1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * 4 * Pentium III FXSR, SSE support 5 * Gareth Hughes <gareth@valinux.com>, May 2000 6 */ 7 8 /* 9 * This file handles the architecture-dependent parts of process handling.. 10 */ 11 12 #include <linux/stackprotector.h> 13 #include <linux/cpu.h> 14 #include <linux/errno.h> 15 #include <linux/sched.h> 16 #include <linux/fs.h> 17 #include <linux/kernel.h> 18 #include <linux/mm.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> 39 #include <linux/io.h> 40 #include <linux/kdebug.h> 41 #include <linux/cpuidle.h> 42 43 #include <asm/pgtable.h> 44 #include <asm/system.h> 45 #include <asm/ldt.h> 46 #include <asm/processor.h> 47 #include <asm/i387.h> 48 #include <asm/desc.h> 49 #ifdef CONFIG_MATH_EMULATION 50 #include <asm/math_emu.h> 51 #endif 52 53 #include <linux/err.h> 54 55 #include <asm/tlbflush.h> 56 #include <asm/cpu.h> 57 #include <asm/idle.h> 58 #include <asm/syscalls.h> 59 #include <asm/debugreg.h> 60 61 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 62 63 /* 64 * Return saved PC of a blocked thread. 65 */ 66 unsigned long thread_saved_pc(struct task_struct *tsk) 67 { 68 return ((unsigned long *)tsk->thread.sp)[3]; 69 } 70 71 #ifndef CONFIG_SMP 72 static inline void play_dead(void) 73 { 74 BUG(); 75 } 76 #endif 77 78 /* 79 * The idle thread. There's no useful work to be 80 * done, so just try to conserve power and have a 81 * low exit latency (ie sit in a loop waiting for 82 * somebody to say that they'd like to reschedule) 83 */ 84 void cpu_idle(void) 85 { 86 int cpu = smp_processor_id(); 87 88 /* 89 * If we're the non-boot CPU, nothing set the stack canary up 90 * for us. CPU0 already has it initialized but no harm in 91 * doing it again. This is a good place for updating it, as 92 * we wont ever return from this function (so the invalid 93 * canaries already on the stack wont ever trigger). 94 */ 95 boot_init_stack_canary(); 96 97 current_thread_info()->status |= TS_POLLING; 98 99 /* endless idle loop with no priority at all */ 100 while (1) { 101 tick_nohz_stop_sched_tick(1); 102 while (!need_resched()) { 103 104 check_pgt_cache(); 105 rmb(); 106 107 if (cpu_is_offline(cpu)) 108 play_dead(); 109 110 local_irq_disable(); 111 /* Don't trace irqs off for idle */ 112 stop_critical_timings(); 113 if (cpuidle_idle_call()) 114 pm_idle(); 115 start_critical_timings(); 116 } 117 tick_nohz_restart_sched_tick(); 118 preempt_enable_no_resched(); 119 schedule(); 120 preempt_disable(); 121 } 122 } 123 124 void __show_regs(struct pt_regs *regs, int all) 125 { 126 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; 127 unsigned long d0, d1, d2, d3, d6, d7; 128 unsigned long sp; 129 unsigned short ss, gs; 130 131 if (user_mode_vm(regs)) { 132 sp = regs->sp; 133 ss = regs->ss & 0xffff; 134 gs = get_user_gs(regs); 135 } else { 136 sp = kernel_stack_pointer(regs); 137 savesegment(ss, ss); 138 savesegment(gs, gs); 139 } 140 141 show_regs_common(); 142 143 printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n", 144 (u16)regs->cs, regs->ip, regs->flags, 145 smp_processor_id()); 146 print_symbol("EIP is at %s\n", regs->ip); 147 148 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", 149 regs->ax, regs->bx, regs->cx, regs->dx); 150 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n", 151 regs->si, regs->di, regs->bp, sp); 152 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n", 153 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss); 154 155 if (!all) 156 return; 157 158 cr0 = read_cr0(); 159 cr2 = read_cr2(); 160 cr3 = read_cr3(); 161 cr4 = read_cr4_safe(); 162 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", 163 cr0, cr2, cr3, cr4); 164 165 get_debugreg(d0, 0); 166 get_debugreg(d1, 1); 167 get_debugreg(d2, 2); 168 get_debugreg(d3, 3); 169 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n", 170 d0, d1, d2, d3); 171 172 get_debugreg(d6, 6); 173 get_debugreg(d7, 7); 174 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n", 175 d6, d7); 176 } 177 178 void release_thread(struct task_struct *dead_task) 179 { 180 BUG_ON(dead_task->mm); 181 release_vm86_irqs(dead_task); 182 } 183 184 /* 185 * This gets called before we allocate a new thread and copy 186 * the current task into it. 187 */ 188 void prepare_to_copy(struct task_struct *tsk) 189 { 190 unlazy_fpu(tsk); 191 } 192 193 int copy_thread(unsigned long clone_flags, unsigned long sp, 194 unsigned long unused, 195 struct task_struct *p, struct pt_regs *regs) 196 { 197 struct pt_regs *childregs; 198 struct task_struct *tsk; 199 int err; 200 201 childregs = task_pt_regs(p); 202 *childregs = *regs; 203 childregs->ax = 0; 204 childregs->sp = sp; 205 206 p->thread.sp = (unsigned long) childregs; 207 p->thread.sp0 = (unsigned long) (childregs+1); 208 209 p->thread.ip = (unsigned long) ret_from_fork; 210 211 task_user_gs(p) = get_user_gs(regs); 212 213 p->thread.io_bitmap_ptr = NULL; 214 tsk = current; 215 err = -ENOMEM; 216 217 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 218 219 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { 220 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, 221 IO_BITMAP_BYTES, GFP_KERNEL); 222 if (!p->thread.io_bitmap_ptr) { 223 p->thread.io_bitmap_max = 0; 224 return -ENOMEM; 225 } 226 set_tsk_thread_flag(p, TIF_IO_BITMAP); 227 } 228 229 err = 0; 230 231 /* 232 * Set a new TLS for the child thread? 233 */ 234 if (clone_flags & CLONE_SETTLS) 235 err = do_set_thread_area(p, -1, 236 (struct user_desc __user *)childregs->si, 0); 237 238 if (err && p->thread.io_bitmap_ptr) { 239 kfree(p->thread.io_bitmap_ptr); 240 p->thread.io_bitmap_max = 0; 241 } 242 return err; 243 } 244 245 void 246 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) 247 { 248 set_user_gs(regs, 0); 249 regs->fs = 0; 250 regs->ds = __USER_DS; 251 regs->es = __USER_DS; 252 regs->ss = __USER_DS; 253 regs->cs = __USER_CS; 254 regs->ip = new_ip; 255 regs->sp = new_sp; 256 /* 257 * Free the old FP and other extended state 258 */ 259 free_thread_xstate(current); 260 } 261 EXPORT_SYMBOL_GPL(start_thread); 262 263 264 /* 265 * switch_to(x,yn) should switch tasks from x to y. 266 * 267 * We fsave/fwait so that an exception goes off at the right time 268 * (as a call from the fsave or fwait in effect) rather than to 269 * the wrong process. Lazy FP saving no longer makes any sense 270 * with modern CPU's, and this simplifies a lot of things (SMP 271 * and UP become the same). 272 * 273 * NOTE! We used to use the x86 hardware context switching. The 274 * reason for not using it any more becomes apparent when you 275 * try to recover gracefully from saved state that is no longer 276 * valid (stale segment register values in particular). With the 277 * hardware task-switch, there is no way to fix up bad state in 278 * a reasonable manner. 279 * 280 * The fact that Intel documents the hardware task-switching to 281 * be slow is a fairly red herring - this code is not noticeably 282 * faster. However, there _is_ some room for improvement here, 283 * so the performance issues may eventually be a valid point. 284 * More important, however, is the fact that this allows us much 285 * more flexibility. 286 * 287 * The return value (in %ax) will be the "prev" task after 288 * the task-switch, and shows up in ret_from_fork in entry.S, 289 * for example. 290 */ 291 __notrace_funcgraph struct task_struct * 292 __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 293 { 294 struct thread_struct *prev = &prev_p->thread, 295 *next = &next_p->thread; 296 int cpu = smp_processor_id(); 297 struct tss_struct *tss = &per_cpu(init_tss, cpu); 298 bool preload_fpu; 299 300 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ 301 302 /* 303 * If the task has used fpu the last 5 timeslices, just do a full 304 * restore of the math state immediately to avoid the trap; the 305 * chances of needing FPU soon are obviously high now 306 */ 307 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5; 308 309 __unlazy_fpu(prev_p); 310 311 /* we're going to use this soon, after a few expensive things */ 312 if (preload_fpu) 313 prefetch(next->fpu.state); 314 315 /* 316 * Reload esp0. 317 */ 318 load_sp0(tss, next); 319 320 /* 321 * Save away %gs. No need to save %fs, as it was saved on the 322 * stack on entry. No need to save %es and %ds, as those are 323 * always kernel segments while inside the kernel. Doing this 324 * before setting the new TLS descriptors avoids the situation 325 * where we temporarily have non-reloadable segments in %fs 326 * and %gs. This could be an issue if the NMI handler ever 327 * used %fs or %gs (it does not today), or if the kernel is 328 * running inside of a hypervisor layer. 329 */ 330 lazy_save_gs(prev->gs); 331 332 /* 333 * Load the per-thread Thread-Local Storage descriptor. 334 */ 335 load_TLS(next, cpu); 336 337 /* 338 * Restore IOPL if needed. In normal use, the flags restore 339 * in the switch assembly will handle this. But if the kernel 340 * is running virtualized at a non-zero CPL, the popf will 341 * not restore flags, so it must be done in a separate step. 342 */ 343 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl)) 344 set_iopl_mask(next->iopl); 345 346 /* 347 * Now maybe handle debug registers and/or IO bitmaps 348 */ 349 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV || 350 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT)) 351 __switch_to_xtra(prev_p, next_p, tss); 352 353 /* If we're going to preload the fpu context, make sure clts 354 is run while we're batching the cpu state updates. */ 355 if (preload_fpu) 356 clts(); 357 358 /* 359 * Leave lazy mode, flushing any hypercalls made here. 360 * This must be done before restoring TLS segments so 361 * the GDT and LDT are properly updated, and must be 362 * done before math_state_restore, so the TS bit is up 363 * to date. 364 */ 365 arch_end_context_switch(next_p); 366 367 if (preload_fpu) 368 __math_state_restore(); 369 370 /* 371 * Restore %gs if needed (which is common) 372 */ 373 if (prev->gs | next->gs) 374 lazy_load_gs(next->gs); 375 376 percpu_write(current_task, next_p); 377 378 return prev_p; 379 } 380 381 #define top_esp (THREAD_SIZE - sizeof(unsigned long)) 382 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) 383 384 unsigned long get_wchan(struct task_struct *p) 385 { 386 unsigned long bp, sp, ip; 387 unsigned long stack_page; 388 int count = 0; 389 if (!p || p == current || p->state == TASK_RUNNING) 390 return 0; 391 stack_page = (unsigned long)task_stack_page(p); 392 sp = p->thread.sp; 393 if (!stack_page || sp < stack_page || sp > top_esp+stack_page) 394 return 0; 395 /* include/asm-i386/system.h:switch_to() pushes bp last. */ 396 bp = *(unsigned long *) sp; 397 do { 398 if (bp < stack_page || bp > top_ebp+stack_page) 399 return 0; 400 ip = *(unsigned long *) (bp+4); 401 if (!in_sched_functions(ip)) 402 return ip; 403 bp = *(unsigned long *) bp; 404 } while (count++ < 16); 405 return 0; 406 } 407 408