended state
275 free_thread_xstate(current);
277 EXPORT_SYMBOL_GPL(start_thread);
280 * switch_to(x,yn) should switch tasks from x to y.
282 * We fsave/fwait so that an exception goes off at the right time
283 * (as a call from the fsave or fwait in effect) rather than to
284 * the wrong process. Lazy FP saving no longer makes any sense
285 * with modern CPU's, and this simplifies a lot of things (SMP
286 * and UP become the same).
288 * NOTE! We used to use the x86 hardware context switching. The
289 * reason for not using it any more becomes apparent when you
290 * try to recover gracefully from saved state that is no longer
291 * valid (stale segment register values in particular). With the
292 * hardware task-switch, there is no way to fix up bad state in
293 * a reasonable manner.
295 * The fact that Intel documents the hardware task-switching to
296 * be slow is a fairly red herring - this code is not noticeably
297 * faster. However, there _is_ some room for improvement here,
298 * so the performance issues may eventually be a valid point.
299 * More important, however, is the fact that this allows us much
302 * The return value (in %ax) will be the "prev" task after
303 * the task-switch, and shows up in ret_from_fork in entry.S,
306 __notrace_funcgraph struct task_struct *
307 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
309 struct thread_struct *prev = &prev_p->thread,
310 *next = &next_p->thread;
311 int cpu = smp_processor_id();
312 #ifndef CONFIG_X86_NO_TSS
313 struct tss_struct *tss = &per_cpu(init_tss, cpu);
316 #if CONFIG_XEN_COMPAT > 0x030002
317 struct physdev_set_iopl iopl_op;
318 struct physdev_set_iobitmap iobmp_op;
320 struct physdev_op _pdo[2], *pdo = _pdo;
321 #define iopl_op pdo->u.set_iopl
322 #define iobmp_op pdo->u.set_iobitmap
324 multicall_entry_t _mcl[8], *mcl = _mcl;
326 /* XEN NOTE: FS/GS saved in switch_mm(), not here. */
329 * If the task has used fpu the last 5 timeslices, just do a full
330 * restore of the math state immediately to avoid the trap; the
331 * chances of needing FPU soon are obviously high now
333 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
336 * This is basically '__unlazy_fpu', except that we queue a
337 * multicall to indicate FPU task switch, rather than
338 * synchronously trapping to Xen.
340 if (task_thread_info(prev_p)->status & TS_USEDFPU) {
341 __save_init_fpu(prev_p); /* _not_ save_init_fpu() */
343 mcl->op = __HYPERVISOR_fpu_taskswitch;
348 #if 0 /* lazy fpu sanity check */
349 else BUG_ON(!(read_cr0() & 8));
354 * This is load_sp0(tss, next) with a multicall.
356 mcl->op = __HYPERVISOR_stack_switch;
357 mcl->args[0] = __KERNEL_DS;
358 mcl->args[1] = next->sp0;
362 * Load the per-thread Thread-Local Storage descriptor.
363 * This is load_TLS(next, cpu) with multicalls.
366 if (unlikely(next->tls_array[i].a != prev->tls_array[i].a || \
367 next->tls_array[i].b != prev->tls_array[i].b)) { \
368 mcl->op = __HYPERVISOR_update_descriptor; \
369 *(u64 *)&mcl->args[0] = arbitrary_virt_to_machine( \
370 &get_cpu_gdt_table(cpu)[GDT_ENTRY_TLS_MIN + i]);\
371 *(u64 *)&mcl->args[2] = *(u64 *)&next->tls_array[i]; \
378 if (unlikely(prev->iopl != next->iopl)) {
379 iopl_op.iopl = (next->iopl == 0) ? 1 : (next->iopl >> 12) & 3;
380 #if CONFIG_XEN_COMPAT > 0x030002
381 mcl->op = __HYPERVISOR_physdev_op;
382 mcl->args[0] = PHYSDEVOP_set_iopl;
383 mcl->args[1] = (unsigned long)&iopl_op;
385 mcl->op = __HYPERVISOR_physdev_op_compat;
386 pdo->cmd = PHYSDEVOP_set_iopl;
387 mcl->args[0] = (unsigned long)pdo++;
392 /* If we're going to preload the fpu context, make sure clts
393 is run while we're batching the cpu state updates. */
395 mcl->op = __HYPERVISOR_fpu_taskswitch;
400 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
401 set_xen_guest_handle(iobmp_op.bitmap,
402 (char *)next->io_bitmap_ptr);
403 iobmp_op.nr_ports = next->io_bitmap_ptr ? IO_BITMAP_BITS : 0;
404 #if CONFIG_XEN_COMPAT > 0x030002
405 mcl->op = __HYPERVISOR_physdev_op;
406 mcl->args[0] = PHYSDEVOP_set_iobitmap;
407 mcl->args[1] = (unsigned long)&iobmp_op;
409 mcl->op = __HYPERVISOR_physdev_op_compat;
410 pdo->cmd = PHYSDEVOP_set_iobitmap;
411 mcl->args[0] = (unsigned long)pdo++;
416 #if CONFIG_XEN_COMPAT <= 0x030002
417 BUG_ON(pdo > _pdo + ARRAY_SIZE(_pdo));
419 BUG_ON(mcl > _mcl + ARRAY_SIZE(_mcl));
420 if (unlikely(HYPERVISOR_multicall_check(_mcl, mcl - _mcl, NULL)))
423 /* we're going to use this soon, after a few expensive things */
425 prefetch(next->xstate);
428 * Now maybe handle debug registers
430 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
431 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
432 __switch_to_xtra(prev_p, next_p);
435 * Leave lazy mode, flushing any hypercalls made here.
436 * This must be done before restoring TLS segments so
437 * the GDT and LDT are properly updated, and must be
438 * done before math_state_restore, so the TS bit is up
441 arch_end_context_switch(next_p);
444 __math_state_restore();
447 * Restore %gs if needed (which is common)
449 if (prev->gs | next->gs)
450 lazy_load_gs(next->gs);
452 percpu_write(current_task, next_p);
457 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
458 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
460 unsigned long get_wchan(struct task_struct *p)
462 unsigned long bp, sp, ip;
463 unsigned long stack_page;
465 if (!p || p == current || p->state == TASK_RUNNING)
467 stack_page = (unsigned long)task_stack_page(p);
469 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
471 /* include/asm-i386/system.h:switch_to() pushes bp last. */
472 bp = *(unsigned long *) sp;
474 if (bp < stack_page || bp > top_ebp+stack_page)
476 ip = *(unsigned long *) (bp+4);
477 if (!in_sched_functions(ip))
479 bp = *(unsigned long *) bp;
480 } while (count++ < 16);