xref: /linux/arch/x86/kernel/process_32.c (revision 32786fdc9506aeba98278c1844d4bfb766863832)
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/export.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/syscalls.h>
53 #include <asm/debugreg.h>
54 #include <asm/switch_to.h>
55 #include <asm/vm86.h>
56 
57 void __show_regs(struct pt_regs *regs, int all)
58 {
59 	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
60 	unsigned long d0, d1, d2, d3, d6, d7;
61 	unsigned long sp;
62 	unsigned short ss, gs;
63 
64 	if (user_mode(regs)) {
65 		sp = regs->sp;
66 		ss = regs->ss & 0xffff;
67 		gs = get_user_gs(regs);
68 	} else {
69 		sp = kernel_stack_pointer(regs);
70 		savesegment(ss, ss);
71 		savesegment(gs, gs);
72 	}
73 
74 	printk(KERN_DEFAULT "EIP: %pS\n", (void *)regs->ip);
75 	printk(KERN_DEFAULT "EFLAGS: %08lx CPU: %d\n", regs->flags,
76 		smp_processor_id());
77 
78 	printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
79 		regs->ax, regs->bx, regs->cx, regs->dx);
80 	printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
81 		regs->si, regs->di, regs->bp, sp);
82 	printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
83 	       (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
84 
85 	if (!all)
86 		return;
87 
88 	cr0 = read_cr0();
89 	cr2 = read_cr2();
90 	cr3 = read_cr3();
91 	cr4 = __read_cr4();
92 	printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
93 			cr0, cr2, cr3, cr4);
94 
95 	get_debugreg(d0, 0);
96 	get_debugreg(d1, 1);
97 	get_debugreg(d2, 2);
98 	get_debugreg(d3, 3);
99 	get_debugreg(d6, 6);
100 	get_debugreg(d7, 7);
101 
102 	/* Only print out debug registers if they are in their non-default state. */
103 	if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
104 	    (d6 == DR6_RESERVED) && (d7 == 0x400))
105 		return;
106 
107 	printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
108 			d0, d1, d2, d3);
109 	printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
110 			d6, d7);
111 }
112 
113 void release_thread(struct task_struct *dead_task)
114 {
115 	BUG_ON(dead_task->mm);
116 	release_vm86_irqs(dead_task);
117 }
118 
119 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
120 	unsigned long arg, struct task_struct *p, unsigned long tls)
121 {
122 	struct pt_regs *childregs = task_pt_regs(p);
123 	struct fork_frame *fork_frame = container_of(childregs, struct fork_frame, regs);
124 	struct inactive_task_frame *frame = &fork_frame->frame;
125 	struct task_struct *tsk;
126 	int err;
127 
128 	frame->bp = 0;
129 	frame->ret_addr = (unsigned long) ret_from_fork;
130 	p->thread.sp = (unsigned long) fork_frame;
131 	p->thread.sp0 = (unsigned long) (childregs+1);
132 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
133 
134 	if (unlikely(p->flags & PF_KTHREAD)) {
135 		/* kernel thread */
136 		memset(childregs, 0, sizeof(struct pt_regs));
137 		frame->bx = sp;		/* function */
138 		frame->di = arg;
139 		p->thread.io_bitmap_ptr = NULL;
140 		return 0;
141 	}
142 	frame->bx = 0;
143 	*childregs = *current_pt_regs();
144 	childregs->ax = 0;
145 	if (sp)
146 		childregs->sp = sp;
147 
148 	task_user_gs(p) = get_user_gs(current_pt_regs());
149 
150 	p->thread.io_bitmap_ptr = NULL;
151 	tsk = current;
152 	err = -ENOMEM;
153 
154 	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
155 		p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
156 						IO_BITMAP_BYTES, GFP_KERNEL);
157 		if (!p->thread.io_bitmap_ptr) {
158 			p->thread.io_bitmap_max = 0;
159 			return -ENOMEM;
160 		}
161 		set_tsk_thread_flag(p, TIF_IO_BITMAP);
162 	}
163 
164 	err = 0;
165 
166 	/*
167 	 * Set a new TLS for the child thread?
168 	 */
169 	if (clone_flags & CLONE_SETTLS)
170 		err = do_set_thread_area(p, -1,
171 			(struct user_desc __user *)tls, 0);
172 
173 	if (err && p->thread.io_bitmap_ptr) {
174 		kfree(p->thread.io_bitmap_ptr);
175 		p->thread.io_bitmap_max = 0;
176 	}
177 	return err;
178 }
179 
180 void
181 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
182 {
183 	set_user_gs(regs, 0);
184 	regs->fs		= 0;
185 	regs->ds		= __USER_DS;
186 	regs->es		= __USER_DS;
187 	regs->ss		= __USER_DS;
188 	regs->cs		= __USER_CS;
189 	regs->ip		= new_ip;
190 	regs->sp		= new_sp;
191 	regs->flags		= X86_EFLAGS_IF;
192 	force_iret();
193 }
194 EXPORT_SYMBOL_GPL(start_thread);
195 
196 
197 /*
198  *	switch_to(x,y) should switch tasks from x to y.
199  *
200  * We fsave/fwait so that an exception goes off at the right time
201  * (as a call from the fsave or fwait in effect) rather than to
202  * the wrong process. Lazy FP saving no longer makes any sense
203  * with modern CPU's, and this simplifies a lot of things (SMP
204  * and UP become the same).
205  *
206  * NOTE! We used to use the x86 hardware context switching. The
207  * reason for not using it any more becomes apparent when you
208  * try to recover gracefully from saved state that is no longer
209  * valid (stale segment register values in particular). With the
210  * hardware task-switch, there is no way to fix up bad state in
211  * a reasonable manner.
212  *
213  * The fact that Intel documents the hardware task-switching to
214  * be slow is a fairly red herring - this code is not noticeably
215  * faster. However, there _is_ some room for improvement here,
216  * so the performance issues may eventually be a valid point.
217  * More important, however, is the fact that this allows us much
218  * more flexibility.
219  *
220  * The return value (in %ax) will be the "prev" task after
221  * the task-switch, and shows up in ret_from_fork in entry.S,
222  * for example.
223  */
224 __visible __notrace_funcgraph struct task_struct *
225 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
226 {
227 	struct thread_struct *prev = &prev_p->thread,
228 			     *next = &next_p->thread;
229 	struct fpu *prev_fpu = &prev->fpu;
230 	struct fpu *next_fpu = &next->fpu;
231 	int cpu = smp_processor_id();
232 	struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
233 
234 	/* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
235 
236 	switch_fpu_prepare(prev_fpu, cpu);
237 
238 	/*
239 	 * Save away %gs. No need to save %fs, as it was saved on the
240 	 * stack on entry.  No need to save %es and %ds, as those are
241 	 * always kernel segments while inside the kernel.  Doing this
242 	 * before setting the new TLS descriptors avoids the situation
243 	 * where we temporarily have non-reloadable segments in %fs
244 	 * and %gs.  This could be an issue if the NMI handler ever
245 	 * used %fs or %gs (it does not today), or if the kernel is
246 	 * running inside of a hypervisor layer.
247 	 */
248 	lazy_save_gs(prev->gs);
249 
250 	/*
251 	 * Load the per-thread Thread-Local Storage descriptor.
252 	 */
253 	load_TLS(next, cpu);
254 
255 	/*
256 	 * Restore IOPL if needed.  In normal use, the flags restore
257 	 * in the switch assembly will handle this.  But if the kernel
258 	 * is running virtualized at a non-zero CPL, the popf will
259 	 * not restore flags, so it must be done in a separate step.
260 	 */
261 	if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
262 		set_iopl_mask(next->iopl);
263 
264 	/*
265 	 * Now maybe handle debug registers and/or IO bitmaps
266 	 */
267 	if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
268 		     task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
269 		__switch_to_xtra(prev_p, next_p, tss);
270 
271 	/*
272 	 * Leave lazy mode, flushing any hypercalls made here.
273 	 * This must be done before restoring TLS segments so
274 	 * the GDT and LDT are properly updated, and must be
275 	 * done before fpu__restore(), so the TS bit is up
276 	 * to date.
277 	 */
278 	arch_end_context_switch(next_p);
279 
280 	/*
281 	 * Reload esp0 and cpu_current_top_of_stack.  This changes
282 	 * current_thread_info().
283 	 */
284 	load_sp0(tss, next);
285 	this_cpu_write(cpu_current_top_of_stack,
286 		       (unsigned long)task_stack_page(next_p) +
287 		       THREAD_SIZE);
288 
289 	/*
290 	 * Restore %gs if needed (which is common)
291 	 */
292 	if (prev->gs | next->gs)
293 		lazy_load_gs(next->gs);
294 
295 	switch_fpu_finish(next_fpu, cpu);
296 
297 	this_cpu_write(current_task, next_p);
298 
299 	return prev_p;
300 }
301