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/cpu.h> 13 #include <linux/errno.h> 14 #include <linux/sched.h> 15 #include <linux/fs.h> 16 #include <linux/kernel.h> 17 #include <linux/mm.h> 18 #include <linux/elfcore.h> 19 #include <linux/smp.h> 20 #include <linux/stddef.h> 21 #include <linux/slab.h> 22 #include <linux/vmalloc.h> 23 #include <linux/user.h> 24 #include <linux/interrupt.h> 25 #include <linux/delay.h> 26 #include <linux/reboot.h> 27 #include <linux/init.h> 28 #include <linux/mc146818rtc.h> 29 #include <linux/module.h> 30 #include <linux/kallsyms.h> 31 #include <linux/ptrace.h> 32 #include <linux/personality.h> 33 #include <linux/percpu.h> 34 #include <linux/prctl.h> 35 #include <linux/ftrace.h> 36 #include <linux/uaccess.h> 37 #include <linux/io.h> 38 #include <linux/kdebug.h> 39 40 #include <asm/pgtable.h> 41 #include <asm/ldt.h> 42 #include <asm/processor.h> 43 #include <asm/i387.h> 44 #include <asm/fpu-internal.h> 45 #include <asm/desc.h> 46 #ifdef CONFIG_MATH_EMULATION 47 #include <asm/math_emu.h> 48 #endif 49 50 #include <linux/err.h> 51 52 #include <asm/tlbflush.h> 53 #include <asm/cpu.h> 54 #include <asm/idle.h> 55 #include <asm/syscalls.h> 56 #include <asm/debugreg.h> 57 #include <asm/switch_to.h> 58 59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 60 asmlinkage void ret_from_kernel_thread(void) __asm__("ret_from_kernel_thread"); 61 62 /* 63 * Return saved PC of a blocked thread. 64 */ 65 unsigned long thread_saved_pc(struct task_struct *tsk) 66 { 67 return ((unsigned long *)tsk->thread.sp)[3]; 68 } 69 70 void __show_regs(struct pt_regs *regs, int all) 71 { 72 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; 73 unsigned long d0, d1, d2, d3, d6, d7; 74 unsigned long sp; 75 unsigned short ss, gs; 76 77 if (user_mode_vm(regs)) { 78 sp = regs->sp; 79 ss = regs->ss & 0xffff; 80 gs = get_user_gs(regs); 81 } else { 82 sp = kernel_stack_pointer(regs); 83 savesegment(ss, ss); 84 savesegment(gs, gs); 85 } 86 87 printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n", 88 (u16)regs->cs, regs->ip, regs->flags, 89 smp_processor_id()); 90 print_symbol("EIP is at %s\n", regs->ip); 91 92 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", 93 regs->ax, regs->bx, regs->cx, regs->dx); 94 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n", 95 regs->si, regs->di, regs->bp, sp); 96 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n", 97 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss); 98 99 if (!all) 100 return; 101 102 cr0 = read_cr0(); 103 cr2 = read_cr2(); 104 cr3 = read_cr3(); 105 cr4 = read_cr4_safe(); 106 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", 107 cr0, cr2, cr3, cr4); 108 109 get_debugreg(d0, 0); 110 get_debugreg(d1, 1); 111 get_debugreg(d2, 2); 112 get_debugreg(d3, 3); 113 get_debugreg(d6, 6); 114 get_debugreg(d7, 7); 115 116 /* Only print out debug registers if they are in their non-default state. */ 117 if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) && 118 (d6 == DR6_RESERVED) && (d7 == 0x400)) 119 return; 120 121 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n", 122 d0, d1, d2, d3); 123 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n", 124 d6, d7); 125 } 126 127 void release_thread(struct task_struct *dead_task) 128 { 129 BUG_ON(dead_task->mm); 130 release_vm86_irqs(dead_task); 131 } 132 133 int copy_thread(unsigned long clone_flags, unsigned long sp, 134 unsigned long arg, struct task_struct *p) 135 { 136 struct pt_regs *childregs = task_pt_regs(p); 137 struct task_struct *tsk; 138 int err; 139 140 p->thread.sp = (unsigned long) childregs; 141 p->thread.sp0 = (unsigned long) (childregs+1); 142 143 if (unlikely(p->flags & PF_KTHREAD)) { 144 /* kernel thread */ 145 memset(childregs, 0, sizeof(struct pt_regs)); 146 p->thread.ip = (unsigned long) ret_from_kernel_thread; 147 task_user_gs(p) = __KERNEL_STACK_CANARY; 148 childregs->ds = __USER_DS; 149 childregs->es = __USER_DS; 150 childregs->fs = __KERNEL_PERCPU; 151 childregs->bx = sp; /* function */ 152 childregs->bp = arg; 153 childregs->orig_ax = -1; 154 childregs->cs = __KERNEL_CS | get_kernel_rpl(); 155 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED; 156 p->thread.fpu_counter = 0; 157 p->thread.io_bitmap_ptr = NULL; 158 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 159 return 0; 160 } 161 *childregs = *current_pt_regs(); 162 childregs->ax = 0; 163 if (sp) 164 childregs->sp = sp; 165 166 p->thread.ip = (unsigned long) ret_from_fork; 167 task_user_gs(p) = get_user_gs(current_pt_regs()); 168 169 p->thread.fpu_counter = 0; 170 p->thread.io_bitmap_ptr = NULL; 171 tsk = current; 172 err = -ENOMEM; 173 174 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 175 176 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { 177 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, 178 IO_BITMAP_BYTES, GFP_KERNEL); 179 if (!p->thread.io_bitmap_ptr) { 180 p->thread.io_bitmap_max = 0; 181 return -ENOMEM; 182 } 183 set_tsk_thread_flag(p, TIF_IO_BITMAP); 184 } 185 186 err = 0; 187 188 /* 189 * Set a new TLS for the child thread? 190 */ 191 if (clone_flags & CLONE_SETTLS) 192 err = do_set_thread_area(p, -1, 193 (struct user_desc __user *)childregs->si, 0); 194 195 if (err && p->thread.io_bitmap_ptr) { 196 kfree(p->thread.io_bitmap_ptr); 197 p->thread.io_bitmap_max = 0; 198 } 199 return err; 200 } 201 202 void 203 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) 204 { 205 set_user_gs(regs, 0); 206 regs->fs = 0; 207 regs->ds = __USER_DS; 208 regs->es = __USER_DS; 209 regs->ss = __USER_DS; 210 regs->cs = __USER_CS; 211 regs->ip = new_ip; 212 regs->sp = new_sp; 213 regs->flags = X86_EFLAGS_IF; 214 /* 215 * force it to the iret return path by making it look as if there was 216 * some work pending. 217 */ 218 set_thread_flag(TIF_NOTIFY_RESUME); 219 } 220 EXPORT_SYMBOL_GPL(start_thread); 221 222 223 /* 224 * switch_to(x,y) should switch tasks from x to y. 225 * 226 * We fsave/fwait so that an exception goes off at the right time 227 * (as a call from the fsave or fwait in effect) rather than to 228 * the wrong process. Lazy FP saving no longer makes any sense 229 * with modern CPU's, and this simplifies a lot of things (SMP 230 * and UP become the same). 231 * 232 * NOTE! We used to use the x86 hardware context switching. The 233 * reason for not using it any more becomes apparent when you 234 * try to recover gracefully from saved state that is no longer 235 * valid (stale segment register values in particular). With the 236 * hardware task-switch, there is no way to fix up bad state in 237 * a reasonable manner. 238 * 239 * The fact that Intel documents the hardware task-switching to 240 * be slow is a fairly red herring - this code is not noticeably 241 * faster. However, there _is_ some room for improvement here, 242 * so the performance issues may eventually be a valid point. 243 * More important, however, is the fact that this allows us much 244 * more flexibility. 245 * 246 * The return value (in %ax) will be the "prev" task after 247 * the task-switch, and shows up in ret_from_fork in entry.S, 248 * for example. 249 */ 250 __visible __notrace_funcgraph struct task_struct * 251 __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 252 { 253 struct thread_struct *prev = &prev_p->thread, 254 *next = &next_p->thread; 255 int cpu = smp_processor_id(); 256 struct tss_struct *tss = &per_cpu(init_tss, cpu); 257 fpu_switch_t fpu; 258 259 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ 260 261 fpu = switch_fpu_prepare(prev_p, next_p, cpu); 262 263 /* 264 * Reload esp0. 265 */ 266 load_sp0(tss, next); 267 268 /* 269 * Save away %gs. No need to save %fs, as it was saved on the 270 * stack on entry. No need to save %es and %ds, as those are 271 * always kernel segments while inside the kernel. Doing this 272 * before setting the new TLS descriptors avoids the situation 273 * where we temporarily have non-reloadable segments in %fs 274 * and %gs. This could be an issue if the NMI handler ever 275 * used %fs or %gs (it does not today), or if the kernel is 276 * running inside of a hypervisor layer. 277 */ 278 lazy_save_gs(prev->gs); 279 280 /* 281 * Load the per-thread Thread-Local Storage descriptor. 282 */ 283 load_TLS(next, cpu); 284 285 /* 286 * Restore IOPL if needed. In normal use, the flags restore 287 * in the switch assembly will handle this. But if the kernel 288 * is running virtualized at a non-zero CPL, the popf will 289 * not restore flags, so it must be done in a separate step. 290 */ 291 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl)) 292 set_iopl_mask(next->iopl); 293 294 /* 295 * If it were not for PREEMPT_ACTIVE we could guarantee that the 296 * preempt_count of all tasks was equal here and this would not be 297 * needed. 298 */ 299 task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count); 300 this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count); 301 302 /* 303 * Now maybe handle debug registers and/or IO bitmaps 304 */ 305 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV || 306 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT)) 307 __switch_to_xtra(prev_p, next_p, tss); 308 309 /* 310 * Leave lazy mode, flushing any hypercalls made here. 311 * This must be done before restoring TLS segments so 312 * the GDT and LDT are properly updated, and must be 313 * done before math_state_restore, so the TS bit is up 314 * to date. 315 */ 316 arch_end_context_switch(next_p); 317 318 /* 319 * Restore %gs if needed (which is common) 320 */ 321 if (prev->gs | next->gs) 322 lazy_load_gs(next->gs); 323 324 switch_fpu_finish(next_p, fpu); 325 326 this_cpu_write(current_task, next_p); 327 328 return prev_p; 329 } 330 331 #define top_esp (THREAD_SIZE - sizeof(unsigned long)) 332 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) 333 334 unsigned long get_wchan(struct task_struct *p) 335 { 336 unsigned long bp, sp, ip; 337 unsigned long stack_page; 338 int count = 0; 339 if (!p || p == current || p->state == TASK_RUNNING) 340 return 0; 341 stack_page = (unsigned long)task_stack_page(p); 342 sp = p->thread.sp; 343 if (!stack_page || sp < stack_page || sp > top_esp+stack_page) 344 return 0; 345 /* include/asm-i386/system.h:switch_to() pushes bp last. */ 346 bp = *(unsigned long *) sp; 347 do { 348 if (bp < stack_page || bp > top_ebp+stack_page) 349 return 0; 350 ip = *(unsigned long *) (bp+4); 351 if (!in_sched_functions(ip)) 352 return ip; 353 bp = *(unsigned long *) bp; 354 } while (count++ < 16); 355 return 0; 356 } 357 358