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/mc146818rtc.h>
28 #include <linux/module.h>
29 #include <linux/kallsyms.h>
30 #include <linux/ptrace.h>
31 #include <linux/personality.h>
32 #include <linux/percpu.h>
33 #include <linux/prctl.h>
34 #include <linux/ftrace.h>
35 #include <linux/uaccess.h>
36 #include <linux/io.h>
37 #include <linux/kdebug.h>
38
39 #include <asm/pgtable.h>
40 #include <asm/ldt.h>
41 #include <asm/processor.h>
42 #include <asm/fpu/internal.h>
43 #include <asm/desc.h>
44 #ifdef CONFIG_MATH_EMULATION
45 #include <asm/math_emu.h>
46 #endif
47
48 #include <linux/err.h>
49
50 #include <asm/tlbflush.h>
51 #include <asm/cpu.h>
52 #include <asm/idle.h>
53 #include <asm/syscalls.h>
54 #include <asm/debugreg.h>
55 #include <asm/switch_to.h>
56 #include <asm/vm86.h>
57
58 #include "process.h"
59
60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
61 asmlinkage void ret_from_kernel_thread(void) __asm__("ret_from_kernel_thread");
62
63 /*
64 * Return saved PC of a blocked thread.
65 */
thread_saved_pc(struct task_struct * tsk)66 unsigned long thread_saved_pc(struct task_struct *tsk)
67 {
68 return ((unsigned long *)tsk->thread.sp)[3];
69 }
70
__show_regs(struct pt_regs * regs,int all)71 void __show_regs(struct pt_regs *regs, int all)
72 {
73 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
74 unsigned long d0, d1, d2, d3, d6, d7;
75 unsigned long sp;
76 unsigned short ss, gs;
77
78 if (user_mode(regs)) {
79 sp = regs->sp;
80 ss = regs->ss & 0xffff;
81 gs = get_user_gs(regs);
82 } else {
83 sp = kernel_stack_pointer(regs);
84 savesegment(ss, ss);
85 savesegment(gs, gs);
86 }
87
88 printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
89 (u16)regs->cs, regs->ip, regs->flags,
90 smp_processor_id());
91 print_symbol("EIP is at %s\n", regs->ip);
92
93 printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
94 regs->ax, regs->bx, regs->cx, regs->dx);
95 printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
96 regs->si, regs->di, regs->bp, sp);
97 printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
98 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
99
100 if (!all)
101 return;
102
103 cr0 = read_cr0();
104 cr2 = read_cr2();
105 cr3 = read_cr3();
106 cr4 = __read_cr4_safe();
107 printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
108 cr0, cr2, cr3, cr4);
109
110 get_debugreg(d0, 0);
111 get_debugreg(d1, 1);
112 get_debugreg(d2, 2);
113 get_debugreg(d3, 3);
114 get_debugreg(d6, 6);
115 get_debugreg(d7, 7);
116
117 /* Only print out debug registers if they are in their non-default state. */
118 if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
119 (d6 == DR6_RESERVED) && (d7 == 0x400))
120 return;
121
122 printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
123 d0, d1, d2, d3);
124 printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
125 d6, d7);
126 }
127
release_thread(struct task_struct * dead_task)128 void release_thread(struct task_struct *dead_task)
129 {
130 BUG_ON(dead_task->mm);
131 release_vm86_irqs(dead_task);
132 }
133
copy_thread_tls(unsigned long clone_flags,unsigned long sp,unsigned long arg,struct task_struct * p,unsigned long tls)134 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
135 unsigned long arg, struct task_struct *p, unsigned long tls)
136 {
137 struct pt_regs *childregs = task_pt_regs(p);
138 struct task_struct *tsk;
139 int err;
140
141 p->thread.sp = (unsigned long) childregs;
142 p->thread.sp0 = (unsigned long) (childregs+1);
143 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
144
145 if (unlikely(p->flags & PF_KTHREAD)) {
146 /* kernel thread */
147 memset(childregs, 0, sizeof(struct pt_regs));
148 p->thread.ip = (unsigned long) ret_from_kernel_thread;
149 task_user_gs(p) = __KERNEL_STACK_CANARY;
150 childregs->ds = __USER_DS;
151 childregs->es = __USER_DS;
152 childregs->fs = __KERNEL_PERCPU;
153 childregs->bx = sp; /* function */
154 childregs->bp = arg;
155 childregs->orig_ax = -1;
156 childregs->cs = __KERNEL_CS | get_kernel_rpl();
157 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
158 p->thread.io_bitmap_ptr = NULL;
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.io_bitmap_ptr = NULL;
170 tsk = current;
171 err = -ENOMEM;
172
173 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
174 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
175 IO_BITMAP_BYTES, GFP_KERNEL);
176 if (!p->thread.io_bitmap_ptr) {
177 p->thread.io_bitmap_max = 0;
178 return -ENOMEM;
179 }
180 set_tsk_thread_flag(p, TIF_IO_BITMAP);
181 }
182
183 err = 0;
184
185 /*
186 * Set a new TLS for the child thread?
187 */
188 if (clone_flags & CLONE_SETTLS)
189 err = do_set_thread_area(p, -1,
190 (struct user_desc __user *)tls, 0);
191
192 if (err && p->thread.io_bitmap_ptr) {
193 kfree(p->thread.io_bitmap_ptr);
194 p->thread.io_bitmap_max = 0;
195 }
196 return err;
197 }
198
199 void
start_thread(struct pt_regs * regs,unsigned long new_ip,unsigned long new_sp)200 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
201 {
202 set_user_gs(regs, 0);
203 regs->fs = 0;
204 regs->ds = __USER_DS;
205 regs->es = __USER_DS;
206 regs->ss = __USER_DS;
207 regs->cs = __USER_CS;
208 regs->ip = new_ip;
209 regs->sp = new_sp;
210 regs->flags = X86_EFLAGS_IF;
211 force_iret();
212 }
213 EXPORT_SYMBOL_GPL(start_thread);
214
215
216 /*
217 * switch_to(x,y) should switch tasks from x to y.
218 *
219 * We fsave/fwait so that an exception goes off at the right time
220 * (as a call from the fsave or fwait in effect) rather than to
221 * the wrong process. Lazy FP saving no longer makes any sense
222 * with modern CPU's, and this simplifies a lot of things (SMP
223 * and UP become the same).
224 *
225 * NOTE! We used to use the x86 hardware context switching. The
226 * reason for not using it any more becomes apparent when you
227 * try to recover gracefully from saved state that is no longer
228 * valid (stale segment register values in particular). With the
229 * hardware task-switch, there is no way to fix up bad state in
230 * a reasonable manner.
231 *
232 * The fact that Intel documents the hardware task-switching to
233 * be slow is a fairly red herring - this code is not noticeably
234 * faster. However, there _is_ some room for improvement here,
235 * so the performance issues may eventually be a valid point.
236 * More important, however, is the fact that this allows us much
237 * more flexibility.
238 *
239 * The return value (in %ax) will be the "prev" task after
240 * the task-switch, and shows up in ret_from_fork in entry.S,
241 * for example.
242 */
243 __visible __notrace_funcgraph struct task_struct *
__switch_to(struct task_struct * prev_p,struct task_struct * next_p)244 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
245 {
246 struct thread_struct *prev = &prev_p->thread,
247 *next = &next_p->thread;
248 struct fpu *prev_fpu = &prev->fpu;
249 struct fpu *next_fpu = &next->fpu;
250 int cpu = smp_processor_id();
251 struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
252 fpu_switch_t fpu_switch;
253
254 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
255
256 fpu_switch = switch_fpu_prepare(prev_fpu, next_fpu, cpu);
257
258 /*
259 * Save away %gs. No need to save %fs, as it was saved on the
260 * stack on entry. No need to save %es and %ds, as those are
261 * always kernel segments while inside the kernel. Doing this
262 * before setting the new TLS descriptors avoids the situation
263 * where we temporarily have non-reloadable segments in %fs
264 * and %gs. This could be an issue if the NMI handler ever
265 * used %fs or %gs (it does not today), or if the kernel is
266 * running inside of a hypervisor layer.
267 */
268 lazy_save_gs(prev->gs);
269
270 /*
271 * Load the per-thread Thread-Local Storage descriptor.
272 */
273 load_TLS(next, cpu);
274
275 /*
276 * Restore IOPL if needed. In normal use, the flags restore
277 * in the switch assembly will handle this. But if the kernel
278 * is running virtualized at a non-zero CPL, the popf will
279 * not restore flags, so it must be done in a separate step.
280 */
281 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
282 set_iopl_mask(next->iopl);
283
284 switch_to_extra(prev_p, next_p);
285
286 /*
287 * Leave lazy mode, flushing any hypercalls made here.
288 * This must be done before restoring TLS segments so
289 * the GDT and LDT are properly updated, and must be
290 * done before fpu__restore(), so the TS bit is up
291 * to date.
292 */
293 arch_end_context_switch(next_p);
294
295 /*
296 * Reload esp0 and cpu_current_top_of_stack. This changes
297 * current_thread_info().
298 */
299 load_sp0(tss, next);
300 this_cpu_write(cpu_current_top_of_stack,
301 (unsigned long)task_stack_page(next_p) +
302 THREAD_SIZE);
303
304 /*
305 * Restore %gs if needed (which is common)
306 */
307 if (prev->gs | next->gs)
308 lazy_load_gs(next->gs);
309
310 switch_fpu_finish(next_fpu, fpu_switch);
311
312 this_cpu_write(current_task, next_p);
313
314 return prev_p;
315 }
316