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/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> 41 #include <linux/io.h> 42 #include <linux/kdebug.h> 43 44 #include <asm/pgtable.h> 45 #include <asm/system.h> 46 #include <asm/ldt.h> 47 #include <asm/processor.h> 48 #include <asm/i387.h> 49 #include <asm/desc.h> 50 #ifdef CONFIG_MATH_EMULATION 51 #include <asm/math_emu.h> 52 #endif 53 54 #include <linux/err.h> 55 56 #include <asm/tlbflush.h> 57 #include <asm/cpu.h> 58 #include <asm/idle.h> 59 #include <asm/syscalls.h> 60 #include <asm/ds.h> 61 62 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 63 64 /* 65 * Return saved PC of a blocked thread. 66 */ 67 unsigned long thread_saved_pc(struct task_struct *tsk) 68 { 69 return ((unsigned long *)tsk->thread.sp)[3]; 70 } 71 72 #ifndef CONFIG_SMP 73 static inline void play_dead(void) 74 { 75 BUG(); 76 } 77 #endif 78 79 /* 80 * The idle thread. There's no useful work to be 81 * done, so just try to conserve power and have a 82 * low exit latency (ie sit in a loop waiting for 83 * somebody to say that they'd like to reschedule) 84 */ 85 void cpu_idle(void) 86 { 87 int cpu = smp_processor_id(); 88 89 /* 90 * If we're the non-boot CPU, nothing set the stack canary up 91 * for us. CPU0 already has it initialized but no harm in 92 * doing it again. This is a good place for updating it, as 93 * we wont ever return from this function (so the invalid 94 * canaries already on the stack wont ever trigger). 95 */ 96 boot_init_stack_canary(); 97 98 current_thread_info()->status |= TS_POLLING; 99 100 /* endless idle loop with no priority at all */ 101 while (1) { 102 tick_nohz_stop_sched_tick(1); 103 while (!need_resched()) { 104 105 check_pgt_cache(); 106 rmb(); 107 108 if (cpu_is_offline(cpu)) 109 play_dead(); 110 111 local_irq_disable(); 112 /* Don't trace irqs off for idle */ 113 stop_critical_timings(); 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 const char *board; 131 132 if (user_mode_vm(regs)) { 133 sp = regs->sp; 134 ss = regs->ss & 0xffff; 135 gs = get_user_gs(regs); 136 } else { 137 sp = (unsigned long) (®s->sp); 138 savesegment(ss, ss); 139 savesegment(gs, gs); 140 } 141 142 printk("\n"); 143 144 board = dmi_get_system_info(DMI_PRODUCT_NAME); 145 if (!board) 146 board = ""; 147 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n", 148 task_pid_nr(current), current->comm, 149 print_tainted(), init_utsname()->release, 150 (int)strcspn(init_utsname()->version, " "), 151 init_utsname()->version, board); 152 153 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n", 154 (u16)regs->cs, regs->ip, regs->flags, 155 smp_processor_id()); 156 print_symbol("EIP is at %s\n", regs->ip); 157 158 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", 159 regs->ax, regs->bx, regs->cx, regs->dx); 160 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n", 161 regs->si, regs->di, regs->bp, sp); 162 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n", 163 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss); 164 165 if (!all) 166 return; 167 168 cr0 = read_cr0(); 169 cr2 = read_cr2(); 170 cr3 = read_cr3(); 171 cr4 = read_cr4_safe(); 172 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", 173 cr0, cr2, cr3, cr4); 174 175 get_debugreg(d0, 0); 176 get_debugreg(d1, 1); 177 get_debugreg(d2, 2); 178 get_debugreg(d3, 3); 179 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n", 180 d0, d1, d2, d3); 181 182 get_debugreg(d6, 6); 183 get_debugreg(d7, 7); 184 printk("DR6: %08lx DR7: %08lx\n", 185 d6, d7); 186 } 187 188 void show_regs(struct pt_regs *regs) 189 { 190 __show_regs(regs, 1); 191 show_trace(NULL, regs, ®s->sp, regs->bp); 192 } 193 194 /* 195 * This gets run with %bx containing the 196 * function to call, and %dx containing 197 * the "args". 198 */ 199 extern void kernel_thread_helper(void); 200 201 /* 202 * Create a kernel thread 203 */ 204 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 205 { 206 struct pt_regs regs; 207 208 memset(®s, 0, sizeof(regs)); 209 210 regs.bx = (unsigned long) fn; 211 regs.dx = (unsigned long) arg; 212 213 regs.ds = __USER_DS; 214 regs.es = __USER_DS; 215 regs.fs = __KERNEL_PERCPU; 216 regs.gs = __KERNEL_STACK_CANARY; 217 regs.orig_ax = -1; 218 regs.ip = (unsigned long) kernel_thread_helper; 219 regs.cs = __KERNEL_CS | get_kernel_rpl(); 220 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2; 221 222 /* Ok, create the new process.. */ 223 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL); 224 } 225 EXPORT_SYMBOL(kernel_thread); 226 227 void release_thread(struct task_struct *dead_task) 228 { 229 BUG_ON(dead_task->mm); 230 release_vm86_irqs(dead_task); 231 } 232 233 /* 234 * This gets called before we allocate a new thread and copy 235 * the current task into it. 236 */ 237 void prepare_to_copy(struct task_struct *tsk) 238 { 239 unlazy_fpu(tsk); 240 } 241 242 int copy_thread(unsigned long clone_flags, unsigned long sp, 243 unsigned long unused, 244 struct task_struct *p, struct pt_regs *regs) 245 { 246 struct pt_regs *childregs; 247 struct task_struct *tsk; 248 int err; 249 250 childregs = task_pt_regs(p); 251 *childregs = *regs; 252 childregs->ax = 0; 253 childregs->sp = sp; 254 255 p->thread.sp = (unsigned long) childregs; 256 p->thread.sp0 = (unsigned long) (childregs+1); 257 258 p->thread.ip = (unsigned long) ret_from_fork; 259 260 task_user_gs(p) = get_user_gs(regs); 261 262 tsk = current; 263 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { 264 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, 265 IO_BITMAP_BYTES, GFP_KERNEL); 266 if (!p->thread.io_bitmap_ptr) { 267 p->thread.io_bitmap_max = 0; 268 return -ENOMEM; 269 } 270 set_tsk_thread_flag(p, TIF_IO_BITMAP); 271 } 272 273 err = 0; 274 275 /* 276 * Set a new TLS for the child thread? 277 */ 278 if (clone_flags & CLONE_SETTLS) 279 err = do_set_thread_area(p, -1, 280 (struct user_desc __user *)childregs->si, 0); 281 282 if (err && p->thread.io_bitmap_ptr) { 283 kfree(p->thread.io_bitmap_ptr); 284 p->thread.io_bitmap_max = 0; 285 } 286 287 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR); 288 p->thread.ds_ctx = NULL; 289 290 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR); 291 p->thread.debugctlmsr = 0; 292 293 return err; 294 } 295 296 void 297 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) 298 { 299 set_user_gs(regs, 0); 300 regs->fs = 0; 301 set_fs(USER_DS); 302 regs->ds = __USER_DS; 303 regs->es = __USER_DS; 304 regs->ss = __USER_DS; 305 regs->cs = __USER_CS; 306 regs->ip = new_ip; 307 regs->sp = new_sp; 308 /* 309 * Free the old FP and other extended state 310 */ 311 free_thread_xstate(current); 312 } 313 EXPORT_SYMBOL_GPL(start_thread); 314 315 316 /* 317 * switch_to(x,yn) should switch tasks from x to y. 318 * 319 * We fsave/fwait so that an exception goes off at the right time 320 * (as a call from the fsave or fwait in effect) rather than to 321 * the wrong process. Lazy FP saving no longer makes any sense 322 * with modern CPU's, and this simplifies a lot of things (SMP 323 * and UP become the same). 324 * 325 * NOTE! We used to use the x86 hardware context switching. The 326 * reason for not using it any more becomes apparent when you 327 * try to recover gracefully from saved state that is no longer 328 * valid (stale segment register values in particular). With the 329 * hardware task-switch, there is no way to fix up bad state in 330 * a reasonable manner. 331 * 332 * The fact that Intel documents the hardware task-switching to 333 * be slow is a fairly red herring - this code is not noticeably 334 * faster. However, there _is_ some room for improvement here, 335 * so the performance issues may eventually be a valid point. 336 * More important, however, is the fact that this allows us much 337 * more flexibility. 338 * 339 * The return value (in %ax) will be the "prev" task after 340 * the task-switch, and shows up in ret_from_fork in entry.S, 341 * for example. 342 */ 343 __notrace_funcgraph struct task_struct * 344 __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 345 { 346 struct thread_struct *prev = &prev_p->thread, 347 *next = &next_p->thread; 348 int cpu = smp_processor_id(); 349 struct tss_struct *tss = &per_cpu(init_tss, cpu); 350 bool preload_fpu; 351 352 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ 353 354 /* 355 * If the task has used fpu the last 5 timeslices, just do a full 356 * restore of the math state immediately to avoid the trap; the 357 * chances of needing FPU soon are obviously high now 358 */ 359 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5; 360 361 __unlazy_fpu(prev_p); 362 363 /* we're going to use this soon, after a few expensive things */ 364 if (preload_fpu) 365 prefetch(next->xstate); 366 367 /* 368 * Reload esp0. 369 */ 370 load_sp0(tss, next); 371 372 /* 373 * Save away %gs. No need to save %fs, as it was saved on the 374 * stack on entry. No need to save %es and %ds, as those are 375 * always kernel segments while inside the kernel. Doing this 376 * before setting the new TLS descriptors avoids the situation 377 * where we temporarily have non-reloadable segments in %fs 378 * and %gs. This could be an issue if the NMI handler ever 379 * used %fs or %gs (it does not today), or if the kernel is 380 * running inside of a hypervisor layer. 381 */ 382 lazy_save_gs(prev->gs); 383 384 /* 385 * Load the per-thread Thread-Local Storage descriptor. 386 */ 387 load_TLS(next, cpu); 388 389 /* 390 * Restore IOPL if needed. In normal use, the flags restore 391 * in the switch assembly will handle this. But if the kernel 392 * is running virtualized at a non-zero CPL, the popf will 393 * not restore flags, so it must be done in a separate step. 394 */ 395 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl)) 396 set_iopl_mask(next->iopl); 397 398 /* 399 * Now maybe handle debug registers and/or IO bitmaps 400 */ 401 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV || 402 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT)) 403 __switch_to_xtra(prev_p, next_p, tss); 404 405 /* If we're going to preload the fpu context, make sure clts 406 is run while we're batching the cpu state updates. */ 407 if (preload_fpu) 408 clts(); 409 410 /* 411 * Leave lazy mode, flushing any hypercalls made here. 412 * This must be done before restoring TLS segments so 413 * the GDT and LDT are properly updated, and must be 414 * done before math_state_restore, so the TS bit is up 415 * to date. 416 */ 417 arch_end_context_switch(next_p); 418 419 if (preload_fpu) 420 __math_state_restore(); 421 422 /* 423 * Restore %gs if needed (which is common) 424 */ 425 if (prev->gs | next->gs) 426 lazy_load_gs(next->gs); 427 428 percpu_write(current_task, next_p); 429 430 return prev_p; 431 } 432 433 int sys_clone(struct pt_regs *regs) 434 { 435 unsigned long clone_flags; 436 unsigned long newsp; 437 int __user *parent_tidptr, *child_tidptr; 438 439 clone_flags = regs->bx; 440 newsp = regs->cx; 441 parent_tidptr = (int __user *)regs->dx; 442 child_tidptr = (int __user *)regs->di; 443 if (!newsp) 444 newsp = regs->sp; 445 return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr); 446 } 447 448 /* 449 * sys_execve() executes a new program. 450 */ 451 int sys_execve(struct pt_regs *regs) 452 { 453 int error; 454 char *filename; 455 456 filename = getname((char __user *) regs->bx); 457 error = PTR_ERR(filename); 458 if (IS_ERR(filename)) 459 goto out; 460 error = do_execve(filename, 461 (char __user * __user *) regs->cx, 462 (char __user * __user *) regs->dx, 463 regs); 464 if (error == 0) { 465 /* Make sure we don't return using sysenter.. */ 466 set_thread_flag(TIF_IRET); 467 } 468 putname(filename); 469 out: 470 return error; 471 } 472 473 #define top_esp (THREAD_SIZE - sizeof(unsigned long)) 474 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) 475 476 unsigned long get_wchan(struct task_struct *p) 477 { 478 unsigned long bp, sp, ip; 479 unsigned long stack_page; 480 int count = 0; 481 if (!p || p == current || p->state == TASK_RUNNING) 482 return 0; 483 stack_page = (unsigned long)task_stack_page(p); 484 sp = p->thread.sp; 485 if (!stack_page || sp < stack_page || sp > top_esp+stack_page) 486 return 0; 487 /* include/asm-i386/system.h:switch_to() pushes bp last. */ 488 bp = *(unsigned long *) sp; 489 do { 490 if (bp < stack_page || bp > top_ebp+stack_page) 491 return 0; 492 ip = *(unsigned long *) (bp+4); 493 if (!in_sched_functions(ip)) 494 return ip; 495 bp = *(unsigned long *) bp; 496 } while (count++ < 16); 497 return 0; 498 } 499 500