xref: /openbmc/linux/arch/x86/kernel/process_32.c (revision fea88a0c)
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 
61 /*
62  * Return saved PC of a blocked thread.
63  */
64 unsigned long thread_saved_pc(struct task_struct *tsk)
65 {
66 	return ((unsigned long *)tsk->thread.sp)[3];
67 }
68 
69 void __show_regs(struct pt_regs *regs, int all)
70 {
71 	unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
72 	unsigned long d0, d1, d2, d3, d6, d7;
73 	unsigned long sp;
74 	unsigned short ss, gs;
75 
76 	if (user_mode_vm(regs)) {
77 		sp = regs->sp;
78 		ss = regs->ss & 0xffff;
79 		gs = get_user_gs(regs);
80 	} else {
81 		sp = kernel_stack_pointer(regs);
82 		savesegment(ss, ss);
83 		savesegment(gs, gs);
84 	}
85 
86 	show_regs_common();
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 	printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
115 			d0, d1, d2, d3);
116 
117 	get_debugreg(d6, 6);
118 	get_debugreg(d7, 7);
119 	printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
120 			d6, d7);
121 }
122 
123 void release_thread(struct task_struct *dead_task)
124 {
125 	BUG_ON(dead_task->mm);
126 	release_vm86_irqs(dead_task);
127 }
128 
129 /*
130  * This gets called before we allocate a new thread and copy
131  * the current task into it.
132  */
133 void prepare_to_copy(struct task_struct *tsk)
134 {
135 	unlazy_fpu(tsk);
136 }
137 
138 int copy_thread(unsigned long clone_flags, unsigned long sp,
139 	unsigned long unused,
140 	struct task_struct *p, struct pt_regs *regs)
141 {
142 	struct pt_regs *childregs;
143 	struct task_struct *tsk;
144 	int err;
145 
146 	childregs = task_pt_regs(p);
147 	*childregs = *regs;
148 	childregs->ax = 0;
149 	childregs->sp = sp;
150 
151 	p->thread.sp = (unsigned long) childregs;
152 	p->thread.sp0 = (unsigned long) (childregs+1);
153 
154 	p->thread.ip = (unsigned long) ret_from_fork;
155 
156 	task_user_gs(p) = get_user_gs(regs);
157 
158 	p->fpu_counter = 0;
159 	p->thread.io_bitmap_ptr = NULL;
160 	tsk = current;
161 	err = -ENOMEM;
162 
163 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
164 
165 	if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
166 		p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
167 						IO_BITMAP_BYTES, GFP_KERNEL);
168 		if (!p->thread.io_bitmap_ptr) {
169 			p->thread.io_bitmap_max = 0;
170 			return -ENOMEM;
171 		}
172 		set_tsk_thread_flag(p, TIF_IO_BITMAP);
173 	}
174 
175 	err = 0;
176 
177 	/*
178 	 * Set a new TLS for the child thread?
179 	 */
180 	if (clone_flags & CLONE_SETTLS)
181 		err = do_set_thread_area(p, -1,
182 			(struct user_desc __user *)childregs->si, 0);
183 
184 	if (err && p->thread.io_bitmap_ptr) {
185 		kfree(p->thread.io_bitmap_ptr);
186 		p->thread.io_bitmap_max = 0;
187 	}
188 	return err;
189 }
190 
191 void
192 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
193 {
194 	set_user_gs(regs, 0);
195 	regs->fs		= 0;
196 	regs->ds		= __USER_DS;
197 	regs->es		= __USER_DS;
198 	regs->ss		= __USER_DS;
199 	regs->cs		= __USER_CS;
200 	regs->ip		= new_ip;
201 	regs->sp		= new_sp;
202 	/*
203 	 * Free the old FP and other extended state
204 	 */
205 	free_thread_xstate(current);
206 }
207 EXPORT_SYMBOL_GPL(start_thread);
208 
209 
210 /*
211  *	switch_to(x,y) should switch tasks from x to y.
212  *
213  * We fsave/fwait so that an exception goes off at the right time
214  * (as a call from the fsave or fwait in effect) rather than to
215  * the wrong process. Lazy FP saving no longer makes any sense
216  * with modern CPU's, and this simplifies a lot of things (SMP
217  * and UP become the same).
218  *
219  * NOTE! We used to use the x86 hardware context switching. The
220  * reason for not using it any more becomes apparent when you
221  * try to recover gracefully from saved state that is no longer
222  * valid (stale segment register values in particular). With the
223  * hardware task-switch, there is no way to fix up bad state in
224  * a reasonable manner.
225  *
226  * The fact that Intel documents the hardware task-switching to
227  * be slow is a fairly red herring - this code is not noticeably
228  * faster. However, there _is_ some room for improvement here,
229  * so the performance issues may eventually be a valid point.
230  * More important, however, is the fact that this allows us much
231  * more flexibility.
232  *
233  * The return value (in %ax) will be the "prev" task after
234  * the task-switch, and shows up in ret_from_fork in entry.S,
235  * for example.
236  */
237 __notrace_funcgraph struct task_struct *
238 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
239 {
240 	struct thread_struct *prev = &prev_p->thread,
241 				 *next = &next_p->thread;
242 	int cpu = smp_processor_id();
243 	struct tss_struct *tss = &per_cpu(init_tss, cpu);
244 	fpu_switch_t fpu;
245 
246 	/* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
247 
248 	fpu = switch_fpu_prepare(prev_p, next_p, cpu);
249 
250 	/*
251 	 * Reload esp0.
252 	 */
253 	load_sp0(tss, next);
254 
255 	/*
256 	 * Save away %gs. No need to save %fs, as it was saved on the
257 	 * stack on entry.  No need to save %es and %ds, as those are
258 	 * always kernel segments while inside the kernel.  Doing this
259 	 * before setting the new TLS descriptors avoids the situation
260 	 * where we temporarily have non-reloadable segments in %fs
261 	 * and %gs.  This could be an issue if the NMI handler ever
262 	 * used %fs or %gs (it does not today), or if the kernel is
263 	 * running inside of a hypervisor layer.
264 	 */
265 	lazy_save_gs(prev->gs);
266 
267 	/*
268 	 * Load the per-thread Thread-Local Storage descriptor.
269 	 */
270 	load_TLS(next, cpu);
271 
272 	/*
273 	 * Restore IOPL if needed.  In normal use, the flags restore
274 	 * in the switch assembly will handle this.  But if the kernel
275 	 * is running virtualized at a non-zero CPL, the popf will
276 	 * not restore flags, so it must be done in a separate step.
277 	 */
278 	if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
279 		set_iopl_mask(next->iopl);
280 
281 	/*
282 	 * Now maybe handle debug registers and/or IO bitmaps
283 	 */
284 	if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
285 		     task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
286 		__switch_to_xtra(prev_p, next_p, tss);
287 
288 	/*
289 	 * Leave lazy mode, flushing any hypercalls made here.
290 	 * This must be done before restoring TLS segments so
291 	 * the GDT and LDT are properly updated, and must be
292 	 * done before math_state_restore, so the TS bit is up
293 	 * to date.
294 	 */
295 	arch_end_context_switch(next_p);
296 
297 	/*
298 	 * Restore %gs if needed (which is common)
299 	 */
300 	if (prev->gs | next->gs)
301 		lazy_load_gs(next->gs);
302 
303 	switch_fpu_finish(next_p, fpu);
304 
305 	percpu_write(current_task, next_p);
306 
307 	return prev_p;
308 }
309 
310 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
311 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
312 
313 unsigned long get_wchan(struct task_struct *p)
314 {
315 	unsigned long bp, sp, ip;
316 	unsigned long stack_page;
317 	int count = 0;
318 	if (!p || p == current || p->state == TASK_RUNNING)
319 		return 0;
320 	stack_page = (unsigned long)task_stack_page(p);
321 	sp = p->thread.sp;
322 	if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
323 		return 0;
324 	/* include/asm-i386/system.h:switch_to() pushes bp last. */
325 	bp = *(unsigned long *) sp;
326 	do {
327 		if (bp < stack_page || bp > top_ebp+stack_page)
328 			return 0;
329 		ip = *(unsigned long *) (bp+4);
330 		if (!in_sched_functions(ip))
331 			return ip;
332 		bp = *(unsigned long *) bp;
333 	} while (count++ < 16);
334 	return 0;
335 }
336 
337